The Use of Gas-Sensor Arrays in the Detection of Bole and Root Decays in Living Trees

http://www.sensorsportal.com/HTML/DIGEST/Submission.htm

SSeennssoorrss && TTrraannssdduucceerrss

IInntteerrnnaattiioonnaall OOffffiicciiaall JJoouurrnnaall ooff tthhee IInntteerrnnaattiioonnaall FFrreeqquueennccyy SSeennssoorr AAssssoocciiaattiioonn ((IIFFSSAA)) DDeevvootteedd ttoo

RReesseeaarrcchh aanndd DDeevveellooppmmeenntt ooff SSeennssoorrss aanndd TTrraannssdduucceerrss

VVoolluummee 119933,, IIssssuuee 1100,, OOccttoobbeerr 22001155

Editor-in-Chief Prof., Dr. Sergey Y. YURISH

IFSA Publishing: Barcelona Toronto

http://www.sensorsportal.com/

Copyright 2015 IFSA Publishing, S. L. All rights reserved. This journal and the individual contributions in it are protected under copyright by IFSA Publishing, and the following terms and conditions apply to their use: Photocopying: Single photocopies of single articles may be made for personal use as allowed by national copyright laws. Permission of the Publisher and payment of a fee is required for all other photocopying, including multiple or systematic copyright, copyright for advertising or promotional purposes, resale, and all forms of document delivery. Derivative Works: Subscribers may reproduce tables of contents or prepare list of articles including abstract for internal circulation within their institutions. Permission of the Publisher is required for resale or distribution outside the institution. Permission of the Publisher is required for all other derivative works, including compilations and translations. Authors' copies of Sensors & Transducers journal and articles published in it are for personal use only. Address permissions requests to: IFSA Publisher by e-mail: [email protected] Notice: No responsibility is assumed by the Publisher for any injury and/or damage to persons or property as a matter of products liability, negligence or otherwise, or from any use or operation of any methods, products, instructions or ideas contained in the material herein. Printed in the USA.

SSeennssoorrss && TTrraannssdduucceerrss

Volume 193, Issue 10, October 2015 www.sensorsportal.com

e-ISSN 1726-5479 ISSN 2306-8515

Editors-in-Chief: Professor, Dr. Sergey Y. Yurish, tel.: +34 93 4137941, e-mail: [email protected]

Editors for Western Europe Meijer, Gerard C.M., Delft Univ. of Technology, The Netherlands Ferrari, Vittorio, Universitá di Brescia, Italy Mescheder, Ulrich, Univ. of Applied Sciences, Furtwangen, Germany

Editor for Eastern Europe Sachenko, Anatoly, Ternopil National Economic University, Ukraine

Editors for North America Katz, Evgeny, Clarkson University, USA Datskos, Panos G., Oak Ridge National Laboratory, USA Fabien, J. Josse, Marquette University, USA

Editor for Africa Maki K., Habib, American University in Cairo, Egypt

Editors South America Costa-Felix, Rodrigo, Inmetro, Brazil Walsoe de Reca, Noemi Elisabeth, CINSO-CITEDEF UNIDEF (MINDEF-CONICET), Argentina Editors for Asia Ohyama, Shinji, Tokyo Institute of Technology, Japan Zhengbing, Hu, Huazhong Univ. of Science and Technol., China Li, Gongfa, Wuhan Univ. of Science and Technology, China Editor for Asia-Pacific Mukhopadhyay, Subhas, Massey University, New Zealand

Editorial Board

Abdul Rahim, Ruzairi, Universiti Teknologi, Malaysia Abramchuk, George, Measur. Tech. & Advanced Applications, Canada Aluri, Geetha S., Globalfoundries, USA Ascoli, Giorgio, George Mason University, USA Atalay, Selcuk, Inonu University, Turkey Atghiaee, Ahmad, University of Tehran, Iran Augutis, Vygantas, Kaunas University of Technology, Lithuania Ayesh, Aladdin, De Montfort University, UK Baliga, Shankar, B., General Monitors, USA Barlingay, Ravindra, Larsen & Toubro - Technology Services, India Basu, Sukumar, Jadavpur University, India Booranawong, Apidet, Prince of Songkla University, Thailand Bousbia-Salah, Mounir, University of Annaba, Algeria Bouvet, Marcel, University of Burgundy, France Campanella, Luigi, University La Sapienza, Italy Carvalho, Vitor, Minho University, Portugal Changhai, Ru, Harbin Engineering University, China Chen, Wei, Hefei University of Technology, China Cheng-Ta, Chiang, National Chia-Yi University, Taiwan Cherstvy, Andrey, University of Potsdam, Germany Chung, Wen-Yaw, Chung Yuan Christian University, Taiwan Cortes, Camilo A., Universidad Nacional de Colombia, Colombia D'Amico, Arnaldo, Università di Tor Vergata, Italy De Stefano, Luca, Institute for Microelectronics and Microsystem, Italy Ding, Jianning, Changzhou University, China Djordjevich, Alexandar, City University of Hong Kong, Hong Kong Donato, Nicola, University of Messina, Italy Dong, Feng, Tianjin University, China Erkmen, Aydan M., Middle East Technical University, Turkey Fezari, Mohamed, Badji Mokhtar Annaba University, Algeria Gaura, Elena, Coventry University, UK Gole, James, Georgia Institute of Technology, USA Gong, Hao, National University of Singapore, Singapore Gonzalez de la Rosa, Juan Jose, University of Cadiz, Spain Goswami, Amarjyoti, Kaziranga University, India Guillet, Bruno, University of Caen, France Hadjiloucas, Sillas, The University of Reading, UK Hao, Shiying, Michigan State University, USA Hui, David, University of New Orleans, USA Jaffrezic-Renault, Nicole, Claude Bernard University Lyon 1, France Jamil, Mohammad, Qatar University, Qatar Kaniusas, Eugenijus, Vienna University of Technology, Austria Kim, Min Young, Kyungpook National University, Korea Kumar, Arun, University of Delaware, USA Lay-Ekuakille, Aime, University of Lecce, Italy Li, Fengyuan, HARMAN International, USA Li, Jingsong, Anhui University, China Li, Si, GE Global Research Center, USA Lin, Paul, Cleveland State University, USA Liu, Aihua, Chinese Academy of Sciences, China Liu, Chenglian, Long Yan University, China Liu, Fei, City College of New York, USA Mahadi, Muhammad, University Tun Hussein Onn Malaysia, Malaysia

Mansor, Muhammad Naufal, University Malaysia Perlis, Malaysia Marquez, Alfredo, Centro de Investigacion en Materiales Avanzados, Mexico Mishra, Vivekanand, National Institute of Technology, India Moghavvemi, Mahmoud, University of Malaya, Malaysia Morello, Rosario, University "Mediterranea" of Reggio Calabria, Italy Mulla, Imtiaz Sirajuddin, National Chemical Laboratory, Pune, India Nabok, Aleksey, Sheffield Hallam University, UK Neshkova, Milka, Bulgarian Academy of Sciences, Bulgaria Passaro, Vittorio M. N., Politecnico di Bari, Italy Patil, Devidas Ramrao, R. L. College, Parola, India Penza, Michele, ENEA, Italy Pereira, Jose Miguel, Instituto Politecnico de Setebal, Portugal Pillarisetti, Anand, Sensata Technologies Inc, USA Pogacnik, Lea, University of Ljubljana, Slovenia Pullini, Daniele, Centro Ricerche FIAT, Italy Qiu, Liang, Avago Technologies, USA Reig, Candid, University of Valencia, Spain Restivo, Maria Teresa, University of Porto, Portugal Rodríguez Martínez, Angel, Universidad Politécnica de Cataluña, Spain Sadana, Ajit, University of Mississippi, USA Sadeghian Marnani, Hamed, TU Delft, The Netherlands Sapozhnikova, Ksenia, D. I. Mendeleyev Institute for Metrology, Russia Singhal, Subodh Kumar, National Physical Laboratory, India Shah, Kriyang, La Trobe University, Australia Shi, Wendian, California Institute of Technology, USA Shmaliy, Yuriy, Guanajuato University, Mexico Song, Xu, An Yang Normal University, China Srivastava, Arvind K., Systron Donner Inertial, USA Stefanescu, Dan Mihai, Romanian Measurement Society, Romania Sumriddetchkajorn, Sarun, Nat. Electr. & Comp. Tech. Center, Thailand Sun, Zhiqiang, Central South University, China Sysoev, Victor, Saratov State Technical University, Russia Thirunavukkarasu, I., Manipal University Karnataka, India Thomas, Sadiq, Heriot Watt University, Edinburgh, UK Tian, Lei, Xidian University, China Tianxing, Chu, Research Center for Surveying & Mapping, Beijing, China Vanga, Kumar L., ePack, Inc., USA Vazquez, Carmen, Universidad Carlos III Madrid, Spain Wang, Jiangping, Xian Shiyou University, China Wang, Peng, Qualcomm Technologies, USA Wang, Zongbo, University of Kansas, USA Xu, Han, Measurement Specialties, Inc., USA Xu, Weihe, Brookhaven National Lab, USA Xue, Ning, Agiltron, Inc., USA Yang, Dongfang, National Research Council, Canada Yang, Shuang-Hua, Loughborough University, UK Yaping Dan, Harvard University, USA Yue, Xiao-Guang, Shanxi University of Chinese Traditional Medicine, China Xiao-Guang, Yue, Wuhan University of Technology, China Zakaria, Zulkarnay, University Malaysia Perlis, Malaysia Zhang, Weiping, Shanghai Jiao Tong University, China Zhang, Wenming, Shanghai Jiao Tong University, China Zhang, Yudong, Nanjing Normal University China

Sensors & Transducers Journal is a peer review international journal published monthly by International Frequency Sensor Association (IFSA). Available in both: print and electronic (printable pdf) formats. Copyright © 2015 by IFSA Publishing, S. L. All rights reserved.


SSeennssoorrss && TTrraannssdduucceerrss JJoouurrnnaall

CCoonntteennttss

Volume 193 Issue 10 October 2015

www.sensorsportal.com ISSN 2306-8515e-ISSN 1726-5479

Research Articles

Smart and Customized Electrical Conductivity Sensorfor Measurements of the Response Time from SprayersBased on Direct Injection Heitor V. Mercaldi, Caio H. Fujiwara,Elmer A. G. Peñaloza, Vilma A. Oliveira, Paulo E. Cruvinel ............................................................................................................... 1 Dynamical Capillary Rise Photonic Sensor for Testing of Diesel and Biodiesel Fuel Michal Borecki, Michael L. Korwin-Pawlowski, Mariusz Duk, Andrzej Kociubiński, Jarosław Frydrych, Przemyslaw Prus, Jan Szmidt ............................................................ 11 Novel Smart Glove Technology as a Biomechanical Monitoring Tool Brendan O’Flynn, J. T. Sanchez, S. Tedesco, B. Downes, J. Connolly, J. Condell, K. Curran ............................................................................................................................ 23 Wide Spectral Sensitivity of Monolithic a-SiC:H pi’n/pin Photodiode Outside the Visible Spectrum Manuela Vieira, Manuel Augusto Vieira, Isabel Rodrigues,Vitor Silva, Paula Louro, A. Fantoni ........................................................................................................................... 33 Sub-nanosecond Gating of Large CMOS Imagers Octavian Maciu, Wilfried Uhring, Jean-Pierre Le Normand, Jean-Baptiste Kammerer, Foudil Dadouche, Norbert Dumas ...................................................................................... 41 Superpixel Compressive Sensing Recovery of Spectral Images Sensed by Multi-patterned Focal Plane Array Detectors Yuri H. Mejia, Fernando A. Rojas, Henry Arguello ............................................................. 50 Advanced Controlled Cryogenic Ablation Using Ultrasonic Sensing System Assaf Sharon, Gabor Kosa ................................................................................................ 57 Cavity Enhanced Absorption Spectroscopy in Air Pollution Monitoring Janusz Mikołajczyk, Zbigniew Bielecki, Jacek Wojta, Sand Sylwester Chojnowski .......... 63 Design and Analysis of a Collision Detector for Hybrid Robotic Machine Tools Dan Zhang, Bin Wei ........................................................................................................... 67 Numerical Signal Analysis of Thermo-Cyclically Operated MOG Gas Sensor Arrays for Early Identification of Emissions from Overloaded Electric Cables Rolf Seifert, Hubert B. Keller, Navas Illyaskutty, Jens Knoblauch and Heinz Kohler ........ 74 Analysis of the Planar Electrode Morphology Applied to Zeolite Based Chemical Sensors Luiz Eduardo Bento Ribeiro, Glaucio Pedro de Alcântara, Cid Marcos Gonçalves Andrade, Fabiano Fruett .................................................................................................... 80


An Empirical Study for Quantification of Carcinogenic Formaldehyde by Integrating a Probabilistic Framework with Spike Latency Patterns in an Electronic Nose Muhammad Hassan, Amine Bermak, Amine Ait Si Ali and Abbes Amira .......................... 86 Alternative Processes for Manufacturing of Metal Oxide-based Potentiometric Chemosensors Winfried Vonau, Manfred Decker, Jens Zosel, Kristina Ahlborn, Frank Gerlach, David Haldan and Steffen Weissmantel ........................................................................... 93 Improvement in Humidity Sensing of Graphene Oxide by Amide Functionalization Sumita Rani, Dinesh Kumar, Mukesh Kumar .................................................................... 100 PbS Infrared Detectors: Experiment and Simulation S. Kouissa, A. Djemel, M. S. Aida, M. A. Djouadi .............................................................. 106 Amplitude to Phase Conversion Based on Analog Arcsine Synthesis for Sine-cosine Position Sensors Mohieddine Benammar, Antonio Jr. Gonzales .................................................................. 114 New Design-methodology of High-performance TDC on a Low Cost FPGA Targets Foudil Dadouche, Timothé Turko, Wilfried Uhring, Imane Malass, Norbert Dumas, Jean-Pierre Le Normand .................................................................................................... 123 Experiences in Automation and Control in Engineering Education with Real-world Based Educational Kits Filomena Soares, Celina Pinto Leão,José Machado and Vítor Carvalho .......................... 135 Improving Systems Dynamics by Means of Advanced Signal Processing – Mathematical, Laboratorial and Clinical Evaluation of Propofol Monitoring in Breathing Gas Dammon Ziaian, Philipp Rostalski, Astrid Ellen Berggreen, Sebastian Brandt, Martin Grossherr, Hartmut Gehring, Andreas Hengstenberg and Stefan Zimmermann ... 145 The Use of Gas-Sensor Arrays in the Detection of Bole and Root Decays in Living Trees: Development of a New Non-invasive Method of Sampling and Analysis Manuela Baietto, Sofia Aquaro, A. Dan Wilson, Letizia Pozzi, Daniele Bassi ................... 154 Motor Bourn Magnetic Noise Filtering for Magnetometers in Micro-Rotary Aerial Vehicles Nathan J. Unwin, Adam J. Postula..................................................................................... 161 Reflection from Disordered Silver Nanoparticles on Multilayer Substrate Victor Ovchinnikov ............................................................................................................. 170 Performance Analysis of Commercial Accelerometers: A Parameter Review Stephan Elies ..................................................................................................................... 179

Authors are encouraged to submit article in MS Word (doc) and Acrobat (pdf) formats

by e-mail: [email protected]. Please visit journal’s webpage with preparation instructions: http://www.sensorsportal.com/HTML/DIGEST/Submition.htm

International Frequency Sensor Association (IFSA).

http://www.iaria.org/conferences2016/ALLSENSORS16.html

Sensors & Transducers, Vol. 193, Issue 10, October 2015, pp. 1-10

1

Sensors & Transducers

© 2015 by IFSA Publishing, S. L. http://www.sensorsportal.com

Smart and Customized Electrical Conductivity Sensor for Measurements of the Response Time from Sprayers

Based on Direct Injection

1, 2 Heitor V. MERCALDI, 1, 2 Caio H. FUJIWARA, 1, 2 Elmer A. G. PEÑALOZA, 2 Vilma A. OLIVEIRA,

1 Paulo E. CRUVINEL 1 Embrapa Instrumentação, Rua XV de Novembro 1452, São Carlos, SP, 13560-970, Brazil

2 Universidade de São Paulo, Av. Trabalhador Sãocarlense 400, São Carlos, SP, 13566-590, Brazil 1 Tel.: (+55)1621072800, fax: (+55)1621075754

1 E-mail: [email protected]

Received: 31 August 2015 /Accepted: 5 October 2015 /Published: 30 October 2015 Abstract: In the application of herbicides on the basis of direct injection systems, spraying response time plays an important role for the quality of spraying, particularly when operating in real time. The response time is defined as the time elapsed from the time of injection until the concentration of the mixture (water mixed with herbicide) reaches 95 % of its regime value in the sprayer nozzles. In the response time, the transport delay and the rise time for achieving the desired concentration are considered. This paper describes an intelligent sensor mounted near the sprayer nozzles to measure the concentration response time in an herbicide direct injection system, which uses a highly stable sinusoidal excitation signal. The sensor calibration was performed with NaCl solutions at concentrations similar to those found in actual application conditions. Using an integrated system based on the Arduino platform, an algorithm was developed to relate the measurements to the response time. The integrated system comprises the sensor with its own sensing hardware, A/D converter, processing and storage capabilities, software drivers, self-assessment algorithms and communication protocols. The immediate application of the integrated system is in the monitoring of the response time of a precision herbicide application. The results point to the next generation of smart devices that have embedded intelligence to support decision making in precision agriculture. Copyright © 2015 IFSA Publishing, S. L. Keywords: Intelligent sensor, Electrical conductivity, Direct injection, Response time. 1. Introduction

Brazil has experienced in the last two decades a

significant increase in the use of pesticides for agricultural production. Despite the significant growth of the area cultivated with transgenic seeds, a technology that promises to reduce chemical use in agricultural production, sales of these products increased by over 72 % between 2006 and 2012 and

is still rising up according to data from the Brazilian National Union of the Industry of Agricultural Defense Products [1], association which represents the pesticide manufacturers in the country.

In In the same period, the cultivated area with grains, fiber, coffee and sugar cane grew by less than 19 %, from 68.8 million to 81.7 million hectares, according to the Brazilian National Company for Supply [2]. This means that the average consumption

http://www.sensorsportal.com/HTML/DIGEST/P_2729.htm



2

of pesticides, which was just over 7 kilograms per hectare in 2005, rose to 10.1 kilograms in 2011, an increase of 43.2 %. Although this amount indicates more protection for products and higher incomes, the uniform rate of application leads to soil and water contamination. A key approach to reduce environmental pollution is to use variable-rate application.

An approach to develop variable-rate sprayer technologies is to install automation and control procedures in conventional sprayers. In order to adjust the sprayer operation, reference for variables such as working pressures, travelling speeds, and spraying concentration rates can be selected to achieve uniform drop size distribution.

The agricultural machinery and technologies available today allow chemical application variable rate based on prescription maps or sensors [3]. Variable-rate application can be performed by varying the concentration of the chemical on-the-go using a direct injection system [4]. The direct injection system is an electronically controlled system in which the chemical is injected into the carrier stream. The direct injection system has separated chemical and carrier reservoirs and the chemical can be injected into the carrier stream in different positions.

In the literature, reports of systems to inject concentrated pesticides into the carrier stream began to appear in the 70th decade [5]. In [6], Vidrine and collaborators tested the feasibility of injecting concentrated pesticides. In [7], Reichard and Ladd developed a field sprayer which included injection of pesticides at specific rates accounting for variations in travel speed. In [8], Chi and collaborators developed a flow rate control system which allowed the measurements of concentrated pesticides. In [9], Ghate and Perry developed a field sprayer based on the use of a compressed air to inject chemical into the carrier stream. In [10], Miller and Smith reported the development of a direct injection system. In general, during the spraying process errors can be observed. Research works on the evaluation of the application rate errors have shown that errors are not only due to the deviations from the target flow rates but also due to interaction between the dynamics of the systems and sprayer response time. By now, is quite well known that the direct injection system sprayer response time depends on the sprayer dynamics and on the transport delay [11].

The transport delay is due to flow rates and distance of the nozzle from the injection point. The farther from the injection point the nozzle is, the larger the uniformity of the mixture, but the higher the transport delay of the sprayer. Several studies on the performance of direct injection sprayers and the response time have appeared [12-20]. Therefore, the conventional implements can be reorganized to operate in variable-rate using control systems [21].

An advantage of the injection rate application over pressure-based variable rate application is the ability to change the herbicide type as well as to

perform on-line changes in the concentration [22]. The direct injection systems advantage is in the mixing of the required amount of chemicals with water, saving the excess amount for later use [23]. A key indicator to determine the precision of a direct injection sprayer is the control system response time. For sprayers, how much shorter the response time, much higher will be its field precision.

This paper presents the complete version of a smart and customized conductivity sensor (SCCS) for the evaluation of the response time of direct injection sprayers based on the electrical conductivity measurements. Previous discussions related to its development were presented in [24], and [25]. With the response time measurements in variable rate sprayers, a looking-ahead approach, which is useful to increase competitiveness and support sustainability in agriculture can be performed.

After this introduction, this paper is organized as follows. Section 2 presents the theoretical background on electrical conductivity; Section 3 presents the materials and methods for the development of the SCCS and the procedures for its validation. Finally, the results and discussions are presented in Section 4, followed by the conclusion in Section 5.

2. Theoretical Background

The electrical conductivity, also called specific conductance, is the ability of a solution to conduct an electric current. The mechanism for the electrical current conduction in electrolyte solutions is not the same as for metals. In liquids, this process is based on the movement of solvated ions, which are attracted by an electrical field. Therefore, the physical-chemical process is related to the occurrence of combination between the molecules of a solvent with molecules or ions of the dissolved substance. As electrolyte solutions obey Ohm’s law in the same way as the metallic conductors, when powered by direct current passing through the body of the solution, the conductance denoted G is defined as the inverse of the resistance expressed in 1−Ω or Siemens (S). The conductance G of a homogeneous body having uniform section is proportional to the cross-sectional area of the conductor A and inversely proportional to the length of the conductor denoted by l, that is:

A

Gσ= , (1)

where the proportionality constant σ is the electrical conductivity given in S/m. The ratio l /A is called the conductivity cell constant and depends on the instrumentation used. The conductivity increases with increasing temperature. Furthermore, the conductivity of a solution depends on the number of ions present and for this reason the most common is the use of the molar conductivity defined as:


3

Mm

σ=Λ , (2)

where Λm is the equivalent conductivity or the molar conductivity in Sm2/mol and M is the molarity or molar concentration in mol/L. The molar conductivity varies with the concentration of the electrolytes. The main reason for this effect is the change in the number or mobility of the ions present. The first case occurs in weak electrolytes, where the dissociation of ions in a solution is not complete. The second case occurs on strong electrolytes, where in the solution the dissociation of the molecule into ions is total, resulting in a very strong interaction between the oppositely charged ions, and can reduce its mobility.

The measurements of electrical conduction in ionic solutions are useful for a quick and routine analysis of solutions, since it is a simple measure related to the properties of the solution. In this context, the conductivity of a solution in a cell having an arbitrary dimension can be obtained by measuring the resistance of a solution of known concentration to determine the cell constant. After the cell constant is determined, the values of conductivities of different solutions can be obtained from experimental measurements data. For devices without automatic temperature compensation, the conductivity must be determined at the reference temperature.

The measurement of absolute values of conductivity requires the use of linear temperature compensation. Therefore, an electrical conductivity measured at room temperature can be corrected to one reference temperature, such per example, 25°C as follows:

( )( )25100125 −+=

θαθG

G , (3)

where θ is the room temperature, Gθ is the conductivity measured at room temperature and α is the temperature coefficient of variation in %/°C. Typical values for temperature coefficients are given in Table 1 [26].

Table 1. Typical temperature coefficients of substances.

Substance α (%/°C)

Acids 1.0 to 1.6 Bases 1.8 to 2.2 Salts 2.2 to 3.0 Potable water around 2.0

In solutions, yet it is necessary to correct the conductivity observed by subtracting the conductivity of the solvent, to get the value of σcorrected. Therefore, the molar conductivity Λm shall be written as:

Mcorrected

m

σ=Λ (4)

Thus, turning the unit concentration mol/L to mol/cm3, the equivalent conductivity Λm between two electrodes spaced 1 cm away due to 1 mol of substance may be given as:

Mcorrected

m

σ1000=Λ (5)

Then, for a parallel plate sensor, the conductance

G can be determined based on the molar conductivity Λm. The corrected specific conductivity of the electrolyte is then given in terms of the total ionic concentration M (mol/cm3) of the substance in the electrolyte solution and the equivalent conductivity. Therefore, by using the Equation (1), the conductance G can be found as:

( ), , ,1000

mm

MG A M

A

Λ Λ = ×

(6)

Peck and Roth defined response time (tT) as the

period from the instant the injection begins until the chemical concentration rate reaches 95 % of the equilibrium rate [27]. The rise time (tr) and transport delay (td) characteristics of a sprayer proposed by these authors are shown in Fig. 1. A 95 % concentration rate corresponds to the chemical concentration of the spraying, which is necessary for satisfactory weed control [28].

Fig. 1. Delay time (td), rise time (tr), and response time (tT) of a typical injection system as described in [27]. The

dotted line indicates the time behavior of the concentrated mixture (water-NaCl) as a response to an injection input.

The response illustrated in Fig. 1 can be identified as a first order system plus delay time. The time response is given by:

cdr TTT 3+= , (7)

where 3Tc is the requested time to reach 95 % in concentration in relation of the steady state value, i.e., after Td seconds [29].


4

3. Material and Methods

The injection sprayer systems can reduce applicator exposure to chemicals during the mixing and loading process. A direct injection sprayer has a typical control loop as shown in Fig. 2. In this figure, the upper blocks indicate the direct injection components and corresponding variables qhref, Vh, and qh, which represent the set point for the chemical flow, controlled, and measured variables respectively. In the lower blocks, at the some figure, is possible to observe the sprayer components, which

are described as qfref, Vf, and qm, which represent the set point for the mixture flow, controlled, and measured variables respectively. In this type of direct injection sprayer, the injection point is located upstream from the sprayer pump as presented in [30], and [31]. The water flow qW is dependent of both the flow mixture qm and the injection flow qh. The customized smart sensor is assembled at the nozzle in the end of the boom to measure the flow mixture concentration, which is proportional to its output denoted vSCCS.

Fig. 2. Block diagram of herbicide and mixture control.

The components of the customized smart sensor designed to measure the response time in spraying systems using direct injection of pesticides are shown in Fig. 3. For the implementation of the smart sensor, voltage regulators, opto-isolators and filters, as well as an integrated circuit for signal generator having the capability of frequencies adjustment were used. Active analogue filters, non-inverting amplifier drives and isolators circuits were implemented with operational amplifiers.

Fig. 3. Block diagram of the SCCS for response time measurements in spraying systems based on direct injection

of pesticides.

3.1. Excitation Circuits

An excitation circuitry (Fig. 4) for the SCCS was implementing to provide a sinusoidal signal with

appropriate frequency and magnitude. Such module was designed to produce signals with high stability and accuracy in an operating frequency range of 0.01 Hz to 1 MHz (Fig. 5).

Fig. 4. Block diagram for the excitation circuitry.

Fig. 5. The circuit of the excitation signal generator.


5

For the generation of the sinusoidal signal, an XR2206 integrated circuit, which produces sinusoidal signals with considerably low harmonic distortion, was used [32]. The oscillator frequency was set to 1.0684 kHz, which is suitable for the application and reducing the electrolysis and the polarization of the solution. In order to tailor the sinusoidal signal for the sensor application considered, a high-pass filtering and a signal amplification module were also used. After conditioning, the signal was appropriate for use presenting voltage limits as Vmax = 4.96 V and Vmin = -4.92 V.

The output signals of each of the stages of the excitation circuitry were analyzed using an oscilloscope (model TDS2012B, Tektronix®) and graphics were later built in Matlab® software.

3.2. Signal Conditioning

A signal conditioning circuitry was implemented to obtain a precise continuous voltage level with appropriate magnitude to be connected to an Arduino Uno platform. The Arduino Uno is an electronic prototyping platform, hardware open and single board, designed with an Atmel AVR microcontroller with built-in input-output support and a standard programming language with origin in Wiring projects, essentially based on C/C ++ [33]. The circuitry comprises a precision rectifier, a second order low-pass filter and an isolator buffer circuit (Fig. 6).

Fig. 6. Block diagram of the signal conditioning circuitry.

For signal conditioning (Fig. 7), the operational amplifier integrated circuit (model LF347), which presents broad bandwidth range (4 MHz), high Slew-Rate (13 V/s), high impedance input (1012 Ω), and fast settling time (2 µs) was used [34]. A buffer circuit was implemented with a low voltage operational amplifier (model OPA344) with an output type Rail to Rail [35] to isolate and protect the Arduino Uno platform against voltage surges or malfunctioning of the designed circuits.

The output of the buffer circuit is limited to voltages from 0 V to 5 V, safe input voltage range for the analogue/digital converter (ADC) of the Arduino Uno platform. This ADC has 6 channel 10-bit resolution with absolute accuracy of ±2 LSB (≈15 mV), and maximum sample rate of 15 kS/s.

The frequency response and impedance characteristics of the buffer circuit were analyzed through computer simulations performed in LTspice® software.

Fig. 7. The circuit for signal conditioning.

3.3. Sensor Mounting and Calibration

Based on the theoretical backgrounds previously presented, a set of parallel plates conductivity type transducers was built and analyzed for measuring the response time based on the electrical conductivity of the mixed flow in an agricultural direct injection spraying system. Two stainless steel electrodes with a diameter of 5 mm were used. The transducer was constructed with a polyacetal base and assembled direct into the nozzle body equipped with a diaphragm check valves. Fig. 8 illustrates the positioning and location of the SCCS for response time measurements.

To analyze the sensor with a static fluid, the electrodes were coupled to the base and spaced at three different distances chosen as 0.5 mm, 1.5 mm and 1 mm, resulting in constant cells equal to 0.255 cm-1, 0.500 cm-1 and 0.764 cm-1, respectively. The calibration was performed using a commercial conductivity meter (model mCA150, Tecnopon), with an operating range of 0 to 200 µS/cm, resolution of 0.1 µS/cm, 2 % of full scale accuracy and 1 % of full scale precision [37]. The measurements performed with the commercial conductivity meter were checked with a standard KCl solution (0.02000 mol/L). Static tests were carried out for the analysis of solutions consisting of water and NaCl. The procedures were conducted for three different cell constants at 25 oC. 3.4. Response Time Measurements

In order to validate the developed sensor, a real experiment to measure the response time of a sprayer system based on direct injection was performed. The experiment was set for a cell constant of 0.500 cm-1.

For real time analysis, the conductivity was measured and the results were processed via LabVIEW® software, i.e., using the Arduino Uno platform, which allows the computational processing and intelligence aggregation [36].

The initial concentration of the injected NaCl solution was 50 g of salt for 16 L of potable water. This NaCl solution was injected upstream from the


6

sprayer pump at 2.3 L/min during 30 seconds. The mainstream flow was regulated at 16 L/min and 23 L/min with pressure at 200 kPa and 400 kPa, respectively. For each regulated flow, 3 repetitions of the response time measurements were conducted.

To obtain a time constant Tc in a range of variation lower than 5 %, the liquid temperature

could not vary more than ±2.5oC during the conductivity acquisitions. This threshold was determined based on simulation using the Equations (3) and (7), as well as calibration curves. For validation, and in order to consider a general scenario, it was used a temperature coefficient of variation (α) equal to 3.0 %/°C.

Fig. 8. Details of the customized smart sensor (SCCS) to measure the response time of direct injection sprayers. The individual time responses and the transport

delay time for each repetition were analyzed in an actual spraying system. Fig. 9 illustrates the sensor location in the actual sprayer system.

Fig. 9. Instrumental arrangement for validation of the intelligent sensor to measure response time

of a direct injection sprayer with TeeJet® QJS Multiple Nozzle Bodies e-ChemSaver.

4. Results and Discussion The computational processing of data and

analysis with the developed conductivity sensor was performed using the LabVIEW® software, after signal conditioning and the analogue/digital

conversion with the Arduino Uno platform. The use of the electronic Arduino Uno platform and the LabVIEW® interface allowed the aggregation of intelligence for self-diagnostic of the SCCS, as well as its self-assessment based on the use of a specific algorithm (Fig. 10). The data valid flag was determined experimentally based on the dynamic range of the SCCS defined by 0.50 V ±2LSB < vSCCS ≤ 4.90 V ±2LSB, which is related with the accuracy allowed by the internal ADC of the Arduino Uno platform.

The calibration results identified the electrodes faces distance for a better accuracy of the conductivity sensor which is dependent on the conductivity cell constant of the transducer. The shorter the distance between the faces of the electrodes for the same cross-sectional area, the greater will be the accuracy of the measurements. The calibration curves are shown in Fig. 11, Fig. 12 and Fig. 13 for constant cells given respectively by 0.255 cm-1, 0.500 cm-1 and 0.764 cm-1. Table 2 shows the experimental values obtained for determining the response time tT.

A first set of experiments were conducted and the results are shown in Fig. 14. In this case, a water-NaCl solution flow of 16 L/min and pressure of 200 kPa were used. The dynamic responses obtained for the three repetitions demonstrated the sensor accuracy and reliability.


7

Fig. 10. The flow-diagram of the algorithm for self-assessment and self-diagnostic.

Fig. 11. Calibration curves and comparison of measurements of the electrical conductivity

from experimental solutions obtained with the SCCS with constant cell of 0.255 cm-1.


from experimental solutions obtained with the SCCS with cell constant of 0.500 cm-1.


from experimental solutions obtained with the SCCS with cell constant of 0.764 cm-1.

Table 2. Experimental values of delay and response times (Mean is Average Values, Std is the Standard Deviation,

and CV is the Coefficient of Variation).

Flow (L/min)

Repetitions and

statistical analysis

Vmin (0%)

Vmax (100%)

td (s) tT (s)

16

1st 1.58 4.16 29.56 42.48 2nd 1.57 4.17 28.80 41.55 3rd 1.57 4.15 28.36 41.40

Mean 1.58 4.16 28.91 41.81 Std 0.008 0.010 0.607 0.585 CV 0.005 0.002 0.021 0.014

23

1st 1.56 3.99 22.53 32.81 2nd 1.53 3.96 22.52 33.01 3rd 1.55 3.97 23.09 33.77

Mean 1.55 3.97 22.71 33.20 Std 0.012 0.013 0.326 0.506 CV 0.008 0.003 0.014 0,015


8

Fig. 14. Transients and transport delay times of the sprayer with the SCCS assembled in an actual spraying system

with water-NaCl solution flow of 16 L/min and pressure equals to 200 kPa.

For this experimental arrangement, the average delay time was 28.91 s and the average response time 41.81 s.

A second set of experiments were conducted and the results are shown in Fig. 15. In this case, a water-NaCl solution flow of 23 L/min and pressure of 400 kPa were used. The results obtained for the sprayer system having a flow rate equal to 23 L/min and pressure of 400 kPa, as occurred previously, have shown again, accuracy and reliability in the measurements obtained with the developed sensor. The average delay time was of 22.71 s and the average response time 33.20 s.

Fig. 15. Transients and the transport delay times obtained with the SCCS assembled in an actual spraying system

with water-NaCl solution flow of 23 L/min and pressure equal to 400 kPa.

Also, it is important to notice that the use of faster circuits implies higher current densities, lower voltage tolerances and higher electric fields, which

make integrated circuits more vulnerable to electrical failure.

The integrated circuits used in the intelligent sensor are new generations of electronic devices and they provided good performance.

Furthermore, the use of the polyacetal base was adequate and results have shown a robust mechanical design.

The reliability assessments are crucial for the end user as adjustments to electrical conditions and thermal management, since the electrical conductivity is dependent on the temperature of the flows related to the mixture of water plus pesticide.

5. Conclusions

A smart and customized sensor to measure the response time of spray systems based on direct injection was presented. The results have shown its usability in real time applications. The decision to embed the smart sensor directly in the sprayer nozzle provides a scenario where the input data from the physical sensor could be analyzed by various knowledge-based routines. The sensor output could be raw data or preprocessed information. This information could be in the form of a flag, which shows a confidence level of the response time for pesticide applications.

The results based on the calibration curves for the sensor in three different assemblies showed that the accuracy of measurements depends directly on the conductivity cell constant. However, to determine the response time of a direct injection system of pesticides, a customized sensor with shorter spacing between its electrodes can provide an adequate sensitivity for sensing the level of the concentration in the mixture involving pesticide plus water.

The results of the sprayer system response time with direct injection obtained in this research work have shown that the smart sensor developed has good repeatability, reliability and practicality. Furthermore, the results show the decreasing of the response time with the increasing of the flows in consequence of the increased speed in which the mixture of water and pesticide travels through the system.

The use of an intelligent sensor provides more additional information than that of traditional sensors. The information provided by an intelligent sensor can include actual data, corrected data, validity of the data, and reliability of the sensor.

Furthermore, such SCCS development meets future prospects in practical applications, bringing potential benefits for sustainability, as well as precision agriculture processes.

Acknowledgements The authors acknowledge the financing

support from the Brazilian Agricultural Research


9

Corporation (Embrapa Instrumentation, Process MP2 No. 02.11.07.0.25.00.00) as well as from the CNPq Brazilian agency under grant 306477/2013-0. References [1]. SINDAG. SINDAG and Embrapa study agchem

aerial application in Brazil, Accessed at March 5, 2015. [Online]. Available:

http://news.agropages.com/News/NewsDetail—13170.htm (2014).

[2]. CONAB, Monitoring of the Brazilian grain crops - season 2013/14 (original in Portuguese), Brazilian National Company for Supply, Tech. Rep., 2014.

[3]. M. Sökefeld, Variable rate technology for herbicide application, in Precision Crop Protection - the Challenge and Use of Heterogeneity. Springer Verlag, 2010, pp. 335-347.

[4]. J. Vondrica, Study on the process of direct nozzle injection for realtime site-specific pesticide application, Ph.D. dissertation, Institut für Landtechnik der Rheinschen Friedrich-Wilhelms-Universität, 2008.

[5]. K. L. Hughes, A. R. Frost, A review of agricultural spray metering, Journal of Agricultural Engineering Research, Vol. 32, Issue 3, 1985, pp. 197-207.

[6]. C. G. Vidrine, C. E. Goering, C. L. Day, M. R. Gebhardt, D. B. Smith, A constant pesticide application rate sprayer model, Transaction of the ASAE, Vol. 18, Issue 3, 1975, pp. 439-443.

[7]. D. L. Reichard, T. L. Ladd, Pesticide injection and transfer system for field sprayers, Transactions of the ASAE, Vol. 26, Issue 3, 1983, pp. 683-686.

[8]. L. Chi, R. L. Kushawara, F. W. Gigsby, Chemical flow rate control in injection-type sprayers, Canadian Agricultural Engineering, Vol. 30, Issue 1, 1988, pp. 19-26.

[9]. S. R. Ghate, C. D. Perry, Ground speed control of pesticide application rates in a compressed air direct injection sprayers, Transaction of the ASAE, Vol. 37, Issue 1, 1994, pp. 33-38.

[10]. M. S. Miller, D. B. Smith, A review of application error for sprayers, Transactions of the ASAE, Vol. 35, Issue 3, 1992, pp. 787-791.

[11]. P. S. Lammers, J. Vondricka, Direct injection sprayer, in Precision Crop Protection - the Challenge and Use of Heterogeneity, Springer Verlag, 2010, pp. 295-310.

[12]. R. S. Budwig, J. M. McDonald, T. J. Karsky, Evaluation of chemical injection system for mobile agricultural spray equipment, Transaction of the ASAE, 1988, Paper No. 88-1592.

[13]. F. D. Tompkins, K. D. Howard, C. R. Mote, R. S. Freeland, Boom flow characteristics with direct chemical injection, Transaction of the ASAE, Vol. 33, Issue 3, 1990, pp. 737-743.

[14]. K. A. Sudduth, S. C. Borgelt, J. Hou, Performance of chemical injection sprayer system, Applied Engineering in Agriculture, Vol. 11, Issue 3, 1995, pp. 343-348.

[15]. Y. M. Koo, H. R. Sumner, Total flow control for a direct injection sprayer, Applied Engineering in Agriculture, Vol. 14, Issue 4, 1998, pp. 363-367.

[16]. H. Zhu, R. D. Fox, H. E. Ozkan, R. D. Brazee, R. C. Derksen, A system to determine lag time and mixture uniformity for inline injection sprayers,

Applied Engineering in Agriculture, Vol. 14, Issue 2, 1998, pp. 103-110.

[17]. E. A. Anglund and P. D. Ayers, Field evaluation of response times for a variable rate (pressured-based and injection) liquid chemical applicator, Applied Engineering in Agriculture, Vol. 19, Issue 3, 2003, pp. 273-282.

[18]. P. Hloben, Study on the response time of direct injection systems for variable rate application of herbicides. PhD dissertation, Universitäts und Landesbibliothek Bonn, 2007.

[19]. A. EL Aissaoui, F. Lebeau, M.-F. Destain, K. Houmy, Applying control volume finite element for modeling direct injection boom spraying, in Proceedings of the Congress 7th World Congress of Computer in Agriculture, Reno, USA, 22-24 June 2009, pp. 190-198.

[20]. N. S. Hassen, C. S. Nor Azwadi, J. M. Sheriff, Advanced techniques for reducing spray losses in agrochemical application system, Life Science Journal, Vol. 11, Issue 3, 2014, pp. 56-66.

[21]. M. Jafari, A. Hemmat, M. Sadeghi, Development and performance assessment of a DC electric variable-rate controller for use on grain drills, Computers and Electronics in Agriculture, Vol. 73, Issue 1, 2010, pp. 56-65.

[22]. M. L. Stone, D. K. Giles, K. J. Dieball, Distributed network system for control of spray droplet size and application rate for precision chemical application, Transaction of the ASAE, Canada, 1999, Paper no. 99-3112.

[23]. A. Landers, The theory and constituent parts of a direct injection sprayer, Accessed at January 27, 2015. [Online]. Available:

http://vivo.cornell.edu/individual/AI-24241250305 (1998).

[24]. H. V. Mercaldi, C. H. Fujiwara, E. A. G. Peñaloza, V. A. Oliveira, P. E. Cruvinel, An intelligent and customized electrical conductivity sensor to evaluate the response time of a direct injection system, in Proceedings of the 6th International Conference on Sensor Device Technologies and Applications (SENSORDEVICES’15), Venice, Italy, 23-28 August 2015, pp. 19-24.

[25]. C. H. Fujiwara, H. V. Mercaldi, E. A. G. Peñaloza, P. E. Cruvinel, V. A. Oliveira, Direct injection system response time determination using a conductivity smart sensor directly coupled into herbicide sprayer nozzle, Abstracts Proceedings of the XLIV Brazilian Congress of Agricultural Engineering (CONBEA’15), Sao Pedro-SP, Brazil, 13-17 September 2015, pp. 171.

[26]. A guide to conductivity measurement: Theory and practice of conductivity applications (MELTER/TOLEDO), Accessed at September 27, 2015. [Online]. Available:

http://www.tau.ac.il/~chemlaba/Files/conductivity_guide_EN%20(2).pdf (2013)

[27]. D. R. Peck, L. O. Roth, Field sprayer induction system development and evaluation, Transaction of the ASAE, 1975, Paper No. 75-1541.

[28]. K. A. Bennet, R. B. Brown, Direct nozzle injection and precise metering for variable rate herbicide application, Transaction of the ASAE, 1997, Paper No. 97-1046.

[29]. K. Ogata, Modern Control Engineering, 4th ed., Prentice Hall PTR, 2001.

[30]. U. R. Antuniassi, P. Miller, M. Paice, Performance evaluation of injection metering systems, Revista


10

Brasileira de Engenharia Agricola e Ambiental, Vol. 6, Issue 1, 2002, pp. 159-165.

[31]. K. P. Gillis, D. K. Giles, D. C. Slaughter, D. Downey, Injection mixing system for boomles, target-activated herbicide spraying, Transaction of the ASAE, Vol. 46, Issue 4, 2003, pp. 997-1008.

[32]. Exar Corporation, Datasheet XR2206, Accessed at February 24, 2015. [Online]. Available:

https://www.sparkfun.com/datasheets/Kits/XR2206 104 020808.pdf (2008)

[33]. Arduino, Arduino Uno, Accessed at March 16, 2015. [Online]. Available:

http://arduino.cc/en/Main/ArduinoBoardUno (2015)

[34]. Texas Instruments, Datasheet LF147, LF347-N. Accessed at February 24, 2015. [Online]. Available: http://www.ti.com/lit/ds/symlink/lf147.pdf (2013)

[35]. Texas Instruments, Datasheet OPA344. Accessed at March 04, 2015. [Online]. Available:

www.ti.com/lit/ds/symlink/opa344.pdf (2008) [36]. R. Taymanov, K. Sapozhnikova, Metrological self-

check and evolution of metrology, Measurement, Vol. 43, Issue 7, 2010, pp. 869-877.

[37]. Tecnopon, Instructions manual mCA150. Accessed at February 24, 2015. [Online]. Available:

http://www.tecnopon.com.br (2014).

___________________

2015 Copyright ©, International Frequency Sensor Association (IFSA) Publishing, S. L. All rights reserved. (http://www.sensorsportal.com)

http://www.sensorsportal.com/HTML/DIGEST/april_2012/USTI_EXT.htm





11

Sensors & Transducers© 2015 by IFSA Publishing, S. L.

http://www.sensorsportal.com

Dynamical Capillary Rise Photonic Sensor for Testing of Diesel and Biodiesel Fuel

1 Michal BORECKI, 2 Michael L. KORWIN-PAWLOWSKI,

3 Mariusz DUK, 3 Andrzej KOCIUBIŃSKI, 4 Jarosław FRYDRYCH, 5 Przemyslaw PRUS, 1 Jan SZMIDT

1 Institute of Microelectronics and Optoelectronics, Warsaw University of Technology, 75 Koszykowa Str., 00-662 Warsaw, Poland

2 Département d’informatique et d’ingénierie, Université du Québec en Outaouais, 101 rue Saint-Jean-Bosco, J8X 3X7 Gatineau Québec, Canada

3 Institute of Electronics and Information Technology, Lublin University of Technology, 38A Nadbystrzycka Str., 20-618 Lublin, Poland

4 Automotive Industry Institute, 55 Jagiellońska Str., 03-301 Warsaw, Poland 5 Blue Oak Inventions, 31/101 Lwowska Str., 56-400 Oleśnica, Poland

1 Tel.: +48 22 234 77 49, fax: +48 22 234 60 63 E-mail: [email protected]

Received: 31 August 2015 /Accepted: 5 October 2015 /Published: 30 October 2015 Abstract: There are many fuel quality standards introduced by national organizations and fuel producers. Usual techniques for measuring fuel parameters like cetane number, cetane index, fraction composition, viscosity, density, and flash point, require relatively complex and expensive laboratory equipment. On the fuel user side, fast and low cost sensing of useful state of biodiesel fuel is important. The main parameters of diesel fuel compatibility are: density, viscosity and surface tension. These three parameters define indirectly the quality of the fuel atomization process and the injected portion of energy that affect the quality of the fuel.

In the presented paper the purposefulness of fuel testing using measurements of separable parameters is discussed. On this base, a sensor which enables the examination of relation of the mentioned parameters in one arrangement is proposed, analyzed and tested. The sensor uses the dynamic capillary rise method with photonic multichannel data reading in an inclined capillary. The principle of the sensor’s operation, the construction of the sensor head, and the experimental results are presented. The capillary is a disposable element. The sensor testing was performed with freshly prepared biodiesel fuels, and fuels stored for 2 years. We conclude that the proposed construction may be in future the base of low cost commercially marketable instruments for basic fuel classification: fit for use or not. That classification includes initial fuel composition and fuel parameters change during storage. Therefore, the proposed sensor is intended to use in fuel buying/selling point rather than used as part of a diesel engine automated system. Copyright © 2015 IFSA Publishing, S. L. Keywords: Biodiesel fuel, Diesel fuel, Fuel quality, Fuel storage, Viscosity, Density, Surface tension, Multiparametric sensor, Capillary sensor, Capillary rise.




12

1. Introduction

The paper consists of 5 sections. The first section is the introduction, where the diesel fuel, diesel engine properties as well as surface tension measurements and known sensors for diesel fuel testing are discussed. The second section describes the monitoring of capillary dynamical rise as an idea of multiparametric sensor for diesel fuel testing. The construction of the developed sensor is presented in section three. In section four the discussion of obtained results of diesel fuels examination using the developed sensor is presented and a method of fuels classification is proposed. The conclusion is gathered in section five.

1.1. Diesel Fuel

Classical diesel fuels are made from distillated products of crude oil with addition of improvers. Biodiesel fuel is a mixture of classical diesel fuel, bio-components and additives, which include, for example, ignition improvers, injector cleanliness, lubricants and antioxidants. Present-day bio-components are obtained from vegetal oils, or the bio-mass. Important disadvantages of biodiesel fuel can be overcome by fuel processing. A new generation of biodiesel fuel can be made using bio-component isomerization or hydrogenation.

The quality of biodiesel fuel is defined in relation to four main processes: a) fuel storage stability, b) fuel transfer dynamics from tank to injection unit, c) combustion in the chamber, and d) gas emission.

Nowadays, producers define the useful state of diesel fuel by several parameters: cetane number (CN), cetane index, density, distillation parameters and kinematic viscosity. One of the most important diesel fuel quality parameters is its ignition quality. The ignition quality depends on the molecular composition of the fuel and is characterized by the ignition delay time, which is the time between the start of injection and the start of combustion. Other diesel fuel parameters, not directly related to engine performance, characterize its operability, as for example: amount of carbon residue, water and sediment, cloud point, conductivity, oxidation stability, acidity, copper corrosion, flashpoint, lubricity, appearance, and color [1].

1.2. Diesel Fuel and Engine

Diesel engine performance, fuel consumption, and emitted pollutants result from the combustion process. The environment of combustion, the injected fuel’s form and the fuel quality play primary role in the diesel combustion process [2]. Therefore, contemporary diesel engines may be equipped with the set of sensors that are aimed to monitor several parameters: engine speed, fuel temperature, fuel pressure at the injector/common rail and at the fuel supply pump, oil pressure, oil temperature, air inlet

temperature, air flow/mass, coolant temperature, and oil and fuel levels as well as exhaust gas – nitrogen oxide presence. These sensors are often integrated with on board diagnostic (OBD) bus.

The main challenge for diesel engine designers is to adjust the combustion chamber size design according to the diesel fuel injection characteristics [3]. The diesel fuel must be introduced into the combustion chamber, vaporize, and react with oxygen at an assumed speed. The fuel properties that have the greatest effect on the injection process are viscosity, density and surface tension. If the injection is made at a constant pressure, the viscosity affects the fuel spray formation by limiting the speed of fuel flow. If the injection system is designed to meter the volume of injected fuel, the density of fuel defines the fuel injected mass that is linked with the useable portion of energy [4]. The surface tension is the one of main factors that describes the fuel tendency to form drops, known also as the fuel atomization process or spray forming. Spray ignition phenomena are described by dynamic spray characterization, as for example: droplets concentrations, droplets diameters, droplets velocities [5]. In standard diesel engines, the expected injected droplet velocities are from 300 to 400 m/s, while the expected droplet diameters are in the range from 0 to 40 µm. Therefore, the costly diesel engines sometimes are equipped with fluid condition sensors that measure the density and the viscosity of fuel or oil [6]. The most costly diesel engines, for example for oceanic ships, sometimes are equipped with combustion temperature or pressure sensor as well as with spray form monitors which enables control of advanced combustion strategies [7].

1.3. Measurement of Diesel Fuel Parameters The standardization process of measurement of

diesel fuel parameters is still in progress [8]. Nowadays, the basic idea of fuel characterization is a selective and specific measurement of sequent parameters.

Diesel fuel operability parameters measurements are not full standardized. The procedures and measured equipment are set to a series of parameters, for example: - acid-number, used to describe the engine’s

corrosion susceptibility by the fuel acidity level using chemical methods,

- flashpoint, used to describe the lowest temperature at which the fuel is sufficiently vaporized to form a flammable mixture with air under standard conditions, can be measured using the Pensky-Martens closed cup tester,

- lubricity of a substance is not a material property and cannot be measured directly, but a standard test method for evaluating the lubricity of diesel fuels uses a high frequency reciprocating rig and gives the value of wear trace diameter called WS 1.4 and expressed in [μm].


13

Measurements of ignition quality of fuel (CN) have to be carried out in the Cooperative Fuel Research Engine (CFR-5) or the Ignition Quality Tester (IQT™). The distillation parameters may to be measured using the distillation unit, the basic components of which are: a distillation container, a condenser and an associated cooling bath, a heater and the temperature measuring device, as well as a receiving cylinder, or may be measured with automated distillation process analyzer [9]. The diesel fuel kinematic viscosity is a measure of the time for a fixed volume of liquid to flow by gravity through a capillary. The kinematic viscosity is calculated from calibrated device as

η = C ∙ t, (1)

where: η is the viscosity, t is measured flow time, C is the calibration constant.

In a European study, it was observed that using the biodiesel fuel at low environment temperatures can lead to the degeneration of the engine [10]. Therefore, production standards for biodiesel fuel were introduced: density at 15 °C (ISO3675) and low-temperature fluidity for the transitional seasonal periods and winter (DIN EN 116).

The basic disadvantage of such laboratory approach to fuel measurements is the high cost of the set of measurement devices and the complexity of the procedures. For the ordinary fuel user, the analysis of such collection of parameters is also too complex. Moreover, the surface tension of diesel fuel measurement is not standardized. Instead, the ASTM D971 standard is set to determine the possible contaminants of hydrocarbon fluids with water as the purity of hydrocarbon fluids is important factor of diesel fuels quality.

1.4. Surface Tension Measurement Methods There are a few basic methods of surface tension

measurements: the capillary rise method, the drop weight method, the ring or plate method, the maximum bubble pressure method, the method of analyzing the shape of the hanging liquid drop, and the dynamic methods [11].

The capillary rise method is the oldest method used for determination of the surface tension. To perform the surface tension measurement, a capillary is first dipped into the tested liquid. If the interaction forces of the liquid with the capillary walls are stronger than those between the liquid molecules, the liquid wets the walls and rises in the capillary to a defined level. If the cross-section area of the capillary is circular and its radius is sufficiently small, then the meniscus is semispherical. When the capillary is filled in the vertical position, the surface tension may be calculated with following equation:

γ ≅ r ∙ g ∙ h ∙ ρ ρ , (2)

where: γ is the surface tension, θ is the contact angle, r is the inner radius of capillary tube, g is the gravity constant, h is the height of meniscus, ρl is the density of liquids, and ρv is the density of vapor [11]. For standard temperatures, the density of liquid is much greater than of the density of vapor, therefore the presence of the vapor phase is ignored. The contact angle may have different values in the static or dynamic situations. Such situations for a capillary being filled are presented in Fig. 1.

Fig. 1. Static and advancing contact angle in capillary flow. The advancing contact angle value may be

measured with analysis of the shape of the sessile drop positioned on sloping surface [13]. In dynamic situations, when the liquid is moving over the surface, the velocity gradients in the fluid may be large and the viscous forces may control the shape of the fluid surface interface [14]. When the liquid is a multi-component solution, the main component surface tension is modified with the surface excess of other components related to the effective surface concentration, dependent on the nature of the solvents and solutes [15]. Many additives of diesel fuel may be considered as surfactant particles, which alter the surface properties of fuel, even when present in small quantities, Fig. 2.

Fig. 2. Surface tension in multi-component liquid.


14

The diffusion of surfactants to the surface continues with a determined and limited velocity, until equilibrium is established. The time to equilibrium varies depending on surfactant concentrations and type, and may range from 0.01 s to few hours. Therefore, when the liquid flows up the capillary, the surfactants particles also flow, but not necessary maintaining the original surface concentrations, resulting in modifications of the temporary surface tension and the temporary advancing contact angle (2) [16].

The simplest technique for measuring the capillary rise is using a ruler with an optical reading, but its automation requires a camera, a mechanical device as a lift, or a liquid injector coupled with an additional light source to improve the optical reading of the meniscus position.

The readings of data in the capillary rise method to measure the surface tension may be performed for semi-transparent liquids also using the light attenuation effect. The lensing effect of the meniscus may be used for measurements of the contact angle [17]. The principle of the measuring set-up for this method is presented in Fig. 3. The lensing effect of the meniscus is registered as dimensions in the camera grid, while the capillary rise controls the amount of registered light.

Fig. 3. Principle of surface tension and contact angle measurement lengthwise of capillary.

The ring method and the Wilhelmy plate method

are similar. In these methods an object is moved perpendicularly into or out of the liquid. The plate is moved towards the surface until the meniscus connects with it, or the submerged ring is pulled out of the liquid. The additional force acting on the plate or ring

due to its wetting by liquid is measured. The ring method is unsuitable for dynamic surface tension measurements, but can be extended with captured image analysis to simultaneous measurements of surface tension and contact angle [18].

The drop weight method is one of the most common methods used for surface tension automated measurements. In this method, drops formed at the tip of the glass capillary are weighted and counted. The pendant drop at the tip starts to detach when its weight reaches the value balancing the surface tension of the liquid. The advantage of the method is the possibility to measure surface tensions between liquids and other than optically transparent materials [19]. The disadvantage of this method is the necessity to calculate the corrections for the capillary tip parameters and the volume of the drop, which are characteristic of a given device. Also evaporation of the drops may be a problem for some measurements.

The shape analyzing method of liquid drop is based on the effect of the liquid’s deformation caused by the gravitation force action. The hanging drop shape can be analyzed in set-up presented in Fig. 4.

Fig. 4. Pendant drop tensiometer.

When the drop lies on a plane, the surface area of a drop is proportional to its squared radius, and the gravitational deformation depends on its volume. This deformation is proportional to the third power of the drop radius and to the liquid’s surface tension as well as on the liquid – plane contact angle, or the plane’s surface energy [20].

In the maximum bubble pressure method, an air or gas bubble is blown at constant rate through a capillary submerged in the tested liquid. The maximum measured pressure that is required to insert the bubble of gas into the liquid may be used for surface tension determinations.

The analysis of the shape of an oscillating liquid jet method bases on optical observations of characteristic shapes of the liquid that flushes out from an elliptic orifice.

The methods that are suitable to measure the dynamic surface tension are the maximum bubble pressure and the oscillating liquid jet methods, but they require more complex experimental laboratory set-ups.


15

1.5. Surface Tension and Advancing Angle of Diesel Fuel

It has been proven that surface tension is one of the most important of inhomogeneous fluid properties. It reflects the range of interactions in a fluid more directly than the bulk properties do [21]. The relationship between the surface tensions of different real vegetable oils and their fatty acid composition was postulated in [22]. Predicting the surface tension of biodiesel fuels from their fatty acid composition showed that the differences in surface tension between biodiesel types are not the only cause of the reported differences in engine tests [23].

The surface tension and advancing angle are interesting case in diesel fuel characterization. On one hand the surface tension and ASTM D971 standard are proposed as useful indicator of cetane number [24] and fuel atomization process [25]. On the other hand, it was postulated that the surface tension of most liquid hydrocarbons is very similar. For example, for HydroCal 300 - a hydrotreated naphthenic medium grade lubricant oil and IFO-120, an intermediate fuel oil, the measured surface tensions at 25 °C are the same and equal 31.8 mN/m, while their viscosities differ significantly: 162mPa·for HydroCal and 487 mPa for IFO-120 [26].

Results of investigation of surface tension values of diesel fuels and their components, with static capillary rinse method performed according (2), are presented in Table 1, where EN-590 depicted premium diesel fuel, FAME is fatty acids methyl ester - bio component, x% is concentration of bio component in fuel. For experiment VitroCom CV3040Q capillaries of inner diameter equal 300 µm were used. Since a film of oil remains on the inner surface of the capillary after the receding phase, representing complete wetting, the static contact angle in (2) was considered to be equal to 0° [27].

Obtained data shows that commercial premium diesel fuel, standard diesel fuel and 100 % biodiesel component surface tension values are considerably differ. It should be noted that contemporary diesel fuels components are not only pure hydrocarbons, it consist significant concentration of fatty acids esters. Moreover FAME requires different fuel additives than petrodiesel. Therefore, such changes in fuel composition may significantly change surface tension, contact angle and other diesel fuel parameters [28].

Table 1. Surface tension values of standard fuels and their components, obtained with static capillary rinse method.

Surface tension, (mN/m),

of fuel type

Temp (°C)

Premium, EN-590,

FAME 0%

Standard, EN-590,

FAME 7%

BioFuel, EN 14214

FAME 100 %

10 28.0 29.5 31.5 20 27.0 28.4 30.6 30 26.1 27.3 29.7

Advancing contact angles between a glass slide and different hydrocarbon oils differ significantly depending on the temperature, the speed of liquid creep and the type of liquid hydrocarbon. For example, measured at 25 °C and 264 µm/s the advancing angle of HydroCal 300 is 36° while of IFO-120 is 54° [26].

1.6. Sensors for Diesel Fuel Testing Generally, diesel fuel analysis has taken place in

certified laboratories. Though, there is a rising demand for on-line and real-time diesel fuel monitoring. To meet such requirements, series of different type sensors are introduced, nowadays, into market.

Some sensors base on surface acoustic waves examination [29]. They are claimed as liquid/oil quality monitors, while they are used to measure acoustic impedance at monitored liquid surface:

Z = ω ∙ ρ ∙ η, (3)

where: ω is the radian frequency, ρ is the density, η is the viscosity. Due to impedance dependence on density and viscosity, the postulated diesel fuel quality monitor has to consist of two sensor heads. But, the measurement tricky of proposed construction is determination of diesel fuel dynamical coupling to examined surface, which depends on surfactants properties. Therefore, mentioned sensors can be used for monitoring simpler than diesel fuel liquids as for example lubricant oil. Still the works on critical coupling surface parameters improving are in progress [30]. The density of fuel is measured with the use of resonant frequency sensors [31] that are capable of electronic integration [32].

Optical sensors are also popular in diesel fuel examination. Infrared sensors are used for two proposes, to measure the blend concentration in blended fuel and to measure predict fuel quality [33]. Some smart ultraviolet sensors enable the examination of a set of fuel parameters, but the possibility of easy measurement comes with a high price of the devices [34]. On the other hand, some proposed multiparametric sensors may be low cost, but using them requires trained operating personnel. The possibility of diesel fuel quality testing using an optoelectronic set-up implementing a multi-parametric method was shown in [35] and [36]. The components for such sensors are under technological optimization [37] [38] [39].

2. Monitoring of the Capillary Dynamical Rise as Idea of Multiparametric Sensor for Diesel fuel Testing

Diesel fuel users require the simplest possible

answer to a question: is that fuel useful for my engine?


16

This answer may be deducted from capillary rinse sensor depicted in [40].

The idea of the sensor was inspired by the need to come up with a measurement method of a set of diesel fuel parameters in one system that would require the minimum of automated mechanical elements. The aimed at set of diesel fuel parameters includes: density, viscosity, surface tension and wetting represented by the advancing contact angle between fuel and glass. To obtain the dynamical data of advancing contact angle one needs to measure the capillary rise speed. An increase of the speed of the speed of the liquid’s movement and its distance in the capillary in order to improve the precision of the measurement is possible by inclination of the capillary. When the capillary’s axis is inclined at the α angle to the horizontal, a liquid is drawn in by capillary forces according to equation (4).

= ∙ ∙ ∙ ∙ ∙∙ ∙ ∙ , (4)

where γ is the surface tension, θ is the advancing contact angle, r is the inner radius of the capillary tube, g is the gravity constant, l is the length coordinate of meniscus at capillary axis, t is the time of measurement, α is the angle of inclination, ρl is the density of liquid, and η is its viscosity. In the general case depicted by equation (4) the advancing contact angle θ is a function of the liquid’s speed. This results in an involved dependence for dynamical calculation of liquid parameters. Assuming that the advancing contact angle is characterized by semi constant values, equation (4) may be solved for the dynamical diesel fuel analysis in two ways.

The first solution of equation (4) is for local speed measurement at a set length coordinate of meniscus with the difference equation:

= ∙ ∙ ∙ ∙ ∙∙ ∙ , (5)

where: θs is the semi constant value of advancing contact angle at coordinate ls, while ls and Δls are presented in Fig 5.

The second solution of equation (4) is for average speed measurement at a set distance and is based on integration:

=∙ ∙∙ ∙ ∙ ∙ ∙ , (6)

where: θ21 is the average and semi constant value of advancing contact angle between coordinate l1 and l2 presented in Fig. 6. For the sensor application, equation (6) may be solved numerically by inputting in the measured time of rise t2-t1 at corresponding coordinates of meniscus.

Fig. 5. Parameters position for local speed determination with (5).

Fig. 6. Parameters position for local speed determination with (6).

3. Sensor Construction

The sensor construction used in dynamical capillary rise application analysis was firstly presented at the SENSORDEVICES 2015 Conference [40].

3.1. Sensor Head

The sensor’s head consists of two functional blocks: the base and the disposable capillary optrode, shown on Fig. 7.

The base is used to integrate the removable vessel for the examined liquid, the three optical paths of the source and the receiver, as well as for positioning the disposable capillary optrode at one of the three possible inclination angles. The ramp geometry used in presented experiments is shown in Fig. 8. The capillary optrode is dipped for 3 mm into the liquid.

Fig. 7. The dynamical capillary rise sensor head.


17

The disposable optrode is made from 15 cm sections cut from the TSP 700-850 capillary from Polymicro Inc. In the optical paths, large core optical fibers BFH 37-800 from Thorlabs were used. They are characterized by the core radius of 800 µm and the hard clad outer diameter of 830 µm, which is similar to the capillary’s outer diameter. The tips of the fibers are positioned at the distance of 1mm, where in the middle of the distance between the tips is positioned the capillary (Fig. 9).

Fig. 8. Ramp geometry used in experiments.

Fig. 9. Optical path. The geometry of the optical path defines the value

of Δls, required in equation (3). The approximate value of Δls may be interpreted as, visible by fiber, section of capillary at its axis. Therefore it may be calculated using numerical aperture of fiber and its core diameter. As the optical paths are the same Δls=Δl1=Δl2=Δl3≈ 1.2 mm.

3.2. Optoelectronic Signal Processing As light source, three fiber coupled LEDs with

three different emission wavelengths were used. The lowest-power LEDs were selected from Thorlabs list of high power devices: M490F2, blue with the 490 nm wavelength; M565F1, green/yellow - 565 nm wavelength; and M625F1, red - 625 nm wavelength. The diodes were connected to DC2100 controllers operating in the light modulation mode. The receiving fibers were inserted into the optoelectronic detection unit of our own design, presented previously in [41]. The optoelectronic unit was connected to a PC by an analog input of IOtech Personal DAQ 3000 data acquisition system. That system was also used to monitor the ambient temperature with two LM35DT circuits and control it with a radiant heater at 25 °C.

To operate the system, at the 0.1 s sampling rate, a script in DasyLab 10 was designed. The script algorithm is based on the signal filtration and signal derivative analysis [42].

The raw data collected for acetone are presented in Fig. 10. The initial values of signals for different paths are off-set on Fig. 10 for clarity. Moreover, the differences of the initial signal values do not matter, as the sensor operation is based on the differences of the measured time values. The measured time differences for acetone increase with capillary length coordinate: Δt1<Δt2<Δt3 and [(t2-t1) = 0.9 s] < [(t3-t2) = 2.6 s]. Therefore obtained results are in agreement with theoretical model represented with equations (5) and (6).

Fig. 10. Acetone characteristics.

The results of measurement of fresh PD presented in Fig. 11 show that, contrary to acetone, the Δts are measurable in each optical patch Δt1=0.2 s, Δt2=0.3 s, Δt3=0.7 s, and that the time intervals (t2-t1) = 6.5 s and (t3-t2) = 13 s are greater than for acetone, as fuel viscosity is much greater than that of acetone.

Fig. 11. Measurement signals of fresh premium quality

petrodiesel PD.

In the presented set-up, the t1 measurement value is uncertain, as the filling of the vessel may not be repeatable when a hand held pipetor is used.


18

4. Experimental Results In this section are presented the diesel and

biodiesel fuels parameters examination as well as experimental classification procedure results.

4.1. Diesel and Biodiesel Fuels Used for Examination

The operation of the sensor was examined with

fuels that are mixtures prepared from components according to the European Union standards. The fuels were fresh, or stored in room condition for two years.

Base oil (BO) was prepared from crude oil distillation products. The petrodiesel (PD) was prepared with additives according to the EN-590 norm. Biodiesel fuels (BDx – x is the volume ratio of the biocomponent) were prepared from PD with addition of fatty acids methyl esters (FAME) and other additives according to the EU standard. The lubricants were added in different volumes, but in the standard range, to different fuel mixtures. This enables testing the lubricant effects on CI and the fuel classification. The results of laboratory examination of prepared fuels are presented in Table 2. The accuracy of the density measurement was ±2.0 kg/m3 while the resolution was ±0.1 kg/m3. Kinematic viscosity, lubricity (ISO 12156) and oxidation stability (EN 16091) measurement accuracy was ±1 %. Kinematic viscosity was measured at 40 °C, while the density at 15 ºC.

The density and kinematic viscosity increase monotonic with FAME fuel component concentration (Fig. 12), while cetane index decreases monotonically. The fuels BO, BD30 BD50, BD70 and BD100 do not meet the standards of density, viscosity and FAME ratio, but meet the quality test of CN. In our opinion (tests on CFR-5), fuels from PD up to BD30 can be used without problems. The characteristics of lubricity and oxidation stability of examined fuels are not monotonic versus FAME fuel component concentration, as depicted in Fig. 13.

The calculated values of surface tension of fuels, basing on equation (6), as well as on the assumption that average advancing contact angle is 0 and using data collected in Table 2, are presented in Fig. 14. The obtained results show that the assumption for the advancing angle is not correct when absolute values are under consideration. But experiment show, that when fuel speed decreases, the calculated surface tension increases and tends to the static measured values presented in Table 1.

Fig. 12. Density and kinematic viscosity of examined

fuels vs. FAME component concentration.

Fig. 13. Lubricity and oxidation stability of examined

fuels vs. FAME component concentration.

Table 2. The parameters of investigated fuels.

Fuel acronym

Parameter

FAME [%] Density [kg/m3] Kinematic viscosity [mm2/s]

Lubricity WS 1.4

[µm]

Oxidation stability

[min] CN

BO 0 805.0 1.581 734 55.02 49.8 PD 0 832.6 3.367 321 37.56 59.6

BD02 2 833.6 3.3825 209 41.36 58.4 BD04 4 834.5 3.394 202 50.28 58.3 BD06 5.8 835.4 3.401 215 50.19 58.6 BD08 7.8 836.4 3.413 207 50.18 57.3 BD10 9.7 837.4 3.432 214 33.08 57.3 BD30 28.8 847.0 3.595 210 32.58 54.9 BD50 48.9 857.4 3.813 216 27.23 53.6 BD70 68.6 867.3 4.058 209 23.50 53.7 BD100 100 883.2 4.509 196 19.02 51.2


19

The non-monotonic character of surface tension versus FAME concentration is visible, but should not disturb further considerations as the lubricity of examined fuels also does not depend monotonically. These results show also that the values of surface tensions are probable and that the advancing contact angles differ significantly for the analyzed fuels as differs the lubricity.

Fig. 14. Surface tensions of fresh fuels calculated from equation (6).

The comparison of surface tension calculated with above mentioned assumption for methods based on local (5) and average measurement (6) is presented in Fig. 15. Obtained results confirm previous conclusion of the speed influence on calculated surface tensions. It is thus evident that standardization of dynamic surface tension measurements of diesel fuel is not a trivial task.

Fig. 15. Surface tensions of fresh fuels calculated from

equations (5) and (6).

4.2. Procedure of Diesel Fuel Classification

On the base of previous results, it is clear that the direct calculation of physical values of fuel parameters including dynamical values of surface tension is methodologically difficult. But, methods represented

by equations (5) and (6) in tX-tX-1 and Δt3 contain information on surface tension dynamics, density and viscosity. The difference in (t3-t2) or (t2-t1) times varies much more from one fuel to another than the fuels densities or viscosities.

For the purpose of diesel fuel classification with the use of dynamical capillary rise measurement, freshly prepared fuels from PD to BD30 meet quality requirements. On the base of oxidation stability investigation, all fuels stored for two years in closed dark tanks do not meet quality requirements. The volumes of stored fuels seemed to be constant, no presence of resins was observed, but the fuels were more transparent, which indicated that chemical reactions occurred in the fuels under storage.

Data of time intervals obtained for fresh fuels are presented in Fig. 16. The amplitude difference of (t3-t2) is greater than (t2-t1). Therefore, the allowed range of (t3-t2) is set. It should be pointed that this determination of fuel quality is not always accurate, as BD50 biodiesel fuel looks here acceptable.

Fig. 16. Measurement data of fresh diesel fuels related

for method described by (6) – time intervals. The time interval values of fuels stored for two

years are presented in Fig. 17.

Fig. 17. Measurement data of stored diesel fuels related

for method described by (6) – time intervals.


20

The results presented in Fig. 17 in comparison to the results seen in Fig. 16 indicate that the measured times intervals increase for stored fuels, but not monotonically. The proposed time difference range discrimination is again not conclusive, as spoiled BD02 is classified as fulfilling the quality requirements.

The results of examination of time differences Δt3 for fresh and stored fuels are presented in Fig. 18. The set range of allowed range of time difference to meet proper quality fuel is again not conclusive as according to it the stored base oil meets requirements.

Conclusive examination can be obtained with simultaneous examination of time difference and time interval data as presented in Table 3.

Therefore, on the base of data collected in the experiments, the set of parameters determining the useful state of biodiesel fuel should include the limits of the time intervals t3-t2, the limits of time differences Δt3, and the results of logical operator [(t3-t2) and Δt3].

Fig. 18. Measurement data of fresh and stored diesel fuels related for method described by (5) – time

differences.

Table 3. Comparison of fuels classification results.

Fuel acronym

Fuel condition

Classification result of fuels fit for use

Time interval (Ti) Time difference (Td) Ti and Td Assumed quality

BO Fresh N N N N Old N Y N N

PD Fresh Y Y Y Y Old N N N N

BD02 Fresh Y Y Y Y Old Y N N N

BD04 Fresh Y Y Y Y Old N N N N





BD50 Fresh Y N N N Old N N N N

BD70 Fresh N Y N N Old N N N N

BD100 Fresh N N N N Old N N N N

N – Wrong classification result; Y – Proper classification result.

5. Conclusions

We proposed a sensor working on the principle of optical examination of diesel fuel under dynamical capillary rise conditions. The presented device has been tested with fuels of different compositions and

storage age, and it shows great promise for practical implementation.

The experiment data indicate that stored fuels may flow significantly slower than fresh ones, and their flow speeds can be determined by measurements of time intervals. The change in flow speed modifies the conditions of fuel injection and spray formation. The


21

direct analysis of dynamical phenomenon of surface tension of diesel fuel brings about interpretation difficulties.

The analysis of the measured signals of diesel and biodiesel fuels showed the relationship of the times of fuel flow in the capillary with the useful state of diesel fuels. Especially, the summarized results of the experiments led us to the conclusion that Δt3 time measurement data differences and t3-t2 time intervals relate more clearly to the acceptable quality fuels than independent set of surface tension, viscosity and density data.

We showed that the information on useful state of diesel fuel could be presented in the form of recommended ranges of times of the fuel crossing sections of an inclined capillary during the capillary fill. Specific results can be obtained applying logical and operator to the capillary fill analysis using two methods: time difference and time interval. This approach is not yet as accurate as other laboratory methods, but its intrinsic advantages of simplicity of design and low instrumentation costs could make the technology viable in the near future.

Acknowledgements

This work was partially supported by the NCBiR/PGNiG grant Polish Technology for Shell Gas, task T3.1.

References [1]. Department of Industry, Science and Resources,

Setting national fuel quality standards - Discussion paper 4, in Operability fuel parameters (petrol and diesel), Environment Australia, 2001.

[2]. E. Torres-Jimenez, M. Pilar-Dorado, B. Kegl, Experimental investigation of injection characteristics of bioethanol-diesel fuel and bioethanol-biodiesel blends, Fuel, Vol. 90, 2011, pp. 1968-1979.

[3]. G. E. Totten ed., T. W. Rayan, Diesel Fuel Combustion Characteristics, in Fuels and Lubricants Handbook, properties, performance, and testing, ASTM International, 2003, pp. 717–720.

[4]. Z, Hu, L. T. M. Somers, T. Davies, A. McDougall, R. F. Cracknell, A study of liquid fuel injection and combustion in a constant volume vessel at diesel engine conditions, Fuel, Vol. 107, 2013, pp. 63-73.

[5]. S. K. Aggrawal, A review of spray ignition phenomena: present status and future research, Progress in Energy and Combustion Science, Vol. 24, 1998, pp. 565-600.

[6]. K. Durdag, Solid state acoustic wave sensors for real‐time in‐line measurement of oil viscosity, Sensor Review, Vol. 28, 2008, pp. 68 – 73.

[7]. R. A. Atkins, J. H. Gardner, W. N. Gibler, C. E. Lee, M. D. Oakland, M. O. Spears, V. P. Swenson, H. F. Taylor, J. J. McCoy, and G. Beshouri, Fiber-optic pressure sensors for internal combustion engines, Applied Optics, Vol. 33, 1994, pp. 1315-1320.

[8]. C. Schaschke, I. Fletcher, N. Glen, Density and Viscosity Measurement of Diesel Fuels at Combined High Pressure and Elevated Temperature, Processes, Vol. 1, 2013, pp. 30-48.

[9]. G. Fannin, Distillation process analyser with ASTM D86 Compliance, Petro Industry News, Vol. August/September, 2013, pp. 40-41.

[10]. M. Balat, H. Balat, A critical review of biodiesel as a vehicular fuel, Energy Conversion and Management, Vol. 49, 2008, pp. 2727–2741.

[11]. P. G. de Gennes, F. Brochard-Wyart, D. Quéré, Capillary and Wetting Phenomena–Drops, Bubbles, Pearls, Waves, Springer-Verlag, 2004.

[12]. D. L. Katz, W. Saltman, Surface Tension of Hydrocarbons, Industrial Engineering Chemistry, Vol. 31, 1, 1939, pp. 91–94.

[13]. Shi‐Yow Lin, Hong‐Chi Chang, Lung‐Wei Lin, Pao‐Yao Huang, Measurement of dynamic advancing receding contact angle by video‐enhanced sessile drop tensiometry, Review of Scientific Instruments, Vol. 67, 1996, pp. 2852-2858.

[14]. Q. Chen, E. Ramé, S. Garoff, The breakdown of asymptotic hydrodynamic models of liquid spreading at increasing capillary number, Physics of Fluids, Vol. 7, 1995, pp. 2631-2639.

[15]. D. O. Topping, G. B. McFiggans, G. Kiss, Z. Varga, M. C. Facchini, S. Decesari, and M. Mircea, Surface tensions of multi-component mixed inorganic organic aqueous systems of atmospheric significance: measurements, model predictions and importance for cloud activation predictions, Atmospheric Chemistry and Physics, Vol. 7, 2007, pp. 2371–2398.

[16]. J. D. Pandey, T. Srivastava, Surface tension - A theoretical study of multicomponent solutions, Journal of Molecular Liquids, Vol. 155, 1, 2010, pp. 51–56.

[17]. M. G. Cottingham, C. D. Bain D. J. T. Vaux, Rapid method for measurement of surface tension in multiwell plates, Laboratory Investigation, Vol. 84, 2004, pp. 523–529.

[18]. Ruey‐Yug Tsay, Shaw‐Chern Yan, Shi‐Yow Lin, A video‐enhanced plate method for simultaneous measurements of surface tension and contact angle, Review of Scientific Instruments, Vol. 66, 1995, pp. 5065-5069.

[19]. M. Borecki and J. Kruszewski, Intelligent high resolution sensor for detecting of liquid mediums, Optica Applicata, Vol. 31, 2001, pp. 691–699.

[20]. K. D. Danov, et al., Capillary meniscus dynamometry – Method for determining the surface tension of drops and bubbles with isotropic and anisotropic surface stress distributions, Journal of Colloid and Interface Science, Vol. 440, 2015, pp. 168–178.

[21]. H. W. Xiang, Surface Tension, in The Corresponding-States Principle and its Practice Thermodynamic, Transport and Surface Properties of Fluids, Elsevier, 2005, pp. 215–228.

[22]. E. A. Melo-Espinosa, et al., Surface tension prediction of vegetable oils using artificial neural networks and multiple linear regression, Energy Procedia, Vol. 57, 2014, pp. 886 – 895.

[23]. C. A. W. Allen, K. C. Watts, R. G. Ackman, Predicting the surface tension of biodiesel fuels from their fatty acid composition, Journal of the American Oil Chemists' Society, Vol. 76, 3, 1999, pp. 317–323.

[24]. Ö. L. Gülder, G. Moroz, Surface tension as an indicator of cetane number of diesel fuels, Warrendale, PA: Society of Automotive Engineers, 1986.

[25]. P. Krause, O. Klyus, Application of surface tension parameter for estimation of fuel atomization process in Diesel engines, Scientific Journals - Maritime University of Szczecin, Vol. 36, 2013, pp. 98–101.


22

[26]. A. A. Keller, V. Broje, K. Setty, Effect of advancing velocity and fluid viscosity on the dynamic contact angle of petroleum hydrocarbons, Journal of Petroleum Science and Engineering, Vol. 58, 2007, pp. 201–206.

[27]. B. Jańczuk, E. Chibowski, T. Białopiotrowicz, Influence of N-alkanes on contact angle and flotability of quartz, Journal of Materials Science, Vol. 25, 1990, pp. 1353–1356.

[28]. F. Wang, J. Wu, and Z. Liu, Surface Tensions of Mixtures of Diesel Oil or Gasoline and Dimethoxymethane, Dimethyl Carbonate, or Ethanol, Energy Fuels, Vol. 20, 6, 2006, pp. 2471–2474.

[29]. K. Durdag, Solid state acoustic wave sensors for real‐time in‐line measurement of oil viscosity, Sensor Review, Vol. 28, 2008, pp. 68 – 73.

[30]. P. Firek, et al., Electric Characterization and Selective Etching of Aluminum Oxide, Plasma Processes and Polymers, Vol. 6, 2009, pp. 840-843.

[31]. D. Sparks, K. Kawaguchi, M. Yasuda, D. Riley, V. Cruz, N. Tran, A. Chimbayo, N. Najafi, Embedded MEMS‐based concentration sensor for fuel cell and biofuel applications, Sensors and Actuators A, Vol. 145‐146, 2008, pp. 9‐13.

[32]. S. Y. Yurish, N. V. Kirianaki, R. Pallàs‐Areny, Universal frequency‐to‐digital converter for quasi‐digital and smart sensors: specifications and applications, Sensor Review, Vol. 25, 2, 2005, pp. 92 – 99.

[33]. P. Baptista, P. Felizardo, J. C. Menezes, M. J. N. Correia, Multivariate near infrared spectroscopy models for predicting the methyl esters content in biodiesel, Analytica Chimica Acta, Vol. 607, 2008, pp. 153-159.

[34]. M. Wlodarski, A. Bombalska, M. Mularczyk-Oliwa, M. Kaliszewski, and K. Kopczyński, Fluorimetric techniques in analysis and classification of fuels, Proc. SPIE of the 10th Symposium on Laser Technology, Vol. 8703, 2012, pp. 87030B.

[35]. M. Borecki and M. L. Korwin-Pawlowski, Optical capillary sensors for intelligent microfluidic sample classification, in Nanosensors: Theory and Applications in Industry, Healthcare and Defence, T. C. Lim, Ed., CRC Press, 2011, pp. 215– 245.

[36]. M. Borecki, P. Doroz, P. Prus, P. Pszczółkowski, J. Szmidt, M. L. Korwin-Pawlowski, J. Frydrych, A. Kociubiński and M. Duk, Fiber Optic Capillary Sensor

with Smart Optode for Rapid Testing of the Quality of Diesel and Biodiesel Fuel, International Journal of Advances in Systems and Measurements, Vol. 7, 2014, pp. 57–67.

[37]. A. Baranowska, P. Miluski, M. Kochanowicz, J. Zmojda, D. Dorosz, Capillary optical fibre with Sm3+ doped ribbon core, Proc. SPIE of the Conference Photonics Applications in Astronomy, Communications, Industry, and High-Energy Physics Experiments, Vol. 9662, 2015, pp. 96620Z.

[38]. M. Geca, M. Borecki, M. L. Korwin-Pawlowski, A. Kociubiński, Local liquid sample heating: integration and isolation of a micro-heater, Proc. SPIE of the Conference Photonics Applications in Astronomy, Communications, Industry, and High-Energy Physics Experiments, Vol. 9662, 2015, pp. 96620E.

[39]. P. Prus, M. Borecki, M. L. Korwin-Pawlowski, M. Duk, Software detection of characteristics data of optical signals received in multiparametric capillary sensors of diesel fuel, Proc. SPIE of the Conference Photonics Applications in Astronomy, Communications, Industry, and High-Energy Physics Experiments, Vol. 9662, 2015, pp. 966209.

[40]. M. Borecki, J. Szmidt, M. L. Korwin-Pawlowski, A. Kociubinski, M. Duk, J. Frydrych, P. Prus, Capillary Rise Multiparametric Sensor for Testing of Diesel and Biodiesel Fuel, in Proceedings of the 6th International Conference on Sensor Devices Technologies and Applications (SENSORDEVICES’15), Venice, Italy, 23-29 August 2015, pp. 37-43.

[41]. M. Borecki, J. Szmidt, T. Niemiec, M. Szmidt, K. Urbańska, M. L. Korwin-Pawlowski, M. Duk, A. Kociubiński, Sensing of Essential Amino Acids Behaviour Under Fast Thermal Shocks in Liquid Water Environment, in Proceedings of the 5th International Conference on Sensor Devices Technologies and Applications (SENSORDEVICES’14), Lisbon, Portugal, 16–20 November, 2014, pp. 32–38.

[42]. P. Prus, M. Borecki, M. L. Korwin-Pawlowski, A. Kociubiński, and M. Duk, Automatic detection of characteristic points and form of optical signals in multiparametric capillary sensors, Proc. SPIE of the Conference Photonics Applications in Astronomy, Communications, Industry, and High-Energy Physics Experiments, Vol. 9290, 2014, pp. 929009.

___________________

2015 Copyright ©, International Frequency Sensor Association (IFSA) Publishing, S. L. All rights reserved (http://www.sensorsportal.com)


23



Novel Smart Glove Technology as a Biomechanical Monitoring Tool

1 Brendan O’FLYNN, 1 J. T. SANCHEZ, 1 S. TEDESCO, 2 B. DOWNES,

3 J. CONNOLLY, 3 J. CONDELL, 3 K. CURRAN 1 Tyndall National Institute, University College Cork, Cork, Ireland

2 Waterford Institute of Technology, Waterford, Ireland 3 Computing & Engineering, Magee College, Ulster University, Derry, N. Ireland

1 Tel.: +353 21 2346041, fax: +353 21 49004958 E-mail: [email protected]

Received: 31 August 2015 /Accepted: 5 October 2015 /Published: 30 October 2015 Abstract: Developments in Virtual Reality (VR) technology and its overall market have been occurring since the 1960s when Ivan Sutherland created the world’s first tracked head-mounted display (HMD) – a goggle type head gear. In society today, consumers are expecting a more immersive experience and associated tools to bridge the cyber-physical divide. This paper presents the development of a next generation smart glove microsystem to facilitate Human Computer Interaction through the integration of sensors, processors and wireless technology. The objective of the glove is to measure the range of hand joint movements, in real time and empirically in a quantitative manner. This includes accurate measurement of flexion, extension, adduction and abduction of the metacarpophalangeal (MCP), Proximal interphalangeal (PIP) and Distal interphalangeal (DIP) joints of the fingers and thumb in degrees, together with thumb-index web space movement. This system enables full real-time monitoring of complex hand movements. Commercially available gloves are not fitted with sufficient sensors for full data capture, and require calibration for each glove wearer. Unlike these current state-of-the-art data gloves, the UU / Tyndall Inertial Measurement Unit (IMU) glove uses a combination of novel stretchable substrate material and 9 degree of freedom (DOF) inertial sensors in conjunction with complex data analytics to detect joint movement. Our novel IMU data glove requires minimal calibration and is therefore particularly suited to multiple application domains such as Human Computer interfacing, Virtual reality, the healthcare environment. Copyright © 2015 IFSA Publishing, S. L. Keywords: Data glove, IMU, Virtual reality, Arthritis, Joint Stiffness, Hand Monitoring. 1. Introduction

Data gloves contain strategically placed sensors controlled by circuitry that communicates finger joint movement to an end device. In recent years data gloves have been evaluated by researchers as an effective replacement for the universal goniometer (UG) [12–17]. Results showed comparable

repeatability to the UG with the added advantage of simultaneous angular measurement and removal of intra-tester and inter-tester reliability problems associated with the UG. Data gloves however have several drawbacks; they require laborious calibration, are difficult to don and doff; and are designed to fit specific hand sizes and so require small, medium and large gloves to fit all hand variations. The first




24

iteration of our system was developed using a state-of-the-art 5DT Ultra 14 data glove [18]. In this paper, our inertial measurement unit (IMU) Smart Glove is evaluated against this data glove for accuracy and repeatability and further validated using the Vicon motion capture system [19].

Virtual reality (VR) systems can be segmented into one of three experiences: non-immersive, semi-immersive, and fully immersive. Non-immersive systems would be those that can be visualized on a desktop computer. Semi-immersive VR environments incorporate images projected on the walls (e.g., cave automatic virtual environment, better known by the acronym CAVE). For a period of time, the user may superficially succumb to the perception of “being there”, but all the while still be aware of their real world surroundings. Finally, there is fully-immersive technology. In these systems, real-world visual and auditory cues are completely blocked out and the user has a sensory experience of being inside the computer-generated world. The experience is made ever more real through the use of hand-held and/or wearable devices that in some cases deliver haptic feedback which invoke sensations of touch. To enable Human Computer Interaction (HCI) in this immersive fashion, high precision data acquisition systems need to be developed which are accurate, require minimal calibration and which provide real-time data streams wirelessly. The development of such a glove based system lends itself to multiple use cases including the Gaming environment and hand healthcare (e.g., Rheumatoid Arthritis (RA) monitoring).

This paper is organized as follows. Section 2 describes the glove hardware. Section 3 addresses the system implementation, the calibration of the glove anddescribes the graphical user interface (GUI). Section 4 describes the Data analytics and Post processing while Section 5 gives an account of the tests and results. Section 6 goes into the conclusions. The acknowledgment section closes the paper.

1.1. Virtual Reality (VR)

To be compatible with the Virtual Reality use case, it is important that any glove system developed for HCI adheres to the requirements detailed below:

Accuracy & Precision. Accuracy is the degree of closeness to a quantity's actual true value. Precision is the degree to which repeated measurements give the same quantity. Here, we define accuracy and precision to consist of position and orientation. Different parts of the hand should have priority for accuracy: a) Mapping of the center of the virtual hand is most important for many VR applications, b) The finger tips are next most important for accuracy as these joints can be estimated via inverse kinematics and other constraints, c) The skeleton/joints of the hand are next most important for accuracy.

Consistent recognition of gestures. Like speech recognition, if a gesture recognition system occasionally misinterprets signals then a break in

presence occurs and users can become frustrated. Accidental gestures (known as false positives) are also a problem (e.g., accidentally signaling a command when unconsciously “talking with the hands”).

Low latency. The faster the response of the system, then the more pleasant the user experience and the more easily users can enter a state of flow.

Simulation of button presses. Some applications will greatly benefit from simulation of button presses that provide a sense of self-haptic feedback (e.g., by touching two fingers together) and to control the game and system.

1.2. Rheumatoid Arthritis Assessment RA is an auto-immune disease which attacks the

synovial tissue lubricating skeletal joints and is characterized by pain, swelling, stiffness and deformity. This systemic condition affects the musculoskeletal system including bones, joints, muscles and tendons that contribute to loss of function and Range of Motion (ROM). Early identification of RA is important to initiate treatment, reduce disease activity, restrict its progression and ultimately lead to its remission. Clinical manifestations of RA can be confused with similar unrelated musculo-skeletal and muscular disorders. Identifying its tell-tale symptoms for early diagnosis has been the long-term goal of clinicians and researchers. Classifiers such as the Disease Activity Score (DAS) and Health Assessment Questionnaire (HAQ) provide an outcome measurement that reflects a patient’s severity of RA disease activity. Such measurements are subjective and can be influenced by other factors such as depression or unrelated non-inflammatory conditions. Traditional objective measurement of RA using the universal goniometer (UG) and visual examination of the hands is labour intensive, open to inter rater and intra-rater reliability problems.

The DAS and HAQ [2, 3] are commonly used to measure disease onset and to assess disease status during clinical assessment [1]. Joint Stiffness is a common condition of RA that affects a patients’ ability to perform basic activities and daily functions [4, 5]. Several objective measurement systems have been devised by researchers and assessed in clinical trials for effectiveness as a joint stiffness measurement device [6–11].

2. Tyndall Glove HW Description

The objective of our IMU Smart Glove is to quantitatively measure finger joint ROM including flexion, extension, adduction and abduction of the MCP, PIP and DIP joints of the fingers and thumb in degrees, together with thumb-index web space, palmar abduction to assist medical clinicians with the accurate measurement of the common condition of loss of movement in the human hand in patients with arthritis. All Smart Glove functionality is maintained,


25

controlled and analyzed by our in-house developed software system.

The described glove is a second generation iteration of the system by the authors as described in previous work [20].

Fig. 1. The IMU Smart Glove rev 2.

2.1. System HW Description

The IMU glove, shown in Fig. 1, has been manufactured using a mix of stretchable & flexible technology. Stretchable areas of the device cross each finger joint so they can conform to the human hand.

The glove includes 16 9-axes IMU’s (each including 3-axis accelerometer, 3-axis gyroscope and 3-axis magnetometer) strategically placed to account for the degrees of freedom (DOF) of each finger joint of the hand. IMUs are positioned on the stretchable interconnect and are located on the phalange of each finger segment to measure orientation and biomechanical parameters.

Each IMU provides 6 DOF motion information (3 translational + 3 rotational) and 3D orientation information. By placing an IMU at both sides of each finger joint, (that is one per each finger bone and

another one on the palm of the hand), the relative orientation of each IMU is calculated and used to generate angular and velocity movement throughout flexion and extension exercise of each finger joint and to calculate splaying of each finger.

2.2. Microcontroller

The processor selected for use in the system is an Atmel AVR32 UC3C 32 Bit Microcontroller. This is a high performance, low power 32-bit AVR microcontroller with built in single precision floating point unit. It was selected to enable complex embedded algorithms focused on motion analysis to be developed for real time low power consumption operation. 2.3. Wireless Communication

The RS9110-N-11-22 [21] module shown in Fig. 2 is an IEEE 802.11b/g/n WLAN device that directly provides a wireless interface to any equipment with a UART or SPI interface for data transfer. It integrates a MAC, baseband processor, RF transceiver with power amplifier, a frequency reference, and an antenna in hardware. It also provides all WLAN protocols and configuration functionality. A networking stack in embedded in the firmware to enable a fully self-contained 802.11n WLAN solution for a variety of applications.

The module incorporates a highly integrated 2.4 GHz transceiver and power amplifier with direct conversion architecture, and an integrated frequency reference antenna. The RS9110-N-11-22 comes with flexible frameworks to enable usage in various scenarios including high throughput networking applications.

Fig. 2. RS9110-N-11-22 System Block Diagram.

The system operates according to a low complexity standard 4-wire SPI interface with the capability of operation up to a maximum clock speed of 25MHz. The communications module conforms to IEEE 802.11b/g/n standards and includes hardware

accelerated implementation of WEP 64/128-bit and AES in infrastructure and ad-hoc modes. Themodule supports multiple security features such as WPA/WPA2-PSK, WEP, TKIP which makes it compatible with all medical ERP systems.


26

2.4. Sensors

The MPU-9150 [22] is a full three axis inertial

measurement system incorporating tri-axis angular rate sensor (gyroscope) with sensitivity up to 131 LSBs/dps and a full-scale range of ±250, ±500, ±1000, and ±2000 dps, tri-axis accelerometer with a programmable full scale range of ±2 g, ±4 g, ±8 g and ±16 g and a tri-axis compass with a full scale range of ±1200 μT. This module incorporates embedded algorithms for run-time bias and compass calibration, so no user intervention is required. The MPU-9150 features three 16-bit analog-to-digital converters (ADCs) for digitizing gyroscope outputs, three 16-bit ADCs for digitizing accelerometer outputs, and three 13-bit ADCs for digitizing magnetometer outputs. For precision tracking of both fast and slow motions, the module features a user programmable gyroscope full-scale range of ±250, ±500, ±1000, and ±2000°/sec (dps), a user programmable accelerometer full-scale range of ±2 g, ±4 g, ±8 g, and ±16 g, and a magnetometer full-scale range of ±1200 μT.

2.5. Additional Features

To make the system adaptable in operation and

compatible with a wide range of use cases outside the immediate application of RA monitoring, the IMU Smart Glove system also incorporates such features as optional storage via a micro SD card, battery monitoring and recharge facility, a USB bootloader, USB communication interface, and 15 Analogue inputs for optional resistive sensors (e.g., bend sensors or force sensors). The analogue front end is a buffered voltage divider to enable additional sensing functionality.

2.6. Flex Technology

The IMU Smart Glove PCB is a combination of

flexible and stretchable PCB technology [33]. The stretchable material enables the microsystem to closely replicate mechanical properties of the human hand more accurately than standard flexible technology. As shown in Fig. 3, stretchable PCB sections are incorporated on hand areas crossing several finger joints to enable flexion at the knuckles and provide an interconnect mechanism between the “islands” of rigid PCB substrates which incorporate the WIMU technology.

The stretchable PCB technology is available from the company “Q.P.I. Group”. The substrate material is polyurethane. it is possible to obtain a stretch factor of up to 30 % to enable wearable sensor system interconnect, depending on the design implemented on the copper pattern.

Fig. 3. PCB top view and flexible areas.

3. System Implementation

All the system embedded code is implemented using the Atmel Studio 6 IDE. Currently the implementation includes full application code that continuously reads sensor outputs and wirelessly transmits their data through a TCP socket.

The accuracy of IMU-based real time motion tracking algorithms is highly influenced by sensor sampling rate. Therefore a fundamental design requirement of the IMU Smart Glove was high application throughput to facilitate the development of algorithms using suitable PC SW such as MATLAB, C# and Unity. In addition, it was envisaged that once the algorithms would have been fully developed and tested, they would be fully implemented on the embedded platform. This eliminates the requirement for a high throughput device and allows for a low power implementation for example using BLE in a potential third generation of the glove.

To ensure maximum achievable sampling rates and computation time are compatible with the application scenario envisaged as specified in conjunction with clinical partners regarding signal temporal granularity, it was decided not to share the I2C bus between each of the 16 MPU9150’s. Instead, dedicated I2C lines are provided to each one of the sensors and are driven in parallel. This provides the added advantage of ensuring synchronization between all IMU sensors.

3.1. Case 1. Raw Data Transmission

The embedded processor enables multiple modes of operation depending on the use case and degree of data granularity required. Having the wireless system transmitting raw data at the highest achievable data rate is desirable for the development of the analytics as it is more practical to develop them using PC based SW (real time or post processing) and then porting them to the embedded system than develop them directly within the embedded system.

3.2. Case 2. Transmission of Raw Data and Information

The wireless system transmits raw data and quaternions/rotation matrix (from gyros) at the highest


27

achievable data rate. Quaternions then will be subject to drift/errors and the analytics to correct for this are implemented within the controlling software. At this stage we have a clear idea of the maximum processing time that could be allocated in the embedding to this task and that is taken into consideration when designing these algorithms.

3.3. Case 3. Transmission of Processed Data

With the wireless system with full analytics embedded, the internal sampling rate of the sensors should be kept to a maximum achievable SR, the high wireless data rate might no longer be required.

The processing time (per sample cycle) allocated to the embedded tasks and estimated maximum sampling rate / Application Throughput with the microcontroller running at 48 MHz are shown in table I Depending on the computational complexity of the drift correction algorithm, (which are under development at the Tyndall Institute), different application data throughputs are achievable as shown in Table 1.

Table 1. Processing time requirements for motion data analysis.

Raw data Raw Data + Quaternions

Full analytics

embedded Sensor Sampling (16 IMUs)

~ 0.9 ms ~ 0.9 ms ~ 0.9 ms

Wireless Communications

~0.3 ms 0.4-0.5 ms ~0.4-0.5 ms

Quaternions / Rotation Matrix Algorithm

None ~ 0.3-0.5 ms ~0.3-0.5 ms

Drift correction algorithms

None None ~

8/3/2/1.33/0.5 ms

Estimated application throughput /data rate

~750 Hz / 2 Mbps

~500 Hz / 2 Mbps

100/200/250/300/400 Hz

3.4. Calibration using Accelerometry and Gyroscope Datasets

Data glove accuracy and repeatability is affected by the non-linear nature of glove sensor output and any misalignment between the wearers hand and data glove sensor positioning. Data glove sensor calibration improves sensor accuracy and matches the boundaries of each sensor to those of each finger joint. A calibration routine requires the glove wearer to position groups of finger joints such as MCP’s and PIP’s at specific poses. Each pose places a finger joint group and relevant data glove sensors at their minimum and maximum boundaries. The IMU Smart Glove uses on-board sensors to automatically calibrate each glove sensor, regardless of the wearer’s joint

flexibility. Each glove accelerometer sensor is sampled when the hand is in a neutral position to calculate finger joint thickness and slope offset, and used during angular calculation. Accelerometers placed on each one of the finger’s phalanges provide information with regards to the inclination to gravity of the phalanx. The output response of each sensor provides information on the orientation of the sensor to gravity as shown in Fig. 4. The orientation to gravity of each one of the sensors placed on adjacent phalanges can be used to estimate the flexion of the finger.

Fig. 4. Output response vs. Orientation to gravity

For example, if the measured acceleration for a specific finger from the medial phalanx accelerometer is (Xout, Yout, Zout) = (-1,0,0) g and from the proximal phalanx accelerometer is (Xout, Yout, Zout) = (0,0,1) g, it indicates a flexion of the PIP joint of 90 degrees. The inclination to gravity is determined according to the standard formulas (1), (2) and (3):

(1)

(2)

(3)

where θ is the angle between the horizon and the x-axis of the accelerometer, ψ is the angle between the horizon and the y-axis of the accelerometer, and φ is the angle between the gravity vector and the z-axis. 3.5. GUI/User Interface

Data is streamed in real-time according to the use cases outlined above and post processed by our controlling software. This software is called

+= −

2,

2,

,1tanOUTZOUTY

OUTX

AA

Aθ

+= −

2,

2,

,1tanOUTZOUTX

OUTY

AA

Aψ

+= −

OUTZ

OUTYOUTX

A

AA

,

2,

2,1tanφ


28

‘DigitEase’ [34] shown in Fig. 5. A pivotal role of DigitEase is its ability to encapsulate movement

associated with finger joints in real time. Fig. 5 shows an example of DigitEase’s user interface.

Fig. 5. Angular output from the data glove is displayed in 3D.

Algorithms segment recorded data to extract relevant flexion and extension movement information. Each piece of sensor data is categorised into pre-repetition, flexion, sustain, extension and post-repetition movement.

Fig. 6 demonstrates a typical flexion and extension angular movement profile used by DigitEase. Segmentation of finger joint movement is required to isolate flexion and extension movement data from unrequired pre-rep, port-rep and sustain time. Flexion and extension movement is analysed for initial and final angles. Both values represent minimum and maximum ROM information for movement

repetitions and are indicators of completion and initialisation of flexion and extension movement.

DigitEase’s data analysis dashboard presents post-segmented patient movement information. Fig. 7 shows an example of the dashboard.

The dashboard displays summary information on patient-completed exercise routines. It details individual repetitions and constituent sub-elements for each exercise routine. Colour coding of each repetition segment indicates performance information. Line charts graphically depict velocity, angle-angle and relative phase information for individually selected repetitions.

Fig. 6. Chart demonstrating segments that characterise segmentation of finger joint movement.


29

Fig. 7. DigitEase data analysis dashboard.

4. Data Analytics and Post Processing

Each angular calculation is low-pass filtered to remove sensor noise. A complementary filter with error control is implemented to combine accelerometer output with gyroscope rotation angle. Gyroscope rotational angle is initially accurate and drifts over time. Accelerometer angle cannot distinguish between lateral acceleration and rotation. The complementary filter acts as a high-pass and low-pass filter on both signals. It combines estimated gyroscope rotation and accelerometer angle to create an angular output.

4.1. Algorithms for Joint Angle Estimation

Fig. 8 shows the joints of the hand along with their number of degrees of freedom. The joint angles can be calculated as the result of the relative orientation of adjacent phalanges one to another so the algorithms to estimate the joints angles from the IMUs are based on the orientation estimation of the sensors themselves. This orientation is commonly represented by quaternions. Equations (4) - (7) represent the orientation / quaternions of adjacent phalanges that are linked by each joint.

q , q (4) , (5) , (6) , (7)

Over the years, a variety of algorithms for the estimation of orientation have been developed. The majority of these are quaternion-based algorithm and

they can be divided in three main approaches: a) the deterministic approach (least-squares), b) the frequency–based approach (Complementary Filter) and c) the stochastic (Kalman filtering).

Fig. 8. Degrees of freedom of hand.

The deterministic approach was originally introduced in 1965, in the so-called Wahba’s problem [30], which is a constrained least-squares optimization problem for finding the rotation matrix between two coordinate systems from a set of weighted vector measurements. Some of such algorithms are the TRIAD (Tri-axial Attitude Determination), QUEST (Quaternion ESTimator) and FQA (Factored Quaternion Algorithm). These algorithms are based on the concept of vector matching and require measurements of constant reference vectors.

The frequency–based approach fuses the orientation estimated from accelerometers and


30

magnetometers with the orientation estimated from gyroscope with a complementary filter. This filter blends the static low-frequency information provided by accelerometers and magnetometers, and the dynamic high frequency information provided by gyroscopes. The aim of the complementary filter is to ensure a compromise between the accuracy provided by short-term integration of the gyroscope data and the long-term measurements precision obtained by the accelerometer and magnetometer [31].

Stochastic estimation algorithms use a dynamic model for predicting aspects of the time behaviour of a system and a measurement model in order to produce the most accurate estimate possible of the system state. Among all stochastic algorithms, the Kalman filter is one of the most often used algorithms for tasks that involve multisensory fusion, filtering and motion prediction [32]. The filter works in a two-step process consisting of a prediction step and an update step. In the prediction step, Kalman filter produces estimates of the current state variables, along with their uncertainties. Once the outcome of the next measurement (corrupted with error and noise) is observed, these estimates are updated using a weighted average. Because of the algorithm's recursive nature, it can run in real time using only the present input measurement and the previously calculated state and its uncertainty matrix.

5. Testing Strategies and Results

Our new data glove was assessed for accuracy and repeatability and was compared with the 5DT state-of-the-art data glove. The Vicon MX motion capture system was used during accuracy testing to independently measure angular values generated at each finger joint. Movement was recorded by Vicon and simultaneously by DigitEase whilst each glove was placed on blocks of wood cut to specific angles. Angular readings were assessed using Root Mean Square Error (RMS) to provide an indicator of the variance between each estimated angular repetition value and the expected true value influenced by the angle on each block of wood. RMS error is influenced by both positive and negative errors which are either above or below the expected true value. Therefore RMS output is a measure of the angular error. Repeatability testing examined the ability of each data glove to consistently replicate angular readings when the subjects hand was held in a repeatable position. Testing strategies were originally developed to assess data glove suitability as a replacement for the UG. Although no formal set of repeatability testing strategies exist, the strategies used by [12] have been adopted by subsequent research groups [13, 16, 23–26] and are used in this study to allow comparison between study results.

The ‘flat hand’ test examines each data glove’s ability to maintain a minimum repeatable value after full stretch of each data glove sensor. The plaster mould test examines the ability of each data glove to

reproduce angular readings when positioned in a repeatable position. In all tests, our data glove was not calibrated for the subject, the 5DT data glove was calibrated.

5.1. ‘Flat Hand’ Results

The ‘flat hand’ test results demonstrated in Table 2 show that the IMU data glove outperformed the 5DT data glove. Mean MCP readings for the IMU glove were near-perfect -0.38°, with PIP readings of -2.53°. The 5DT produced readings of 4.17° for MCP and 2.27° for PIP. The results for our IMU glove are based on a system which is not calibrated before use.

Table 2. Comparison of mean angular readings recorded during ‘flat hand’ testing.

5DT (Angle / SD)

IMU (Angle / SD)

Index MCP 2.34 (1.59) -0.59 (1.87) Index PIP 2.04 (1.05) -2.74 (0.90) Middle MCP 5.9 (0.55) 1.32 (2.26) Middle PIP 3.27 (1.13) -2.94 (1.25) Ring MCP 5.14 (0.59) -2.33 (1.21) Ring PIP 1.02 (0.52) -2.7 (1.11) Little MCP 3.32 (0.88) 0.07 (2.56) Little PIP 2.76 (1.32) -1.75 (1.31) Mean MCP 4.17 (0.90) -0.38 (1.98) Mean PIP 2.27 (1.0) -2.53 (1.14) Overall mean 3.22 (0.95) -1.46 (1.56)

5.2. Plaster Mould Test Results

Table 3 shows comparison results for plaster mould testing for the 5DT and our IMU data glove. Readings showed the IMU Smart Glove produced better repeatability for MCP and PIP joints and better overall repeatability as indicated by the lower mean range angular reading. Comparison of mean range and SD readings from plaster mould testing for each data glove.

Table 3. Plaster mould test results.

Glove MCP PIP Mean

Range SD Range SD Range SD 5DT 8.85 2.13 6.23 2.09 7.54 2.11 IMU 5.99 1.89 5.10 1.58 5.55 1.74

5.3. Glove Finger Position Accuracy Results

Table 4 shows comparison of results for the 5DT and our IMU Smart Glove compared with the Vicon motion capture system and the UG.

Results showed the goniometer had greatest overall accuracy of 93.23 % with overall RMS of 2.76°. This is in agreement with typical findings on goniometric accuracy with 95 % of intratester reliability within 5° of measurement and intertester


31

reliability in the range of 7° to 9° [27–29]. The Vicon system provided mean accuracy of 89.33 % with RMS of 5.19°. Mean accuracy percentage for each sensor including mean error and overall accuracy percentage. Table 4. Comparison between Tyndall WIMU Glove

and 5DT Glove.

Sensor Vicon 5DT Goniometer IMU Index MCP 93.31 94.20 97.95 89.57 Index PIP 91.23 92.01 90.75 91.47 Middle MCP 91.46 79.66 95.83 82.40 Middle PIP 84.08 74.97 88.96 77.29 Ring MCP 87.20 70.46 97.37 82.02 Ring PIP 86.99 91.99 90.70 89.51 Little MCP 86.14 85.83 91.28 83.38 Little PIP 94.23 74.56 93.03 86.27 Overall accuracy %

89.33 82.96 93.23 85.24

RMS 5.19 7.15 2.76 5.95

This inaccuracy was most likely caused by noise, marker occlusion, and distance of reflective markers from Vicon cameras. Our IMU data glove provided best accuracy measurement of both data gloves and demonstrated similar accuracy to the Vicon measurement system. RMS results show that readings obtained from sensors contained approximately 5.95° of error. Results shown in Table 3 indicate that all sensors demonstrated accuracy between 82 % to 91 % except for the Middle PIP sensor that had accuracy of 77.29 %. This decreased accuracy may have been caused by slight stretch of sensor cable for this particular sensor.

5.4. Comparison with Previous Trials

The results shown in Table 5 compare ‘flat hand’ and plaster mould tests for the 5DT and our IMU data glove with previous research studies involving data gloves.

Table 5. Comparison of ‘flat hand’ and plaster mould tests with previous data glove studies.

Study Flat hand

test (Range / SD)

Plaster mould test

(Range / SD) Wise et al. [12] 4.4 (2.2) 6.5 (2.6) Dipietro et al. [13] 3.84 (1.23) 7.47 (2.44) Simone et al. [15] 1.49 (0.5) 5.22 (1.61) Gentner and Classen [26]

2.61 (0.86) 6.09 (1.94)

5DT (this study) 2.27 (0.995) 7.54 (2.11) IMU (this study) 4.86 (1.56) 1.74)

The 5DT data glove demonstrated range readings that out-performed data glove findings by [12] [13] and were similar to [26]. The data glove examined by [15] provided better results than all studies including the 5DT and our IMU glove. However this glove contained only 5 sensors that recorded movement of

the MCP joints. The IMU glove performed better than all other data glove studies. Readings recorded by earlier studies are averaged for several subjects. This can hide higher inaccurate results for some subjects. For example, [12] recorded range readings from 5 subjects that varied between 2.5° to 6.7°. Results were averaged to 4.4°. Similarly, results from ‘flat hand’ testing from the study by [13] were summarised from a group of 6 male and female participants. Mean male range results went from 2.37° to 5.49° and mean female from 3.90° to 4.75°.

6. Conclusions

Data gloves have been proven to be a viable replacement for the UG and can offer unbiased finger joint ROM measurement. However their dependence on calibration reduces their usefulness in the many application spaces. The novel IMU based wireless Smart Glove detailed in this paper removes the requirement for sensor calibration using accelerometers and gyroscopes teamed with intelligent software techniques. Test results showed our IMU data glove had comparable repeatability to the UG with the added advantage of simultaneous angular measurement and removal of intra-tester and inter-tester reliability. Accuracy testing results showed the IMU data glove provided better accuracy and less overall error than the 5DT data glove with which it was compared. Of note the IMU glove required no calibration before use whilst maintaining results which demonstrated it had similar accuracy to the Vicon system.

Acknowledgements

The support of Science Foundation Ireland (SFI) as well as the National Access Program (NAP) support provided by the Tyndall National Institute is gratefully acknowledged. This work was also supported by Department of Education and Learning (DEL).

References [1]. D. M. van der Heijde, van ’t H. M, P. L. van Riel, and

L. B. van de Putte, Development of a disease activity score based on judgment in clinical practice by rheumatologists, J. Rheumatol., Vol. 20, No. 3, 1993, pp. 579–81.

[2]. J. F. Fries, P. Spitz, R. G. Kraines, and H. R. Holman, Measurement of patient outcome in arthritis, Arthritis Rheum., Vol. 23, No. 2, 1980, pp. 137–145.

[3]. J. F. Fries, P. W. Spitz, and D. Y. Young, The dimensions of health outcomes: the health assessment questionnaire, disability and pain scales, J. Rheumatol., Vol. 9, No. 5, 1982, p. 789-793.

[4]. P. Emery, F. C. Breedveld, M. Dougados, J. R. Kalden, M. H. Schiff, and J. S. Smolen, Early referral recommendation for newly diagnosed rheumatoid arthritis: evidence based development of a clinical guide, Ann. Rheum. Dis., Vol. 61, No. 4, Apr. 2002, pp. 290–297.


32

[5]. F. C. Arnett, S. M. Edworthy, D. A. Bloch, D. J. McShane, J. F. Fries, et al., The American Rheumatism Association 1987 revised criteria for the classification of rheumatoid arthritis, Arthritis and Rheumatism, Vol. 31, No. 3, Mar-1988, pp. 315–324.

[6]. J. T. Scott, Morning stiffness in rheumatoid arthritis, Ann. Rheum. Dis., Vol. 19, Dec. 1960, pp. 361–368.

[7]. V. Wright and R. J. Johns, Quantitative and qualitative analysis of joint stiffness in normal subjects and in patients with connective tissue diseases, Ann. Rheum. Dis., Vol. 20, Mar. 1961, pp. 36–46.

[8]. M. L. Ingpen and P. H. Kendall, A simple apparatus for assessment of stiffness, Ann. Phys. Med., Vol. 9, No. 5, Feb. 1968, pp. 203–205.

[9]. a Unsworth, P. Yung, and I. Haslock, Measurement of stiffness in the metacarpophalangeal joint: the arthrograph, Clin. Phys. Physiol. Meas., Vol. 3, No. 4, Nov. 1982, pp. 273–81.

[10]. A. Howe, D. Thompson, and V. Wright, Reference values for metacarpophalangeal joint stiffness in normals, Ann. Rheum. Dis., Vol. 44, No. 7, Jul. 1985, pp. 469–76.

[11]. E. Dionysian, J. M. Kabo, F. J. Dorey, and R. a Meals, Proximal interphalangeal joint stiffness: measurement and analysis, J. Hand Surg. Am., Vol. 30, No. 3, May 2005, pp. 573–9.

[12]. S. Wise, W. Gardner, E. Sabelman, E. Valainis, Y. Wong, et al., Evaluation of a fiber optic glove for semi-automated goniometric measurements, J. Rehabil. Res. Dev., Vol. 27, No. 4, 1990, p. 411.

[13]. L. Dipietro, A. M. Sabatini, and P. Dario, Evaluation of an instrumented glove for hand-movement acquisition, J. Rehabil. Res. Dev., Vol. 40, No. 2, 2003, pp. 179–189.

[14]. L. K. Simone and D. G. Kamper, Design considerations for a wearable monitor to measure finger posture, J. Neuroeng. Rehabil., Vol. 2, No. 1, Mar. 2005, p. 5.

[15]. L. K. Simone, N. Sundarrajan, X. Luo, Y. Jia, and D. G. Kamper, A low cost instrumented glove for extended monitoring and functional hand assessment, J. Neurosci. Methods, Vol. 160, No. 2, Mar. 2007, pp. 335–348.

[16]. G. Saggio, S. Bocchetti, C. A. Pinto, G. Orengo, and F. Giannini, A novel application method for wearable bend sensors, in Proceedings of the 2nd International Symposium on Applied Sciences in Biomedical and Communication Technologies, Bratislava, Slovakia, Nov-2009, pp. 1–3.

[17]. K. Li, I.-M. Chen, S. H. Yeo, and C. K. Lim, Development of finger-motion capturing device based on optical linear encoder, J. Rehabil. Res. Dev., Vol. 48, No. 1, 2011, p. 69.

[18]. 5DT, 5DT Data Glove 14 Ultra, 2011. [Online], Available at:

http://www.5dt.com/products/pdataglove14.html [retrieved: 01-2012].

[19]. Vicon Motion Systems, Vicon, 2013. [Online], Available at: http://www.vicon.com/. [retrieved: 01-2012].

[20]. B. O’Flynn, J. Sanchez, P. Angrove, J. Connolly, J. Condell, and K. Curran, Novel smart sensor glove for arthritis rehabiliation, in Proceedings of the IEEE

International Conference on Body Sensor Networks, (BSN 2013), May-2013, pp. 1–6.

[21]. Redpine Signals, RS9110-N-11-22: 802.11BGN wireless device server, RS9110-N-11-22 Product brief, 2008, [Online] Available at: http://www.redpinesignals.com/pdfs/RS9110-N-11-22 Wlan Module.pdf [retrieved: 01-2012].

[22]. MPU-9150 Product Specification Revision 4.0, InvenSense, California, Vol. 1, No. 408, 2012, pp. 1–52.

[23]. G. Kessler, N. Walker, and L. Hodges, Evaluation of the CyberGlove (TM) as a whole hand input device, GVU Technical Report, GIT-GVU-95-05, Georgia Institute of Technology, 1995.

[24]. M. Mentzel, F. Hofmann, T. Ebinger, B. Jatzold, L. Kinzl, and N. J. Wachter, Reproducibility of measuring the finger joint angle with a sensory glove, Handchir Mikrochir Plast Chir, Vol. 33, No. 1, 2001, pp. 9–64.

[25]. N. W. Williams, J. M. T. Penrose, C. M. Caddy, E. Barnes, D. R. Hose, and P. Harley, A goniometric glove for clinical hand assessment, J Hand Surg Br., Vol. 25, No. 2, 2000, pp. 200–207.

[26]. R. Gentner and J. Classen, Development and evaluation of a low-cost sensor glove for assessment of human finger movements in neurophysiological settings, J. Neurosci. Methods, Vol. 178, No. 1, Mar. 2009, pp. 138–47.

[27]. A. Hellebrandt, E. Duvall, and M. Moore, The measurement of joint motion. Part III : Reliability of goniometry, Phys Ther Rev, Vol. 29, No. 6, 1949, pp. 302–307.

[28]. E. Lewis, L. Fors, and W. J. Tharion, Interrater and intrarater reliability of finger goniometrie measurements, Am. J. Occup. Ther., Vol. 64, 2010, pp. 555–561.

[29]. N. B. Reese and W. D. Bandy, Measurement of Range of motion and muscle length: background, history and basic principles, Joint Range of Motion and Muscle Length testing, 2nd ed., Saunders/Elsevier, St. Louis, MO, 2010. pp. 3–29.

[30] G. Wahba, A least-square estimate of spacecraft attitude, SIAM Rev., 8, 1965, p. 409.

[31] H. Fourati, N. Manamanni, Position Estimation Approach by Complementary Filter-aided IMU for Indoor Environment, in Proceedings of the 12th biannual European Control Conference, Zurich, Switzerland, July 2013.

[32] G. F. Welch, The use of the Kalman filter for human motion tracking in virtual reality, Presence, Vol 18, 2009.

[33] Brendan O’Flynn, J. T. Sanchez, James. Connolly, Joan. Condell, Kevin. Curran, Philip Gardiner, Barry Downes, Integrated Smart Glove for Hand Motion Monitoring, in Proceedings of the 9th International Conference on Sensor Technologies and Applications (SENSORCOMM’15), Venice, Italy, August 23 - 28, 2015, pp. 45-50.

[34] Connolly, J., Condell, J., Curran, K. and Gardiner, P., Towards Joint Stiffness measurement of Rheumatoid Arthritis sufferers, in Proceedings of the 2nd European Conference on Design 4 Health (Design4Health), Sheffield Hallam University, UK, SheffieldPub, 2013, pp. 150.

___________________



33



Wide Spectral Sensitivity of Monolithic a-SiC:H pi’n/pin Photodiode Outside the Visible Spectrum

1, 2, 3 Manuela Vieira, 1, 2 Manuel Augusto Vieira, 1 Isabel Rodrigues,

1, 2 Vitor Silva, 1, 2 Paula Louro, 1, 2 A. Fantoni 1 Telecommunication and Computer Dept. ISEL, R. Conselheiro Emídio Navarro,

1959-007 Lisboa, Portugal 2 CTS-UNINOVA, Quinta da Torre, Monte da Caparica, 2829-516, Caparica, Portugal 3 DEE-FCT-UNL, Quinta da Torre, Monte da Caparica, 2829-516, Caparica, Portugal

1 Tel.: +351218317150, fax: +351218317144 1 E-mail: [email protected]

Received: 31 August 2015 /Accepted: 5 October 2015 /Published: 30 October 2015 Abstract: In this paper, we experimentally demonstrate the use of near-ultraviolet steady state illumination to increase the spectral sensitivity of a double a-SiC/Si pi’n/pin photodiode beyond the visible spectrum (400 nm-880 nm). The concept is extended to implement a 1 by 4 wavelength division multiplexer with channel separation in the visible/near infrared ranges. The device consists of a p-i'(a-SiC:H)-n/p-i(a-Si:H)-n heterostructure, sandwiched between two transparent contacts. Optoelectronic characterization of the device is presented and shows the feasibility of tailoring the wavelength and bandwidth of a polychromatic mixture of different wavelengths. Results show that the spectral current under steady state ultraviolet irradiation depends strongly on the wavelength of the impinging light, and on the background intensity and irradiation side allowing controlled high-pass filtering properties. If several monochromatic pulsed lights, in the visible/near infrared (VIS/NIR) range, separately or in a polychromatic mixture illuminate the device, data shows that, front background enhances the light-to-dark sensitivity of the medium, long and infrared wavelength channels, and quench strongly the low wavelengths channels. Back background has the opposite behavior; it enhances only channel magnitude in short wavelength range and strongly reduces it in the long ones. This nonlinearity provides the possibility for selective tuning of a specific wavelength. A capacitive optoelectronic model supports the experimental results. A numerical simulation is presented. Copyright © 2015 IFSA Publishing, S. L. Keywords: Amorphous SiC technology, Optoelectronics, Spectral sensitivity, UV irradiation, Photodiode, Multiplexer device, VIS/NIR decoding, Numerical simulation. 1. Introduction

The LED is a very effective lighting technology due to its high brightness, long life, energy efficiency, durability, affordable cost, optical spectrum and its colour range for creative purposes. Their application as communication device with a

photodiode as receptor, has been used for many years in hand held devices, to control televisions and other media equipment, and with higher rates between computational devices [1]. This communication path has been employed in the near infra-red (NIR) range, but due to the increasing LED lighting in homes and offices, the idea to use them for visible light




34

communications (VLC) has come up recently. Newly developed technologies, for infrared telecommunication systems, allow increase of capacity, distance, and functionality, leading to the design of new reconfigurable active filter [2-4]. To enhance the transmission capacity and the application flexibility of optical communication, efforts have to be considered, namely the fundamentals of Wavelength Division Multiplexer (WDM) based on a-SiC:H light controlled filters, when different visible signals are encoded in the same optical transmission path [5]. They can be used to achieve different filtering processes, such as: amplification, switching, and wavelength conversion.

In this paper, it is demonstrated that the same a-SiC:H device under front and back controlled near ultraviolet optical bias acts as a reconfigurable active filter in the visible and near infrared ranges, making the bridge between the infrared and the red spectral ranges. In consequence, bridging the visible spectrum to the telecom gap offers the opportunity to provide alternative and additional low cost services to improve operative production processes in office, home and automotive networks.

In Section 1, an introduction is given and in Section 2, some experimental results are presented. In Section 3, the bias controlled selector is analyzed and in Section 4, the Wavelength Division Multiplexed (WDM) based on SiC technology is described. In Section 5, the optoelectronic model gives insight the physics of the device, the decoding algorithm is presented in Section 6, and finally, in Section 7, the conclusions are presented.

2. Experimental Details

2.1. Device Configuration

The light tunable filter is realized by using a double pi’n/pin a-SiC:H photodetector produced by Plasma Enhanced Chemical Vapor Deposition (PECVD).

The device has Transparent Conductive Oxide (TCO) front and back biased optical gating elements as depicted in Fig. 1.

Fig. 1. Device configuration and operation.

The active device consists of a p-i'(a-SiC:H)-n/p-i(a-Si:H)-n heterostructure with low conductivity doped layers. The deposition conditions and optoelectronic characterization of the single layers were described elsewhere [5].

The thicknesses and optical gap of the front i'- (200 nm; 2.1 eV) and back i- (1000 nm; 1.8 eV) layers are optimized for light absorption in the blue and red ranges, respectively [6].

2.2. Device Operation

Monochromatic (infrared, red, green, blue and violet; λIR,R,G,B,V) pulsed communication channels (input channels) are combined together, each one with a specific bit sequence, impinge on the device and are absorbed accordingly to their wavelengths (see arrow magnitudes in Fig. 1).

The combined optical signal (multiplexed signal; MUX) is analyzed by reading out the generated photocurrent under negative applied voltage (-8 V), without and with near ultraviolet background (λBackground=390 nm) and different intensities, applied either from the front (λF) or the back (λB) sides. The device operates within the visible range using as input color channels the square wave modulated low power light supplied by near-infrared/visible (VIS/NIR) LEDs. In Fig. 2(a), the 524 nm input channel is displayed under front, back and without UV irradiation. The arrows indicate the enhancement (solid line) or quenching (dot line) of the dark signal, respectively under front and back irradiation.

0,0 0,5 1,0 1,5 2,0 2,50,0

0,2

0,4

0,6

0,8λ input=524 nmλBackground=390 nm

Dark

Back

Front

Pho

tocu

rren

t (μA

)

Time (ms)0,0 0,5 1,0 1,5 2,0 2,5

0,0

0,5

1,0

1,5

Back

Front

697 nm524 nm850 nm400 nm

Pho

tocu

rren

t (μA

)

Time (ms)

(a)

(b)

Fig. 2. (a) 524 nm input channel under front, back and without (dark) background irradiation; (b) MUX signals and under front and back λ=390 nm irradiation and different bit sequences.


35

In Fig. 2 (b), the polychromatic mixture of four different input channels (400 nm, 524 nm, 697 nm and 850 nm) under front and back 2800 µWcm-2 irradiation, is displayed. At the top, the input channels wavelengths and their bit sequences guide the eyes.

3. Bias Controlled Selector

3.1. Optical Bias Controlled Filter

The spectral sensitivity was tested through spectral response measurements [7] without and under 390 nm front and back backgrounds of variable intensities. The spectral gain (α), defined as the ratio between the signal with and without irradiation was inferred.

In Fig. 3, the spectral gain (α) is displayed under steady state irradiations. In Fig. 3 (a), the light was applied from the front (λF) and in Fig. 3 (b), the irradiation occurs from the back side (λB). The background intensity (φ) was changed between 5 µWcm-2 and 3800 µWcm-2.

Results show that, the optical gains have opposite behaviors. Under front irradiation (Fig. 3(a)) and low

flux, the gain is high in the infrared region, presents a well-defined peak at 725 nm and strongly quenches in the visible range. As the power intensity increases, the peak shifts to the visible range and can be deconvoluted into two peaks, one in the red range that slightly increases with the power intensity of the background and another in the green range that strongly increases with the intensity of the ultraviolet (UV) radiation.

In the blue range, the gain is much lower. This shows the controlled high-pass filtering properties of the device under different background intensities. Under back bias (Fig. 3(b)) the gain in the blue/violet range has a maximum near 420 nm that quickly increases with the intensity. Moreover, it strongly lowers for wavelengths higher than 450 nm, acting as a short-pass filter. Thus, back irradiation, tunes the violet/blue region of the visible spectrum whatever the flux intensity, while front irradiation, depending on the background intensity, selects the infrared or the visible spectral ranges. Here, low fluxes select the near infrared region and cuts the visible one, the reddish part of the spectrum is selected at medium fluxes, and high fluxes tune the red/green ranges with different gains.

400 450 500 550 600 650 700 750 8001

2

3

4

5

6Φ (μW/cm2)

5 20 56 100 160 250 575 1270 2600 3800

Φ

λF=390nm 3500 Hz

Gai

n (α

F)

Wavelength (nm)400 450 500 550 600 650 700 750 800

0.5

1.0

1.5

2.0

2.5

3.0Φ (μW/cm2)

5 20 55 100 160 250 575 1270 2600 3800

Φ

λB=390 nm 3500 Hz

Gai

n (α

B)

Wavelength (nm)

(a)

(b)

Fig. 3. Front (λF) and back (λB) spectral gains (αF,B) respectively, under λ=390 nm irradiations.

3.2. Nonlinear Spectral Gain

To analyze the effect of the background intensity in the input channels, several monochromatic pulsed lights separately (850 nm, 697 nm, 626 nm, 524 nm, 470 nm, 400 nm; input channels) or combined (MUX signal) illuminated the device at 12000 bps [8].

Steady state optical bias with different intensities was superimposed separately from the front and back sides and the photocurrent measured. For each individual channel the photocurrent gain under irradiation was determined. In Fig. 4, these gains are displayed as a function of the background lighting under front (Fig. 4(a)) and back (Fig. 4(b)) irradiation.

Results show that, even under transient conditions and using commercial visible and NIR LEDs, the

background side and intensity alters the signal magnitude of the input channels.

The gain depends mainly on the channel wavelength and to some extent on the lighting intensity. Even across narrow bandwidths, the photocurrent gains are quite different. This nonlinearity allows identification of the different input channels in the visible/infrared ranges.

4. Wavelength Division Multiplexer

4.1. Input Channels

Four monochromatic pulsed lights with different intensities, separately (400 nm, 470 nm, 697 nm and 850 nm; input channels) or combined (MUX signal) illuminated the device at 12000 bps.


36

Steady state 390 nm front and back optical bias with 2800 µWcm-2 intensity was superimposed separately and the photocurrent was measured. In

Fig. 5 (a), the blue and violet transient signals are presented under front and back irradiations while in Fig. 5 (b), the red and infrared signals are displayed.

1,0x101 1,0x102 1,0x103

1

2

3

4

5

470 nm 524 nm 626 nm 697 nm 850 nm

λFront=390 nm

α F

Φ (μWcm-2)

1,0x101 1,0x102 1,0x103

0,5

1,0

1,5

2,0

470 nm 524 nm 626 nm 697 nm 850 nm

λBack=390 nm

α B

Φ (μWcm-2)

(a)

(b)

Fig. 4. Front (a) and back (b) optical gains as a function of the background intensity for different input wavelengths in the VIS/NIR range.

0.0 0.5 1.0 1.5 2.0 2.50.0

0.3

0.6

Front

Back

Back

Front

λB=470 nmλV=400 nm

Phot

ocur

rent

(μΑ

)

Time (ms)0,0 0,5 1,0 1,5 2,0 2,5

0,0

0,1

0,2

Back

Front

Back

Front

λIR=850 nm λR=697 nmP

hoto

curr

ent (

μΑ)

Time (ms)

(a)

(b)

Fig. 5. Input signals under front and back 390 nm background irradiation; a) violet and blue channels. b) red and infrared channels.

In Table 1, the measured optical gains for five

different input channels are displayed.

Table 1. Optical gains under 390 nm front (αFront) and back (αBack) irradiations.

λ=400

nm λ=470

nm λ=524

nm λ=697

nm λ=850

nm αBack 11.6 1.8 0.61 0.46 0.44

αFront 0.9 1.5 3.2 4.3 3.5

Back irradiation enhances, differently, the input signals in the short wavelength range (Fig. 5(a)) while front irradiation increases them otherwise in the long wavelength range (Fig. 5(b)). This side dependent effect is used to enhance or to quench the

input signals allowing their recognition and providing the possibility for selective tuning of the visible and IR input channels.

4.2. MUX Signal

In Fig. 6, two MUX signals due to the input signals of Fig. 2(a) and Fig. 5 are displayed without (dark) and under front and back irradiation. On top, the signals used to drive the input channels are shown to guide the eyes into the on/off channel states.

Results show that, the background side alters the form of the MUX signal, enhancing or quenching different spectral ranges. In Fig. 6(a) all the on/off states are possible so, without optical bias, 24 ordered levels are detected and correspond to all the possible combinations of the on/off states.


37

0.0 0.5 1.0 1.5 2.0 2.50.0

0.3

0.6

0.9

1.2

1.5

FrontBack Dark

697 nm524 nm470 nm400 nm

Pho

tocu

rrent

(μA)

Time (ms)0.0 0.5 1.0 1.5

0.0

0.2

0.4

0.6

0.8

1.0

Back

Front

697 nm524 nm850 nm400 nm

Phot

ocur

rent

(μA

)

Time (ms)

(a)

(b)

Fig. 6. MUX signals: (a) without and under front and back λ=390 nm irradiation and different bit sequences. (b) Front and back irradiation and two channels (400 nm and 697 nm) with the same bit sequence.

Under, either front or back irradiation, each of

those four channels, by turn, are enhanced or quenched differently (Fig. (6), Table 1) resulting in an increase magnitude of red/green under front irradiation or of the blue/violet one, under back lighting. Since the gain of the four input channels is different (αF,B; Table 1) this nonlinearity allows identifying the different input channels in a large visible/infrared range. In Fig. 6 (b), both 400 nm and 697 nm channels have the same bit sequence which corresponds to only 23 ordered levels, however once the optical gains of both channels are quite different under front and back irradiation (Table 1) it is possible to identify them. Under back irradiation the MUX signal receive its main contribution from the 400 nm channel while under front irradiation it is mainly weighed buy the long wavelength channels. By comparing front and back irradiation is possible to decode the transmitted information.

Under front irradiation, near-UV radiation is absorbed at the beginning of the front diode and, due to the self-bias effect, increases the electric field at the back diode where the red/infrared incoming photons (see Fig. 1) are absorbed accordingly to their wavelengths (see Fig. 3) resulting in an increased collection. Under back irradiation the electric field decreases mainly at the back i-n interface enhancing the electric field at the front diode quenching it at the back one. This leads to an increased collection of the violet/blue input signals.

So, by switching between front to back irradiation the photonic function is modified from a long- to a short-pass filter allowing, alternately selecting the red/infrared channels or the blue and violet ones, thus, making the bridge between the visible and the infrared regions.

5. Optoelectronic Model Based on the experimental results and device

configuration a two connected phototransistors model (Fig. 7 (a)), made out of a short- and a long-pass filter

was developed [5] and upgraded to include several input channels. The ac circuit representation is displayed in Fig. 7 (b) and is supported by the complete dynamical large signal Ebers-Moll model with series resistances and capacities.

i´p np

nn ip

Q1

Q2

IB,IG

I IR, IR IG

(a)

I4

Q2

I1

I2

Q1

C2 R2 R1 (V)

C1I3

(b)

λV α1/C1

α2/C2

i1(t)

i2(t)

-1/R1C1

1/R1C2

1/R1C1

dt

dt

-1/R1C2 -1/R2C2

1/R2

i (t)λB

λG

λR,IR

v1

+

+ +

++

+

v2

v1

v2

.

.

(c)

Fig. 7. a) Two connected transistor model; b) equivalent electric circuit; c) block diagram

of the optoelectronic state model.


38

The charge stored in the space-charge layers is modelled by the capacitor C1 and C2. R1 and R2 model the dynamical resistances of the internal and back junctions under different dc bias conditions. The operation is based upon the following strategic principle: the flow of current through the resistor connecting the two transistor bases is proportional to the difference in the voltages across both capacitors (charge storage buckets). The modified electrical model developed is the key of this strategic operation principle. Two optical gate connections ascribed to the different light penetration depths across the front (Q1) and back (Q2) phototransistors were considered to allow independent blue (I1), red/infrared (I2) and green (I3, I4) channels transmission. Four square-wave current sources with different intensities are used; two of them, I1 and I2, with different frequencies to simulate the input blue and red channels and the other two, I3 and I4, with the same frequency but different intensities, to simulate the green channel due to its asymmetrical absorption across both front and back phototransistors.

In Fig. 7(c), the block diagram of the optoelectronic state model is displayed. The resistors (R1, R2) and capacitors (C1, C2) synthesize the desired filter characteristics. The input signals, λIR,R,G,B,V

model the input channels and i(t) the output signal. The amplifying elements, α1 and α2 are linear combinations of the optical gains of each impinging channel, respectively into the front and back phototransistors and account for the enhancement or quenching of the channels (Fig. 3) due to the steady state irradiation. Under front irradiation we have: α2>>α1 and under back irradiation α1>>α2. This affects the reverse photo capacitances, (α1,2/C1,2) that determine the influence of the system input on the state change.

A graphics user interface computer program was designed and programmed within the MATLAB® programming language, to ease the task of numerical simulation. This interface allows selecting model parameters, along with the plotting of bit signals and both simulated and experimental photocurrent results. To simulate the input channels we have used the individual magnitude of each input channel without background lighting (Fig. 2 and Fig. 5), and the corresponding gain at the simulated background intensity (Table 1). Fig. 8, presents results of a numerical simulation with 3000 µW/cm2 front and back λ=390 nm irradiation and the experimental outputs of Fig. 2(b) and Fig. 6(b), respectively.

0,0 0,5 1,0 1,5 2,0 2,50,0

0,5

1,0

1,5 697 nm 524 nm 850 nm 400 nm

Back Experimental Front

MU

X s

igna

l (μA

)

Time (ms)

Simulation

0.0 0.5 1.0 1.5 2.0 2.50.0

0.2

0.4

0.6

0.8

1.0 697 nm 524 nm 850 nm 400 nm

Back

Experimental

Front

MU

X s

igna

l (μA

)

Time (ms)

Simulation

(a)

(b)

Fig. 8. Numerical simulation with front and back λ=390 nm irradiation, and different channel wavelength combinations

and bit sequences. Values of R1=10 kΩ, R2=1 kΩ, C1=1000 pF,

C2=20000 pF were used during the simulation process (Fig. 7(c)). On top of the figures, the drive input LED signals guide the eyes into the different on/off states and correspondent wavelengths

A good fitting between experimental and simulated results was achieved. The plots show the ability of the presented model to simulate the sensitivity behavior of the proposed system in the visible/infrared spectral ranges. The optoelectronic model with light biasing control has proven to be a good tool to design optical filters. Furthermore, this model allows for extracting theoretical parameters by fitting the model to the measured data (internal resistors and capacitors). Under back irradiation

higher values of C2 were obtained confirming the capacitive effect of the near-UV radiation on the device that increases the charge stored in the space charge layers of the back optical gate of Q2 modelled by C2 [9].

6. Decoding Algorithm Results show that the background side changes

the shape of the MUX signal, enhancing or quenching different spectral ranges. In Fig. 8(a) all the on/off states are possible so, 24 ordered levels are detected and correspond to all possible combinations of the on/off states. Under, either front or back


39

irradiation, each of those four channels, by turn, are enhanced or quenched differently (Fig. 5, Table 1) resulting in an increased magnitude of red/green under front irradiation or of the blue/violet one, under back lighting. Since the gain of the input channels is different (αF,B; Table 1) this nonlinearity allows identifying the different input channels in a large visible/infrared range. Under front irradiation the MUX signal presents sixteen separate levels each one ascribed to one of the of the 24 possible combinations of the on/off states and pondered by their optical gains. So, by assigning each output level to a four digit binary code weighted by the optical gain of the each channel, the signal can be decoded. A transmission capability of 15 kbps was achieved.

The decoding algorithm is based on a proximity search [10]. Each time slot is translated to a vector in multidimensional space. The vector components’ are computed as a function of the sampled currents I1 and I2, where I1 and I2 are the currents measured under front and back optical bias in the respective time slot. The result is then compared with all vectors obtained from a calibration sequence. The color bits of the nearest calibration point are assigned to the time slot. An Eucledian metric is applied to measure distances. We have used this simple algorithm to perform 1-to-16 demultiplexer (DEMUX) function and to decode the multiplex signals. As proof of concept the decoding algorithm was implemented in Matlab [11] and tested using different binary sequences. In Fig. 9 a random MUX signal under front and back irradiation is displayed as well as the decoding results. A good agreement between the signals used to drive the LED’s and the decoded sequences is achieved. In all tested sequences tested the RGBV signals were correctly decoded.

0 0.2 0.4 0.6 0.8 1 1.2

x 10

-3

0

0.5

1

1.5

2

2.5

3

3.5

Time (ms)

Ph

oto

cu

rre

nt (u

A)

Front

Back

Fig. 9. DEMUX signals and decoded RGBV binary bit sequences.

The DEMUX sends the input logic signal to one of its 2n (n is the number of color channels) outputs, according to the optoelectronic demux algorithm. So, by means of optical control applied to the front or back diodes, the photonic function is modified,

respectively from a long-pass filter to pick the red/infrared channels to a short-pass filter to select the violet channel, giving a step reconfiguration of the device. The green and blue channels are selected by combining both active long- and short-pass filters into a band-pass filter. In practice, the decoding applications far outnumber those of demultiplexing. Multilayer SiC/Si optical technology can provide a smart solution to communication problem by providing a possibility of optical bypass for the transit traffic by dropping the fractional traffic that is needed at a particular point.

7. Conclusions

We experimentally and theoretically demonstrate the use of near-ultraviolet steady state illumination to increase the spectral sensitivity of a double a-SiC/Si pi’n/pin photodiode beyond the visible spectrum (400 nm-880 nm). The concept is extended to implement a 1 by 4 wavelength division multiplexer with channel separation in the visible/near infrared ranges.

Results show that, the pi´n/pin multilayered structure becomes reconfigurable under front and back irradiation, acting as data selector in the VIS/NIR ranges. The device performs WDM optoelectronic logic functions providing photonic functions such as signal amplification, filtering and switching. The opto-electrical model with light biasing control has proven to be a good tool to design optical filters in the VIS/NIR. An optoelectronic model was presented and proven to be a good tool to design optical filters in the VIS/NIR range. A decoding algorithm to decode

Acknowledgements

This work was supported by FCT (CTS multi annual funding) through the PIDDAC Program funds (UID/EEA/00066/2013) and PTDC/EEA-ELC/120539/2010. References [1]. T. Komiyama, K. Kobayashi, K. Watanabe,

T. Ohkubo, Y. Kurihara, Study of visible light communication system using RGB LED lights, in Proceedings of the IEEE SICE Annual Conference (SICE’11), 2011, pp. 1926-1928.

[2]. S. S. Djordjevic, et al., Fully Reconfigurable Silicon Photonic Lattice Filters With Four Cascaded Unit Cells, IEEE Photonics Technology Letters, Vol. 23, No. 1, 1 January 2011, pp. 41-44.

[3]. P. P. Yupapin, P. Chunpang, An Experimental Investigation of the Optical Switching Characteristics Using Optical Sagnac Interferometer Incorporating One and Two Resonators, Optics & Laser Technology, Vol. 40, No. 2, 2008, pp. 273-277.

[4]. S. Ibrahim, et al., Fully Reconfigurable Silicon

R [0111010110010100]

V [0100010100101111] B [0101110011001010] G [0101001001111100]


40

Photonic Lattice Filters with Four Cascaded Unit Cells, in Proceedings of the Optical Fibre Communications Conference (OSA/OFC/NFOEC), San Diego, 21 Mar. 2010, paper OWJ5.

[5]. M. Vieira, P. Louro, M. Fernandes, M. A. Vieira, A. Fantoni, J. Costa, Three Transducers Embedded into One Single SiC Photodetector: LSP Direct Image Sensor, Optical Amplifier and Demux Device, in Advances in Photodiodes, Chap. 19, InTech, 2011, pp. 403-425.

[6]. M. A. Vieira, P. Louro, M. Vieira, A. Fantoni, A. Steiger-Garção, Light-activated amplification in Si-C tandem devices: A capacitive active filter model, IEEE Sensor Jornal, Vol. 12, No. 6, 2012, pp. 1755-1762.

[7]. M. A. Vieira, M. Vieira, P. Louro, V. Silva, A. S. Garção, Photodetector with integrated optical thin film filters, Journal of Physics: Conference Series, Vol. 421, No. 1, 25 March 2013, 012011.

[8]. M. Vieira, M. A. Vieira, I. Rodrigues, V. Silva, P. Louro, Tuning optical a-SiC/a-Si active filters by

UV bias light in the visible and infrared spectral ranges, Phys. Status Solidi, Vol. 11, 2014, pp. 1-4.

[9]. M. Vieira, M. A. Vieira, I. Rodrigues, V. Silva, P. Louro, A. Fantoni, UV Irradiation to Increase the Spectral Sensitivity of a-SiC:H pi’n/pin Photodiode Beyond the Visible Spectrum Light, in Proceedings of the 6th International Conference on Sensor Devices, Technologies and Applications (SENSORDEVICES’15), Venize, Italy, 2015, pp. 44-49.

[10]. M. A. Vieira, M. Vieira, P. Louro, V. Silva, J. Costa, A. Fantoni, SiC Multilayer Structures as Light Controlled Photonic Active Filters, Plasmonics, Vol. 8, No. 1, 2013, pp. 63-70.

[11]. M. A. Vieira, M. Vieira, J. Costa, P. Louro, M. Fernandes, A. Fantoni, Double pin Photodiodes with two Optical Gate Connections for Light Triggering: A capacitive two-phototransistor model, Sensors & Transducers, Vol. 10, Special Issue, February 2011, pp. 96-120.

___________________


http://www.sensorsportal.com/HTML/E-SHOP/PRODUCTS_4/UFDC_1.htm






41



Sub-nanosecond Gating of Large CMOS Imagers

Octavian Maciu, Wilfried Uhring, Jean-Pierre Le Normand, Jean-Baptiste Kammerer, Foudil Dadouche, Norbert Dumas

ICube, UMR 7357, University of Strasbourg and CNRS 23, rue du Loess BP 20, F-67037 Strasbourg Cedex 2, France

Tel.: +33 (0)3 88 10 68 27, fax: +33 (0) 3 88 10 65 48 E-mail: [email protected]

Received: 31 August 2015 /Accepted: 5 October 2015 /Published: 30 October 2015 Abstract: Ultra-fast gating of large array imagers can be quite challenging to implement due to the distributed RC (Resistance Capacitance) nature of the metal wires used in all ICs (Integrated Circuits) for electrical connections. For the transmission of a signal across a long path, the metal line reduces the electrical bandwidth and adds a delay. The behavior of these lines has been modeled and a new solution is presented to circumvent these limitations. In this paper, we present an edge-based approach to the gating circuitry that allows sub nanosecond gating with a very low skew across the whole imager. Simulation data shows that our solution is an efficient way of reducing the effect of the distributed RC line delay with a small penalty on surface area and consumption. Copyright © 2015 IFSA Publishing, S. L. Keywords: SPAD, Ultra-fast gating, Edge-triggered, Fast pulse, Skewless. 1. Introduction

Over the last decade, ultra-fast imaging has been a booming field with many considerable breakthroughs. A key advancement to this technology has been the ability to design image sensors with a sub-nanosecond temporal resolution. These imagers could be configured in either single-shot mode in [1] and [2], or can be repetitive with photodiode or photogate in [3] and [4] or based on Single-Photon Avalanche Diode (SPAD) arrays rather than classical photodiodes in [5] and [6]. In this case, the temporal resolution can be even lower. Currently, integrated streak cameras operate at the fastest frequency, therefore also requiring robust acquisition signals. These fast signals are used in order to have a sub-nanosecond shutter speed. Depending on the design, the temporal resolution of an integrated streak camera can vary from a few hundred picoseconds to several

nanoseconds. A 1D or 2D approach of integrated streak camera solution can be found in [3], where a delay generator based on the propagation delay of logic gates is used for sub-nanosecond shuttering. The delay could be customized using current starved inverters. In [1], [4] and [7], we see the use of edge-based control signals for fast gating.

Similar to a clock signal in a synchronous sequential circuit, the gating signals are distributed to the entire imager array. Due to the distributed RC (Resistance capacitance) line delay, the integrity of the signal is compromised along the row of pixels. This would degrade the performance as a pixel at the end of the array would not perceive the same signal as another near the beginning of the array. Moreover, it would be completely unusable for ultra-fast gating purposes as the rise time and fall time would be far greater than the gating time. This is especially true for ultra-fast image sensors as a signal commutes at the




42

order of 100 picoseconds. It behooves us then to ensure that the signal is identical and useable for every single pixel in the array. Therefore, a more reliable approach would be to reduce the dependency on the pulse width and have an edge triggered gating. This solution was implemented in a 0.18 μm CMOS (Complementary Metal Oxide Semiconductor) process. These approaches have been applied for smaller dimension image sensors and it is therefore interesting to propose a solution that is extendable to full resolution imagers.

2. Theoretical Approach

As mentioned before, in an imager array, the gating signal is being driven at the beginning of a row. Therefore, the signal perceived at the end of the row will no longer be a clean square pulse. Hence, it is

interesting to study what the maximum row length for which the pulse will still be useable. Firstly, we can model this problem by looking at a single distributed RC model. Fig. 1 shows the equivalent distributed line representation of a row containing N pixels where RD and CD are the resistance and the capacitance per length unit of the metal line respectively. The localized pixel input capacitances Cpixel act like a distributed capacitance according to the pixel pitch

pixelDC where,

(1)

The total distributed capacitance totalDC is thus given

by the sum of the equivalent distributed input pixel capacitance and the line distributed capacitance,

total pixelD D DC C C= + .

Fig. 1. Distributed RC line representation of a sensor row including N Pixels. The open-circuited Laplace transfer function from the beginning of the line V1 to the end of the line V2 can be written as [8, 9]:

, (2)

where s is the Laplace variable, and l is the length of the row. The behavior of the line can be approximated by the simplified circuit model depicted in Fig. 2 [8].

C/2R

V1 V2C/2

Fig. 2. Simplified circuit of entire line.

Where DR R l= ⋅ and totalDC C l= ⋅ are the total

resistance and capacitance of the line. The unit step response gives us a clear indication that while not perfect, this approximation is sufficient to illustrate the distributed RC delay problem [8]. The distributed RC effect acrosss the line therefore affects the driver rise

time. The fastest rise time of V2 can thus be obtained by the following expression:

(3)

Equation (3) states that the rising time increases with the square of the row length l and then can dramatically reach a value that makes it impossible to transport a nanosecond pulse across a large array sensor. 2.1. Parameters Extraction The parameters RD and CD can generally be found in the design kit documentation. Otherwise, the unit line and surface capacitance can be obtained by an analog extracted view simulation. Based on variants (square and rectangular) of the diagram in Fig. 3, theses parameters can be deduced with two different sets of equations (4).

, (4)

where w and l are the respectively the width and the length of the line, Cline is the extracted line capacitance, Cedge is the edge capacitance, given in

Pixel PitchPixel

pixel

D

CC =

RDCD

RDCD

RDCD

Pixel 1

Cpixel

RDCD

Pixel 2

Cpixel

RDCD

RDCD

Pixel N

Cpixe l

V1 V2

2

1

1( )

coshtotalD D

VH s

V s R C l= =

⋅ ⋅

( ) 20.35pixelR D D DT R C C lπ= ⋅ + ⋅

( )2line edge surfaceC C C w ll w= ⋅ + ⋅ ⋅+


43

F/m, and Csurface the surface capacitance of the line, given in F/m².

l

w

Fig. 3. Diagram of layout model variants used

(line and rectangle).

Then, the distributed capacitance for a long line can be approximated by:

, (5)

where w is the width of the considered line. This capacitance did not take account of the parasitic capacitance added by the neighboring wires. This last cannot be easily determined a priori. A first rule of thumbs should be to multiply about 2 or 3 times this capacitance. A post extracted simulation, see section 6, is mandatory to obtain an accurate value.

The resistivity can be extracted in the same way or it can be obtained by process data, such as the metal resistivity ρ and thickness T. Therefore, the resistance per length unit RD of a line of width w is given by:

(6)

3. Ultra-Fast Gating for Large Array Sensor

Fig. 4 shows the shape of a 1 ns pulse propagating in a sensor row at the beginning of the line, where the pulse is applied, in the middle and at the end of the line. If we assume the classic case where the logic of the pixel reshapes the pulse with a threshold voltage of half the power supply (solid line in Fig. 4), we clearly see that after a certain length, the conventional gating is no longer adequate.

Furthermore, it is interesting to create a model in order to anticipate this signal distortion with respect to the length of the sensor. Based on the distributed model and a voltage threshold of half the power supply, Fig. 5 compares the Full-Width at Half Maximum (FWHM) ratio of the in-pixel reshaped pulse versus the original pulse according to the original pulse FWHM to the rise time. The effective FWHM inside the pixel decreases as soon as the original FWHM pulse is below the rising time and is reduced to zero for a ratio of 30 %.

Fig. 4. Illustration of lost signal integrity across a row of pixels.

Fig. 5. Final to original pulse width ratio versus the original pulse width to rising time ratio.

Using (3), the relationship between the increasing

rise time and the length of the array is shown in Fig. 6 for a metal line, level 2, of a typical 0.18 µm CMOS process with a resistivity ρ of 2.65⋅10-8 Ω.m, a thickness T of 425 nm, a pixel pitch of 35 µm, an input pixel capacitance of Cpixel of 2 fF, a surface capacitance Csurface of 0.015 fF/µm², an edge capacitance Cedge of 0.032 fF/µm and a width w of 3 µm. Sub-nanosecond rise times are not allowed for sensor dimensions above 16 mm.

A way to enhance the line bandwidth is to increase the line width w because it reduces the resistance of the line. Fig. 7 shows the calculated rise time in the middle of the previous line according to the metal width w. Increasing the width of the metal track above 3 μm has minimal impact on rise time and thus becomes irrelevant.

2D edge surfaceC C C w= ⋅ + ⋅

DRw T

ρ=

⋅


44

Fig. 6. Rise Time of a signal seen across a pixel row through a metal line of up to 20 mm and a pixel pitch of

35 µm with the following parameters: ρ = 2.65.10-8 Ω.m, T = 425 nm, Cpixel = 2 fF, Csurface = 0.015 fF/µm²,

Cedge = 0.032 fF/µm and w = 3 µm.

Fig. 7. Rise Time with respect to metal track width at midway along a row of pixels.

Therefore, generating 200 ps FWHM gating within large sensor arrays over 10 mm is impossible with pulse propagation techniques. To ensure signal integrity, the use of edge-sensitive logic is mandatory.

4. Proposed Design of Edge-Triggered Gating

4.1. Edge-based Circuit for Ultra-fast Gating

In SPAD array image sensors, for ultra-fast gating mode there are three critical signals needed for optimal operation: Quench, Reset and Gate. Moreover, SPADs could be gated immediately after the reset to avoid the detection of photons arriving before the investigation time slot. While Quench is a much slower signal, the Reset and Gate signals can be sub-nanosecond pulses.

Having shown clear limitations for ultra-fast gating in large SPAD array image sensors, a different

approach is needed. This approach consists of an additional signal GATE_DELAY (cf. Fig. 8). GATE_DELAY is delayed thus creating the effective pulse of width d seen above through a logic gate. It can therefore be extended to the Reset signal of the SPAD as seen in Fig. 9. Hence, we are able to achieve gating in the range of 100 ps up to 1 ns while maintaining signal integrity throughout the SPAD array. Quenching time is much longer than the reset and gating signals and do not require an ultrafast signal generation.

Quench Reset GatingRequired Timing

d

GATE

GATE_DELAY

Fig. 8. Timing diagram of typical SPAD quench, reset and gating with delayed (d) edge-based signals below.

SPAD_RSTSPAD_RST_DELAY

GATEGATE_DELAY

RQ

uenc

h

VDD_SPAD

SPAD

V_ANODE (-)

Quench

Vref-+

SPAD OUT

Fig. 9. SPAD Edge-triggered circuit with active quench, reset and gating.

4.2. Edge-based Driver for Ultra-fast Gating

In Fig. 10, it can be seen that the GATE signal is delayed to obtain a GATE_DELAY signal. In order to generate the delay, we used a delay generator to drive the signals. The delay generator is constructed using a series of current starved double inverters as seen in [10].

The delay of both signals (cf. Fig. 10) can be determined by a modifiable control voltage (Vctrl). We are able to modify the pulse width by splitting the signal in two branches with one providing minimal propagation delay and the other being delayed with respect to the other. Both signals are then fed into an in-pixel logic gate to obtain the desired pulse derived from the delay between the two signals.


45

Fig. 10. Delay generator block made up of double current starved inverters used to create

GATE and GATE_DELAY signals.

4.3. Simulation of Edge-based Technique

The presented solution can be demonstrated and validated through a Cadence simulation with two buffered lines of 20 mm long and a width of 3 µm based on the model presented in this paper in order to obtain a 300 ps pulse through a logic gate.

In Fig. 11, the FWHMs of the two pulses at the beginning (solid line) and at the end (dashed line) of a 20 mm length line are both equal to 300 ps. Our solution performs well in a Cadence simulation using the model presented in Fig. 3 and the defining equations shown previously. Consequently, thanks to the edge pulse technique generation, we can ensure that all the pixels of the sensor exhibit the same temporal gate width. However, a non-negligible skew is still present.

Fig. 11. Simulation of a 300 ps pulse created using the edge-based technique (line: 20 mm length, 3 µm width). Signals at the beginning of the line (solid line) and at the

end (dotted line).

5. Analysis and Elimination of Skew

For large array ultra-fast image sensors, ensuring signal integrity across the array is only half of the challenge. As can be seen in Fig. 6, a 20 mm sensor

will introduce a skew of over 1.5 ns. Hence, the edge based technique has to be improved for large dimensions if nanosecond or sub-nanosecond gating are targeted.

In order to present a more robust approach on ultra-fast gating, we proceed to eliminate the large skew present across the array mentioned. This can be done by introducing two skewed inverters on each branch (GATE and GATE_DELAY) before the logic gate. The threshold voltage of each inverter can be modified by altering the beta ratio effects inside each inverter. By compensating the voltage drop at a given time along the sensor lines thanks to the inverted threshold voltage, the skew across the entire sensor array can be compensated.

Fig. 12 shows the simulation of needed VTH variation across a 20 mm image sensor row using above mentioned parameters. The threshold voltage at the beginning of the sensor is arbitrary set at 1.1 V (above VDD/2 where VDD = 1.8 V) in order to maintain threshold levels across the array well above the |VT| of each transistor. For a required voltage VTH, the inverter should be adjusted by modifying the geometry of the PMOS and NMOS transistors, thus changing the beta ratio.

Fig. 12. Simulation of needed VTH variation across a 20 mm image sensor row using above

mentioned parameters.

This can be obtained by solving the following equation:

(7)

In this case, both the NMOS and the PMOS transistor are in the saturated operation region, thus the currents are given by:

(8)

GATE

GATE_DELAY

VMIN_DELAY

GATE VMOD

VctrlIN

VctrlIN

, for 2

in th

n p

out

V VI I

V VDD

==

=

( )

( )

2

2

11

2 2

11

2 2

N

N

n DDn n ox TH T

n

p DDp p ox DD TH T

p

W VI C V V

L

W VI C V V V

L

μ λ

μ λ

= − +

= − − +


46

Replacing In and Ip in equation (7) leads to:

( ) ( )22 pnn TH T p DD TH Tp

n p

WWV V V V V

L Lμ μ− = − −

(9)

Hence, we obtain the ratio ρ:

(10)

Furthermore, in order to keep the best fill factor, the inverter area A given by:

(11)

has to be kept as low as possible. For a threshold voltage Vth above VDD/2, the PMOS transistor has to be more conductive than the NMOS one, thus the ratio ρ should be greater than one. For this, we can consider using Wn=Wmin and Lp=Lmin as the best choice. The inverter area becomes:

2 2 2 2

min min min min

1n n

n n

W LA L W L W

L Wρ ρα

α= + = +

(12)

Minimizing A according to the variable α leads to best sizes of the MOS transistors for the smallest use of surface area:

min min

minmin

and pn

n p

WW W W

L L LL

ρ

ρ

⋅= =

⋅ (13)

In a similar way, the optimal sizes of the transistors for a threshold voltage VTH under VDD/2 can be computed as:

min min

min min

and pn

n p

WW W W

L L L L

ρρ

⋅= =

⋅ (14)

In both case, the area of each threshold adjusted inverter is given by:

min min 4A W L ρ= (15)

The more ratio ρ is distant from 1, the higher the surface area.

It would be ideal to designate an area of a pixel for the placement of automatically generated skewed inverters based on their position in the sensor row.

By employing this technique, it can be seen in Fig. 13 that the pulse at the beginning of the sensor row and the pulse at the end are synchronized and the

skew seen in Fig. 11 is eliminated. The adjusted threshold voltage are indicated with doted and dashed line. This ensure that a sub-nanosecond gated operation can be proceeded uniformly across the whole sensor.

Fig. 13. Simulation of skewless pulse. Solid line: signal at the beginning of the sensor, dotted line: signal

at the end of a 20 mm sensor. 5.1. Variable Slew Rate Driver

To be able to apply the previously described method to eliminate skew across the sensor, a custom driver is used. This driver should have a variable slew rate to further compensate skew and ensure that the skew eliminating current starved inverters are effective. This is due to the fact that when operating at sub-nanosecond time intervals, the skew elimination is only possible for a relatively slow driving signal as can be seen in Fig. 13. Otherwise, it would not be possible to compensate the threshold voltage across the sensor array as the signal’s rise time would be too fast. Furthermore, it is desirable to be able to control the slew rate of the signal being driven across the sensor array. This can be done using a voltage controlled current starved inverter element as seen in Fig. 14.

V Slew Rate

VDD

IN OUT

Fig. 14. Voltage-controlled slew driver made up of a current starved inverter cell.

( )( )

2

2

p

p TH Tn

pn DD TH Tp

n

W

L V V

W V V VL

μρ

μ−

= =− −

2 2n n

p p n n p n

n n

W LA W L L W L W

L Wρ= + = +


47

As only the rise time of the edge signals is used to generate the fast pulse generation, the current starved inverter is degenerated only on the PMOS branch .A symmetrical slow driver would create significant limitations in operating frequency due to the slow fall time. Hence it is best to design an asymmetrical driver with a slow rise time and fast fall time. This can be done by altering the geometry of the NMOS transistor in Fig. 14. As can be seen in Fig. 15, the rise time is just over 9 ns and the fall time is three times faster at around 3 ns for a maximum period of 20 ns (50 MHz).

Fig. 15. Variable slew rate driver response with a period of 20 ns.

6. Application to an Ultra-fast Pixel

This new concept has been applied to a new chip

designed to operate with a sub-nanosecond gating

taking into account the aforementioned problems and solutions that arise during the design process. This new chip has a total row length of approximately 13 mm split into 10 sub-row. Each sub-row consists of 33 pixels (1.3 mm).

In Fig. 16, we have shown the layout for our SPAD gating system and pixel. For this application, there are three critical signals, therefore six lines needed for edge-based pulse propagation (seen in cyan and red). It can be seen that while they have a considerable impact on the pixel fill factor, it remains reasonable for pixel pitch of 35 µm.

6.1. Post Extracted Parameters (Additive Parasitic Capacitances)

For a more accurate picture of total parasitic

capacitance across a row of pixels, a post extracted simulation is needed. This is due to the fact that there are more Cedge parasitic capacitances in parallel than originally accounted for in our theoretical model. In the design presented in Fig. 16, these extra parasitic elements triple the total parasitic capacitance for a line. Therefore, the impact of neighbouring parasitic elements inside the pixel are also very significant. Thus, the adjustment of the threshold voltage of the skew inverter has to be done after a post extracted simulation. Even if the pixel capacitance remains low compare to the line capacitance, altering the inverter’s transistor geometry will have a little impact on the distributed pixel capacitance and consequently, the post extracted simulation has to be reiterate. A good rule of thumbs for the first iteration is to considered that the distributed capacitance is about 3 times the single line capacitance described in section 2.

Fig. 16. Layout of SPAD pixel using ultra-fast technique.


48

6.2. Monte-Carlo Simulations

After performing a Monte-Carlo simulation on the possible dispersion existing between the two most spaced pulses within a single line (beginning and end of line), we obtain a sigma of only 18.42 ps rms. However, it is important to take into account the possible dispersion found between lines, i.e. across the

entire sensor. By doing a similar simulation, we obtain a sigma of 37.32 ps rms, i.e. a mismatch of less than 90 ps FWHM. This result is illustrated in the histogram in Fig. 17.

In order to circumvent the inevitable mismatch, it is conceivable to modify (post-fab) the supply voltages of the inverters added in order to fine tune the additional skew.

Fig. 17. Interline Monte Carlo simulation showing possible dispersion due to fabrication mismatches.

7. Conclusion

In the increasingly relevant field of ultra-fast imaging, gating signals play a key role in assuring optimal operation. When designing large resolution sensors, acquisition pulses must be uniformly distributed throughout the entire array. Due to the delay effect of a distributed RC line, this becomes troublesome. For this, a conventional driver at the beginning of the line is no longer adequate and a more robust solution needs to be implemented. This new approach is independent to pulse-based gating and instead uses an edge-triggered gating. Moreover, the designer could place such edge-based drivers on either side of the sensor to obtain a near-perfect gating signal distribution. While this technique is still hindered by signal skew, it can be eliminated using the method presented in this paper. As mentioned earlier, these effects can as much as triple the parasitic effects seen by one line. Therefore, our approach is crucial in ensuring the effective transmission of a sub-nanosecond pulse across a large sensor array. The maximal pixel operating frequency can be in the range of 50 MHz and the gating operation well below 1 nanosecond. The process mismatch show skew spread of about 90 ps FWHM on the entire sensor.

Acknowledgements We thank the French agency “Agence Nationale

de la Recherche” for financial support under the grant ANR-14-CE26-0024-01.

References [1]. F. Morel, J.-P. Le Normand, C.-V. Zint, W. Uhring, Y.

Hu, and D. Mathiot, A new spatiotemporal CMOS imager with analog accumulation capability for nanosecond low-power pulse detections, Sensors Journal, Vol. 6, No. 5, 2006, pp. 1200-1208.

[2]. M. Zlatanski and W. Uhring, Streak-mode optical sensor in standard BiCMOS technology, in Proceedings of the IEEE Sensors, Limerick, Ireland, 2011, pp. 1604-1607.

[3]. M. Zlatanski, W. Uhring, J. Le Normand, C. Zint, and D. Mathiot, Streak camera in standard (Bi)CMOS (bipolar complementary metaloxide-semiconductor) technology, Measurement Science and Technology, Vol. 21, No. 11, 2010, p. 115203.

[4]. Keita Yasutomi, SangMan Han, Min-Woong Seo, Taishi Takasawa, Keiichiro Kagawan, Shoji Kawahito, A time-resolved image sensor for tubeless streak cameras, Proc. SPIE, 9022, 2014, 902202.


49

[5]. Maruyama, Y., Blacksberg, J., Charbon, E., A 1024 x 8, 700-ps Time-Gated SPAD Line Sensor for Planetary Surface Exploration With Laser Raman Spectroscopy and LIBS, IEEE Journal of Solid-State Circuits, Vol. 49, No. 1, 2014, pp. 179-189.

[6]. L. Pancheri, N. Massari, F. Borghetti and D. Stoppa, A 32x32SPAD pixel array with nanosecond gating and analog readout, in Proceedings of the Int. Image Sensor Workshop, 2011, R40.

[7]. Octavian Maciu, Wilfried Uhring, Jean-Pierre Le Normand, Jean-Baptiste Kammerer and Foudil Dadouche, Edge-Based Technique for Ultra-Fast Gating of Large Array Imagers, in Proceedings of the 9th International Conference on Sensor Technologies and Applications (SENSORCOMM ’15), Venice, Italy, 23-28 August 2015, pp. 24-28.

[8]. V. B. Rao, Delay Analysis of the Distributed RC Line, in Proceedings of the 32nd Conference on Design Automation (DAC '95), 1995, pp. 370-375.

[9]. Ogata, M., Okabe, Y., Nishi, T., Simple RC models of distributed RC lines in consideration with the delay time, in Proceedings of the International Symposium on Circuits and Systems (ISCAS '04), Vol. 4, May 2004, pp. 23-26.

[10]. Zlatanski, M. Uhring, W. Le Normand, and J.-P. Zint, V., A new high-resolution Time-to-Digital Converter concept based on a 128 stage 0.35 µm CMOS delay generator, in Proceedings of the Joint IEEE North-East Workshop on Circuits and Systems and TAISA Conference (NEWCAS-TAISA '09), 2009, pp. 1-4.

___________________ 2015 Copyright ©, International Frequency Sensor Association (IFSA) Publishing, S. L. All rights reserved. (http://www.sensorsportal.com)

http://www.sensorsportal.com/HTML/BOOKSTORE/Human_Eye.htm







50



Superpixel Compressive Sensing Recovery of Spectral Images Sensed by Multi-patterned Focal Plane

Array Detectors

1 Yuri H. MEJIA, 2 Fernando A. ROJAS, 2 Henry ARGUELLO 1 Department of Electrical Engineering, Universidad Industrial de Santander, Colombia

2 Department of Systems Engineering and Computer Science, Universidad Industrial de Santander, Colombia

Tel.: +57 7 6344000 E-mail: [email protected], [email protected], [email protected]

Received: 31 August 2015 /Accepted: 5 October 2015 /Published: 30 October 2015 Abstract: Conventional spectral imaging systems capture spectral and spatial information from a scene to produce a spectral data cube by scanning procedures. Photolithography technology development enables the production of complex filters by combining patterning techniques with optical coatings. These filters can be directly deposited onto detector arrays in order to measure spectral information with a unique snapshot. Nevertheless, recovering the spectral image with traditional methods following a demosaicing approach is impractical. State-of-the-art establishes multispectral demosaicing for recovering images with a specific spatio-spectral resolution depending on the number of pixels in the detector and the filter mosaic. Recently compressive sensing technique has been developed that allows recovering signals with few measurements than the traditional methods by using the sparse representation of the underlying signal. The selection of superpixels in the multi-patterned focal plane array detectors to calculate the spectral response of a single pixel in the reconstructed spectral images could improve the reconstruction, based on exploiting the sparse representation of the spectral images. This paper presents a model for spectral images recovering from superpixels formed with multi-patterned focal plane array detectors measurements using the concept of compressive sensing. This model selects subsets of the superpixels measurements following a downsampling matrix operation, therefore a reconstruction model is formulated by directly reconstruct a spectral image with the spectral resolution given by the number of filters. The superpixel size selection leads to a variable recovered spatial resolution preserving the filters spectral resolution. Multi-patterned focal plane array detectors measurements for real spectral images are simulated in order to verify the effectiveness of the recovery model. An ensemble of random dichroic and band pass filters is used. The superpixel compressive sensing reconstruction approach and the demosaicing scheme reconstruction are compared. Copyright © 2015 IFSA Publishing, S. L. Keywords: Compressive sensing, Spectral images, Multi-patterned focal plane array detectors, Superpixel. 1. Introduction

Spectral images applications cover ocean

research, food safety, geology, and medical demands.

For instance, the characterization of phytoplankton in the ocean [1], quality evaluation in the area of food safety [2], plant stress assessment [3], characterization of different bacterial colonies [4],




51

disease diagnosis, and image-guided surgery [5]. In some spectral imagers, the scene is beamsplit

into the wavelength components for example using a prism assembly, and each of these images is captured in a separate detector array. In this method the sensing devices have significant size and weight disadvantages [6]. One of the most intuitive multispectral scanning techniques is the tunable filter, where a complete spectral image is produced after a sequence of exposures by capturing an image of one spectral band at time. For instance, the spectral image can be sensed by using a filter wheel where some optical filters are installed in a rotatory mechanical structure [7]. Most of the methods are related to scanning operations where multiple exposures are used causing motion artifacts.

On the contrary, some techniques use multi-patterned filter arrays detectors and collect multiple wavelength spectra from a single snapshot [8]. Nowadays, optical coatings technologies have been miniaturized and optimized such that the creation of multi-patterned arrays of different optical filters, with traditional design and manufacturing methods, is allowed [6]. The optical coatings production methodology combines modern optical thin film deposition techniques with microlithographic procedures. This process enables micron-scale precision patterning of optical thin film dichroic coatings on a single substrate. A dichroic filter is an accurate color filter used to selectively pass light of a small range of wavelengths while reflecting other wavelengths.

For example, Miao, et al. [9] generate a multi-patterned filter arrays following a binary tree-based method, which starts from a checkerboard pattern. By recursively separating the original checkerboard, the algorithm generates the multi-patterned filter arrays given the number of spectral bands and the probability of appearance of each band. Then, they design a demosaicing algorithm based on the same binary tree. Brauers and Aach [10] propose a multi-patterned filter arrays that consists of color filter blocks of the size 3×2 pixels, this configuration allows to use a fast bilinear interpolation with a reconstruction up to 6 spectral bands. Monno, et al. [11] propose a five-band multi-patterned filter arrays. In the pattern, the green-like channel is distributed as in the Bayer color filter array (CFA), and other channels are arranged following a binary-tree approach. For demosaicing an adaptive kernel can be estimated directly from the raw data. Common to these systems is that the multi-patterned filter arrays design is application and number of bands dependent, which reach at most 6.

On the other hand, Compressive Sensing (CS) has emerged as a rising research area that allows the acquisition of signals at sampling rates below the Nyquist-criterion. In CS traditional sampling is substituted by measurements of random projections of the signal. The signals are then reconstructed by solving an l1 and l2 minimization problem in a basis where admits sparse representations. CS exploits the

fact that hyperspectral images can be sparse in some basis representation.

Mathematically, a multispectral image LNN ××ℜ∈F in its vector representation Mℜ∈f with

M=N2L, can be expressed as f=Ψθ, where θ is the coefficients sequence of S elements that represents f, with S<<M, and Ψ is a representation basis. Here, N×N represents the spatial dimensions, and L the number of spectral bands in the data cube. Compressive sensing allows f recovering from m random projections when m ≥ S log (M) << M.

Assuming that the multi-patterned filter arrays detectors perform a linear measurement process that calculate m<<M internal products between f and a

collection of vectors Hij, as ,, jiy Hf= then

y=Hf, where the set of yi projections forms the vector y of m elements, H is the measurement matrix with dimensions m×M, with Hj

T rows, and f is the original signal of size M. For recovering f from y, there exist infinite solutions due to the size of y is much less than the size of f.

Following the sparse representation of the signal and the multi-patterned filter arrays detectors measurements can be expressed as y=Hf =HΨθ=Aθ, where Mm×ℜ∈= HΨA is the sensing matrix. This underdetermined equation system can be solved if it is satisfied that the measurement matrix H is incoherent with the sparse transformation Ψ. It is possible to exploit the capabilities of multi-patterned filter selecting measurements subsets to form superpixels that have spectral information of a single reconstructed pixel. That is, spectral information of a single pixel can be reconstructed based on superpixel measurements. The superpixel size selection leads to a variable reconstructed spatial resolution preserving the filters spectral resolution, reconstructing a spatial decimated data cube. This information can be used in applications requiring higher spectral than spatial image quality, also for a quick view of the scene, for example for purposes of transmission and communication applications, because increasing superpixel side reduces the time of data cube recovery.

This paper presents a model for spectral images reconstruction from superpixels formed with multi-patterned filter arrays detectors measurements using the principle of compressive sensing. This model selects subsets of the superpixels measurements following a downsampling matrix operation, therefore a reconstruction model is formulated by directly reconstruct a spectral image of variable spatial resolution. The maximum spatial resolution is limited by the detector resolution. The number of different filters limits the spectral resolution.

The data cube is then reconstructed as

),ΨH Ψ~

12θ

θyargmin(f θτ+−= S where y is the

measurement selection, HS is the measurement

matrix defined as H,])D...()D[(H TTqTS

10 2 −= where


52

D is the th downsampling matrix used for measurement selection, θ is an S-sparse representation of a low resolution version of f on the basis Ψ, and τ is a regularization constant.

The rest of this paper is organized as follows. Section 2 describes the mathematical model of the spectral image acquisition system using multi-patterned filter arrays detectors. Section 3 describes the traditional demosaicing process. Section 4 addresses the superpixel CS reconstruction approach. Section 5 describes the mathematical model of the dichroic and band pass filters. Section 6 shows the simulation results. The conclusion closes the article.

2. Spectral Image Acquisition by Multi-patterned Focal Plane Array Detectors Fig. 1 shows the functions of the sensor-

multispectral filter array system following the physical sensing phenomena for L=6 spectral bands and focusing in the jth-slice.

Fig. 1. Sensing phenomena representation of the multi-patterned focal plane array detectors. The jth slice of the data cube is coded by a row of the multi-patterned filters

and then integrated onto the detectors.

For purposes of illustration the Fig. 1 shows two optical elements separately, but the device is a focal plane array covered by multi-patterned filters. First, the multi-patterned filters T(x, y, λ) modulates the spatial-spectral data cube f0(x, y, λ), resulting in the coded field f1(x, y, λ), where (x, y) are the spatial coordinates, and λ is the wavelength. Then the coded density impacts on the sensor. Eq. (1) represents the coded density integrated into the detector.

,'')','(),','(),','(),,( 12 −−= dydxyyxxhyxfyxTyxf λλλ (1)

where ),','( λyxT is the transmission function representing the multi-patterned filters, and

)','( yyxxh −− is the optical response of the system. Each pixel in the sensor is a discretized

measurement. The source f0(x, y, λ) can be written in discrete form as k j, i,F where i and j index the spatial

coordinates, and k determines the kth spectral plane.

Let Ti,j,k ∈ 0,1 be the discretization of the multi-patterned filter arrays.

Then, the discretized multi-patterned filter arrays detectors measurements can be expressed as

−

=

+=1

0

,,,,, ,L

k

jikjikjiji, TFY ω (2)

where Yi,j is the intensity at the (i, j)th position of the detector, i, j=0, 1, …, N-1, and the dimensions of the detector are N×N. F is an N×N×L spectral data cube, and ωi, j is the white noise of the sensing system.

The measurements Yi,j in (2) can be written in matrix notation as

y=Hf+ω, (3)

where y is the N2-long vector representation of Yi,j, f=vect([f0, …, fL-1]) is the vector representation of the data cube F where fk is the vectorization of the kth spectral band.

The output y in (3) can be extended as

[ ] ,)(diag)(diag

1

1

0

10 ω+

=

−

−

L

L

f

ff

y

H

tt (4)

where kt is the vectorization of the kth multi-patterned

filter arrays spectral plane, diag(tk) is an N2×N2 diagonal matrix whose entries are tk, more

specifically ( ) kNNiiNiik T ,/,/ −=t

for i=0, …, N2-1. Fig. 3 depicts a random multi-patterned filter arrays based matrix H for N=6, and L=4. Colored squares represent unblocking light elements related to a specific wavelength.

N2 (1st band) N2 (2nd band) N2 (3rd band) N2 (4th band)

diag(t0) diag(t1) diag(t2) diag(t3)

N2

Block Pass λ0 Pass λ1 Pass λ2 Pass λ3

Fig. 2. The matrix H is shown for N=6, and L=4. Colored squares represent unblocking light elements related to a

specific wavelength.

3. Traditional Demosaicing

Given the set of measurements y a traditional demosaicing algorithm estimates for each reconstructed pixel the intensities for all wavelength components. Common approach minimizes the linear mean square error between the measurements and the vector estimation multiplied by the sensing matrix, that is


53

,minarg~

2f-yf

fH= (5)

A closed-form solution to (5) is given by

( ) y,yf +−== HHHH TT 1~

(6)

where H+ is known as the pseudoinverse of H, and HT is its transpose. For comparison purpose, this approach is implemented.

4. Recovery by Superpixel CS Approach In this model, the superpixel definition is based on

the assumption that q×q neighboring pixels, in an N×N×L spectral image, could have a similar spectral response, Fig. 3 illustrates this premise. Then, the superpixel, which is formed of q×q measurement pixels in the sensor, is taken as the spectral response of a single pixel for a decimated reconstruction [12]. For instance, Fig. 4 shows an example of the measurement selection for a superpixel side size of q=2. In total are taken q2 subsets of measurements in a single shot of the multi-patterned filter arrays detectors.

L

N

N

qq

Fig. 3. Illustration of the assumption that for q×q neighboring pixels, in a N×N×L spectral image, the spectral

response is simillar.

In the acquisition model, the matrix product between a downsampling matrix and the total sensor measurements forms each subset of measurements. More specifically, each subset is given by

,fy HD = (7)

where D does a downsampling in each q×q square of pixels for taken q2 different subsets of the total measurements, and y is the ht subset of

measurements where .1,0 2 −∈ q Fig. 5 shows an example of a subset measurement selection for q=2. Precisely, the function of the decimation matrix D is selecting in each q×q block of the measurements the ht - element for forming

the ht -subset of measurements. The decimation matrix element-by-element can be expressed as:

−

++

−+==

otherwise,,0

),()1( if,1,

qNqN

iqNqiqj

ji

D (8)

where ,12

,,1,0 −=q

Ni

2 j=0,1,…,N2-1, and

.1,,1,0 2 −= q Fig. 6 depicts the downsampling

matrix D for q=2, N=6, and .3,2,1,0= The white squares represent one-valued elements.

y0

y10

y11 y1

3

y12

1st subset of measurements

2nd subset of measurements

3rd subset of measurements

4th subset of measurements

y1 y2 y3

N

N

q

q

A macropixel

Fig. 4. Example of a subset selection for a superpixel size of q×q = 2×2 that forms 4 subsets of measurements in a single shot of the multi-patterned filter arrays -

based sensor.

In this case, the complete set of measurements is given by

f,f

D

DD

y

yy

y S

qq

HH =

=

=

−− 1

1

0

1

1

0

22

(9)

where subjacent data cube projection is reconstructed solving an l1 and l2 minimization problem, where the decimation process is taken into account. The optimization problem is given by

),argmin(~

12θθyf

θτ+−= ΨH Ψ S where y is given

by (10), HS is the measurement matrix defined as

H,DDH TTqTS ])...()[( 10 2 −= θ is an S-sparse

representation of a low resolution version of f on the basis Ψ, and τ is a regularization constant [13].

HS is the measurement matrix defined as

H,DDH TTqTS ])12

...()0[( −= where D is the th

downsampling matrix used for measurement selection, θ is an S-sparse representation of a low


54

resolution version of f on the basis Ψ, and τ is a regularization constant.

Fig. 5. The downsampling matrix D is shown for q=2, N=6 and .3,2,1,0= White squares represent ones and the

black elements are zero.

5. Filters For developing this work, two multi-patterned

filter arrays were selected. First, a spectral response for pixel that can be selected randomly from a set of band-pass filters; and second, dichroic filters, that is a special case of random band-pass where its spectral response let pass only one spectral band for each sensor pixel.

The spectral response of a ),( Hi

Li λλ band-pass

filter can be defined as

otherwise,

, if0,

1 Hi

Li

ikk λλ ≤≤

=,

)(t (10)

for 10 −= Lk ,..., , 10 2 −= Ni ,..., , and

,..., 10 −∈≤ LHi

Li λλ . For instance, L=4, 2=L

iλ ,

and 3=Hiλ define the spectral response of the

position 1=i , as ][)(t 11001 =k . Fig. 6(a)

depicts an example of a coded field column filtered by band-pass filters, with representations of the spectral response of three band-pass filter pixels.

ℒ ℋ λ 5 λ 5

100%

λ8 λ8

100%

λ1 λ1

100%

ℒ ℋ

ℒ ℋ

λ 5

100%

λ8

100%

100%

!

!

!

λ1

(a) (b)

Fig. 6. Band-pass and dichroic filters representations. (a) Example of a coded field column filtered by band-pass

filter pixels, with the spectral response representations of three band-pass filter pixels; (b) Example of a coded field column filtered by dichroic filter pixels, with the

spectral response of three dichroic filters.

Dichroic filters are a special case of band-pass filters that let pass only one spectral band. Then the spectral response of a )( D

iλ dichroic filter pixel can

be defined as

otherwise,

, if0,

1 kDi

ik=

= λ,)(t (11)

for ,..., 10 −∈ LD

iλ , and 10 2 −= Ni ,..., . For example,

if L=4, and 35 =Dλ , then the spectral response of the

position 5=i in the vectorized filter array is ][)(t 10001 =k . Fig. 6(b) depicts an example of a

coded field column filtered by dichroic filter pixels, with representations of the spectral response of three dichroic filter pixels.

6. Simulations and Results To verify the multi-patterned filter array detectors

reconstructions, a set of compressive measurements is simulated using the model of Eq. (2). These measurements are constructed employing two spectral images captured with a CCD camera Apogee Alta U260 and a VariSpec liquid crystal tunable filter, in the range of wavelength 400 nm-560 nm, with steps of 10 nm [14]. The resulting test data cubes have 512512 × pixels of spatial resolution and L=16 spectral bands. The RGB images mapped versions of the selected data cubes are shown in Fig. 7.

Fig. 7. The RGB images mapped versions of the selected

data cubes (left) Feathers, and (right) Glass Tiles.

The experiments were carried out using the images Feathers and Glass Tiles. Compressive sensing reconstruction is realized using the GPSR algorithm [15]. Simulations results are analyzed in terms of PSNR (Peak-Signal-to-Noise-Ratio) of the reconstructed images. The representation basis Ψ is a Kronecker product 21 Ψ⊗Ψ=Ψ , where 1Ψ is the

two-dimensional-wavelet Symmlet-8 basis and 2Ψ is

the cosine basis. The simulations are performed in a desktop architecture with an Intel Core i7 3.6 GHz processor, 32 GB RAM, and using Matlab R2014a. Each experiment is repeated ten times and the respective results are averaged.


55

The measurements are simulated using the first L spectral bands of the real data cubes. Ensembles of dichroic and band pass filters based on a random selection of spectral bands are used for obtaining measurements for the compressive sensing approach. Fig. 8 shows a comparison of the average PSNR reconstruction for the Feathers data cube. Fig 9 shows similar results for PSNR reconstructions in the Glass Tiles data cube.

(a)

(b)

(c)

(d)

Fig. 8. Reconstructions of the Feathers data cube using (a) 1=q , (b) 2=q , (c) 4=q and (d) 8=q .

(a)

(b)

(c)

(d)

Fig. 9. Reconstructions of the Glass Tiles data cube using

(a) 1=q , (b) 2=q , (c) 4=q and (d) 8=q .

For the model that perform CS reconstructions, increasing the size of the superpixel q improves the PSNR. Additionally, the reconstruction based on superpixel measurements selection (Dichroic and

Band pass) approaches have a better performance than the demosaicing approach. For q>1 the Dichroic random filters has a better performance than the others. Finally, increasing spectral bands decreases the PSNR for all reconstruction methods. Furthermore, the simulation time is halved at each q increase.

7. Conclusions A model for CS recovery of spectral images

sensed by multi-patterned focal plane array detectors using a superpixel approach is presented. The model performs a selection of measurements subsets to form superpixels that have spectral information of a single recovery pixel. The CS reconstruction approaches using dichroic and band pass filters are compared with a traditional demosaicing reconstruction. For the CS reconstruction, the PSNR increases with the superpixel side size. For instance, the improvements range from 0.5 dB to 4 dB with respect to the traditional approach in real data cubes. Results show that increasing spectral bands decreases the PSNR for all reconstruction methods.

Acknowledgments The authors gratefully acknowledge to

Vicerrectoría de Investigación y Extension of the Universidad Industrial de Santander for supporting this research registered under the project title: Detección y Clasificación en imágenes espectrales obtenidas a través de un sistema de adquisición compresivo con un detector de un solo pixel, (VIE 1802 code).

References [1]. J. Ryan, C. Davis, N. Tufillaro, R. Kudela,

B.-C. Gao, Application of the Hyperspectral Imager for the Coastal Ocean to Phytoplankton Ecology Studies in Monterey Bay, CA, USA, Remote Sensing, Vol. 6, No. 2, January 2014, pp. 1007-1025.

[2]. Z. Xiong, A. Xie, D.-W. Sun, X.-A. Zeng, D. Liu, Applications of Hyperspectral Imaging in Chicken Meat Safety and Quality Detection and Evaluation: A Review, Critical Reviews in Food Science and Nutrition, 55, 9, 2015, pp. 1287-1301.

[3]. G. J. Bellante, S. L. Powell, R. L. Lawrence, K. S. Repasky, T. A. O. Dougher, Aerial detection of a simulated CO2 leak from a geologic sequestration site using hyperspectral imagery, International Journal of Greenhouse Gas Control, Vol. 13, March 2013, pp. 124-137.

[4]. M. Mehrübeoglu, G. W. Buck, D. W. Livingston, Differentiation of bacterial colonies and temporal growth patterns using hyperspectral imaging, in SPIE Optical Engineering + Applications, 2014, 922206.

[5]. G. Lu, B. Fei, Medical hyperspectral imaging: a review, Journal of Biomedical Optics, Vol. 19, No. 1, January 2014, 10901.


56

[6]. J. D. Barrie, K. A. Aitchison, G. S. Rossano, M. H. Abraham, Patterning of multilayer dielectric optical coatings for multispectral CCDs, Thin Solid Films, Vol. 270, No. 1-2, December 1995, pp. 6-9.

[7]. Z. Frentress, L. C. Young, H. D. Edwards, Field Photometer with Nine-Element Filter Wheel, Applied Optics, Vol. 3, 1964, pp. 303-309.

[8]. P. J. Lapray, X. Wang, J. B. Thomas, P. Gouton, Multispectral Filter Arrays: Recent Advances and Practical Implementation, Sensors, Vol. 14, No. 11, 2014, pp. 21626-21659.

[9]. L. Miao, H. Qi, R. Ramanath, W. E. Snyder, Binary tree-based generic demosaicking algorithm for multispectral filter arrays, IEEE Transactions on Image Processing, Vol. 15, No. 11, 2006, pp. 3550-3558.

[10]. J. Brauers, T. Aach, A Color Filter Array Based Multispectral Camera, in Proceeding of the 12th Workshop Farbbildverarbeitung, 2006.

[11]. Y. Monno, M. Tanaka, M. Okutomi, Multispectral Demosaicking Using Adaptive Kernel Upsampling, in Proceedings of the 18th IEEE International Conference on Image Processing (ICIP'11), 2011, pp. 3218-3221.

[12]. Y. Mejia, F. Rojas, H. Arguello, Macropixel Compressive Sensing Reconstruction of Spectral Images Sensed by Multispectral Filter Array-based Sensors, in Proceeding of the 6th International Conference on Sensor Device Technologies and Applications (SENSORDEVICES’15), Venice, Italy, 23-28 August 2015, pp.158-163.

[13]. H. Arguello, G. R. Arce, Colored coded aperture design by concentration of measure in compressive spectral imaging, IEEE Transactions on Image Processing, Vol. 23, No. 4, April 2014, pp. 1896-1908.

[14]. CAVE | Projects: Multispectral Image Database. [Online]. Available at: http://www.cs.columbia.edu/CAVE/databases/multispectral/ [Accessed: 24-Feb-2015].

[15]. M. A. Figueiredo, R. D. Nowak, S. J. Wright, Gradient projection for sparse reconstruction: Application to compressed sensing and other inverse problems, Journal of Selected Topics in Signal Processing, Vol. 1, No. 4, 2007, pp. 586-597.

___________________


http://www.sensorsportal.com/HTML/BOOKSTORE/Sensors_and_Signals.htm





57



Advanced Controlled Cryogenic Ablation Using Ultrasonic Sensing System

* Assaf Sharon, Gabor Kosa

Robots and BioMedical Micro Systems (RBM2S) research laboratory, School of Mechanical Engineering, Faculty of Engineering, Tel Aviv University, Tel Aviv, Israel

* Tel.: + 972523404144, fax: 972-3-6407334 * E-mail: [email protected]

Received: 31 August 2015 /Accepted: 5 October 2015 /Published: 30 October 2015 Abstract: Cryoablation process is one of the methods for treating various tissue abnormities. Cryoablation devices are mostly minimally invasive and are used in open loop control, monitored by additional imaging devices. In this study, we monitor the growth of the ablated area by using a miniature ultrasonic transducer that is collocated with the tip of the cryogenic device. The 20 MHz ultrasonic sensor is capable of measuring the size of the ice sphere that is created in front of the needle. In addition to real time monitoring of the ablation process, the ultrasonic sensor will be able to determine the local thickness of the tissue prior to the treatment (thus enabling the setting of the power of the ablation treatment). The combined device will shorten the ablation treatment and will eliminate the need for additional ablation treatments or monitoring devices. The proof of concept was done in water, ultrasonic gel and muscle tissue. In the experiments we found that, in the frequency domain one can identify at 10-12 MHz the increase of the intensity of the returned echo in the ice and the decrease of the signal after the ice-tissue boundary. One can correlate the increase of the intensity with the growth of the ice sphere. Copyright © 2015 IFSA Publishing, S. L. Keywords: Cryogenic, Ablation, Control, Ultrasound, Piezo, Ice. 1. Introduction

Cryosurgery, also referred to as Cryotherapy or Cryoablation, is a minimally invasive surgical technique in which freezing is used to destroy undesirable tissue. Cryoablative techniques have persistently improved over the past forty years with the development of successive generations of devices including Cryoneedles, Cryoballoons, intraoperative ultrasound and vast knowledge of the mechanisms by which cells are affected by low temperature exposure [2]. We now recognize two mechanisms causing cell death following a freezing cycle: direct mechanism adjacent to the ablating device of cell rupture due to intracellular ice crystal formation and cellular

dehydration with associated osmotic damage, and indirect mechanism of ischemia and necrosis throughout the tissue/tumor peripheral zone [3-4]. To perform a cryosurgical procedure successfully, it is important to monitor precisely and evaluate accurately the extent of freezing. Failure to do so can lead to either insufficient or excessive freezing, and consequently, to recurrence of malignancies treated by cryosurgery or to destruction of healthy tissues [5]. Most of the Cryoablation devices are used in an open loop control. The results of the treatment are inspected by additional imaging devices such as ultrasound, camera, temperature sensors and other sensors according to the relevant application [2].




58

In this study, we intend to add a miniature ultrasonic sensor to a cryoablation device in order to determine the treatment progress in real time, observing the ice sphere’s boundary growth. This capability allows controlling directly the ablation process by closed loop control. Such a device, will be able to determine much more effectively (faster, more accurately and more precisely). The close loop controlled cryogenic device will be more efficient and safe than the current treatment. In addition to real time monitoring of the ablation process, the ultrasonic sensor will be able to determine the local thickness of the target area before treatment and will enable a more accurate setting of the device’s parameters (freezing power and period). The system will shorten the ablation treatment and eliminate the need for reoccurrence treatments.

To the best of our knowledge, there is no Cryoablation device with collocated sensor that monitors in-situ the progress of the target tissue’s freezing.

The purpose of the present study is to show feasibility of detecting the ice sphere growth from within the ice sphere during Cryoablative therapy without using additional monitoring devices. We assume that by transducing a high frequency ultra-sound wave we will be able to determine the location

of the ice sphere outer boundary due to the impedance differences between frozen and unfrozen tissue. We expect to be able to distinguish the frozen tissue’s returning echo’s (inside the ice sphere) from the unfrozen tissue’s returning echo’s (at the outer rim of the ice sphere).

In Section 2 (Method), we will present an overview of the entire system with detailed description of its two main components; Ablation device and monitoring device, following with an explanation of the monitoring method and the analysis done. Section 3 (Results) will include a summary of the results of different analysis steps and our conclusions will be presented in Section 4. 2. Method 2.1. System Overview

The controlled Cryoablation system that we developed is a device for Cryoablation therapy with an ultrasound transducer attached to it as shown in Fig. 1. In the future the ultrasonic sensor will be integrated into the cryogenic needle.

(a

(a)

(b)

(c)

Fig. 1. CryoProbe to sensor connection before (a) and after (b) freezing; Schematic overview of the system (c).

The main components of the system are:

1) A cryoablation device with a Cryoprobe reaching extreme low temperatures (about -170 ºC) at its tip. Acronymed as CAS.

2) A forward looking ultrasonic sensor that can measure regular and frozen tissue up to 10 mm in depth. Acronymed as USS.

The detailed description of the components is described in the following sections.

Ultrasound Transducer

Cryogenic Needle

Ice here


59

2.2. Ablation Device

We used the Cryo-ablative device designated as IceSense3™ system (Fig. 2) manufactured by IceCure Medical. IceCure Medical developed a minimally invasive Cryo-ablation therapy for the Women health market [6].

Fig. 2. IceCure's IceSense3™ System (Left)

and its CryoProbes with an example of Ice Sphere at their tip (Right).

The IceSense3™ system, provides a minimally invasive, in-office, definitive treatment which uses low temperatures (about -170°C) to destroy (ablate) the targeted tissue in situ. The system uses a closed loop cryogen which reaches a Cryoprobe tip at the center of the ablate tissue and cooled to sub-zero temperatures, removing heat from the targeted tissue by conduction [7] (See Fig. 1(b) and Fig. 2). During the Cryo-ablative procedure, an ice spheroid (for convenience we regard to it as a sphere) is formed around the Cryoprobe tip. The ice sphere size varies in time and can reach a diameter of 40 mm and length of around 55 mm after 10 minutes.

2.3. Monitoring Device 2.3.1. 18.5 MHz Transducer (matching

Water)

We have used an Ultrasound transducer manufactured by Vermon, France as our monitoring element (Fig. 3).

Fig. 3. Vermon ultrasonic sensor -Ø2×5.5 mm.

The transducer is a single element transducer with outer dimensions of Ø2×5.5 mm and central frequency of 18.5 MHz as shown in the frequency response in Fig. 4.

Fig. 4. Vermon US sensor frequency response with central frequency of 18.5 MHz.

This frequency is equivalent to 0.126 mm axial resolution [8] in water. The frequency response (Fig. 4), has a central frequency of 18.5 MHz (at -3 dB) bandwidth frequency of 8.7 MHz, low cut frequency 14.2 MHz and high cut frequency 22.9 MHz.

The ultrasound transducer time response in water is noted in Fig. 5.

Fig. 5. USS time response.

The transducer is controlled by USBox system of Lecoeur Electronique Company, transmitting 230 V (1 volts step), square pulse. The monitoring was done in A-mode, transmitting and receiving from the same transducer. The returning Echo’s were monitored by Matlab software, sampling in 80 MHz in order to observe returning Echo’s of up to 40 MHz.

2.3.2. 11.5 MHz Transducer (matching ICE)

We have used PiezoCAD software [9], first to compare our calculations to the current 18.5 MHZ transducer (Fig. 6) and then, custom designed using the same S/W an 11.5 MHz Ultrasound transducer matching Ice which was later manufactured by ImaSonic, France.


60

Fig. 6. 18.5 MHz sensor calculated Transmit efficiency using PiezoCAD S/W.

The transducer is a single element transducer with outer dimensions of Ø1.5×5.0 mm and central frequency of 11.5 MHz with two matching layers fitting Ice as our medium (Fig. 7).

Fig. 7. 11.5 MHz transducer, ImaSonic, France.

2.4. Experimental Method

Our main objective, was to recognize the outer contour of the ice sphere in real time (during treatment) when our monitoring device is located at the center of the freezing zone hence allowing us to have both treating and monitoring elements in a single device. The transducer was chosen to be minimally in size to allow in the future positioning inside the ablating element (less than Ø3 mm). In this experiment, we connected it to the center of the CryoProbe freezing zone with additional connector placing the transducer adjacent to the CryoProbe outer surface (Fig. 1). We have used US Parker gel inside a Standard 1000 ml beaker as our tissue model (ablated medium) and compared it to water and chicken breast tissue.

Several cryogenic treatment simulations were done. The freezing process duration was up to 10 minutes. The ice sphere size was measured using an external camera in time periods of 30 seconds as a reference to the transducer measurements. All returning echoes were received in a time to voltage raw data manner and several analyses were done in order to observe the ice sphere growth as detailed below.

2.5. Data Analysis

Looking at the time response of the echo (Fig. 8), one can observe the large attenuation of the signal in the ice.

The ice-water boundary, cannot be distinguished from direct A-mode inspection of the ice sphere’s outer contour.

In order to detect the boundary we transferred the signal to the frequency domain using Matlab software’s signal analysis toolbox. We are using the following steps to estimate accurately the ice sphere’s contour’s distance:

1) The sound velocity in ice is estimated by correlating the signal response near the exit of the transducer (called ringing in the US jargon) in water and ice.

2) Using the sound velocity, the time response is converted into distance and an A-mode US image is derived (Fig. 8).

3) The time response is also converted to the frequency domain using the short time Fourier transformation. The representation of the data is in a spectrogram that enables identification of significant features in the image (See Fig. 9).

Fig. 8. Returning US Echo of 5 mm thick frozen meat slice.

Fig. 9. The spectogram at 3 mm (a) and 4 mm (b) radius Ice sphere (In Parker Gel). The area of interest

of 10-12 MHz is in the black frames. The intensity of the echo is depicted by jet colormap.

4) In order to emphasize the growth of the ice sphere we used a binary conversion. This calculation is based on the reduction of the spectrogram

Expected ice-water boundary

0.000 Frequency, MHz 40000

750000

Po/Pi

mW/W

0.000


61

according to the derivative of the intensity in the spectrogram (growing intensity black, and vice versa).

5) We also calculated, the sum of the total intensity at the targeted frequency range and found that it is a good indicator of the growth of the ice sphere.

3. Results 3.1. 18.5 MHz Transducer (matching Water)

Using the correlation analysis, we were able to determine that the frozen gel used has a sound velocity of 2520 m/s (the literature value of ice sound velocity is 3600 m/s and water 1430 m/s) with this property determine we were able to match the returning Echo’s with their correct location.

The dynamic process of the growing ice sphere is characterized by the increase of the intensity from the ice layer and the reduction of the intensity of the echo from the water or tissue layer beyond the ice sphere. Fig. 9 demonstrates that this process is the most apparent in the bandwidth of 10-12 MHz.

Focusing on the bandwidth we identified in Fig. 9 we are able to show a consistence advancing increase of intensity of the echo in time as shown in Fig. 10 (red and yellow in the jet colormap).

Fig. 10. 10-12 MHz Range at different ice sphere sizes: radius of – 0 mm (a), 2 mm (b), 3 mm (c), 4 mm (d)

and 5 mm (e) - (In Parker Gel The intensity of the echo is depicted by jet colormap).

This dynamic process can be correlated to the growth of the ice sphere. The increase in the echo from the ice is more emphasized than the decrease in the gel.

To better distinguish the intensity increments, we did additional binary analysis (Fig. 11).

Each signal received at different ice sphere size was compared to the base signal (prior to the freezing cycle) – areas with increased intensity than the base signal are shown in black and areas with lower intensity than the base intensity are shown in white. The advance of the boundary can be detected by the increase of the black area with the growth of the ice

sphere. This binary intensity contrast is easier to detect using computerized models in real time.

Fig. 11. 10-12 MHz Range Binary differences analysis between the base line (0 Sec) and different ice sphere sizes:

radius of – 1 mm (a); 2 mm (b); 3 mm (c); 4 mm (d); and 5 mm (e) - (In Parker Gel).

Finally, we calculated the total intensity of the returning Echo’s of different ice sphere sizes during its growth and received an increasing graph for each experiment. We compared all experiments in gel and water in a single graph and receive a linear trend line with R-squared value of 0.9656 (Fig. 12).

Fig. 12. Retrieved signal Total Intensity Vs. different ice sphere sizes – 10-12 MHz Range (All Experiments in water or Parker Gel) with additional combined Linear Trendline

(in black).

One can see that the total intensity is a good quantitative measure for the increase of the ice sphere.

The results in the chicken breast tissue show similar results to the gel. There were visible increments with time of the returning Echo’s at 10-12 MHz frequency range, better observed when looking at the binary analysis (Fig. 13).

We also repeated the returning Echo’s total intensity calculation of different ice sphere sizes during its growth and received an increasing graph similar in values to all other experiments (Fig. 14).

Combining all the information gathered from the results, we receive a very good indication that measuring the ice ball growth from inside the ice ball is possible even in real time.


62

Fig. 13. 10-12 MHz Range Binary differences analysis between the base line (0 Sec) and different ice sphere sizes:

radius of – 1 mm (a); 2 mm (b); 3 mm (c); 4 mm (d); and 5 mm (e) - (In chicken breast).

Fig. 14. Retrieved signal Total Intensity Vs. different ice sphere sizes – 10-12 MHz Range (In chicken breast).

3.2. 11.5 MHz Transducer (Matching ICE)

Using the new custom made transducer we can observe the returning echo even in the time domain, due to the lower frequency transmitted pulse, and the Ice matching layers, both reducing the attenuation. Thus the ice-water boundary, can be distinguished from direct A-mode inspection of the ice sphere’s outer contour. Using the sound velocity, the time response is converted into distance and an A-mode US image is derived (Fig. 15).

Returning echoes of the ice-water

Real boundry

Fig. 15. Returning US Echo of 2 mm thick frozen Ice cube.

4. Conclusions

Examining our results, using the 2nd phase ultrasound transducer, we can clearly distinguish the ice-water boundary in the time domain. With the 1st phase transducer, we can clearly distinguish a growing intensity of the signal with the ice sphere growth. We also noticed, that our finding are similar when using ultrasound gel and breast tissue as our ablate model.

With further experiments (further using and analyzing the improved ultrasound sensor), we will be able to determine the ice sphere real size with high accuracy and precision (both with the time domain echo and the intensity correlation method).

This capability will allow us to monitor the Cryoablation treatment from within the Cryo-ablative device and in the future have a closed loop combined device both treating and monitoring in real time. References [1]. A. Sharon, G. Kosa, Controlled Cryogenic Ablation

Using Ultrasonic Sensing, in Proceeding of the of the 6th International Conference on Sensor Device Technologies and Applications (SENSORDEVICES’15), Venice, Italy, 23-28 August 2015, pp. 116-120.

[2]. B. Rubinsky, CRYOSURGERY, Annual Review of Biomedical Engineering, Vol. 2, 2000, pp. 157-187.

[3]. A. A. Gage, J. G. Baust, Cryosurgery-a review of recent advances and current issues, Cryoletters, Vol. 23, No. 2, 2002, pp. 69-78.

[4]. M. Maccini, D. Sehrt, A. Pompeo, F. A. Chicoli, W. R. Molina, F. J. Kim, Biophysiologic considerations in cryoablation: a practical mechanistic molecular review, International Braz J Urol, Vol. 37, No. 6, 2011, pp. 693-696.

[5]. J. G. Baust, A. A. Gage, A. T. Robilottto, J. M. Baust, The pathophysiology of thermoablation: optimizing cryoablation, Current Opinion in Urology, Vol. 19, No. 2, 2009, pp. 127-132.

[6]. IceCureMedical Website, Available: http://icecure-medical.com/

[7]. J. Baust, A. Gage, T. B. Johansen, J. Baust, Mechanisms of cryoablation: clinical consequences on malignant tumors, Cryobiology, Vol. 68, No. 1, 2014, pp. 1-11.

[8]. D. Ensminger, L. J. Bond, Ultrasonics: Fundamentals, Technologies, and Applications, CRC Press, Taylor & Francis Group, 2011.

[9]. Sonic Concepts Website: Available: http://www.sonicconcepts.com/index.php/products-1/piezocad

___________________



63



Cavity Enhanced Absorption Spectroscopy in Air Pollution Monitoring

Janusz MIKOŁAJCZYK, Zbigniew BIELECKI, Jacek WOJTAS

and Sylwester CHOJNOWSKI Institute of Optoelectronics, Military University of Technology,

2 Kaliskiego Str., 00908Warsaw, Poland Tel./fax.: 0048261837943

E-mail: [email protected]

Received: 31 August 2015 /Accepted: 5 October 2015 /Published: 30 October 2015 Abstract: The paper presents some practical aspects of cavity enhanced absorption spectroscopy application in detection of nitrogen dioxide (NO2), nitrous oxide (N2O), nitric oxide (NO) and carbon monoxide (CO). These gases are very important for monitoring of environment. There are shown results of lab-setups for N2O, NO, CO detection and portable sensor of NO2. The portable instrument operates in the UV spectral range and reaches a level of single ppb. The lab–devices use high precision mid-infrared spectroscopy and they was demonstrated during testing the laboratory air. These sensors are able to measure concentration at the ppb level using quantum cascade lasers, high quality optical cavities and modern MCT detection modules. It makes it possible to apply such sensors in monitoring the atmosphere quality. Copyright © 2015 IFSA Publishing, S. L. Keywords: Laser absorption spectroscopy, Cavity enhanced spectroscopy, CEAS, Gas sensors, QCL. 1. Introduction

Detection of various gases and measurement of their concentration are very important for monitoring of industrial processes and investigation of their environmental impact. Nitrous oxide is so-called greenhouse gas emitted from agriculture, transportation, fossil fuel combustion, wastewater management, and industrial processes agriculture. This gas is naturally present in the atmosphere and has a variety of natural sources e.g. plants, animals, and microorganisms. Nitrous oxide is removed from the atmosphere by absorption of some bacteria or by ultraviolet radiation destruction or by chemical reactions. However, N2O can stay in the atmosphere for an average of 114 years before being “natural” removed. In practice, it is observed ca. 300 times stronger impact of N2O on warming the atmosphere in comparison with carbon dioxide [1].

Carbon monoxide and oxides of nitrogen are ones of the most important air pollutants from petrol, diesel, and alternative-fuel engines. The concentration of these pollutants is regulated by the Euro emissions standards. Modern cars are not significant problem for air quality but their large quantity. Carbon monoxide and oxides of nitrogen are generally invisible, and in comparison to CO2 their concentrations are less dependent on fuel consumption. Carbon monoxide reduces the blood’s oxygen-carrying capacity and it can be fatal in the case of extreme levels of exposure. At lower concentrations CO may increase a health risk, particularly to those suffering from heart disease.

Nitrogen oxide NO reacts in the atmosphere to form nitrogen dioxide which can have adverse effects on people health with respiratory illness. The gas concentration have been linked with increased hospital admissions due to respiratory problems and




64

the exposure time may affect lung function and increase the response to allergens in sensitive people.

Within the last decades, many methods were developed for detection of volatile substances. There is a few classifications of gas sensors. For example, they can be classified into semiconductor-type, solid electrolyte-type, electrochemical-type and catalytic combustion-type, capacitor and a simple NDIR type. A separate group of gas sensors are gas analyzers. In this group, the main roles play mass spectrometry and gas chromatography. Their main inconveniences are size, cost of the apparatus and the maintenance, complicated exploitation. But the designing process of sensitive pollution air sensors is very dynamic. It is shown, that optoelectronic sensors employing laser absorption spectroscopy are very useful in the effort to minimize the level of the environment contamination [2]. They use the phenomenon of optical radiation interaction with specific compounds to detect and to measure the concentrations of the molecules, provide achieving low detection limits and high selectivity. For this purpose, it is necessary to apply radiation, the wavelength of which is precisely tuned to the spectral range characterized by strong absorption of the tested molecules. Such sensors are more sensitive and selective than many others [2].

2. Cavity Enhanced Absorption Spectroscopy in Atmosphere Monitoring Cavity enhanced absorption spectroscopy (CEAS)

was proposed by R. Engeln in 1998. The principle of its operation is very similar to cavity ring down spectroscopy (CRDS). In both setups there is applied an optical cavity with a high quality factor that is made up of two concave mirrors with very high reflectivity R. This results in a long optical path, even up to several kilometers [5]. A pulse of optical radiation is injected into the cavity through one of the mirrors. Then inside the cavity multiple reflections are observed. After each reflection, part of the radiation exiting from the cavity is registered with a photodetector. The output signal from the photodetector determines the intensity of radiation propagated inside the optical cavity. If the laser wavelength is matched to the absorption spectra of gas filling the cavity, the cavity quality decreases. Thus, parameters of the photodetector signal are changed. Thanks to this, the absorption coefficient and concentration of gas can be determined.

The main difference between CEAS and CRDS relates to the laser beam and the optical cavity alignment. In CEAS the light is injected at a very small angle in respect to the cavity axis (Fig. 1). As a result, dense structure of weak radiation modes is obtained or they overlap. Sometimes, a piezoelectric-driven mount that modulates the cavity length (position of the output mirror) is used in order to

prevent the establishment of a constant mode structure within the cavity [3].

Fig. 1. The scheme of CEAS setup.

The mode structure causes that the entire system is much less sensitive to instability in the cavity and to instability in laser frequencies. Additionally, due to off-axis illumination of the front mirror, the source interference by the optical feedback from the cavity is eliminated. CEAS sensors attain the detection limit of about 10-9cm-1 [5]. Therefore, this method creates the best opportunity to develop a portable optoelectronic sensor of nitrogen oxides.

In the applied methods, determination of the gas concentration is performing by measuring the decay time of the photodetector signal [3-4]. If the laser pulse duration is negligibly short and only the main transverse mode of the cavity is excited, then exponential decay of radiation intensity can be measured. The decay time of signal in the cavity (τ) depends on the reflectivity of mirrors, diffraction losses and the extinction coefficient α:

( )LRc

L

ατ

+−=

1, (1)

where L is the length of the resonator, c is the speed of light. Determination of the examined gas concentration is a two-step process. First, measurement of the decay time (τ0) of radiation in the optical cavity without tested gas is performed. During next step, the same measurements is made (decay time τ) for the cavity filled with the gas. Knowing the absorption cross section (σ) of the examined gas, its concentration can be calculated from the formula

−=

01

111

ττσcC , (2)

where


65

( )Rc

L

−=

10τ (3)

3. Laboratory Experiments Nitrogen oxides are very important to control in

the air because they cause accelerate corrosion of stone buildings and metal structures, threaten human health, irritate the respiratory system and general weaken the body's resistance to infectious diseases. That is why, the first research was focused on the development of nitrogen oxides sensors providing the detection limit as low as possible. In our experimental setups, visible and mid-IR semiconductor lasers were applied. The first one was constructed using a blue-violet semiconductor laser (414 nm) developed at the Institute of High Pressure of the Polish Academy of Sciences. For detection of NO, N2O, the mid-IR lasers operating at the wavelength located in the infrared region were applied (Fig. 2) [6].

Fig. 2. The results of the calculations of the absorption cross section for selected gases.

There were applied quantum cascade lasers (4.53 µm and 5.27 µm) from Alpes Lasers SA, Switzerland.

In the case of CO, the prototype quantum cascade laser (4.78 µm) from the Institute of Electron Technology, Poland was used.

The views of the constructed NO and NO2 sensors are presented in Fig. 3.

(a)

(b)

Fig. 3. Photos of NO2 sensor (a) and prototype NO sensor (b).

During the sensor investigation, concentration measurements of reference gas samples were carried out. Reference samples were prepared using the 491M type gas standards generator from KIN-TEK Laboratories, Inc. (La Marque, TX, USA, Fig. 4).

Fig. 4. Photo of 491M type gas standards generator.

The achieved results are summarized in Table 1.

Table 1. Measured detection limit of the designed sensors [7].

Type of sensor

Operation Wavelength

Detection Limit

NO2 414 nm 1 ppb N2O 4.53 µm 45 ppb NO 5.27 µm 70 ppb CO 4.78 µm Approx. 150 ppb

4. Outdoor Tests

The outdoor tests were performed with portable NO2 sensor. The measurements were made in the


66

neighborhood of the university camp for different kind of environment. The results are listed in Table 2.

Table 2. The measured NO2 concentration for environments.

The lowest concentration value was obtained for forest area. This value increases with movement in the more urban area. The highest level is observed for area with large car traffic. That is why, the addition tests were performed for direct diesel car exhaust. The mean value was 980 ppb. These results shown that the NO2 sensor can be applied in many scenarios of environment monitoring.

5. Conclusions CEAS sensors are able to measure concentration

of atmosphere gases at the ppb level. Their sensitivity is comparable with the sensitivities of other detecting instruments. But in comparison, they offer quick response, no-impact on tested air samples, online operation, high selectivity and ease of use.

It is observed, that detection of NO2 molecules can be performed using broadband multimode lasers. It is possible because a relatively large mean absorption cross section within the range of several nanometers resulting from analyses of electronic transitions.

However, for other compounds (like N2O, NO and CO) mid-IR absorption lines are very narrow. That is why, these sensors require to apply precisely

defined lasers wavelengths. The presented results shown, that application of mid-IR QCL lasers in such sensors is very promising. Currently, several works related to the development of these sensors is still continuing. Acknowledgements

The presented work was supported by The National Centre for Research and Development in the scope of Project ID: 179616. References [1]. Web Portal United States EPA, Overview of

Greenhouse Gases, (http://www3.epa.gov/ climatechange/ghgemissions/gases/n2o.html).

[2]. M. Miczuga, K. Kopczyński, J. Pietrzak, R. Owczarek, Measuring system for detection and identification of hazardous chemicals, SPIE, Vol. 8703, 2012.

[3]. T. Stacewicz, J. Wojtas, Z. Bielecki, M. Nowakowski, J. Mikołajczyk, R. Mędrzycki, B. Rutecka, Cavity ring down spectroscopy: detection of trace amounts of substance, Opto-Electron. Rev., Vol. 20, No. 1, 2012, pp. 34-41.

[4]. J. B. Paul, L. Lapson, J. G. Anderson, Ultrasensitive absorption spectroscopy with a high-finesse optical cavity and off-axis alignment, Applied Optics, Vol. 40, Issue 27, 2001, pp. 4904-4910.

[5]. I. Courtillot, J. Morville, V. Motto-Ros, D. Romanini. Sub-ppb NO2 detection by optical feedback cavity-enhanced absorption spectroscopy with a blue diode laser, Applied Physics: B, Vol. 85, Issue 2, 2006, pp. 407-412.

[6]. J. Wojtas, J. Mikolajczyk, Z. Bielecki, Aspects of the Application of Cavity Enhanced Spectroscopy to Nitrogen Oxides Detection, Sensors, Vol. 13, No. 6, 2013, pp. 7570-7598.

[7]. Z. Bielecki, J. Wojtas, J. Mikolajczyk, S. Chojnowski, Application of Cavity Enhanced Absorption Spectroscopy in Detection of Selected Gas Pollutants, in Proceedings of the 6th International Conference on Sensor Technologies and Applications (SENSORCOMM’15), Venice, Italy, 23-28 August 2015, pp. 83-84

___________________


Place of measurement Concentration [ppb]

Min. Max. Mean

Wooded area 5 30 14 In the neighborhood of garbage dumps

23 35 30

Intersection 32 55 42 Rondo with high traffic 65 87 75 Overpass the road S8 30 57 44 In the neighborhood of the shooting range

28 51 35


67



Design and Analysis of a Collision Detector for Hybrid Robotic Machine Tools

Dan ZHANG, Bin WEI

University of Ontario Institute of Technology, 2000 Simcoe Street North, Oshawa, Ontario, L1H 7K4, Canada

Tel.: 905.721.8668 ext. 5721, fax: 905.721.3370 E-mail: [email protected], [email protected]

Received: 31 August 2015 /Accepted: 5 October 2015 /Published: 30 October 2015 Abstract: Capacitive sensing depends on the physical parameter changing either the spacing between the two plates or the dielectric constant. Based on this idea, a capacitive based collision detection sensor is proposed and designed in this paper for the purpose of detecting any collision between the end effector and peripheral equipment (e.g., fixture) for the three degrees of freedom hybrid robotic machine tools when it is in operation. One side of the finger-like capacitor is attached to the moving platform of the hybrid robotic manipulator and the other side of the finger-like capacitor is attached to the tool. When the tool accidently hits the peripheral equipment, the vibration will make the distance of the capacitor change and therefore trigger the machine to stop. The new design is illustrated and modelled. The capacitance, sensitivity and frequency response of the detector are analyzed in detail, and finally, the fabrication process is presented. The proposed collision detector can also be applied to other machine tools. Copyright © 2015 IFSA Publishing, S. L. Keywords: Machine tools, Collision detection, Sensor, Capacitance, Vibration, Modelling. 1. Introduction

A capacitor is defined as two conductors that can

hold opposite charges. If the distance and relative position between two conductors change due to the external force, the capacitance value will be changed. This is the basic principle of capacitive sensing, which belongs to electrostatic sensing. The major advantages of electrostatic sensing can be concluded as follows, firstly, it is simple. No special functional materials (such as piezoresistive and piezoelectric materials) are required. The sensing principle is easy to implement, requiring only two conducting surfaces. Secondly, it has the characteristic of fast response. Capacitor-based sensing has high response speed due to the fact that the transition speed is

controlled by the charging and discharging time constants that are small for good conductors [1-2]. There are two kinds of capacitive electrode geometries: parallel plate capacitor and interdigitated finger capacitor. For the interdigitated finger capacitor, it can be regarded as many parallel plate capacitors combining together. One side of the finger is fixed and the other side is suspended and can move in one or more axes.

Parallel mechanisms have been widely used in different kinds of areas [3-4], such as machine tools as shown in Fig. 1. Parallel robotics machine tools are the new trend for the manufacturing and automations. When machine tools are in operations, it is sometimes unavoidable to hit the peripheral equipment (e.g. fixture) by the tool, which can




68

damage cutting tools, clamps and fixtures, or cause damage to the machine itself as shown in Fig. 2, which undermines the performance of the whole parallel robotic system and delays the productions.

Fig. 1. 3-DOF parallel robotic machine tool developed in the R&A Lab in UOIT.

Fig. 2. Collision occurred during machining [10].

Therefore, how to prevent the collisions during machining is very important for machining processes. When the end-effector/tool of parallel robotic machine tool accidently hits the peripheral equipment, e.g. fixture, in a better case scenario, say an operator finds it and hits the emergency stop button, then sometimes it is too later. The machine tool should ideally stop immediately when the tool hits the peripheral equipment and stop causing

further damage [5]. Based on this need and motivation, a sensor should be developed to address this goal. Some computer aided manufacturing software have the ability to perform a machine collision check, but some machine tools do not have this function. Most computer aided manufacturing programs determine the cutter paths considering sometimes just the tool. In machine tools, it is likely to drive the end effector outside the bounds, resulting in a collision with others. Many machine tools do not aware about the surrounding environment. The end effector just follows the code and it is totally dependent on an operator to detect if a crash occurs.

Let us image the following, when the tool hits the fixture, it is a collision, then there should be a vibration produced, so how can we harness this vibration and convert the vibration to an electrical signal that can be recognized by a computer. Capacitive based sensor depends on the physical parameter changing, e.g. the space between the plates and the dielectric constant between them, etc. For example, in the vehicle air bag deployment system, a crash acceleration makes one plate closer to the other and therefore trigger the bag to deploy. Inspired by this idea, we design a capacitance-based collision detector/sensor that can sensor the vibration and convert the vibration to an electrical signal.

In [6], vibration on a rotating spindle is generated by the sum of the variations in weight distribution. The corrective action is needed to have a force with an equal but opposite direction to cancel the imbalance condition. The first step to achieve this is to measure the vibration. There, a vibration sensor is installed in a grinding machine spindle, and the vibration is measured using a vibration sensor that is composed of a seismic mass that is connected to a piezoelectric transducer which converts the vibration into an electric signal. The above one is mainly measure the amount of vibration that the rotating spindle produced and it is not appropriate to use here as a collision detector. In [7], the chatter vibration is detected by using three different acceleration sensors that attached to three different axis of machine tool. In [8], a web learning tool with 3D simulation for axial table collision detection was proposed, but no device has been designed. In [9], a vibration detection algorithm was proposed and a speed regulator was designed for the backlash vibration of a machine tool. In [10], a new approach was presented to detect and avoid hard and soft collisions caused by user errors, and a capacitance based sensor was briefly mentioned for the collision of the machine tools, but it did not explicitly design and propose the capacitance based collision sensor. In [11], a six-dimensional wrist force/torque sensor based on E-type membranes is designed and fabricated, and it is applied onto the five-axis parallel machine tool to measure the tool forces and torques, previous one is force/torque sensor (used to detect forces and moments), but that is not a vibration sensor.

In this paper, a collision detection sensor is designed that can sense the vibration that the end-


69

effector caused when the tool accidently hits the fixture. When the tool accidently hits the ground or something, it will produce vibration, the vibration will make the distance of the capacitor change and therefore trigger the machine to stop immediately. No one has ever designed the collision detector that gears towards the situation that the machine tool should stop immediately when the end-effector hits the peripheral equipment.

The structure of the paper is as follows. In Section 2, the new design will be illustrated. The capacitance and sensitivity will be discussed in Section 3. Section 4 analyzes the resonant frequency of the detector, and finally fabrication process is presented in Section 5.

2. New Design Capacitive sensing depends on the physical

parameter changing either the spacing between the two plates or the dielectric constant. Our vibration sensing method is based on this idea. Capacitive sensing needs external electronics to make the changes in capacitance to an output voltage that can be read by a computer.

One side of the finger-like capacitor is attached to the moving platform of the hybrid robotics manipulator and the other side of the finger-like capacitor is attached to the tool. When the tool accidently hits the ground or something, it will vibrate and the distance between the fingers will change and therefore, the capacitance will change and trigger the machine to stop. The capacitance between a pair of fingers is contributed by the surface of fingers in the overlapped region. Capacitance derived from multiple pairs are connected in parallel, so the total capacitance is the summation of capacitance contributed by neighboring fingers.

3. Capacitance and Sensitivity

For a single fixed finger and its two neighboring moving fingers. There are two capacitances associated with each finger pair, one is the left-side

of the finger, denoted as lC , and the other is the

right-side of the finger, denoted as rC . When the tool

is not vibrating, the values of these two capacitance is the same, i.e.:

0l r

ltC C

d

ε= = , (1)

where 0ε is the permittivity of the vacuum, l is the

engaged overlapping distance of the fingers, t is the thickness of the fingers, d is the distance between a fixed-comb finger and its neighboring movable finger.

3.1. Movement Along y Direction When the tool accidently hits the fixture, the tool

will vibrate, the free finger will move by a distance, say x, and then the capacitance values of these two capacitors become the following:

0l

ltC

d x

ε=−

(2)

and

0r

ltC

d x

ε=+

(3)

The total value of capacitance is:

C = lC + rC = 0lt

d x

ε−

+ 0lt

d x

ε+

= 02 2

2d lt

d x

ε−

(4)

For a case study, suppose there are 13 fingers,

which means there are 12 capacitors, so the 12 capacitors will contribute the total capacitance of the device. So the above can be rewritten as follows:

C = 12 02 2

2d lt

d x

ε−

(5)

This change can be transferred to the electrical

signal, and under a certain value, then it means the machine tool is in the process of manufacturing, even though there is small vibrations, the capacitance change is under that value, the capacitance will not trigger the electrical controller to stop the machine, but when the capacitance change is very large, then the capacitor will trigger the controller to stop the machine immediately. This is under the condition that when the pieces are softer than fixture. When the pieces are stiffer than fixture, then we need to set the condition that when the capacitance under that value, then the capacitor needs to trigger the controller to stop the machine, and when the vibration is above that value, then let the capacitor not trigger the controller to stop the machine. It can also be put as this way, under certain value range, the capacitance change is not sensitive (big) enough to trigger the controller to stop the machine, which is under the condition that the piece is softer than fixture. And also that certain value needs to be determined by experimentation.

If: C= 12 02 2

2d lt

d x

ε−

< Value 1

Value 1 needs to be determined by experiment. The capacitor will not trigger the controller to

stop the machine

If: C= 12 02 2

2d lt

d x

ε−

> Value 1


70

The capacitor will trigger the controller to stop the machine.

Above certain value range, the capacitance change is not sensitive (small) enough to trigger the controller to stop the machine, which is under the condition that the piece is stiffer than fixture. And also that certain value needs to be determined by experimentation.

If: |C= 12 02 2

2d lt

d x

ε−

|< |Value 1|

The capacitor will not trigger the controller to stop the machine.

If: |C= 12 02 2

2d lt

d x

ε−

|> |Value 1|

The capacitor will trigger the controller to stop the machine.

Set the decision logic to certain value, decision logic will receive a signal from the capacitor/sensor to determine if it is actually manufacturing or collision. We can set the logic to negative value when the pieces are stiffer than fixture, it is like a mathematic logic. The sensor will not decide if the contact is the beginning of a collision or simply defines the manufacturing pieces. This is the function of the decision logic [6], and set certain value to the logic, if the value larger or smaller than certain value that was given to the logic, then the detector/sensor will trigger the machine to stop or not to stop, it is related to the decision logic module design, which has been out of author’s research scope, and can be done as a future work. This report is mainly proposes the idea that use the capacitive principle based method to design the collision sensor.

In terms of when the cutting tool breaks the moment it hits the fixture, this must be the condition that the fixture is harder than the piece, if the piece is harder than the fixture, then the tool will also break when manufacturing pieces, so it is not realistic at all, so the pieces must be softer than fixture at than case. If it is in that case, as being said above, i.e. under certain value range, the capacitance change is not sensitive (big) enough to trigger the controller to stop the machine when the tool is in the process of manufacturing. And also that certain value needs to be determined by experiment. Ideally, when the tool hit the fixture, the detector/sensor will trigger the machine to stop immediately, so the tool will not break. Say a worst case scenario, the tool breaks, the capacitance will also change, so it will trigger the machine to stop. Either way, no matter the tool breaks or not, if the capacitance change is above that value, then it will trigger the machine to stop.

However during motions, the rate of capacitance change can be measured, this rate of change can also be called the displacement sensitivity, it is obtained by taking the derivative of C with respect to x, and we can have the following,

S(x) = C

x

∂∂

= 02 2 2

48

( )

d ltx

d x

ε−

(6)

The above is under the movement along the y

direction (transverse). Transverse comb drive devices are frequently used for sensing for the sensitivity and they are ease of fabrication.

3.2. Movement Along x Direction

When the movement is along the x direction, we have the following, please note that the movement along z direction is very small or none because the suspension beam is along the z direction which blocks the movement along z direction. There are 13 fingers, which means there are 12 capacitors, so the 12 capacitors will contribute the total capacitance of the device. At rest, the total capacitance is:

012lt

Cd

ε= (7)

When there is force in x direction, which will

make the fingers move in the x direction, and therefore will cause the effective thickness t’ to change. Suppose the change value is x, under the above changed condition, the capacitance will change to the following:

0 ( )12

l t xC

d

ε −= (8)

The relative change of capacitance w.r.t.

displacement x (i.e. displacement sensitivity, or the change of capacitance as a function of applied displacement) can be expressed as follows:

012 lC

x d

ε∂ = −∂

(9)

4. Frequency Response

The device can be seen as a fixed-free cantilever

beam, and the resonant frequency can be expressed as:

1 31

1.732

2

EIgf

Flπ= (10)

3w

12

tI = (11)

There are two suspension beams, so the force/spring constant can be expressed as follows [1]:


71

F k x= ⋅ (12)

3

31

w2

4

E tk

l= × (13)

Plug in the above F and I , resonant frequency

can be finally derived as follows:

1 31

1.732 1.732

2 2 6

EIg gf

Fl xπ π= = , (14)

where 1l is the suspension beam length. E is the

Young’s modulus. w is the width of the finger, t is the thickness of the finger. The resonant frequency value is depend on the above parameters, different parameter will result different values. 5. Fabrication

Silicon bulk micromachining is the process that

involve partial removal of bulk material in order to create three dimensional structures or free suspended devices. Etching a subtractive process that remove materials. Etching can be divided into two categories, one is wet etching and the other is dry etching. For the wet etching, the liquid etchants can be acids and hydroxides; for the dry etching, we have the physical etching (impact of atoms/ions), reactive ions and enhanced by RF energy. And also isotropic etching can give rounded profiles and anisotropic etching can yield flat surfaces.

Prototyping and fabrication processes are illustrated in the Fig.3-7. If we draw a vertical line that cuts across both sets of fingers in Fig. 3, we will get the cross section as shown in Fig. 6. Here we only drew two of the 13 fingers in the cross section for the purpose of clearly illustrating the fabrication process. These are the two floating rectangles in final step.

Fig. 3. Vibration detector/sensor used in 3-DOF hybrid robotic manipulator.

Fig. 4. Dimensions of the detector.

Fig. 5. Capacitance based collision detector.

Silicon substrate

Deposition

(a) Deposition of oxide

Silicon substrate

Buried oxide

Deposition

(b) Deposition of Si

Silicon substrate

Buried oxideSi

Patterning

(c) Patterning

Silicon substrate

Buried oxideSi

Fig. 6. Fabrication process.


72

And

Glass substrate

Deposition

(d) Deposition of Epoxy

Glass substrate

Epoxy

Patterning

(e) Patterning

Glass substrate

Epoxy

Bond

(f) Bond

Glass substrate

Epoxy SiBuried oxide

Silicon substrate

Etch in TMAH

(g) Etch

Glass substrate

Epoxy

SiBuried oxide

Silicon substrate

(h) Etch off silicon substrate

Glass substrate

Epoxy

SiBuried oxide

Etch off the oxide

(i) Etch off oxide

Glass substrate

Epoxy Si

Fig. 6. Fabrication process (Continued).

Fabrication method 2:

Glass wafer

Deposition

20 um epoxy

Glass wafer

Deposition

20 um Si20 um epoxy

Glass wafer

Patterning

20 um Si20 um epoxy

Glass wafer

The epoxy is etched in an oxygen plasma (dry

etching), which undercuts the silicon to free the moving parts;

Dry etching

20 um Si20 um epoxy

Glass wafer

Etch epoxy, undercutting Si

Fig. 7. Fabrication process.

Compared with parallel-plate capacitors, the capacitance between two neighboring set of fingers are relatively small. However, one can achieve large capacitance and force by increasing the number of comb pairs. The proposed collision detector can also be used in other machine tools.

7. Conclusions

In this paper, a capacitance based collision detector is proposed and designed for the purpose of


73

detecting any collision between tool and peripheral equipment (e.g. fixture). One side of the finger-like capacitor is attached to the moving platform of the hybrid robotic manipulator and the other side of the finger-like capacitor is attached to the tool. When the tool accidently hits the peripheral equipment, the vibration will make the distance of the capacitor change and therefore trigger the machine to stop. The new design is illustrated and modelled. The capacitance, sensitivity and frequency response are analyzed in details, and finally the fabrication process are presented. The proposed collision detector can also be applied to other machine tools. Future work will build the prototype to test the proposed detector in the real application scenarios. Acknowledgements

The authors would like to gratefully acknowledge the financial support from the Natural Sciences and Engineering Research Council of Canada (NSERC) and Canada Research Chairs program. References [1]. Chang Liu, Foundation of MEMS, Pearson Prentice

Hall, 2006. [2]. L. E. B. Ribeiro, F. Fruett, Analysis of the Planar

Electrode Morphology for Capacitive Chemical Sensors, in Proceedings of the 6th International Conference on Sensor Device Technologies and Applications (SENSORDEVICES’15), Venice, Italy, 23-28 August 2015, pp. 179-182.

[3]. Tsai L. W., Joshi S., Kinematic analysis of 3-DOF position mechanisms for use in hybrid kinematic

machines, Journal of Mechanical Design, Vol. 124, No. 2, 2002, pp. 245-253.

[4]. D. Zhang, Kinetostatic analysis and optimization of parallel and hybrid architecture for machine tools, Ph.D. thesis, Laval University, Canada, 2000.

[5]. Dan Zhang, B. Wei, Design, Analysis and Modelling of a Capacitive-Based Collision Detector for 3-DOF Hybrid Robotic Manipulator, in Proceedings of the 6th International Conference on Sensor Device Technologies and Applications (SENSORDEVICES’15), Venice, Italy, 23-28 August 2015, pp. 31-36.

[6]. http://www.marposs.com/technology.php/eng/ sensors_grinders_monitoring

[7]. Dong-Hoon Kim, Jun-Yeob Song, Suk-Keun Cha, etc., Real-Time Compensation of Chatter Vibration in Machine Tools, International Journal of Intelligent Systems and Applications, Vol. 5, No. 6, 2013, pp. 34-40.

[8]. Chia-Jung Chen, Rong-Shine Lin, Rong-Guey Chang, Efficient Web-Learning Collision Detection Tool on Five-Axis Machine, World Academy of Science, Engineering and Technology, Vol. 7 No. 7, 2013, pp.1053-1057.

[9]. Ebrahim Mohammadiasl, Vibration Detection and Backlash Suppression in Machine Tools, in Proceedings of the IEEE/RSJ International Conference on Intelligent Robots and Systems, St. Louis, USA, October 2009, pp. 972-977.

[10]. T. Rudolf, C. Brecher, F. Possel-Dölken, Contact-based Collision Detection – A New Approach to Avoid Hard Collisions in Machine Tools. A. Donmez, L. Deshayes (Eds.), in Proceedings of the International Conference on Smart Machining Systems, 2007, pp. 1-4.

[11]. Qiaokang Liang, Dan Zhang, Quanjun Song, etc., Design and fabrication of a six-dimensional wrist force/torque sensor based on E-type membranes compared to cross beams, Measurement, Vol. 43, Issue 10, December 2010, pp. 1702-1719.

___________________


http://www.sensorsportal.com/HTML/DIGEST/Submission.htm


74



Numerical Signal Analysis of Thermo-Cyclically Operated MOG Gas Sensor Arrays for Early Identification

of Emissions from Overloaded Electric Cables

1 Rolf Seifert, Hubert B. Keller, 2 Navas Illyaskutty, Jens Knoblauch and Heinz Kohler

1 Institute of Applied Informatics (IAI), Karlsruhe Institute of Technology, Hermann-von-Helmholtz-Platz1, Eggenstein-Leopshafen, D-76344, Germany

2 Institute for Sensor and Information Systems (ISIS), Karlsruhe University of Applied Sciences, Moltkestr. 30, Karlsruhe, D-76133, Germany

1 Tel.: +49-721-6082-4411, fax: +49-721-6082-5702 E-mail: [email protected], [email protected]

Received: 31 August 2015 /Accepted: 5 October 2015 /Published: 30 October 2015 Abstract: A thermo-cyclically operated multi metal oxide gas sensor (MOG) array is introduced together with a novel signal analysis approach (SimSens) for identifying the emissions from overheated isolation cable materials thereby detecting the fires originated in electrical cabinets at early stages. The MOG array can yield specific conductance signatures appropriate to specifically identify gases. The obtained results bear good capability for detection and identification of pyrolysis gas emissions at relatively low sample heating temperatures even before a visible color-change of the polyvinyl chloride (PVC)-isolation material. The dynamic conductance signals were evaluated using SimSens, a numerical analysis tool designed for simultaneous evaluation of conductance profiles. The results show promising pyrolysis gas identification and concentration determination capabilities in relation to the conductance profile shapes of model gases like carbon monoxide (CO) and propene. Copyright © 2015 IFSA Publishing, S. L. Keywords: Oxide gas sensor, Sensor array, Early fire detection, Pyrolysis, Data analysis.

1. Introduction In the current scenario, developing sensor systems

for early detection of fires instigated by overloaded electrical circuits has gained great attention due to increased sensibility for security aspects. Metal oxide gas sensors (MOG) can be used as appropriate candidates for detection of conventional fires and smoke [1]. This type of gas sensor could be utilized for early detection of fires in electrical installations, as pyrolysis of cable materials leads to emission of distinct gas mixtures depending on insulation

material composition [2-3]. Identification of those typical gas mixtures by thermo-cyclic operation of MOG sensor arrays and simultaneous sampling of conductance over time profiles (CTPs) together with numerical analysis of these profiles has been shown in the past to be an elegant and reliable method for detection of fires [3-8].

2. Outline of the SimSens Program

In many applications, a multitude of different gases may occur, which have to be identified and




75

analyzed simultaneously. One application of great interest is the abovementioned detection of fires due to electrical overload of cables at early stages of development. Depending on the various coating materials of the cables a variety of gases or gas mixtures can be emitted. These gases or gas mixtures according to the related coating materials have to be simultaneously identified and analyzed for early detection of such developing risk. The term “simultaneously” in this context means that the measurements are performed with only one sensor system and the instantaneous analysis is performed with only one analysis procedure covering all the different gases or gas mixtures, which may occur at a certain time but not at once.

The calibration and evaluation program SimSens was designed to meet these demands on simultaneous gas identification capability. SimSens is an extension of the well-introduced program ProSens [10-11], which was designed to analyze only one target gas or gas mixture under consideration.

Like ProSens, SimSens consists of a calibration part and an evaluation part. In the calibration part SimSens provides the mathematical calibration models for every gas to be analyzed. Each calibration model consists of functions for determination of the component concentration und of functions for calculating the related so-called theoretical CTP for substance identification. This means, the calibration part of SimSens provides calibration models for every gas or gas mixture under consideration, whereas ProSens calculates only one calibration model. The functions included in the calibration part of SimSens are parametric functions and the parameters are determined by multiple linear regressions of the CTP sample values versus concentration.

These parameters are transferred to the evaluation part of SimSens for the analysis of an unknown gas sample. Based on these parameters and the CTP of an unknown gas sample the evaluation part of SimSens calculates theoretical CTPs for each calibrated gas and compares these CTPs with the measured CTP.

If the measured CTP and one of the theoretical CTPs are close together, i.e., a difference value calculated from the sum of quadratic differences of every sample point of the measured CTP and the theoretical CTP is smaller than a pre-determined decision value, SimSens identifies the unknown gas sample as the related calibrated target gas.

Otherwise SimSens recognizes that the gas sample is none of the calibrated target gases. In case of identification, SimSens calculates the concentration of the gas sample based on the related calibration model.

Using these algorithms of numerical sensor signal analysis, SimSens has the capability to identify and analyze a variety of target gases or gas mixtures. This is due to the fact that in the calibration part of SimSens more than only one calibration model can be determined, namely one calibration model for each gas or gas mixture, which may occur in the

considered application. Furthermore, in the evaluation part of SimSens, not only one gas or gas mixture can be identified, but all gases or gas mixtures, which are calibrated. 3. Experimental Set

The principal sensing element used for pyrolysis gas identification studies is a four-fold sensor array on a 4×4 mm2 alumina chip (Fig. 1), which comprises four micro-dispensed thick-film sensing layers of different SnO2/additive-composites [3]. In operation, the sensor-chip is thermally modulated by applying a steady slope heater-voltage, which results in a periodic, nearly triangular temperature profile between 100 °C and 450 °C, at a cycle time of three minutes (Fig. 2).

Fig. 1. Multi-sensor-array with four different layers dispensed on thin-film Inter-Digital-Electrodes. The chip

is mounted on TO 8 header.

Fig. 2. Triangular heater voltage profile applied to operate the sensor array at thermo-cyclic mode (left ordinate)

and corresponding temperature monitored by IR camera (right ordinate).

Operating MOG sensors thermo-cyclically and sampling of the conductance simultaneously yields gas specific CTPs [3-6], which enable identification


76

of the gas atmosphere from the characteristic CTP shapes.

Pure SnO2, SnO2/2 %ZnO-, SnO2/1 %PtO- and SnO2/1 %PdO-composites were selected systemati-cally from a variety of material combinations by investigating their sensitivity performance. The various material combinations were investigated towards their specific conductance behavior to propene (C3H6), carbon monoxide (CO) and methane (CH4) at different concentrations in synthetic air and 37.5 % (21 °C) relative humidity (r.H.) and were later exposed to pyrolysis gases as described in the following sections. These model gases were particularly selected as they are found to be the main components evolved during the pyrolysis of standard PVC-insulation cable materials. The selected sensing materials show optimal sensitivity, stability and gas identification capabilities towards model gases like CO and propene and to pyrolysis gases. The sensitivity to methane is very low and not further reported.

A detailed description of the sensor fabrication, experimental set up and generation of pyrolysis gases is depicted in refs [4], [6-7]. Pyrolysis experiments were conducted with 3.5 g of PVC isolated litz copper wires (LiY – 0.14 mm2, AWG26, 2A current rating, yellow). The pyrolysis gases were generated by heating the samples in a quartz tube reactor and the evolving gas is carried to the sensor by a constant synthetic air flow. After the reactor, the constant flow is mixed with another adjustable flow of synthetic air in order to get varied gas concentrations. The reactor temperature was increased in a stepwise fashion starting from room temperature up to 170 °C. This fixed temperature was selected after several systematic experiments conducted below and above 170 °C. Evidently, the wire sample showed no visible change up to temperatures of 150 °C and even at 170 °C only a slight change of the sample by shrinking of insulation diameter is observed. At 200 °C discoloring takes place, the sample turns brown.

4. Results and Discussion 4.1. Sensor Response

An overview of the pyrolysis experiment with

sensing pattern of a pure SnO2 layer towards pyrolysis gas at different concentrations is given in Fig. 3. The pyrolysis gas was transported by a constant primary gas flow and diluted by a dilution flow of synthetic air. It is demonstrated that at 170 °C, an almost constant emission from the polymer material is observed over more than five hours. The absolute conductance-values presented here vary greatly depending on experimental parameter values. Several CTPs were recorded at a reactor temperature of 170 °C, while changing the dilution level to set defined relative concentrations. The CTPs at each concentration are highly

reproducible at the repeated cycling by showing similar baseline and peak conductance values (Fig. 3, inset).

Fig. 3. Measurement sequence of a pure SnO2 layer exposed with pyrolysis gas of a heated PVC-coated wire with temperature profile and gas flows. The numbers relate the pyrolysis gas concentrations in percent. At each concentration, the heater cycles 10 times. The inset and the oval in green show the reproducibility of the profiles with stable baseline and maximum conductance at repeated temperature cycles.

The CTPs measured by different sensitive materials to model gases (CO and propene) and pyrolysis gases at different concentrations are visualized in Fig. 7 for comparison. In the very most cases they show very specific features with respect to gas identification capabilities.

The thermo-cyclically driven array of four different gas sensitive SnO2/additive-composites exhibits four different profiles to an exposed gas, which in fact can enhance the gas identification capability compared to conventional mono-sensors operated at isothermal mode. For example, in consideration of propene and CO response, the different sensing materials show completely different CTP shapes (Fig. 7, 1st and 2nd column). In case of pyrolysis gas, although the profile shapes of all materials look similar with two distinctive conductance peaks at the temperature rising and dropping regions, the conductance peak positions are dependent on temperature. Also, by the CTP-features of each sensor the pyrolysis gases can be clearly distinguished from the model gases (Fig. 4, rows).

4.2. Data Analysis

To demonstrate the performance of the sensor system and the procedure SimSens, it was assumed that besides pyrolysis gases propene and CO as interfering atmospheres may occur. The goal is to distinguish between the different gas atmospheres, this means to recognize whether an unknown gas sample is one of the considered gases. In case of identification, additionally, the concentration of the gas has to be determined. The analysis model is


77

based on the CTP-data of all four gas sensitive layers of the chip (Fig. 4), which are simply

linked together to one “extended CTP” (ECTP) as visualized in Figs. 5-7.

Fig. 4. Matrix representation of the comprehensive sensor responses: Conductance over time profiles (CTP) of the four sensing layers (in columns) when exposed to various concentration of propene, CO and pyrolysis gas (in rows). The propene and CO concentrations are given in ppm and the pyro gas concentration is given in percentages in relation to the dilution flow used, (see caption of Fig. 3 for more details).

For the calibration model of pyrolysis gas, the

CTPs of pyrolysis gases at 30 %, 40 % and 60 % relative gas concentrations (Fig. 3 and Fig. 4) were used and for the calibration model of propene and CO, the CTPs were taken at the corresponding concentrations 125 ppm, 250 ppm and 500 ppm.

To test the capability for identifying the gas component/mixture and to determine the associated concentrations, three gas samples were measured and evaluated by SimSens. These are pyrolysis gas at 50 % relative concentration, propene and CO at 375 ppm.

The following figures show that the pyrolysis gas sample at 50 % relative concentration could be

clearly recognized as a pyrolysis gas. This is due to the fact that in Fig. 5 the difference between measured CTP and calculated CTP on the basis of the pyrolysis calibration model is very small. In Fig. 6 and in Fig. 7, the differences between measured CTP and calculated CTP based on the calibration model of propene, respectively CO, are much higher. Analogous results were obtained when evaluating the propene sample and the CO sample.

Of course, the decision of identification is not based on a visual impression. For substance identification, a difference value D is calculated as a measure of the difference between measured CTP and calculated CTP.


78

Fig. 5. Measured ECTP of pyrolysis sample at 50 % dilution and associated calculated ECTP based on the

pyrolysis calibration model.

Fig. 6. Measured ECTP of pyrolysis sample at 50 % dilution and associated calculated CTP based on the

propene calibration model.

Fig. 7. Measured ECTP of pyrolysis sample at 50 % dilution and associated calculated CTP based on the CO

calibration model.

D is the relative sum of quadratic differences of every sample point of the measured CTP and the theoretical CTP. If the difference value is smaller than a so-called decision threshold value eps, the sample under consideration is identified as the associated gas. In this investigation, the decision threshold value was set to eps=4.e-01.

In Table 1, only the red bold numbers are smaller than the decision threshold value. Therefore, the pyrolysis sample could be identified as a pyrolysis gas, the propene sample as a propene gas and the CO sample as a CO gas.

After identifying the gas samples, SimSens calculates the associated gas concentrations. The evaluation results are given in Table 2.

Table 1. Difference values between measured CTPs and theoretical CTPs.

Pyrolysis Model

Propene Model

CO Model

Pyrolysis 50 %

2.2e-03 6.6e+02 8.7e+06

Propene 375 ppm

9.9e+01 1.3e-02 1.5e+05

CO 375 ppm

1.0e+00 4.3e+02 1.7e-01

Table 2. Comparison of dosed and analyzed concentration values and relative deviation. About the

meaning of concentrations in % see Fig. 3.

Sample Dosed Conc.

Calculated Conc.

Relative Error

Pyrolysis 50 %

50 % 47,6 % 4,7 %

Propene 375 ppm

375 ppm 388,8 ppm 3,7 %

CO 375 ppm

375 ppm 379,6 ppm 1,2 %

The concentration values estimated by the calibration model deviate from the experimentally adjusted values by less than 5 % in all cases.

5. Conclusions and Outlook The thermo-cyclic operation of MOG array

combined with simultaneous numerical analysis of the CTPs has been shown to be an elegant way for identification of gas mixtures. In this study, the pyrolysis gases emitted by heated PVC-based insulation materials are recognized by well reproducible CTPs, even at temperatures where no color change of the sample material could be observed. These CTPs are well distinguishable from those obtained for two model gases, CO and propene. The results look promising considering the aim of early fire detection with high sensitivity.


79

The acquired CTPs were numerically analyzed employing the SimSens algorithm and the results showed very good identification capabilities and concentration estimation accuracy, which can lead to better incident identification and a very sensitive, more robust detection with a low false alarm rate.

In case of field applications, gas detection experiments for fire or any other dangerous gases using MOG sensors may have to be conducted in a real environment, where indefinite and unspecific but typical gas mixtures are present. In such cases, the numerical analysis of data using SimSens of a thermo-cyclically operated MOG-array will be of great advantage because the CTPs of four sensitive layers provide numerous gas specific features. To enable these advanced field applications, a microprocessor-based sensor system has to be devised, featuring the necessary means. Acknowledgements

The authors would like to thank German Federal Ministry of Education and Research (BMBF) for financial support of this work (Project Nr. 16N12262).

References [1]. D. Gutmacher, C. Foelmli, W. Vollenweider,

U. Hoefer, J. Wöllenstein, Comparison of gas sensor technologies for fire gas detection, Procedia Engineering, Vol. 25, 2011, pp. 1121-1124.

[2]. G. Korotchenkov, V. Brynzari, S. Dmitriev, SnO2 thin film gas sensors for fire-alarm systems, Sensors and Actuators B: Chemical, Vol. 54, Issue 3, 1999, pp. 191-196.

[3]. R. Seifert, H. B. Keller, N. Illyaskutty, J. Knoblauch, H. Kohler, Early Detection of Emissions Preceding Fires from Overloaded Electric Cables: Approach

with Thermo-Cyclically Operated MOG Sensor Arrays and Numerical Signal Analysis, in Proceedings of the Sixth International Conference on Sensor Device Technologies and Applications (SENSORDEVICES ‘15), pp. 91-96.

[4]. N. Illyaskutty, J. Knoblauch, M. Schwotzer, H. Kohler, Thermally modulated multi sensor arrays of SnO2/additive/electrode combinations for gas identification, Sensors and Actuators B: Chemical, 217, 2015, pp. 2-12.

[5]. K. Frank, V. Magapu, V. Schindler, H. Kohler, H. B. Keller, R. Seifert, Chemical analysis with tin oxide gas sensors: choice of additives, method of operation and analysis of numerical signal, Sensor Letters, Vol. 6, No. 6, 2008, pp. 908-911.

[6]. J. Knoblauch, N. Illyaskutty, H. Kohler, Early detection of fires in electrical installations by thermally modulated SnO2/additive-multi sensor arrays, Sensors and Actuators B: Chemical, 217, 2015, pp. 36-40.

[7]. J. Knoblauch, et al., Sensor system applying thermally modulated MOG for early detection of fires in electrical cabinets, in Proceedings der 15. Internationale Tagung über Automatische Brandentdeckung (AUBE 2014), Duisburg, Vol. II, 14-16 October 2014, pp. 105-112.

[8]. H. Kohler, R. Becker, N. Link, J. Röber, Apparatus and process for identifying and/or determining the concentration of at least one gas component, Patent No. EP 0 829 718 B1, 2004.

[9]. H. B. Keller, R. Seifert, H. Kohler, SimSens – a New Mathematical Procedure for Simultaneous Analysis of Gases with Resistive Gas Sensors, Sensors & Actuators B: Chemical, 2015, pp. 203-207.

[10]. R. Seifert, H. B. Keller, K. Frank, H. Kohler, ProSens - an Efficient Mathematical Procedure for Calibration and Evaluation of Tin Oxide Gas Sensor Data, Sensor Letters, Vol. 9, No. 1, 2011, pp. 7-10.

[11]. R. Seifert, H. B. Keller, Verfahren zur Klassifikation und zur Bestimmung der Einzelkonzentrationen eines Stoffgemisches (Procedure for Classification and Determination of the Concentration of the Components of a Gas Mixture), Patentschrift DE10 2004 057 350.

___________________



80



Analysis of the Planar Electrode Morphology Applied to Zeolite Based Chemical Sensors

1 Luiz Eduardo BENTO RIBEIRO, 2 Glaucio Pedro de ALCÂNTARA,

2 Cid Marcos GONÇALVES ANDRADE, 1 Fabiano FRUETT 1 School of Electrical and Computer Engineering of State University of Campinas, Campinas, Brazil

2 Department of Chemical Engineering of State University of Maringá, Paraná, Brazil E-mail: [email protected], [email protected]

Received: 31 August 2015 /Accepted: 5 October 2015 /Published: 30 October 2015 Abstract: In order to improve the zeolite chemical sensors sensibility, three different electrode structures are compared in this work: conventional interdigitated electrodes (IDE), serpentine electrodes (SRE) and ring-shaped electrodes (RSE). Simulation results and experimental characterization of these electrodes showed that ring-shaped electrodes have a slight capacitance increase per unit of area. When used as a zeolite chemical sensor, the ring-shaped electrodes prove to be more suitable since they take advantage of a better usage of the drop distribution and a better capacitance per area ratio. Copyright © 2015 IFSA Publishing, S. L. Keywords: Interdigitated electrodes, Ring-shaped electrodes, Serpentine electrodes, Chemical sensors, Electrode structure, Interdigital electrodes, Zeolite. 1. Introduction

IDE as capacitive microstructures have been widely used in electronics applications such as surface acoustic wave devices [1-2], thin-film acoustic electronic transducers [3], tunable devices [4], dielectric spectroscopy [5], dielectric studies on thin films [6], humidity and chemical sensors [7-9], etc. They possess interesting features, such as signal strength control by changing its dimensions, multiple physical effects in the same structure, simplified modeling in two dimensions when the aspect ratio of the electrode length to the space wavelength IDE is large, and can be used in a wide range of frequencies [10]. Moreover, it can be manufactured using inert substrates with multiple materials with different fabrication processes, or even microfluidic compatible. Capacitive microstructures used as chemical sensor, typically have one sensitive layer deposited over the electrodes. Polymers have been

used for organic vapor sensing because they exhibit rapid reversible vapor sorption and are easy to apply as thin or thick films by a variety of techniques [10]. The polymer layer can be chosen according to its affinity to a particular molecule or set of molecules one wishes to detect. If several sensors with different polymer layers are used to make a sensor array it is then possible to evaluate complex organic vapor samples. These sensor arrays can be part of a so-called electronic nose. Another possibility is the use of flexible substrates such as sensing element. As an example the use of polyimide [11] and plastic foil [12].

Interdigitated electrodes analytical characteri-zation have been received many efforts in order to improve their capacitance by exploring their geometrical parameters [13]. Igreja et al developed a theoretical model of capacitance for this structure [14]. These capacitors have also been simulated using different tools. They have typically been adopted as a




81

sensor because of the low-power consumption of capacitive transduction mechanism and being compact with a large contact area and relative ease of manufacture. Some authors have analyzed other morphologies such as serpentine, spiral and concentric rings electrodes, in order to improve design performance. Moreover, aside from the different morphologies, several strategies have been employed to increase and improve its ability as a sensor, for example the deposition of a more sensitive layer on the electrodes. The sensor can also have its selectivity improved by the deposition of compounds such as zeolite for detection of air humidity or gases [15].

In this work, we study different geometries of electrodes instead of the materials used to maximize capacitance per area. We can thus improve the sensitivity of capacitive sensors by increasing the total capacitance of capacitive microstructures. Here, we compare the conventional IDE, the serpentine electrodes (SRE) and ring-shaped electrodes (RSE). These electrodes are compared in detail as generic capacitive transducers by numerical simulations. In these simulations, the geometric parameters that most influence on the total capacitance are shown. We measure the capacitance of these thin-film electrodes

made of titanium and gold on a glass substrate. After that, we compare the experimental results with the theoretical analysis, including simulations. Finally, we compared the influence of such geometries under a LTA structured zeolite layer.

In Section 2, the different electrodes designs and their geometrical parameters are shown. They are the main data to the numerical simulation, which is explained in the Section 3. The fabrication process is described in Section 4, while the zeolite properties and deposition method are described in Section 5. Subsequently, the values of electrical and geometrical characterization are shown in Section 6. The results are compared and discussed in Section 7. Finally, the conclusion is presented in Section 8.

2. Methods and Materials

The layouts of some interdigitated electrode pairs are shown in Fig. 1. Although they were designed with the same area, each structure has a particular capacitance. Therefore, they have different sensitivities as capacitive transducer only because of their different geometric structure.

Fig. 1. Ring-shape electrode (a); serpentine electrode (b); and interdigitated electrode (c) structures and used parameters. We simulate the capacitive structures of IDE,

SRE and RSE with the same area, same substrate and the same top layer, calculating and comparing the capacitance of each structure using the multiphysics numerical simulator: COMSOL Multiphysics (Comsol, Inc., Stockholm, Sweden). This software, based on partial differential equations with the finite element method has been used in the literature to calculate distribution of potential field in similar structures. Fig. 2 depicts the simulated structure showing the 3D multislice view of the electrical potential distribution around RSE with 20 fingers.

Multiple simulations were performed to compare dimensions that are relevant to increase the difference between the capacitances. The main geometrical parameters analyzed are the electrode length, the gap between the electrodes, the electrode width (always

kept as same value as gap), the thickness of electrodes and the number of fingers.

Fig. 2. Distribution of potential field in a 3D RSE structure.


82

The electrical properties of the substrate and the top layer are also included in the simulator. Because of the differences between the vertical and horizontal dimensions (millimeters to micrometers) critical to reduce the time simulation, we have applied the extrapolation method, presented in the Rivadeneyra work [16]. The thickness has been set to 20 % of electrode width.

3. Numerical Simulation

We have performed different numerical simulations comparing the calculated DC capacitance of IDE, SRE and RSE structures always with the same distance between digits, surface area and same materials and manufacturing process. There are multiple geometrical factors that can be varied in the simulations, but for the sake of clarity, we focus on some of them, keeping fixed the rest. Since the goal is maintaining reduced the surface area of electrodes, we define different widths and distances between digits of the structures given that the lowest safe distance to our manufacturing process is 10 µm. Therefore, we have used the number, length and width of fingers and the thickness of the deposited metal film as simulation parameters. Table 1 is a summary of these geometric parameters used for simulation. Remember that fabricated structures for each type of electrode, SRE, IDE and RSE, was made with three distances between digits (10 µm, 20 µm and 50 µm) but the number of electrodes and the thickness of the metal thin film was kept constant.

Table 1. Geometrical Parameters of Planar Electrodes Pairs.

The first important result was that the capacitance to RSE was the highest in all comparisons. For example, simulating the capacitor with w = 10 µm, t = 2 µm and 20 electrodes we obtain 2.47 pF for RSE while the capacitance for the SRE was 2.45 pF and the capacitance for IDE was 2.44 pF. This means

an increase of capacitance of 30 fF between the RSE and the IDE. Whereas electrodes are used as capacitive transducers, the fact of the geometrical factor RSE be higher than in other structures means that the sensitivity of the sensor will always be larger using the same area. Furthermore, the capacitance and the increase in sensitivity will increase proportionally. After that, we calculate the numerical capacitances in order to improve the performance of RSE.

Influence of the width of the electrodes (10 to 50 μm) with aspect ratio of the electrode/thickness of the thin film of 5/1 was evaluated for the RSE, SRE and IDE with 20 digits. The results showed a slope of 161.1 fF for each micrometer added to the width of the electrodes to the RSE, a slope of 160.7 fF/µm for the SRE and 159.3 fF/µm for the IDE. Remembering that the surface area of each structure was kept the same for each electrode width.

Electrode length contribution (400 to 800 µm) was been evaluated for IDE and SRE electrode with 20 fingers and 10 µm of finger width. Results showed the slope of 4.99 fF/µm for the SRE and 4.74 fF for the IDE.

We also have performed the contribution of finger numbers (n) in each structure. The results are presented in Fig. 3, as this contribution is nonlinear.

Fig. 3. Capacitance vs. number of fingers of RSE, SRE and IDE.

4. Fabrication Process The electrodes fabrication started with the

production of the electrode masks where the IDE, SRE and RSE were designed using high-resolution direct writing photolithography with a laser beam. In the next step, a photoresist layer was deposited onto a square optical glass plate (60 mm side Kodak 1A High Resolution Glass) and patterned by conventional ultraviolet light (UV) photolithographic method following the electrodes masks. The UV exposures were carried out in a MJB-3 UV300 contact mask aligner (Karl-Suss, Garching, Germany). Titanium-gold (TiAu) thin films, deposited with a Leybold Univex 300 ebeam

Para-meter

s

Electrode type Descripti-on RSE SRE IDE

w 10-50 µm

10-50 µm

10-50 µm

Finger width

g 10-50 µm

10-50 µm

10-50 µm

Gap between fingers

t 2-10 µm 2-10 µm 2-10 µm Metal film thickness

l 600-800 µm

600-800 µm

- Structure length

d 600-800 µm

600-800 µm

- Structure width

r - - 0.6-3.02 mm

Electrode external radium

n 20-50 20-50 20-50 Number of electrodes


83

evaporator (Cologne, Germany), were used as electrode materials. After thin-film depositions, by the lift-off technic, the devices were immersed in acetone to remove the photoresist layer and excess of metal, leaving the patterned electrodes on the glass surface [7].

5. Characterization

For the experimental characterization the nominal capacitance target was 2 pF, taking into account the minimum distance between electrodes of 10 µm.

Structures with electrode width of 50µm have a total area of 28.26 mm², and their dimensions are g = 50 µm, l = 5.95 mm, d = 4.75 mm and r = 3.02 mm. The area is 4.52 mm² for structures with w = 20 µm, g = 20 µm, l = 2.38 mm, d = 1.90 mm and r = 1.21 mm. Structures with w = 10 µm, minimum secure dimension to our fabrication process, the area is 1.13 mm² (g = 10 µm, l = 1.19 mm, d = 0. 95 mm and r = 0.60 mm).

The geometric inspection of the fabricated structures was performed using the Olympus BX60 microscope as shown in Fig. 4.

Fig. 4. Fabricated RSE (a); SRE (b); and IDE (c) structures.

Electrical characterization has been carried out by measuring their capacitance and parallel resistance using the four-wire measurement with a HP 4284A impedance analyzer (Agilent Tech., Santa Clara, CA, USA). The applied excitation voltage was VAC = 1 V and VDC = 0 V. The frequency sweep of analysis was from 100 Hz to 1 MHz. The four-point method was used to minimize the contribution of stray capacitances. 6. Zeolite Deposition

The zeolite layers were prepared using micro-

drop deposition. A top view of zeolite over electrodes can be observed in the Fig. 5. The process first step involves suspension of the crystals in alcohol. We placed 0.1 g of the zeolite crystals with 10 ml of isopropyl alcohol into a vessel. After that, the vessel was placed into a sonic bath for 10 minutes. The bath make the suspension homogenous, allowing agglomerations to sink to the bottom of the vessel leaving single crystals dispersed.

Using a micro-syringe, a single droplet (10 µl) of the solution was carefully deposited on to the electrodes under a low magnification microscope. After 10 minutes, the isopropyl alcohol evaporated and the crystals adhered to the surface by electrostatic forces. The evaporation of this droplet takes place leaving a zeolite deposit on the electrodes. The process was repeated until around 100 µl was achieved (usually repeated 10 times) [15].

The used LTA structured zeolite (zeolite A) is built up of two types of arrangements; 8 tetrahedra in

a cubic shape and 24 tetrahedra in an octahedron shape. The framework is formed by the connection of every cube corner to the octahedrons creating caged cavities in-between. Zeolite A is a low-silica based zeolite and therefore has a high cation concentration to compensate for the reduction in charge [17]. The Si/Al ratio is around 1 making this type of zeolite a fairly hydrophilic material. The pore size is around 4 angstroms, the cage diameter is 11.5 angstroms and the pore volume is around 0.30 cc/g [18].

Fig. 5. Microstructure during the zeolite deposition process.

7. Experimental Results

Measured and the simulated values of the capacitances are in agreement, as shown in Table 2.


84

The measurements were carried out in room conditions (50 % RH and 25ºC) at 100 kHz.

Table 2. Numerical vs. Experimental Capacitance.

Finger width (µm)

Numerical simulation (pF) RSE SRE IDE

10 2.4667 2.4446 2.4362 20 4.0906 4.0653 4.0345 50 8.9122 8.8723 8.8091

Experimental measurement (pF) RSE SRE IDE

10 2.2332 2.2141 2.1917 20 4.2949 4.2667 4.2412 50 10.5385 9.9979 9.7105

Capacitance measured values have a maximum of 12 % discrepancy compared to the simulated results. This difference is due to the capacitance of contacts pads that were not included in the simulation and the variability of the fabrication process.

After the zeolite deposition, we also measured the percentage of the capacitance increment when the structures are with and without the zeolite layer. For the ring-shape electrode the capacitance with zeolite layer incremented 139 % (2.39 times) of that without zeolite. In case of the interdigitated electrodes, the capacitance with zeolite layer incremented 44 % (1.44 times) in comparison of that without zeolite.

This preliminary study indicates that the sensibility of a zeolite chemical sensor can be also improved by changing the capacitive microstructure. However, when in contact with the liquid or gas under analysis, the total impedance must be modeled to comprise effects such as double layer capacitance and interfacial polarization phenomenon. 8. Conclusion

A comparison between the capacitances under different geometric parameters of RSE, SRE and the conventional IDE was shown on this work. Numerical simulations of the capacitance have been carried out to calculate the differences between them due exclusively to their geometrical characteristics at constant area. In these conditions, we have shown a slight capacitance increase for the ring-shaped electrodes against the serpentine electrodes and the conventional interdigitated electrodes.

We have validated the numerical results by experimental characterization of ring-shaped, serpentine and interdigitated structures capacitance. Experimental results verified the capacitive differences between the three structures with same area, for 10 µm, 20 µm and 50 µm of finger width. Moreover, ring-shape electrode presents a geometrical morphology that allows the better usage of its area. It is a promising base electrode mainly when used as sensor in application that involves

dripping of substances under analysis or dripping of selective/sensitive substances like zeolite.

While the capacitance of the interdigitated electrodes with zeolite layer is 1.44 times greater, the ring-shape electrodes is 2.39 times greater. This preliminary study with zeolite deposition over the electrodes show us that the total capacitance can be further enhanced by the electrode geometry.

Acknowledgements

The authors acknowledge the Center for Semiconductor Components (CCS), the Multi-User Laboratory of IFGW (LAMULT), the Device Research Laboratory (LPD) and the Brazilian Synchrotron Light Laboratory (LNLS). Financial support for this project was provided by the National Council for Scientific and Technological Development (CNPq) and the São Paulo State Research Support Foundation (FAPESP) inside the National Institute for Science and Technology of Micro and Nanoelectronic Systems (INCT NAMITEC) project.

References [1]. M. I. Rocha-Gaso, C. March-Iborra, Á. Montoya-

Baides, A. Arnau-Vives, Surface generated acoustic wave biosensors for the detection of pathogens: A review, Sensors, Vol. 9, No. 7, 2009, pp. 5740-5769.

[2]. N. A. Ramli, A. N. Nordin, Design and modeling of MEMS SAW resonator on Lithium Niobate, in Proceedings of the 4th International Conference on Mechatronics (ICOM’11), 2011, pp. 1-4.

[3]. M. W. Kim, Y. H. Song, J. B. Yoon, Modeling, fabrication and demonstration of a rib-type cantilever switch with an extended gate electrode, Journal of Micromechanics and Microengineering, Vol. 21, 2011, 115009.

[4]. R. Mahameed, A. M. El-Tanani, G. M. Rebeiz, A zipper RF MEMS tunable capacitor with interdigitated RF and actuation electrodes, Journal of Micromechanics and Microengineering, Vol. 20, No. 3, 2010, 035014.

[5]. Heileman, Khalil, Jamal Daoud, Maryam Tabrizian, Dielectric spectroscopy as a viable biosensing tool for cell and tissue characterization and analysis, Biosensors and Bioelectronics, Vol. 49, 2013, pp. 348-359.

[6]. Z. Chen, A. Sepúlveda, M. D. Ediger, R. Richert, Dielectric spectroscopy of thin films by dual-channel impedance measurements on differential interdigitated electrode arrays, The European Physical Journal B - Condensed Matter and Complex Systems, Vol. 85, No. 8, 2012, pp. 1-5.

[7]. L. E. B. Ribeiro, M. H. Piazzetta, A. L. Gobbi, J. S. Costa, J. A. F. da Silva, and F. Fruett, Fabrication and Characterization of an Impedance Micro-Bridge for Lab-on-a-Chip, ECS Transactions, Vol. 31, No. 1, 2010, pp. 155-163.

[8]. Y. Kim, B. Jung, H. Lee, H. Kim, K. Lee, H. Park, Capacitive humidity sensor design based on anodic aluminum oxide, Sensors and Actuators B: Chemical, Vol. 141, No. 2, 2009, pp. 441-446.


85

[9]. A. Rivadeneyra, J. Fernández-Salmerón, J. Banqueri, J. A. López-Villanueva, L. F. Capitan-Vallvey, A. J. Palma, A novel electrode structure compared with interdigitated electrodes as capacitive sensor, Sensors and Actuators B: Chemical, Vol. 204, 2014, pp. 552-560.

[10]. J. W. Grate, G. C. Frye, Acoustic wave sensors, Sensors, Vol. 2, Issue 1, 1996, pp. 37-83.

[11]. J. Virtanen, L. Ukkonen, T. Bjorninen, L. Sydanheimo, Printed humidity sensor for UHF RFID systems, in Proceedings of the IEEE Sensors Applications Symposium (SAS’10), 2010, pp. 269-272.

[12]. D. Briand, A. Oprea, J. Courbat, N. Bârsan, Making environmental sensors on plastic foil, Materials Today, Vol. 14, No. 9, 2011, pp. 416-423.

[13]. L. E. B. Ribeiro, F. Fruett, Analysis of the Planar Electrode Morphology for Capacitive Chemical Sensors, in Proceedings of the 6th International Conference on Sensor Device Technologies and Applications Analysis (SENSORDEVICES’15), Venice, Italy, 23-28 August 2015, pp. 179-182.

[14]. R. Igreja, C. J. Dias, Analytical evaluation of the interdigital electrodes capacitance for a multi-layered structure, Sensors and Actuators A: Physical, Vol. 112, No. 2-3, 2004, pp. 291-301.

[15]. M. Urbriztondo, I. Pellejero, A. Rodriguez, M. P. Pina, J. Santamaria, Zeolite-coated interdigital capacitors for humidity sensing, Sensors and Actuators B: Chemical, Vol. 157, Issue 1, 2011, pp. 450-459.

[16]. A. Rivadeneyra, J. Fernández-Salmerón, M. Agudo, J. A. López-Villanueva, L. F. Capitan-Vallvey, A. J. Palma, Design and characterization of a low thermal drift capacitive humidity sensor by inkjet-printing, Sensors and Actuators B: Chemical, Vol. 195, 2014, pp. 123-131.

[17]. Scott M. Auerbach, Kathleen A. Carrado, Prabir K. Dutta, Handbook of Zeolite Science and Technology, Marcel Dekker Inc., New York, Basel, 2003.

[18]. D. W. Breack, Zeolite molecular sieves-structure, chemistry and use, Robert E. Krieger Publishing Company, Malabar, Florida, 1984.

___________________


http://www.sensorsportal.com/HTML/BOOKSTORE/MEMS_Pressure_Sensors.htm


86



An Empirical Study for Quantification of Carcinogenic Formaldehyde by Integrating a Probabilistic Framework

with Spike Latency Patterns in an Electronic Nose

1 Muhammad HASSAN, 1, 2 Amine BERMAK, 3 Amine Ait Si ALI and 3 Abbes AMIRA

1 School of Engineering, Hong Kong University of Science and Technology, Kowloon, Hong Kong 2 College of Science and Engineering, Hamad bin Khalifa University, Education City, Doha, Qatar

3 College of Engineering, Qatar University, Doha, Qatar 1 Tel.: 00852 23588526


Received: 31 August 2015 /Accepted: 5 October 2015 /Published: 30 October 2015 Abstract: Recently, exposure to formaldehyde has appeared as a major concern since it has been listed as a human carcinogen. Conventional methods for its long-term monitoring are not feasible due to their high operational cost, long analysis time and the requirement of specialized equipment and staff. In this paper, we develop an electronic nose, containing an array of commercially available low cost Figaro gas sensors, to support autonomous and long-term monitoring of formaldehyde. Hardware friendly gas quantification without requiring any manual tuning of parameters is the major challenge with the electronic nose. We handle this challenge by treating it as a classification problem because data acquisition at continuously varying concentrations may incur large expense and a great deal of time. Instead, twenty different concentrations of formaldehyde with 0.25 ppm increment step in the target range between 0.25 to 5 ppm, spanning commonly found formaldehyde levels in indoor and outdoor environments, are input to obtain its signatures in order to quantify/classify its levels within this target range. A computationally efficient bio-inspired spike latency coding scheme, in which spike latencies corresponding to sensitivity patterns of the sensors in the array shift with the change in concentration, is targeted for this purpose. However, stochastic variability in the spike latency patterns, corresponding to repeated exposure to the same formaldehyde concentration level, is observed. We target two Bayesian inference methods, namely multivariate Bayesian and naive Bayes, to express the uncertainty about the spike latency patterns in terms of a probability encoding framework. These methods do not require any manual tuning of parameters, in contrast to other state of the art methods. A best performance of 95.75 % is achieved with the naive Bayes method on the experimentally obtained data set of formaldehyde. Copyright © 2015 IFSA Publishing, S. L. Keywords: Carcinogenic formaldehyde, Sensor array, Spike latency pattern, Bayesian inference. 1. Introduction

Formaldehyde (CH2O) is one of the most ubiquitous and reactive aldehydes in the environment. It is a colorless and strong-smelling chemical, which

is widely used in building and furniture construction materials [1]. Building residents may be exposed to CH2O gas when it is emitted from materials containing this chemical upon thermal or chemical decomposition. Unvented fuel burning appliances and




87

tobacco smoke may also cause CH2O inhalation. One scientific study reveals that overall CH2O emission from different building sources may exceed 2 parts per million (ppm) [2]. Short-term exposure to CH2O may cause skin, nasal, throat, and eye irritation. The United States Occupational Safety and Health Administration (OSHA) has set its short-term exposure limit (15-30 minutes) at 2 ppm and permissible exposure limit (up to 8 hours) to 0.75 ppm [2]. Precautionary measures are required for healthy living when the CH2O concentration reaches one half of the permissible exposure limit.

Formaldehyde received great attention in 1980, when its carcinogenicity was reported in rats and mice after its long-term inhalation [3]. Since that time, its carcinogenicity has also been studied in humans. In 2004, the International Agency for Research on Cancer (IARC) classified CH2O as a human carcinogen based on sufficient evidence that its long-term exposure may cause nasopharyngeal cancer and leukemia in humans [4]. A recent study reported increasing concentration levels of CH2O in the indoor environment of urban areas [5]. This alarming situation highlights the importance of formaldehyde monitoring with a low cost and robust solution on a long-term basis for healthy living. Unfortunately, traditional methods of gas chromatography and spectro-fluorimetry [5-7] cannot be used for the long-term monitoring of formaldehyde because specialized equipment and staff are required for the analysis of air samples collected from the area being monitored. Moreover, the cost and analysis time associated with these methods is very high.

Electronic nose systems, containing an array of gas sensors, emerged as a successful platform for the fast identification of gases in the last two decades, and they are targeted at many applications like food quality checking [8], diseases diagnosis [9], bacteria identification [10], environmental monitoring [11], beverages classification [12], paper quality inspection [13] and identification of health endangering indoor gases [14]. In this paper, we develop an electronic nose system, containing an array of six commercially available Figaro gas sensors, to quantify CH2O concentration.

Motivated by the recent experimental findings in the field of neuroscience which report a logarithmic relationship between odor concentration and the spike latency of mitral cells [15], a logarithmic time domain scheme has been previously presented for gas classification by translating a sensor array response into a spike latency pattern [16]. Hardware friendly rank order based classifiers have then been developed for gas identification by using this technique [16-18]. In these classifiers, the temporal sequence of spikes is utilized to distinguish gases. We adopt this scheme to retrieve concentration information by utilizing the shifts in the spike latencies with the change in concentration. However, there is no straightforward relationship between the shift in the spike latency and the formaldehyde concentration as there is in the rank order based classifiers, where the change in relative

times between spikes does not change the classification performance as long as their temporal order is not changed. Fig. 1 demonstrates the basic concept of this approach. The figure shows that the spike latency of each sensor is changed with the change in concentration but the arbitrary temporal sequence remains fixed.

Fig. 1. Demonstration of change in spike latency of each

sensor with the change in concentration.

Generally, gas sensors exhibit randomness in their responses due to inherent issues in this technology (such as drift, aging, change in operating conditions etc.); and as a result, stochastic variability is observed in the latency patterns, which makes gas quantification more challenging.

In this paper, we formulate this problem as a classification problem because the use of discrete values of concentration, spanning regular intervals within the target range of interest, is the most feasible option for obtaining an experimental data set in order to save time and cost. Data corresponding to each discrete value of concentration is treated as single class data.

A Bayesian inference approach [19] is targeted to deal with the randomness in the latency patterns because this provides an analytical solution and no manual tuning of parameters is required. This approach has been successfully used in neuroscience to build computational theories for perception and action [20]. There are two major steps in this approach. The first step is to learn the probability encoding model or the tuning curve for the spike latency patterns at each predefined concentration value of CH2O from the experimental data obtained through the sensor array.

The second step is to use a Bayesian decoding model to estimate the formaldehyde concentration for a newly test latency pattern by using the learned probability encoding model. Multivariate Bayesian or quadratic discriminant analysis (QDA) and a naive Bayes classifier (NBC) are typically used under this framework when there are no shared parameters between the classes. Multivariate Bayesian assumes a relationship between features, while the naive Bayes classifier considers independence among features. The performance of both of these methods, along with those of other state of the art approaches, is evaluated by acquiring CH2O data at twenty different concentration values, spanning from 0.25 to 5 ppm.


88

The paper is organized as follows. Section 2 explains the probabilistic inference approach for CH2O concentration estimation. Next, Section 3 describes the experimental setup for data acquisition and evaluates the performance of the Bayesian inference methods along with other state of the art methods. Finally, the conclusion is drafted in Section 4. 2. Probabilistic Framework

A logarithmic time-domain encoding scheme has

been used for gas identification in rank order based classifiers [16-18]. In these classifiers, the spike latency li (expressed in arbitrary units) of the i-th sensor corresponding to a target gas is represented as = , (1)

where xi denotes the sensitivity of the i-th sensor and ai is a sensor dependent parameter which is extracted through linear regression between the average log sensitivity of the sensors across the array as an explanatory variable and the sensitivity of the i-th sensor as an output variable. The resultant spike latency patterns carry information about the gas identity and its concentration. In rank order based classifiers [16-18], the temporal sequence of spikes is used for gas identification. In this paper, we utilize the change in spike latency of each sensor to estimate the concentration level of CH2O.

The potential challenge with this scheme is that gas sensors usually exhibit randomness in their responses, which results in stochastic variability in the latency patterns. We use a probabilistic inference approach to retrieve concentration information from the random latency patterns.

The main objective of using probabilistic inference is to find the most probable concentration class or level of the newly test latency pattern by learning the distribution of latency patterns corresponding to each concentration level from the available measurements taken with the electronic nose. Let us consider the following notations for this probabilistic inference problem: Suppose we have a set of concentrations c=c1, c1, . . ., cn, and the experimentally obtained latency pattern l=l1, l2, . . ., ld, where d denotes the total number of sensors and n represents the total number of concentration classes or levels. For gas quantification algorithms, li corresponds to one single feature and l to a d-dimensional feature vector.

Probabilistic inference is a two-step process [20]. In the first step, we learn a model fitting that captures the mapping from l to c from the available sensor array measurements. In the second step, we use Bayesian decoding to estimate the concentration level cj for the newly observed test latency pattern.

In order to learn the model fitting, we need to know the distribution or probability encoding model of the latency patterns conditioned on the CH2O odor

intensity. With a particular model, parameterized by a vector ɵ, we can use maximum likelihood (ML) to

obtain the optimal estimate ɵ for which the latency patterns are most likely:

= argmax ( | , ) (2)

In the multivariate Bayesian method [21], we

assume that latency patterns follow multivariate Gaussian distribution N( , Σ) and hence second order statistics, that is, mean and covariance, are sufficient to learn this distribution. We use ML to estimate these parameters from the sensor array measurements. If m is the ML estimate of the true mean ( ) and S is the ML estimate of the true covariance matrix (Σ), then the conditional density of the latency patterns with a given CH2O odor intensity class cj is given by

= 1(2 ) exp − 12 − −

(3)

For a new latency pattern, Bayesian decoding is used to compute the posterior probability p(cj|l) of each concentration class cj in the set with a given latency pattern l. It can be described as

= ( ) , (4)

where p(cj) is the prior probability of the j-th class. In our case, we consider the same prior for each class. The denominator term p(l) is for normalization, and the same for all classes, and hence can be ignored. As a result, the posterior probability (p(cj|l)) only depends on the likelihood term (p(l| cj)).

The multivariate Bayesian approach considers correlations between latencies, and hence involves substantial computations and requires large storage for estimated parameters, i.e., n ⨯ (d ⨯ d) for the covariance matrix and (n ⨯ d) for the mean vector for n classes. Moreover, the sample covariance matrix is considered as a poor estimate of the true covariance matrix in the case when the available measurements and feature vector size are comparable or their ratio is not extremely large [22]. On the other hand, the naive Bayes classifier assumes that features are independent and require comparatively few parameters, i.e., (n ⨯ d) for the mean vector and (n ⨯ d) for the variance. Due to the independent assumption, the likelihood term can be written as:

= (5)

After estimating the sample mean (denoted as mij)

and sample variance (denoted as vij) through ML for the i-th sensor and j-th class, the p(li|cj) can be written as for univariate Gaussian distribution:


89

= 12 exp − −2 (6)

Finally, the concentration class with the maximum

posterior probability (represented as c) is selected as

an estimated concentration of the newly observed latency pattern:

= argmax (7)

3. Experimental Setup and Performance Evaluation

We use six commercially available Figaro gas

sensors to build an array for CH2O concentration estimation. The description of these sensors is listed in Table 1, along with the names of the target compounds for which they are mainly marketed. Different part numbers are used in an attempt to achieve a unique latency pattern with varying sensitivity.

Table 1. Gas sensors used to acquire CH2O signatures

No. Figaro Part

Number Target Compounds

1. TGS 826 Ammonia 2. TGS 2600 Air contaminants 3. TGS 2602 Volatile organic compounds 4. TGS 2610 Liquefied petroleum gas 5. TGS 2611 Methane 6. TGS 2620 Solvent vapors

The experimental setup for acquiring the response of CH2O at different concentrations is shown in Fig. 2. The sensor array is embedded in a glass container with an inlet valve for CH2O exposure and outlet valve for its outflow. The cylinders of CH2O and air are connected to mass flow controllers (MFCs), which are used to control the CH2O concentration by mixing it with air in different proportions. A computer with a data acquisition board is used for MFCs programming in order to achieve the desired concentration of CH2O and digitize the response of the sensor array.

Fig. 2. (a) Block diagram of experimental setup to acquire data of CH2O at different concentrations; (b) Gas cylinders; (c) MFCs; (d) Sensor array embedded in a glass chamber; (e) Data acquisition system.

We expose the sensor array to twenty different concentration values of CH2O in the range between 0.25 ppm to 5 ppm, with a 0.25 ppm increment step. At the start of the experiment, the sensor array is firstly exposed to air for 750 seconds to obtain the response without any target gas vapors. This response is referred to as a baseline response. The sensor array is then exposed to CH2O at a specified concentration for 500 seconds to obtain its response. Then air is again

injected for 750 seconds to remove the CH2O molecules trapped on the sensors in the previous gas injection phase and to recover the baseline response, i.e., the response without CH2O vapors. All the sensors in the array respond to the target concentrations of CH2O with different values of sensitivity. A typical response of the sensors in the array to CH2O at two different concentrations is shown in Fig. 3. In the figure, the sensor array response corresponding to air


90

exposure is shown up to 750 seconds. Upon injection of CH2O, with concentration value of 0.25 ppm, at 751 seconds, the resistance of each sensor starts decreasing. At 1250 seconds, air is again injected to allow the sensors to recover their baseline response. At 2000 seconds, CH2O is again injected with increased concentration value of 0.5 ppm, which results in a further drop in the resistance of the sensors. We repeat this process to acquire 200 response patterns for the whole concentration range.

Fig. 3. Sensor array response to air and at two different concentrations of CH2O.

From the resistance values, the sensitivity of each sensor is computed by dividing the value of the sensor resistance at the end of gas exposure by the sensor resistance at the end of air exposure. The regression coefficient ai of each sensor is computed through linear regression between the log sensitivity of the i-th sensor and the average of the log sensitivity across the sensor array, as shown in Fig. 4.

Fig. 4. Extraction of regression coefficients for each sensor

in the array through linear regression.

The resultant regression parameters are used to transform the sensitivity pattern of the sensor array into a spike latency code or pattern by using

Equation (1). Fig. 5 shows the latency values at each concentration level, where the latency is represented in arbitrary units resulting from Equation (1). There is no fixed value of spike latency at any concentration level. Instead, variability is shown at each concentration level and the mean value of the latency varies with the change in concentration.

Fig. 5. Spike latency of each sensor at twenty different concentrations of CH2O.

Bayesian inference methods along with other state of the art methods, including Gaussian mixture models (GMM), multi-layer perceptron (MLP) and support vector machines (SVM) with a linear and radial basis function (RBF) kernel, are used to estimate the CH2O concentration class with the resultant data set. A 5 × 2 cross validation technique is used to evaluate the performance of these methods by dividing the experimental data into training, validation and testing sets. The performances of all these methods are summarized in Table 2. Table 2. Performance comparison of algorithms for CH2O

concentration quantification/classification

Classification Method

Classification Performance

(%)

Maximum Concentration Error (ppm)

GMM 91.25 0.25 MLP 89.25 0.50 SVM (Lin) 88.5 0.25 SVM (RBF) 92.5 0.25 QDA 92.75 0.25 NBC 95.75 0.25

In terms of maximum concentration error, the performance of most of the classifiers is comparable, but a maximum accuracy of 95.75 % is achieved with naive Bayes classifier (NBC) to correctly classify the true concentration class. The better performance of the naive Bayes classifier among other state of the art methods has been reported in many different domains


91

[23-27], even when the independence assumption does not hold, which is justified in [26-27] by showing its optimality under dependence conditions. The simpler implementation of the naive Bayes classifier as compared to other methods also facilitates the integration of a simple hardware solution with the sensor array.

A simplified gas quantification system can be developed with the naive Bayes classifier as shown in

Fig. 6. Parameters for latency formation and univariate Gaussian distribution are computed from the training data and stored in memory. A newly test sensitivity pattern is first transformed to a latency pattern by using the corresponding regression coefficients and then the stored distribution parameters of each class are used to compute the conditional probability. Finally, the concentration class with maximum probability is assigned to the new test pattern.

Fig. 6. Block diagram for CH2O quantification with naive Bayes classifier by utilizing spike latency patterns.

4. Conclusion

In this paper, we have proposed a low cost and compact solution in the form of an electronic nose, containing an array of commercially available low cost gas sensors, to estimate the concentration of health endangering formaldehyde. As compared to this solution, commercial methods for formaldehyde monitoring are more accurate but cannot be adopted for continuous monitoring on a long-term basis due to the time, cost and complex sequence of procedures involved in these methods. Instead of focusing on exact knowledge of concentration, we treat this problem as a classification problem by dividing the target concentration range at evenly spaced intervals. A bio-inspired spike latency coding scheme is used to extract knowledge of concentration. In order to estimate concentration in the presence of stochastic variability in the latency patterns, different classification methods are explored. The best performance is achieved with a naive Bayes classifier. Besides providing an analytical solution and not requiring any manual tuning of parameters, the naive Bayes classifier also facilitates hardware friendly gas quantification/classification which can be realized with a simple hardware platform. A building resident can continuously monitor formaldehyde concentration

levels with the electronic nose and can set a threshold to indicate when to take preventive action, such as removal of the formaldehyde source or provision of proper ventilation to reduce its concentration. Acknowledgements

This paper was made possible by the National Priorities Research Program (NPRP) grant No. 5 - 080 - 2 - 028 from the Qatar National Research Fund (a member of the Qatar Foundation). The statements made herein are solely the responsibility of the authors. References [1]. M. N. Indang, A. S. Abdulamir, A. A. Bakar, A. B.

Salleh, Y. H. Lee, Y. N. Azah, A review: methods of determination of health-endangering formaldehyde in diet, Research Journal of Pharmacology, Vol. 3, No. 2, 2009, pp. 31-47.

[2]. T. Salthammer, S. Mentese, R. Marutzky, Formaldehyde in the indoor environment, Chemical Reviews, Vol. 110, No. 4, 2010, pp. 2536-2572.

[3]. W. D. Kerns, K. L. Pavkov, D. J. Donofrio, E. J. Gralla, J. A. Swenberg, Carcinogenicity of


92

formaldehyde in rats and mice after long-term inhalation exposure, Cancer Research, Vol. 43, No. 9, 1983, pp. 4382-4392.

[4]. International Agency for Research on Cancer, IARC classifies formaldehyde as carcinogenic to humans, Press release. June15, 2004, URL: http://www.iarc. fr/en/mediacentre/pr/2004/pr153.html [accessed: 2015-09-09].

[5]. P. Guo, K. Yokoyama, F. Piao, K. Sakai, Sick building syndrome by indoor air pollution in Dalian, International Journal of Environmental Research and Public Health, Vol. 10, No. 4, 2013, pp. 1489-1504.

[6]. J. G. Dojahn, W. E. Wentworth, S. D. Stearns, Characterization of formaldehyde by gas chromatography using multiple pulsed-discharge photoionization detectors and a flame ionization detector, Journal of Chromatographic Science, Vol. 39, No. 2, 2001, pp. 54-58.

[7]. H. L. Pinheiro, M. V. de Andrade, P. A. de Paula Pereira, J. B. de Andrade, Spectrofluorimetric determination of formaldehyde in air after collection onto silica cartridges coated with Fluoral P, Microchemical Journal, Vol. 78, No. 1, 2004, pp. 15-20.

[8]. S. Capone, P. Siciliano, F. Quaranta, R. Rella, M. Epifani, L. Vasanelli, Analysis of vapours and foods by means of an electronic nose based on a sol-gel metal oxide sensors array, Sensors and Actuators B-Chemical, Vol. 69, No. 3, 2000, pp. 230-235.

[9]. A. P. F. Turner, N. Magan, Electronic noses and disease diagnostics, Nature Reviews Microbiology, Vol. 2, No. 2, 2004, pp. 161-166.

[10]. M. Trincavelli, S. Coradeschi, A. Loutfi, B. Söderquist, P. Thunberg, Direct identification of bacteria in blood culture samples using an electronic nose, IEEE Transactions on Biomedical Engineering, Vol. 57, No. 12, 2010, pp. 2884-2890.

[11]. R. E. Baby, M. Cabezas, E. N. Walsöe de Reca, Electronic nose: a useful tool for monitoring environmental contamination, Sensors and Actuators B-Chemical, Vol. 69, No. 3, 2000, pp. 214-218.

[12]. M. Penza, G. Cassano, Classification of food, beverages and perfumes by WO3 thin-film sensors array and pattern recognition techniques, Sensors and Actuators B-Chemical, Vol. 73, No. 1, 2001, pp. 76-87.

[13]. M. Holmberg, F. Winquist, I. Lundström, J. W. Gardner, E. L. Hines, Identification of paper quality using a hybrid electronic nose, Sensors and Actuators B-Chemical, Vol. 27, No. 1, 1995, pp. 246-249.

[14]. M. Hassan, A. Bermak, Threshold detection of carcinogenic odor of formaldehyde with wireless electronic nose, in Proceedings of the IEEE Sensors, Valencia, November 2014, pp. 1368-1371.

[15]. T. W. Margrie, A. T. Schaefer, Theta oscillation coupled spike latencies yield computational vigour in a mammalian sensory system, Journal of Physiology, Vol. 546, No. 2, 2002, pp. 363-374.

[16]. H. T. Chen, K. T. Ng, A. Bermak, M. K. Law, D. Martinez, Spike latency coding in biologically inspired microelectronic nose, IEEE Transactions on Biomedical Circuits and Systems, Vol. 5, No. 2, 2011, pp. 160-168.

[17]. M. Hassan, S. Brahim Belhaouari, A. Bermak, Probabilistic rank score coding: A robust rank-order based classifier for electronic nose applications, IEEE Sensors Journal, Vol. 15, No. 7, 2015, pp. 3934-3946.

[18]. J. A. Yamani, F. Boussaid, A. Bermak, D. Martinez, Glomerular latency coding in artificial olfaction, Frontiers in Neuroengineering, Vol. 4, art. 18, 2012, pp. 1-9.

[19]. E. Alpaydin, Introduction to Machine Learning, MIT Press, 2010.

[20]. D. C. Knill, A. Pouget, The Bayesian brain: the role of uncertainty in neural coding and computation, Trends in Neuroscience, Vol. 27, No. 12, 2004, pp. 712-719.

[21]. M. Hassan, A. Bermak, A. A. S. Ali, A. Amira, Bayesian inference using spike latency codes for quantification of health endangering formaldehyde, in Proceedings of the 6th International Conference on Sensor Device Technologies and Applications (SENSORDEVICES’ 2015), Venice, Italy, 23-28 August 2015, pp. 79-82.

[22]. A. P. Dempster, Covariance Selection, Biometrics, Vol. 28, No. 1, 1972, pp. 157-175.

[23]. B. Turhan, A. Bener, Analysis of Naive Bayes’ assumptions on software fault data: An empirical study, Data & Knowledge Engineering, Vol. 68, No. 2, 2009, pp. 278-290.

[24]. J. M. A. Luz, M. S. Couceiro, D. Portugal, R. P. Rocha, Hélder Araújo, Gonçalo Dias, Comparison of classification methods for golf putting performance analysis, in Computational Intelligence and Decision Making, Springer, Netherlands, 2013, pp. 35-45.

[25]. Y. Li, R. Anderson-Sprecher, Facies identification from well logs: A comparison of discriminant analysis and naive Bayes classifier, Journal of Petroleum Science and Engineering, Vol. 53, No. 3, 2006, pp. 149-157.

[26]. H. Zhang, Exploring conditions for the optimality of naive Bayes, International Journal of Pattern Recognition and Artificial Intelligence, Vol. 19, No. 2, 2005, pp. 183-198.

[27]. P. Domingos, M. Pazzani, Beyond independence: conditions for the optimality of the simple Bayesian classifier, in Proceedings of the International Conference on Machine Learning, 1996, pp. 105-112.

___________________



93



Alternative Processes for Manufacturing of Metal Oxide-based Potentiometric Chemosensors

1 Winfried VONAU, 1 Manfred DECKER, 1 Jens ZOSEL,

1 Kristina AHLBORN, 1 Frank GERLACH, 2 David HALDAN and 2 Steffen WEISSMANTEL

1 Kurt-Schwabe-Institut für Mess- und Sensortechnik, Kurt-Schwabe-Straße 4, 04736 Waldheim, Germany

2 Hochschule Mittweida, Fakultät Ingenieurwissenschaften, Technikumplatz 17, 09648 Mittweida, Germany 1 Tel.: +49 34327 608 0, fax: +49 34327 608131

1 E-mail: [email protected]

Received: 31 August 2015 /Accepted: 5 October 2015 /Published: 30 October 2015

Abstract: New possibilities for the preparation of partially selective redox electrodes based on passivated metals of the subgroups IV to VI of the periodic system are presented by the example of vanadium. The gas phase oxi-dation at controlled oxygen partial pressures (CPO) and the pulsed laser deposition (PLD) as an high-vacuum method are utilised as alternative methods beside the well-established chemical and electrochemical passivation which usually lead to the highest possible oxidation state of the passivated metal. These newly available methods enable in principle the tailoring of oxidation states in the sensitive layer and therefore the optimisation of the electrochemical sensitivity and selectivity of sensors equipped with it. The use of vanadium as basic electrode material is crucial because it shows in several matrices a remarkable corrosion susceptibility. This problem can be solved by the introduction of stable alloys with high vanadium contents. These materials can be efficiently produced by pulsed laser deposition (PLD). Copyright © 2015 IFSA Publishing, S. L. Keywords: Partial selective redox electrode, Chemical passivation, Electrochemical passivation, Gas phase oxi-dation, Pulsed laser deposition.

1. Introduction

Measurements of the oxidation/reduction potential

(ORP) are suited for online determination of oxidising agents like halogens or hydrogen peroxide e.g. in pro-cess and waste waters, if the electrode materials pro-vide a partial selectivity [1]. It is already known that systems made of passivated metals of high purity

(> 99.5 wt.-%) of the subgroups IV to VI of the peri-odic system are suited for this purpose. The favourable sensory behaviour is related to the semiconducting and corrosion properties of the oxide layers with electronic and ionic defects. Investigations have shown that for the quality of the sensor functionality an n-type con-ducting mechanism is favourable, accordingly the pas-sivating sensing oxide should be an n-type semicon-ductor. The sensor is working like a Schottky-diode




94

[2]. Fig. 1 presents in this context exemplarily the elec-trochemical process for a chlorine determination at a passivated titanium electrode.

Fig. 1. Electron migration at a titanium/ titanium oxide electrode.

The preparation of the oxide layers was carried out

so far by chemical passivation [M(ethod) 1)] using air or pure oxygen as oxidising agent or by anodic oxida-tion [M 2)] in half concentrated acids (e.g. sulfuric and nitric acid) [3], leading usually to the highest possible valence state of the metal at the outer electrode sur-face. The inner oxide layer near the metal bulk can contain also lower valence states.

In Table 1, appropriate metals for the construction of potentiometric sensors with partial selectivity ver-sus selected analytes are presented. Fig. 2 shows a pos-sible design for the described electrode according to the state of the art.

Table 1. Membrane materials for the fabrication of partial selective electrodes.

Generally, electrodes made of these materials ex-

hibit a cross sensitivity to the pH value which should be known and considered. As an example for such a behaviour Fig. 3 shows results for the determination of nitrite by means of an anodically passivated tungsten based indicator electrode (according to Fig. 2) at pH values between -1.4 … 9. There are two ranges with negligible influence of the pH value on the nitrite con-centration related electrode potential (range 1: pH= -1.4 … 1; range 2: pH= 4 … 9). Therefore, a sim-ultaneous pH determination is always advantageous, whereupon the existing problems of pH measurement

with glass based or other pH electrodes in strong acidic media should be considered.

To tailor the most important parameters sensitivity and selectivity of the described partial selective elec-trochemical sensors it might be advantageous to pre-pare oxide layers in contact with the electrolyte which also possess several lower oxidation states with suffi-cient chemical stability.

Fig. 2. Redox electrode with passivated metal disk as partial selective membrane.

Fig. 3. Nitrite determination with tungsten oxide elec-trodes vs. Ag/AgCl, Cl-

sat .

In this connection in section 2, the preparation, characterisation and application of membrane materi-als for partial selective electrodes based on passivated pure metals of subgroups IV-VI is described exempla-rily by using vanadium as source material. Here, the final goal was to develop a potentiometric electrode for the determination of H2O2 in process-relevant ap-plications. In addition to the known use of M 1 and M 2 for this purpose electrode preparations also were car-ried out by means of gas phase oxidation at controlled temperature and oxygen partial pressure [CPO (M 3)] and pulsed laser deposition [PLD (M 4)].

Based on the knowledge gained during the first po-tentiometric sensory applications of these novel mate-rials in section 3 new perspectives are presented. The innovative production technologies allow a cost-and material-saving manufacturing of tailored passivated electrodes on the base of alloys of the above-men-tioned subgroup metals. This has particular relevance for the fabrication of vanadium oxide electrodes which show remarkable corrosion in aggressive media result-ing in a reduced service life.

electrode materialanalyte in aqueous phase Ti Ta W Nb V

Cl2 x x xBr2 x x x xI2 x

NO2 xH2O2 xFe3+ x

-

-3 -2 -1 0 1 2 3 4 5 6 9-300

-200

-100

0

100

200

300

400

500

600

700

1000 ppm nitrite

Poten

tial[

mV]

pH value87


95

2. Electrode Preparation, Characterisa-tion and Application

2.1. Preparation

M 1 Vanadium disks were fabricated from a metal bar

(purity: 99.8 wt.-%, Ø: 1 cm) using a diamond saw, contacted with a Pt-wire, encased by an insulating pol-ymer (e.g. polypropene) and wet grinded with sandpa-per (grain size: 4000). After cleaning of the metal sur-face by back sputtering passivation was carried out us-ing air or pure oxygen (low content of carbon dioxide and water vapour) as oxidising agent at temperatures < 100 °C.

M 2

Fig. 4 shows the electrochemical polarisation curve of a high-purity vanadium disk (prepared as de-scribed before) in sulfuric acid obtained within the range of UP = 0 ... 6 V vs. Ag/AgCl, Cl-

(sat.) at the scan rate 100 mV/s which was recorded in the course of the production of a related partial selective electrode. The observed colour changes of the vanadium surface seem to be in correlation with the formation of VO (green), VO2 (blue-black) and V2O5 (orange) as the fi-nal state. The anodic oxidation process was carried out using a potentiostat and an electrochemical cell ac-cording to Fig. 5.

Fig. 4. Anodic polarisation curve of pure vanadium in 50 wt.-% H2SO4 measured vs. Ag/AgCl, Cl-

(sat.).

Fig. 5. Electrochemical cell for the anodic oxidation and potentiometric measurements of partial selective elec-

trodes; 1 partial selective electrode, 2 reference electrode, 3 Pt-counter electrode, 4 Haber-Luggin-capillary,

5 gas injection system, 6 gas outlet system.

M 3 The gas phase oxidation method (CPO) offers the

possibility to establish tailored oxidation states of va-nadium oxides smaller than +5 by adjusting tempera-ture and pO2. The thermodynamic ranges of the differ-ent oxidation states of vanadium given in Fig. 6 pro-vide an orientation for the parameter adjustment in the gas phase.

Fig. 6. Phase diagram of the system vanadium-oxygen published in [4].

As illustrated in Fig. 7, the oxygen partial pressure of a mixture of N2/air/H2 is adjusted precisely within the range p(O2) = 10-30 ... 0.2 bar by a combination of a solid electrolyte pump cell and a solid electrolyte measuring cell [5]. After gas passage of the vanadium sample positioned in a separately heated transparent flow through cell, the oxygen partial pressure can be controlled again to measure the oxygen uptake by the vanadium surface [6]. Polished vanadium discs were placed inside the transparent cell and treated at differ-ent partial pressures and temperatures up to the point, where an oxide layer was visible by the unaided eye. Since the melting point of V2O5 amounts to 690 °C in contrast to that of VO2 at 1970 °C the change between the two oxidation states could easily be noticed by the observer in the transparent cell at probe temperatures above 690 °C. This change from liquid V2O5 to solid VO2 occurred at T = 1000 K at pO2 ≈ 100 Pa which is about on order of magnitude higher than indicated in Fig. 6.

Fig. 7. Schematic drawing of the experimental setup

for gas phase oxidation of vanadium.


96

This pO2-shift is caused by a kinetic phenomenon, the oxygen mass transfer from the gas phase to the ox-ide layer, enabling significant oxygen diffusion to the non-oxidized vanadium surface. This diffusion leads to a continuous increase of the thickness of the oxide layer and the corresponding oxygen consumption of the inner vanadium surface establishes a gradient of oxygen partial pressure within the oxide layer and the adjacent gas phase according to Fig. 8.

Fig. 8. Establishment of an oxygen pressure gradient at a heated vanadium surface due to oxygen diffusion through the oxide layers with different oxygen content x, calculation from thermodynamic data, published in [7].

Cooling the vanadium discs after heating to T = 1000 K over 15 min at constant p(O2) leads to dif-ferent oxidation states, as results of XPS analysis of the prepared oxide layers given in Fig. 9 clearly indi-cate. Adjusting oxygen partial pressure at pO2 ≈ 1 Pa the oxide surface is dominated by VO2 while pO2 ≈ 2·104 Pa leads to a surface consisting of V2O5 exclusively.

CPO with tailored temperature, oxygen partial pressure and treatment time enables therefore the sep-arate adjustment of surface oxidation state and thick-ness of the oxide layer without introducing additional chemicals, essentially important for the electrode be-haviour.

Fig. 9. XPS spectra of two different vanadium samples treated at 700 °C and (a): pO2 = 2·104 Pa, (b): pO2 = 4·10-9 Pa compared to a spectrum of (c): polished metallic surface.

M 4 Vanadium oxide films with defined oxidation

states were also prepared by using pulsed laser depo-sition (PLD) either in an oxygen background gas or with oxygen ion beam bombardment of growing films (see Fig. 10) [8].

Fig. 10. High vacuum pulsed laser deposition system used for the preparation of vanadium oxide films.

The method allows the preparation of metastable phases of vanadium oxide. For the preparation of the films, a KrF-excimer laser with 248 nm wavelength, 1 J laser pulse energy and 30 ns laser pulse duration was used for ablation from a pure vanadium target. For this, the laser beam was introduced into a high vacuum chamber and focused onto the target surface. A rela-tively high laser fluency of 8 J/cm² was used for abla-tion, the laser pulse repetition rate was 50 Hz. The films were prepared in an oxygen background gas us-ing partial pressures in the range of p(O2) = 5 × 10-4 ... 5 × 10-2 mbar. The substrate temperature during film deposition was constantly 400 °C.

The method is characterised by a high degree of more than 50 % of ionised species having high mean kinetic energy of several 10 eV in the ablated particle current. The ablated vanadium particles collide on their way to the substrate with the oxygen molecules of the background gas resulting in the dissociation of the oxygen molecules. Thus, highly reactive oxygen atoms form and can combine with the vanadium atoms and ions at the substrate surface. In a proper energy range the metastable vanadium oxide VO phase is formed. This was established by measuring the ratio of oxygen and vanadium in the films by using EDX (see Table 2), which is nearly 1. The variation of the oxy-gen partial pressure did not lead to significant changes in this ratio, so metastable VO is formed at all applied background gas pressures. 2.2. Characterisation

The preparation of vanadium oxides by method M 1 showed only irreproducible results. This was in accordance with [1], where investigations concerning


97

the passivation of Ti, Ta and W yielded in comparable observations. For that reason the following experi-ments have been executed with vanadium oxide-based electrodes prepared with the other oxidising methods. Table 2. Composition of vanadium oxide films deposited

by PLD at various oxygen partial pressures (Si results from the substrate and Pt from a thin conduction layer sputtered on the films. The thickness of the vanadium oxide films in-creased with increasing partial pressure so that the Si signal

becomes smaller.)

p(O2) [mbar]

O (W[%])

Si (W[%])

V (W[%])

Pt (W[%])

5.0E-04 22.93 47.2 26.45 3.5 5.0E-03 30.92 27.11 38.04 3.93 5.0E-02 41.63 13.63 43.03 1.71

Fig. 11 shows microscopic/ SEM shootings of the surfaces of the vanadium oxide based membranes gen-erated on metallic structures demonstrating that the methods result in functional films that cover the whole surface but differ significantly in their microscopic ap-pearance. It has to be stated that only M 3 and M 4 resulted in metal oxides with vanadium valences less than +5.

(a)

(b)

(c)

Fig. 11. SEM images of vanadium oxide films fabri-

cated by different preparation methods; (a) V2O5 formed by M 2 in 50 % H2SO4 with a scan rate UP= 4.3 V vs.

Ag/AgCl, Cl-sat at ϑ = 25 °C; (b) VO2 prepared by M 3;

(c) VO formed by M 4.

Apart from the method-specific created different chemical compositions of the oxides determined by XPS measurements (an example was given in Fig. 9) these changes are mainly caused by variations of layer thicknesses. Furthermore, the layer presented in Fig. 11(c) contains a number of droplets, a known drawback of the PLD method.

2.3. Application

Fig. 12 shows that with all discussed methods sen-sory active systems of the type V/VxOy with partial sensitivity to H2O2 are realisable. The investigated measuring range is fitted to the requirements of the galvanotechnical industry. The knowledge of H2O2-concentration is relevant for the correct adjustment of the ratio of Cu+/Cu2+ in the electrolyte [9]. The elec-trode functions vary widely. While with M 2 there are method-conditioned restrictions in relation to the stoi-chiometry of the functional oxide, M 3 and M 4 offer clearly more options concerning the sensor optimisa-tion. The highest sensitivity was obtained by the mem-brane produced with the PLD technology.

Fig. 12. Potentiometric determinations of H2O2 with sys-tems of the type V/VxOy realised by different preparation methods (M 2, M 3, M 4) at ϑ = 25°C in 5% Na2SO4 with additives of H2O2 at pH ≈2 (all vs. Ag/AgCl, Cl-

sat); M 2: anodic oxidation, M 3: CPO; M 4: PLD.

3. Perspectives

Experiences gained from measurement campaigns with indicator electrodes for the determination of H2O2 based on anodic oxidation of vanadium have demonstrated that a use of the pure transition group metal is unfavourable since the functional layers cor-rode quickly during the measurement (s. Fig. 13).

Fig. 13. Corroded electrode membrane after two week dwell time in a nickel electrolyte (ENTHONE) (mag. 500).

20 µm

20 µm

20 µm

-2 -1 0 1

0

20

40

60

80

100

120

140

160

ΔU[m

V]

log cH2O2 [mol/L]

M(ethod) 2 ---> 2V O5

M 3 ---> VO2

M 4 ---> VO

20 µm

20 µm

20 µm


98

It has been proven to be advantageous to alloy va-nadium with titanium before passivation playing an important role for the use of e.g. orthopaedic implants because of their corrosion stability and the improve-ment of the biocompatibility [10]. By adjusting an op-timal alloy-ratio of both metals, it is possible to retain the functionality of the electrode over a very long pe-riod. In this context, results of corrosion measure-ments given in Fig. 14 clearly indicate that the corro-sion stability of the electrode membrane can be in-creased significantly by adding titanium to the alloy.

For the choice of an appropriate alloy composition, it is necessary to investigate the upper alloying level of Ti at which the partial selectivity for hydrogen per-oxide is maintained in appropriate degree. In the case

of the investigated V-Ti-alloys the optimal titanium content for the hydrogen peroxide determination with partially selective electrodes can be amounted to 20 wt.-% according to Fig. 14 and 15.

Investigations in [11] give rise to the assumption that the application of ternary alloys may lead to even better results. The method 4 offers in this field the best requirements. For instance by an introduction of dif-ferent targets (V, Ti, Nb) in the PLD-chamber and a supply of an appropriate gas mixture for the oxidation the fabrication of ternary mixtures should be cheaper than a melting of the alloys with additional electro-chemical oxidation [12].

Fig. 14. Current density for activation ja and passivation jp of V-Ti-alloys.

Fig. 15. Electrode functions of electrodes of V and V-Ti al-loys vs. Ag/AgCl, Cl-

sat in 5 wt.-% Na2SO4 at ϑ = 25 °C in a H2O2 concentration range of b = 1 g/L … 20 g/L.

4. Conclusions

Beside anodic passivation, also CPO and PLD de-liver suitable vanadium oxide based membranes for the potentiometric determination of H2O2. The last two mentioned methods allow for vanadium in principle a largely variable realisation of oxidation states differ-ing from the value +5. This has an impact on the per-formance of resulting electrochemical indicator elec-trodes. In future work, these effects shall be examined more closely and the new manufacturing methods for the specific application will be improved, too.

Because on pure vanadium based electrodes are not suitable for aggressive matrices the application of corrosion stable passivated binary or ternary vanadium alloys can be a reasonable alternative. This will require an optimisation process concerning the corrosion re-sistance and the necessary electrode sensitivity and se-lectivity. For the manufacturing process the PLD tech-nology offers the best preconditions. References [1]. W. Habermann, Partial selective redox electrodes for

the on-line determination of oxidising agents, Ph. D. Thesis, Martin Luther University, Halle, Germany, 1990.

[2]. W. Habermann, P. John, H. Matschiner, and H. Spähn, Partially selective semiconductor redox electrodes, J. Anal. Chem., Vol. 356, 1996, pp. 182-186.

[3]. T. Bachmann, U. Guth, and W. Vonau, Physical and electrochemical properties of vanadium alloy oxide layers prepared by anodic passivation, Ionics, Vol. 7, 2001, pp. 172-177.

[4]. W. Brückner, H. Oppermann, W. Reichelt, J. I. Terukow, F. A. Tschudnowski, and E. Wolf, Vanadiumoxide. Darstellung, Eigenschaften, Anwendung, Akademie-Verlag, Berlin, 1983.

[5]. M. Schelter, J. Zosel, W. Oelßner, U. Guth, and M. Mertig, A solid electrolyte sensor for trace gas analysis, Sens. Actuators B: Chemical, Vol. 187, 2013, pp. 209–214.


99

[6]. V. Vashook, L. Vasylechko, J. Zosel, W. Gruner, H. Ullmann, and U. Guth, Crystal structure and electrical conductivity of lanthanum–calcium chromites–titan-ates La1−xCaxCr1−yTiyO3−δ (x=0–1, y=0–1), J. Solid State Chem., Vol. 177, 2004, pp. 3784-3794.

[7]. H. Endo, M. Wakihara, M. Taniguchi, T. Katsura, Phase Equilibria in the V2O3-VO2 System at High Temperatures, Bull. Chem. Soc. Japan, Vol. 46, 1973, pp. 2087-2090.

[8]. G. Reiße, S. Weissmantel, B. Keiper, B. Steiger, H. Johanson, T. Martini, and R. Scholz, Influence of ion bombardment on the refractive index of laser pulse deposited oxide films, Appl Surf. Sci., Vol. 86, 1995, p. 107.

[9]. T. Bachmann, W. Vonau, and P. John, Hydrogen Per-oxide Concentration in Acid Etching Baths, GIT La-boratory Journal, Vol. 7, pp. 2003, 149-150.

[10]. T. Bachmann, Electrochemical investigations of oxide layers consisting of vanadium and vanadium alloys, PhD. Thesis, University of Technology, Dresden, Germany, 2008.

[11]. P. John, Indicator electrode with sensitive surface and application therefor, German Patent, DE 199 53 218 C2.

[12]. W. Vonau, M. Decker, J. Zosel, K. Ahlborn, F. Ger-lach, and S. Weissmantel, New methods for the prep-aration of partial selective redox electrodes for the de-termination of H2O2, in Proceedings of the 6th Interna-tional Conference on Sensor Device Technologies and Applications (SENSORDEVICES '15), Venice, Italy, 28-30, August 2015, pp. 28-30.

___________________


http://www.sensorsportal.com/HTML/BOOKSTORE/Nanosensors_IFSA.htm





100



Improvement in Humidity Sensing of Graphene Oxide by Amide Functionalization

Sumita RANI, Dinesh KUMAR, Mukesh KUMAR

Electronic Science Department, Kurukshetra University Kurukshetra, Haryana, India


Received: 31 August 2015 /Accepted: 5 October 2015 /Published: 30 October 2015 Abstract: In this work, the effect of water adsorption on the electrical properties of graphene oxide (GO) and Amide functionalized graphene oxide (AGO) was studied using direct current measurements. AGO was synthesized by chemical method at room temperature. Fourier transform infrared spectroscopy, X-ray diffraction and scanning electron microscopy measurements were carried out to verify the functionalization of GO. The films of GO and AGO between aluminum electrodes on SiO2/p-Si (100) substrate were formed by drop casting method. The variation in the I-V characteristics was recorded at different humidity level. It has been observed that the interaction between water molecules with AGO was more as compared to GO. It was reported that electrical properties of GO and AGO are humidity and applied voltage dependent. At low humidity level the response of GO sensor was poor, however at high humidity the conductivity of GO increases. Compared to GO, AGO shows good response. The resistance of the AGO film was approximately 9.87 kΩ at 10 % relative humidity (RH), and decreases to 1.5 kΩ at 90 % RH. Copyright © 2015 IFSA Publishing, S. L. Keywords: Amide functionalized graphene oxide, Graphene oxide, Relative humidity sensing, I-V characteristics. 1. Introduction

Sensor technology plays vital role in industry and agricultural production, public safety, military etc. Due to low power consumption and ease of resistance measurements chemiresistive sensor become important from last few decades [1]. The measurement in variation in humidity is necessary to monitor, detect and control the ambient humidity by precise sensors [2]. Humidity sensors are important for fabrication of integrated circuits [3], in semiconductor industry it is essential to monitor moisture levels constantly. In medical field, humidity sensors are essential for respiratory equipment, sterilizers, incubators and biological processing. In agriculture, humidity sensors are used for green-

house air-conditioning, plantation protection, soil moisture monitoring, etc. In general, humidity sensors are used for moisture detection by various paper, textile and food processing industries. In modern humidity sensor there is requirement of high sensitivity and wide detection range for fast response and short recovery time to meet industry applications. Many sensing materials such as silicon nanostructures, ceramic nanomaterials, semiconductor metal oxide [1], carbon nanotubes (CNTs), metal oxide etc. are widely used for moisture sensing application due to high surface to volume ratio [4-5].

Two dimensional, graphene analog to CNTs, has emerged out as an important material for sensing due to its extraordinary electronic, thermal, chemical




101

properties [2, 4-5]. The sensors based on graphene shows very high sensitivity to gases like NH3, NO2 [6-8] etc. The studies reported on humidity sensing based on graphene shows long response time hinder its application in humidity sensing [9].

The graphene precursor, graphene oxide (GO) shows advantage as a material for humidity detection because of its large surface area [2-5]. GO has hydroxyl, epoxy and carboxylic acid functional group bonded in two-dimensional network of sp2 and sp3 hybridized carbon atoms arranged in a honeycomb structure [3]. The oxygen containing functional groups of GO enhance hydrophilic properties of it but these groups makes it an insulator by decreasing its conductivity [2, 8]. But oxygen functional groups of GO allow fast passage of water within the GO layers [10], which makes fast response of the sensor based on GO. But improving the electrical properties of GO by surface modification via noncovalent or covalent functionalization is a promising way for fabrication of high performance GO-based sensor [11-12]. Already functionalized GO have been successfully used for detection of acetone, hydrogen sulfide, nitrogen dioxide (NO2), etc [13-14]. Covalent functionalization of GO enhance its physicochemical properties, for instance, isocyanate-treated GO has been exfoliated and form a stable dispersion in polar aprotic solvents [15]. Such functionalization improves the mechanical, electrical, thermal properties and dispersion of functionalization GO in to organic solvents [16]. Recently, dodecylamine and ethylenediamine functionalized GO was used for hydrogen sulphide gas detection [17].

In view of above observations, heteroaryl/phenyl amine was grafted onto GO sheets by the amide formation between amine functionality of heteroaryl/phenyl amine and oxygen-containing groups (e.g., carboxyl and lactone groups) of GO to give amide functionalized GO (AGO). Afterwards, the fabrication and characterization of AGO chemiresistors was reported to understand effect of AGO on the sensor response. It has been observed that AGO has strong electrical response for water as compared to GO.

2. Experimental 2.1. Materials

Graphite powder (purity 99.99 %), sodium nitrate

(99.0 %), sulphuric acid, potassium permanganate (99 %), hydrogen peroxide, hydrochloric acid, sodium hydroxide (NaOH), hydroxybenzotriazole (HOBt), 2-aminothiazole, N,N'-dicyclohexyl-carbodiimide (DCC) were used to prepare GO and AGO.

2.2. Synthesis of GO To synthesize GO, graphite oxide was prepared

by the oxidation treatment of graphite with KMnO4.

For this, Graphite (2 g) and NaNO3 (1 g) were put in cooled concentrated sulphuric acid (46 ml) under stirring in ice bath. KMnO4 (6 g) was gradually added to the above placed mixture with stirring and cooling so that the temperature of mixture was maintained between 10–15°C. The reaction mixture was then stirred at 40°C for 30 minutes. Afterward, 80 ml of high purity water was added to the formed paste, followed by another 90 minutes stirring at 90°C. Successively, the oxidative reaction was terminated by addition of 200 ml water. 6 ml of 30 % H2O2 was added in above mixture sequentially to destroy the excess KMnO4. The complete removal of KMnO4 was indicated by color changed to yellow. Sometimes the solution’s color was yellow before addition of H2O2 which indicated complete reduction of KMnO4. The solution was then washed with HCl (10 %) to remove sulphate. Subsequently, it was filtered and washed several times with DI water. The filtered paste was dissolved in 100 ml DI water. The solution was ultrasonicated for 1 hour and centrifuged for 20 minutes at 4000 rpm. GO powder thus obtained was collected and dried at room temperature.

2.3. Amide Functionalization of GO The AGO was obtained by condensation of amine

group of heteroaryl/phenyl amine with lactone group of GO. The oxygenated GO sheet prepared by the chemical method was treated with NaOH to open the lactone groups (–CO-O–) on the basal plane and convert them into hydroxyl and carboxyl groups. Treatment of resulting reaction mixture with organic amines in the presence of DCC and HOBt leads to the amidation of the carboxyl groups to give AGO. AGO was synthesized by dispersing GO (0.3 g) in 30 ml DMF by ultrasonication for 60 minutes at room temperature. Then, NaOH (0.3 g; 7.5 mmol) was added and resulting solution was stirred for 60 minutes at room temperature. Afterward, 2-aminothiazole (3.1 mmol), HOBt (3.1 mmol) followed by DCC (3.1 mmol) addition to the above reaction mixture and stirred for 24 hours at room temperature. AGO powder collected by centrifugation was added to pure DMF and the resulting suspension was again centrifuged to remove side products. This process was repeated twice with DMF and then with water to remove DMF to give pure AGO. The prepared AGO was dried at 60°C overnight.

2.4. Preparation of Sensor The sensor was obtained by depositing a pair of

aluminum electrodes of thickness of 200 nm was deposited on SiO2/p-Si substrate by thermal vacuum coating unit. To prepare a sensing layer GO and AGO was dispersed into water through ultrasonication for 1 hour. The sensing layer was prepared by drop casting method on SiO2/Si substrate


102

between the aluminum electrodes. The prepared sensors of GO and AGO2 were kept in the chamber one by one and a particular humidity environment was maintained inside the glass chamber by introducing water vapors into chamber by inlet. The sensor I–V characteristics were measured at different humidity level using Keithley 2400 Series Source Measurement Unit via two probe method by stepping the voltage in the range of 0 V to 5 V.

2.5. Characterization The crystal phase was characterized by X-ray

diffraction (XRD) XPERT-PRO diffractometer (45 kV, 40 mA) equipped with a Giono-meter PW3050/60 working with Cu Kα radiation of wavelength 1.5406Å in the 2θ range from 5 to 80°). The functional surface group was studied by Perkin Elmer Fourier transform infrared (FTIR) model SPETRUM 65 system. Dried solid samples were mixed with KBr powder and were pelletized before performing the scan from wave number 4000 to 400 cm-1. SEM characterization was carried out by using JSM-6510LV Series Scanning Electron Microscope (SEM) having pre centered W hairpin filament (with continuous auto bias) and equipped with accelerating voltage of 500 V to 30 kV with high magnification of 300,000.

3. Results and Discussion The amidation occurs in the GO after treatment

with amine can be verified easily by FTIR, XRD, and SEM techniques. Structural changes of AGO were investigated by comparing the FTIR spectra of GO and AGO (Fig. 1). FTIR spectrum of GO and AGO is plotted in the range of 2000 cm-1 to 1000 cm-1. The most characteristic features in the FTIR spectrum of GO was the adsorption bands corresponding to the C=O carbonyl stretching at 1721 cm-1, the stretching bands for C=C bonds at 1591 cm-1, the O–H deformation vibration at 1392 cm-1 and the C–O stretching at 1051 cm-1 [18-19]. As depicted in FTIR spectra of AGO the peak intensity and position of these peaks are changed after amidation. The FTIR spectra of AGO, the C–O stretching peak present in GO at 1051 cm-1 was disappear and a new peak at about 1634 cm-1 corresponding to the amide carbonyl (C=O) stretch (amide I) and the peak at about 1585 cm-1 for amide II (C–N in-plane stretching and CHN deformation) [20], which demonstrate that amines has been grafted onto GO as amide bond.

XRD patterns for graphite, GO and AGO are shown in Fig. 2. XRD of graphite represent peak at 26.3˚, which shifted towards lower value of angle for GO (2θ = 11.42˚), which is the characteristic peak of the GO, with increase in interspacing from 0.34 nm (graphite) to 0.77 nm. This confirms the oxidation of graphite.

Fig. 1. FTIR spectra of GO and AGO.

Fig. 2. XRD of graphite, GO and AGO.

After functionalization of GO the peaks shift to the lower value of angle (2θ) and increase in the interlayer spacing from 0.77 nm (GO) to 0.84 nm for AGO, was observed. The larger value of interlayer spacing suggests the incorporation of additional functionality on the surface of the basal plane of GO.

The SEM images represent the surface morphology of the GO and AGO (Fig. 3). These images show that the resulting sample of AGO was obviously different from GO. Morphology of GO was observed to have flaky texture indicating its layered microstructure. Different surface morphology of AGO clearly indicates the surface modification of GO after functionalization.

Fig. 3. SEM image of GO and AGO.


103

The DC electrical properties, i.e., current–voltage (I-V) characteristics of the GO and AGO films were measured with a voltage sweeping mode at various humidity points. To measure the I-V characteristics one electrode was loaded with the sweeping voltage bias, and the other electrode is grounded. I-V characteristics of GO and AGO films was investigated for 2 to 5 V sweeping voltages at various humidity levels, respectively. The different percentage of humidity level was achieved in sensing chamber by introducing the water vapors.

Fig. 4 (a) shows the measured I-V characteristics of GO film in the relative humidity (RH) range of 10 % to 90 %. From Fig. 4 (a) it is clear that the channel current of the GO film increases with increasing RH, indicating that the water adsorption results in a decrease in the resistance of GO films [21]. Water adsorption easily takes place in GO because of presence of large number of oxygen containing functional groups in it. The result revealed that GO is a weak conductor with a continuous water adsorption onto GO films. It is also clear that the current increases with increase in humidity level. At higher humidity (>60 %) GO films gives good response.

The sensor resistance (Rs), as a function of humidity is calculated by:

Rs=δV/δI,

where δV is the incremental voltage and δI is the incremental current.

The variation of Rs with humidity is shown in Fig. 5. The Rs of the GO film was 0.76 MΩ at 10 % RH. High value of resistance at low RH can be explained on the basis of interaction of water molecules on the GO film. At low RH, water molecules are adsorbed on to the available sites of GO by double hydrogen bond, called first layer adsorption of water. The first layer water molecules are unable to move freely due to strong hydrogen bonding. Thus at low RH, GO film exhibits strong electrical resistance. At high RH, multilayer adsorption of water molecules occurs. The second adsorbed layer of water molecules was attached by single hydrogen bond on hydroxyl groups of GO. In higher adsorbed layers water molecules become free and move like bulk liquid [4]. This adsorbed water can be ionized and produce large number of hydrogen ions which involves in the reduction of GO. Therefore, the resistance of GO films decreases in case of high applied voltage at high humidity level. GO + 2H+ + 2e- reduced GO + H2O

The resistance in GO film decreases to 0.48 MΩ at 90 % RH. Further, at high RH, water molecules penetrate in to interlayer of GO and hydrolyzed the oxygen containing functional groups. These ions contribute to ionic conductivity, this explain the humidity sensing behavior of GO at low and high humidity level.

However, it can be observed from Fig. 4(b) that for AGO, the variation in current is more as

compared to GO. This may be due to increase in surface defects of GO with functionalization [12].

Fig. 4. I-V variation in (a) GO and (b) AGO with different humidity from 2V to 5V sweeping voltage.

Fig. 5. Variation in sensor resistance with humidity.

So, AGO is preferred for moisture sensing as compared to GO. The resistance of the AGO film was 9.87 kΩ at 10 % RH, which decreases to 3.52 kΩ at 30 % RH. The resistance decrease further with increase in humidity and it becomes 1.5 kΩ at 90 % RH. The resistance of film at 90 % RH was approximately seven times smaller than that at 10 %


104

RH. The smaller resistance of AGO at low RH is due to amidation which decreases the number of oxygen containing groups as compared to GO, so AGO gives better sensitivity. When ambient RH is high, the numbers of adsorbed water molecules are large. As a result, the ionization process generates more hydrogen ions, result in reduction of AGO film. Thereby, AGO film shows high conductivity for higher applied voltage at high humidity level.

4. Conclusions An efficient and easy approach has been used to

synthesize the covalent functionalized GO by a simple amidation reaction using 2-aminothiazole. DC measurement method was used to investigate the effect of humidity on the electrical properties of AGO films. Through electrical characterizations, the strong interaction of water molecules with AGO films was observed. The electrical properties of GO and AGO films were affected by humidity and the amplitude of applied voltage. At low RH (<60 %), GO films exhibited small variation in current due to the presence of sp3-bonded hybridized carbon atoms and presence of oxygen containing functional groups. AGO shows much better response at lower and higher RH (10 % to 90 %) indicates its effectiveness as compared to GO. The results are useful for the development of graphene-based sensors.

Acknowledgments One of author Sumita Rani is thankful to

INSPIRE, Department of Science and Technology (DST), India for funding support. References [1]. N. Hu, Y. Wang, J. Chai, R. Gao, Z. Yang,

E. S. Kong, Y. Zhang, Gas sensor based on p-phenylenediamine reduced graphene oxide, Sensors and Actuators B, Vol. 163, 2012, pp. 107-114.

[2]. S. Borini, R. White, D. Wei, M. Astley, S. Haque, E. Spigone, N. Harris, J. Kivioja, T. Ryha¨nen, Ultrafast Graphene Oxide Humidity Sensors, ACS Nano, Vol. 7, 2013, pp. 11166-11173.

[3]. D. Kumar, S. Rani, Synthesis of Amide Functionalized Graphene Oxide for Humidity Sensing Application, in Proceedings of the 6th

International Conference on Sensor Device Technologies and Applications (SENSORDEVICES’15), Venice, Italy, 23-28 August, 2015, pp. 142-145.

[4]. H. Bi, K. Yin, X. Xie, J. Ji, S. Wan, L. Sun, M. Terrones, M. S. Dresselhaus, Ultrahigh humidity sensitivity of graphene oxide, Scientific Reports, Vol. 3, 2013, 2714.

[5]. U. Mogera, A. A. Sagade, S. J. George, G. U. Kulkarni, Ultrafast response humidity sensor using supramolecular nanofibre and its application in

monitoring breath humidity and flow, Scientific Reports, Vol. 4, 2014, 4103.

[6]. J. T. Robinson, F. K. Perkins, E. S. Snow, Z. Q. Wei, P. E. Sheehan, Reduced graphene oxide molecular sensors, Nano Letters, Vol. 8, No. 10, 2008, pp. 3137-3140.

[7]. R. Arsat, M. Breedon, M. Shafiei, P. G. Spizziri, S. Gilje, R. B. Kaner, K. Kalantar-Zadeh, W. Wlodarski, Graphene-like nano-sheets for surface acoustic wave gas sensor applications, Chemical Physics Letters, Vol. 467, No. 4, 2009, pp. 344-347.

[8]. I. Jung, D. Dikin, S. Park, W. Cai, S.L. Mielke, R. S. Ruoff, Effect of water on electrical properties of individual reduced graphene oxide sheets, Journal of Physical Chemistry C, Vol. 112, No. 51, 2008, pp. 20264-20268.

[9]. A. Ghosh, D. J. Late, L. S. Panchakarla, A. Govindaraj, C. N. R. Rao, NO2 and humidity sensing characteristics of few-layer graphenes, Journal of Experimental Nanoscience, Vol. 4, No. 4, 2009, pp. 313-322.

[10]. R. R. Nair, H. A.Wu, P. N. Jayaram, I. V. Grigorieva, A. K. Geim, Unimpeded Permeation of Water Through Helium-Leak–Tight Graphene-Based Membranes, Science, Vol. 335, No. 6067, 2012, pp. 442-444.

[11]. C. Marichy, P. A. Russo, M. Latino, J. Tessonnier, M. Wilinger, N. Donato, G. Neri, N. Pinna, Tin dioxide-carbon heterostructures applied to gas sensing: structure-dependent properties and general sensing mechanism, Journal of Physical Chemistry C, Vol. 117, Issue 38, 2013, pp. 19729-19739.

[12]. R. Sumita, K. Mukesh, S. Sumit, K. Dinesh, Characterization and dispersibility of improved thermally stable amide functionalized graphene oxide, Material Research Bulletin, Vol. 60, 2014, pp. 143-149.

[13]. S. Mao, S. Cui, G. Lu, K. Yu, Z. Wei, J. Chen, Tuning gas-sensing properties of reduced graphene oxide using tin oxide nanocrystals, Journal of Material Chemistry, Vol. 22, No. 22, 2012, pp. 11009-11013.

[14]. P. A. Russo, N. Donato, S. G. Leonardi, S. Baek, D. E. Conte, G. Beri, N. Pinna, Room-temperature hydrogen sensing with heteronanostructures based onreduced graphene oxide and tin oxide, Angewandte Chemie International Edition, Vol. 51, Issue 44, 2012, pp. 11053-11057.

[15]. S. Stankovich, R. D. Piner, S. T. Nguyen, R. S. Ruoff, Synthesis and exfoliation of isocyanate-treated graphene oxide nanoplatelets, Carbon, Vol. 44, No. 15, 2006, pp. 3342-3347.

[16]. E. Fuente, J. A. Menéndez, D. Suárez, M. A. Montes-Morán, Basic surface oxides on carbon materials: A global view, Langmuir, Vol. 19, No. 8, 2003, pp. 3505-3511.

[17]. M. M. Alaie, M. Jahangiri, A. M. Rashidi, A. Haghighi Asl, N. Izadi, A novel selective H2S sensor using dodecylamine and ethylenediamine functionalized graphene oxide, Journal of Industrial and Engineering Chemistry, Vol. 29, 2015, pp. 97-103.

[18]. G. I. Titelman, V. Gelman, S. Bron, R. L. Khalfin, Y. Cohen, H. Bianco-Peled, Characteristics and microstructure of aqueous colloidal dispersions of graphite oxide, Carbon, Vol. 43, No. 3, 2005, pp. 641-649.

[19]. J. I. Paredes, S. Villar-Rodil, A. Martı´nez-Alonso, J. M. D. Tascón, Graphene oxide dispersions in organic


105

solvents, Langmuir, Vol. 24, No. 19, 2008, pp. 10560-10564.

[20]. H. Gunzler, H. U. Gremlich, IR spectroscopy, Wiley-VSH, Winheim, 2002, pp. 223-227.

[21]. H. F. Teoh, Y. Tao, E. S. Tok, G. W. Ho, C. H. Sow, Electrical current mediated interconversion between graphene oxide to reduced graphene oxide, Applied Phyical Letters, Vol. 98, 2011, 173105.

___________________


http://www.sensorsportal.com/HTML/CONFERENCES/SEIA_2015/index.htm


106



PbS Infrared Detectors: Experiment and Simulation

1 S. KOUISSA, 1A. DJEMEL, 1 M. S. AIDA, 2 M. A. DJOUADI 1 Department of Physics, University of Constantine1, 25000, Algeria

2 Laboratoire ID2M, Institut des Matériaux Jouhn-Rouxel-CNRS, Université de Nantes, France 1 Tel.: (213) 31 81 11 07, fax: (213) 31 81 11 08

E-mail: [email protected], [email protected], [email protected], [email protected]

Received: 31 August 2015 /Accepted: 5 October 2015 /Published: 30 October 2015 Abstract: The present work deals with the characterization and simulation of lead sulfide (PbS) photoconductors infrared detectors growth by Chemical Bath Deposition (CBD) method. Three different solutions bath are used in order to explore the doping effect and oxidant agent on detection capabilities. Photoelectrical characterization indicates that detectors performances depend strongly on oxidant and doping agents. A simulation study with surface state model is also presented. The physical parameters are deduced and are found to be in agreement with those published in the literature. Copyright © 2015 IFSA Publishing, S. L. Keywords: Chemical Bath Deposition, Photoconductors, Infrared detectors, Surface state model, PbS. 1. Introduction

Thin film lead sulfide detectors have been widely used over the past years for radiation sensing in 1 to 3 µm spectral region. They are mainly very useful in academic, commercial and military applications. In military application, PbS detectors are used for both tactical and strategic systems, with a very strong emphasis towards an increasing requirement for large area multiple element arrays.

Unlike most other semiconductors IR detectors, lead sulfide materials are used in the form of polycrystalline films approximately 1 µm thick and with individual crystallites ranging in size from approximately 0.1 µm to 1 µm. They are usually prepared by chemical bath deposition (CBD), which generally yields better uniformity of response and more stable results than the evaporate methods [1-6].

As-deposited PbS films exhibit very low photoconductivity, however, a post deposition process are used to achieve final sensitization. To obtain high performance detectors, lead chalcogenide

films need to be sensitized by oxidation. This oxidation may be carried out using additives in the deposition bath, post-deposition heat treatment in the presence of oxygen, or chemical oxidation of the films.

Others impurities added to the chemical – deposition solution for PbS have a considerable effect on photosensitivity films characteristics [6-10]. They may increase the photosensitivity by some order of magnitude more than films prepared without these impurities [9-10].

This work deals with characterization and simulation of PbS photoconductors infrared detectors prepared by CBD. The effect of oxidant agent and Bismuth Nitrate additives on the performances of PbS detectors is also examined. Finally, a simulation study with surface state model proposed in [11-12] is also presented. This paper is organized as follows: Section 1 presents the particularities associated with PbS infrared photodetectors development. Section 2 details the experimental procedure for devices fabrication and characterization. After analysis of




107

surface morphology and film structure in Subsection 3.1, we performed optical measurements (3.2) in order to understand the effect of additives of oxidation and doping on structural and optical properties of deposited PbS films. Measurements of frequency response and time constant of detectors prepared by solution Bath2 are presented in Subsection 3.4.1. Based on surface state model reviewed in Subsection 3.3, a comparison between photoelectrical measurements of developed PbS photoconductors and theoretical prediction is proposed in Subsection 3.4.2. The conclusions and perspectives are described in Section 4.

2. Experimental Details The setup deposition of E. Pentia, et all. [10] was

used for growing the PbS films over three bath solutions, summarized in Table 1. The Bath 1 contain the basic precursors which are Lead Nitrate Pb(NO3)2, Sodium Hydroxide (NaOH) and Thiourea SC(NH2)2, the Bath 2 contain in addition an oxidant agent, named Hydroxylamine Hydrochloride (NH2OH-HCl), and the Bath 3 contain in addition to Bath 2, the Bismuth Nitrate Bi(NO3)3 as doping agent.

Table 1. Solutions bath.

Bath 1 Pb(NO3)2 NaOH SC(NH2)2

Bath 2 Bath1 + NH2OH.HCl Bath 3 Bath2 + Bi(NO3)3

Aqueous solution of 0.069 M lead nitrate, 0.69 M NaOH and 0.24 M thiourea were used. In order to prepare PbS films, the following procedure was adopted: 20 ml of lead nitrate solution was mixed with 20 ml of NaOH with constant stirring. The initial color solution was milky, after it became transparent, 20 ml of thiourea solution was gradually added followed by addition of oxidant with 20 ml of 0.086 M Hydroxylamine Hydrochloride (NH2OH.HCL), again with constant stirring, the global solution was diluted with 20 ml of water. Finally, a small quantity of Bismuth nitrate with 1.5 ml of 2.0610-4 M was added to some reactions. The PbS films were deposited on microscope glass substrate cleaned, for about 48 h, in a mixture of (HNO3, (K2Cr2O7: H2SO4; 1:10), 1 % EDTA followed by rinsing in distilled water.

After drying, one facet of this substrate was stuck with an inert paste on a support having T format introduced vertically in the reaction bath containing the chemical mixture. After some time, the transparent color solution started to change to become completely black after one (01) hour.

The measured film thicknesses, using DEKTAK profilometer, were about 150 nm for both films elaborated with and without oxidant. Thicker films

were obtained by repeated deposition. Gold electrodes were evaporated on the surface of PbS films for electric and photoelectric measurements in a coplanar configuration. The films characterizations were performed after annealing in air at 80 °C for approximately 70 h.

Structural properties were evaluated by x-rays diffraction using a D5000 Siemens diffractometer. The scans were carried out at room temperature, in the conventional θ/2θ mode using Cu-Kα radiation (0.1542 nm). The morphology was observed by scanning electron microscopy (SEM) using a JEOL 6400F microscope. The optical properties were studied with a CARY 5000 UV-Vis-NIR double beam spectrometer. The photoconductivity measurements were performed with a system constituted by a standard IR light source, an Oriel MS257 monochromator operated in the range 1-20 µm, a chopper fixed at 400 Hz and the acquisition equipment composed with spectrum analyzer, lock-in amplifier and oscilloscope.

3. Results and Discussions

3.1. Surface Morphology and Film Structure Fig. 1 and Fig. 2 show SEM micrographs of PbS

films deposited with and without oxidant. It appears that the average grain size increases with the used oxidant.

Fig. 1. SEM image of PbS films prepared without oxidant agent.

Concerning the films structural properties, Fig. 3 and Fig. 4 show the effect of annealing treatment (T=80 °C, for 72 h) and Bismuth doping on the XRD (X Rays Diffraction) patterns of deposited PbS films. As can be seen, films deposited with hydroxylamine hydrochloride additive are less textured when compared to films prepared without this additive, which are (200) preferentially oriented, the grains become oriented quasi-equally with (200) and (111) crystallographic direction. The effect of annealing at


108

80°C for 72 h and doping with bismuth nitrate, affects in opposite manner the XRD patterns. As seen, the annealing increases, while the doping decreases the films texture.

Fig. 2. SEM image of PbS films prepared with oxidant agent.

20 30 40 50 60 70 80

0

500

1000

1500

2000

2500

20 30 40 50 60 70 80

0

100

200

300

400

500

600

700

Inte

ns

ity

(a

,u)

1

2θ

PbS with additives

2

(222)(211)(220)

(200)

1:As deposited (without

annealing)

Inte

nsi

ty(a

,u)

2θ

PbS Without additives

12

2:Annealed at 80° C

for 72 h

(111)

Fig. 3. Effect of annealing on the XRD patterns of PbS films.

20 30 40 50 60 70 80

0

100

200

300

400

500

600

700

Inte

nsi

ty(a

,u)

2θ

With Bi(NO2)3

PbS with NH2OH.HCl

Without Bi(NO2)3

Fig. 4. Effect of doping with Bi(NO2)3 on the XRD patterns of PbS films.

3.2. Optical Properties Fig. 5 and Fig. 6 show the transmittances (Tcorr)

and (αhν)2 plots of PbS films prepared with different baths and annealed at 80°C for 72 h respectively.

0,5 1,0 1,5 2,0 2,5 3,0 3,5

0,0

0,2

0,4

0,6

0,8

1,0

0,4 0,6 0,8 1,0 1,2 1,4 1,6 1,8

104

105

hν (eV)

α (c

m-1)

2

1 3

1: without additives_0,7µm:Eg=0.42eV2: NH

2OH.HCl_1µm:Eg=0.50eV

3: NH2OH.HCl + Dopage Bi_450nm:Eg=0.56eV

λ(µm)

Τ co

rr

2

1

3

Fig. 5. Transmittances of PbS thin films prepared with different bath solution.

0,35 0,40 0,45 0,50 0,55 0,60 0,65 0,700

1x108

2x108

3x108

4x108

5x108

1: without additive :Eg=0.42eV2: NH

2OH.HCl :Eg=0.50eV

3: NH2OH.HCl + Dopage Bi :Eg=0.56eV

hν(eV)

(α.h

ν)2

1

2

3

Fig. 6. Plot of (αhν)2 for PbS thin films prepared with different bath solution.

The inset shows the absorption coefficient of these films. As shown, oxidant and doping affect the optical properties. The optical band gap was calculated from the spectral absorption near the fundamental absorption edge. The direct band gap of all synthesized PbS films was estimated using the Tauc relation given as follow

( )2

1

Eh gh −= ννα , (1)

where α is the parameter which depends on the transition probability. For direct transition in the fundamental absorption, (αhν)2 have linear dependence on the photon energy (hν). The intercept on energy axis gives the direct band gap energy. Based on the optical transmission measurements, we


109

have obtained the direct band gap energy for PbS films grown by different baths, which is 0.42 eV for films prepared without additive (referred as 1), it becomes 0.5 eV for films prepared with oxidant (referred as 2) and 0.56 eV for films prepared with doping agent (referred as 3).

3.3. Surface State Model

Photoconductivity in PbS films can be explained by two models depending upon whether the semiconductor is single-crystalline or polycrystalline, which are recombination and barrier models respectively. The recombination model assumes that change in conductivity on illumination results from the change in the number of conducting electrons or holes per unit volume. In the barriers model, it is assumed that illumination produces little or no change in the density of charge carriers but an increase in their effective mobility. A large number of surface defects are produced along the crystalline boundaries. These surface defects capture electrons from the interior of the single-crystalline and produce space charge barriers. Illumination reduces the number of electrons in the surface defects and thereby lowers the barrier height. In real polycrystalline material, the observed photoconductivity may be due to a combination of both recombination and barrier processes. One of these combination models is the surface state model proposed by [11-12] (Fig. 7).

Zd Ei

Ea

d

W l

Z

Y

X

O

RL

IPH

VPH Vapp

Ohmic contacts

surface depletion region Zd

Bulk

regi

on

Infrared incident radiation IR

Appl

ied

elec

tric

fiel

d

surface (internal) electric field

Fig. 7. Schematic structure of surface state model [11].

This model treats the free surface with different manner of precedent model [13-14]. It was characterized by surface state density (Nt), unique energy level (Et) associated with defect localized in the forbidden gap and effective cross section (σ) rather than the surface recombination velocity. In this model, the barrier height at the semiconductor free surface is modified under photonic excitation. It was suggested that recombination influence directly the quantum efficiency of detector. All models accept that, the role of oxidant is assumed to introduce a trapping state that inhibits recombination; these traps

capture the minority carriers and thereby extending the life time of material. As mentioned above, without the sensitization step, lead sulfide has very short life time and a low response.

The free surfaces of semiconductor constitute a localized discontinuity in comparison to bulk electronic states. The introduction of surfaces gives rise to new eigenstates for which the wave function is localized at the surface, the so-called surface states. The presence of free surface constitutes an intrinsic defect, which altered dramatically the lifetime of device. The extrinsic defect is characterized by the oxidation, adsorption and contamination. In semiconducting materials, the defects is generally charged induced a bend bending in the energetic diagram, however, a potential barrier is formed under the free surface, the electric field separate the charge carriers (é-h) in two opposite direction, and create a depletion region (Fig. 8). The majority carriers created by irradiation have two ways: scanned by the é field and contribute to photocurrent, or trapped with surface state and contribute to noise.

BC

NT

BV Eb

EFI

EFP

Et P-PbS

0.41eV

≅ 3µm

Fig. 8. Energetic diagram.

The concentration of excess carriers density; the key parameter of this model, allows us to explore all the theoretical equations of detectors figure of merit. This parameter is calculated with a self-consistent way, taking into account the resolution of continuity equation of majority and minority carriers in the depletion and neutral (Bulk) region of material, given by [11-12].

( )'

2'expexp)'(

2

exp)(

dzL

zzz

L

zzzG

D

L

L

zzBzn

z

z n

d

n

d

n

n

n

dn

d

++−−−−

+

−−=Δ

, (2)

where Bn is the concentration of excess carriers density at free surface (z=0), which is determined by the appropriate conditions of limits [15], Zd, Ln, Dn and G(z) are width of depletion region, diffusion length, diffusion coefficient and generation rate respectively. Detailed calculation of the surface analysis, optical generation and photoconductor performances (Signal, Spectral response or responsivity and specific detectivity) is presented respectively in the appendix.


110

3.4. Photo Detection Performances Measurement and Simulation

3.4.1. Measurement of Frequency Response

and Time Constant of Detector

Fig. 9 shows the setup measurement test to determine the variation of signal voltage as a function of chopping frequency. Select a low chopping frequency (50 Hz). Tune the wave analyzer to this frequency and record the signal voltage. Repeat this process at successively higher frequencies. All measurements are normalized to the initial low

frequency measurement and plotted as shown in Fig. 10. This curve can be used to determine time constant of the detector, which provides a convenient number to describe the speed with which a detector responds to a change in incident flux. The variation of response with frequency can be described by

222

0

41 τπ f

RR f

+= ,

(3)

where τ is the time constant.

Blackbody Temperarure

Controller

Blackbody Source (500°K)

Chopper variation speed

Detector Preamp

Bias Supply

Wave analyzer

Oscilloscope

Detection synchrone

Frequencecontroller

Radian power

Fig. 9. Basic test set for frequency response measurement.

Taking f1 as the frequency at which the responsivity is 0.707 times its low-frequency value, the time constant is given by

f 12

1

πτ =

(4)

From the frequency response curve for the photo-detector developed by solution Bath 2, shown in Fig. 10.

10 100 1000

90

100

R%

f(Hz)

Photoconductor of Bath2

S:0.3 Cm2

RL=100KΩΔf=50HzOuverture:0.2inch

f1=640Hz

Fig. 10. Frequency response of PbS photoconductor

prepared by Bath 2.

The value of f1 is about 640 Hz: the calculated value of time constant is τ=0.25 ms, the corresponding carriers diffusion length is about Lp=617 µm the calculated value of time constant is in agreement of those published in the literature of PbS detectors at ambient temperature, which varied between 0.1 ms and 0.5 ms [16-20].

3.4.2. Signal, Spectral Response and Detectivity Measurement and Simulation

In order to investigate the influence of oxidant and doping on detection and capabilities of PbS films, three photoconductive detectors prepared with and without additives (Table 1) have been analyzed. To avoid the Flicker noise, manifested at low frequency, the chopping frequency at 400 Hz and the total polarization voltage at 50 V have been fixed. The test set used provides radiant flux in a very narrow spectral band centered about any desired wavelength (λ). There are three functional controls on the monochromator:

• The centered λ of the exciting beam; • The width of the spectral interval centered at λ; • The amount of flux passing through the

monochromator.


111

A constant flux per unit λ interval should be maintained for any selected λ center in the interval of [1-4 µm]. According to application data sheet of the infrared source, the power of infrared radiation is calculated manually by assuming the IR source as a blackbody at 1230 °C (the IR source is a Silicon Carbide SiC emitted in the range of 0.7-28 µm, the irradiance of this source is nearly the same of blackbody heated at 1230 °C, particularly in 0.7-3 µm spectral range) [21].

It is interesting to notice, here, that the noise is measured in the absence of light, with lock-in amplifier and spectrum analyzer under equivalent noise band width of Δf=50, the effective value of noise is done by Vn/ (Δf )1/2 [22]. The values of Vn (noise voltage in rms) for photoconductors prepared with three baths are 55.15, 86.3 and 63.6 µV respectively.

This measurement of noise is crucial in the calculating of spectral specific detectivity, which depends of the ratio of spectral response and noise.

Fig. 11, Fig. 12 and Fig. 13 show the signal, spectral response and specific detectivity of these photoconductive detectors fitted with surface state model [11-12]. The oxidant enhances the capabilities of detection comparatively to detectors prepared without oxidant. The doping increases again these capabilities and decreases the peak wavelength.

0 1 2 3 4 5

0

1

2

3

4

5

0 1 2 3 4 5-0,5

0,0

0,5

1,0

1,5

2,0

2,5

3,0

3,5 Bath1

Vp

h(m

V)

λ(µm)

Vp

h (

V)

Bath2

λ(µm)

Bath3

Experimental pointsSurface state model

Fig. 11. Simulation of signal with surface state model.

0 1 2 3 4 5

0,0

4,0x104

8,0x104

1,2x105

1,6x105

0 1 2 3 4 5

0

20

40

60

80

100

R(V

/ W)

λ(µm)

Bath1

Bath2

R(V

/W)

λ(µm)

Bath3

Experimental pointsSurface state model

Fig. 12. Simulation of spectral response with surface state model.

0,1 1 10109

1010

1106

107

λ(µm)

Bath1

Bath2

D* (C

m,

Hz1/

2 W-1)

λ(µm)

Bath3

Experimental points

Surface state model

Fig. 13. Simulation of spectral detectivity with surface state model.

The best adjustments are obtained with the data reported in Table 2. It should be noted that, the confrontation between experimental and calculated data of signal is excellent, but a small discrepancy in spectral response and specific detectivity for the first point of measurement is observed. It is probably due to high fluctuation of signal at low wavelength. Also, the parameters issued with this simulation are in agreement with the published ones in the literature [16-20, 22, 24].

Table 2. Simulation results.

Detector Bath 1 Bath 2 Bath 3Simulated parameters

Eg (eV) 0.43 0.43 0.48 Lp (µm) 500 500 400 Zd (nm) 30 10 10 Na (cm-3) 51017 51017 51017 Et (eV) 0.42 0.42 0.47 σ (cm2) 2.5510-14 3.410-16 2.4510-17 Nt (cm-2) 1.51012 51011 51011

Introduced parameters d(µm) 0.7 1 0.45

Ad (cm2) 0.4 0.3 0.4 RL (MΩ) 0.1 0.1 0.25

4. Conclusions In the present work, detectors based on sensitized

thin films, growth by chemical bath deposition are studied. The photoelectrical characterization allowed that, detectors developed without oxidant has approximately very low performances (Signal, Responsivity and Specific Detectivity) compared with those developed with oxidant. The use of doping agent increases again these performances along with decreasing the peak position wavelength of detector.

A simulation study of proposed infrared photoconductors, using surface state model, has been also presented. The plots show the ability of this model to adjust their performance behavior. The extracted physical parameters are in agreement with those published in literature.


112

Appendix A. Surface Analysis

Assuming that Q is the absolute charge at the surface absorber material, then

)1( feNNeZQ tad −==

)1( fN

NZ

a

t

d−= ,

(A1)

where Zd is the depletion region width and f the occupation probability of the donor energy level given by [11] [15]:

)cosh(2)0()0(

)exp()0(

00

0

kTEE

nppnn

kTEE

nnnf

tii

iti

−++Δ++Δ

−++Δ=

(A2)

where Δn(0)(Δp(0)),ni, Et and Nt are the electron (holes) excess carriers concentrations at the surface, intrinsic carriers concentration, energy level of surface states and surface states density, respectively. The electron (hole) concentration at the surface n0 (p0) is given by:

)exp(2

0 kT

E

N

nn b

a

i= (A3)

)exp(0 kT

ENp b

a −= (A4)

The barriers height is given by:

2

2 da

b ZeN

Eε

= (A5)

B. Optical Generation

The optical generation rate is given by:

eFzG fphz)( αηα −= , (A6)

where Ff is the front surface flux modeled by Plank function, α is the absorption coefficient and the η is the quantum efficiency calculated by [17]:

[ ]τ

τη α

αeff

d

d

er

er−

−

−−−=

.1

1).1( (A7)

In which r is the reflection coefficient and τeff is the effective lifetime given by [17]:

dseff z.

111

τττ+= (A8)

With τ, and τs are the bulk and surface electron

lifetime respectively. C. Signal, Responsivity and Detectivity

The conductance of film (Fig. 7) is given by:

( )dzznl

eZg

dn Δ=Δ

0

μ (A9)

The signal in current and voltage is given by:

Lphph RIV

VgIph

.

.

=Δ=

, (A10)

where V is the bias voltage.

The spectral response is given by:

( )2

0

21 τπf

RR

+= (V/W),

(A11)

where R0:

inc

ph

P

VR =0

(V/W), (A12)

f is the modulation frequency and τ the bulk carrier lifetime (τ = LP

2/DP).

The specific detectivity is given by:

n

dV

n

di

V

fAR

i

fARD

Δ=

Δ=* (A13)

The total noise expression is done by

22

/1

22

johfRGniiii ++=

− (A14)

fR

kTf

f

iB

fAdqgAEqqGi

d

f

dthdqn

Δ+Δ

+Δ+=

4

).(4

21

22 η

(A15)

The 1/f noise is given by [25] where

dtf AN

CB =/1

, C=0.1, Ad is the detector area, Nt is

the surface state density and Δf is the electrical bandwidth of detector.

At room temperature and high modulation frequency, the dominant compound of noise is the thermal generated-recombination part given by [26]:


113

figqin Δ= 02

4 , (A16)

where i0 is the dark current, Δf is the noise equivalent frequency bandwidth given by 1/(2tint) where tint is the electronic integration time and g is the gain of photoconduction. References [1]. S. Kumar, T. P. Sharma, M. Zulfequar, M. Husain,

Characterization of vacuum evaporated PbS thin films, Physica B, Vol. 325, 2003, pp. 8-16.

[2]. J. N. Humphrey, Optimum utilization of lead sulphide detectors under diverse operating conditions, Appl. Opt., Vol. 4, Issue 6, 1965, pp. 665-675.

[3]. T. H. Johnson, H. T. Cozine, B. N. McLean, Lead selenide detector for ambient temperature operation, Appl. Opt., Vol. 4, Issue 6, 1965, pp. 693-696.

[4]. D. E. Bode, Lead Salt Detectors, in Physics of Thin Films, Vol. 3, (G. Hass and R. E. Thun, Eds.), Academic Press, New York, 1966, pp. 275-301.

[5]. T. S. Moss, G. J. Burrel, B. Ellis, Semiconductor Optoelectronics, Butterworths, London, 1973.

[6]. S. Kouissa, A. Djemel, M. S. Aida, M. A. Djouadi, Characterization and Simulation of PbS Photoconductors Prepared by Chemical Bath Deposition, in Proceedings of the 6th International Conference on Sensor Device Technologies and Applications (Sensordevices’15), Venice, Italy, 23-29 August 2015, pp. 1-6.

[7]. E. Pentia, L. Pintilie, C. Tivarus, I. Pintilie, T. Botila, Influence of Sb3+ ions on photoconductive properties of chemically deposited PbS films, Materials Science and Engineering, B80, 2001, pp. 23-26.

[8]. R. Dalven, A review of semiconductors properties of PbTe, PbSe, PbS and PbO, Infrared Physics, Vol. 9, No. 4, 1969, pp. 141-184.

[9]. V. M. Simic, Z. B. Marinkomc, Influence of impurities on photosensitivity in chemically deposited lead sulphide layers, Infrared Physics, Vol. 8, 1968, pp. 189-195.

[10]. E. Pentia, L. Pintilie, T. Botila, I. Pintilie, A. Chaparro, C. Maffiotte, Bi influence on growth and physical properties of chemical deposited PbS films, Thin Solid Films, Vol. 434, 2003, pp. 162-170.

[11]. S. Kouissa, M. S. Aida, A. Djemel, Surface state simulation model for photoconductors infrared detectors, Journal of Materials Science: Materials in Electronics, Vol. 20, 2009, pp. 400-406.

[12]. S. Kouissa, A. Djemel, M. S. Aida, Surface state dependence of PbS and PbSe infrared noise and detectivity, Journal of Materials Science: Materials in Electronics, Vol. 23, No. 12, 2012, pp. 2083-2088.

[13]. J. C. Slater, Barrier theory of the photoconductivity of lead sulfide, Phys. Rev, Vol. 103, No. 6, 1956, pp. 1631-1644.

[14]. R. L. Petritz, Theory of photoconductivity in semicon-ductors films, Phys. Rev, Vol. 104, No. 6, 1956, pp. 1508-1516.

[15]. A. Djemel, R. J. Taronto, J. Castnaing, Y. Marfaing, A. Nouiri, Study of electronic surface properties of GaAs in cathodoluminescence experiment, Phys. Stat. Solid (a), 1998, Vol. 168, No. 2, pp. 425-432.

[16]. E. Dereniak, G. D.Boreman, Infrared Detector and Systems, Wiley Interscience, 1996.

[17]. A. Rogalski, Infrared photon detectors, SPIE Optical Engineering Press, 1995.

[18]. William L. Wolfe, George J. Zissis (Eds.), The infrared Handbook, IRIA, 1993.

[19]. J. Piotrowski, A. Rogalski, High-operating-temperature infrared photodetectors, SPIE Optical Engineering Press, 2007.

[20]. R. Hudson, Infrared system engineering, John Wiley & Sons. Inc, 1969.

[21]. Calculation of Radiant Flux for Restricted Wavelength Regions, SLS. APP 0005, Application Data Sheet, Infrared Industries, Inc., 1989 (Private Document).

[22]. J. D. Vincent, Fundamentals of Infrared Detector Operation and Testing, Wiley-Interscience, 1990.

[23]. E. Dereniak, G. D. Boreman, Infrared Detector and Systems, Wiley Interscience, 1996.

[24]. N. B. Kotaiya, A. J. Kothari, D. Tiwari, T. K. Chaudhuri, Photoconducting nanocrystalline lead sulphide thin films obtained by chemical bath deposition, Appl. Phys A, Vol. 108, Issue 4, 2012, pp. 819-824.

[25]. W. Z. Shen, A. G. U. Perera, Low frequency noise and interface states in GaAs homojunction far- infrared detectors, IEEE Trans-Electron Devices, Vol. 46, No. 4, 1999, pp. 811-814.

[26]. E. Rosencher, B. Vinter, Optoélectronique, Edit. Masson, Paris, 1998.

___________________



114



Amplitude to Phase Conversion Based on Analog Arcsine Synthesis for Sine-cosine Position Sensors

* Mohieddine BENAMMAR, Antonio Jr. GONZALES College of Engineering, Qatar University, P. O. Box 2713, Doha, Qatar

* Tel.: +974-4403-4203, fax: +974-4403-4201 * E-mail: [email protected]

Received: 31 August 2015 /Accepted: 5 October 2015 /Published: 30 October 2015 Abstract: Sinusoidal encoders, including Hall effect sensors, are position sensors that provide analog sine and cosine signals of angular position. All schemes used for converting these signals into measure of the angle require either trigonometric or inverse trigonometric function implementation. The proposed converter is based on the use of the alternating pseudo-linear segments of the sensor signals together with a simple and effective linearization technique. The theoretical absolute error of non-linearity of the converter is 0.05 degree over the full 360 degree range. The converter may be implemented numerically or electronically. The paper describes the proposed method, full details of its analog implementation, and experimental results obtained using both a computer-based sensor emulator and a Hall effect sensor. Results demonstrate agreement between theory and experimental results. Copyright © 2015 IFSA Publishing, S. L. Keywords: Amplitude-phase conversion, Inverse sine synthesis, Linearization, Position measurement, Sinusoidal encoder, Hall effect sensor. 1. Introduction

Rotational speed and position measurement and

control is often required in various applications in industry, military, avionics, communication and other fields. Sinusoidal encoders whether operating on optical, inductive, Hall effect or magnetoelectric principles produce quadrature electrical signals in which the unknown angle is encoded [1-10]. In practice, it is common to observe phase and amplitude imbalances in the sensor signals; therefore the sensor outputs may be written as:

( )

+×+=

×=

)β(θ cos α1)θ(

)(θ sin )θ(

AU

AU

C

S, (1)

where A is the maximum amplitude of the sinusoidal component of US(θ), θ is the shaft angle of the rotor

of the sensor, and α and β are the amplitude and phase imbalances respectively. Note that some sensors may produce signals with dc offset components; however, these may easily be removed before further processing. Amplitude and phase balancing may be achieved in a number of ways [11-14]. In this work, we propose balancing by generating a perfect cosine signal, from US(θ) and UC(θ), with the same amplitude as the sine signal,

×=×++

=

×==

)(θ cos )θ(βtanβcos)α1(

)θ()θ(ˆ

)(θ sin )θ()θ(ˆ

AUU

U

AUU

SC

C

SS

(2)

The sensor requires a suitable converter in order

to determine θ from its signals (2); many open loop and closed loop conversion schemes have been described in literature. Some open loop converters




115

based on the linearization of the pseudo-linear segments of sinusoidal signals have been described [10, 15-23]. Ratiometric techniques based on arctangent method are also used in open loop converters [24-27]; these require the use of look up tables or linearization techniques. Other open loop converters based on the use of reference ac signals and time measurement techniques have also been described [8-9, 28-30]. Closed loop converters employ the phase-locked loop (PLL) technique [31-38].

In the present work, we present an open-loop amplitude-to-phase converter that uses a simple arcsine synthesis technique which can be easily implemented with few standard electronic components. The remainder of the paper is organized as follows. In Section 2, the principle and theory of operation of the proposed converter, including a novel linearization technique, are described in details. A dedicated signal shaping technique used for implementing the converter is also described. In Section 3, the practical implementation of the converter and experimental results are described. Section 4 concludes the paper. 2. Proposed Converter

The basic principle of the proposed converter (Fig. 1) is based on making use of the alternating pseudo-linear segments of the sensors signals (2) in order to produce a signal U0(θ) which is almost proportional to the unknown angle θ in each of its four quadrants (Fig. 2), as has been reported in previous works [5-19, 23]. The four quadrants are identifiable using two binary outputs whose states depend on the signs of the sum and difference of the transducer signals (i.e., LOW state for negative and HIGH state for positive values):

( )( )

⊕=

+=

)θ(ˆ-)θ(ˆsign )(θ1)(θ0

)θ(ˆ)θ(ˆsign )(θ1

CS

CS

UUBitBit

UUBit (3)

The signal Bit0(θ) is used to control the

multiplexer that selects the pseudo-linear segments of the sensor signal to produce U0(θ),

( ) ( ))θ(ˆ)(θ0 )θ(ˆ)(θ0)θ( SC0 UBitUBitU ×+×= (4) The sign of ÛS(θ) - ÛC(θ) is used together with a

synchronous rectifier to generate a rectified signal U0R(θ), a sawtooth-like waveform made up of four identical and positive-slope sections as shown in Fig. 1 and Fig. 2. The signal U0R(θ) requires further linearization in order to obtain a piecewise linear output U0RL(θ). The angle may be determined from the linearize signal V0RL(θ) using a simple linear equation within each quadrant of input angle with minimal non-linearity error. Within the full 360 degree range, the computed angle noted θC (in degree) is determined from:

( )[ ])θ(4)θ(2)θ(245 100 BitBitU/Aθ RLc ×+×+××= ° (5)

Note that in the range 315° to 360°, the angle

determined using (5) is negative (i.e., measured clockwise). Evidently, the residual error (θC-θ) depends on the quality of linearization scheme. Previous works have presented various linearization methods with different degrees of complexity and precision [10, 15-22]. These schemes that use multipliers/dividers add complexity and cost, particularly, in analog implementation. In this work, the proposed linearization method is based on signal shaping techniques.

Signal shaping networks are usually associated with trigonometric and inverse trigonometric function synthesis. In this application, the pseudo-linear segments of U0R(θ) belong to sinusoidal signals. Hence, linearization requires arcsine function synthesis, or in other words sine-to-triangle conversion. It is well known that signal shaping can be implemented using well established piece-wise linear approximation techniques involving diodes and/or transistors [39-41]. Other techniques based on differential pairs of transistors offer attractive and simpler solutions that closely approximate sine function [42-48]. In the present application, a signal converter based on the non-linear I-V characteristics of the base-emitter junction of low power bipolar junction transistors. The basic circuit diagram of the proposed linearization scheme is shown in Fig. 3. This is a translinear sine-triangle converter inspired from the triangle-sine scheme presented in [48].

Fig. 1. Basic diagram of the proposed converter.


116

Fig. 2. Converter signals: input signals ÛS(θ) and ÛC(θ), non-linearized signals U0(θ) and U0R(θ) and binary outputs

Bit0(θ) and Bit1(θ).

Fig. 3. Basic principle of the proposed linearizer.

The following analysis assumes that the transistors in the scheme of Fig. 3 are matched. Because of symmetry, the current source of magnitude 2I ensures that the individual bias currents I in the two trans-diodes are equal. The sinusoidal segments of the rectified voltage U0R(θ) applied to the linearization circuit result in a current i= -U0R(θ)/R. Note that this current should always be lower than I; the maximum value should correspond to the maximum amplitude A of the sinewaves from which the segments of U0R(θ) are extracted and therefore imax=A/R=I. By invoking the Shockley equation ( TBE VV

SE eII /≈ where VT is the thermal voltage,

approximately 26mV at room temperature, and IS is the saturation current) for the base-emitter junctions of the transistors, we can write:

( )

( )

20 1 2 1 2

1

21 2

1

1 2 / /

1 2 / / ln ln

RL BE BE

T TS S

RU R R i V V

R

R I i I iR R i V V

R I I

= − + × + −

+ −≈ − + × + −

( )

( )

21 2

1

21 2

1

0R 0R1 2 2 1

1 1 2 0R

0R 0R22

1

1 2 / / ln

1 /1 2 / / ln

1 /

1 /2ln

1 /

1 / ( )2 1 ln

1

T

T

T

T

R I iR R i V

R I i

R i IR R i V

R i I

U U IRR R R RV

R R R R U IR

U U IRRR V

R R U

+ ≈ − + × + − + ≈ − + × + −

++≈ − × + + −

+≈ − − + − 0R / ( )IR

( )[ ]

−+

+−−≈

−+×

+−=

)IR/(U1

)IR/(U1lnV

R

R1U

R

R2

VViR//R2R

R1U

T1

22

2BE1BE211

2RL0

0R

0R0R

(6)

Note that the emitter currents for Q1 and Q2 are

(I+i) and (I-i) respectively. The Maclaurin series expansion of the logarithmic part of (6), which is of the form ( ) ( )( )xx −+ 1/1ln with x<1, yields:

+

+

+−

++−=

+

+−−=

∞

=

+

...5

1

3

112

1)(

2

12

))/((12

2

5

0R

3

0R

1

2

1

22

0R

0

120R

1

20R

20RL

RI

U

RI

UV

R

R

VR

RIR

RI

U

n

IRUV

R

RU

R

RU

T

T

n

n

T

(7)

The objective is to convert the sinusoidal signal

U0R(θ) into a sinusoidal current which is in turn converted into a triangular output voltage U0RL(θ):

( )

+

+

+

×−=

+−=

×−=

∞

=

+

...40

3

6

1

))/(()12()!(4

)!2(

)/(arcsin

5

0R

3

0R0R1

0

120R21

0R10RL

RI

U

RI

U

RI

UK

RIUnn

nK

RIUKU

n

nn

(8)

By comparing (7) and (8), and considering the

first three terms of the expansions, we can deduce the conditions for a successful conversion:

5/1240/ 3or 3/126/

4/π

2/1

1

21

1

21

10RLmax

121

2

+=

+=

=

=+

+

TT

T

VR

RKV

R

RK

KU

KIRVR

R

(9)

Note that the maximum amplitude of U0RL(θ)

occurs at the peak of U0R(θ) which is equal to 2-½A; therefore U0RLmax=K1×arcsin(2½A/(RI))= K1π/4. Since K1 cannot satisfy both requirements defined by the last condition of (9), a compromise should be sought to keep K1 within the boundaries defined by the last condition of (9):


117

( ) ( ) 15//80 /14 12112 TT VRRKVRR ≤≤+ (10)

Straightforward analysis shows that when considering (10), the two conditions in (9) may be re-written as:

1π

1π20

15

15

25//

maxRL0

1

2maxRL0

21

−≤≤−

≤≤

TT

TT

V

U

R

R

V

U

I

VRR

I

V

(11)

Assuming operation at room temperature

(VT=26 mV), a current I= 0.25 mA, sensor signals amplitude A=10 V, and choosing a peak output voltage U0RLmax=10 V, simple analysis yields:

126/90

173//104

12

21

≤≤

Ω≤≤Ω

RR

RR (12)

Since A=RI, the required value for

R=10 V/0.25 mA=40 kΩ. Simulation of U0RL(θ) according to (6) suggest that with R=40.04 kΩ, R1=130 Ω and R2=14130 Ω, the residual error in the determination of θ using (5) is within ±0.048° in the full 360° range as shown in Fig. 4.

Fig. 4. Simulation of the performance of the proposed linearization scheme. U0RL(θ) is computed according

to approximation in (6) with A=10 V, I=0.25 mA, R=40.04 kΩ, R1=130 Ω, R2=14130 Ω, and VT=26 mV.

Note that: 1) The inversion of U0RL(θ) with respect to U0R(θ)

is due to the inverting configuration of the operational amplifier;

2) The amplitude A of the sensor signals is assumed to be 20 Vpp and therefore U0R(θ) is 20/2½ Vpp.

It is important to note that the operating temperature usually affects signal shaping schemes; this can be seen in (6) through the thermal voltage term. However, some degree of temperature compensation may be achieved by selecting a suitable current source element with appropriate positive temperature coefficient.

It is important to note that since the proposed scheme is open loop, the bandwidth is only limited by the dynamics of the components used, mainly the operational amplifiers and analog switches. In all cases, the bandwidth should be much higher than what is required in practical positioning applications: e.g., for a maximum rotational speed of 24000 rpm (i.e., considered to be very high for real applications), the corresponding signals’ frequency would be 24000/60=400 Hz only which does not pose any problem even for standard components.

3. Experiment The practical converter is a straightforward

implementation of the scheme of Fig. 1 using few standard electronic components. Fig. 5 depicts the detailed circuit implementation of the proposed converter. Matched dual NPN transistors (MAT01) and ±10 V reference source (REF01 IC with an op-amp inverter with a gain of -1) have been used in the implementation of the linearizer. In order to simplify implementation of the linearizer without affecting performance, the current sources shown in Fig. 3 have been omitted, and the required currents (I and 2I) have been derived from a reference voltage as shown in Fig. 5. The current I is set by RI (i.e., I=10 V/RI). Because the 10 V reference voltage is much greater than the voltage VB at the bases of transistors Q1 and Q2, the current through R2I is nearly constant. Extensive computer simulation using Multisim software showed that the best performance of the linearizer (i.e., lowest absolute error of the converter) is achieved with RI =39.5 kΩ and R2I = 18.7 kΩ, resulting in I≈252 μA. The remaining components for the converter are straightforward implementation of the diagram of Fig. 1. The offset cancellation, amplification, amplitude equalization and phase correction of the encoder output signals are not shown. The amplitudes of the converter’s input signals ÛC(θ) and ÛS(θ) are 20 Vpp, and the linearized output is 20 Vpp.

The proposed converter has been characterized under controlled environment using a sensor emulator built around LabVIEW and a high performance USB-6259 DAQ board from National Instruments, characterized by four 16-bit, 2.8 MS/s analog output channels (Fig. 6).


118

Fig. 5. Detailed implementation of the proposed open loop converter.

Fig. 6. Experimental set-up incorporating a sensor emulator for the characterization of the converter.

The sensor emulator was created under LabVIEW

environment and provided the two input signals ÛC(θ) and ÛS(θ) of the converter and an additional analog sawtooth signal UREF used as reference for assessing the linearity of the converter output signal U0RL(θ). The sensor emulator enables characterization of the proposed converter under various controlled conditions, including transient condition which is not practically possible to perform with a real sensor. The various signals have been recorded using a Tektronix MDO3024 oscilloscope. Further tests of the converter have been conducted with a real sensor as shown below.

Fig. 7 shows the converter input signals ÛS(θ) and ÛC(θ), the signal U0(θ) obtained from the multiplexer, the rectified signal U0R(θ) and the linearized analog output U0RL(θ) and binary outputs Bit0(θ) and Bit1(θ); the sensor was simulated to rotate at a fixed forward rotational speed of 600 rpm (i.e., equivalent to sensor signals frequency of 10 Hz).

Fig. 8 depicts the effectiveness of linearization achieved by the proposed linearizer whose inputs and outputs are -U0R(θ) and U0RL(θ), respectively. Notice that the scales used for the two traces are different.

Fig. 9 depicts the input and output signals of the converter when the sensor rotates at a reverse speed of 1200 rpm. By comparing traces of Fig. 7 and Fig. 9, it is clear that the converter provides unambiguous and absolute measurement of the angle.

Fig. 7. Converter signals at forward speed of 600 rpm.

In order to assess the linearity of the segments of the 20 Vpp output U0RL(θ) of the converter, the setup of Fig. 6 has been used to generate a reference sawtooth signal UREF synchronized with ÛC(θ) and ÛS(θ), and therefore with U0RL(θ). An instrumentation amplifier was used to determine the deviation of U0RL(θ) from UREF as per the diagram of the setup in Fig. 6. The results are depicted in Fig. 10 where U0RL(θ) and UREF are shown using the same scale but with slightly different ground levels for clarity (i.e., in order to distinguish both signals). The deviation of U0RL(θ) from UREF is also shown; this


119

residual error of linearity is within 100 mV peak-to-peak without fine adjustment of the various components of the converter, especially in relation to the linearizer.

Fig. 8. Comparison between the linearized signal U0RL(θ) and the non-linear segment U0R(θ) showing the effect

of linearization.

Fig. 9. Converter signals at reverse speed of 1200 rpm.

Fig. 10. Characterization of the resolver using the setup of Fig. 6.

Fig. 11 depicts the excellent transient response of the converter by stepping the angle from 0° to 150° using the sensor emulator of Fig. 6. Clearly this would not be possible using a real sensor. Note that the step change from 0° to 150° has been chosen to demonstrate the performance of the converter even when the resulting output of the converter angle changes from one quadrant to the another; in this case from the first (θ=0°) to the third (θ=150°)

segments of U0RL(θ), see Fig. 4. The results suggest an almost immediate response of the output, as expected for an open loop converter.

Fig. 11. Transient response of the input and output signals of the converter by stepping the angle from 0° to 150°.

The transient response performance depicted at a time scale of 10μs/division in Fig. 12 indicate that the input signals ÛC(θ) and ÛS(θ), which are generated by the sensor emulator, presented some jitter and did not change in a step fashion due to slow response LabVIEW and associated DAQ card. Hence the apparent first-order-like response of the converter (with a time constant of merely 10μs) may not be attributed to the converter itself.

After its characterization with the sensor emulator, the converter was tested with a Hall effect sensor (model HSCB22) as shown in Fig. 13. This sensor produces two quadrature sine and cosine signals with a nominal peak-to-peak amplitude of 4 V and an offset of 2.5 V. Appropriate analog signal conditioning (not shown in Fig. 5) is applied to the sensor signals in order to

1) Remove the offset; 2) Correct the phase between its signals using (2); 3) Adjust their amplitudes to 20 Vpp. The sensor was mounted on a miniature rotary

table (model A5990TS with an integral 90:1 gear ratio and 100 arc-second accuracy) that enabled precise manual control of the angle θ.

Fig. 12. Transient response of the input and output signals of the converter at a time scale of 10 μs/div.


120

Fig. 13. Experimental setup for testing the converter with commercial Hall Effect encoder HSCB22.

Fig. 14 show the experimental results obtained with the setup of Fig. 13 in the angle range 0° to 180°. The results indicate that the residual error of non-linearity of the converter and associated encoder is within 0.15° peak-to-peak.

-10V

-5V

0V

5V

10V

Output, U0RL(θ)

-0.10°

-0.05°

0.00°

0.05°

0.10°

0° 45° 90° 135° 180°Angle, θ

Error, θC-θ

Fig. 14. Characterization of the converter using the experimental setup of Fig. 13.

The overall absolute error in the determination of the angle θC from the converter output U0RL(θ) was lower than 0.09° in the range 0° to 180° of input angle; this compares well with the theoretical error estimated above in Fig. 4. Despite the simplicity and low cost of the proposed scheme, its error compares well with those of other more complex closed loop and open loop schemes reported in literature, e.g., 0.01° in [15], 0.10° in [17], 0.04° in [20], 0.18° in [38].

4. Conclusions

In this paper, low-cost and simple-to-implement open-loop method for amplitude-to-phase conversion was proposed for use with sinusoidal Hall effect sensors. The conversion was based on a simple and effective linearization technique, full theory of which was given. The theoretical error of non-linearity of the converter is below 0.05°. The scheme was implemented using standard electronic components

and was experimentally tested. The results have been excellent and showed that the proposed scheme can be used to measure angles in the full 360° range with an overall error lower than 0.09°. References [1]. N. H. Duc, B. D. Tu, N. T. Ngoc, V. D. Lap,

D. T. H. Giang, Metglas/PZT-Magnetoelectric 2-D Geomagnetic Device for Computing Precise Angular Position, IEEE Transactions on Magnetics, Vol. 49, Issue 8, 2013, pp. 4839-4842.

[2]. D. T. Huong Giang, P. A. Duc, N. T. Ngo, N. T. Hien, N. H. Duc, Spatial angular positioning device with three-dimensional magnetoelectric sensors, Review of Scientific Instruments, Vol. 83, Issue 9, 2012, pp. 095006-1 - 095006-6.

[3]. Y. Netzer, Linear electric encoder with facing transmitter and receiver, US Patent US2004/02525032A1, Dec. 16, 2004.

[4]. P. Kejik, S. Reymond, R. S. Popovic, Purely CMOS angular position sensor based on a new hall microchip, in Proceedings of the IEEE 34th Annual Conference on Industrial Electronics, Orlando, FL, Nov. 10-13, 2008, pp. 1777-1781.

[5]. P. Kejik, S. Reymond, R. S. Popovic, Circular hall transducer for angular position sensing, in Proceedings of the Conference on Transducers Eurosensors XXI, Lyon, France, Jun. 10-14, 2007, pp. 2593-2596.

[6]. K. Bienczyk, Angle measurement using a miniature hall effect position sensor, in Proceedings of the International Students Conference on Electrodynamics and Mechatronics, Opole, Poland, 19-21 May 2009, pp. 21-22.

[7]. D. Wang, J. Brown, T. Hazelton, J. Daughton, 360 angle sensor using spin valve materials with SAF structure, IEEE Transactions on Magnetics, Vol. 41, No. 10, Oct. 2005, pp. 3702-3705.

[8]. C. S. Anoop, B. George, Electronic Scheme for Computing Inverse-Cosine and its Application to a GMR Based Angle Sensor, IEEE Transactions on Instrumentation and Measurement, Vol. 61, No. 7, July 2012, pp. 1991-1999.

[9]. C. S. Anoop, B. George, Jagadeesh K. V., Tunneling Magneto-Resistor based Angle Transducer, in Proceedings of the 5th International Conference on Sensing Technology (ICST’11), Palmerston North, New Zealand, 28 Nov. – 1 Dec. 2011, pp. 431-435.

[10]. G. Ye, et. al, Design of a precise and robust linearized converter for optical encoders using a ratiometric technique, Measurement Science and Technology, Vol. 25, 2014, pp. 125003-125011.

[11]. D. C. Hanselman, Techniques for Improving Resolver-to-Digital Conversion Accuracy, IEEE Transactions on Industrial Electronics, Vol. 38, No. 6, Dec. 1991, pp. 501-504.

[12]. A. Bünte, S. Beineke, High-performance speed measurement by suppression of systematic resolver and encoder errors, IEEE Transactions on Industrial Electronics, Vol. 51, No. 1, 2004, pp. 49-53.

[13]. J. Zhang, Z. Wu, Automatic calibration of resolver signals via state observers, Measurement Science and Technology, Vol. 25, No. 9, 2014.

[14]. S. Balemi, Automatic Calibration of Sinusoidal Encoder Signals, in Proceedings of the World


121

Congress of the International Federation of Automatic Control (IFAC), Prague, Jul. 2005.

[15]. M. Benammar, L. Ben-Brahim, M. A. Alhamadi, A high precision resolver-to-DC converter, IEEE Transactions on Instrumentation and Measurement, Vol. 54, No. 6, Dec. 2005, pp. 2289-2296.

[16]. M. Benammar, L. Ben-Brahim, Mohd A. Alhamadi, A Novel Resolver-to-360 Degree Linearized Converter, IEEE Sensors Journal, Vol. 4, No. 1, 2004, pp. 96-101.

[17]. M. Benammar, A novel amplitude-to-phase converter for sine/cosine position transducers, International Journal of Electronics, Vol. 94, No. 4, 2007, pp. 353-365.

[18]. M. Benammar, M. Bagher, M. Al Kaisi, Digitally-tuned resolver converter, in Proceedings of the Eurosensors XXIII Conference, Procedia Chemistry, Vol. 1, Issue 1, 2009, pp. 449-452.

[19]. D. Živanović, J. Lukić, D. Denić, A Novel Linearization Method of Sin/Cos Sensor Signals Used for Angular Position Determination, Journal of Electronics in Engineering Technology, Vol. 9, No. 4, 2014, pp. 1437-1445.

[20]. M. Benammar, L. Ben-Brahim, M. A. Alhamadi, M. Al-Naemi, A novel method for estimating the angle from analog co-sinusoidal quadrature signals, Sensors and Actuators-A, Vol. 142, No. 1, March 2008, pp. 225-231.

[21]. V. Riewruja, A. Kaewpoonsuk, OTA-based sine-to-triangular wave converter, in Proceedings of the Conference on Circuits Systems Signal Processing, Vol. 25, 2006, pp.753-765.

[22]. A. Kaewpoonsuk, W. Petchmaneelumka, A. Rerkratn, S. Tammaruckwattana, V. Riewruja, A novel resolver-to-DC converter based on OTA-based inverse-sine function circuit, in Proceedings of the Conference on Society of Instrument and Control Engineers (SICE) of Japan, Japan, 20-22 Aug. 2008, pp. 609-614.

[23]. M. Benammar, A. Gonzales, Simple and Precise Analog Arcsine Synthesis Applied to Amplitude to Phase Conversion for Hall Effect Position Sensors, in Proceedings of the 6th International Conference on Sensor Device Technologies and Applications, (SENSORDEVUCES’15), Venice, Italy, 23-28 August 2015, pp. 13-18.

[24]. M. Benammar, M. Bagher, M. Al Kaisi, Novel Linearizer for Tangent/Cotangent Converter, in Proceedings of the 16th IEEE International Conference on Electronics, Circuits, and Systems (ICECS’09), Hammamet, Tunisia, 13-16 Dec. 2009, pp. 575-578.

[25]. S. C. M. Reddy, K. N. Raju, Inverse Tangent Based Resolver to Digital Converter - A Software Approach, International Journal of Advanced Engineering and Technology, Vol. 4, Issue 2, Sept. 2012, pp. 228-235.

[26]. A. Attaianese, G. Tomasso, Position measurement in industrial drives by means of low-cost resolver-to digital converter, IEEE Transactions on Instrumentation and Measurement, Vol. 56, No. 6, Dec. 2007, pp. 2155-2159.

[27]. S. Sarma, V. K. Agrawal, S. Udupa, Software-based resolver-to-digital conversion using a DSP, IEEE Transactions on Industrial Electronics, Vol. 55, No. 1, Jan. 2008, pp. 371-379.

[28]. L. Ben-Brahim, M. Benammar, M. A. Alhamadi, A resolver angle estimator based on its excitation

signal, IEEE Transactions on Industrial Electronics, Vol. 56, No. 2, Feb. 2009, pp. 574-580.

[29]. L. Ben-Brahim, M. Benammar, M. Alhamadi, N. Alhamadi, M. Alhitmi, A new low cost linear resolver converter, IEEE Sensors Journal, Vol. 8, No. 10, Oct. 2008, pp. 1620-1627.

[30]. M. Benammar, L. Ben-Brahim, K. Besbes, An apparatus for the determination of the angle from its sine/cosine values, UK Patent Application 2471458 (A), 5 Jan. 2011.

[31]. J. Bergas-Jané, C. Ferrater-Simón, G. Gross, R. Ramírez-Pisco, S. Galceran-Arellano, J. Rull-Duran, High-Accuracy All-Digital Resolver-to-Digital Conversion, IEEE Transactions on Industrial Electronics, Vol. 59, No. 1, 2012, pp. 326-333.

[32]. G. A. Woolvet, Digital transducers, Journal of Physics E: Scientific Instruments, Vol. 15, No. 12, Dec. 1982, pp. 1271-1280.

[33]. C. H. Yim, I. J. Ha, M. S. Ko, A resolver-to-digital conversion method for fast tracking, IEEE Transactions on Industrial Electronics, Vol. 39, No. 5, Oct. 1992, pp. 369-378.

[34]. D. A. Khaburi, Software-Based Resolver-to-Digital Converter for DSP-Based Drives Using an Improved Angle-Tracking Observer, IEEE Transactions on Instrumentation and Measurement, Vol. 61, No. 4, April 2012, pp. 922-929.

[35]. N. Noori, D. A. Khaburi, A new software-based method for rotor angle calculation, in Proceedings of the 5th Power Electronics, Drive Systems and Technologies Conference (PEDSTC’14), Teheran, Iran, 5-6 Feb. 2014, pp. 305-310.

[36]. L. Idkhajine, E. Monmasson, M. W. Naouar, A. Prata, K. Bouallaga, Fully Integrated FPGA-Based Controller for Synchronous Motor Drive, IEEE Transactions on Industrial Electronics, Vol. 56, No. 10, Oct. 2009, pp. 4006-4017.

[37]. L. Ben-Brahim, M. Benammar, A New PLL Method for Resolvers, in Proceedings of the International Power Electronics Conference (IPEC’10), Sapporo, Japan, 21-24 June 2010, pp. 299-305.

[38]. N. Al-Emadi, L. Ben-Brahim, M. Benammar, A New Tracking Technique for Mechanical Angle Measurement, Measurement, Vol. 54, 2014, pp. 58-64.

[39]. C. J. Paull, W. A. Evans, Waveform shaping techniques for the design of signal sources, Radio and Electronic Engineer, Vol. 44, No. 10, 1974, pp. 523-532.

[40]. V. Schiffer, W. A. Evans, Approximations in sinewave generation and synthesis, Radio and Electronic Engineer, Vol. 48, No. 3, 1978, pp. 113-121.

[41]. D. Kubanek, Diode piecewise-linear function approximation circuit with current input and output, in Proceedings of the 34th International Conference on Telecommunications and Signal Processing (TSP’11), 2011, pp. 284-287.

[42]. B. Gilbert, Circuits for the Precise Synthesis of the Sine Function, Electronics Letters, Vol. 13, August 1977, pp. 506-508.

[43]. O. Ishizuka, Z. Tang, H. Matsumoto, MOS Sine Function Generator Using Exponential-Law Technique, Electronics Letters, Vol. 27, Issue 21, October 1991, pp. 1937-1939.

[44]. W. A. Evans, V. Schiffer, A low distortion tri-wave to sine converter, Radio and Electronic Engineer, Vol. 47, No. 5, 1977, pp. 217-224.


122

[45]. C. A. dos Reis Filho, M. P. Pessatti, J. P. C. Cajueiro, Analog triangular-to-sine converter using lateral-pnp transistors in analogue pCMOS process, in Proceedings of the 9th International Conference on Electronics, Circuits and Systems, 2002, Vol. 1, 2002, pp. 253-256.

[46]. H. Hassan, FET differential amplifier as a tri-wave to sine converter, in Proceedings of the 36th Southeastern Symposium on System Theory, 2004, pp. 427-430.

[47]. S. Galecki, Multiphase sine-shaper circuit, IEEE Journal of Solid-State Circuits, Vol. 32, No. 1, 1997, pp. 126–129.

[48]. R. G. Meyer, W. M. C. Sansen, S. Lui, S. Peeters, The differential pair as a triangle-sine converter, IEEE Journal of Solid-state Circuits, Vol. SC-11, No. 3, 1976, pp. 418-420.

___________________



http://www.sensorsportal.com/HTML/E-SHOP/PRODUCTS_4/USTI.htm






123



New Design-methodology of High-performance TDC on a Low Cost FPGA Targets

Foudil Dadouche, Timothé Turko, Wilfried Uhring, Imane Malass,

Norbert Dumas, Jean-Pierre Le Normand ICube, UMR 7357, University of Strasbourg and CNRS

23, rue du Loess BP 20, F-67037 Strasbourg Cedex 2, France Tel.: +33 (0)3 88 10 68 27, fax: +33 (0)3 88 10 65 48


Received: 31 August 2015 /Accepted: 5 October 2015 /Published: 30 October 2015 Abstract: This work aims to introduce a design methodology of Time-to-Digital Converters (TDCs) on low cost Field-Programmable Gate Array (FPGA) targets. First, the paper illustrates how to take advantage of the presence of carry chains in elementary logic elements of the FPGA in order to enhance the TDC resolution. Then, it describes how to use the Chip Planner tool to place the partitions composing the system in user specified physical regions. This allows the placement of TDC partitions so that the routing paths are constrained. As a result, the user controls the propagation delay effectively through the connection network. The paper ends by applying the presented methodology to a case study showing the design and implementation of high resolution TDC dedicated to time correlated single photon counting system. The resolution of 42 ps as well as the INL, DNL and mean Jitter values (22 ps rms, 13 ps rms and 26 ps rms, respectively) obtained using a low cost FPGA target Cyclone family are very promising and suitable for a large amount of fast applications. Copyright © 2015 IFSA Publishing, S. L. Keywords: Time-to-digital converter, FPGA, Chip planner, Carry chain logic, Time correlated single photon counting. 1. Introduction

Nowadays, numerous applications require a

precise measurement of time duration separating two or several physical events. 3D scanners or 3D console games represent typical application requiring precise time quantification of the interval time to reconstitute a three-dimensional scene. Such systems are generally based on Time of Flight (TOF) measurement of the light emitted by a laser diode or a light-emitting diode (LED) and detected by a suitable light sensors after reflection by an object. The TOF of the light is proportional to the distance traveled by

the latter. The measurement is made independently by several pixels allowing the reconstitution of the 3D scene [1-2].

To measure this duration, we use devices capable of converting extremely low time durations (some tens of picoseconds) into digital values understandable for downstream processing and conditioning chain. These devices are commonly known as Time-to-Digital Converters (TDCs) [3]. The latter is largely used for several years in numerous smart sensor systems, particle and high-energy physics applications as well as measurement and instrumentation applications such as digital scopes and logic analyzers [3-4].




124

In order to design such systems there are several techniques which are proposed in literature. Some of the techniques that can be readily identified are [3, 5-6]: Tapped Delay Lines (TDL), Delay Locked Loop (DLL), Vernier Delay Line (VDL), Multilevel TDC, etc. All of these Time-to-Digital conversion techniques are usually designed as Application-Specific Integrated Circuits (ASICs) [7]. The latter have the advantage to have high performances but suffer from a higher cost, slow time to market and limited reconfiguration possibilities. It is also worth noting that the ASIC solutions are not suitable for integration into reconfigurable digital designs mostly described in Hardware Description Languages (HDL). As a result, numerous solutions for implementing TDCs on FPGA circuits have emerged [8-13]. However, the most significant limitation of these architectures is the difficulty to predict the placement and routing delays as well as the time delay of the logic gates itself. The consequence of this inevitable hardware restriction is a non-stable resolution of the designed TDC [10].

In this work, we aim to extend our contribution presented at SENSORCOMM 2015, consisting of introducing a design methodology for high resolution TDC on low cost FPGA targets [14] by including some improvements leading to new interesting results. This methodology enables the mastering of the network routing delays as well as the delays of the gates themselves. Therefore, it leads to an optimized TDC design with stable and accurate resolutions.

In order to give some background, the functional principle as well as the structure of the studied TDC in given in the second section. We also point out some associated difficulties encountered while using classical inverters as delay cells. The third section is dedicated to present our approach of implementing a TDC on an FPGA. Firstly, we show how to take advantage of the Carry Chain Logic to enhance and optimize TDC resolution. Secondly, we illustrate how to use the Chip Planner to define the exact physical layout location in the Chip. Therefore, we point out the importance of this operation. Section 4 provides a detailed case study consisting of implementing a 42 ps resolution TDC on Altera Cyclone IV low cost FPGA. The implemented TDC is associated to an FTDI (Future Technology Devices International) USB interface circuit operating in parallel mode with transfer rates reaching up to 40 Mbytes per second. Finally, we end our work by providing some final observations.

2. Setting in the Context

2.1. Functional Principle of the Studied TDC A TDC is an electronic system that measures the

interval time between two occurring events of a given signal. Its main purpose is to convert temporal

information to binary sequence understandable for a downstream processing chain.

For an accurate time duration measurement, generally, a TDC is composed of three blocks: two fine measurement blocks and a coarse one. The coarse one counts the number (N) of clock periods between enabling to disabling the measured interval, and the fine blocks evaluate the uncertainties in both sides that cannot be counted since their duration is shorter than the clock period.

To understand the role of each one of the three blocks, we illustrate by the timing diagram of Fig. 1 the functioning principle of a generic TDC.

Fig. 1. Functioning principle of a generic TDC.

As we can see from this timing diagram, the time interval to be measured (Tm) is a combination of three individual durations:

1) TCoarse, which represents the number of clock periods from enabling to disabling coarse measurement;

2) TFine1, representing the time between the measured signal active edge and the first following rising clock edge;

3) TFine2 which is the time between the falling edge of the measured interval and the following rising clock edge. Accordingly to this timing diagram, the measured time will be expressed as follows:

21 FineCoarseFinem TTTT −+= (1)

However, given the fact that the:

clkCoarse TNT .= (2)

we obtain the following expression:

1 2.m Fine clk FineT T N T T= + − (3)

In practice, TDCs are mostly used in fast imaging

systems needing to know the delay separating a photon emission by a laser diode and the detection of that photon by a Single-Photon Avalanche Diode (SPAD). Therefore, the events can be represented by


125

two signals: (1) a START signal, which can be synchronized with the coarse counter clock, and (2) a STOP signal that means that the SPAD has detected a photon. In that case, the whole TDC can be reduced to a coarse counter associated to a fine TDC measuring TFine2. Consequently, the measured interval time will be given by the expression hereafter:

2.m clk FineT N T T= − (4)

Since the coarse block is a simple counter

incremented by the system clock, we will focus in the following section on the implementation of the fine TDC.

2.2. Structure of the Studied TDC As mentioned previously, there are different

techniques of designing TDCs. In this work, we focus on the commonly used Tapped Delay Lines (TDL) architecture depicted in Fig. 2.

D Q

Clk

D Q

Clk

D Q

Clk

Q0 Q1 QN

HIT

Starttd td td

Fig. 2. Tapped delay line TDC.

A TDL TDC consists of N cascaded delay elements whose inputs are stored in D Flip Flops (DFFs). We would then have as many DFFs as there are delay elements. Therefore, each delay element can be regrouped with its associated DFF to form an elementary cell of the TDC.

The number (N) of these elementary cells depends of the common DFF clock frequency, as well as the propagation time of the delay element (td). This is given by the ratio of clock period to propagation time td. Since the value of the previous parameter is not provided, it is determined experimentally.

2.3. Design and Validation of the Elementary Cell

In order to design the elementary cell of the

studied TDC, we first used a simple invertor as a delay element associated with a DFF as illustrated by Fig. 3.

However, implementing a TDC chain on an FPGA by duplication of this cell leads to a simplified

circuit entirely different from the desired function. Indeed, if the input signal Start_in and the output signal Start_out have the same logic equation, the used HDL software (Quartus II) will simplify the logical equation giving the output versus the input so that it saves place and time.

D Q

Clk

Q

HIT

Start_in Start_out

Fig. 3. Simple TDC elementary cell.

To illustrate this phenomenon, we represent in Fig. 4 the RTL view resulting from the implementation of a simple TDC chain composed of four elementary cells.

Fig. 4. RTL view of the implementation of a simple 4-cells TDC.

It can readily be seen that, in spite of the presence of inverters, the software has simplified the logical equations. Consequently, all the inverted signals are grouped independently of the non-inverted ones. It is thus evident that this method is not suitable for designing a TDC. Nevertheless, it is worth to notice


126

that, if it is not possible to prevent Quartus II software to optimize data path, it is quite possible to create this path manually by operating directly on the logical resources of the FPGA. Indeed, the Quartus II Chip Planner tool allows physical access to logical resources available on the chip.

Using this tool, we can perform a customized configuration of the logic elements and impose the data path. However, the manual configuration of logic elements is tedious and time consuming in particular for systems with a certain complexity such as TDCs.

Even if we can use this technique to implement a TDC on an FPGA, given the large number of logic elements to be configured individually, it is still difficult to set up. Moreover, the TDC chain size can vary from an application to another; it will be therefore preferable to automate the configuration so that the solution will be generic and adaptable. Hence, we propose an appropriate design methodology in the following section.

3. Design Methodology

In order to provide solutions to the above raised issues, in this section we suggest an alternative

approach that can carry out a TDC structure fulfilling the following needs:

• Avoid the software data path simplification; • Increase TDC resolution by reducing the

propagation time through delay elements; • Automate the elementary cells set-up process

to optimize the design time and make possible the development of generic and adaptable structures;

• Use a low cost FPGA target to implement the TDC.

This method is focused on two main areas: • Using adders as delay elements and utilization

of the Carry Chain Logic of the FPGA; • Using the Chip Planner tool.

3.1. Using Adders and Carry Chain Logic The implementation of digital circuits on FPGA

targets depends on the architecture of the logical resources of the target. In this work, we are aiming to use a low cost FPGA from Altera Cyclone family. The selected target is the Cyclone IV (EP4CE55F23C8) based on the logic element shown by Fig. 5 [15].

Fig. 5. Cyclone IV logic element structure.

The Cyclone IV logical element, provides a dedicated path for fast carry propagation. The role of this carry chain is to use specific fast paths for carry propagation instead of general-purpose routing network. By doing so, it makes it possible to drastically optimize the propagation time. This is ideal for the enhancement of the TDC resolution. Moreover, it allows harmonizing the delays of the TDC elementary cells.

The problem is that customized handling of carry chains is reserved to high performance FPGAs as such as the Stratix family from Altera whose cost is outstandingly high. However, it is possible to configure the Quartus II synthesis tool to optimize speed. In this case, the synthesis tool uses the carry chain logic automatically when synthetizing an HDL model involving adders.


127

It is therefore possible to use the carry chain logic to minimize and harmonize propagation delays for components involving adders. It is precisely the idea that is exploited here to design TDC elementary cells based on simple adders. This was done by developing a simple behavioral VHDL model for an adder with a customizable number of elementary cells. The number of the cells depends on the data width modeled by a generic parameter called DATA_WIDTH. The whole model is given by Fig. 6.

Fig. 6. Adder VHDL model.

The fine TDC using adders can be performed by: 1) Applying the TDC input signal STOP to the

carry input signal (cin) of the adder; 2) Choosing values for adder operand inputs

(a and b) so that an output carry is generated (cout=‘1’) if input carry is equal to ‘1’.

The output carry is then an exact replication of the input carry delayed by a transmission time through the cell. To do so, all it takes is to set all the bits of the first operand to ‘1’ and the bits of the second operand to ‘0’. For each bit (i) the arithmetic sum a(i)+b(i) gives ‘1’. When the input carry is activated (cin=‘1’) by the TDC input signal (STOP), the arithmetic sum a+b+cin gives ‘0’ and the carry output moves to ‘1’.

The Fig. 7 illustrates the implementation of one elementary cell of a TDC by a logic element of the Cyclone IV target. The adder cell is obtained by the look up table (LUT) and the DFF by the sequential configurable output register.

Theoretically, to obtain a TDC chain similar to the TDL structure shown by Fig. 2, it is sufficient to duplicate the structure of Fig. 7 as often as necessary to reach the number of desired cells. However, when implementing such a chain on the FPGA, some DFFs of the TDC elementary cells are dissociated of their corresponding 1-bit adder cells even if the data path is perfectly respected.

This phenomenon occurs randomly and leads to the placing of the DFF and the delay element of the

same TDC’s elementary cell in different logic elements, as shown in Fig. 8. The direct consequence of component misplacing is that the delay time is no longer identical for all cells. This inevitably generates unpredictable artifacts. To ensure a reliable operation, it is necessary to overcome this problem by constraining the placement tool to bring together the components of the same cell in the same logic element. This is the purpose of the next section.

Fig. 7. Implementation of a TDC elementary cell by a logic element.

Fig. 8. Random placing of DFFs on the chip.

3.2. Using Chip Planner Using a TDC in fast imaging systems requires the

measurement of very short time durations. It is therefore necessary to master all of the signal propagation delays through the cells as well as the routing network.

As we have seen in the previous section, unconstrained automatic implementation of a TDC on an FPGA usually leads to an inhomogeneous and irreproducible structure. Consequently, the measurement results are tainted by these uncertainties. Therefore, it is necessary to control the exact physical location of TDC cells on the chip.

This could be achieved by using the Chip Planner tool provided by Altera. The latter, according to the user's needs, allows the defining of specified implementation regions on the chip for blocks constituting the whole system. In addition, it supports incremental compilation to preserve the well-implemented parts and reduce the compilation time. This operation takes place in three distinct steps:

• Creating Design Partitions: the first step consists of dividing the design in individual partitions according to system complexity as well as user needs.


128

• Defining logic regions: after partitioning the design, it is necessary to define logical zones that will be associated to the partitions. This allows individual compiling and optimizing of each region. The tool used to perform this operation is LogicLock Region (LLR) within Chip Planner.

• Physical assignment of logic regions: in order to physically preserve the logic regions defined in the previous step, by means of the LLR tool, physical regions of the chip are assigned to implemented partitions.

The physical delimitation of regions permits to constrain the placing and root tool to put partitions in their specified regions defined by the user. Doing so, it allows not only avoiding the random placement of certain DFFs away from their associated delay elements, but also implementing the concerned partitions as close as possible to input signal pins (HIT and STOP). The purpose of the latter operation is to reduce the propagation delays of input signal before reaching the blocks to which they are intended to be applied. For illustrative purposes, we represent on Fig. 9 the assignments of physical allocations of the partitions defined using LLR and a close-up view of the layout of a 16-cells fine TDC implemented using the method presented above on Fig. 10.

Fig. 9. Layout of implemented partitions of a TDC.

Fig. 10. Physical implementation of 16-cells TDC.

The TDC fits perfectly within the reserved region that would be assigned to it. Consequently, the DFF and the delay element of each TDC elementary cell are now implemented by the same logic element. The transmission delays are then identical for all cells.

4. Case Study: Implementation of A 42 ps TDC on A cyclone IV FPGA In this section we aim to apply the presented

methodology to a realistic case: 1) We begin by introducing the experimental

measurement conditions; 2) Then we discuss the effect of constraining cell

placement by using the Chip Plannar tool; 3) And finally we present improved results

obtained by overcoming a hardware limitation of the Dallas Logic FPGA design kit.

4.1. Experimental Measurement Conditions The proposed TDC design has been implemented

within the Cyclone IV (EP4CE55F23C8) FPGA target. The coarse counter clock is 200 MHz, i.e., the clock period is 5 ns. The delay line for the fine TDC, based on carry chain adder architecture, comprises 128 cells in order to cover a dynamic of more than 5 ns. The signal that needs to be measured propagates through the delay chain, until the FPGA clock disables the DFFs to block their outputs and then memorizes their states. The value of these DFFs describes the time spent between the signal STOP and Clock.

The data is then transmitted to a USB port via an FTDI FT232H operating in parallel mode with transfer rates reaching up to 40 Mbyte per second. To acquire data measurements, we developed a specific application using LabVIEW software.

In order to reduce the size of data transmitted to the USB port, we developed a VHDL model of a specific encoder converting the 128 bits to a one byte data. Moreover it filters potential errors. The functioning of the latter is described in the section 4.1.1 here after.

Finally, to avoid some failure and misleading measurements, it was necessary to complete the principal test bench described below by synchronizing the fine and coarse counters as introduced in section 4.1.2.

All this blocks are summarized in Fig. 11 showing the synoptic view of the whole system.

The TDC has been characterized on its whole dynamic, i.e., from 0 to 640 ns with a step of 5 ps. A Stanford research DG 645 digital delay generator has been used to generate the START and STOP signals. At this range of delay, the jitter of the delay generator is lower than 25 ps rms.


129

8

8

8128

EnableFine

Counter

Fine

Counter

CoarseCounter

FIFO

Encoder

FTDI

EnableCoarse

Counter 0

1

WriteRequest

ReadRequest

CLK 200 Mhz

STARTSTOP

USBPort

CL

K 60 M

hz

TX

E

Fifo Empty

Fifo Full

Fig. 11. Synoptic view of the implemented TDC system.

4.1.1. Encoding Fine Counter Output Binary

Stream As described previously the output binary data

stream of the fine counter, representing the measured time, is applied to an encoder. The latter prepares the data before saving it into a FIFO (First In First Out) memory. Indeed, at the output of the delay chain, data are presented as a string made of zeros (‘0’) on the left and ones (‘1’) on the right. The encoder’s role is to count the number of consecutives bits switched to ‘1’ and generate the corresponding 8-bits binary code. So the first idea is to use a simple priority encoder to detect the position of the most significant bit moved to ‘1’ and gives the corresponding binary code. To illustrate this we show on Fig. 12 a 14-cells TDC output when a half of the cells were crossed by the measured signal. As we can see the input of the encoder is set to “00000001111111” and its associated output is set to “111”. However, the problem with such an approach is that, because of manufacturing variations, such as Setup time mismatch, flip-flops located further in the delay chain can sometimes react before other flip-flops or vice versa. Consequently, the TDC’s output can be erroneous as illustrated by Fig. 13. Indeed, the input of the encoder can be set to “00000101111111” for the same delay of the previous case because a flip flop present a shorter setup time and thus detect the data prematurely. In this case, the output of the encoder will be the wrong code “1001”, instead of the good code “0111”.

To overcome similar situations it is necessary to build a robust encoder. The latter, in addition to encoding the 128 bits in one byte data, is designed such that it detects failed measures due to the flip-flops Setup and Hold Times.

TDC Fine

CounterEncoder14-bits

TDC output: 00000001111111

1118-bits

‘1’ to ‘0’ transition

Encoder output:

Fig. 12. The theoretical string FOR 14-cells TDC.

TDC Fine


TDC output: 00000101111111

10018-bits

Expected ‘1’ to ‘0’ transition

Encoder output:

Wrong ‘1’ to ‘0’ transition

Fig. 13. Illustration of potential errors in the data string.

The method adopted here is to add a supplementary detection condition. The last one consists of detecting a sequence of ‘011’ instead of ‘01’. In this way all the sequences including ‘1’ between two or more zeros (‘0’) are identified as wrong behavior of corresponding DFF. Such events are ignored and only the events including at least two consecutive following ones (‘1’) are considered. Thanks to this method, we can guarantee the data coherence, even if a false code appears as it shown on Fig. 14. For this encoder, a code “00000101111111”, will generate a good output equal to “0111”.


130

TDC Fine


TDC output: 00000101111111

1118-bits

Validated ‘11’ to ‘0’ transition

Encoder output:

Ignored ‘1’ to ‘0’ transition

Fig. 14. Correction of potential errors by the encoder.

4.1.2. Synchronization of Fine and Coarse Counters

The last step in the process is the synchronization

of the Fine and the Coarse counter. Without this step, measure noise could promptly be equal to the least significant bit of the coarse counter during its state changes.

To prevent this problem, a second clock is instantiated. Its operating frequency is the same as the main clock in the circuit, except its phase will be slightly shifted.

One of the two clock rate the coarse counter and the other one rate the fine counter. By changing the second clock phase, it is possible to perfectly synchronize the fine and coarse counter.

4.2. Effect of Constraining Cell Placement by the Chip Planer Tool

To show the effect of constraining cell placement

by the Chip Planer tool we report on Fig. 15 the detail of the unconstrained and constrained fine TDC measurements between two reference clock edges, i.e., on a range of 5 ns.

1500 2000 2500 3000 3500 4000 4500 5000 5500 6000

0

20

40

60

80

100

120

Time (ps)

Fin

e TD

C R

AW

data (lsb)

y = 0.0241*x - 34.1

Unconstrainted TDC raw data

Constrainted TDC raw data

linear fit

Fig. 15. Responses of Fine unconstrained and constrained TDC.

The unconstrained fine TDC response (blue) shows a large discrepancy of the LSB value indicating that some DFFs have been randomly placed. The resulting large steps make the unconstrained fine TDC unusable for sub nanosecond timing. Consequently, the use of the Chip Planner tool as described in section III is mandatory to obtain the behavior of the constrained fine TDC represented by the green curve. A linear fit is then used to assess the LSB value of the fine TDC which is given by the inverse of the linear fit slope, i.e., 41.5 ps in this study case.

4.3. Jitter, INL and DNL Evaluation

The noise visible on the fine TDC response, depicted on Fig. 15 bellow, is due to the jitter. The latter adds uncertainty on each measurement and it can be evaluated by computing the standard deviation

of a set of measurements at a given fixed delay between the START and STOP signals.

The jitter depicted in Fig. 16 has been characterized for different delays corresponding to a given signal propagation along the fine TDC line. As each fine TDC elementary cell adds its own jitter [16], the global jitter will then increase as a square root of the number N cells as given by the following expression:

2 2 Nα β+ ⋅ , (5)

where α is the initial jitter present at the input of the first cell and β the single cell jitter.

A curve, following this law is fitted on the jitter profile to underline the jitter’s variation relationship in the delay line. The extraction of this parameters leads to an initial jitter α of 62 ps rms and a single cell jitter β of 5.8 ps rms.


131

The accumulated jitter across the fine TDC delay line leads to a mean jitter of 90 ps rms. Thus, the line length has to be kept as low as possible in order to obtain the best accuracy. This can be done by using the fastest achievable frequency for the coarse counter.

The integral non linearity error (INL) and the differential non linearity (DNL) have been measured over the entire range of the TDC. For illustrative purposes, the results from a delay of 0 to 160 ns are represented by Fig. 17 hereafter. It can be seen that, the implemented system shows an INL of 132 ps rms and a DNL of 50 ps rms.

The measured Jitter is quite high compared to those extracted from literature (Table 1). Moreover the distribution of jitter does not show a Gaussian shape, indicating that this noise is probably correlated to a parasitic signal.

Different verification tests have been investigating on the effect of the FPGA input/output

buffer delays, the FPGA oscillator frequency and the USB communication interface in order to determine the origin of the discrepancy. Finally, most of the jitter arise from a noise present on the 1.2 core voltage FPGA power supply. Indeed, to generate this voltage the Dallas Logic FPGA design kit uses a switching DC/DC converter. The last one presents a periodic noise of about 10 millivolts at a frequency close to 42 kHz. To overcome this limitation the provided power supply has been unsoldered and replaced by an external linear regulated power supply generator. The new measurement of jitter, INL and DNL are reported on Fig. 18 and Fig. 19. It can readily be seen that the mean jitter is improved and reaches 26 ps rms, the INL and the DNL are reduced to 22 ps rms and 13 ps rms, respectively. The periodic behavior of the jitter is due to the transition between two stages of the tapped delay line. The spikes visible in the INL at 3100 and 8100 ns correspond to the toggle of a coarse counter bit.

Fig. 16. Jitter measurement according to the elementary level, the jitter increases as the signal propagates along the fine TDC cells.

0 2 4 6 8 10 12 14 16

x 10

4

-200

0

200

Time (ps)

IN

L (ps)

0 2 4 6 8 10 12 14 16

x 10

4

-200

0

200

Time (ps)

DN

L (ps)

Fig. 17. INL and DNL errors of the implemented TDC over a range of 160 ns.


132

Table 1. Comparison of recent implementations of TDC on different FPGA targets.

Reference Date Architecture FPGA Resolution Jitter DNL INL

[18] 2011 Multi Channel

Tapped Delay Line Xilinx

Virtex-6 10 ps [7.38;14.24] ps [-1;1.5] LSB [-2.25;1.61] LSB

[19] 2015 Dual Phase Tapped

Delay Line Xilinx

Virtex-6 10 ps 12.83 ps-rms [-1;1.8] LSB [-2.20;2.60] LSB

[20] 2010 Vernier Delay Line Xilinx

Virtex 4 321.5 ps / [-0.28;0.3] LSB [-0.3;0.65] LSB

[13] 2008 Ring Oscillator Altera

Stratix II 40 ps / <1 LSB <1 LSB

[21] 2006 Vernier Delay Line Altera

ACEX 1 K91.5 ps / [-0.416;0.783] LSB [-0.567;0.687] LSB

[21] 2006 Vernier Delay Line Xilinx

Virtex-II 68.5 ps / [-0.953;1.051] LSB [-2.003;1.855] LSB

[22] 2012 Tapped Delay Line Xilinx

Virtex-5 30 ps 56.5 ps rms [-1;3] LSB [-4;4] LSB

[23] 2010 Tapped Delay Line Virtex-II

Pro 17 ps 24 ps rms [-1;2] LSB [-1.5;3.5] LSB

[9] 2013 Vernier Delay Line

Manual Rooting Xilinx

Virtex-5 9 ps <1 LSB < 0.11 LSB /

[11] 2015 Hybrid Delay Line Xilinx

Spartan 3E30 ps ± 154 ps / /

[12] 2009 Tapped Delay Line Xilinx

Virtex-5 55 ps / [-2.5;1] LSB [-4.5;3] LSB

[8] 2010 Pulse Shrinking Xilinx

Spartan 3E42 ps 24 ps [-0.98;0.417] LSB [-4.21;3.36] LSB

This work 2015 Tapped Delay

Line Altera

Cyclone IV42 ps

0.62 LSB 26 ps rms

[-0.4; 0.5] LSB 22 ps rms

[-1;2] LSB 13 ps rms

0 20 40 60 80 100 120

0

10

20

30

40

50

Elementary TDC cell

Jitter (ps rm

s)

Fig. 18. New Jitter measurement after replacing the Dallas Logic power supply core by a DC regulated power supply.

0 1000 2000 3000 4000 5000 6000 7000 8000

-100

-50

0

50

100

Time (ps)

INL (ps)

0 1000 2000 3000 4000 5000 6000 7000 8000

-20

-10

0

10

20

Time (ps)

DNL (ps)

Fig. 19. New INL and DNL errors of the implemented TDC over a range of 160 ns after disabling the Dallas Logic supply core.


133

5. Conclusions

This paper has proposed a global methodology to design and implement Time-to-Digital Converters on low cost FPGA targets. It presents how to use different tools to enhance the TDC resolution by reducing propagation delays through the connection network as well as the logic gates themselves. First, the use of adders as delay elements, to benefit from a dedicated carry chain logic path, is presented. Then we detailed how to take advantage of the chip planner, to constrain the placing and root tool to put the partitions of the system in user specified physical regions. Doing so, it allowed the mastering of propagation delays and consequently improved the resolution and the stability of the TDC.

The work is ended by a case study that applied this methodology to design a TDC with a resolution of about 42 ps on a Cyclone IV FPGA. The implemented TDC presents a jitter of only 26 ps rms, and the DNL and the INL has been measured respectively to 22 and 13 ps rms. As we can see from the state of the art summarized in Table 1, comparable values of our results are obtained but by using high performances FPGA targets as such as Virtex 6 of Xilinx which are highly expensive.

The highlighted results in this paper are very promising, not only because they are suitable for domains requiring high performances, but also because they are achieved by using a low cost FPGA family which opens the door to a broader use in a great amount of fast application fields.

As a perspective in the near future, we plan to integrate the presented TDC in different applications such as image photon counting devices and microfluidic experimentations [17].

References [1]. L. Li, Time-of-flight camera – an introduction, Texas

Instruments, SLOA190B – Technical White Paper, January 2014, revised May 2014.

[2]. E. Charbon, M. Fishburn, R. Walker, R. K. Henderson, C. Niclass, SPAD-based sensors TOF Range-Imaging Cameras, F. Remondino and D. Stoppa (Eds.), Springer-Verlag, Berlin Heidelberg, 2013, pp. 11-38.

[3]. S. Henzler, Time-to-Digital Converters, Springer Science+Business Media B. V., 2010.

[4]. S. Y. Yurish, Smart Optoelectronic Sensors and Intelligent Sensor Systems, Sensors & Transducers, Vol. 14-1, Special Issue, March 2012, pp. 18-31.

[5]. J. Kalisz, Review of methods for time interval measurements with picoseconds resolution, Metrologia, Vol. 41, No. 1, 2004, pp. 17-32.

[6]. C. S. Hwang, P. Chen, H. W. Tsao, A high-precision Time-to-Digital converter using a two-level conversion scheme, IEEE Transactions on Nuclear Science, Vol. 51, No. 4, 2004, pp. 1349-1352.

[7]. I. Malasse, W. Uhring, J. Le Normand, N. Dumas, F. Dadouche, 10-ps Resolution Hybrid Time to Digital Converter in a 0.18 µm CMOS Technology,

in Proceedings of the IEEE 12th International New Circuits and Systems Conference (NEWCAS), June 2014, pp. 105-108.

[8]. J. Kalisz, R. Szplet, J. Pasierbinski, A. Poniecki, Field-programmable-gate-array-based time-to-digital converter with 200-ps resolution, IEEE Transactions on Instrumentation and Measurement, Vol. 46, Issue 1, 1997, pp. 51-55.

[9]. I. Vornicu, R. Carmona-Galán, Á. Rodríguez-Vázquez, Wide range 8-ps incremental resolution time interval generator based on FPGA technology, in Proceedings of the IEEE 21st International Conference on Electronics, Circuits and Systems (ICECS’14), 2014, pp. 395-398.

[10]. M. Lin, G. Tsai, C. Liu, S. Chu, FPGA-Based high area efficient Time-to-Digital IP design, in Proceedings of the IEEE Region 10 Conference (TENCON’06), 2006, pp. l-4.

[11]. R. Narasimman, A. Prabhakar, N. Chandrachoodan, Implementation of a 30 ps resolution Time-to-Digital Converter in FPGA, in Proceedings of the IEEE International Conference on EDCAV, 2015, pp. 12-17.

[12]. A. Aloisio, P. Branchini, R. Giordano, V. Izzo, S. Loffredo, High-precision Time-to-Digital converter in a FPGA device, in Proceedings of the IEEE Nuclear Science Symposium Conference Record (NSS/MIC), Vol. 13, 2009, pp. 290-294.

[13]. S. S. Junnarkar, P. O’Connor, R. Fontaine, FPGA based self calibrating 40 picosecond resolution, wide range Time-to-Digital converter, in Proceedings of the IEEE Nuclear Science Symposium Conference Record (NSS '08), 2008, pp. 3434-3439.

[14]. F. Dadouche, T. Turko, W. Uhring, I. Malass, J. Bartringer, J.-P. Le Normand, Design Methodology of TDC on Low Cost FPGA Targets, in Proceedings of the 9th International Conference on Sensor Technologies and Applications (SENSORCOMM’ 2015), Venice, Italy, 23-27 August 2015, pp. 29-34.

[15]. Cyclone IV Device Handbook, Vol. 1, Chapter 2, Altera Corporation, April 2014.

[16]. M. Zlatanski, W. Uhring, J.-P. Le Normand, D. Mathiot, A Fully characterizable asynchronous multiphase delay generator, IEEE Transactions on Nuclear Science, Vol. 58, No. 2, 2011, pp. 418-425.

[17]. J. Léonard, N. Dumas, J. Causse, S. Maillot, N. Giannakopoulou, S. Barre, W. Uhring, High-throughput time-correlated single photon counting, Lab on a Chip - Royal Society of Chemistry, Vol. 14, No. 22, November 2014, pp. 4338-4343.

[18]. H. Menninga, C. Favi, M. W. Fishburn, E. Charbon, A multi-channel, 10ps resolution, FPGA-based TDC with 300MS/s throughput for open-source PET applications, in Proceedings of the IEEE Nuclear Science Symposium and Medical Imaging Conference (NSS/MIC), 23-29 Oct. 2011, pp. 1515-1522.

[19]. J. Y. Won, S. I. Kwon, H. S. Yoon, G. B. Ko, J.-W. Son, J. S. Lee, Dual-Phase Tapped-Delay-Line Time-to-Digital Converter with On-the-Fly Calibration Implemented in 40 nm FPGA, IEEE Transactions on Biomedical Circuits and Systems, Vol. PP, No. 99, p. 1.

[20]. L. Arpin, M. Bergeron, M-A.Tetrault, R. Lecomte, R. Fontaine, A Sub-Nanosecond Time Interval Detection System Using FPGA Embedded I/O Resources, IEEE Transactions on Nuclear Science, Vol. 57, No. 2, April 2010, pp. 519-524.

[21]. J. Song, Q. An, S. Liu, A high-resolution time-to-digital converter implemented in field-


134

programmable-gate-arrays, IEEE Transactions on Nuclear Science, Vol. 53, No. 1, February 2006, pp. 236-241.

[22]. L. Zhao, X. Hu, S. Liu, J. Wang, Q. An, A 16-channel 15 ps TDC implemented in a 65 nm FPGA, in Proceedings of the 18th IEEE-NPSS Real Time Conference (RT), 9-15 June 2012, pp. 1-5.

[23]. M. Daigneault, J.-P. David, A novel 10 ps resolution TDC architecture implemented in a 130 nm process FPGA, in Proceedings of the 8th IEEE International NEWCAS Conference (NEWCAS), 20-23 June 2010, pp. 281-284.

___________________


http://www.sensorsportal.com/HTML/BOOKSTORE/Digital_Sensors.htm






135



Experiences in Automation and Control in Engineering Education with Real-world Based Educational Kits

1 Filomena SOARES, 1 Celina Pinto LEÃO, 2 José MACHADO

and 1, 3 Vítor CARVALHO 1 Algoritmi R&D-University of Minho, Campus of Azurém, 4800-058 Guimarães, Portugal 2 MEtRICs R&D-University of Minho, Campus of Azurém, 4800-058 Guimarães, Portugal

3 EST-IPCA, Campus of IPCA, 4750-810 Barcelos, Portugal 1 Tel.: +351253510180, fax: +351253510189


Received: 31 August 2015 /Accepted: 5 October 2015 /Published: 30 October 2015 Abstract: The well-known paradigm learning by doing is particularly important in engineering courses. Still, in some situations, there is a lack of real-world didactic workbenches due to the absence of financial support, human resources or maintenances restrictions. The authors of this paper have been overcome this difficulty by designing and implementing virtual and remote laboratories in Process Monitoring, Control and Automation teaching applied to Mechanical, Electronics and Biomedical Engineering. The goal of this paper is to present the work developed regarding the real-world workbenches to be used in automation and control practical classes as an integrated virtual and remote laboratory. Some important points include the modelling and control of Discrete Event Systems, Continuous Systems and Real-Time Systems as well as Industrial Control Networks. The physical parts were developed and connected, in a closed-loop configuration, with the respective controllers. The developed kits and systems were geared towards the engineering students’ needs. This integrated approach is very useful for providing students with a global set of skills in this domain. Quantitative and qualitative studies are continuously applied not only for obtaining students feedback but also to gather information to devise strategies for future virtual and remote laboratory applications development suitable for the target public. The positive results achieved so far are very encouraging attesting its efficiency not only in terms of students’ learning but also as a first contact to face real-world problems. The less positive identified point is concerned with technical aspects. Copyright © 2015 IFSA Publishing, S. L. Keywords: Automation, Control design, Control engineering, Control equipment, Engineering education. 1. Introduction

The design of new teaching/learning

methodologies implies the definition of tools and environments that promote students and teachers’ engagement in the knowledge acquisition process. These tools will serve as a complement to the traditional face-to-face lectures.

In engineering courses the laboratory work is very important since it is when students have the opportunity to apply the theoretical concepts learned earlier. So, to install and maintain several working positions in the laboratory is sometimes a challenge: the equipment is expensive, laboratorial space and qualified personnel are not enough. And this problem is replicated at each school.




136

The virtual and remote (V&R) laboratories may overcome this limitation [1-7]. The students may access the V&R laboratory at any time and place testing the real-world case-study through a graphical animation or a remote webcam.

Nevertheless, a good understanding of the pedagogical aspects is crucial for an efficient remote labs use [8]. Remote laboratories can also be used as a link between the University and the Industry, by providing remote monitoring for some industrial applications [9].

With this in mind and regarding Automation and Control education, a multidisciplinary group of teachers with different backgrounds, from the University of Minho (UM) and from the Polytechnic Institute of Cávado and Ave (IPCA), at North of Portugal, have been centering their attention in designing and developing workbenches in order to enable a practical learning environment. The goal was to provide to the engineering students different pedagogical tools/examples in order to have real-world interactive situations, for instance virtual processes and interactive animations, as well as local and remote experiences.

Considering the main domains, concerning Control and Automation, considered at UM and IPCA - namely: Discrete Event Systems modeling and control, Continuous Systems modeling and control, Real-Time Systems modeling and control and control industrial networks - several workbenches have been developed. For teaching these domains several software tools and controllers can be used. It was, in this spirit, that the workbenches were developed. Also, the physical part of the workbenches were developed and connected, in a closed-loop configuration, with the respective controllers [1].

Some (five) of the developed solutions include: the velocity control of a Direct Current (DC) motor (modeled as a Continuous System, controlled by a microcontroller in which was implemented a discrete Proportional, Integral, Derivative (PID) algorithm); two examples of systems controlled by Programmable Logic Controller (PLC): an automation workbench for testing and simulating PLC programs (modeled as a Discrete Event System), and the implementation of a small intelligent house (modeled as a Real-Time System); and, finally, two workbenches controlled by Industrial Computers modeled, both, as Real-Time Systems: one, of them - the level control of a two-tank system - using LabVIEW, and the other one - the acquisition of physiological signals – using, first, LabVIEW and, finally, C# and Java. This paper presents the developed solutions as an integrated V&R laboratory. Among others, this approach allows students to support and consolidate the traditional classes’ model allowing an efficient learning methodology in the specific areas of Control, Automation, Domotics, and Biomedical Data Acquisition fully integrated in one laboratory [1].

In this way, in order to describe the work developed and exchanged experience on best practice in the use of V&R laboratory for educational purpose over several years this paper is organized as follows: Section 2 provides a brief review of Virtual and Remote Labs, Section 3 presents the five developed real-world didactic kits used in the work. On Section 4 the workbench developed to the test and to validate the PLC programs is described. Section 5 depicts the students’ qualitative and quantitative feedback obtained during several experiences and finally, Section 6 enunciates the final remarks.

2. Virtual and Remote Labs

In the past years there has been a special concern from the scientific community regarding the teaching and learning process. And here, virtual labs gained particular important tools as they allow for a virtual perception of the physical phenomenon and reducing the material and equipment costs and maintenance.

In [10] Brinson presents a review synthesizing post-2005 empirical studies that focus on directly comparing learning outcome achievement using traditional lab (TL) and non-traditional lab (NTL, virtual and remote). The study comprised the scientific works (1291 articles) in Elsevier/ScienceDirect, EBSCO Suite (Academic Search Premier, Applied Science and Technology Source, Education Research, Science Reference Center), JSTOR, EdITLib, and ERIC databases. Findings suggest that most studies reviewed (n ¼ 50, 89 %) demonstrate student learning outcome achievement is equal or higher in NTL versus TL across all learning outcome categories (knowledge and understanding, inquiry skills, practical skills, perception, analytical skills, and social and scientific communication).

Flowers [11] presents the students’ perceptions of biology virtual laboratories. The sample size consisted of 19 undergraduate non-science majors (n=19) enrolled in an introductory biology course at a postsecondary 4-year university. In order to allow the comparison the students participated in a standard face-to-face laboratories followed by participation in similar technology-based laboratories. Research findings indicate that students prefer to participate in virtual labs compared to face-to-face labs. Data also indicated that students perceived higher learning gains as a result of participating in virtual labs compared to traditional hands-on labs.

Santos-Martin, et al. [12] present the experience of a problem-based learning approach on teaching wind energy conversion systems for electricity generation at an Electrical and Electronic Master's degree level - problem of finding the response of a wind turbine to a grid fault. Groups of three students worked on a cooperative project for 15 weeks, supervised by the teacher. A virtual wind turbine simulator and a real wind turbine setup were available to students. The results show that the


137

students valued the learning experience, considering the simulator and the experimental set-up important tools.

Goodwin, et al. [13] propose the use of emulation-based virtual laboratories in control engineering education allowing students (who are geographically dispersed) to acknowledge the reality of industrial control design. The laboratories are cost-effective solution to respond to students’ preferences and learning needs.

There are available in the Internet several virtual laboratories where the User has access to a vast number of didactic experiments in different thematic areas. We present some examples among a vast list a virtual labs: Virtuallaboratory (http://virtuallaboratory.colorado.edu/) helps students to recognize, confront, correct, and expand their understanding of scientific ideas and techniques in chemistry and biology; Biomodel (http://biomodel.uah.es/en/lab/inicio.htm) has available several virtual labs as: Cybertory a molecular biology lab, the simulation of cellulose acetate electrophoresis of proteins, and the simulation of column chromatography of proteins; Random (http://www.math.uah.edu/stat/) is a website devoted to probability, mathematical statistics, and stochastic processes, and is intended for teachers and students of these subjects; BioInteractive (http://www.hhmi.org/biointeractive/about-biointeractive) has multimedia resources, including apps, animations, videos, interactives, and virtual labs in Biology, in particular to identify bacteria based on its DNA, construct a transgenic fruit fly, and measure traits in fossil fish to record evolutionary change; LabSim (http://dei-s1.dei.uminho.pt/labsim/SimLab/Home.html) is a virtual laboratory in control and numerical methods, written in Portuguese; NeCTAR Virtual Laboratories (https://www.nectar.org.au/virtual-laboratories-1) are web environments that draw together research data, models, analysis tools and workflows to support collaborative research in geophysics, climate, marine among other areas.

A step forward in the development of educational tools was to enable students to remotely visualize the real experiments, having the perception of the physical experiment. The equipment was placed in a laboratory and shared with different institutions all over the world - Remote Laboratories.

The government of India as explored this resource developing a platform (http://vlab.co.in/index.php) which aggregates a several virtual laboratories in various disciplines of Sciences and Engineering.

Eguíluz [14] explores the use of virtual reality in the area of remote and virtual laboratories, enunciating the main challenges and opportunities in the teaching of electronic and electrical engineering. The project “Circuit Warz” was presented.

Nedev [15] shows the development of a computer-based system for data acquisition and LabVIEW virtual instruments for humidity and temperature as Internet-based virtual laboratories.

This setup enables to undertake remotely real experiments allowing virtual instruments in training of engineering students where it is of utmost importance to take practice hours.

Santana, et al. [16] describes the experiences using remote laboratories for education and research in the area of Control Engineering. The aim of this work is to apply remote laboratories in research applications to share equipment with researchers. The implementation of remote experimentation to control a 3-DOF robot by the Distance Laboratory System (SLD) is presented.

3. Real-World Didactic Kits

The real-world engineering problems are from automation and control areas. The goal was to provide students with practical real-world examples previously learned at class. These examples were used in engineering courses of different areas Mechanical, Electronics and Biomedical, at two institutions, Minho University (Braga and Guimarães, Portugal) and IPCA (Barcelos, Portugal) with 3rd and 4th year’s students. In the next subsections there are presented five developed kits: the velocity control of a DC motor, controlled by a Microcontroller; two examples of systems controlled by PLC: an automation workbench for testing and simulating PLC programs and the implementation of a small intelligent house; and, finally, two workbenches controlled by Industrial Computers: the level control of a two-tank system and the acquisition and treatment of physiological signals. The booking of students to access the developed kits is done by a queue defined by a set of conditions for proper priority (i.e., role of the participant – administrator or general user, user with/without previous scheduled access, etc).

3.1. Velocity Control of a DC Motor

A remote controlled DC motor velocity was developed for undergraduate control studies [17], allowing discrete PID algorithm testing and simulation [18]. A DC Motor “Maxon RE36” is inserted with an encoder “HEDS-5540 A11” for velocity measurement. The control methodology was implemented in a microcontroller. Four different digital versions of the PID algorithm are available. The microcontroller PIC16F876 from Microchip is used to directly control the motor; the control board is linked to the local personal computer by a serial RS-232 communication link for monitoring.

The user interface was developed in a LabVIEW (Fig. 1). Twelve motor velocity values can be selected (from 7 m/min to 335 m/min). The user must select the PID algorithm to be tested as well as the proportional, integral and derivative controller parameters.


138

Fig. 1. User interface.

3.2. Automation Workbench

The main focus of the workbench is to allow faster testing and validation of PLC programs, modeled as discrete event systems. The global goal of the developed system is to simulate and test the control programs in order to be sure that, when tested on the real system, those programs will not damage the physical part of the platform. It must be highlighted that this platform can be accessed from outside by different users – using Internet access - from different origins.

Thus, it was decided to use a PLC as the controller device and another PLC that will interact with the first one, using a developed network for that, as the plant simulator. For this purpose, the PLC that will run the program that simulates the plant behavior generates all the input signals for the PLC that runs the program of the controller.

The best configuration to meet the needs of the proposed workbench is presented in Fig. 2. OMRON equipment was used due to its availability in the laboratory. Both the controller and the plant models were implemented in the controllers OMRON CP1L M30 with CX-Programmer. The physical connection between components (the convertor MDW-45 and Personal Computer, PC and the converter and the PLC) was done using a serial port RS-422; in the connection PC-Converter an adaptor USB to serial port RS-232 was included.

The illustration of the scheme for connection between the real controller (PLC_1) and the simulator of the plant (PLC_2) is presented in Fig. 3. In Fig. 3 the configuration of the represented lines means:

• Black line: 24 V Conductor for PLC’s; • Orange line: 230 V Conductor for power

supply and MDW-45; • Green line: connection between the PLC and

buttons on the bench; • Red line: connection between the PLC

Controller inputs and PLC Plant outputs; • Blue line: connection between the PLC

Controller outputs and PLC Plant inputs; • Red arrow: RS-422 connection between the

MDW-45, PLC Controller and PLC Plant.

The connection between workbench and PC is made by RS-232. The RS-232 will be connected to MDW-45; and with RS-422 from MDW-45 to both PLCs. There is a second type of connection but this is only to see one single PLC. This is made by a USB 2.0 A-to-B cable. Both PLC can use this connection.

Fig. 2. Hardware in the Loop Workbench circuit.

Fig. 3. The connection scheme of all components from the workbench.

The control software used in this prototype is from OMRON. There were used the CX-One software and CX-Supervisor. CX-One allows using CX-Programmer to edit and simulate programs and CX-Designer that allows creation of simulation scenarios (drawings, schemes, among others) and using these scenarios associated to the program developed in CX-Programmer the user can understand the simulation of the system. Together with these tools, CX-Supervisor is devoted to the design and operation of PC visualization and machine control. It is not only simple to use, for small supervisory and control tasks, but it also offers a wealth of power for the design of the most sophisticated applications. CX-Supervisor boasts powerful functions for a wide range of PC based HMI (Human-Machine Interface) requirements. Simple applications can be created rapidly with the aid of a large number of predefined functions and libraries.


139

CX-Supervisor allows interaction between the user and the PLC program that is running on the controller and also interaction with the program that is in the PLC that simulates the plant. Fig. 4 shows how it is simulated. First the button isn’t pressed, the program from PLC is not running, the led is Red; after that, when the button is pressed, the program from PLC is running and the led is Green.

Fig. 4. CX-Supervisor – view of the simulation of PLC.

At this point it is presented a simple case study, which main purpose is to illustrate how it must be done the plant modeling, in order to make useful the workbench developed specifically for the test and validation of PLC programs.

For the physical system to be simulated it was chosen a simple electropneumatic system composed by three simple-acting cylinders, three respective monostable directional valves associated and, also, two respective position sensors (Fig. 5).

The physical system to be modeled and simulated is composed by:

• Compressor (1 unit); • Multiple distributor with 3 out way (1 unit); • Directional valve 3/2 way (3 units); • Cylinder of a simple effect (3 units); • Limit switches (6 units).

Fig. 5. Illustration of the physical system to be simulated by PLC_2.

The input/output circuit designed for the workbench is presented in Fig. 6.

Fig. 6. Inputs and outputs of the considered system.

There were used the CX-One software and CX-Supervisor from OMRON. CX-One allows using CX-Programmer to edit and simulate programs and CX-Designer that allows creation of simulation scenarios (drawings, schemes, among others) and using these scenarios associated to the program developed in CX-Programmer the user can understand the simulation of the system. Together with these tools, CX-Supervisor is devoted to the design and operation of PC visualization and machine control. It allows interaction between the user and the PLC program that is running on the controller and also the interaction with the program that is in the PLC that simulates the plant.

3.3. Small Intelligent House

A test kit for a “small intelligent house” [2] with the following functionalities was developed: alarm control, temperature control, entrance door open/close, and illumination control (Fig. 7). The sensors were positioned in order to allow control of alarm intrusion, main and internal illumination, main door opening/closing and attic temperature, using the PLC CQM1H-CPU61 from Omron. The simulation of the “small intelligent” house was developed in LabVIEW (Fig. 8). The LabVIEW interface enables both simulation and “small intelligent” house monitoring and actuation, but direct control is performed by the PLC. The interface allows also monitoring and testing different proportional, integral and derivative parameters of the attic temperature control algorithm, as well as real time monitoring using a Webcam.


140

Fig. 7. Small Intelligent House kit.


3.4. Level Control of a Two-Tank System

This work considers the objective of the design and implementation of a remote experiment for controlling the water level in a two-tank system (Fig. 9) [8].

Fig. 9. Two tanks system layout.

The system includes two tanks made in acrylic, a pump to circulate the water from the lower to the upper tank, two ultrasonic sensors for measuring both

tank levels, an electrovalve to stop the flow between the upper and the lower level tank and a manual valve for security purposes. The upper tank is being controlled and the lower tank serves as a buffer.

An 8-bit Atmega16 microcontroller from Atmel is used for data acquisition, control and pump actuation. A custom made LabVIEW tool, the Watch Tank, was developed for monitoring purposes. The control algorithms, on-off and PID are implemented in the microcontroller that is connected to a PC by a RS232 protocol.

The user interface, Watch Tank, developed in LabVIEW allows the monitoring of the system (Fig. 10). In the Watch Tank program, the user must select the PID type algorithm to be tested, as well as the controller parameters, Kp, Ki and Kd, the proportional, integral and derivatives constants, respectively.


The user can test the digital algorithms and parameters, change level reference values and register the output data. The system can be set available with either local or remote control configurations for teaching/learning purposes. A Webcam is used for real-time monitoring.

3.5. Remote Physiological Signals

The study of human physiology allows understanding the human body’s structure and the processes that are carried out within it, which involves mechanical, electrical and chemical forces. So, and in order to support the biomedical engineering undergraduate students’ learning process on physiological data acquisition studies, an innovative remote laboratory has been developed: RePhyS (Remote Physiological Systems) laboratory. RePhyS is capable of engaging biomedical engineering students, providing the real-time remote acquisition of physiological signals from the human body through a web platform. The educational goals should be achieved, namely the knowledge of acquisition methods, the recognition of the signals and the identification of important components of each one [19].


141

RePhyS was developed using the commercial BioStarter® kit, with sensors’ modules: Electrocardiogram (ECG), Electromyography (EMG), Galvanic Skin Response (GSR), Strain Gauge and an accelerometer. The system includes different remote experiments for the acquisition and study of different physiological signals [20]. The data processing, storage and transmission are made by a computer Bluetooth connection. The general architecture of the RePhyS lab is presented in Fig. 11. The ShimmerTM device captures the data through the electrodes placed on the human body. The Bluetooth (IEEE 802.15.1) connection established between the computer, in the real laboratory, and the experiment allows the control and the configuration of the device, and the transfer of measured physiological data. The students, located in a distance place, access the experience through an interface, on an Internet browser, facilitating the interaction with the device through various control buttons. A webcam allows monitoring of the experience.

Fig. 11. General architecture of RePhyS.

Fig. 12 shows the user interface for the acquisition of ECG remote data acquisition. The experiments developed were initially implemented with LabVIEW being others user-side solutions (C and Java) under development [20].

Fig. 12. RePhyS user interface during an ECG remote acquisition.

Thus, the student/user can acquire, visualize and analyze, in real-time, the results of practical experiences being able to interact and to control the

measurement parameters. The stored data is always available for further processing.

Currently, the architecture defined for the first version of the remote laboratory is being improved. Thus, RePhyS will allow simultaneous access up to three users, each having a control experiment, and will also allow the simultaneous use of several modules that is, in the same experiment, able to study various physiological conditions of the system.

4. Students’ Feedback: Qualitative

and Quantitative Study

The authors of this paper have been V&R laboratories as a tool in using the teaching/learning process of Automation and Control subjects in engineering courses since 2006, where simulations tools were developed showing their importance in the students’ learning process on these subjects [21-23].

As stated by [24] virtual laboratories can be used as a student’s first contact with the studied problem and remote laboratories complement to real hands-on experiments.

So, if V&R laboratories are to be used as an educational tool, they must fulfill the pedagogical purpose that they were developed.

For all educational kits and systems described herein, and in order to obtain the students’ feedback, questionnaires were always presented at the end of each semester course and results analyzed by the teachers involved in the process. Recently, the questionnaire also allows the identification of students’ perception in relation to their own learning style employed during the experiment performance, in accordance with the students’ perspective. The analysis of this factor becomes an important indicator on how these kits or the V&R laboratories procedure is appropriate for the students’ learning process and if they are suitable for the engineering learners (the end users) [17, 25].

The use of these questionnaires enabled the improvement of the educational kits in order to accomplish their objective. The questionnaires had open questions in order to obtain students feedback towards specific points as well as quantitative questions defined in the Likert scale (1, strongly disagree, to 5, completely agree). The questionnaires were divided in sections: after the student characterization, close, open and multiple choices responses, and the evaluation of several statements for the experiment evaluation, the simulator assessment, and to analyze students’ perspectives, fillings and knowledge before and after carried out the experiment were used. The analysis was performed using SPSS software.

The students that participated in this study were from Mechanical, Electronics and Biomedical engineering courses (n=200, total number of students), having an average of 21 years old, of which 20 % were female (Biomedical presents the


142

highest percentage of female students with 40 %). Concerning the experiment evaluation, on average, students evaluate it with a positive score (values higher than 3 for Q1, Q2, Q4, Q5, Q6 and lower than 3 for Q3), as described in Table 1.

Table 1. Kits average students’ evaluation by item.

Item Average Q1: In general, I was motivated for the use of these kits in the course context.

4.4

Q2: In general, I can say that the performance of the kits…

1: … helped me assimilate the concepts presented during the course.

3.6

2: … made my learning more objective. 3.7 3: … increased my chances of getting a

high final evaluation. 3.5

4: … motivated me towards the course. 3.8 5: … raised my expectations relative to

the assessment. 3.5

Q3: Running the kits had nothing to do with my motivation and my interest in this course.

2.2

Q4: The kits are suitable for my Control/Automation learning process.

4.2

Q5: I recommend the implementation of these kits and activities, in the next school year, as a teaching/learning tool.

4.3

Q6: The implementation of the remote experiment …

1: … has increased understanding of the operation of an on-off controller.

3.5

2: … has increased understanding of the operation of the PID controller.

3.1

3: … allow the visualization of the effect of the off-set reduced when going from a controller P to PI.

3.0

4: motivated to learn the subjects under study.

3.5

Based on a protocol, Biomedical engineering students performed several remote experiments in physiological signals acquisition (RePhyS: ECG, body acceleration and body temperature acquisition). All students understood the ECG signal acquisition and they were able to identify all the waves’ constituents during the data real-time visualization (as demonstrated in Fig. 12). For the other two experiments, 89 % identified to have understood and visualized the experiments.

Regarding students’ perception in relation to their own learning style and the learning style employed during the experiment performance, it is possible to observe that, and based on the Biomedical engineering students, 67 % prefer to process information more actively (2nd and 3th quadrants, Fig. 9), and 67 % also prefer to understand the concepts through concrete experiences (1st and 2nd quadrant, Fig. 13). This behavior (expected in some way) is coherent to an engineering behavior: motivated to investigate how situations are processed and to put into practice ideas. This information is

important as it suggests that the development efforts of these didactics kits meet engineering students’ learning styles.

Fig. 13. Learning Styles diagram obtained: 1st quadrant – diverging style,

2nd quadrant – accomodating style, 3rd quadrant – covengent style,

4th quadrant – assimilating style.

The qualitative analysis of the data helped to recognize the necessity of devising and implementing new learning strategies to fulfill students’ needs. The students replied to the questionnaire on a voluntary basis and according to the experiment conducted. A majority, if not all students who performed the experience, has accepted this challenge.

The strategy designed that V&R laboratory development is appropriated for the target public’s learning style (converging learning style) where students do the things in an active way and they have a more abstract perception of the situations.

In general, the students’ feedback about the use of the kits and system in Automation and Control Engineering studies is very positive stating that these kits and activities should be used in the coming years, and being suitable for the Automation and Control learning process. The majority of students felt encouraged in using these kits and sentences like: “an innovative and useful tool”; “easy to understand”, and “rewarding experience” were found.

The less positive issues pointed out by students were concerned to more technical aspects, like specific problems with the Internet connection, real-time monitoring of an experiment. For a continuous improvement, new alternatives are under studied to overcome the identified problems. The feedback from students to these real-world engineering problems was very positive allowing the researchers to continue and improve this methodology. 7. Final Remarks

This work describes the work undertaken at two Higher Education Institutions, from north of Portugal, on the development of V&R laboratories as


143

facilitators in the students’ study. The students can be from different engineering branches, aiming towards a final objective of optimizing their learning about different important subjects that must be highlighted in the context of teaching of Control and Automation for different future Engineers in many areas, including mechanical engineering, biomedical engineering, and electronics engineering that will need Control and Automation knowledge during their professional life.

Although some of the didactics kits developed and presented in this paper could be considered comparable to some described in the literature (while it has been authors’ choice to include as they are part of the automation and control laboratory described), the acquisition and treatment of physiological signals correspond to an innovative remote laboratory. So, this paper describes the development of the virtual and remote laboratory since its creation until today, highlighting the strengths of the didactics kits selected.

A real added value of this project is the use of dedicated workbenches – covering, in a complementary way, the domains related with teaching of control and automation fields - for dedicated needs of real learning, as commercial tools available are very specific mismatching important concepts, together, in the Control and Automation domains.

Based on the different views expressed by students, the final users of the kits and system in Automation and Control Engineering studies, alongside positive points, there are also a few negative points to be identified. The positive points state that these kits and activities are suitable for the Automation and Control learning process and should be used in the coming years. The less positive point identified is concerned to technical aspects, namely specific problems with the Internet connection and real-time monitoring of an experiment.

Finally, it is important to highlight that strong physical and human resources have been optimized because students from both institutions can use this remote access laboratory and experiences have been developed taking into account their specific Curricular Unit’s needs. Acknowledgements

This work has been supported by FCT – “Fundação para a Ciência e Tecnologia” in the scope of the project: PEst-UID/CEC/00319/2013. The authors are grateful to the students that participated in this project as designers and as end users. References [1]. F. Soares, C. P. Leão, J. Machado, V. Carvalho,

Automation and Control in Engineering, A Global Approach with Educational Kits, in

Proceedings of the 6th International Conference on Sensor Device Technologies and Applications (SENSORDEVICES’15), Venice, Italy, 23-28 August 2015, pp. 7-12.

[2]. L. Gomes, S. Bogosyan, Current Trends in Remote Laboratories, IEEE Transactions on Industrial Electronics, Vol. 56, Issue 12, December 2009, pp. 4744-4756.

[3]. P. M. Silva, P. C. Oliveira, N. C. Carvalho, C. P. Leão, F. Soares, Virtual and Remote Control: an experience by students for students, in Proceedings of the International Conference on Remote Engineering and Virtual Instrumentation (REV’07), Porto, Portugal, June 2007, p. 3.

[4]. S. Santos, B. Alves, F. Soares, Remote Laboratory for Control Studies – A Case-study Developed by Students for Students, in Proceedings of the 9th Portuguese Conference on Automatic Control (Controlo’10), Coimbra, Portugal, September 2010.

[5]. T. Karakasidis, Virtual and remote labs in higher education distance learning of physical and engineering sciences, in Proceedings of the IEEE Global Engineering Education Conference (EDUCON’13), Berlin, Germany, 13-15 March 2013, pp. 798-807.

[6]. T. Machet, D. Lowe, Issues Integrating Remote Laboratories into Virtual Worlds, in Proceedings of the 30th Australian Society for Computers in Learning in Tertiary Education (ASCILITE’13), Sydney, Australia, 1-4 December, 2013, p. 521-525.

[7]. X. Chen, G. Song, Y. Zhang, Y., Virtual and Remote Laboratory Development: A Review, Earth and Space, 2010, pp. 3843-3852.

[8]. N. Brito, et al., A Remote System for Water Tank Level Monitoring and Control - a Collaborative Case-study, in Proceedings of the 3rd IEEE International Conference on e-Learning in Industrial Electronics (ICELIE’09), Porto, Portugal, November 2009, pp. 19-23.

[9]. A. Mokhtar, G. R. Mikhail, C. Seong-Joo, A Survey on Remote Laboratories for E-learning and Remote Experimentation, Contemporary Engineering Sciences, Vol. 7, Issue 29, 2014, pp. 1617-1624.

[10]. J. R. Brinson, Learning outcome achievement in non-traditional (virtual and remote) versus traditional (hands-on) laboratories: A review of the empirical research, Computers & Education, Vol. 87, September 2015, pp. 218-237.

[11]. L. O. Flowers, Investigating the Effectiveness of Virtual Laboratories in an Undergraduate Biology Course, Journal of Human Resources & Adult Learning, Vol. 7, Issue 2, December 2011, pp. 110-116.

[12]. D. Santos-Martin, J. Alonso-Martínez, J. Eloy-Garcia Carrasco, S. Arnaltes, Problem-Based Learning in Wind Energy Using Virtual and Real Setups, IEEE Transactions on Education, Vol. 55, Issue 1, February 2012, pp. 126-134.

[13]. G. C. Goodwin, A. M. Medioli, W. Sher, L. B. Vlacic, J. S. Welsh, Emulation-Based Virtual Laboratories: A Low-Cost Alternative to Physical Experiments in Control Engineering Education, IEEE Transactions on Education, Vol. 54, No. 1, February 2011, pp. 48-55.

[14]. M. J. Callaghan, A. G. Eguíluz, G. McLaughlin, N. McShane, Opportunities and Challenges in Virtual Reality for Remote and Virtual Laboratories, in Proceedings of the 12th International Conference on Remote Engineering and Virtual Instrumentation


144

(REV’15), Bangkok, Thailand, 25-27 February 2015, pp. 235-237.

[15]. N. Nedev, Z. Nenova, S. Ivanov, Virtual Instruments for Humidity and Temperature Measurements, Information Technology Based Higher Education and Training (ITHET’14), 11-13 September 2014, York, England, pp. 1-6.

[16]. I. Santana, M. Ferre, E. Izaguirre, R. Aracil, L. Hernández, Remote Laboratories for Education and Research Purposes in Automatic Control Systems, IEEE Transactions on Industrial Informatics, Vol. 9, No. 1, Feb. 2013, pp. 547-556.

[17]. F. Soares, C. P. Leão, V. Carvalho, R. M. Vasconcelos, S. Costa, Automation & Control Remote Laboratory: A Pedagogical Tool, International Journal of Electrical Engineering Education, Vol. 51 Issue 1, January 2014, pp. 54-67.

[18]. K. Ogata, Modern Control Engineering, 4th ed., Prentice-Hall, New York, 2001.

[19]. C. Barros, C. P. Leão, F. Soares, G. Minas, J. Machado, RePhyS: A Multidisciplinary Experience in Remote Physiological Systems Laboratory, International Journal of Online Engineering, Vol. 9, No. S5, June 2013, pp. 21-24.

[20]. C. Barros, C. P. Leão, F. Soares, G. Minas, J. Machado, Issues in remote laboratory developments for biomedical engineering education, in Proceedings of the International Conference on Interactive Collaborative Learning (ICL’13), Kazan, Russia, Sep. 2013, pp. 290-295.

[21]. C. P. Leão, F. O. Soares, Application of new methodologies in an industrial electronics engineering course: case study, iNEER Special Volume: INNOVATIONS 2006 - World Innovations in Engineering Education and Research, Chapter 28, A. Win, et al., Eds., 2006, pp. 323-332.

[22]. C. P. Leão, F. O. Soares, V. Carvalho, R. M. Vasconcelos, Automation and Control Engineering Laboratory: Students Perspectives, Lectures Notes in Engineering and Computer Science, Vol. 2185, Issue 1, 2010, pp. 2017-2021.

[23]. C. P. Leão, F. Soares, S. Costa, Insights on Using WALC Platform as a Learning Tool, International Journal of Engineering Pedagogy (iJEP), Vol. 2, Issue 1, 2012, pp. 12-16.

[24]. C. A. Jara, F. A. Candelas, S. T. Puente, F. Torres, Hands-on Experiences of Undergraduate Students in Automatics and Robotics Using a Virtual and Remote Laboratory, Computers & Education, Vol. 57, Issue 4, 2011, pp. 2451-2461.

[25]. C. Barros, C. P. Leão, F. Pereira, F. Soares, J. Machado, D. Matos, V. Carvalho, Assessing Remote Physiological Signals Acquisition experiments, in Proceedings of the International Mechanical Engineering Congress & Exposition (IMECE’14), Montreal, Canada, Nov. 2014, pp. V005T05A049-7.

___________________


http://www.sensorsportal.com/HTML/BOOKSTORE/Handbook_of_Measurements.htm





145



Improving Systems Dynamics by Means of Advanced Signal Processing – Mathematical, Laboratorial and Clinical Evaluation of Propofol Monitoring

in Breathing Gas

1, 2 Dammon ZIAIAN, 2, 3 Philipp ROSTALSKI, 4 Astrid Ellen BERGGREEN, 4 Sebastian BRANDT,

4, 5 Martin GROSSHERR, 4 Hartmut GEHRING, 2, 6 Andreas HENGSTENBERG and 1 Stefan ZIMMERMANN

1 Institute of Electrical Engineering and Measurement Technology, Appelstraße 9a, 30167 Hannover, Germany

Tel.: +49 511 762 4671, 2 Research Unit, Draegerwerk AG & Co.KGaA, Moislinger Allee 53-55, 23558 Luebeck, Germany

3 Now at Universitaet zu Luebeck, Institut fuer Medizinische Elektrotechnik, Ratzeburger Allee 160, 23562 Luebeck, Germany

Tel.: +49 451 3101 6200, 4 University of Luebeck, Dept. of Anesthesiology and Intensive Care Medicine,

Ratzeburger Allee 160, 23538 Luebeck, Germany Tel.: +49 451 500 4057,

5 Segeberger Kliniken, Abteilung für Kardioanästhesie,Am Kurpark 1, 23795 Bad Segeberg, Germany Tel.: +49 451 593 608

6 Now at SICK AG, Merkurring 20, 22143 Hamburg, Germany E-mail: [email protected], [email protected],

[email protected], E-mail: [email protected], [email protected], [email protected], [email protected],

[email protected]

Received: 31 August 2015 /Accepted: 5 October 2015 /Published: 30 October 2015 Abstract: Electrochemical sensors are used in various gas measurement applications and are available for different gases. Depending on the application, the sensor might need to be installed far away from the actual measurement site, requiring the use of long sampling lines. Examples are portable gas measurement devices in which remote locations like tanks and chemical reactors need to be monitored. But also medical applications, where the sensors cannot be positioned in close vicinity to the patient, are common like, e.g., the side-stream measurement of breathing gas. Due to the characteristics of electrochemical sensors and to the adsorption and desorption behavior of sampling lines for different gases, the electrical sensor signal may indicate long response times. In this paper, we propose an on-line signal processing algorithm which is capable to significantly improve the performance. After characterizing the dynamic behavior of the sensor system, a properly designed deconvolution filter is used to reduce response time and signal noise. Within this article, we also provide an




146

example of this algorithm for a novel electrochemical sensor for the measurement of the anesthetic agent propofol in exhaled air. For this application, the acceleration is prerequisite for the measurement chain to be of practical use in a clinical setting. Our goals, to establish measurement dynamics to record the physiologic parameter and to reduce non-physiological disturbances, were achieved with additional reserves. This article is based on [1] and is extended by original clinical data. As an example, we present propofol monitoring in breath of one patient in order to demonstrate the performance of the introduced algorithm in a real clinical application. We proved that the electrochemical sensor, associated with the provided algorithm, is capable for real-time monitoring in a clinical setting. Copyright © 2015 IFSA Publishing, S. L. Keywords: Deconvolution, Electrochemical sensor, Propofol, Response time, Noise reduction. 1. Introduction

Electrochemical sensors [2] are widely used for measuring gases in various industries, most notably process industry, oil and gas, but also in medical applications. Various research and development activities have gone into optimizing the design, material and electrochemical properties of this type of gas sensor in order to improve response time, selectivity, accuracy, precision, minimizing the drift and other adverse effects. However, a certain delay in the response due to the diffusion, chemical reaction but also adsorption/desorption is inherent in any sensor.

We will discuss a specific medical application for an electrochemical gas sensor, but similar techniques may be used in other areas. After describing the specific application and measurement chain for the physiological signal considered for the remainder of this article, we will discuss the design of a deconvolution filter to accelerate the system response while filtering non-physiological disturbances. The performance of the filter is derived mathematically and analyzed, both based on laboratory measurements and in a clinical setting. The intention of this work is to assess how far the usage of this electrochemical sensor for patients is feasible under real clinical conditions. For that reason breathing gas was sampled of a number of patients which received propofol for anesthesia and analyzed online. In Section 2, the specific application and the experimental setup are described. Furthermore, mathematical explanations on modeling and the design of the algorithm are given and the clinical setup for data collection at the bed side is introduced. Results of the signal acceleration and noise reduction are presented in Section 3 together with breath measurements of one patient during the awakening phase of anesthesia. At the end, the application of advanced signal processing is discussed and concluded in Section 4.

2. Material and Methods

2.1 Specific Application As a practically relevant example, an electro-

chemical gas sensor which is used to quantify the

concentration of 2,6-Diisopropylphenol, also known as propofol, is considered [3]. Propofol is applied as an anesthetic agent for humans and animals. It is intravenously administered in the formulation of a lipid emulsion. Its volatile characteristics allow detecting propofol in the breathing gas after injection [4]. During anesthesia the exhaled concentration appears in the order of around 20 ppb (parts per billion) [5, 6]. As a result, the required sensitivity and accuracy of the sensor need to be suitable to ensure a reliable measurement.

When handling substances in such low concentrations, effects of adsorption and desorption along the measurement chain have a particularly strong impact. In the case discussed here, the carrier gas is drawn through a 2.5 m long sampling line from the propofol source which happened to be a gas cylinder (carrier gas: N2, propofol concentration: 40 ppb ±30 % rel. standard deviation). Since the true concentration of one cylinder lies between 28 ppb and 52 ppb according to the vendor, the sensitivity of the sensor cannot be stated based on the actual data. The schematic of the setup is presented in Fig. 1. A sampling flow of 180 ml/min is generated with the help of a pump. Hence, the carrier gas is led through the sampling line to interact with the sensor.

Fig. 1. Schematic of the experimental setup. The side stream is driven through the sampling line to pass the

electrochemical sensor for detection. A T-piece connector is used for switching.


147

The saturation of propofol on the inner surface of the line leads to a major dynamic delay in time before the electrochemical sensor is able to detect the absolute concentration. Besides the delay caused by the sampling system, the response characteristics of the electrochemical sensor itself adds further major delay. It is assumed that these two effects are responsible for the main delay. Another cause refers to the volume of the line which the gas has to pass before reaching the sensor. This volume leads to a constant, non-dynamic delay. Considering an inner tubing diameter of 3.2 mm, given flow rate and given length of the sampling line, a dead time of 5.4 sec is created. However, 5.4 sec are negligible in contrast to the major dynamic delay reasons.

Recorded signals are usually corrupted by noise. Owing to the pump, to thermal noise within the electronic components and other effects, all measured values possess a specific variance. Besides the minimization of the response time, another objective of the signal processing algorithm proposed in this article is to increase signal quality in real-time. We have been aware of the fact that signal acceleration might lead to over-proportional amplification of non-physiological disturbances. In the light of the aforesaid, it is mandatory to seek for this goal.

2.2. System Identification

Any signal processing procedure needs to be tailored to the specific application. This can be either done heuristically by following tuning rules or by using a model-based procedure. In this publication we follow the latter route. System identification is the necessary first step to identify the underlying system model.

2.2.1. Recording the Step Response A step change of propofol gas was applied in order

to excite the measurement system. Data was recorded using the setup illustrated in Fig. 1. At defined times the propofol-free sampling line was connected and disconnected to the main stream which contained propofol saturated gas of approximately 40 ppb. One standardized cycle consists of three minutes of recording the baseline with propofol-free room air, followed by 30 minutes of connection to the main stream. And finally, the system was purged for 30 minutes with room air by disconnection from the T-piece connector. In Fig. 2, the input excitation is shown in green and the resulting step response of the sensor system in blue. Dotted lines indicate the 90 % value and the steady state of the response. In this particular measurement it lasts

sec40190 =t (1) before 90 % of the steady state value is reached.

Fig. 2. Result of one measurement cycle. Excitation sequence is shown in green and sensor signal in blue. 90 %

of the ultimate value is reached within 401 sec after connecting to the main stream.

To quantify the precision of each measurement the signal to noise ratio (SNR), here defined as

deviation standard

amplitude=SNR (2)

is calculated. The amplitude derives from the mean value of a short time period towards the end of exposure to propofol gas and is thus equal to the steady state value. The SNR may be understood as an intra-measurement precision, whereas the evaluation of a set of multiple measurements leads to the overall precision of the sensor system. A higher SNR reflects a better precision. For the measurement shown in Fig. 2 the resulting ratio appears as 367, indicating a rather low noise situation. Again, the actual noise situation is not the reason for seeking a better SNR, but the signal acceleration, explained in what follows, makes it mandatory. 2.2.2. Modeling and Parameter Identification

System identification can be done using different

methods, ranging from white box modeling based on first-principles with parameters derived using physically measures to black box modeling, where no prior knowledge about the model is available. We will follow the latter approach with two assumptions on the model structure. The step response from Fig. 2 suggests a first or second order response, without overshoot and no oscillatory components. A reasonable model structure (second order) in the time domain is thus given by

.1)(12

2

21

1mod

21

−−+

−−+=

−−

TT

TTe

TT

TTektf zT

t

zT

t

el (3)


148

Therein k represents the static gain. zTTT and, 21 characterize the dynamics of the system and t is set to be the time variable. By means of the Laplace transformation [7] the same relationship may be stated in the frequency domain as

)()(mod sFtf modelel ⇔ (4)

)()()(

21

11

1

21 TT

Tzmodel ss

s

TT

TksF z

++

+= (5)

,)1()1(

1

21 +++=

sTsT

sTk z (6)

where s is defined as the complex angular frequency.

The parameters of this model are computed using an optimized fitting procedure. With the help of a least squares method the set of parameters ,k ,zT 1T and 2T

are determined which yield the smallest sum of squares error between the modeled and the actual response. The best values found for this particular setup are as follows:

1=k (7)

sec03.413=zT (8)

sec95.5361 =T (9)

sec49.522 =T (10) In Fig. 3 the result of the modeling and parameter

identification is illustrated.

Fig. 3. Result of the modeling and parameter identification. The measured time course is displayed in blue

and the modeled signal in orange. The blue line represents the measurement as

displayed in Fig. 2. After finding an appropriate model structure and reasonable parameters the modeled

signal, drawn in orange, can be calculated. The visual matching indicates that the fitted curve agrees well with the measurement. The remaining mismatch is likely to be a result of the inherent non-linearity of different dynamics for rising and falling concentrations. With a hybrid model including two switching dynamics for rising and falling signal response respectively the fitting curve would show a better match, albeit at the price of mathematical complexity and a non-linear behavior. 2.3. Design of Algorithm

In [8], the physiological lung dynamics regarding the propofol exchange from blood plasma to breathing gas are described by a first order differential equation. Its time constant T, which is defined as the time to reach 63 % of the final propofol concentration in the lung due to a sudden change in the blood plasma propofol level, is estimated using clinical patient data to be sec414T = in mean, approximately corresponding to a respond time of 90, 952 sec.breatht = We expect that a proper metering for anesthesia monitoring is performed when the detection happens three times faster than the parameter might change. Considering this, the sensing system should not exceed a maximal response time of 90,max 317 sect =

for a clinically relevant measurement of propofol in breath.

As mentioned before, the main issue of the electrochemical measurement system for propofol is its long response time. Major causes for this are adsorption/desorption effects in the sampling system as well as the inherent measurement dynamics of the electrochemical sensor itself. All of these elements lead to a delayed response between the propofol concentration in breath and the signal provided by the sensor with a typical response time of .sec40190 =t Fortunately in a clinical environment most of the factors determining the delay are almost constant over time and depend only on the measurement chain. The delay can thus be compensated by using linear signal processing. The design of such an algorithm is the content of this section.

2.3.1. Deconvolution The measured signals )(tφ are the result of the

time course of the propofol concentration in the breathing gas )(tcbreath and ),(tfsystem which is meant

to be the unknown response characteristic of the measuring system,

),()()( tctft breathsystem ∗=φ (11)

where ∗ denotes the convolution operator.


149

During a measurement, )(tfmodel (identified in

Section 2.2.2) and all past values of )(tφ are known.

The aim is to compute the original source signal )(tcbreath with these known information. It is possible

to estimate the delayed signal by inverting the slow dynamic components of the measurement chain. Mathematically, this means that we need to invert the effect of the convolution through deconvolution with all transfer elements between source and electrical signal of the sensor. Deconvolution is best understood in frequency domain. Each frequency component is delayed and damped individually by the measurement chain

).()()( sCsFs breathsystem=Φ (12)

The idea of deconvolution is simply to shift and

amplify each frequency component accordingly to reverse this effect. Based on the model identified in Section 2.2.2, we can approximate )(sCbreath through

deconvolution as

( )(s)inv)()(ˆmodelbreath FssC ⋅Φ= (13)

,(s)

1)()(

modelbreathsystem F

sCsF= (14)

where )(sΦ denotes the frequency transform of the

measured signal, (s)modelF is the model identified in

the previous section and )(ˆ sCbreath provides the

estimated propofol concentration. This procedure, however, is not realizable for a

number of reasons in practical setting requiring real-time computation in a medical device. First and foremost, the deconvolution equation (14) as given above cannot be realized, at least not without modification. Any causal system satisfies the property that its numerator order is equal or lower than its denominator order. This translates into the fact that at each point in time we only measure the next signal value but not its time derivatives.

( ) ( )( )( )1

11

(s)

1(s)inv 21

+++==

sTk

sTsT

FF

zmodelmodel (15)

Not keeping causality would incorrectly imply that

an effect may appear before its cause. Another potential issue of a simple inversion is

noise. In reality, (12) can be rewritten as

),()()()( sRsCsFs breathsystem +=Φ (16)

where )(sR denotes the Laplace transform of the

noise. The estimation of )(ˆ sCbreath considering the

noise term results in

.

(s)

)(

(s)

)()()(ˆ

modelmodel

breathsystembreath F

sR

F

sCsFsC += (17)

Most real systems, including the measurement

chain in question, have a low-pass behavior which dampens high frequency noise. Inverting the transfer function of such a system would result in a high-pass behavior which highly amplifies non-physiological signal components such as noise and distorts the actual propofol signal. Our aim, to increase the SNR, addresses this over-proportional amplification of noise during deconvolution.

2.3.2. Causality and Noise Treatment One potential solution to overcome the issues

mentioned above is to augment the deconvolution filter in (17) by a low-pass filter

(s)

)()(

(s)

)()()()(ˆ

estimation

model

filter

model

filterbreathsystembreath

F

sFsR

F

sFsCsFsC

+

= (18)

( )

(s)

)()()()(

model

filterbreathsystem F

sFsRsCsF += (19)

.(s)

)()(

algorithmtmeasuremen model

filter

F

sFsΦ=

(20)

If the filter order is chosen high enough, the

causality of the overall system is satisfied. Since ( ),(s)inv modelF see equation (15), has a numerator

order of two and a denominator order of one, a low-pass filter )(sFfilter with an order of at least one would

therefore fulfill the need for causality.

2.3.3. Low-pass Filter Dedicated to the electrochemical measurement

system and to the application requirements the following low-pass filter –a second order Butterworth filter [9] – has shown sufficient performance. Its flatness and linearity in the pass band and the uncomplicated design are beneficial in our case. Other types of filtering may be more favorable depending on the application.

In order to determine a suitable cut-off frequency we considered the appearing dynamics. The dynamics of propofol exhalation in breathing gas can be derived from [8, 10]. In a subsequent adjustment procedure our filter parameters have been fine-tuned during application to obtain a satisfying compromise between


150

noise rejection and response time. With the resulting cut-off angular frequency of

sec

rad1042 2−⋅== fπω (21)

the transfer function of the Butterworth filter is given as

.

sec

rad0016.0

sec

rad05657.0s

sec

rad0016.0

)(

2

22

2

2

++=

s

sFfilter (22)

2.3.4. The Resulting Algorithm and its

Software Implementation As implied in (20), the algorithm for the signal

processing is composed as

.

(s)

)()(

model

filteralgorithm F

sFsF = (23)

For our particular case, the algorithm results in ( )( )

( )

2

22

2

2

21

sec

rad0016.0

sec

rad05657.0s

sec

rad0016.0

1

11)(

++⋅

+++=

s

sTk

sTsTsF

zalgorithm

(24)

due to the model characterized in Section 2.2.2 and due to the filter described in Section 2.3.3. All poles of

)(sFalgorithm are negative, thus stability is ensured.

The continuous frequency domain is helpful for design matters. But for the implementation as a real-time capable algorithm few more steps are required. Since the sensors output is available digitally it exists of discrete values appearing in a specific rate. Therefore, it is necessary to transform into the discrete time domain. With the use of the bilinear transform [11] and the properties of the z-transform [12] the algorithms output at a certain point in time can be stated as a sum. (25) depicts a general description. Therein, )k(y expresses the algorithms output and

)k(x the measured value, while k denotes the discrete

time variable.

[

])nk()1k(

)nk()1k()k(1

)k(

1

100

−++−−

−++−+=

yaya

xbxbxba

y

n

n

(25)

Hence, )k(y is calculated as a linear combination of

previous calculations and measurements, whereas

n0a and n0b denote the coefficients characterizing

the algorithm. Applied on )(sFalgorithm of (24), we find a compact

description to be implemented in any software as a real-time capable signal processing algorithm:

[

])3k()2k(

)1k()3k(

)2k()1k()k(1

)k(

32

13

2100

−+−−−−−+

−+−+=

yaya

yaxb

xbxbxba

y

(26)

After transformation for a sampling rate of 1 Hz

the coefficients can be found in Table 1. The given precisions of decimal places are required for stability, when the algorithm is programmed to perform.

Table 1. Discrete Time Coefficients for a 1 Hz-Clocked Deconvolution Filter.

Coefficient Value

0a 1

1a -2.9551616730526264

2a 2.9113070363222864

3a -0.95614323255658584

0b 0

1b 0.060676477761667597

2b -0.12009490846751848

3b 0.059420561418924996

2.4. Clinical Data

As clinical patient data are part of our evaluations, we will give an overview of how this data was collected.

With written informed consent and approval of the local ethnic committee patients undergoing surgeries demanding anesthesia were monitored. Anesthesia was performed using propofol and remifentanil. Their application was performed by means of target controlled infusion pumps. During a standardized course of anesthesia, breathing gas was sampled in order to quantify the propofol concentration in real-time. Standardly, the blood plasma target concentration was maintained at 2 µg/ml for 15 min before the intake of propofol and remifentanil were stopped. This phase took place after the actual surgical procedure. The connection to the ventilation circuit for the propofol measurement happened postoperatively. In Fig. 4, the clinical setup is referred to. While similarity to the laboratorial setup is given, the main difference regards the T-piece connector which is placed between the patient and the ventilator machine. The quantification of the propofol concentration occurred in real-time using the technologies explained in the previous sections.


151

Fig. 4. Clinical setup for the measurement of propofol in breathing gas of a ventilated patient.

3. Results

Due to the concise description of the algorithm presented in the previous section it is possible to calculate an accelerated signal in real-time as well as retrospectively. Further on, results are presented for a step change of the propofol concentration as displayed in Fig. 2 of Section 2.2.2 and for repeated measurements.

In Fig. 5, the post-processed signal is illustrated in red. This example visualizes the improvement possible through signal processing. In this particular case, the response time 90t is notably reduced from

401 seconds to 104 seconds. In Section 2.3, the maximal tolerated response time is mentioned to be

.sec317max,90 =t Therefore, this requirement is

fulfilled with an additional reserve. The secondary objective stated is an increase of precision. By noise treatment consideration during the filter design the SNR is enhanced from 367 to remarkably 1482. As the measuring system is afflicted with non-linearity the algorithm shows a different result for rising and falling signals. Both overshooting and undershooting lead to higher (102.3 %) and lower (-1.1 %) values. However, the error stays below ±5 %, which is an acceptable result compared to the enhancements in response time and noise reduction.

To express the performance and the repeatability of the algorithm, repeated measurements were evaluated. The same setup was used at different times whereas the algorithm was executed online in real-time. As an example, the results for three repetitions are displayed in Fig. 6. Pairwise-colored curves denote sensor signals and their related real-time processed estimation of the input excitation. It might be that a drift of the propofol source concentration or of the sensor sensitivity have occurred during the repetitions. Even though, the algorithm has performed stable and with expected results. Summarized, we observe that

results are highly reproducible and that the algorithm performs similarly in real-time as long as main parameters of the system do not significantly change and linearity is given for the tolerated concentration range. This result is valid for the discussed clinical practice and also for several other applications of electrochemical and other sensors.

Fig. 5. The result of the signal post-processing is displayed in red, the sensor signal used for the calculation in blue.

Fig. 6. Three reproduced measurements 1- 3 are illustrated to express the repeatability. The algorithm was applied in

real-time during each measurement. Pairs of the same color represent related sensor and algorithm signals.

In addition to mathematical and laboratorial considerations, we present the performance of the algorithm in a field application while measuring the exhaled propofol concentration of a narcotized patient. In Fig. 7, the time series of the raw sensor signal together with the output of the algorithm are illustrated. The sampling line of the sensor module was connected to the main circulatory ventilation right after the plasma target concentration was set to 2 µg/ml. About 15min later, the target was set to 0 µg/ml leading to an instant stop of infusion. As soon as the intubated patient woke up, the sampling line was disconnected from the respiratory circuit.


152

Consequently, room air was sampled hereafter. During that phase, it is possible to visually recognize that the use of the algorithm leads to an accelerated signal when connecting and disconnecting the sampling line. It performed stable and the noise level was reduced as expected.

Fig. 7. Schematic of the experimental setup. The side stream is driven through the sampling line to pass the

electrochemical sensor for detection. A T-piece connector is used for switching.

4. Discussion and Conclusion

In various practical applications of electrochemical

sensors a sampling line is required to transport the gas from the sampling site to the sensor. This together with the dynamics of the sensor itself might lead to significant delays and adverse measurement dynamics, rendering the electrochemical measurement signal useless for the application. To overcome this obstacle, a solution in form of an accelerating algorithm is presented.

In this article, we have demonstrated that the application of advanced signal processing can help to optimize the performance of electrochemical measurement systems with long sampling lines. For the example of an electrochemical propofol sensor, the response time could significantly be reduced by a factor of 9.2 while the SNR could be increased at the same time by a factor of 4. Furthermore, the proposed procedure involves only straight forward model-based design steps and should thus easily be transferable to other applications. Starting with a modeling and system identification step, the characteristics of the sensor system are identified. Here, a second order equation is used to model the sensor response. Noise considerations lead to the specification of a second order low-pass Butterworth filter and to the design of a deconvolution algorithm.

Our primary objective has been to realize a detecting system able to observe the propofol concentration in patients’ exhaled breath at least three times faster than this physiological parameter might change. With the help of the presented algorithm the accelerated response time of sec10490 =t is 9.2 times

faster than the patients average breath propofol

concentration change with 90, 952 sec.breatht =

The patient with the fastest exhalation dynamic observed in [8] does have a time constant of

.sec227T = This implies a maximal permitted response time of sec174max,90 =t for the measuring

device. Thus, using the presented acceleration algorithm the propofol sensors system is even suitable to monitor such, probably exceptional, fast exhalation dynamics.

Another issue discussed in the paper relates to noise. On the one hand more accurate signals are advantageous in general and on the other hand we are aware of the fact that deconvolution might lead to exceedingly higher noise levels. With a proper choice of a low-pass filter, this effect could be coped with, resulting in a significantly improved SNR, albeit the SNR has been acceptable before processing.

With the decision to run measurements in a real clinical environment, we have achieved insights for propofol monitoring in clinical practice. First of all, we showed that an electrochemical sensor is capable to quantify the propofol concentration in breathing gas in real-time and with an acceptable response time. So far, clinical circumstances have been a significant obstacle leading to large response times in all practical settings. Additionally the interaction with the ventilator machine, disposable articles and clinical air, for example, were questionable. As a result of our study, we observed a stable, accelerating and noise reducing performance of the algorithm following the predictions of our laboratory measurements.

One difference to a clinical setup is regarded to the relative humidity conditions of the gas, which appear much higher when patients’ breathing gas is sampled. Influences of humidity were not observable during other investigation. Thus, the impact is not part of this work and might be a topic to address in future.

It is worth mentioning that the application of the presented signal processing is not limited to the clinical setting. Especially, portable gas detection devices are often used in conjunction with long samplings reaching up to 30 m and thus leading to remarkable delays beside the dead time delays due to volume.

Technologically, other more advanced signal processing algorithms come to mind such as Wiener Filter [13], (linear/nonlinear) Kalman Filter [14] or moving horizon estimation, however at a price of a higher complexity. It will be part of ongoing research activities to evaluate these techniques in the context of electrochemical sensors with long sampling lines and to compare the results against the surprisingly simple and effective solution provided here. References [1]. D. Ziaian, P. Rostalski, A. Hengstenberg and

S. Zimmermann, Reducing System Response Time and Noise of Electrochemical Gas Sensors - Discussed


153

for Propofol Monitoring in Breathing Gas, in Proceedings of the 6th International Conference on Sensor Device Technologies and Applications (SENSORDEVICES’15), Venice, Italy, August 23 - 28, 2015, pp. 85–90.

[2]. P. J. Gellings and H. J. Bouwmeester, The CRC Handbook of Solid State Electrochemistry, Chapter 10, CRC Press, 1997.

[3]. Package leaflet: Information for the user Diprivan 10 mg/ml (1 %) Emulsion for Injection or Infusion, 2015, URL: https://www.medicines.org.uk/emc/ medicine/8375 [accessed: 2015-06-19].

[4]. M. Grossherr, A. Hengstenberg, T. Meier, L. Dibbelt, K. Gerlach, and H. Gehring. Discontinuous Monitoring of Propofol Concentrations in Expired Alveolar Gas and in Arterial and Venous Plasma during Artificial Ventilation, Anesthesiology, Vol. 104, 2006, pp. 786–790.

[5]. M. Grossherr et al. Propofol concentration in exhaled air and arterial plasma in mechanically ventilated patients undergoing cardiac surgery, British Journal of Anesthesia, Vol. 102, No. 5, 2009, pp. 608–613.

[6]. A. Taketa, K. Masui, and T. Kalama, On-line Monitoring of End-tidal Propofol Concentration in Anesthetized Patients, Anesthesiology, Vol. 106, pp. 659–664, 2007.

[7]. H. Dwyer, The Laplace Transform: Motivating the Definition, CODEE Journal, Vol. 125, pp. 1231–1238, 2011, URL: http://www.codee.org/ref/CJ11-0116 [accessed: 2015-06-19].

[8]. D. Ziaian et al., Pharmacokinetic modeling of the transition of propofol from blood plasma to breathing gas, in Proceedings of the IEEE International Symposium on Medical Measurements and Applications (MeMeA’14), Lisboan, Portugal, June 11-12, 2014, pp. 80–84.

[9]. D. Gordon, E. Robertson, and J. J. Dowling, Design and responses of Butterworth and critically damped digital filters, Journal of Electromyography and Kinesiology, Vol. 13, 2003, pp. 569–573.

[10]. C. Horns, D. Wieck, S. Prawn, M. Dolce, C. Patel, and G. Schelling, Time course of expiratory propofol after bolus injection as measured by ion molecule reaction mass spectrometry, Analytical and Bioanalytical Chemistry, Vol. 403, No. 2, April 2012, pp. 555–561.

[11]. G. Zhang, X. Chen, and T. Chen, Digital redesign via the generalized bilinear transformation, International Journal of Control, Vol. 82, No. 4, April 2009, pp. 741–754.

[12]. J. R. Magazine and L. A. Zadie, The analysis of sampled-data systems, Transactions of the American Institute of Electrical Engineers, Part II: Applications and Industry, Vol. 71, No. 5, Nov 1952, pp. 225–234.

[13]. N. Wiener, Extrapolation, Interpolation, and Smoothing of Stationary Time Series, Wiley, New York, 1949.

[14]. R. E. Kalman, A New Approach to Linear Filtering and Prediction Problems, Journal of Fluids Engineering, Vol. 82, No. 1, March 01 1960, pp. 35–45.

___________________


http://www.sensorsportal.com/HTML/E-SHOP/PRODUCTS_4/Evaluation_board.htm

http://www.sensorsportal.com/HTML/E-SHOP/PRODUCTS_4/Evaluation_board.htm


154



The Use of Gas-Sensor Arrays in the Detection of Bole

and Root Decays in Living Trees: Development of a New Non-invasive Method of Sampling and Analysis

1 Manuela BAIETTO, 1 Sofia AQUARO, 2 A. Dan WILSON,

3 Letizia POZZI, 1 Daniele BASSI 1 Dipartimento di Scienze Agrarie e Ambientali (DISAA), Università degli Studi di Milano,

via Giovanni Celoria 2, 20133 Milano, Italy 2 Forest Insect and Disease Research, USDA Forest Service, Southern Hardwoods Laboratory, 432

Stoneville Road, Stoneville, MS 38776, USA 3 Demetra Società Cooperativa Sociale ONLUS, Via Visconta 75, 20842 Besana in Brianza, Italy

1 Tel.: +39025031656, fax: +390250316553 E-mail: [email protected]

Received: 31 August 2015 /Accepted: 5 October 2015 /Published: 30 October 2015 Abstract: Wood rot is a serious fungal disease of trees. Wood decay fungi penetrate and gain entry into trees through pruning cuts or open wounds using extracellular digestive enzymes to attack all components of the cell wall, leading to the destruction of sapwood which compromises wood strength and stability. On living trees, it is often difficult to diagnose wood rot disease, particularly during extreme weather conditions when trees can fail, causing tree parts to fall onto people and property. Today, tree stability evaluation and inner decay detection are performed visually and by the use of commercial instruments and methods that are often invasive, time-consuming and sometimes inadequate for use within the urban environment. Moreover, most conventional instruments do not provide an adequate evaluation of decay that occurs in the root system. A long-term research project, initiated in 2004, was aimed at developing a novel approach for diagnosing inner tree decays by detecting differences in volatile organic compounds (VOCs) released by wood decay fungi and wood from healthy and decayed trees. Different commercial electronic noses (ENs) were tested under laboratory conditions and directly in the field, on healthy and artificially-inoculated stem wood chips, and root fragments. The first stage of the research was focused on testing different commercially available electronic noses (e-noses) for the capabilities of discriminating between different strains and species of wood decay fungi as well as sapwood belonging to different tree species. In the second stage, sapwood of different tree species was artificially inoculated with decay fungi to test the diagnostic ability of the e-noses to detect differences in aroma bouquets emitted by healthy and inoculated woods. Root fragments were then inoculated with specific root decaying fungi and incubated under different types of soils to assess whether soil odors could influence the ability of the e-nose to discriminate between non-inoculated and diseased root fragments. For the final stage, soil air was evaluated for the presence of VOCs released by root-decaying fungi on diseased standing trees cultivated in the urban environment. Copyright © 2015 IFSA Publishing, S. L. Keywords: Electronic nose, Decay detection, Urban forestry, VOCs, Tree.




155

1. Introduction Trees in urban environments are cultivated under

conditions that are extremely hostile (causing many stresses and negative effects) and consequently inadequate to sustain healthy plant life. Frequently, trees must face high levels of air, soil and water pollution [1]. Root development is often limited by permanent water stress and very small soil volume for root expansion, providing inadequate support of above-ground plant parts [2-3]. Moreover, road works, hasty or poor pruning methods, and vandalism increase tree stress in the urban environment. These adverse environmental factors dramatically increase physiological stresses that decrease tree fitness and increase susceptibility to attack by pathogenic agents [4].

Wood decay fungi are some of the worst microbial pathogens because they can take advantage of tree physiologic stresses by attacking and destroying all woody components, reducing tree structural stability, leading to failure (breaks) especially during severe weather events [5-6]. Root rots (decays) are even more dangerous and severe, due to difficult detection and the possibility of causing wind throw to the ground (complete tree loss). Trunk and root rot diagnoses in standing trees currently are performed primarily by electrical conductivity meters, constant feed drills, single pulse sonic and ultrasonic techniques, core samples, computerized tomography, and molecular methods for identification of decay fungi [7]. These tools and methods are pensive, invasive, require very skilled personnel, and do not provide systemic information. For these reasons, a multi-year study aimed at developing a novel approach for diagnosing inner tree decays using several gas sensor arrays or ENs was tested.

2. VOCs Emitted by Wood Decaying Fungi

Live standing trees containing decayed wood release a particular mixture of volatile organic compounds (VOCs) consisting of fungal metabolites, tree metabolites, and fungus-induced tree antimicrobial defense compounds (e.g. phenolic metabolites, terpenoids, isoprenoids, and phytoalexins). The composition of metabolites released by individual fungi is controlled largely by the types and combinations of metabolic pathways specific to microbial species, which are regulated by genetic, substrate and environmental factors [8]. Korpi, et al. [9] found microbes that released pinenes, acrolein, ketones and acetylenes that were irritants to mice.

Other investigations have focused on the identification of VOCs released by food spoilage fungi [10-11]. The compound 1-octen-3-ol was detected in damp houses containing various mold fungi [12]. Numerous other chemical species have

been reported as fungal metabolites, including complex acids, sesquiterpenes, methyl ketones and alcohols [13].

Relatively few recent studies have reported on the release of VOCs by healthy and decayed trees. An analysis of healthy Populus spp. and Pinus spp. indicated the presence of mainly monoterpenes, acetone and small amounts of isoprene [14]. Other studies have indicated increases in toluene and α-pinene emissions associated with P. sylvestris under pathogen attack [15], and a decrease in isoprene emissions from diseased Quercus fusiformis L. and Q.virginiana L. [16].

The bacteriostatic role of plant VOCs was studied by Gao, et al. [17] who found emissions of terpenoids, alcohols, aldehydes, organic acids, and esters released by five healthy coniferous species in which α-pinene, β-pinene, 2,(10)-pinene, myrcene and d-limonene represented more than 95 % of total VOC emissions. Increased levels of α-pinene, limonene, nonaldehyde and benzaldehyde also were found in artificially-inoculated wood shaves in the same study

3. Electronic Noses

The EN is an instrument that mimics the human olfactory apparatus to detect VOCs through a series of sensors (sensor array) that provide digital signatures (sensor patterns) of all volatile chemicals present in the aroma bouquet of the sample analyte. It typically consists of a multisensor array, an information-processing unit such as an artificial neural network (ANN), software with digital pattern-recognition algorithms, and reference-library databases. The cross-reactive sensor array is composed of incrementally-different sensors chosen to respond to a wide range of chemical classes and discriminate diverse mixtures of possible analytes. The output from individual sensors are collectively assembled and integrated to produce a distinct digital response pattern. Identification and classification of an analyte mixture is accomplished through recognition of this unique aroma signature of collective sensor responses. A reference library of digital aroma signature patterns for known samples is constructed prior to analysis of unknowns. The ANN is configured through a learning process (neural net training) using pattern recognition algorithms that look for differences between the patterns of all analyte types included in the reference library. This process continues until a previously selected level of discrimination is met. The results are validated and assembled into the reference library to which unknown samples can be compared. Identification of unknowns is based on the distribution of aroma attributes or elements that the analyte pattern has in common with patterns present in databases of the reference library.

In this experiment, we employed three different commercially available ENs.


156

3.1. AromaScan A32S Electronic Nose The AromaScan 32S (Osmetech Inc., Wobum,

MA, USA) is a conducting polymer (CP) EN that contains an organic matrix-coated polymer-type 32-sensor array designed for general use applications with 15 V across sensor paths. The sensor array response to different VOCs was tested previously [8].

Sensors responses are measured as a percentage of electrical resistance changes to current flow in the sensors relative to baseline resistance (%ΔR/Rbase). The sorption of headspace volatiles, composed of specific VOC mixtures, to the CP sensor surfaces induces a change in the electrical resistance to current flow, which is detected and measured to produce the sensor array output. Sensor responses varied with the type of plastic polymer used in the sensor matrix coating, produced by electropolymerization of either polypyrrole, polyaniline or polythiophene derivatives, which have been modified with ring substitutions of different functional groups and with the addition of different types of metal ions to the polymer matrix in order to improve and modulate sensor response.

All measurements were statistically compared using normalized sensor outputs from the sensor array. Conducting polymer analysis methods used with this instrument employ application-specific reference libraries for aroma pattern recognition and neural-net training algorithms.

3.2. Lybranose 2.1 Electronic Nose

Operation of this EN is based on the quartz crystal microbalance (QCM) technology, which can be described as an ultrasensitive sensor capable of measuring small changes in mass on a quartz crystal recorded in real-time. The heart of the QCM is the piezoelectric AT-cut quartz crystal sandwiched between a pair of electrodes. The electrodes are attached to an oscillator. When an AC voltage is applied over the electrodes, the quartz crystal starts to oscillate at its resonance frequency due to the piezoelectric effect. If sample volatiles are evenly deposited onto one or both of the electrodes, the resonant frequency will decrease proportionally to the mass of the adsorbed layer according to the Sauerbrey equation [18].

The LibraNose 2.1 (Technobiochip, Pozzuoli, NA, Italy) sensor array consists of eight 20 MHz AT-cut QCM sensors with a gold surface (Gambetti Kenologia, Binasco, PV, Italy) coated with either metalloporphyrines, deposited by solvent casting, or by polypyrrole polymer) films (Technobiochip patent. No. 04425560.2-2102) deposited by means of Langmuir-Blodgett technology using a KSV 5000 film-deposition device (KSV Instruments, Helsinki, Finland). This process utilizes 0.3 mg/mL polymers dissolved in chloroform and ultrapure, distilled water as a subphase.

3.3. PEN3 Electronic Nose

The PEN3 EN (Airsense Analytics GmbH, Schwerin, Germany) is a very compact instrument (255 × 190 × 92 mm), light-weight (2.1 kg) and portable olfactory system. It consists of an array of 10 different doped metal-oxide semi-conductive (MOS) gas sensors positioned into a very small chamber with a volume of only 1.8 ml.

The instrument operates with filtered, ambient air as a carrier-gas at a flow rate of 10-400 ml min-1, sample-chamber temperature of 0-45°C, and sensor array operating temperature of 200-500°C. The sensing reaction is based on an oxygen exchange between the volatile gas molecules and the metal coating material. Electrons are attracted to the loaded oxygen and result in decreases in sensor conductivity. Instrument sensitivity to various gas analytes ranges from 0.1-5.0 ppm.

4. Main Goals

This research project, a pioneer in the field of plant pathology and urban forestry, is based on following steps starting from basic research to be applied to finalized research solution. In every phase we have formulated a hypothesis derived from a question, with the aim to verify if the instrument could give positive answers.

4.1. Can the EN Discriminate Between Healthy and Inoculated Wood Samples?

The first step of the research was aimed at determining if wood decay fungi emit certain combinations of VOCs that can be detected and recognizable by ENs [19]. 11 wood decay fungi (WDF) strains were selected, cultivated and inoculated on wood chips samples (sapwood) taken from 19 tree species: Fraxinus pennsylvanica Marsh., Liquidambar styraciflua L., Pinus taeda L., Platanus occidentalis L., Populus deltoids Bartr. ex Marshall, Quercus nuttallii Palm., Quercus lyrata Walt., Thuia occidentalis L., Taxodium distichum L. Acer negundo L., A. saccharinum L., Aesculus hippocastanum L., Castanea sativa Mill., Cedrus deodara (D. Don) G. Don fil., Celtis australis L., Platanus x acerifolia Brot., Quercus rubra L., Robinia pseduoacacia L., and Tilia spp.

Species were selected from among the hardwood and conifer species most common in the lower Mississippi Delta and Northern Italy urban and forest environment, where the experiments were conducted.

After 6, 12 and 24 months of incubation under standard conditions we evaluated the discrimination ability of all three selected ENs. Fig. 1 – Fig. 3 show some results of this step.

Fig. 1(a) reports about the ability of Lybranose 2.1 in discrimining healthy and inoculated wood samples of all tree species with all fungal species.


157

(a)

(b)

Fig. 1. Discrimination of volatiles from healthy

and decayed wood block using PCA. Labels are as follows: yellow and green labels indicate volatiles from healthy

controls and red and blue labels indicate volatiles from decayed samples.

Although some zones of the graph show some overlaps between the two types, it is possible to assert that WDF emit volatiles which are clearly discriminable for the instrument. PEN3, differently, but clearly discriminated healthy and inoculated wood samples (Fig. 1(b)).

Running Principal Component Analysis (PCA) on samples belonging to one single tree species inoculated with different fungal strains, it is clear as the EN (Fig. 2) can easily discriminate the different WDF species.

4.2. Can the EN Discriminate Between Healthy and Inoculated Root Samples Under Soil Conditions?

The root system is the most important organ for

initiating plant growth as it is dedicated to the uptake

of water and minerals. In trees, structural roots give mechanical support to the heavy woody structure. The root system is by nature the least known of all tree organs as it is not assessable other than by destructing methods. As far as root decay diagnoses are concerned, there are not commercially available tools capable of assessing and diagnosing decays in the root system.

Fig. 2. Discrimination of volatiles from artificially-inoculated decayed wood samples of Tilia spp. by PCA.

Different color labels indicate different wood decay fungi responsible for decay. Undecayed (control) wood block are

labeled in yellow.

Fig. 3. Discrimination of VOCs from healthy controls (green labels) and artificially-inoculated root chips

after 6 months (red labels) and 12 months (blue labels) from inoculation using PCA.

This phase of the research, aimed at determining if the presence of VOCs emitted by wood decaying fungi or decayed living wood can be detected even under-soil conditions, utilized root tissue form four species of shade trees [7]. Parts of 1 cm healthy roots were sampled from each tree in which roots were prepared and inoculated with four different WDF strains (two strains of Armillaria mellea, one of Ganoderma lucidum and one of Heterobasidion annosum). Inoculated root chips were then incubated under two different kinds of soils (very poor urban soil and professional soil for horticulture) for 12 months at standard laboratory conditions.

12 months controls

6 months


158

Our results show that: 1) PEN 3 EN could easily discriminate between

inoculated and non-inoculated root chips after 12 months of incubation;

2) PEN 3 EN could not discriminate the inoculated samples from the healthy ones after only 6 months from the inoculation. This means that under soil conditions, wood decay fungi take a little more time to develop enough VOCs to be detected from the EN (Fig. 3);

3) Soil type does have an influence on the discrimination capability of the instrument. This is probably due to the fact that professional soil type, which is still rich in microorganisms, emits a strong aroma bouquet.

4.3. Can the EN Detect the Presence of a Decay in the Root System of Standing Trees Directly in the Field ?

All previous steps of this long research were

aimed at evaluating the diagnostic feasibility of EN under the stable and standardized conditions of the laboratory environment. In this stage, the EN was employed directly in the field to detect the presence of VOCs emitted by wood decay fungi attacking the root system. A very important postulate of this research is that the diagnostic system (tool as well as sampling method) should be totally non-invasive for the plant. Wounds caused by sampling, diagnosis or analysis method could eventually be preferred entry points for further pathogenic attack. According to this, a revolutionary sampling method was tested based on detecting decay fungi that emit VOCs which diffuse in soil air macropores. To sample and analyze soil air, a pump system was designed and built as seen in Fig. 4, in which soil air is sucked in by the pump and directly carried to Nalophan bags for e-nose analysis.

Fig. 4. The automatic pump employed in the field to put directly soil air in Nalophan bags.

For this final stage of the research, more than 60 trees planted in the city of Milano, Italy,

belonging to five different species [Acer negundo L., A. negundo ‘Variegatum’, A. pseudoplatanus L., Aesculus hippocastanum L., and Platanus x acerifolia (Aiton) Willd were sampled]. All of these trees were previously assessed via conventional methods for the presence of stem and root decays. Soil air was sampled six times over a period of two years.

Our preliminary results, shown in Fig. 5 – Fig. 7, demonstrate that WDF VOCs can be found in soil macropores, and that their concentration in the zone of the root system is high enough to be detected by EN sensors. Fig. 5 show a linear discriminant analysis (LDA) performed on soil air samples taken about 30 cm from the bole of healthy and decayed trees. The etiologic agent causing decays in sample trees, previously recognized via traditional methods, was also recognized by the EN. Fig. 6 shows the diagnostic feasibility of the PEN 3 EN in discriminating between different WDF species.

Fig. 5. Linear Discriminant Analysis performed on volatiles in soil air samples taken 30 cm from the bole

of healthy (red labels) and decayed (red labels) standing trees.

Fig. 6. PCA performed on volatiles in soil air samples taken 30 cm from the bole of healthy (green labels)

and decayed standing trees. Different colors correspond to main etiologic decay agent: Armillaria spp. (red labels),

Meripilus giganteus (orange), Ganoderma spp. (pink) and Perenniporia ssp. (blue labels).

Soil air or healthy control trees also were used to check if healthy root systems release the same VOCs


159

as those released by trunk sapwood; and if ENs can discriminate between species on the basis of those VOCs. Fig. 7 shows the discrimination feasibility of the PEN 3 EN between different healthy tree species on the basis of VOC analysis in the aroma bouquet released by root systems sampled at about 30 cm from the bole by the use of an air pump.

Fig. 7. PCA performed on volatiles in soil air samples taken 30 cm from the bole of healthy standing trees. Different colors correspond to different species as

following: Metasequoia glyptostroboides (pink labels), Fagus sylvatica (blue), F. sylvatica ‘Pendula’ (red),

Aesculus hippocastanum (green) and Quercus rubra (orange).

7. Conclusions

Tree cultivation in the urban environment requires some agronomic works, which are particularly important and expensive for Public Administrators. Among these, pruning and maintaining tree stability most influence the annual budget of ordinary management. Tree stability assessment, a fundamental duty to prevent sudden tree failures so ensure citizen’s safety, is performed by very skilled personnel who employ commercial instruments and tools which are, in most of cases, invasive and very expensive. Decay assessment of the root system is not currently performed, as there are no commercially available instruments besides ENs capable of these assessments.

Our experimental research started about 10 years ago, was aimed at developing a sampling and analysis methodology to determine the presence of active wood decay and root rots on standing trees, in a rapid and non-invasive way, applicable in all situations and usable by non-skilled operators. Through multiple stages of research, we have demonstrated that three different commercial ENs can discriminate: between different tree species and WDF species by analyzing the VOCs in the aroma bouquet released by healthy (non-inoculated) and inoculated trunk wood chips; between healthy (non-inoculated) and inoculated root chips incubated under two different kinds of soils; between healthy and decayed living and standing trees, between different species of healthy standing trees, and between etiologic agents of diseased standing trees on the

basis of the analysis of the aroma bouquet present in the soil air (macropores) sampled near the tree bole.

The EN system is not fully ready to be employed daily in the field yet, as it is necessary to build application-specific aroma signature databases of healthy tree species at different stages of growth, development and phonologic phase, as well as diseased tree species, decayed by different wood decay fungi species, in order to train ENs to yield immediate answers directly in the field.

Acknowledgements

This multi-year research was funded by Comune di Milano – Sett.re Tecnico Arredo Urbano e Verde, and from Demetra Società Cooperativa Sociale Onlus (Besana in Brianza, MB, Italy). The Authors would like to thank Dr. Daniele Guarino, Charisse Oberle, Luca Maccabelli, Raffaello Brambilla and Lorenzo Presicce for their help in sampling and analyzing data. References [1]. B. Skrbic, M. Snezana, M. Matavulj, Multielement

profiles of soil, road dust, tree bark and wood-rotten fungi collected at various distances from high frequency road in urban area, Ecological Indicators, Vol. 13, 2012, pp. 168-177.

[2]. T. B Randrup, Urban soils as a growing medium for urban trees, in Urban Forestry in the Nordic countries, in Proceedings of a Nordic Workshop on Urban Forestry, T. B. Randrup, K. Nilsson, P. Nilsson, Eds., Danish Forest and Landscape Research Institute, Hoersholm, 1996, p. 59.

[3]. S. D. Day, J. R. Seiler, N. Persaud, A comparison of root growth dynamics of silver maple and flowering dogwood in compacted soil at differing soil water content, Tree Physiology, Vol. 20, Issue 4, 2000, pp. 257-263.

[4]. C. J. Luley, Wood Decay Fungi Common to Urban Living Trees in the Northeast and Central United States, Urban Forestry LLC, Naples, NY, 2005.

[5]. D. Lonsdale, Principles of tree hazard assessment and management, Forest Commission, Farnham, 1999.

[6]. N. P. Mantheny, J. R. Clark, A Photographic Guide to the Evaluation of Hazard Trees in Urban Areas, 2nd ed., International Society of Arboriculture, Savoy, 1994.

[7]. M. Baietto, L. Pozzi, A. D. Wilson, D. Bassi, Evaluation of a protable MOS electronic nose to detect root rots in shade tree species, Computer and Electronics in Agriculture, Vol. 93, 2013, pp. 117-125.

[8]. A. D. Wilson, D. G. Lester, C. S. Oberle, Development of conductive polymer analysis for the rapid detection and identification of phytopathogenic microbes, Phytopathology, Vol. 94, 2004, pp. 419-431.

[9]. A. Korpi, J. P. Kasanen, Y. Alarie, V. M. Kosma, A. L. Pasanen, Sensory irritating potency of some microbial volatile organic compounds (MVOCs) and a mixture of five MVOCs, Archives of Environmental Health, Vol. 54, Issue 5, 1999, pp. 347-352.


160

[10]. N. Magan, P. Evans, Volatiles as an indicator of fungal activity and differentiation between species, and the potential use of electronic nose technology for early detection of grain spoilage, Journal of Stored Products Research, Vol. 36, Issue 4, 2000, pp. 319-340.

[11]. B. J. P. de Lacy Costello, R. J. Ewen, H. E. Gunson, N. M. Ratcliffe, P. T. N. Spencer-Phillips, Sensors for early warning of postharvest spoilage in potato tubers, British Crop Protection Council, Farnham, UK, 2002, pp. 425-432.

[12]. G. Strom, J. West, B. Wessen, U. Palmgren, Quantitative analysis of microbial volatiles in damp Swedish houses, in Health Implications of Fungi in Indoor Environments, R. A. Samson, B. Flannigan, M. E. Flannigan, A. P. Verhoefl, O. C. G. Adan, E. S. Hoekstra, Eds., Elsevier, Amsterdam, The Netherlands, 1994, pp. 291-305.

[13]. R. J. Ewen, P. R. H. Jones, N. M. Ratcliffe, P. T. N. Spencer-Phillips, Identification by gaschromatography-mass spectometry of the volatile organic compounds emitted from the woodrotting fungi Serpula lacrimans and Coniophora puteana, and from Pinus sylvestris timber, Mycological Research, Vol. 108, 2004, Issue 7, pp. 806-814.

[14]. I. Villanueva-Fierro, C. J. Popp, R. S. Marin, Biogenic emissions and ambient concentration of hydrocarbons, carbonyl compounds and organic

acides from ponderosa pine and cottonwood trees at rural and forested sites in Central New Mexico, Atmospheric Environment, Vol. 38, 2004, pp. 249-260.

[15]. J. Faldt, H. Solheim, B. Langstrom, A. K. Borg-Karlson, Influence of fungal infection and wounding on contents and enatiomeric compositions of monoterpens in phloem of Pinus sylvestris, Journal of Chemical Ecology, Vol. 32, Issue 8, 2006, pp. 1779-1795.

[16]. L. J. Anderson, P. C. Harley, R. K. Monson, Reduction of isoprene emissions from live oak with oak wilt, Tree Physiology, Vol. 20, No. 17, 2000, pp. 1199-1203.

[17]. Y. Gao, Y. Jin, H. Li, H. Chen, Volatile organic compounds and their roles in bacteriostatis in five conifer species, Journal of Integrative Plant Biology, Vol. 47, No. 4, 2005, pp. 499-507.

[18]. G. Sauerbrey, Verwendung von Schwingquarzen zur Wägung dünner Schichten und zur Mikrowägung (The use of quartz oscillators for weighing thin layers and for microweighing), Zeitschrift für Physik, Vol. 155, Issue 2, 1959, pp. 206-222.

[19]. M. Baietto, A. D. Wilson, D. Bassi, F. Ferrini, Evaluation of three electronic noses for detecting incipient wood decay, Sensors, Vol. 10, Issue 2, 2010, pp. 1062-1092.

___________________


http://www.sensorsportal.com/HTML/BOOKSTORE/NDIR_Gas_Measurement.htm


161



Motor Bourn Magnetic Noise Filtering for Magnetometers in Micro-Rotary Aerial Vehicles

Nathan J. UNWIN, Adam J. POSTULA

School of Information Technology and Electrical Engineering University of Queensland, Brisbane, Australia

E-mail: [email protected], [email protected]

Received: 31 August 2015 /Accepted: 5 October 2015 /Published: 30 October 2015 Abstract: Avionics systems of micro aerial vehicles (MAV) pose unique problems in system design, sensor signal handling and control. This is evident in micro-rotary aircraft as their whole body rotates with the sensors of the flight control. The precise calculation of attitude and heading from magnetometer readings is complex due to the body rotation. It is made even more difficult by noise induced in the geomagnetic signal by fluctuating magnetic field of the closely positioned motors. Filtering that noise is challenging since the rotation speed of motors and the vehicle can be very close. This paper presents analysis of motor induced noise, based on experimental data of brushless micro motors. A novel time domain filter is proposed, designed, implemented in FPGA hardware, tested and compared to other filters. This filter provides good performance even when the rotational rate of the motor and vehicle are close and traditional frequency domain filters would perform poorly. Copyright © 2015 IFSA Publishing, S. L. Keywords: Magnetic noise, Magnetometer, Rotary body UAV. 1. Introduction

Rotary body aircraft is unique since it is both a

rotary wing and fixed wing aircraft, which produces lift by spinning like a maple seed. The Papin-Rouilly Gyroptère [1] built in 1915 as a manned airplane is the first example of a “monocoptor”, a type of rotary body aircraft. While the Gyroptère did not fly it is the basis for contemporary designs of rotary body unmanned micro-aircrafts. Fig. 1 shows some of the latest developments of such micro-aircrafts in industry and academia [2-5].

The interesting property of the rotary body aircraft is that the core set of sensors of the flight control system, the inertial measurement unit containing magnetometer, is always rotating as it is affixed to the body of the aircraft.

While this is not a problem for the sensors, it is an issue for calculation of the attitude and heading of the vehicle since this rotation must be filtered out of the geomagnetic signal. This is compounded by the relatively high rotation rate of these vehicles of up to 10 Hz [6-7].

As the scale of a monocoptor decreases, the speed at which it rotates needs to increase if efficiency of flight is to be maintained [7-8]. In a fast spinning and small aircraft the on-board magnetometer placed close to the motors is exposed to high level of magnetic noise generated by the motors which rotate with speed close to the spin. Filtering that noise with traditional frequency domain filters is difficult since frequency separation between noise and signal is small, necessitating a complex high order filter, and raising a question if a standard frequency based filter could be effective at all.




162

(a)

(b)

(c)

(d)

Fig. 1. (a) Lockheed-Martin Samarai prototype [2]; (b) Lockeed –Martin patent drawing [3]; (c) University of Maryland aircraft [4];

(d) University of Queensland aircraft [5].

This paper presents an alternative: using a recoded or constructed signal to null the signal generated by the motor.

The most widely used motor for micro aerial vehicles is the BrushLess Direct Current (BLDC)

motor. The brushless DC motor is a permanent magnet synchronous motor designed to be used with a square wave input generated by a DC powered speed controller [9]. The motor is comprised of a permanently magnetised “rotor” that rotates and the electro-magnetic “stator” that remains stationary. This paper focuses on the most popular out-runner motor type where the rotor is positioned around the outside of the stator as shown in Fig. 2.

Fig. 2. 1- BLDC simplified configuration; 2 - Illustration of localised demagnetisation (not to scale).

The out-runner motor has a number of magnets arranged with the poles alternating on the faces of a ring outside the stator, the ring is then connected to a centre shaft that runs inside the stator (stator sandwiched between the rotor and shaft) [10-11].

As the motor rotates, the magnets are presented to different parts of the stator. By energising the windings to pull the magnet towards the winding or inverting the power to winding to push the magnet away, torque is applied to the rotor [12].

The same movement of the magnets generates an alternating field outside the rotor. This field is used by some speed controllers to sense the position of the rotor to determine the optimum way to energise the stator at that instant. This field also is measured by magnetometers as noise superimposed on the geomagnetic measurements.

The strength of the field is dependent on the construction of the magnet, size of the magnet, construction around the magnet and distance to the magnet.

Permanent magnets exhibit a tendency to demagnetise over time. Demagnetisation exhibits relationships with temperature, time and subjected magnetic fields [13-14]. However it has been noted that there is behaviour where regions of the magnet will demagnetise in preference to surrounding regions resulting in poles with non-uniform strength within the pole region.

Localised demagnetisation is of interest for sensing applications as it adds higher frequency components to the signal generated by the rotation of the motor. Frequencies of these components are approximately odd multiples of the number of poles. As each magnet may not deteriorate identically, this frequency may not be an integer value (but as it is a


163

periodic signal, will be a waveform with an even number of poles).

This paper is structured as follows: in Section 2 analysis of the noise generated by motors is provided, in Section 3 we analyse options for filtering and outline the design of time domain inverse filter, in Section 4 filter performance is discussed, in Section 5 design options and limitations are examined, Section 6 details several position sensing methods, in Section 7 ways to make the filter update on a live system are investigated and finally in Section 8 we conclude and outline possible extensions to this work.

2. External Magnetic Field of Permanent Magnet Synchronous Motors

Before a method of correction could be

attempted, the properties of the interference due to the motors needed to be determined. To do this a motor was operated with a moderate load and the resulting magnetic interference recorded by a magnetometer in close proximity. First the effects of the motor were measured at various speeds. During this test it was noted that the noise was fairly constant across the different speeds. The results of measurements are shown in Fig. 3.

Fig. 3. Raw measured magnetic field of rotating motor.

The external magnetic field with the motor running (for the tested motor) is approximately 0.05 gauss. The field generated purely due to permanent magnetic field is also approximately 0.05 gauss for this motor. The above observation indicates that the field measurements can be performed with the motor rotated by an external drive e.g. a stepper motor. Such an arrangement allows for much better control and more precise measurements.

The result of the magnetic field measurement is a periodic waveform presented in Fig. 4. It can be observed that the motor appears to have a higher and lower frequency components.

Closer analysis using the Fast Fourier Transform identifies three dominate frequency components, as shown in Fig. 5.

Fig. 4. Example magnetic field measurement with a rotating motor.

The lowest frequency component f1 corresponds to the speed of the motor. From this it can be established that the motor forms a pair of strong poles, possibly due to imbalances in the magnets. This pair of poles forms the strongest field present in the motor when considering only the permanent magnetic field.

The medium frequency component f7 corresponds to the magnets embedded inside the rotor of the motor: The motor under test was a 14 pole motor, with each poll corresponding to either a north or south orientation. As the motor is rotated the polls will result in a waveform with a number of peaks equal to the number of poles and a frequency equal to half the number of poles.

Fig. 5. FFT analysis of magnetic field of rotating BLDC motor.


164

The high frequency component f35 is due to localised demagnetisation. Owing to the fact that the demagnetisation is not the same for all magnets, the spectrum had a wider distribution. Different motors exhibited different centre frequencies and distributions, but were all around the 35 times rotation rate.

3. Selection and Design of Filters 3.1. Frequency Domain Filters

Frequency domain filters are the current choice for filtering noise for sensors on UAVs.

The low pass filter is almost universally used on all sensors (typically 2nd or 5th order) to remove high frequency noise from the desired signal. Low pass filters perform poorly when the desired signal is close to the noise and unfortunately, that is the case when the vehicle rotational speed is close to the motor speed.

A more sophisticated method to allow the speed of the vehicle to approach the speed of the motor is to use a combination of tracking notch filter and a low pass filter. Assuming that the motor and vehicle rotation speed don’t overlap for long periods, tracking the motor speed, and centring a notch filter on the motor speed may provide a filter of superior performance. A low pass filter would remove high frequency noise outside the maximum vehicle dynamics.

We developed and tested a novel method based on combination of the notch filter tracking the motor and the band pass filter tracking gyroscope measurements of the vehicle. This is based on the observation that if the gyroscope signal is approximately correct then the gyroscope data can be used to estimate the frequency of changes in the geomagnetic field directly related to the motion of the vehicle. We used a notch filter tracking motor speed (4th order Butterworth) and a band pass filter tracking gyroscope measurements (4th order Butterworth).

3.2. Inverse Filter in Time Domain

The principle of inverse filter is that for periodic noise signals, such as generated by motors, a period of noise signal known not to contain the desired signal is recorded and applied as an inverse signal super-positioned with the measured signal in time domain. The expected result should be the desired signal with only residues of the noise components. This approach has been used in magnetic tape playback [15].

In our application, this method has the potential of providing optimal results assuming that the noise is only dependent on the angular position of the motor’s rotor. The noise signal can be divided into position dependent elements that are stored in a look

up table. Each element corresponds to a small arc of the motors motion; as the motor rotates, successive elements in the table are used to provide a correction. As this method requires accurate position of the rotor, an optical encoder is added to the motor. The encoder used for experiments generates a signal twice every 1/800th of a rotation (two edges, spaced apart by 1/1600th of a rotation). This signal is used to estimate the position of the rotor and to step to the next element in the look up table.

The design shown in Fig. 6 is the core of the filter. The index is the current lookup location, Din is the sensor input and Dout is the corrected output. The filter was implemented on an FPGA, and is optimised to make use of the available resources. The BRAM (Block RAM) acts as the look up table and is loaded with the correction values.

Fig. 6. Inverse filter structure.

Half wave and Quarter wave symmetry were applied to reduce the BRAM usage. Symmetry requires that the index value needs to be folded into a subset of ranges and an offset is needed to shift the waveform. This makes the control part of the design more complex and taking into account the looming accuracy issue in case of asymmetry of the waveform, as shown in Fig. 7, makes the approach less attractive.

Fig. 7. Waveform of half wave symmetric table.


165

Table 1 shows that the logic size doubles when symmetry is used, however as the filter is a very lightdesign, the utilisation of FPGA resources is very small.

Table 1. FPGA implementation resources.

Used Spartan 3e500 total

available Logic Slices

(no symmetry) 33 4656

Logic Slices (symmetry)

61 4656

Block Ram (1024x16bit)

1 20

The throughput of the filter is very high, requiring only 5 clock cycles to perform a correction. The correction time is as follows:

• 1 cycle to load in the sensor data and index address.

• 1 cycle to convert the index address to LUT address.

• 1 cycle to fetch the correction from the LUT. • 1 cycle to add the correction and sensor data. • 1 cycle to output the result. This filter is comparable to the simplest frequency

domain filter (2 point window filter) in both speed and implementation size. Compared to the band stop/pass filters the implementation size is significantly smaller as it requires at most 3 addition operations rather than iterative addition and multiplication operations. Given a moderate 50 MHz clock speed the latency is 100 nanoseconds, which is more than adequate for this application, where filtering in the range of kilohertz is required.

4. Filter Performance

For all filter designs except the look up table approach, the frequency separation between signal and noise is important. The performance was compared for several conditions:

1. “Traditional” separation that could be expected for most vehicles; vehicle rotation is significantly slower than the motor speed (Typically Fn/Fs > 10).

2. Narrow separation where the vehicle rotation is still slower than the motor, but they are with in an order of magnitude. (10 > Fn/Fs > 1).

3. Reversed separation where the motor speed is slower than the vehicle rotational velocity. This would only likely be seen if the motor power was reduced to slow the vehicle (Fn/Fs < 1).

The results in Fig. 8 show that the low pass filters perform poorly when vehicle and motor rotation speeds are close.

Both the band pass and notch filters worked well. Their difference in performance, as shown in Fig. 8,

is explained by the fact that the notch filter handles motor rotations which have some variance while the band pass filters vehicle rotations which are much more stable due to larger inertia. Their main disadvantage is much larger cost of implementation than the inverse filter.

Fig. 8. Results of filters at different noise to signal frequency factors.

The look up table based inverse filter has showed good performance and the exceptional performance was achieved with the addition of a low pass filter for filtering the f35 component.

Another advantage of the look up table based method is constant latency and linear phase delay, which would be achievable with standard FIR filter but at much higher implementation cost.

5. Inverse Filter Considerations As Fig. 5 shows, the spectrum of magnetic noise

from the motor contains high frequency component f35. In a straightforward approach the look up table of inverse filter would need to have sufficient number of samples/elements to cover that spectrum. A more efficient method is to augment inverse filter with a simple 2nd order low pass filter (LPF in Fig. 8) to filter out that part of the spectrum.

Results presented in Fig. 8 were obtained with the highest resolution implemented for the inverse filter, however depending on the platform it may be desirable to decrease the number of elements in the look up table. Tables with 800, 400, 200, 100 and 50 element were tested to determine the impact.

The results in Fig. 9 show that for 800 to 200 look up elements the deterioration of performance is limited, while 100 and 50 element table causes significant decrease in effectiveness. The x axis is the rotor position error measured in the number of elements miscounted by the rotor position encoder. The larger the number of samples, the lesser is the


166

impact of the position error since the angle which each sample represents is smaller. FPGA BRAMs make implementation of look up tables very efficient, even if large number of elements is required.

Fig. 9. Heading error due to rotor encoder miscounts.

Our experiments with a number of motors show significant variations in the rotating magnetic field patterns, even for the same type of motor. This is caused by manufacturing imperfections, demagnetisation and/or wear and tear. The magnets vary in strength (or over time loose strength at differing rates) resulting in the waveform being compressed for a portion of cycle and expanded for another portion, and the individual peaks having varying magnitude. The localised demagnetisation effect adds a high frequency component that does not have an instantaneously constant frequency.

The above phenomena mean that the look up table must be prepared for a specific motor and updated with the aging of the motor. This can be done automatically with help of an additional calibration module in the aircraft control system that

is activated in the start when the vehicle is stationary [19].

6. Motor Rotation Positioning

The 1600 count optical encoder, used in the rig for testing the motors, provided test instrument accuracy and performance, but unfortunately it lacks compactness, robustness and low cost factor required in the micro rotational vehicle. Several alternative sources of motor rotation position were investigated to identify a low cost alternative:

• Position extrapolation from a low resolution encoder.

• Brushless motor speed controller’s estimated position.

• Magnetic or Capacitive encoder.

6.1. Position Sensing Using Extrapolation from a Low Resolution Encoder

With an assumption that the speed of the motor

does not significantly change or oscillate from one rotation to the next, it should be possible to estimate the rotational position by the time passed since the last encoder signal.

As shown in Fig. 10, in practice there is some variance in speed and oscillations after a speed change, but it will be dependent on the speed controller and motor power compared to the rotating inertia so would need to be assessed case by case. For Fig. 10 one combination shows very good damping and position extrapolation would work almost immediately after a change, while the other combination has some oscillations that would need to decay before accurate position estimates could be made (in practice this will be on the order of a second).

Fig. 10. Speed of the same motor controlled by two different ESC [16].

6.2. Position Sensing Using the Speed Controller’s Internal Estimate

Brushless motor Electronic Speed Controller

(ESC) must maintain an estimate of the motors

current rotational angle to correctly energise the winding of the motor [17]. The ESC identifies three key points in one rotation cycle based on Back ElectroMotive Force (BEMF), where the un-energised winding’s voltage crosses zero. Using this


167

estimate for addressing the filter look up table is similar to using the position derived from the low-resolution encoder. The main advantage is that no additional sensors are required since the information is already available from ESC and estimation of angles between the key points is straightforward, assuming stable speed during one turn of the motor.

Several problems were identified that would require solving before this method could be used:

The position is based on back EMF and might not be a stable point for all speeds and throttles.

The position estimate quality would be dependent on the ESC’s internal update rate.

The location is not unique; it is only the position in the current cycle (for the common 14 magnet poll motor, there are 7 ESC cycles to one rotation).

Following Pt 3, if the motor is unpowered or let freewheel, the location would be lost and un-trackable as there will be no measurable BEMF.

While points 1 & 2 may be resolved with careful implementation, points 3 and 4 are likely to not be easy to resolve without some form of external sensor or some form of online filter updating where the loss of position would only be a short term issue while the filter updates.

6.3. Position Sensing Using a Magnetic or Capacitive Encoder

These encoders are essentially a drop in

replacement for an optical encoder but without a large and fragile disc; magnetic or capacitive encoder offers the possibility of a smaller and more robust package. Some encoders such as the CUI inc ATM103 [18] provide 2048 encoder counts per rotation at a maximum velocity of 7500 RPM or 15000 RPM with a reduced count.

It is expected that results would be identical with any encoder given the same counts per rotation and maximum velocity limit so choice would be down to weight, size and robustness. As the capacitive sensors are sealed and can be mounted behind the motor, they might have been a better choice than the optical sensor used.

7. Online Filter Tuning Adjusting the filter to match the motor while in

operation offers the possibility of minimising the motor magnetic noise over longer periods of operation without the need for manual intervention.

Several methods for automatic adjustment of the filter were investigated and can be further developed:

• Baseline offline correction • Updating correction on power up of vehicle • Attempt to minimise difference of magnitude from local geomagnetic field. • Fitting measured signal to a sinusoid.

7.1. Offline Correction

As detailed in Section III, offline correction is performed by collecting magnetometer readings for each encoder location and removing DC bias from the AC waveform formed by the alternating magnets of the motor (the DC bias contains the geomagnetic field signal). This method assumes that the measurement platform is stationary and separated from other time varying magnetic fields so that only the motor based magnetic field varies.

This method is not suitable for real time correction as the method used to remove dc bias relies on the geomagnetic field (and hard/soft iron error sources) to remain stationary: as the method will take any time varying magnetic field as an error to correct, movement will result in the geomagnetic field being interpreted as an error source.

Online methods will need to account for the possibility that the vehicle might be moving. 7.2. Power-up Correction

To allow for correction without user intervention a power-on measure and correct routine was developed. The start-up routine automates the offline method so that when the system is powered on and relatively stationary, the motor’s magnetic field can be measured and an up to date correction generated.

Fig. 11. Modified filter to allow online updating.

This method relies on the platform to be stationary (strictly speaking, only rotation needs to be as low as possible) and the only time variant signal being due to the rotation of the motors. The process for generating the correction is as follows (Fig. 12).

The first step is to have the motor under test (remember there might be multiple motors) accelerate to a stable speed. It is important that this speed be sufficient for the ESC to maintain a constant speed but not too fast that the vehicle is likely to react.


168

Fig. 12. Software flow for power-up correction.

Once the motor has reached the stable speed, measurements are made for each LUT index. This may be difficult on slower processors, so the LUT can be used a sample and hold buffer if the write enable to port one is enabled. Ideally several readings for each index should be made and an average taken.

Removal of the DC bias can be performed a number of ways, but the simplest method is to sum all the LUT index values and subtract the result from all elements.

Updating the LUT was done through the second port of the Block RAM in the FPGA. This simplified the updating process and reduced the logic required. A single port RAM could have been used at the expense of some logic being needed to swap between rotor index and software index.

The performance of the Power-up correction was identical to the offline method (800 elements) as they are identical in operation. While only the full LUT method without symmetry was attempted, it is expected that there would be no issue with using the methods with symmetry.

7.3. Correction Based on Geomagnetic Field

Correction based on the magnitude of the measured magnetic field compared to the mapped local geomagnetic field (geomagnetic field either measured at a site or taken from a geomagnetic field map) was attempted. The idea was to take the difference between the measured field and the mapped field for each motor encoder index and apply this difference as a correction.

Rather than taking the full error as the new correction, a proportional change where the error was scaled and added to the existing correction value was used to attempt to prevent movement (and thus changing magnetic fields) from significantly changing the correction. This should results in the filter being driven over time towards an error minimum. In practice, this method only worked away from any external sources of magnetic field and any deviation away from the geomagnetic field would cause the correction to deviate.

7.4. Sinusoid Fitting Correction

A promising method for online correction is curve fitting with sinusoids. The correction is produced as a superposition of sinusoids with frequencies corresponding to the motor induced signal. An

approach is to perform an FFT, remove all frequency components away from the motor based frequencies and take the IFFT as the new correction.

As the motors tend to change slowly over time, a solution would not need to be calculated under a strict time constraint and could be completed over a period of time. In a vehicle with some form of multitasking, this may prove to be a good approach as the calculation can be done in otherwise idle time.

8. Conclusions and Future Research

We analysed and experimented with the motor induced noise in magnetometer measurements on board micro body rotate aircraft. Methods of filtering that noise were investigated and properties of various filters assessed for this application. An inverse filter in time domain, based on look up table principle, has been designed on FPGA. The filter was tested and proven to have superior performance in filtering signals and noise of very close frequencies. This filter demonstrates a small implementation cost while offering speed either matching or exceeding the performance of optimised frequency domain based filters.

The research presented in this paper focused on steady state operation of the motor. This was justified as the motor speed of a rotary body vehicle does not vary significantly during operation. However for some applications where the motor speed need to change rapidly, further research is needed to assess applicability of our filtering method.

A possible application of our research is also for assessing the health of a motor by monitoring the spectrum of motor induced noise in magnetometer readings. It has been demonstrated that as a motor ages or is subjected to high loads, the localised demagnetisation increases, changing its noise spectrum. This relationship could be possibly used to estimate the health of the motor.

References [1]. W. Pearce, Papin-RouillyGyropter (Gyropter),

[Online]. Available from: https://oldmachinepress. wordpress.com/2012/09/06/papin-rouilly-gyroptere-gyropter/ [Retrieved 03/06/2015]

[2]. K. Fregene, C. L. Bolden, Dynamics and Control of a Biomimetic Single-Wing Nano Air Vehicle, in Proceedings of the American Control Conference (ACC), Baltimore, MD, USA, 30 June -2 July 2010, pp. 51-56.

[3]. S. M. Jameson, B. P. Boesch, E. H. Allen, Active maple seed flyer, United States of America Patent US7766274 B1, Lockheed-Martin, 3 August 2010.

[4]. E. R. Ulrich, D. J. Pines, J. S. Humbert, From Falling to Flying: The Path to Powered Flight of a Robotic Samara Nano Air Vehicle, Bioinspiration and Biomimetics, Vol. 5, No. 4, 2010, pp. 3-16.

[5]. Pounds P., Singh S., Samara: Biologically inspired self-deploying sensor networks, IEEE Potentials, Vol. 34, Issue 2, 2015, pp. 10-14.


169

[6]. C. Hockley, B. Butka, The SamarEye: A Biologically Inspired Autonomous Vehicle, in Proceedings of the 29th IEEE Digital Avionics Systems Conference (DASC’10), Salt Lake City, October 2010, pp. 5.C.1-1 - 5.C.1-9.

[7]. H. Youngren, S. Jameson, B. Satterfield, Design of the SAMARAI Monowing Rotorcraft Nano Air Vehicle, [Online]. Available from: http://www.atl.lmco.com/papers/1628.pdf [Retrieved 14/04/2015].

[8]. A. R. S. Bramwell, G. Done, D. Balmford, Bramwell's Helicopter Dynamics, Butterworth-Heinemann, 2001.

[9]. Microsemi User Guide, Field Oriented Control of Permanent Magnet Synchronous Motors, [Online]. Available from: www.microsemi.com [Retrieved 03/06/2015].

[10]. G. H. Jang, J. H. Chang, D. P. Hong, K. S. Kim, Finite-Element Analysis of an Electromechanical Field of a BLDC Motor Considering Speed Control and Mechanical Flexibility, IEEE Transactions on Magnetics, Vol. 38, No. 2, 2002, pp. 945-948.

[11]. N. Bianchi, S. Bolofa, F. Luise, Analysis and design of a brushless motor for high speed operation, in Proceedings of the Electric Machines and Drives Conference, Madison, Wisconsin, 2003, pp. 44-51.

[12]. J. Rais, M. P. Donsión, Permanent Magnet Synchronous Motors (PMSM). Parameters influence on the synchronization process of a PMSM, in Proceedings of the International Conference Renewable Energies and Power Quality (ICREPQ'08), Santander, March 2008, pp. 409-413.

[13]. M. Ooshima, S. Miyazawa, A. Chiba, F. Nakamura, T. Fukao, A Rotor Design of a Permanent Magnet-Type Bearingless Motor Considering Demagnetization, in Proceedings of the Power Conversion Conference, Naqaoka, 1997, pp. 655-660.

[14]. S. Touati, R. Ibtiouen, O. Touhami, A. Djerdir, Experimental investigation and optimisation of permanent magnet motor based on coupling boundry element method with permeances network, Progress in Electromagnetics Research, Vol. 111, 2011, pp. 71-90.

[15]. R. W. Kruppa, Method, and apparatus and article of manufacture for filtering periodic noise from a magnetic read head, United States of America Patent 5,887,075, 23 March 1999.

[16]. B. Tefay, FPGA-Based Electronic Speed Control for Brushless DC Motors, Bachelor of Engineering Thesis, School of Information Technology and Electrical Engineering, University of Queensland, 2011.

[17]. D. Torres, AN1160: Sensorless BLDC Control with Back-EMF Filtering, 2008. [Online]. Available: http://ww1.microchip.com/downloads/en/AppNotes/01160A.pdf [Accessed 12 September 2015].

[18]. CUI Inc, 14 September 2015. [Online]. Available: http://www.cui.com/product/resource/amt10-v.pdf

[19]. N. Unwin, A. Postula, Filtering of Magnetic Noise Induced in Magnetometers by Motors of Micro-Rotary Aerial Vehicle, in Proceedings of the 8th International Conference on Advances in Circuits, Electronics and Micro-electronics (CENICS’15), Venice, Italy, 23-28 August, 2015, pp. 17-22.

___________________


http://www.sensorsportal.com/HTML/BOOKSTORE/Magnetic_sensors.htm


170



Reflection from Disordered Silver Nanoparticles on Multilayer Substrate

Victor OVCHINNIKOV

Department of Aalto Nanofab, School of Electrical Engineering, Aalto University, Espoo, Finland Tel.: +358503523319, fax: +35894123014


Received: 31 August 2015 /Accepted: 5 October 2015 /Published: 30 October 2015 Abstract: Reflection from disordered silver nanoparticles on dielectric and multilayer substrates is studied. Optical characterization of nanoparticles on transparent substrates is performed and it is shown that positions of localized surface plasmon resonances in reflection spectra cannot be ascribed to spectrum peaks like it is done in case of extinction spectra. To clarify positions of plasmon resonances in total reflection spectra optical properties of samples are analyzed in relation to their design and morphology. The disordered silver nanoparticles arranged in one- or two layers on substrates from quartz, silicon and oxidized silicon are studied. Extinction and reflection from as prepared, plasma etched and SiO2 covered samples are compared. It is concluded that coupling between nanoparticles, phase shift of scattered light and reflection from film interfaces lead to additional features in reflection spectra in comparison with extinction ones. Recommendations for identification of plasmon resonances in reflection spectra of disordered nanoparticles on multilayer substrates are proposed. Copyright © 2015 IFSA Publishing, S. L. Keywords: Ag nanoparticle, Plasmon resonance, Reflectance, Optical cavity, Irregular array. 1. Introduction

Plasmonic nanostructures are widely used in sensors, metamaterials, solar cells, photonics and spectroscopy [1-5]. Effective application of these structures is based on localized surface plasmon resonance (LSPR) demonstrated in ultraviolet, visible and infrared. The wavelength of LSPR depends on material and geometry of nanostructures, their interaction with each other and electromagnetic properties of environment, including substrate and capping layers. Despite on near field nature, LSPR can be observed in far field optical measurements due to variation in optical properties of the studied structures. Extinction is the most popular method of LSPR registration due to its simple implementation and straightforward interpretation, i.e., maximum position

and width of extinction peak correspond LSPR wavelength and damping, respectively. However, extinction can be measured only for non-opaque structures, e.g., for nanostructures on transparent substrates or for plasmonic colloidal particles. Furthermore, extinction spectra are not effective for overlapped peaks, when spectral deconvolution is not obvious.

LSPRs on opaque substrates can be visualized by different kinds of reflection and scattering measurements. However, peaks and troughs of specular reflectance do not correspond to LSPRs and spectrum analysis becomes problematic. Scattering measurements require special arrangement of light illumination (dark field) to separate low scattering signal from strong reflection background. It limits range of measured samples by plasmonic




171

nanostructures on substrate surface and, for example, plasmonic nanoparticles inside of dielectric matrix cannot be analyzed. In case of correlated scattering centers, i.e., when array of coupled nanostructures is analyzed, correspondence between scattered peaks and LSPRs is broken and LSPRs should be observed in specular reflectance. Additionally, scattering results are obtained in arbitrary units and cannot be used for comparison of different experiments. This happens, due to the problems with measurement of scattered reference spectrum for calibration procedure. In contrast, reflection reference spectrum can be easily obtained for any material. Combination of total reflectance and diffuse reflectance is especially useful for analyzing plasmon structures, because the last one provides measurement of scattering in absolute values.

In this paper, we continue our study of reflection from irregular arrays of nanoparticles [6] and propose to use total reflectance for identification of LSPR on opaque and transparent substrates. It is demonstrated that wavelength position of LSPR correlates with peak and trough of reflectance in a clear way. Furthermore, overlapped peaks manifest themselves separately in reflectance spectra and can be easily distinguished.

This paper is organized in a following way. In the subsequent Section 2, the details of sample preparation and the measurement procedures are presented. In Section 3, the results of the work are demonstrated for quartz and silicon substrates by scanning electron microscope (SEM) images as well as reflection and extinction spectra of the fabricated samples. The effect of additional dielectric layers on reflectance of silver nanoparticles is discussed in Section 4. Specific features appearing in UV part of reflection spectra are considered in Section 5. In Section 6, the conclusions are drawn.

2. Method

Quartz or crystalline Si wafers (4” in diameter, 0.5-mm-thick) were used as substrates. An Al2O3 layer was grown on the substrate by atomic layer deposition (ALD) and a SiO2 layer was created by thermal oxidation of the Si wafer. Silver layers with a thickness of 15 nm were deposited by electron-beam evaporation with deposition rate 0.5 nm/s. Nanoparticle arrays were fabricated by ion beam mixing (IBM) or annealing of silver films. In case of IBM, Ag films were irradiated by 400 keV Ar ions at normal incidence and at low (1×1016 Ar/cm-2) or high (2×1016 Ar/cm-2) ion fluences to produce the nanoparticles as reported elsewhere [7]. One sample was processed by IBM with Xe ions at dose 6×1015 Xe/cm-2. In the case of annealing, silver films were heated at 350 ºC during 10 minutes. Annealing was done in a diffusion furnace in nitrogen ambient. Further details about samples and processing can be found elsewhere [8, 9]. To cover the nanoparticles with a SiO2 layer we used a plasma enhanced chemical vapor deposition (PECVD) technique. Metal deposition and ion irradiation were performed at room

temperature. To avoid silver oxidation, ALD and PECVD processes were run at reduced temperatures 200 ºC and 170 ºC, respectively. To examine the nanoparticle formation in the structures created, the images of the samples were taken with a Zeiss Supra 40 field emission scanning electron microscope. Three such images, depicting effect of IBM dose and mixing ions are shown in Fig. 1(a, b, c). One more image of the sample prepared by annealing of silver film is presented in Fig. 1 (d). The details of nanoparticle size distribution and sample surface morphology can be found elsewhere [7, 10].

Fig. 1. Plan SEM images of (a) low and (b) high dose Ar IBM samples, (c) Xe IBM sample and (d) annealed sample.

Scale bar is 200 nm.

Optical extinction and reflection spectra were measured with a Perkin Elmer Lambda 950 UV-VIS spectrometer in the range from 250 to 850 nm. Reflection spectra at the angle of light incidence 8º were obtained by using an integrating-sphere detector incorporated in the spectrometer. Either total reflectance or diffuse reflectance only can be measured by placing spectralon plate at the specular reflectance angle or removing it, respectively.

3. Ag Nanoparticles on a Dielectric Substrate

In this section, we study spectra of silver nanostructures on quartz and silicon substrates without additional sublayers. In subsections 3.1 and 3.2, Ag nanoparticles on quartz substrates are discussed, while the subsection 3.3 is devoted to Ag nanoparticles on silicon.

3.1. Spectra of Ag Nanoparticles on Quartz Substrate

Silver nanostructures on a transparent substrate without any additional layers between nanostructures and the substrate are studied in this subsection. It simplifies spectrum analysis due to excluding from consideration interference effects. In Fig. 2 (a, b) extinction, reflection and scattering of silver nanoislands on quartz substrate are demonstrated for

a)

b)

c)

d)


172

high and low dose of IBM, respectively. The corresponding SEM images of the samples are given in Fig. 1 (a, b). In the spectra, there are distinctly visible two areas: right one (wavelength more than 400 nm) with broad and intense peak in visible (VIS) range and left one (wavelength is less than 400 nm) with weak peak in ultraviolet (UV) range. Further, we call these parts as VIS and UV, respectively. The high amplitude peak is usually attributed to dipolar LSPR, whereas the low amplitude one is connected with quadrupolar LSPR [3, 10, 11]. Theoretically, LSPR exhibits itself at the same wavelength in extinction and scattering [2]. However, it is valid only for isolated nanoparticles without size variation. In Fig. 2 (a, b) we observe difference in peak positions for extinction and diffuse reflection, while coinciding for extinction and total reflection peaks. Standard explanation of the observed difference is the size variation of plasmon nanoparticles. Scattering cross-section is higher for larger nanoparticles possessing lower frequency LSPR, while extinction cross-section is higher for smaller nanoparticles having LSPR at higher frequency. As a result, extinction and scattering peaks are separated. The same argument is used for explaining an increased full width at a half maximum (FWHM) of peaks in comparison with calculated ones [2]. Peak asymmetry is usually explained by shape deviation of nanoparticles from sphere to ellipsoid. It results in splitting of one LSPR in two separate resonances (redshifted and blueshifted), which can lead to observable shape of dipolar peak. In Fig. 2 (a, b) the low dose sample demonstrates redshift of scattering in comparison with the high dose sample, because with increasing of IBM dose particle shape variation diminishes and particle size distribution converges to smaller size.

UV resonance manifestation is usually attributed to valley near 360 nm in total reflection, as well as to peak at 350 nm in extinction [7, 10, 12] and is ascribed to quadrupolar resonance. There is also a peak at 330 nm clearly visible in total reflection of low dose sample (Fig. 2 (b)). As a whole, UV features are more intense in low dose sample than in high dose one, but dipolar peak intensity is practically the same in both samples.

According to Fig. 2, intensity of diffuse reflection is 20 times less than intensity of total reflection. Therefore, we can consider total reflection as specular one and assume that samples scatter most of radiation in direction of specular reflection. It is only possible, if all radiating points, i.e., silver nanoparticles work in phase and have similar radiation patterns. If we consider our samples as diffracting gratings, then specular reflection is possible at the zero-order grating condition on the period Λ, which is expressed as [13]

< , (1)

where λ is the wavelength, θ is the incident angle and n is the refractive index of ambient. For θ = 8º and n = 1 the inequality (1) is simplified to Λ < λ. This

condition is fulfilled for all wavelengths in our experiments and explains specular reflection of the arrays despite of scattering of any individual nanoparticle.

(a)

(b)

(c)

(d)

Fig. 2. Spectra of (a) high and (b) low dose Ar mixed samples, (c) RIE treated sample and (d) SiO2

capped sample.

0

0,2

0,4

0,6

0,8

1

0

0,2

0,4

0,6

200 300 400 500 600 700 800

Opt

ical

den

sity

Refle

ctan

ce

TotalDiffuse, x10Extinction

0

0,2

0,4

0,6

0,8

1

0

0,2

0,4

0,6

200 300 400 500 600 700 800

Opt

ical

den

sity

Refle

ctan

ce

TotalDiffuse, x10ExtinctionAbsorption

0

0,2

0,4

0,6

0,8

1

0

0,2

0,4

0,6

200 300 400 500 600 700 800

Opt

ical

den

sity

Refle

ctan

ce


0

0,2

0,4

0,6

0,8

1

0

0,2

0,4

0,6

200 300 400 500 600 700 800

Opt

ical

den

sity

Refle

ctan

ce

Wavelength (nm)



173

The low dose sample was additionally treated by reactive ion etching (RIE) as reported elsewhere [5, 14, 15]. As a result, oxide between Ag nanoislands was removed and SiO2 pillars with a height of 50 nm were fabricated. Ag nanoparticles were left at the top of pillars. The purpose of experiment was to change the dielectric environment and to reduce possible coupling between nanoparticles. The obtained spectra of pillar sample are shown in Fig. 2 (c). Additionally, in Fig. 2 (b, c) is also shown absorptance

RTA −−= 1 , (2)

where T and R are the transmittance and reflectance, respectively. The dipolar extinction, absorption and scattering peaks were blueshifted on 30 - 50 nm, due to replacing SiO2 (ε=2.1 at 633 nm) between nanoparticles by air (ε=1). However, the peak of total reflection was left at the same position. Intensity of extinction and total reflection decreased 1.5 and 3 times, respectively, but intensity of absorption and scattering increased 2 and 2.5 times, respectively. The UV peak of absorptance (peak and valley in reflectance) was replaced by shoulder in absorptance and broad depression in reflectance.

One more variation of dielectric environment was done by covering (capping) of Ag nanoparticles by oxide layer. It was realized by IBM of silver layer covered by 12 nm of SiO2, which resulted in Ag nanoparticles embedded inside of SiO2 matrix. Fig. 2 (d) shows spectra of Ag nanoparticles capped by SiO2. Intensities of VIS extinction and total reflection became higher than in uncapped sample. Additionally, VIS peaks are much wider, due to increased extinction and reflection at long wavelengths. As in case of RIE treatment, the sharp UV valley in absorptance disappeared. The quadrupolar peak of absorptance and dipolar peak of total reflectance were redshifted on 30 nm and 20 nm, respectively, due to increasing of ambient ε (SiO2 from top and bottom of particles).

Comparison of spectra in Fig. 2, leads to conclusion that peaks of extinction and diffuse are connected with average positions of LSPRs. However, these positions are ascribed to LSPRs of different nanoparticles (small and large, respectively) and the difference exceeds 50 nm. At the same time, behavior of total reflection is still unclear and several moments should be considered in details. First of all, it concerns invariability of total reflection peak in Fig. 2 (a, b, c), despite of changing of dielectric environment. According to LSPR theory [2] the peak must be blueshifted in RIE treated sample due to decreasing of effective dielectric constant ε. Then, it is unclear why dipolar and quadrupoalr LSPRs are observed in different way in total reflection, i.e., as peak and valley, respectively. And at last, it is required to consider the difference in peak positions of absorptance and total reflection of the same sample, which varies from zero (Fig. 2 (a, b)) to 30 nm (Fig. 2 (c)).

3.2. Analysis of Obtained Results In disordered array of nanoparticles LSPR happens

only in part of nanoparticles at any wavelength. Other particles reflect in nonresonant mode like a continuous film consisting from mixture of silver and air. This reflection is strong, creates interference and should be distinguished from LSPRs in reflection spectra. In case of absorption measurements, background from nonresonant particles is weak and does not affect on resonance peak position. The identification of resonance features in total reflection, can be done by comparison of obtained spectrum with nonresonant one, e.g., spectrum of silver film. In Fig. 3 (a) reflectance of low dose IBM sample is given together with reflectance of 8 nm thick silver film on glass. The dipolar resonance is responsible for increasing of reflectance in the range of 380 - 480 nm and decreasing of reflectance for wavelengths longer than 480 nm. In turn, quadrupolar resonance is connected with formation of UV valley at 360 nm. However, LSPR scattering from nanoparticles in isotropic medium cannot explain decreasing of reflectance, because total reflection is sum of contributions from all nanoparticles and cannot decrease. Therefore, possible reflection from quartz substrate leading to interference and intensity reduction must be taken into account.

(a)

(b)

Fig. 3. (a) Comparison of reflectance from nanoparticles

and silver film. (b) Optical properties of Ag/SiO2 interface.

0

0,2

0,4

0,6

200 300 400 500 600 700 800

Refle

ctan

ce

Particles

Film

0

0,5

1

40

90

140

200 400 600 800

Refle

ctan

ce

Phas

e (d

eg)

Wavelength (nm)

-ΔφR


174

Interference is inseparably linked with phase change in the system. In case of plasmon particle on surface, phase variation happens due to LSPR and reflection from surface. Nanoparticle radiates light with phase shift Δφ appearing between incident and emitted radiation [1] ∆ = , (3)

where β is the damping constant, ω0 is the plasmon resonance frequency and ω is the frequency of incident wave. According to (3), Δφ is changed from 0º at low frequency to 180º at high frequency and is equal 90º at the resonance frequency.

Radiation emitted by nanoparticle is reflected by substrate with variable intensity and phase, which depend on nanoparticle and substrate properties. Exact reflection spectra can be obtained with the help of simulation, but rough estimation can be done by replacing of Ag nanoparticle by Ag film. For illustration in Fig. 3 (b) is shown wavelength variation of reflectance intensity and phase for Ag/SiO2 interface (incident side is Ag). The results are obtained using expressions from [16] and optical parameters from [17]. High reflectance and broad phase variation facilitate appearing of interference. Reflected light propagates back to nanoparticle and interact with its electromagnetic field. As a result, substrate and a nanoparticle create an optical cavity in which the nanoparticle is one of the mirrors. Length of the cavity is equal to nanoparticle height (around 45 nm for IBM samples [7]).

The particle emits light during plasmon excitation with phase shift Δφpl according to (3). After that total phase shift of optical cavity is Δφpl + Δφrefl, where Δφrefl is the phase shift upon reflection at the substrate. For example, at the wavelength 360 nm (quadrupolar valley in Fig. 2 (b)) reflectance phase is close to 90º (Fig. 3(b)), what provides conditions for destructive interference if Δφpl = 90º, i.e., at resonance frequency of LSPR. As a result, exciting electromagnetic field near plasmon nanoparticle is reduced, which leads to decreasing of nanoparticle radiation at quadrupolar LSPR and appearing of quadrupolar valley.

Importance of reflecting plane for formation of total reflection is illustrated by Fig. 2 (c, d). In both cases sharp quadrupolar valley at 360 nm is not observed, because back reflection plane disappears. After RIE treatment common supporting plane of Ag nanoparticles is replaced by individual pillars, what leads to diminishing of specular reflection and increasing of nonresonant scattering. In the sample with capping SiO2 layer the reflection plane disappears at all, because substrate and capping layer have the same refractive index. In both cases the quadrupolar resonance is observed only in absorptance as 350 nm shoulder (Fig. 2 (c)) and 380 nm peak (Fig. 2(d)), due to eliminating of optical cavity. Redshift of quadrupolar peak at 380 nm after capping by SiO2 layer is explained by increasing of ε.

The optical cavity modifies also dipolar peak of total reflection (Fig. 3a). For wavelengths longer than 500 nm reflectance decreases due to destructive interference, for shorter wavelengths it increases due to constructive interference in the optical cavity. However, dipolar decreasing of reflectance is less significant than quadrupolar one, because destructive interference happens not at resonance wavelength (Δφpl < 90º). In the sample with capping layer dipolar destructive interference is not possible, what leads to increasing of reflectance in the red part of spectrum (Fig. 2 (d)).

For wavelengths shorter than 430 nm steep decreasing of Δφ is observed for Ag/SiO2 interface (Fig. 3b). Together with drop of silver extinction coefficient, it leads to increasing of destructive interference and decreasing of reflectance in nonresonant mode. Therefore, the left slope of total reflection peak is defined by optical properties of air/Ag and Ag/ SiO2 interfaces. That is why this part of spectra looks similar for all samples in Fig. 2. And that is why the peak position of total reflection is not sensitive to RIE treatment. However, the peak intensity decreases, because most of LSPR happens at wavelengths shorter than 430 nm.

The analysis of samples on quartz substrate demonstrates that total reflection must be carefully used for LSPR identification in disordered arrays of nanoparticles. Dipolar resonance corresponds Δφpl = 90º and can be found between destructive and constructive interference areas, i.e., on the right slope of total reflection peak. In case of known scattering, the position of dipolar LSPR can be considered between peaks of total and diffuse reflection. Quadrupolar LSPR position does not coincide exactly with 360 nm valley of reflection, but corresponds total phase shift 180º in the particle optical cavity. Peak position of total reflection at 430 nm is strongly connected with material properties of Ag and SiO2 and cannot be significantly blushifted by LSPR. Therefore, difference in peak positions of absorptance (sensitive to LSPR) and total reflection (sensitive to optical properties) can be significant.

3.3. Reflection of Ag Nanoparticles on Si Substrate

Fig. 4 demonstrates reflection and scattering of silver nanoparticles on bare silicon substrate. This sample was prepared by annealing of silver film deposited on Si substrate with native oxide. The UV part of total reflection looks similar to reflection of IBM samples on quartz (Fig. 2 (a, b)), but with increased peak intensity and redshifted to 380 nm minimum of reflectance. The VIS part of total reflection is dissimilar to spectra of Fig. 2 and contents only one very broad peak at 720 nm.

Silicon has much higher dielectric constant (ε=15 at 633 nm) than quartz. Due to this and higher surface density of nanoparticles (Fig. 1 (d)), dipolar coupling is very strong in this sample. It leads to


175

spliting of original LSPR peak in blue and red ones. In this case, atypical nanoparticles are not coupled ones and diffusion reflection shows position of original LSPR at 830 nm. Therefore, VIS spectrum is manly formed by blue component of split LSPR. Phase shift of radiated light differs in coupled and isolated nanoparticles. Coupled particles oscillate with Δφpl > 90º and Δφpl < 90º at the frequencies of blue and red peaks, respectively. It means, that constructive interference happens for whole VIS part of total reflection in the optical cavity with nanoparticle and results in observed spectrum. Position of dipolar LSPR cannot be identified in reflection spectrum in case of Si substrate.

Fig. 4. Total and diffuse reflection of silver nanoparticles on silicon substrate.

4. Ag Nanoparticles on a Multilayer Substrates In the previous section, we demonstrated that

nanoparticle and substrate form an optical cavity. It, in turn, leads to intensity variation of electromagnetic field near the nanoparticle. Multilayer substrate provides additional reflection interfaces, what result in multiple optical cavities including the same nanoparticle. Fig. 5 shows spectra of Xe mixed Ag nanoparticles on 100 nm thick Al2O3 layer above c-Si substrate. Here is also given reflectance spectrum of 15 nm thick Ag film above Al2O3 before IBM. This spectrum demonstrates transparency of 15 nm thick silver film and high quality interference in Al2O3 optical cavity with Si and Ag mirrors. The minimum at 290 nm and maximum at 380 nm are close to theoretical interference extrema, calculated with bulk silver optical constants.

After formation of nanoislands, the minimum at 530 nm is replaced by maximum at 520 nm and minimum at 470 nm. In this sample, the new optical cavity is created between reflecting surface of Si substrate and Ag nanoparticle on Al2O3 layer. Therefore, presence of the additional optical cavity creates two local extrema visible in total reflection. When in the optical cavity one of the mirrors is replaced with plasmonic structure, the phase shift balance for extrema is [1]

Fig. 5. Total and diffuse reflection of silver nanoparticles on 100 nm thick Al2O3 layer deposited on silicon substrate.

2∆ + ∆ + ∆ = , (4)

where Δφprop is the phase shift due to propagation of the wave through cavity, Δφrefl is the phase shift upon reflection at the cavity mirror and N is integer. From diffuse reflection spectrum in Fig. 5 follows, that LSPR of Ag nanoparticles on Al2O3 is situated below 510 nm. Wavelength of local minimum in Fig. 5 is 460 nm, what corresponds 2Δφprop=1.47π (nAl2O3=1.69). Additional phase shift Δφpl=0.53π provides condition for destructive interference (Δφrefl=π) in the Al2O3 optical cavity at 460 nm according to (4). The dipolar resonance itself is redshifted on 0.03π from the local minimum, because Δφpl=0.53π and Δφ=0.5π at the resonance.

(a)

(b)

Fig. 6. Total and diffuse reflection of silver

nanoparticles created by (a) IBM and (b) annealing on oxidized Si substrate.

0

0,2

0,4

0,6

200 300 400 500 600 700 800 900

Refle

ctan

ce

Wavelength (nm)

Total

Diffuse, x2

0

0,2

0,4

0,6

200 300 400 500 600 700 800 900

Refle

ctan

ce

TotalDiffuse, x10Al2O3/Agc-Si/Al2O3

0

0,2

0,4

0,6

200 300 400 500 600 700 800 900

Refle

ctan

ce

High dose, totalLow dose, totalHigh dose, diffuse, x10

0

0,2

0,4

0,6

200 300 400 500 600 700 800 900

Refle

ctan

ce

Wavelength (nm)

TotalDiffuse, x2c-Si/SiO2, calculated


176

Fig. 6 (a) demonstrates total and diffuse reflection for Ar mixed Ag nanoparticles prepared on Si substrate covered by 20 nm of SiO2. The fabrication procedure is the same as for samples on quartz substrate in Fig. 2 (a, b). The obtained spectra look similar to spectra in Fig. 2 (b) in UV part. They include the same peak at 330 nm and valley at 360 nm. Nevertheless, VIS parts of the spectra in Fig. 2 (b) and Fig. 6 (a) are different. There are two reasons for this: light propagation delay in SiO2 sublayer and reflection from SiO2/Si interface. The wavelength dependent Δφprop, nonresonant reflection from silver nanoparticles and Δφrefl=π at the SiO2/c-Si interface form increasing part of reflectance from 510 nm to 900 nm. At wavelengths shorter than 510 nm LSPRs of Ag nanoparticles in optical cavities increase reflectance similarly to quartz substrate. The reflection spectrum of high dose sample (Fig. 6 (a)) demonstrates one more quadrupolar resonance redshifted to 390 nm. It is attributed to Ag nanoparticles submerged in SiO2 during IBM [10].

(a)

(b)

Fig. 7. (a) Total reflection and (b) extinction of the

sample with two Ag nanoparticle layers and 73 nm thick SiO2 spacer.

Spectra of Ag nanoparticles prepared by annealing

on oxidized Si substrate (36 nm thick SiO2) are given in Fig. 6 (b). They are similar to spectra in Fig. 6 (a) in UV part. The VIS part of Fig. 6 (b) has a broader dipolar peak and redshifted valley in comparison with

similar features in Fig. 6 (a). Larger size of silver nanoparticles (Fig. 1 (d)) redshifts position of dipolar LSPR to 550 nm, how it is visible from scattering in Fig. 6 (b). Additionally, high surface density of nanoparticles (Fig. 1 (d)), results in splitting of original LSPR in the same way as for Si substrate. Part of optical cavities with Ag nanoparticles acquires Δφpl > 90º at blue peak wavelength and creates constructive interference in the range 450-600 nm, what leads to increased FWHM of the VIS peak.

The next two samples for demonstration of cavity modified LSPRs include two layers of Ag nanoparticles with SiO2 spacer between them on quartz substrate. Both nanoparticle layers were prepared by high dose Ar IBM of silver films. The samples differ only by thickness of SiO2 layers (73 nm and 144 nm). Reflection and extinction of samples with SiO2 thicknesses 73 nm and 144 nm are shown in Fig. 6 and Fig. 7, respectively. The samples were characterized at different stages of fabrication: the first layer of Ag nanoparticles (1), this layer capped by SiO2 and Ag (1+SiO2/Ag), two layers of Ag nanoparticles (2), the whole structure capped by SiO2 (2+ SiO2).

(a)

(b)

Fig. 8. (a) Total reflection and (b) extinction of the

sample with two Ag nanoparticle layers and 144 nm thick SiO2 spacer.

The first layer of Ag nanoparticles on quartz

substrate demonstrates identical for both samples reflectance and extinction spectra (blue curves in

0

0,2

0,4

200 300 400 500 600 700 800 900

Refle

ctan

ce

Wavelength (nm)

11+SiO2/Ag22+SiO2

0

0,5

1

1,5

200 300 400 500 600 700 800 900

Opt

ical

den

sity

Wavelength (nm)

11+SiO222+SiO2

0

0,2

0,4

200 300 400 500 600 700 800 900

Refle

ctan

ce

Wavelength (nm)

11+SiO2/Ag22+SiO2

0

0,5

1

1,5

200 300 400 500 600 700 800 900

Opt

ical

den

sity

Wavelength (nm)

11+SiO222+SiO2


177

Fig. 6, Fig. 7), controlled by nanoparticle in an optical cavity near substrate (quartz cavity). However, capping of Ag nanoparticles by SiO2 and silver leads to difference in spectra for 73 and 144 nm thick SiO2. For thin oxide layer interference picture and reduced intensity of total reflection in UV are observed (Fig. 7 (a)). For thick oxide layer UV valley disappears and VIS peak is redshifted (Fig. 8 (a)). The difference is connected with appearance of additional optical cavity for every nanoparticle. The new cavity is formed by SiO2 layer, Ag mirror and Si mirror (SiO2 cavity). Difference in thickness of oxide layer results in different propagation phases. For example, at wavelength 418 nm 2Δφprop=π and 2Δφprop=2π for thin and thick oxide layers, respectively. Accordingly to destructive and constructive interference in the SiO2 layer, new features appear in the reflection spectra (red curves) at this wavelength: the valley in Fig. 7 (a) and steep increase of reflectance in Fig. 8 (a). Therefore, radiation from optical cavities (quartz cavity and SiO2 cavity) can be added (Fig. 8) or subtracted (Fig. 7) and switching of the mode can be done by Δφprop. In case of addition, contributing components of radiation cannot be easily distinguished (Fig. 8), because all of them work in the same phase. In case of subtraction, quadrupolar LSPR of Ag nanoparticle in SiO2 is designated by peak at 385 nm (quartz cavity) and valley at 365 nm (SiO2 cavity). After IBM of upper Ag layer (black curves), in the sample with thick oxide the additive dipolar peak at 495 nm and shoulder at 418 nm are visible. In the sample with thin oxide four dipolar extrema can be distinguished. Due to splitting in SiO2, the original dipolar LSPR is observed as blue and red components. Additionally, two optical cavities support different resonance modes and select two more wavelengths from both blue and red components, respectively. As a result, two dipolar LSPRs are ascribed to the quartz cavity (480 nm and 650 nm) and two dipolar LSPR are connected with the SiO2 cavity (440 nm and 530 nm). After covering by capping SiO2 layer, only one optical cavity created by two nanoparticle layers is left in both samples. The sample with thick oxide demonstrate one dipolar peak at 460 nm and shoulder at 380 nm corresponding to quadrupolar LSPR. The sample with thin oxide shows dipolar LSPR range between peak at 465 nm and valley at 620 nm (green curves). In this range variation of Δφpl provides switching between constructive and destructive interference in the optical cavity.

5. UV Features of the Spectra

All studied samples, exclude the 100 nm Al2O3/Ag and capped ones, demonstrate 330 nm peak in the UV part of reflection spectra. Moreover, in the RIE processed sample, this feature was observed in as prepared nanoparticle array and disappeared after RIE (Fig. 2 (c)). Therefore, existing of reflective surface with phase shift close to π below Ag nanoparticles is essential for obtained results. Wavelength of UV peak (330 nm) does not depend on nanostructure shape and

substrate ε. The peak can be attributed to variation of reflection phase at Ag/air and Ag/substrate interfaces according with wavelength and coincides with maximum of silver refractive index nAg. Removing of reflective surface by pillars or capping changes reflection conditions and the mentioned UV feature disappears.

Position of UV peak in absorption (360 nm) strictly coincides with the peak of diffuse reflection (Fig. 2 (b), Fig. 4 and Fig. 6). It means that this feature is a quadrupole LSPR [11]. However, its position weakly depends on dielectric environment and for Si substrate with high ε the valley is shifted only to 375 nm (Fig. 4). At the same time, quadrupolar LSPR of submerged in oxide Ag particles is shifted to 390 nm (Fig. 6 (a) and Fig. 7 (b)). Additionally, extinction coefficient of silver has minimum and substrate reflection has maximum at the same wavelength 360 nm (see calculated spectra at Fig. 6 (b)), which in turn, facilitates destructive interference. We believe that all these factors contribute in stable position of 360 nm valley.

6. Conclusions

It has been demonstrated that peaks and valleys in reflection spectrum of disordered nanoparticles on multilayer substrate do not correspond directly to plasmon resonances. The sample reflectance is modulated by constructive and destructive interference in the optical cavity, containing the nanoparticle. Nevertheless, it is possible to identify position of LSPR with accuracy of FWHM of resonance band. This position is situated between two local extrema of reflectance spectrum, corresponding phase shift variation during LSPR. Additionally, LSPR positions can be unmasked by proper design of studied samples, e.g., by eliminating reflecting surfaces or by variation of propagation phase shift in the optical cavity. The spectrum features in UV range may be attributed either to quadrupolar resonance or to variation of material properties. In the first case, the valley position depends on dielectric environment and geometry of plasmonic structures. In the second case, the peak is fixed at 330 nm and is observed only in nanostructures having back reflecting surface. The obtained results can be used in analysis and design of plasmonic nanostructures on opaque substrates.

Acknowledgements

This research was undertaken at the Micronova Nanofabrication Centre, supported by Aalto University.

References [1]. R. Ameling, L. Langguth, M. Hentschel, M. Mesch,

P. V. Braun and H. Giessen, Cavity-enhanced localized plasmon resonance sensing, Applied Physics


178

Letters, Vol. 97, Issue 25, 2010, pp. 253116-1-253116-3.

[2]. E. C. Le Ru and P. G. Etchegoin, Principles of surface-enhanced Raman spectroscopy and related plasmonic effects, Elsevier, 2009.

[3]. S. Pillai, K. R. Catchpole, T. Trupke, and M. A. Green, Surface plasmon enhanced silicon solar cells, Journal of Applied Physics, Vol. 101, Issue 9, 2007, pp. 093105-1-093105-8.

[4]. P. Biagioni, J.-S. Huang, and B. Hecht, Nanoantennas for visible and infrared radiation, Reports on Progress in Physics, Vol. 75, Issue 2, 2012, pp. 024402-1-024402-40.

[5]. A. Shevchenko, V. Ovchinnikov, and A. Shevchenko Large-area nanostructured substrates for surface enhanced Raman spectroscopy, Applied Physics Letters, Vol. 100, Issue 17, 2012, pp. 171913-1-171913-4.

[6]. V. Ovchinnikov, Reflection from irregular array of silver nanoparticles on multilayer substrate, in Proceedings of the 9th International Conference on Quantum, Nano/Bio, and Micro Technologies (ICQNM 2015), Venice, Italy, 23-28 August 2015, pp. 16-21.

[7]. V. Ovchinnikov and A. Priimagi, Anisotropic plasmon resonance of surface metallic nanostructures prepared by ion beam mixing, in Proceedings of the 1st

International Conference on Quantum, Nano and Micro Technologies (ICQNM'07), Guadeloupe, French Caribbean, 2-6 January 2007, pp. 3-8.

[8]. V. Ovchinnikov, Analysis of furnace operational parameters for controllable annealing of thin films, in Proceedings of the 8th International Conference on Quantum, Nano/Bio, and Micro Technologies (ICQNM 2014), Lisbon, Portugal, 16-20 November 2014, pp. 32-37.

[9]. V. Ovchinnikov, Effect of thermal radiation during annealing on self-organization of thin silver films”, in

Proceedings of the 7th International Conference on Quantum, Nano and Micro Technologies (ICQNM 2013), Barcelona, Spain, 25-31 August 2013, pp. 1-6.

[10]. V. Ovchinnikov, Formation and characterization of surface metal nanostructures with tunable optical properties, Microelectronics Journal, Vol. 39, Issue 3-4, 2008, pp. 664-668.

[11]. E. Thouti, N. Chander, V. Dutta, and V. K. Komarala, Optical properties of Ag nanoparticle layers deposited on silicon substrates, Journal of Optics, Vol. 15, Issue 3, 2013, pp. 035005-1-035005-7.

[12]. V. Ovchinnikov and A. Shevchenko, Surface plasmon resonances in diffusive reflection spectra of multilayered silver nanocomposite films, in Proceedings of the 2nd International Conference on Quantum, Nano and Micro Technologies (ICQNM 2008), Sainte Luce, Martinique, 10-15 February 2008, pp. 40-44.

[13]. T. Sang, L. Wang, S. Ji, Y. Ji, H. Chen, and Z. Wang, Systematic study of the mirror effect in a poly-Si subwavelength periodic membrane, Journal of the Optical Society of America A, Vol. 26 Issue 3, 2009, pp. 559-565.

[14]. V Ovchinnikov, A Malinin, S Novikov, and C Tuovinen, Silicon nanopillars formed by reactive ion etching using a self-organized gold mask, Physica Scripta, Vol. T79, 1999, pp. 263-265.

[15]. A. Shevchenko and V. Ovchinnikov, Magnetic excitations in silver nanocrescents at visible and ultraviolet frequencies, Plasmonics, Vol. 4, Issue 2, 2009, pp. 121-126.

[16]. H.W. Edwards, Interference in thin metallic films, Physical Review, Vol. 38, 1931, pp. 166-173.

[17]. RefractiveIndex.INFO (http://refractiveindex.info).

___________________


SensorsGlobalSummit.com

SensorsGlobalSummit.com


179



Performance Analysis of Commercial Accelerometers:

A Parameter Review

Stephan Elies Fraunhofer Institute for High-Speed Dynamics, Ernst-Mach-Institut, EMI, Am Klingelberg 1,

Efringen-Kirchen, 79588, Germany Tel.: + 49 7628 9050-691


Received: 31 August 2015 /Accepted: 5 October 2015 /Published: 30 October 2015 Abstract: This paper points out the performance and limits of state of the art commercial capacitive, resistive, piezoresistive, piezoelectric, thermal and optical accelerometers and can be used as a reference work. Therefore, datasheets of 118 accelerometers from 27 manufacturers of eight countries have been analyzed. Focus of the analysis were the parameters overload shock limit, measurement range, frequency response, resonance frequency, volume, weight, power consumption, operating temperature, storage temperature and cost of both uniaxial and triaxial accelerometers. A strict overload shock limit of 10,000 g for accelerometers with proof mass and a measurement range of less than 2,000 g was found. Also, that the performance of uniaxial and triaxial accelerometers differs. Especially uniaxial piezoelectric accelerometers show a better performance with regard to overload shock, measurement range, resonance frequency, frequency response and operating temperature in contrast to triaxial ones. Piezoelectric accelerometers show the highest overload shock limits, measurement range and operating temperature, capacitive accelerometers the lowest power consumption and volume, piezoresistive accelerometers the widest frequency response. Copyright © 2015 IFSA Publishing, S. L. Keywords: Commercial accelerometers, Review, Market overview, Shock limit, Performance. 1. Introduction

Today, off-the-shelf accelerometers are widely used for numerous applications in industries such as consumer electronics, automotive, biomedical, robotics and military [1, 2]. Within the conducted review, seven technologies of today’s commercial accelerometers could be identified. The capacitive, piezoresistive and piezoelectric technology with charge output and also with voltage output are most commonly used [1-4]. Further technologies are the resistive, optical and thermal ones. All these technologies differ in their performance due to their internal structure according to their different

transduction principles. The term technology is used as a generic term for a clear distinction between piezoelectric accelerometers with charge output and voltage output due to the fact they base on the same (piezoelectric) transduction principle. Technology is the more top level term.

So why a review of well-known and widely used technologies? A literature review revealed that there is a general lack of reviews of accelerometers also stated in [1]. Still, reviews of commercial state of the art accelerometers [1, 2] are mostly focused on commonly used technologies, the first three mentioned ones. Still there are less up to date reviews and papers [1, 2, 5] and they are only focused on one




180

or two commonly used technologies. In addition there are reviews out-of-date or focused solely on special applications [1, 2, 6, 7].

Very comprehensive reviews in regard to their mathematically description of transduction principles, algorithms, electronic control circuits, fabrication techniques and other characteristics are [1, 2] but altogether these reviews analyzed only the following parameters of accelerometers; weight, resonance frequency, measurement range, sensitivity, frequency response and noise floor. For real world applications parameters such as volume, operating temperature, power consumption, and cost are important to fit physical and financial boundary conditions.

A less academic view on accelerometers mostly including such parameters is addressed in reference guides by the manufacturers, e.g., [3, 8, 9]. These guides are made for customers to find they right type of accelerometer if they are not familiar with the topic. But they are only focused on technologies of their own products, mostly two or three technologies.

Hence there is no (known) review or guide of accelerometers focusing on more than two or three commercial technologies including parameters such as volume, power consumption, overload shock limit, operating and storage temperature and cost. Hence in this paper we give an overview of state of the art commercial accelerometers of common used technologies.

Because we analyzed entire seven technologies we do not go into detail with regard to transduction principles, algorithms or electronic control circuits. Instead we will identify upper and lower limits of parameters of accelerometers to get a general overview on different common used commercial technologies by analyzing: A number of 118 accelerometers from

27 manufacturers of eight countries; Piezoresistive, resistive, piezoelectric with charge

and with voltage output, capacitive, optical and thermal accelerometers;

The parameters overload shock limit, measurement range, frequency response, resonance frequency, volume, weight, power consumption, operating, storage temperature and cost for both uniaxial and triaxial accelerometers. Our review covers very different commercial

accelerometers used in numerous fields. We will address this fact in detail in Section 4 but we do not consider accelerometers effected by international traffic in arms regulations and export restrictions.

In Section 2, the transduction principles of the analyzed accelerometers are briefly explained and advantages and drawbacks of each technology are mentioned. The analyzed database and the process of data acquisition is addressed in Section 3. In Section 4, the analyzed parameters measurement range, overload shock limit, power consumption, volume, weight, resonance frequency, frequency response, operating and storage temperature and cost of accelerometers are discussed. Finally, the results are summarized in Section 5.

2. Transduction Principles of Accelerometers The vast majority of accelerometers consist of a

proof mass, acting as a mechanical sensing element, attached to a reference frame by a mechanical suspension system [10]. An acceleration applied to the reference frame leads to a deflection of the proof mass caused by the inertial force according to Newton’s second law. There are many different variations of suspension systems [1, 2, 10] but we identified two general types for the analyzed accelerometers. For piezoelectric accelerometers the proof mass is attached by a spring to a reference frame and for resistive, piezoresistive, capacitive and optical accelerometers the proof mass is attached by a cantilever beam to a reference frame. However they can be described as a spring-mass-damper system [10].

There are a number of transduction principles to detect the displacement of a proof mass or the bending of a cantilever beam [2, 10]. Common commercial accelerometers detect the change of resistance, capacitance, charge, temperature or optical characteristics. Further used transduction principles are electromagnetic, resonance and tunneling principles [2, 6, 10]. But we found no manufacturer selling accelerometers based on this principles off-the-shelf.

Each transduction principle needs electronics to convert the signal, such as the change in resistance, of a transducing element into an analog or digital output, and for signal amplifying, filtering and processing. A transducing element is the part of an accelerometer that accomplishes the conversion of motion into a signal [11]. For example, a strain gauge is the transducing element and the electric resistance is the signal to be converted into a voltage. Mostly electronics are combined in one housing but there are also good reasons for decoupling electronics such as harsh or hazardous environments with high voltage levels, temperatures, electromagnetic interference or environments containing explosive substances. For optical and piezoelectric accelerometers with charge output external electronics have to be connected by cables and can be positioned far from the system to be measured.

General, accelerometers are classified into two types with regard to their frequency response also called bandwidth [3]. An AC-response accelerometer is capable to measure dynamic acceleration (meaning change in acceleration). A DC-response accelerometer can measure static (constant) acceleration, such as gravity acceleration, as well as dynamic acceleration.

There are essential interdependence of parameters of accelerometers due to their internal structure, the spring-mass-damper system. Increasing the proof mass of an accelerometer leads to a more strong deflection of the proof mass under the same loading. In other words sensitivity increases. At the same time


181

the resonance frequency and consequently the bandwidth decreases. In contrast, decreasing the proof mass leads to a diametric behavior. So it’s not possible to maximize sensitivity and bandwidth respective resonance frequency. Subsequent we will briefly explain the seven technologies and highlight the most important fact off every technology.

2.1. Piezoelectric Accelerometers with Charge Output

This type of accelerometer comprise piezoelectric materials as polarized ceramics or quartz crystals bonded with a proof mass [12]. An acceleration leads to a deflection of the proof mass and causes stress in the material subsequently. Under stress, the piezoelectric effect causes a charge transfer in the crystal. The amplitude of charge can be measured on the surface of the crystal. For this type electronics is not integrated so these accelerometers provide a charge output signal which has to be transferred by special low noise cable to external electronics [12]. They work only for AC-response. One benefit of decoupling transducing element and electronics is the wide temperature range of common piezoelectric materials up to 400 °C and more [11][12]. Due to the unmodified charge output signal these accelerometers are interchangeable [9].

2.2. Piezoelectric Accelerometers with Voltage Output

They work exactly the same way as piezoelectric accelerometers with charge output, but the electronics to amplify the charge signal and convert it into a voltage signal are integrated [12]. For power supply a constant current source with current levels from 0.5 to 20 mA is mostly needed [9]. They provide a voltage output signal and can also only be used for AC-response. In contrast to piezoelectric accelerometers with charge output the temperature range is limited to that which the electronic will withstand. Furthermore reliability is lower if the accelerometer is subjected to harsh environments but the noise level is lower because of shorter cable length between transducing element and electronics [11]. Still they may not be interchangeable if the power requirement is not the same [9].

2.3. Resistive Accelerometers

This type of accelerometers detect the change of resistance of a metal foil strain gauge or wire bonded to a cantilever beam [11]. An acceleration leads to a bending of the cantilever beam and thus to a change in geometry of the strain gauge. According to the change in geometry the resistance of the strain gauge alters, described by the piezoresistive effect [10]. Up to four strain gauges are configured in a Wheatstone bridge circuit [11]. They provide a voltage signal and

work for DC-response. In the past wire or foil strain gauges have been used exclusively, today silicon strain gauges have a wider distribution because of their higher sensitivity [11].

2.4. Piezoresistive Accelerometers

They work exactly the same way as resistive accelerometers, but the strain gauge is fabricated from semiconductor materials [10]. Single crystal silicon is up to 100 times more sensitive to strain than a metal foil strain gauge [10]. For semiconductors the change of the specific resistance according to stress is the dominating effect caused by the piezoresistive effect. The change in resistance due to change in geometry can be neglected [10]. Configured in a Wheatstone bridge circuit, they provide a voltage signal and function also for DC-response. They are more sensitive to temperature, show lower linearity and breaking strain than metal foil strain gauges [13].

2.5. Capacitive Accelerometers

These accelerometers detect the change of a capacitor configuration. The electrodes of the capacitor are composed of semiconductor material, such as silicon. Typically the differential change in capacitance, of a moveable electrode in-between two stationary electrodes, is detected [10]. The moveable electrode is attached to a proof mass and deflects due to acceleration which leads to a differential change in capacitance. Configured in a Wheatstone bridge circuit, they provide a voltage signal and work for DC-response [2]. This is the most common technology used for accelerometers today [3]. Some benefits are low cost and low power consumption but in contrast they can be vulnerable to electromagnetic interference [6].

2.6. Optical Accelerometers

A variety of transduction principles to convert a mass displacement into a change in optical characteristics exists [11]. In particular one technology have become widespread use, the optical fiber technology [14]. Still there is a variety of technics for optical fiber measurement [15] but the Fiber Bragg Grating (FBG) principle has become widely known [14] and FBG accelerometers seem to be the popular optical fiber technology today.

Bragg Gratings are interference filters written into optical fibers. The gratings reflect only a narrow spectral component of the induced light. [16] This characteristic is used for FBG accelerometers. An acceleration leads to a deformation of an optical fiber attached to a suspension beam. The deformation of the optical fiber change the reflection characteristic of the Bragg gratings. This change can be detected by comparing the spectral component of the reflected with the induced light. They work for DC-response


182

and one benefit of FBG accelerometers is the immunity of light of the transducing element to electromagnetic interference. So usually, electronics are not integrated and they provide a pure optical signal. Electronics can be connected by optical fibers. As well as for piezoelectric technology with charge output, electronics can be positioned far away from the measurement position.

Drawback is their greater size in contrast to the other analyzed accelerometers. There potential for miniaturization is limited and MEMS-technology favors other fiber techniques [17]. In the present work we only analyzed optical FBG accelerometers because no other commercial optical accelerometers could be found. 2.7. Thermal Accelerometers

Thermal accelerometers based on mass displacement have been studied in [18]. We found no manufacturer selling these accelerometers off-the-shelf. Thermal accelerometers without mass displacement have been reported among others in [19]. These types of accelerometers consist of a heater and thermocouples located around the heater in a hermetic chamber. Without acceleration, the heater creates a symmetric temperature profile in the chamber. When acceleration is applied, the hot air in the chamber moves and the temperature profile gets asymmetric. The asymmetry can be detected by the thermocouples around the heater. There is only one manufacturer that patented and commercialized this technology [5]. Electronics is integrated and these accelerometers provide a voltage signal and work for DC-response. In the present work, we only analyzed thermal accelerometers without moving parts. The greatest benefits is their resilience to very high overload shocks due to the absence of a proof mass.

3. Database and Procedure of Data Acquisition

For this paper, we analyzed accelerometers of

manufacturers worldwide. We did not focused to one special type of accelerometer, e.g., micro-g or high-g accelerometers. We directed our review on several different characteristics of accelerometers; high measurement range and overload shock limit, low power consumption, low volume, wide operating temperature range and so on.

Thereby we analyzed very different accelerometers, e.g., for crash-, shock- and impact testing, inertial navigation, machine vibration and bridge monitoring, modal analysis; for applications in digital cameras, mp3 players, mobile phones, medical instrumentations, wind turbines, airbags, sport diagnostics, marine, motor sports, robotics, embedded applications, bio dynamics, pedestrian crash testing; in domains of consumer, research and

development, industrial, railway, oil, gas, aerospace, military and automotive industries.

So this paper includes accelerometers for very different applications, with measurement ranges from -110,000 g up to 110,000 g, with overload shock limits up to 200,000 g, with rugged metal housings or as simple service mounted devices (SMD). The unit g stands for the gravity acceleration defined as 9.81 meters per square seconds.

All analyzed parameters were stored in a database to identify limits and correlation between parameters by data mining.

3.1. Procedure of Data Acquisition and Significance of Data

Generally the scope of the work was to capture

the state of the art of accelerometers from manufacturers worldwide with a certain focus to German manufacturers. Hence the procedure of data acquisition was as follows:

First, we started analyzing accelerometers from German manufactures and took into account one or two samples of different technologies of each manufacturer, for uniaxial and triaxial measurement. Overall we considered accelerometers of ten German manufacturers. Off course there are much more manufacturers and companies in Germany selling accelerometers but many are only distributers of accelerometers of manufacturers worldwide. From the results we deduced benchmarks of state of the art accelerometers of German manufacturers for further analysis.

Second, we analyzed accelerometers worldwide and for the most part we considered only samples of accelerometers with equal or outstanding performance relative to our deduced benchmarks. During the analysis the benchmarks were updated with every new sample and so it was harder to find accelerometers exceeding these benchmark. Thereby not so many manufacturers of countries worldwide have been taken into account in contrast to the number of German manufacturers.

That means, the number of analyzed countries and manufacturers of a country (Germany as well as worldwide) is not exhaustive, but we found no commercial accelerometers with higher overload shock limit, higher measurement range and so on. So the database is suitable to point out lower and upper limits of parameters. In contrast the database is not suitable for statistics calculations such as calculating the average value of a parameter.

3.2. Database

In Table 1 the number of accelerometers classified by their country of origin into European Union and worldwide are listed. According to our explanations before, there is a great focus on accelerometers of German manufacturers, and the


183

number of manufacturers within a country as well as the number of accelerometers of a manufacturer is different.

We chose the headquarters of the manufacturers as country of origin. Due to globalization, several production steps such as engineering, manufacturing, assembly, testing, etc., could be distributed across

several locations worldwide. We think headquarters is most suitable to compare the performance by countries. Because of 27 manufacturers, 15 produce their accelerometers at the headquarters, 7 produce at the headquarters and additionally at one or more locations worldwide. Only for 5 manufacturers, the location of production is not explicitly specified.

Table 1. Number of analyzed accelerometers classified by country of origin into European Union and Worldwide.

Country of origin Number of

manufactures Number of

accelerometers Classification by country of origin

Germany 10 35 European Union (EU) UK 2 11

Denmark 1 11 USA 7 37

Worldwide (Non EU) Switzerland 3 13 China 1 9 Canada 1 1 Hong Kong 1 1

In Table 2 the number of analyzed accelerometers of German manufacturers and for the remaining manufacturers are listed. In addition the technologies are classified by frequency response.

Most common are the capacitive, piezoelectric and piezoresistive technology. Many manufacturers are specialized to one or more of these technologies. Resistive, optical and thermal accelerometers are not very common in contrast to the other technologies. The thermal technology is patented by only one manufacturer and we took into account two samples of their products. Accelerometers using resistive technology are really hard to find because piezoresistive strain gauges are mostly used. We took into account four samples of optical accelerometers because there are only few manufacturers specialized on this technology, they are made for very special niches and because in matters of some analyzed

parameters their performance is lower in contrast to the other technologies. We will discuss this issues in the next section in detail.

Because of small samples of the thermal, resistive and optical technology we will highlight only special characteristics of these three principles in the next section. But note, for these three technologies we found no accelerometer with higher performances. Due to the fact that the analyzed accelerometers are designed for very different applications we will explain some important facts and differences. The two thermal, nearly half of the capacitive and three piezoresistive accelerometers are manufactured as simple service mounted devices. The remaining accelerometers are integrated in more or less rugged housings of aluminum, titanium or plastics. Thereby volume and weight can be very different within a technology.

Table 2. Number of analyzed accelerometers of German manufacturers and remaining manufacturers. The technologies are

classified by frequency response.

Technology Number of accelerometers of German manufactures

Number of accelerometers of remaining manufacturers

Classification by frequency response

Capacitive 19 21

DC-response Piezoresistive 6 9 Resistive 1 6 Optical 1 3 Thermal - 2 Piezoelectric voltage output 5 25

AC-response Piezoelectric charge output 3 17

4. Analysis of Parameters

In this section, the parameters measurement range, overload shock limit, power consumption, volume, weight, frequency response, resonance

frequency, operating and storage temperature of accelerometers specified on their datasheets are analyzed. Additionally to our previous review [20] in this paper the cost have been analyzed in detail.


184

4.1. General Explanation Before we start with the evaluation of parameters

we will explain the figures in this section in detail. The figures were designed to correlate the analyzed parameters among the measurement range. This is the parameter of choice because accelerometers are made for measuring acceleration.

First, we classified all analyzed accelerometers by technology and the technology by the number of measurement axis, hence there are 13 classes, see Table 3. Second, we sorted all accelerometers within a class by ascending measurement range. Third, we numbered all accelerometers by an identifier (ID) starting with the class “piezoresistive 1-Axis” and ended with the class “thermal 3-Axis”, see Table 3. That means within a class the measurement range increase with ascending identifier. A number of 118 different accelerometers have been analyzed, so there are 118 IDs, where each ID is a unique identifier of an accelerometers. That means an accelerometer, e.g., with the ID 50 in Fig. 1 has the same ID in Fig. 2 to Fig. 8.

Table 3. Classification of accelerometers by technology and number of measurement axis and numbering of the

accelerometers within a class by identifiers.

Range of ID Classes 1 - 10 Piezoresistive 1-Axis 11 - 15 Piezoresistive 3-Axis 16 - 30 Capacitive 1-Axis 31 - 55 Capacitive 3-Axis 56 - 68 Piezoelectric charge output 1-Axis 69 - 75 Piezoelectric charge output 3-Axis 76 - 93 Piezoelectric voltage output 1-Axis

94 - 105 Piezoelectric voltage output 3-Axis 106 - 109 Resistive 1-Axis 110 - 112 Resistive 3-Axis 113 - 114 Optical 1-Axis 115 - 116 Optical 3-Axis 117 - 118 Thermal 3-Axis

Among the 118 accelerometers, there are only three capacitive ones with two measurement axes. Because of that small number they were classified into the class of “Capacitive 3-Axis” accelerometers and no differentiation was made. 4.2. Measurement Range of Accelerometers

In datasheets, the measurement range is mostly specified for a symmetric range of acceleration such as from -500 g to 500 g. Rarely the lower limit is restricted, e.g., from -200 g to 500 g. In this review, the parameter measurement range stands for the upper limit of positive range of acceleration. For the last given example the measurement range is 500 g. In this review there is no need to analyze the negative measurement range, as for 95 of 118 accelerometers

the measurement range is specified for a symmetric range. For 24 accelerometers, the lower limit is not specified and only for one accelerometer, the lower limit is restricted.

In Fig. 1 the measurement range is plotted against the accelerometer-ID. Note that this is a three parameter plot. The visualized parameters are technology, number of measurement axis and measurement range. For the sake of completeness the three capacitive accelerometers with two measurement axis are marked by the dotted circle.

Fig. 1. Measurement range of 118 accelerometers classified by technology, number of measurement axis and sorted by

measurement range in ascending order.

Thermal accelerometers are limited to a measurement range of 5 g, optical to a range of 200 g, capacitive to a range of 500 g and resistive to a range of 10,000 g. Only piezoelectric and piezoresistive accelerometers are capable of measuring more than 10,000 g with the restriction that piezoresistive accelerometers are the only ones for DC-response above 10,000 g up to 60,000 g.

For measuring accelerations of 100,000 g and more, only piezoelectric accelerometers are capable with the restriction that triaxial piezoelectric accelerometers with voltage output are strictly limited to 10,000 g. 4.3. Overload Shock Limit Again, before we continue with the analysis, we will go deeper in detail with the figures in this section since it is important to understand the concept. In Fig. 2 the overload shock limit is plotted against the ID. The arrow illustrate the decrease of overload shock limit by increasing measurement range. Note that this is a four parameter plot. The parameters are technology, number of measurement axis, overload shock limit and measurement range. The parameter


185

measurement range is not explicit presentable. Since always the same numbering of accelerometers from Table 3 is used, measurement range increases within a class with ascending ID. So it’s possible to correlate the overload shock limit (and the parameters in the following figures) among the parameter measurement range.

Besides piezoelectric accelerometers most accelerometers are not made for overload shocks beyond the 10,000 g limit. Especially capacitive accelerometers are strictly limited to the 10,000 g threshold. General the capacitive technology is utilized to withstand shock levels up to 20,000 g [11]. We identified only one manufacturer producing and selling capacitive accelerometers with such overload shock limits [21, 22]. But this product is under export control, so we did not take it into account.

Fig. 2. Overload shock limit of 118 accelerometers

classified by transduction principle, number of measurement axis and sorted by measurement range in

ascending order.

Within this review we want to point to the fact that a general comparison of overload shock limits of different accelerometers is only possible to a limited degree, due to the fact that this parameter is only valid for a specified profile of acceleration and a defined period of time. Usually in datasheets the overload shock limit is specified as the peak of a semi sinusoidal profile of acceleration. But a uniform specification of the period of time on datasheets is missing. In the datasheets we analyzed, the period of time is specified not uniformly but instead in terms of the maximum or the minimum time or the rise time of a semi sinusoidal profile. So the significance is completely different, which makes a comparison meaningless. In addition, for altogether 85 accelerometers the period of time is not even specified on datasheets. Still, often the specification for overload shocks is further restricted, e.g., to positive and negative shocks, to the different

measurement axis and to a powered or unpowered device. For this review we used a conservative approach and considered the lowest shock limit of an accelerometer specified on the datasheet.

4.4. Power Consumption

In Fig. 3 the power consumption is plotted against the ID. Note that piezoelectric accelerometers with charge output and optical accelerometers are excluded because electronics is not integrated and thereby no power is used. Of course, both principles also need electronics, but in this review, we did not focused on external electronics since this is beyond the scope of this review and these do not affect the analyzed parameters. We will explain this fact in detail in the following subsections. Furthermore, for 13 accelerometers no power consumption was specified on datasheets.

Fig. 3. Power consumption of 81 accelerometers classified by technology, number of measurement axis and sorted by


Generally the upper limit of power consumption for the analyzed accelerometers is about 1 W. The majority of capacitive accelerometers show the minimal power consumption in total beside some (ultra low power) piezoelectric accelerometers with voltage output. Our analyses show that there is no significant correlation between power consumption and measurement range. Particularly the power consumption for uniaxial piezoelectric accelerometers with voltage output is nearly constant.

4.5. Volume

In Fig. 4 the volume is plotted against the ID. The arrows show the decrease of volume by increasing measurement range. Again, note that the volume for piezoelectric accelerometers with charge output and


186

optical accelerometers does not comprise integrated electronics as explained in the section before.

The volume of the capacitive technology covers a range of five decades from 1E-3 cm3 up to 100 cm3. Because 21 capacitive accelerometers are fabricated as simple service mounted devices they show a very low volume. The remaining capacitive ones are integrated in rugged housings made of plastic or of aluminum and titanium for applications in harsh environments. However this technology has the minimal volume in total. The thermal accelerometers are also fabricated as service mounted devices but their volume is one magnitude higher than the lower limit of capacitive accelerometers.

There is a correlation of decreasing volume with increasing measurement range for piezoresistive and piezoelectric accelerometers with voltage output.

Fig. 4. Volume of 118 accelerometers classified by technology, number of measurement axis and sorted by


Piezoresistive and optical accelerometers show a slightly higher volume in contrast to the other technologies. Between piezoelectric accelerometers with charge output and piezoelectric accelerometers with voltage output (with integrated electronics) there is no significance difference in volume. 4.6. Weight

In Fig. 5 the weight is plotted against the ID. The arrows show the decrease of weight by increasing measurement range. Note again that electronics for piezoelectric accelerometers with charge output and optical accelerometers are excluded. The weight is an important parameter because the weight of an accelerometer should be less than 10 % the weight of the structure to by measured [12]. Otherwise the accelerometer affects the system to be measured.

For 20 accelerometers, 18 capacitive and the two thermal ones, the weight is not specified on datasheets. They are all fabricated as service mounted devices, so it seems the specification is missing due to their probably low weight.

The remaining 98 accelerometers are integrated in housings and thus are more comparable among each other. The weight of these technologies span a range of three decades from 0.1 grams up to 100 grams.

According to the volume, first, there is a tendency of decreasing weight by increasing measurement range for piezoresistive accelerometers and piezoelectric accelerometers with voltage output. Second, between piezoelectric accelerometers with charge output and piezoelectric accelerometers with voltage output (with integrated electronics) there is no significance difference in weight. Piezoresistive and optical accelerometers show no nameable higher weight in contrast to the other technologies

Fig. 5. Weight of 98 accelerometers classified by technology, number of measurement axis and sorted by


For optical and piezoelectric accelerometers with charge output electronics is separated from the system to be measured. That means additional volume and weight for electronics do not affect the system to be measured. 4.7. Resonance Frequency

In Fig. 6 the resonance frequency is plotted against the ID. The arrow illustrate the decrease of resonance frequency by increasing measurement range. For altogether 40 accelerometers, thereof entire 21 capacitive, 10 piezoresistive, two optical and the two thermal ones, this parameter was not specified on their datasheets. Due to sparse data, we exclude these four technologies from discussion.


187

The upper limit of resonance frequency for uniaxial piezoelectric accelerometers is 200 kHz. For uniaxial piezoelectric accelerometers with voltage output, the resonance frequency correlates with increasing measurement range. The triaxial versions show lower limitations of about 90 kHz.

Fig. 6. Resonance frequency of 78 accelerometers classified by technology, number of measurement axis and

sorted by measurement range in ascending order.

In principle for triaxial accelerometers the resonance frequency is specified individual for each axis. Mostly one or two axis show a lower specification. Due to our conservative approach we considered the lowest resonance frequency of an accelerometers.

4.8. Frequency Response

First a comparison of the frequency responses of accelerometers is only possible to a limited degree because a uniform specification of the tolerance of the output signal on datasheets is missing. For only 54 of 118 accelerometers, the frequency response is specified for a tolerance of the output signal within a range of ±3 dB. For 56 accelerometers, the tolerance of the output signal is specified non-uniformly for one or more of the following tolerances, e.g., -18 %, +15 %, ±10 %, ±7 %, ±5 %, ±10 dB, -3 dB, ±2 dB, or completely not specified. For eight accelerometers the specification of the frequency response is completely missing.

For the analysis we considered the frequency response within a range of ±3 dB if specified, because this is the most used uniform tolerance. If not specified within a range of ±3 dB we took into account the frequency response within the lowest tolerance specified on the datasheet.

Furthermore, for triaxial accelerometers the frequency response is specified individual for each

axis. Again, we considered the lowest frequency response of an accelerometer.

In Fig. 7 the range of the frequency response is plotted against the ID. Notice that the axis of ordinate combines a non-continuous logarithmic and linear scale. Beside few exceptions piezoresistive and resistive accelerometers are restricted to an upper limit of 5 kHz and capacitive accelerometers to an upper limit of 1 kHz. Piezoelectric accelerometers are capable for a frequency range of 1 Hz up to 20 kHz and a few can be used for a frequency response down to 0.2 Hz.

Fig. 7. Range of frequency response of 110 accelerometers classified by technology, number of measurement axis and


By trend uniaxial piezoelectric accelerometers

show a wider frequency response than the triaxial ones. The thermal accelerometers are limited to 17 Hz and by frequency extension circuits, this limit can be pushed up to 100 Hz [5]. 4.9. Operating Temperature

Fig. 8 presents the operating temperature plotted against the ID. Piezoelectric accelerometers with charge output can be used for very high and low temperature ranges. Their limits of operating temperature is in the range of -74 °C up to 200 °C and more. Some are capable to work down to -195 °C and up to 250 °C as the electronics as the limiting factor is excluded.

Beside resistive accelerometers all technologies can be manufactured for operating temperatures in the range of -54 °C up to 120 °C. For capacitive accelerometers, the standard temperature is in the range of -40 °C up to 85 °C. For piezoresistive accelerometers the standard temperature is in the range of -20 °C up to 85 °C.


188

Fig. 8. Range of operating temperature of 118 accelerometers classified by technology, number of

measurement axis and sorted by measurement range in ascending order.

By tendency, the positive operating temperature for uniaxial capacitive and uniaxial piezoelectric accelerometers is higher than for the triaxial ones.

4.10. Storage Temperature

In Fig. 9 the storage temperature is plotted against the ID. In contrast to the operating temperature there is no important performance gain. Only the piezoresistive and capacitive technology show extended storage temperature.

For the remaining technologies the storage temperature is equal the operating temperature. There are only a few samples with storage temperatures down to - 60 °C.

Fig. 9. Range of storage temperature of 118 accelerometers classified by technology, number of measurement axis and


4.11. Cost per Unit

We asked all manufacturers for the cost in content of a technology study. Some manufacturers did not answer the request, but overall we got the costs per unit for 107 accelerometers. Excluding the optical technology, the cost per unit reaches from 1 to 3200 €. All costs include the German value-added tax of 19 %. The cost per unit are valid for the minimum purchase, mostly one piece.

Usually additional accessories such as mounting equipment and cables are excluded. But depending on the technology and type of housing, cables are fixed with the accelerometers and thereby included in the cost.

In Fig. 10 the cost per unit are plotted against the ID. The capacitive and thermal technology show the lowest costs. They are available from 1 € but capacitive accelerometers can cost up to 2700 €. Only the service mounted device accelerometers are low cost.

Fig. 10. Costs per unit of 107 accelerometers classified by

technology, number of measurement axis and sorted by measurement range in ascending order.

The cost for the optical accelerometers have to be analyzed separated because for two samples the costs of about 10,000 € incorporates external electronics. The first one incorporates a single channel sealed, miniaturized and rugged electronics. The second one incorporates a two channel electronics for rack mounting. These two kind of electronics are very different. Without analyzing the performance of external electronics it’s difficult to compare costs but it’s a reference point in contrast with the other analyzed technologies. Furthermore it is not effective to compare totally different electronics. This is one reason why we did not focus on external electronics. The costs of 500 € and 800 € for the remaining two optical accelerometers exclude external electronics.


189

Further analyzes of Fig. 10 are not meaningful because cost are a function of several analyzed parameters (not only measurement range and number of measurement axis) and also parameters we have not considered in our review. 4.12. Performance by Country of Origin

For the last figure of this paper, we chose another design to illustrate two interesting facts. Therefor a distinction between numbers of measurement axis has not been made. Fig. 11 presents the overload shock limit plotted against the measurement range. Accelerometers from German manufacturers are highlighted in green, the one from the European Union are highlighted in orange and the non-EU ones are highlighted in black.

Fig. 11. Overload shock limit versus measurement range of

118 accelerometers. The technologies are highlighted by the shape of the symbols.

First, accelerometers from German manufacturers are limited to a measurement range of 6,000 g and to overload shock limits of 10,000 g. Capacitive accelerometers of German manufacturers are up to date in comparison with accelerometers worldwide. But especially piezoelectric and piezoresistive accelerometers show lower performances of about one order of magnitude in regard to measurement range and overload shock limit. But Fraunhofer Ernst-Mach-Institut, EMI, developed a piezoelectric accelerometer with state of the art performance [23]. Accelerometers from manufacturers of the European Union are limited to overload shocks of 100,000 g and to a measurement range of 80,000 g. This is close to the performance of state of the art accelerometers worldwide.

Second, a strict threshold for accelerometers based on a proof mass is found. Within a measurement range of up to 2,000 g, they are strictly limited to overload shocks of 10,000 g. Remember

that the thermal principle works without a proof mass and so it is less fragile to overload shocks. Therefore, beside accelerometers capable of measuring more than 10,000 g, it is very difficult to find such ones with overload shock limits beyond the 10,000 g limitation. 4.13. Concluding Considerations

Before we come to the conclusion we want to mention some additional facts. We analyzed the performance of accelerometers as they are specified on their datasheets. By request some manufacturers offer a better performance as extended temperature range for extra money. Also in part it is possible to customize accelerometers in a limited range.

Products are usually designed and balanced to special market demands such for automotive applications. Therefore this review can only point out the performance of technologies of state of the art accelerometers. Physical and technical limits of the analyzed technologies could be higher.

Furthermore a comparison of parameters of accelerometers is not always easy due to a lack of uniform or missing specification in datasheets. In this context, important parameters, such as negative measurement range, period of time for overload shock limit, frequency response and resonance frequency have to be mentioned.

5. Conclusions

A variety of very different 118 commercial piezoelectric, piezoresistive, resistive, capacitive, thermal and optical accelerometers have been reviewed. The parameters measurement range, overload shock limit, power consumption, weight, volume, frequency range, resonance frequency, operating and storage temperature and also cost were analyzed.

The paper shows that accelerometers from manufacturers of the European Union are nearly state of the art with regard to measurement range and overload shock.

A strict overload shock limit of 10,000 g was identified for accelerometers with proof mass within a measurement range of up to 2,000 g. We ascertain that the performance of uniaxial and triaxial accelerometers is slightly different. Especially uniaxial piezoelectric accelerometers show a better performance with regard to measurement range, overload shock limit, resonance frequency, frequency response and operating temperature than the triaxial ones.

In summary, piezoelectric accelerometers show the highest measurement range up to 100,000 g, shock limits up to 200,000 g and operating temperatures up to 200 °C (piezoelectric with charge output). Capacitive accelerometers show the lowest


190

power consumption down to 1E-3 W and volume down to 1E-2 cm3. Piezoresistive accelerometers show the widest frequency response from 0 Hz up to 10 kHz. Thermal accelerometers show outstanding shock limits of 50,000 g by low measurement ranges of 5 g. Optical accelerometers show immunity to electromagnetic interferences and are capable for measurement ranges up to 200 g and shock limits up to 2,000 g. The cost per unit are in the range of 1 € to 3200 € and for optical about 10,000 €. This review was meant to be a reference work for choosing the right technology.

Acknowledgements

The author would like to thank Marcel Weber for the support in extracting accelerometer parameters from datasheets and proofreading. The author would like to thank Alexander Stolz and Stefan Ebenhöch for useful hints preparing the manuscript and for reviewing the manuscript. References [1]. V. Narasimhan, H. Li, and M. Jianmin,

Mircomachined high-g accelerometers: a review, Journal of Micromechanics and Microengineering, Vol. 25, 3, Feb. 2015, pp. 1-18.

[2]. G. Krishnan, C. U. Kshirsagar, G. K. Ananthasuresh, and N. Bhat, Micromachined high-resolution accelerometers, Journal of the Indian Institute of Science, Vol. 87, 3, Jul.-Sept. 2007, pp. 333-361.

[3]. A. Chu, Choosing the right type of accelerometers, Application Notes, Measurement Specialties Inc., May 2012.

[4]. O. B. Degani, D. J. Seter, E. Socher, S. Kaldor, E. Scher, and Y. Nemirovsky, Comparative study of novel micromachined accelerometers employing MIDOS, Sensors and Actuators A: Physical, Vol. 80, 2, Mar. 2000, pp. 91-99.

[5]. J. Fennelly, S. Ding, J. Newton, and Y. Zhao, Thermal mems accelerometers fit many applications, Sensor Magazin, Mar. 2012, pp. 18-20.

[6]. N. Yazdi, F. Ayazi, and K. Najafi, Micromachined Inertial Sensors, Special Issue of IEEE Proceedings, Vol. 86, 8, Aug. 1998, pp. 1640-1659.

[7]. G. A. Macdonald, A Review of Low Cost Accelerometers for Vehicle Dynamics, Sensors and Actuators A: Physical, Vol. 21, 1-3, Feb. 1990, pp. 303-307.

[8]. Step to select the right accelerometer, Endevco, 2012. [9]. J. Lally, Accelerometers Selection Considerations

Charge and ICP® Integrated Circuit Piezoelectric,

Technical Note VIB-TN-17-0805, PCB Piezotronics Inc., 2005.

[10]. S. Beeby, G. Ensell, M. Kraft, and N. White, MEMS mechanical sensors, Artech House, Inc., Boston London, 2004.

[11]. C. M. Harris, and A. G. Piersol, Harris’ Shock and Vibration Handbook, Fifth Edition, McGraw-Hill, New York, 2002.

[12]. M. Serridge and T. R. Licht, Piezoelektrische Beschleunigungsaufnehmer und Vorverstärker – Theorie und Anwendung, Brüel & Kjær, 1990.

[13]. J. Vaughan, Application of B & K Equipment to Strain Measurements, Brüel & Kjær, 1975.

[14]. A. Méndez, Fiber Bragg grating sensors: a market overview, in Proceedings of the Conference on Third European Workshop on Optical Fibre Sensors, SPIE, Vol. 6619, July 2007, pp. 661905-1-661905-6.

[15]. K. T. V. Grattan and T. Sun, Fiber optic sensor technology: an overview, Sensors and Actuators A: Physical, Vol. 82, 1-3, May 2000, pp. 40-61.

[16]. K. O. Hill and G. Meltz, Fiber Bragg Grating Technology Fundamentals and Overview, Journal of Lightwave Technology, Vol. 15, 8, Aug. 1997, pp. 1263-1276.

[17]. E. Davies, D. S. George, M. C. Gower, and A. S. Holmes, MEMS Fabry–Pérot optical accelerometer employing mechanical amplification via a V-beam structure, Sensors and Actuators A: Physical, Vol. 215, Aug. 2014, pp. 22-29.

[18]. U. A. Dauderstädt, P. H. S. de Vries, R. Hiratsuka, and P. M. Sarro, Silicon accelerometer based on thermopiles, Sensors and Actuators A: Physical, Vol. 46, 1-3, Jan.-Feb. 1995, pp. 201-204.

[19]. A. M. Leung, J. Jones, E. Czyzewska, J. Chen, and B. Woods, Micromachined accelerometer based on convection heat transfer, in Proceedings of the IEEE 11th Annual International Workshop on Micro Electro Mechanical Systems (MEMS’98), Jan. 1998, pp. 627-630.

[20]. S. Elies, S. Ebenhöch, Performance Analysis of Commercial Accelerometers of Different Technologies, in Proceedings of the 6th International Conference on Sensor Device Technologies and Applications (SENSORDEVICES 2015), Venice/Mestre, Italy, 23.-28 August 2015, pp. 54-59.

[21]. http://www.colibrys.com/product/hs8000-high-shock-inertial-sensor/

[22]. S. Habibi, S. J. Cooper, J. M. Stauffer, and B. Dutoit, Gun Hard Inertial Measurement Unit Based on MEMS Capacitive Accelerometer and Rate Sensor, in Proceedings of the Position Location and Navigation Symposium IEEE/ION (PLANS 2008), May 2008, pp. 232-237.

[23]. R. Kuells, S. Nau, M. Salk, and K. Thoma, Novel piezoresistive high-g accelerometer geometry with very high sensitivity-bandwidth product, Sensors and Actuators A: Physical, Vol. 182, Aug. 2012, pp. 41-48.

___________________


http://www.sensorsportal.com/HTML/BOOKSTORE/Thermometry.htm

The Use of Gas-Sensor Arrays in the Detection of Bole and Root Decays in Living Trees

Documents

Transcript of The Use of Gas-Sensor Arrays in the Detection of Bole and Root Decays in Living Trees