Vol.105 (4) December 2014 SOUTH AFRICAN … (4) December 2014 SOUTH AFRICAN INSTITUTE OF ELECTRICAL...

Vol.105 (4) December 2014 SOUTH AFRICAN INSTITUTE OF ELECTRICAL ENGINEERS 1

December 2014 Volume 105 No. 4www.saiee.org.za

Africa Research JournalISSN 1991-1696

Research Journal of the South African Institute of Electrical EngineersIncorporating the SAIEE Transactions

Vol.105 (4) December 2014SOUTH AFRICAN INSTITUTE OF ELECTRICAL ENGINEERS134

(SAIEE FOUNDED JUNE 1909 INCORPORATED DECEMBER 1909)AN OFFICIAL JOURNAL OF THE INSTITUTE

ISSN 1991-1696

Secretary and Head OfficeMrs Gerda GeyerSouth African Institute for Electrical Engineers (SAIEE)PO Box 751253, Gardenview, 2047, South AfricaTel: (27-11) 487-3003Fax: (27-11) 487-3002E-mail: [email protected]

SAIEE AFRICA RESEARCH JOURNAL

Additional reviewers are approached as necessary ARTICLES SUBMITTED TO THE SAIEE AFRICA RESEARCH JOURNAL ARE FULLY PEER REVIEWED

PRIOR TO ACCEPTANCE FOR PUBLICATIONThe following organisations have listed SAIEE Africa Research Journal for abstraction purposes:

INSPEC (The Institution of Electrical Engineers, London); ‘The Engineering Index’ (Engineering Information Inc.)Unless otherwise stated on the first page of a published paper, copyright in all materials appearing in this publication vests in the SAIEE. All rights reserved. No part of this publication may be reproduced, stored in a retrieval system or transmitted in any form or by any means, electronic, magnetic tape, mechanical photo copying, recording or otherwise without permission in writing from the SAIEE. Notwithstanding the foregoing, permission is not required to make abstracts oncondition that a full reference to the source is shown. Single copies of any material in which the Institute holds copyright may be made for research or private

use purposes without reference to the SAIEE.

EDITORS AND REVIEWERSEDITOR-IN-CHIEFProf. B.M. Lacquet, Faculty of Engineering and the Built Environment, University of the Witwatersrand, Johannesburg, SA, [email protected]

MANAGING EDITORProf. S. Sinha, Faculty of Engineering and the Built Environment, University of Johannesburg, SA, [email protected]

SPECIALIST EDITORSCommunications and Signal Processing:Prof. L.P. Linde, Dept. of Electrical, Electronic & Computer Engineering, University of Pretoria, SA Prof. S. Maharaj, Dept. of Electrical, Electronic & Computer Engineering, University of Pretoria, SADr O. Holland, Centre for Telecommunications Research, London, UKProf. F. Takawira, School of Electrical and Information Engineering, University of the Witwatersrand, Johannesburg, SAProf. A.J. Han Vinck, University of Duisburg-Essen, GermanyDr E. Golovins, DCLF Laboratory, National Metrology Institute of South Africa (NMISA), Pretoria, SAComputer, Information Systems and Software Engineering:Dr M. Weststrate, Newco Holdings, Pretoria, SAProf. A. van der Merwe, Department of Infomatics, University of Pretoria, SA Dr C. van der Walt, Modelling and Digital Science, Council for Scientific and Industrial Research, Pretoria, SA.Prof. B. Dwolatzky, Joburg Centre for Software Engineering, University of the Witwatersrand, Johannesburg, SAControl and Automation:Dr B. Yuksel, Advanced Technology R&D Centre, Mitsubishi Electric Corporation, Japan Prof. T. van Niekerk, Dept. of Mechatronics,Nelson Mandela Metropolitan University, Port Elizabeth, SAElectromagnetics and Antennas:Prof. J.H. Cloete, Dept. of Electrical and Electronic Engineering, Stellenbosch University, SA Prof. T.J.O. Afullo, School of Electrical, Electronic and Computer Engineering, University of KwaZulu-Natal, Durban, SA Prof. R. Geschke, Dept. of Electrical and Electronic Engineering, University of Cape Town, SADr B. Jokanović, Institute of Physics, Belgrade, SerbiaElectron Devices and Circuits:Dr M. Božanić, Azoteq (Pty) Ltd, Pretoria, SAProf. M. du Plessis, Dept. of Electrical, Electronic & Computer Engineering, University of Pretoria, SADr D. Foty, Gilgamesh Associates, LLC, Vermont, USAEnergy and Power Systems:Prof. M. Delimar, Faculty of Electrical Engineering and Computing, University of Zagreb, Croatia Engineering and Technology Management:Prof. J-H. Pretorius, Faculty of Engineering and the Built Environment, University of Johannesburg, SAProf. L. Pretorius, Dept. of Engineering and Technology Management, University of Pretoria, SA

Engineering in Medicine and BiologyProf. J.J. Hanekom, Dept. of Electrical, Electronic & Computer Engineering, University of Pretoria, SA Prof. F. Rattay, Vienna University of Technology, AustriaProf. B. Bonham, University of California, San Francisco, USA

General Topics / Editors-at-large: Dr P.J. Cilliers, Hermanus Magnetic Observatory, Hermanus, SA Prof. M.A. van Wyk, School of Electrical and Information Engineering, University of the Witwatersrand, Johannesburg, SA

INTERNATIONAL PANEL OF REVIEWERSW. Boeck, Technical University of Munich, GermanyW.A. Brading, New ZealandProf. G. De Jager, Dept. of Electrical Engineering, University of Cape Town, SAProf. B. Downing, Dept. of Electrical Engineering, University of Cape Town, SADr W. Drury, Control Techniques Ltd, UKP.D. Evans, Dept. of Electrical, Electronic & Computer Engineering, The University of Birmingham, UKProf. J.A. Ferreira, Electrical Power Processing Unit, Delft University of Technology, The NetherlandsO. Flower, University of Warwick, UKProf. H.L. Hartnagel, Dept. of Electrical Engineering and Information Technology, Technical University of Darmstadt, GermanyC.F. Landy, Engineering Systems Inc., USAD.A. Marshall, ALSTOM T&D, FranceDr M.D. McCulloch, Dept. of Engineering Science, Oxford, UKProf. D.A. McNamara, University of Ottawa, CanadaM. Milner, Hugh MacMillan Rehabilitation Centre, CanadaProf. A. Petroianu, Dept. of Electrical Engineering, University of Cape Town, SAProf. K.F. Poole, Holcombe Dept. of Electrical and Computer Engineering, Clemson University, USAProf. J.P. Reynders, Dept. of Electrical & Information Engineering, University of the Witwatersrand, Johannesburg, SAI.S. Shaw, University of Johannesburg, SAH.W. van der Broeck, Phillips Forschungslabor Aachen, GermanyProf. P.W. van der Walt, Stellenbosch University, SAProf. J.D. van Wyk, Dept. of Electrical and Computer Engineering, Virginia Tech, USAR.T. Waters, UKT.J. Williams, Purdue University, USA

Published bySAIEE Publications (Pty) Ltd, PO Box 751253, Gardenview, 2047, Tel. (27-11) 487-3003, Fax. (27-11) 487-3002, E-mail: [email protected]

President: Dr Pat NaidooDeputy President: Mr André Hoffmann

Senior Vice President: Mr TC Madikane

Junior Vice President:Mr J Machinijke

Immediate Past President: Mr Paul van Niekerk

Honorary Vice President:Mr Mario Barbolini


VOL 105 No 4December 2014

SAIEE Africa Research Journal


December 2014 Volume 105 No. 4www.saiee.org.za

Africa Research JournalISSN 1991-1696

Research Journal of the South African Institute of Electrical EngineersIncorporating the SAIEE Transactions Condition Monitoring of Medium Voltage Electrical Cables

by Means of Partial Discharge MeasurementsH. van Jaarsveldt and R. Gouws ................................................. 136

Requirements Practitioner Behaviour in Social Context:A SurveyA. Marnewick, J-H. C. Pretorius and L. Pretorius...................... 147

Carrier Frequency Synchronization for OFDM-IDMA SystemsM.B. Balogun, O.O. Oyerinde and S.H. Mneney ........................ 156

A Framework for the Transmission of Multimedia Traffic using HM and RS Coding over Nakagami-M ChannelsT. Quazi, H. Xu and A. Saeed....................................................... 166

Landmine Detection by means of Ground Penetrating Radar:A Rule-Based ApproachP.A. van Vuuren ........................................................................... 175

SAIEE AFRICA RESEARCH JOURNAL EDITORIAL STAFF ...................... IFC


CONDITION MONITORING OF MEDIUM VOLTAGE ELECTRICAL CABLES BY MEANS OF PARTIAL DISCHARGE MEASUREMENTS H. van Jaarsveldt* and R. Gouws** School of Electrical, Electronic and Computer Engineering, North-West University, Private bag X6001, Post-point 288, Potchefstroom, South-Africa, 2520. *E-mail: [email protected] **E-mail: [email protected] Abstract: The purpose of this paper is to discuss condition monitoring (CM) of medium voltage electrical cables by means of partial discharge (PD) measurements. Electrical cables are exposed to a variety of operational and environmental stressors. The stressors will lead to the degradation of the cable’s insulation material and ultimately to cable failure. The premature failure of cables can cause blackouts and will have a significant effect on the safety of such a network. It is therefore crucial to constantly monitor the condition of electrical cables. The first part of this paper is focussed on fundamental theory concepts regarding CM of electrical cables as well as PD. The derivation of mathematical models for the simulation of PD is also discussed. The simulation of discharge activity is due to a single void within the insulation material of medium voltage cross-linked polyethylene (XLPE) cables. The simulations were performed in the MATLAB® Simulink® environment, in order to investigate the effects of a variety of parameters on the characteristics of the PD signal. A non-intrusive CM technique was designed for the detection of PD activity within cables. The CM technique was used to measure and analyse practical PD data. Two MATLAB® programs were designed to analyse the PD data in both the time-domain and frequency-domain. Keywords: Condition monitoring (CM), partial discharge (PD), cross-linked polyethylene (XLPE) insulation, MATLAB® Simulink®, time- and frequency domain.

1. INTRODUCTION

Electrical networks are interconnections of expensive electrical equipment. Electrical cables are one of the most important parts of any electrical network, as it is used for the transmission of electricity between the variety of electrical equipment. Electrical cables also form an integral part in the safety of any network, as failing cables can lead to the death of employees as well as the damage of expensive equipment. Different types of medium voltage electrical cables are available, each with their own advantages and disadvantages. The type of cable to use is often influenced by the purpose and the physical environment of the cable. Cables are exposed to a variety of environmental and operational stressors which can lead to the degradation of the insulation material and ultimately to cable failure. The level of aging and degradation of electrical cables can be evaluated by means of condition monitoring (CM) techniques [1]. Partial discharge (PD) is classified as both a symptom of degradation as well as a stress mechanism of degradation and will dominate as one of the main degradation mechanisms for medium voltage electrical cables [2]. Due to this, CM techniques, which make use PD measurements, are often used to monitor the condition of electrical cables. PD is the result of localised gasses breakdowns which will occur if the conditions are suitable. A discharge is classified as partial, if the discharge doesn’t completely bridge the insulation between two terminals [3].

Since the beginning of the 1990s most energy companies invested in the development of CM techniques. The process of performing CM on electrical equipment can be described as: “Condition monitoring can be defined as a technique or process of monitoring the operating characteristics of a machine in such a way that changes and trends of the monitored characteristics can be used to predict the need for maintenance before serious deterioration or breakdown occurs” [4]. CM techniques are used mainly due to the health of employees and the safe operation of electrical equipment. Any CM technique can be described as an interconnection of the following four main parts:

• Sensor. • Fault detection. • Data acquisition. • Diagnosis.

The choice of sensor is influenced by the physical environment of the equipment being monitored as well as the specific technique used to monitor the equipment. Sensors are used to convert a physical quantity to an electrical signal [4]. The main objective of any CM technique is to detect electrical faults. Fault detection can be described as the process of determining whether a fault is present in equipment being monitored. The data acquisition unit of a CM technique is the link between the measuring of data and the analysis of the measured data. The main purposes of a data acquisition unit are: to amplify the measured signals, the correction of sensor failures and in some cases the conversion of measured


data from analog to digital [4]. The diagnosis of measured data is used to identify trends as well as specific degradation mechanisms [1]. The interpretation of analysed data can be seen as the most important part of any CM technique and is generally performed by a specialist. The condition of electrical cables needs to be monitored on a regular basis in order to prevent the cables from failing prematurely and causing blackouts. The physical condition of an electrical cable can be determined by the observation, measurement and trending of condition indicators [1]. The ideal cable CM technique adheres to the following desired attributes of a CM technique [5]:

• Non-intrusive and non-destructive. • Applicable to cable types and materials

commonly used. • Affordable and easy to perform. • Provides trendable data. • Capable of measuring property changes that are

trendable and that can be correlated to functional performance during normal service.

• Able to identify the location of faults within the cables being tested.

• Able to detect aging and degradation before cable failure.

• Allows a well-defined end condition to be established.

It is impossible for a single technique to adhere to all of the above mentioned desired attributes. A single technique may therefore not be able to completely characterize the condition of an electrical cable. It is important to use a CM program with multiple CM techniques to successfully monitor electrical cables [5]. Electrical detection is a technique commonly used for the detection of PD within electrical cables. Electrical detection of PD can be divided into three groups [2]: 1) the measurement of individual discharge pulses, 2) measurement of total, integrated loss in the insulating material due to discharge activity and 3) the measurement of electromagnetic field effects associated with PD. The technique used to measure individual discharge pulses are commonly used as a CM technique for cables. This technique can be performed by means of a high frequency current transformer (HFCT) connected around the cable. The output of the CT is then measured by means of a digital oscilloscope or similar recording device. The use of a HFCT is a simple and affordable method to perform and also has the safety advantage of being isolated from the generally high operating voltages. The purpose of this paper is to discuss the design and implementation of a CM technique which make use of PD measurements. The CM technique is focussed on monitoring the condition of medium voltage cables.

2. DESIGN The purpose of this section is to discuss the design of a cable CM technique, which make use of PD measurements to analyse the condition of electrical cables. The IEEE Guide for Field Testing and Evaluation of the Insulation of Shielded Power Cable Systems will be consulted during the design process of the CM technique [6]. According to the guide PD data can be measured by means of special sensors connected to a splice or termination of the cable. The designed system makes use of a digital oscilloscope to store measured data obtained from the sensors. Designed MATLAB® programs were used to analyse the stored PD data in both the time- and frequency domain. Figure 1 shows an overview of the designed system.

Figure 1: Overview of designed CM technique

An HFCT is used as the sensor device for the detection of PD activity. The physical construction of the HFCT is in the shape of a horseshoe and will allow the user to connect and disconnect the sensor around the cable with ease. The use of an HFCT also isolates the user from the generally high operating voltages and there is also no need for disconnections of cables being tested. The HFCT will be connected directly to a digital oscilloscope. Figure 2 (a) illustrates the commonly used split-core HFCT [7]. An HFCT with the horseshoe structure is shown in Figure 2 (B). Figure 2 illustrates the physical differences between these types of sensors and from this it can be seen that the horseshoe clamp allows for the ease of connection as well as disconnection around cables.

Figure 2: Two HFCT models: (a) split-core [7] and (b)

Horseshoe


The system is designed to take PD measurements and store the data on a PC for analysis. The Handyscope HS5, a product of TiePie engineering, will be used as the data acquisition device. The need for a pre-amp is eliminated, due to the sufficient sensitivity of the HS5. The key specifications of the HS5 are shown in table 2. Table 2: Handyscope HS5 specifications Product Code HS5-220 Max. Sampling Speed 200 Ms/s Bandwidth 250 MHz Resolution 14bit (0.006%), (16bit enhanced) Memory 32 M Samples per channel Accuracy 0.25% DC vertical, 0.1% typical The PC is used as the device to control the system and also for storage of the recorded data. Two MATLAB® programs were designed to analyse the measured data in both the time- and frequency-domain. The time domain MATLAB® program is designed to display the measured signal in the time domain and to calculate the values for the maximum PD amplitude as well as the average PD amplitude. The time domain program is also used to determine the amount of discharges per phase and the exact position of each discharge within each phase. The results obtained from these two functions are used to perform phase resolved partial discharge (PRPD) analysis of the measured data. The functional flow diagram of these two functions is illustrated in Figure 3.

Figure 3: Functional flow diagram of source code

The functional flow illustrates that these two functions will analyse each data point individually. The stored data is in the form of a matrix, with each data point containing two elements: time and amplitude. The time element of each data point is used to determine the specific phase of

the measured discharge. The function to determine the amount of discharges per phase will only determine the amount of discharge for a specific phase, as the position of each pulse is not important for this function. PRPD analysis of PD data can be used to determine the source of the discharges. This is an essential part of any CM technique, as different sources may not affect the degradation of the insulation material at the same rate. A MATLAB® program was also designed for the purpose of analysing the measured data in the frequency domain. The transformation process is used to convert a signal from the time domain to the frequency domain and vice versa, this can also be seen as the most important concept of frequency domain analysis. A built-in fast Fourier transform (FFT) of MATLAB® is used for the transformation purposes of the CM technique. FFT functions are commonly used to convert time domain data to the frequency domain by computing the discrete Fourier transform (DFT) of the signal. The functional flow of a FFT function is shown in Figure 4.

Figure 4: Functional flow diagram of an FFT function

The first step of the FFT function is to decompose an N point signal into N single point signals. This is then used to determine the spectrum of each individual point. The final step of the FFT function is to synthesize the N frequency spectra into a single frequency spectrum. The frequency domain analysis of the measured data will be used to determine at which frequencies the discharge activity is more prolific. The FFT plot can also be used to determine the severity of PD activity as well as the location of the fault. 3. SIMULATED PARTIAL DISCHARGE RESULTS

3.1 Introduction The purpose of this section is to discuss the derivation of mathematical models for the simulation of PD. The development of electrical models to study the behaviour of PD activity within the insulation of electrical cables started in 1932, when Germant and Philippoff developed the simplest electrical representation of a defect within the insulation [8]. Research, over the years, lead to the improvement of this original model. The simulations discussed in this paper will be based on this well-known three-capacitor model. The first simulation model will be designed with only the basic parameters required to simulate the occurrence of discharge activity within the insulation material of an electrical cable. The simplicity of this model will cause a lack in accuracy and will also prevent the investigation of certain specific PD


characteristics. This led to the derivation of a more comprehensive simulation model. It is essential that both models are able to accurately simulate PD activity within medium voltage cables. The chosen parameters of the simulation models are therefore very important and careful consideration is taken before choosing the simulation parameters. Cross-linked polyethylene (XLPE) has become the globally preferred insulation material for medium voltage electrical cables. This is due to the many advantages of XLPE, including: improved impact strength, dimensional stability, tensile strength and an improved resistance to aging [9]. For this reason XLPE will be used as the insulation material for both simulation models. PD in electrical cables generally occurs due to cavities within the insulation material of the cable [10]. Both models simulate PD activity due to a single void within the insulation material of an electrical cable. The dimensions of the void also play an essential part in assuring the accuracy of the simulation models. Voids within XLPE insulation typically have a volume of 10 mm3, with a radius between 1 mm and 2 mm and a height between 2 mm and 3 mm [11]. This was considered when the dimensions of the voids, for both simulation models, were chosen. 3.2 Basic three-capacitor model The type of void will have a unique effect on the PD characteristics and will therefore be the main focus of the basic three-capacitor simulation model. The chosen void parameters are also the most important factors for PD characterization. A capacitor circuit was used to model the insulation material as well as the void for the basic simulation model. The capacitor configuration shown in Figure 5 illustrates the three “zones” of the simulation model, including: the void, the insulation in the vicinity of the void and the rest of the healthy insulation [12].

Figure 5: Capacitor configuration for basic model [12]

The derived capacitor circuit was used to design a simulation model in the Simulink® environment of MATALB®. The electrical circuit, on which the Simulink model is based, is illustrated in Figure 6. The main component of the simulation model is the capacitor configuration consisting of the three capacitors, (𝐶𝐶!), (𝐶𝐶!) and (𝐶𝐶!). The parallel RLC branch in the circuit will act as the input impedance to which the measuring equipment will be connected. A high voltage filter (𝑍𝑍) is

added to the circuit for the reduction of background noise.

Figure 6: Derived circuit for the basic simulation model

The equations for the three capacitors of the basic simulation model will only be influenced by the permittivity of the insulation material (𝜀𝜀!), the permittivity of free space (𝜀𝜀!) and the dimensions of the void. The capacitance value of the capacitor associated with the void is given by [12]: 𝐶𝐶! =

𝜀𝜀!×𝑟𝑟!×𝜋𝜋ℎ

(1)

The capacitance values for the healthy part of the insulation and the part of the insulation material in the vicinity of the void is calculated by means of the following two equations [12]: 𝐶𝐶! =

𝜀𝜀!×𝜀𝜀!× 𝑎𝑎 − 2𝑏𝑏 ×𝑏𝑏𝑐𝑐

(2)

𝐶𝐶! =

𝜀𝜀!×𝜀𝜀!×𝑟𝑟!×𝜋𝜋ℎ

(3)

These three equations were used to determine the values for the capacitors required for the simulation of PD by means of the basic three-capacitor simulation model. The first set of simulations was performed in order to study the effect of the input voltage on the maximum amplitude of the measured PD signal. The input voltage was incremented from 7 kV to 50 kV with a constant void size. The results from these simulations are depicted in the graph shown in Figure 7.

Figure 7: Effect of input voltage on the maximum PD amplitude


From the graph it can be seen that the resulting PD magnitudes, for input voltage between 10 kV and 30 kV, varies between 2 μV and 5 μV. An unusual high maximum PD amplitude is present at an input voltage of 7 kV. The conclusion can be made that the maximum PD amplitude will remain relatively constant for a range of input values. As the input voltage is increased beyond the 30 kV mark, the maximum amplitude also increases. It can therefore be said that an increase in input voltage will result in increased maximum amplitudes of the measured PD signal. The correlation between the void size and the determined apparent charge was also investigated by means of a set of simulations. The void dimensions included a constant height of 3 mm and the radius of the void was incremented in sizes of 0.5 mm from 1 mm to 5 mm. The input voltage was kept at a constant value for all of the simulations. The change in void parameters for each simulation will result in different capacitance values of the test object. Equations (1), (2) and (3), were used to determine the capacitors of the test object for each of the simulations. The results of this set of simulations are shown in Figure 8.

Figure 8: Correlation between void size and apparent charge

A study of the results shown in the graph led to the conclusion that, the correlation between void size and apparent charge is almost linear. A trendline was fitted to the data to be able to determine the coefficient of determination (R2). The determined coefficient of determination was R2 = 0.9656. The R2 value indicates that the trendline is a close representation of the actual data. From this the conclusion can be made that a larger void volume will result in an almost linear increase in calculated apparent charge values; this will then result in more severe PD activity within the cable. The equation of the trendline, y = 28.046x – 40.511, can be used with the calculated apparent charge value to determine an estimate void size. The basic three-capacitor simulation model was also used to study the effect of the input frequency on the apparent charge of the PD signal. The results are shown in Figure 9.

Figure 9: The effect of input frequency on the apparent charge

Published research showed that there is no trend in the effect of input frequencies below 50 Hz on PD characteristics [13]. Input frequencies between 50 Hz and 400 Hz were therefore used for this set of simulations. From the graph it is clear that the input frequency has a definite effect on the apparent charge of the measured signal. As the input frequency is increased above the normal operating frequency of 50 Hz, the apparent charge will decrease. The input frequency will also have an effect on other PD characteristics, such as: PD inception voltage, PD signal magnitude and rise and fall times [13]. 3.3 Comprehensive simulation model The comprehensive simulation model is an improvement of the basic simulation model and will therefore be based on the same principles. The improvements of the comprehensive model resulted in a more complex simulation model. Figure 11 shows the cross sectional view of the basic structure of a medium voltage electrical cable [14].

Figure 11: Capacitor configuration comprehensive simulation

model [14] The model shown in Figure 11 was used to derive an electrical circuit which was used to perform the simulations associated with the comprehensive three-capacitor model.


The electrical circuit which was used to create the Simulink® model for the comprehensive simulations is shown in Figure 12.

Figure 12: Derived circuit for the basic three-capacitor

simulation model It is clear that the circuit used for the comprehensive simulations has distinct similarities to the circuit used for the basic simulations. The void capacitor is denoted by (𝐶𝐶!), the capacitor of the insulation in the vicinity of the void by (𝐶𝐶!) and the capacitor of the healthy insulation by (𝐶𝐶!). The improvements of the comprehensive model led to the addition of resistors associated with each of the capacitors of the test object. The void resistor is given by (𝑅𝑅!), the resistor for the insulation in the vicinity of the void by (𝑅𝑅!) and the resistor for the rest of the healthy insulation by (𝑅𝑅!). The streamer resistance of the circuit is denoted by (𝑅𝑅!"#). A voltage controlled switch was used to initiate the occurrence of PD activity for this simulation model. The insulation in the vicinity of the void is divided into two parts, “a” and “b”, the capacitor values for both parts are calculated by means of the following equations [14]:

𝐶𝐶! =

𝜀𝜀!"#ln 𝑟𝑟!"#$ + 𝑑𝑑! −

𝑟𝑟!2 𝑟𝑟!"#$

. 𝑙𝑙!

(𝑟𝑟!"#$ + 𝑑𝑑!) . 𝑡𝑡!

(4)

𝐶𝐶!=

𝜀𝜀!"#ln ( 𝑟𝑟!"#$ + 𝑑𝑑!"# (𝑟𝑟!"#$ + 𝑑𝑑! +

𝑟𝑟!2))

.𝑙𝑙!

(𝑟𝑟!"#$ + 𝑑𝑑!) . 𝑡𝑡!

(5)

The capacitance value for the insulation in the vicinity of the void is calculated by means of equation (4) and equation (5) as the equivalent series capacitance value of (𝐶𝐶!) and (𝐶𝐶!): 𝐶𝐶! =

𝐶𝐶!𝐶𝐶!𝐶𝐶! + 𝐶𝐶!

(6)

The resistance of part “a” and part “b” of the insulation can be determined by means of the following derived equations [14]:

𝑅𝑅! =

1𝜎𝜎!"#

.𝑟𝑟!"#$ + 𝑑𝑑!

𝑙𝑙!𝑡𝑡!

. ln 𝑟𝑟!"#$ + 𝑑𝑑! − 𝑟𝑟! 2

𝑟𝑟!"#$

(7)

𝑅𝑅! =

1𝜎𝜎!"#

.𝑟𝑟!"#$ + 𝑑𝑑!

𝑙𝑙!𝑡𝑡!

. ln 𝑟𝑟!"#$ + 𝑑𝑑!"#

𝑟𝑟!"#$ + 𝑑𝑑! + 𝑟𝑟! 2

(8)

The resistance value for the insulation material in the vicinity of the void is found by means of the equivalent series resistance value of (𝑅𝑅!) and (𝑅𝑅!): 𝑅𝑅! = 𝑅𝑅! + 𝑅𝑅! (9)

The capacitance value for the rest of the healthy insulation is calculated by means of the following equation [14]: 𝐶𝐶! =

2𝜋𝜋𝜀𝜀!"#ln 𝑟𝑟!"#$ + 𝑑𝑑!"# 𝑟𝑟!"#$

. 𝐿𝐿 (10)

The resistance value for this part of the insulation can be found by means of similar assumptions used to determine the value of (𝐶𝐶!) [14]: 𝑅𝑅! =

12𝜋𝜋𝜋𝜋𝜎𝜎!"#

. 𝑙𝑙𝑙𝑙𝑟𝑟!"#$ + 𝑑𝑑!"#

𝑟𝑟!"#$ (11)

The capacitance value of the capacitor associated with the void is determined with the following equation [14]:

𝐶𝐶!=

𝜀𝜀!ln ( 𝑟𝑟!"#$ + 𝑑𝑑! +

𝑟𝑟!2 (𝑟𝑟!"#$ + 𝑑𝑑! −

𝑟𝑟!2))

.

𝑙𝑙!

(𝑟𝑟!"#$ + 𝑑𝑑!) . 𝑡𝑡!

(12)

The resistance value of the resistor associated with the void is then determined by means of:

𝑅𝑅! =1𝜎𝜎! 𝑟𝑟!"#$ + 𝑑𝑑!

𝑙𝑙!𝑡𝑡! ln

𝑟𝑟!"#$ + 𝑑𝑑! + 𝑟𝑟! 2𝑟𝑟!"#$ + 𝑑𝑑! − 𝑟𝑟! 2

(13)

The first set of simulations performed by means of the comprehensive simulation model was used to study the effect of the input voltage on the apparent charge of the PD signal. For this set of simulations the void size was kept constant and the input voltage was incremented by 0.5 kV from 6.6 kV to the final value of 11 kV. The results are shown in Figure 13


Figure 13: Correlation between input voltage and apparent

charge From Figure 13 it can be seen that an increase in input voltage will result in increased values for the apparent charge. There is however no direct trend between these two parameter. The basic simulation model as well as published work from other researchers yielded similar results. The comprehensive model can be used to study the effect of both void size and position on the apparent charge. Three different void positions with the exact same void volume were used for this investigation. The three positions included: 1) a void in the middle of the insulation, 2) a void close to the conductor as well as 3) a void near the sheath of the cable. The following equations were used to calculate the position of each void [14]: Close to the conductor: 𝑑𝑑! = 0.15𝑑𝑑!"# (14)

Middle of insulation: 𝑑𝑑! = 0.5𝑑𝑑!"# (15)

Near the sheath: 𝑑𝑑! = 0.85𝑑𝑑!"# (16)

The length of the void 𝑙𝑙! was incremented from 1 mm to 5 mm while both the height 𝑟𝑟! and width 𝑡𝑡! were kept constant. The simulations for all three the void positions were conducted and the results obtained from each set of simulations were used to draw a graph to show the effect of void volume and void positions on the same graph. The results are shown in Figure 14.

Figure 14: Effect of void parameters on the apparent charge

The graphs in Figure 14 show that an increased void volume will result in an increased value for the calculated apparent charge of the PD signal. The results obtained by means of the basic simulation model were similar to the results obtained from the comprehensive model. The coefficients of determination for the three positions are: middle of the insulation R2 = 0.9983, close to the conductor R2 = 0.9943 and near the outer sheath of the cable R2 = 0.9958. It is thus confirmed that an increased void volume will result in an almost linear increase in apparent charge of the PD signal. The position of the void has an influence on the amplitude of the apparent charge. The simulations discussed in this paper were used to investigate the effect of a variety of parameters on PD characteristics. The results obtained from both the basic and the comprehensive simulation models show similarities when compared. Results obtained from published research were also compared to the simulation results discussed in this paper. The similarities between results from published research and the results discussed in this paper verify the validity of the simulation models.

4. EXPERIMENTAL SETUP AND RESULTS 4.1 Introduction The designed CM technique discussed in Section 2 of this paper was used to measure and analyse practical PD data. The experimental setup as well as the analysis of the measured data will be discussed in this section. The purpose of the designed CM technique is only to detect the presence of discharge activity within medium voltage cables and to analyse the measured PD data to some extent. It is necessary to utilize a number of CM techniques as well as the expertise of a specialist for the complete analysis of PD data. The complete CM technique can be divided into a number of stages which must be completed in order to be able to detect the presence of PD within the insulation material of electrical cables. A functional flow diagram of the required steps is illustrated in Figure 15.

Figure 15: Flow diagram of the stages for the CM technique


The first stage of the CM technique includes the experimental setup of measuring equipment and the test specimen. The next stage is to perform accurate PD measurements on the test specimen. The data obtained from the measurements is then stored for analysis. MATLAB® programs are used to analyse the PD data in the time- and frequency-domain. The final step of the CM technique is to interpret the analysed data. This is the most important step of the entire process and is also the part where the expertise of a specialist is required for the correct interpretation of analysed PD data. 4.2 Experimental setup The experimental setup of the CM technique includes the combination of measuring equipment and test specimen required for the measuring of practical PD data. The experimental setup of the measuring equipment is shown in Figure 16.

Figure 16: Experimental setup of measuring equipment

The measuring equipment includes a PC, used to control the system as well as to act as storage device for the measured data, and the digital oscilloscope which acts as the data acquisition unit of the CM technique. A cable termination was used as the test specimen for the measurement of practical PD data. The cable termination was taken out of operation due to the fact that severe levels of discharge activity were detected within this section of the cable. The test specimen is shown in Figure 17.

Figure 17: Cable termination

The connection of an HFCT around the cable can be seen in Figure 17. The cable termination is connected to a high voltage variac for the supply of variable input voltages. Sets of measurements were taken at input voltages of 1.6 kV, 3.2 kV and 5 kV. The different levels of input voltage facilitate the study of the effect of input voltage on PD characteristics. 5.3 Analysis of PD data The process of measuring PD data can be seen as a relatively easy process to perform. The analysis of the measured data is however far more difficult. Modern CM techniques can be used to analyse measured PD data to some extent, but CM techniques are still dependent on the expertise of a specialist for successful analysis of the condition of electrical equipment. The HS5 scope was used to store the measured data in “.mat” files, which can be accessed by the designed MATLAB® programs. The time domain analysis of the measured data includes: the study of the PD signal over time, the calculation of maximum and average PD amplitude, the calculation of the amount of discharges per phase as well as the ability to indicate the phase patterns of discharges. The time domain program was used to store and analyse data for three cycles (60 ms) of the input voltage. The graphical user interface (GUI) for the designed time domain MATALAB® program is shown in Figure 18.

Figure 18: Time domain GUI of MATLAB program

The function to determine the number of discharges per phase was used to analyse the measured data. The results obtained from this function are given in Table 2. Table 2: Amount of PD pulses per phase Input voltage (kV)

First phase

Second phase

Third phase

Fourth phase

1.6 93 100 78 112 3.2 140 189 137 214 5 274 356 218 365 The numerical values were also used to construct a histogram of the results in order to be able to study the correlation between the input voltage and the number of discharges per phase. The histogram is shown in Figure 19.


Figure 19: Histogram of amount of PD pulses per phase

From this graph it can be seen that a common trend for all the input voltages are present. The conclusion was made that an increased input voltage will result in increased amounts of discharges per phase. This trend is also present in the results obtained from the simulation models. The graph shows a significant increase in the number of discharges per phase at the 5 kV input voltage. This can be a result of the level of degradation of the cable being tested. When the 5 kV level is reached the input voltage is at such a level that PD activity reaches a critical level and will therefore cause a significant increase in various PD characteristics. The phase pattern of the measured PD data is shown in Figure 20. The obtained PD patterns are compared to characteristic PD patterns in order to perform PRPD analysis of the measured data.

Figure 20: Phase plot of PD patterns at 1.6 kV

The plot shown in Figure 20 was compared to published research as well as characteristic PD patterns to be able to determine the types of discharges present within the test specimen. By means of comparison to the work of Hudon and Bélec [15], it was determined that both internal PD as well as exciter pulses is present within the test specimen. The measurements were taken up to a maximum bandwidth of 50 MHz; it is possible that other PD patterns may be visible if data samples are taken above the 50 MHz bandwidth. PD patterns may be clearer if

more samples are taken at each measurement. The reason for this is that insufficient samples may not be enough to reveal a recognizable pattern. The most important part of the frequency domain analysis is the ability to perform FFT analysis of the recorded time domain signals. The GUI for the frequency domain MATALAB® program is illustrated in Figure 21

Figure 21: Frequency domain GUI of the MATLAB program

The FFT function is an algorithm which is used to compute the DFT of time domain data. The DFT is a useful operation, but computing it directly from the definition can be a time consuming operation. An FFT is used to compute the same result, but with improved speed. The FFT plot of the measured signal, at 1.6 kV, is shown in Figure 22.

Figure 22: FFT plot at 1.6 kV

Frequency analysis of PD data associated with electrical cables is often used to determine the location of the discharge activity. The FFT plot shown in Figure 22 shows that the discharges are spread over the entire frequency range. A large portion of the discharges are however present between 15 MHz and 35 MHz. The fact that discharges are present in large portions of the measured frequency range testifies to the severity of the measured discharge activity. If the severity of PD activity within a power cable increases to reach a high level, the pulses shown in the FFT plot will continue to grow until most of the frequency range is covered.


5. CONCLUSION A CM technique was designed which can be used to monitor the condition of electrical cables by means of PD measurements. The CM technique consists of: a sensor, a data acquisition unit and a PC. The type of sensor used for this technique is an HFCT with a physical structure representing a horseshoe. The use of an HFCT as the sensor ensures the safety of this technique, due to the fact that the measuring equipment is isolated from the typically high operating voltages. The data acquisition unit of this technique was a digital oscilloscope. The Handyscope HS5 was chosen due to a number of advantages of this scope. The final component of the CM technique is the PC. The PC is used to control the system and also to act as storage device for measured PD data. Two MATLAB® programs were designed to perform the analysis of measured PD data in both the time- and frequency-domain. Simulation models were used to simulate the occurrence of PD activity within medium voltage electrical cables. The PD activity is due to a single void within the insulation material of the cable. Research indicated that the well-known three-capacitor model is commonly used to derive mathematical models for the representation of PD activity. The basic three-capacitor simulation model was derived with only the basic required parameters to be able to simulate PD activity. The simplistic approach of this model led to the inability to investigate the effects of a variety of parameters on the PD characteristics. This and reduced accuracy were the main reasons for the development of a more comprehensive model. The comprehensive model is able to investigate a number of crucial PD parameters. The comprehensive model also includes the stochastic nature of PD. The results obtained from the simulations discussed in this paper were used to gain knowledge regarding the phenomenon known as PD. The designed CM technique was used to measure practical PD data. The test specimen used for the experimental results was a cable termination, taken out of operation due to the fact that discharge activity was previously detected in this component. The measured data was analysed in the time domain in order to determine: the maximum and average amplitudes of the recorded signal, the amount of discharges per phase as well as to plot the phase patterns of the PD pulses. The phase patterns are used to perform PRPD analysis of the measured PD data. This is done by comparing the obtained patterns to characteristic PD patterns as well as patterns of published research. The frequency domain program is used to analyse the measured time domain data in the frequency domain. An FFT function is used to convert the time domain data to the frequency domain. The FFT plots of the measured data were used to investigate: the frequencies at which the discharge activity is most prolific and the severity of the discharge activity.

6. REFERENCES [1] M. Villaran and R. Lofaro, Essential Elements of an

Electric Cable Condition Monitoring Program, 1st ed. Upton, NY: Brookhaven National Laboratory, 2010.

[2] A. Haddad and D. Warne, Advances in High Voltage

Engineering. London, UK: Institution of Electrical Engineering, London, 2004, ch. 4, pp. 139-185

[3] S. Chaikin, "Partial Discharge Testing - Decreasing

the Field Failures of High Voltage Components," HT LLC, pp. 1-18, March 2001.

[4] Y. Song and Y.H. Han, "Condition Monitoring

Techniques for Electrical Equipment - A Literature Survey," IEEE Transactions on Power Delivery, vol. 18, no. 1, pp. 4 - 13, January 2003.

[5] Brookhaven National Laboratory, "Assessment of

Environmental Qualification Practices and Condition Monitoring Techniques for Low-Voltage Electric Cables," vol. 2, no. 1, February 2001.

[6] IEEE Standards, "IEEE Guide for Field Testing and

Evaluation of the Insulation of Shielded Power Cable Systems," IEEE Power Engineering Society, New York, NY, Standard 400, 2001.

[7] Techimp. (2013, June) Techimp Systems. [Online].

http://www.techimp.com/systems/Products/Sensors/PDSensors.aspx

[8] D.A. Nattrass, "Partial discharge XVII The early

history of partial discharge research," IEEE Electrical Insulation Magazine, vol. 9, no. 4, 1993.

[9] I. A. Metwally, "High-voltage power cables plug into

the future," Potentials, IEEE, vol. 27, no. 1, pp. 18 - 25, Jan - Feb 2008.

[10] N. Ahmed and N. Srinivas, "Partial Discharge

Severity Assessment in Cable Systems," in Transmission and Distribution Conference and Exposition, 2001, pp. 849 - 852.

[11] C.Y. Ren, Y.H. Cheng, P. Yan, Y.H. Sun, and T.

Shao, "Simulation of Partial Discharge in Single and Double Voids Using SIMULINK," in Twenty-Seventh International Conference Record of the Power Modulator Symposium, May 2006, pp. 120-123.


[12] A. Sabat and S. Karmakar, "Simulation of Partial Discharge in High Voltage Power Equipment," International Journal on Electrical Engineering and Informatics, vol. 3, no. 1, pp. 234-247, 2011.

[13] C. Nyamupangedengu and I.R. Jandrell, "Partial

Discharge Spectral Response to Variations in the Supply Voltage Frequency," IEEE Transactions on Dielectrics and Electrical Insulation, vol. 19, no. 2, pp. 521 -531, April 2012.

[14] F Haghjoo, E Khanahmadloo, and S.M. Shahrtash, "Comprehensive 3-capacitor model for partial discharge in power cables," COMPEL: The International Journal for Computation and Mathematics in Electrical and Electronic Engineering, vol. 31, no. 2, pp. 346 - 368, 2012.

[15] C. Hudon and M. Bélec, "The Importance of Phase

Resolved Partial Discharge Pattern Recognition for On-Line Generator Monitoring," in IEEE International Symposium on Electrical Insulation, Arlignton, Virginia, USA, June 1998, pp. 296 - 300.


REQUIREMENTS PRACTITIONER BEHAVIOUR IN SOCIAL CONTEXT: A SURVEY

A Marnewick*, JHC Pretorius** and L Pretorius***

* Faculty of Engineering and the Built Environment, University of Johannesburg, Corner Kingsway and University Road, Auckland Park, 2006, South Africa. E-mail: [email protected] ** Faculty of Engineering and the Built Environment, University of Johannesburg, Corner Kingsway and University Road, Auckland Park, 2006, South Africa. E-mail: [email protected] *** Department of Engineering and Technology Management, Graduate School of Technology Management, University of Pretoria, Lynnwood Road, Pretoria, 0002, South Africa. E-mail: [email protected]

Abstract: The purpose of this research paper is to discover the social behaviour of practitioners that causes the communication breakdowns during the requirements engineering process. Requirements emerge from the social interaction and communication between the requirements practitioner and the various stakeholders. The main problems with the requirements engineering process are communication and coordination breakdowns, as well as the lack of domain knowledge or understanding of the problem. These challenges are all related to the social interaction during the requirements engineering process that impacts the quality of requirements. In practice, requirements are still produced with errors which then lead to unsuccessful solutions to problems. The ultimate goal of any practitioner is delivering a solution fit for purpose first time around. If the social patterns of practitioners that deliver quality requirements are known and compared with those that do not deliver quality requirements, individual performance can be adjusted. The results of this study confirmed that quality of requirements is dependent on the communication established between the requirements practitioner and relevant stakeholders. The communication is enabled through the trust relationships between the parties. A description of how practitioners behave during the requirements process is provided. By discovering these interaction patterns, communication can be improved and made more effective. Additionally, the relationships between the practitioners and their stakeholders are described. These trust patterns provide insight into the levels of collaboration, communication and sharing of knowledge between the practitioners and their stakeholders. By identifying these relationship patterns, the value each party receives from the relationships could increase, and the communication breakdowns could be minimised. Keywords: Requirements, requirements engineering process, survey, communication, trust, problem solving

1. INTRODUCTION Although the research community acknowledges the importance of requirements, the industry still faces many challenges in practice with the requirements engineering process during solution delivery [1]. Industry reports, surveys and research continuously identify poor requirements as the main contributor to failed projects [2-7]. Requirements engineering is about understanding the problem, and then solving it during a systems engineering, software engineering and business analysis process. Understanding problems involves human interaction [8-10]. A purely technical process cannot solve a problem where customers’ needs are to be met [11]. The successes of an organisation, team or project depend on the knowledge in the organisation and are represented by the relationships between the people and the organisation [12]. If the requirements practitioner does not enable collaboration with stakeholders through relationships, knowledge of the problem could be incomplete and impact the quality of requirements [13-15]. The challenges experienced during the requirements engineering process relate to social interaction challenges, specifically communication breakdowns, and not necessarily technical challenges [11, 16].

Communication between the stakeholders and the requirements practitioners is highlighted by many researchers as one of the important challenges faced in practice in delivering accurate requirements [1, 6, 17-21]. The context in which the requirements engineering process is performed is a human activity system. Requirements emerge from the social interaction and communication between the stakeholders and the requirements practitioner [22]. The importance of effective communication to ensure a mutual understanding between the cross-functional team involved in the requirements engineering process has been emphasised over time [15, 23-30]. Collaboration and negotiation between the requirements practitioner and stakeholders provide the building blocks for strong relationships that allow for an interactive and involved requirements engineering process [29, 31]. Paech et al. [32] have summarised studies reviewed in literature to confirm what the problems with the requirements engineering process are: communication and coordination breakdowns lack of domain knowledge, i.e. understanding the

problem in the world changing and conflicting requirements. The requirements engineering studies consistently raise human interaction, in the form of communication,


collaboration or relationships, as a challenge. The focus of this research was therefore not to identify the challenges experienced during the requirements engineering process but rather to identify the social behaviour that causes the communication breakdowns. The aim of this research was to identify the social behaviour during practitioner communication, as communication is one of the challenges that have an impact on the quality of requirements. The study revealed the practitioners’ communication patterns during the requirements engineering process, and what relationships they establish with various stakeholders. A two-part survey was conducted. The first part of the study was to describe how the requirements engineering process is executed by practitioners and to establish what is used and what is not used in practice. The second part of the study was to enhance the understanding of the social context within which the practitioners execute the requirements engineering process. This part of the survey was undertaken to derive a description of how the practitioner behaves during the requirements engineering process in order to identify the communication and trust patterns between the practitioner and the various stakeholders. Trust has been identified as one of seven success factors influencing the delivery of requirements [33]. It was decided to explore the trust relationships between the requirements practitioners and determine whether these relationships have any impact on the quality of the requirements. According to social studies, if trust is not present in a relationship, effective communication will not take place [34]. This paper’s main objective is to document the communication patterns of practitioners and their consequences. By discovering these patterns the knowledge can be utilised to improve communication during the requirements engineering process. If communication is more effective, the outcome of the requirements engineering process, which is quality requirements, should improve. The first section of this paper provides a summary of the research design. Secondly, the results collected are summarised and discussed to derive patterns on how practitioners behave during the requirements engineering process. Finally, key patterns are highlighted.

2. RESEARCH METHOD Social behaviour could not be researched in isolation, but within the context of how the requirements engineering process is executed by practitioners. A research method was required to derive an understanding of what activities are generally performed during the requirements process and how many of the tools and methodologies available in literature are in use or known by practitioners. Secondly, the data had to provide insight into the level of

communication during the process and the impact of the collaboration on the quality of delivered requirements. A survey was selected as the most appropriate research method to collect information about behaviours, knowledge, opinions, or attitudes of the requirements participants in order to derive a description of the requirements engineering process and the practitioners’ behaviour. To document how the practitioners execute the requirements engineering process, data was collected from the first two sections in the survey. The final section of the survey collected information to derive a description of the social factors that impact on the requirements process outputs. This data was used to describe the communication patterns and trust relationships between the responsible person generating requirements and the stakeholders. The focus of the data collected was on communication patterns and trust relationships between stakeholders and the requirements engineer. Communication is a challenge faced across communities in which the requirements discipline is executed; it is also identified as one of the main impacts on the quality of requirements. The trust relationships between the requirements practitioners and stakeholders were explored to determine whether these relationships have any impact on the quality of the requirements or on the communication. Additionally, as requirements engineering is defined as a problem-solving process, it was decided to explore the sources practitioners consult during this process. The patterns explored are summarised in Table 1.

Table 1: Behaviour patterns

Pattern Behaviour analysed Communication Communicates effectively through personal

interaction. Communication models used to facilitate

communication during requirements engineering process.

Motivation for communication interaction. Trust Reliance-based trust: determine if the

practitioners were prepared to rely on others based upon (i) relying on another’s skill, (ii) knowledge, (iii) judgement, (iv) actions including delegating and (v) giving autonomy.

Disclosure-based trust: determine if the practitioners were prepared to share work-related or personal information that was of a sensitive nature.

Problem solving

Information-seeking behaviour of the respondents during the problem-solving activity, sources of information consulted were investigated based on sourced suggested by literature [35].

As these are all social patterns that were to be observed, existing instruments were utilised where possible. During a project, requirements engineering bridges the communication between the stakeholders and the technical delivery team [36]. Based on this, information regarding the following communication patterns was collected:


Communication with project manager on project impacts.

Communication with sponsor to elicit business requirements.

Communication with a subject matter expert to acquire domain knowledge.

Communication with users to elicit functional requirements.

Literature was researched to identify the most appropriate possible instrument available to measure trust in the workplace. The focus of the trust relationships was intra-organisational relationships among co-workers [37]. It was decided to use the behavioural trust inventory developed by Gillespie [38]. This trust measurement was selected because it has been validated by other researchers and was classified as a best fit instrument to capture trusting behaviours [39]. The behavioural trust inventory measures the work relationships between employees and immediate managers, as well as between employees and immediate work colleagues [37]. Finally, to identify the research pattern of typical practitioners in understanding the problem to be solved, sources of information consulted were established as prescribed by literature [35, 40]. The following section provides a summary of the main observations.

3. SURVEY RESULTS SUMMARY

A total of 127 responses were received from requirements practitioners within the South African community. The survey took between 20 and 30 minutes to complete, which is 5 to 10 minutes longer than the suggested time by Rea and Parker, which suggest that web-based surveys should aim at completion within 15 minutes [41]. Two alternatives were considered: remove questions or separate the questionnaire into subsets [42]. The latter choice was elected. The first subsets of questions dealt with the requirements process to derive a description document of how the practitioners execute the requirements engineering process. The second subset was more explorative of the social interaction patterns. The survey respondents were mainly from the finance and banking (33%), information and communication technology (23%) or government public sector and defence industries (9%). Although the majority of respondents were from these industries, the data confirms a presence of requirements practitioners across all industries. The behaviour review provided a description of the practitioners’ communication behaviour and the trust relationships they established during the requirements engineering process with various stakeholders. It identified the communication model used by practitioners during the execution of the requirements activities. It

also indicated the sources practitioners consulted during the problem-solving process. This provided the social context during the requirements process within which the practitioners operated. A summary of the patterns analysed during this review is presented in Table 2.

Table 2: Behaviour pattern results summary

Patterns Actual behaviour Communicates effectively through personal interaction

In 93.7% of the instances the practitioners established communication with the project manager, and in 82.5% of the cases with the end-users. Communication was established with the subject matter expert in 77.6% of the cases.

Practitioners only established communication with the project sponsors in 61.7% of the cases. They indicated that they did not need to communicate with the project sponsor as this was the responsibility of the project manager.

Communication models used to facilitate the communication during requirements engineering process

The communication direction between the practitioners and project manager indicates that both parties initiated communication interaction, with the project manager initiating communication 41.1% of the time and the practitioner 58.9% of the time. The message flow was multidirectional.

In the communication between the practitioner the project sponsor, subject matter expert and the end-user the practitioner was initiating the communication. The message flow was only a one-directional message flow. In 75% of the cases the communication interactions between the practitioner and the project sponsor were initiated by the practitioner. In 95.6% of the cases the communication interactions between the practitioner and the subject matter expert were initiated by the practitioner, and in 93.5% of the cases between the practitioner and the end-user it was initiated by the practitioner.

Motivation for communication interaction

Practitioners reported that the main motivation for interaction with the project manager was to report on the progress of the analysis effort (64%). Other reasons provided for communicating with project managers were: To raise or identify risks (14%), agree on

analysis approach for project (7%), requirement changes (7%), input for project costs, resource requirements and delivery (5%) and project status reports (3%).

The main reason provided for interaction with the project sponsor was to communicate the expectations and feedback from other senior managers and stakeholders (52.8%). Practitioners appeared to spend time with the project sponsor to ensure that the solution was understood (19.4%) as well as to resolve conflicts and political issues (19.4%). Additional reasons noted were status and progress reporting (3.2%) as well as to raise stakeholder concerns (0.8%). No response was noted by 4.4%.

The main reason for interacting with the subject matter expert was to gather knowledge about the domain in which the application/solution would operate (44.4%). Other reasons provided for communicating with the subject matter expert were: To gain an understanding of the

underlying structure of domain problems (20%), gather knowledge about the activities performed in this domain


(28.9%) and resolve conflicting issues (3.1%). No reason was provided in 3.6% of the responses.

The two main reasons given for the interaction with the end-user were to interview the end-user about the functional requirements (58.7%) or to observe the user while executing the daily activities to understand the domain (30.4%). Only a very small percentage of the respondents (0.8%) spent time with the end-users to validate their understanding and discuss issues that were experienced (0.8%). 1.6% mentioned progress reporting as a reason and 7.7% did not provide any reason.

Trust relationships

The trust relationships between the practitioners and stakeholders indicated the trust is based on willingness to rely. The willingness to disclose information in all the relationships between the practitioner and the relevant stakeholder was low.

Problem solving

The majority of practitioners depended on sources of information such as either personal experience (15%) or conversations with customers (16%) and colleagues (17%).

Fewer practitioners consulted standards (10%) and technical reports (9%) and had conversations with consultants (10%) or with vendors (8%).

Academic, industry and textbook sources were used in a few cases (4%, 5% and 6%, respectively).

The communication behaviour, trust relationships and problem-solving behaviour patterns determined from the survey are discussed in detail in the next section. Additional analysis results to understand any impact on the quality of the requirements delivered by the practitioners are also included in the discussion.

4. SURVEY RESULTS DISCUSSION

Pattern 1 - Communicates effectively through personal interaction This communication pattern highlighted that practitioners did not feel obliged to communicate with the project sponsor. They indicated that this was the project manager’s responsibility. Although the project manager is accountable for the communication management, the requirements practitioner is responsible for the communication management plan [43]. The requirements practitioner still has a direct responsibility to communicate with all relevant stakeholders within the business domain regarding the impact of the solution [43]. Literature constantly identifies communication and its impact as a challenge during the requirements engineering process [17-19, 32]. Further analysis was done to determine whether the communication interaction had any impact on the quality of the requirements delivered by the practitioners. The scale elements used to determine the quality of the requirements during the survey were based on the eight characteristics of a quality specification. These characteristics were traceable, modifiable, verifiable, ranked, consistent, complete, unambiguous and correct as defined by existing standards

[44-46]. The respondents had to rate how frequently the requirements contained the eight quality characteristics. To simplify the analysis, exploratory factor analysis was used as a data reduction technique to validate if the eight quality elements could be combined into fewer factors. Initial factor analysis confirmed that it would not be ideal to combine the eight elements. However, confirmatory factor analysis was done to validate the underlying structure of the quality scale and results suggested that the elements “consistent” and “traceable” should be excluded. By adjusting the quality scale the factor analysis confirmed that the six other elements could be summarised into one factor. A summary value for quality could therefore be calculated by using the median for each of the six elements as presented in Table 3. The cross-table in Table 3 indicates a higher percentage quality of requirements when the practitioners established communication with the relevant stakeholder. The results show a clear pattern of dependency between the quality of the requirements and the communication established by the practitioners. A positive relationship is evident where established communication leads to the delivery of higher quality requirements. The pattern is consistent in each stakeholder’s case.

Table 3: Communication established by quality of requirements

Communication established

Quality Rating 1

Rating 3

Rating 3 to 5

Rating 5 to 7

Total

PM communication established (N = 49) 0% 2% 24% 74% 100% PM no communication established (N = 4 ) 25% 0% 25% 50% 100% PS communication established (N = 31) 0% 0% 16% 84% 100% PS no communication established (N = 20) 5% 5% 35% 55% 100% SME communication established (N = 40) 0% 3% 20% 77% 100% SME no communication established (N = 10) 10% 0% 30% 60% 100% EU communication established (N = 41 ) 2% 2% 10% 86% 100% EU no communication established (N = 8 ) 0% 0% 63% 37% 100% PM (Project manager), PS (Project sponsor), SME (Subject matter expert), EU (End-user) Rating: 1 (Never); 2 (Rarely, less than 10%); 3 (Occasionally, in about 30%); 4 (Sometimes, in about 50%); 5 (Frequently, in about 70%); (Usually, in about 90%); 7 (Every time)


Where communication was established between the practitioner and the relevant stakeholder, high quality requirements were delivered. The quality of delivered requirements where there was no involvement of the practitioners varied between very poor and good quality. There was no pattern here. As the cross-table shows a clear pattern, a Mann-Whitney test was done in each case of the two groups, where communication was established and in the case of no communication established, to validate if these two groups had the same median. A Mann-Whitney test was performed to test the hypothesis that the means of the group where communication was established is the same as the mean as the group where no communication was established for each stakeholder. In the cases of the project sponsor and end-user there is a statistically significant difference in the means of the groups where communication was established and where no communication was established. This shows that where communication was established, higher quality requirements were delivered. The significance can be derived from results for the project sponsor data (U = 191, p = 0.017) and for the end-user (U = 86.5, p = 0.03). This statistical significance did not hold up in the case of the project manager. The sample available for the nocommunication established group was very small and could be the reason why no statistically significant difference between the project manager groups could be found. These tests confirm the relationship between quality of requirements and communication. Where communication has been established, the quality of requirements is higher. Pattern 2 - Communication models In the majority of the cases the practitioners established one-way communication with the project sponsor, subject matter expert and the end-user as presented in Table 2. The communication models used between the practitioners and the project sponsor, subject matter expert and the end-user were linear models, as there was only a one-directional message flow. A linear communication model is a unidirectional model that sends a message from the sender to receiver with or without effect. In non-linear models, the message flow is multidirectional [47]. The fundamental driver for a practitioner to communicate with the various stakeholders is to elicit information about the domain in which the problem should be solved. From the communication motivation as presented in Table 2, it is evident that the practitioners established communication to acquire knowledge. Additionally, the value for stakeholders is to acquire knowledge about how the solution to the problem will affect their business activities.

As emphasised by Sommerville et al. [48], when solving complex problems a socio-technical perspective is required. A socio-technical solution includes technical systems, but also includes knowledge of how the system should be used by stakeholders to achieve the ultimate objective [49]. The one-way linear communication model is not good enough to assist the practitioner in acquiring knowledge about the domain of the problem and in enabling all stakeholders to develop a mutual understanding of the solution. To master the requirements engineering process, a good requirements engineer needs to manage the incompleteness of communication [14]. The practitioner as the owner of the requirements engineering process is responsible for facilitating this knowledge transfer to the various stakeholders. From the behaviour review it is evident that there is no drive from the practitioners’ side to facilitate the knowledge acquisition for the stakeholders about how the solution to the problem will affect business activities. Only a very small percentage of the respondents (0.8%) spent time with the end-users to validate their understanding and discuss issues that were experienced (0.8%) as presented in Table 2. To solve complex problems, companies and people are required to work together to create high quality solutions which they would not be able to produce individually [50]. Collaborations have become common where communication brings groups of people from diverse backgrounds and organisation roles together to work on a common problem [50]. If practitioners implement a more collaborative communication model instead of a linear communication model, they can facilitate the knowledge acquisition cycle for themselves (about the problem) as well as for the stakeholders (about the solution). In this way both parties would derive value from the engagement and a more complete understanding would be reached between practitioners and stakeholders. Pattern 3 - Communication motivation The communication motivation for interaction of the practitioner indicates a low percentage of communication with the project manager to agree on the analysis approach as presented in Table 2. This behaviour confirms the lack of focus on the requirements planning activity to generate a requirements management plan that identifies the required work. According to the literature, the requirements practitioner is responsible for the creation of a requirements management plan, which is a key input to the overall project schedule [43, 51]. Agreement on an analysis approach for the project with the project manager was done in only 7% of the cases. During requirements planning a requirements management plan is generated


that identifies the required work. The plan forms the basis for the project cost and resource estimation. Only 5% of the practitioners communicated with the project manager regarding input into these estimations. The practitioners communicated with the end-users to elicit information or to obtain knowledge about the domain. There was no focus on spending time with the end-users to validate their understanding. When this is compared with the communication model in the previous section, it aligns with the findings that the practitioners communicate with the end-users to elicit information or to obtain knowledge about the domain in which the problem should be solved. Practitioners should obtain knowledge of how the system should be used by stakeholders to achieve the ultimate objective [49]. Pattern 4 - Trust relationships The behavioural trust inventory was used to measure trust relationships between the requirements practitioner and the various stakeholders. The inventory suggests that there are two dimensions of trust, i.e. reliance and disclosure [52]. From the exploratory factor analysis results, it was concluded that the ten elements in the trust scale could be summarised into two factors: one factor summarising five reliance elements and the other summarising five disclosure elements. The mean of each dimension was presented on the original scale used to measure the trust, i.e. a scale from 1 (not at all willing) to 7 (completely willing), with 7 indicating high disclosure or reliance and 1 low disclosure or reliance. A summary of the trust behaviour of the respondents within interpersonal work relationships is given based on their willingness to be vulnerable. This willingness is described by the engaging in reliance and disclosure and is presented in Table 5. This summary separates the trust relationships where communication was established between the practitioner and relevant stakeholder versus where no communication was established.

Table 4: Trust relationships summary

Trustrelationships

Project manager

Project sponsor

Subject matter expert

End-user

1 2 1 2 1 2 1 2

Reliance-based elements mean

4.5 3.95 4.74 3.36 4.56 3.73 4.12 1.77

Disclosure-based elements mean

3.9 4.6 2.91 1.99 3.08 2.7 2.79 1.2

1 Communication established 2 No communication established A relationship characterised by a willingness to both rely and disclose represents a higher level of trust than a

relationship characterised by a willingness to trust in only one dimension [52]. The reliance and disclosure dimensions that characterise the trust relationship between the practitioner and the relevant stakeholder where communication was established were evaluated from Table 5. It can be deduced from the results in Table 5 that practitioners had only a one-dimensional trust relationship with all stakeholders which is based on willingness to rely. The practitioners surveyed were not willing to disclose work-related or personal information that was of a sensitive nature. In all the relationships between the practitioner and the relevant stakeholder the disclosure element was low. To determine whether the reliance trust element was high or low, a comparison with a known population mean was done using a one-sample t-test. The results of the one-sample t-test show that the reliance trust element (mean = 4.5286, SD = 1.51906, N = 56) was significantly different from the hypothesis value of 5.93, t (55) = -6.904, ρ = 0.000. The disclosure trust element (mean = 3.9036, SD = 1.44825, N = 56) was significantly different from the hypothesis value of 5.4, t (55) = -7.732, ρ = 0.000. Although the reliance trust element falls in the high reliance area, it can be confirmed that the trust relationship between the project manager and practitioner reflects low levels of trust. This is the only trust relationship that could be estimated as low or high compared with a known population mean, as no other population data exists for a relationship with the other stakeholders. It can therefore be concluded that the trust relationship between the practitioners is only one-dimensional with all stakeholders. It is based on willingness to rely, but is low. The practitioners’ behaviour was explored to determine if the trust relationship between practitioners and various stakeholders was similar or different when communication was established compared with when no communication was established. The trust elements where communication was established were compared with those where no communication was established, in the case of the end-user, project sponsor and subject matter expert. Where communication was established, there was a higher level of trust in the relationship. The project manager trust relationship does not confirm this pattern; however, there were only four responses, and this is a very small sample to derive any conclusion. These results confirm that where communication is not established, trust is low as suggested by the literature [34]. If the level of trust between the respondents and the various stakeholders is that low, it can be expected that the communication levels and interaction have not been established effectively.

4.

4.

4


In summary, communication was established but there is an indication that there was no two-way communication of information exchange. This is confirmed by the level of trust as indicated. Pattern 5 – Problem solving Problem solving is a fundamental competence that requirements practitioners should have [53, 54]. According to literature, as time passes during the problem-solving process, the sources of information consulted change [35]. During the early stages conversations with colleagues and personal experience are used. During the later stages of the process textbooks, codes and standards, industry newsletters and conversations with academics are used [40]. The preferred sources which were used by the practitioners were either personal experience or conversations with customers and colleagues, which are classified as initial sources [40]. The sources used by the practitioners indicate that the information-seeking behaviour did not follow a typical problem-solving process where the sources changed over time. The results from this knowledge discovery, namely how practitioners behave during the execution of the requirements process, highlighted key findings as summarised in Table 6.

Table 5: Summary of behaviour findings

Theme Finding Social and technical elements impact the quality of requirements

The effectiveness of communication and of the execution of the requirements activities impacts the quality of requirements.

Quality impacted by the effectiveness of communication

The more practitioners establish communication with various stakeholders, the higher the quality of requirements delivered.

Practitioners’ information-seeking behaviour

The problem-solving process of the practitioners depends on sources of information which are either personal experience or conversations; their information-seeking behaviour does not change over time.

Knowledge acquisition

There is no drive from the practitioners’ side to facilitate knowledge acquisition for the stakeholders about how the solution to the problem will affect business activities.

Trust Trust between practitioner and relevant stakeholders is low if communication has not been established.

The findings of the research were evaluated and, based on this evaluation, the conclusions were derived. The conclusions are discussed in the next section.

5. CONCLUSIONS This paper provides a summary of how practitioners gather information about the problem during the requirements process, use the information and share their resulting information. By discovering these interaction patterns, communication interactions can be improved

and made more effective. This integration of the information collected and knowledge generated formed a more complete understanding of the causes of the communication breakdowns and therefore the lack of acquisition of domain knowledge, which is continuously mentioned as a challenge in requirements engineering that affects quality. The communication patterns confirmed that the effectiveness of communication impacts on the quality of requirements. In addition, the trust relationships between the practitioners and the stakeholders affect communication effectiveness. The conclusion from these results is that the success of the requirements process is dependent on how well the process is executed, the creation of communication channels between the practitioner and various stakeholders as well as the development of human relationships based on trust. If practitioners implemented a more collaborative communication model instead of a linear communication model, they would be able to facilitate the knowledge acquisition cycle for themselves (about the problem) as well as for the stakeholders (about the solution). In this way, both parties would derive value from the engagement and a more enhanced understanding would be reached between practitioners. If stakeholders received value from the engagement, a more collaborative environment would be established and fewer communication breakdowns would follow. More efficient communication has a direct impact on the quality of requirements. The sources of information used by the practitioners reveal that the information-seeking behaviour does not follow a typical problem-solving process where the sources change over time. The preferred sources which were used by the practitioners were either personal experience or conversations. If practitioners aligned their information-seeking behaviour with a typical problem-solving process where the sources change over time, the quality of the requirements would be affected as more knowledge will be acquired during the problem-solving process. This would enable practitioners to gather comprehensive information about the problem. This knowledge may be used as a basis for future research to integrate the social and technical activities of the requirements engineering process.

6. REFERENCES

[1] M. I. Kamata and T. Tamai: "How does requirements quality relate to project success or failure?", 15th IEEE International Requirements Engineering Conference, pp. 69-78, 2007.

[2] K. Ellis: "Business Analysis Benchmark: The Impact of Business Requirements on the Success

5.


of Technology Projects", IAG Consulting Delaware. 2008.

[3] K. El Emam and A. G. Koru: "A replicated survey of IT software project failures", IEEE Software, Vol. 25 No. 5, pp. 84-90, 2008.

[4] R. L. Glass: Facts and Fallacies of Software Engineering Addison-Wesley, Boston. 2003.

[5] C. Schwaber: "The Root of the Problem: Poor Requirements", Forrester Research, Cambridge. September 2006.

[6] H. F. Hofmann and F. Lehner: "Requirements engineering as a success factor in software projects", IEEE Software, Vol. 18 No. 4, pp. 58-66, 2001.

[7] D. Leffingwell: "Calculating the Return on Investment from More Effective Requirements Management", American Programmer, Vol. 10 No. 4, pp. 13-16, 1997.

[8] A. Aurum and C. Wohlin: "The fundamental nature of requirements engineering activities as a decision-making process", Information and Software Technology, Vol. 45 No. 14, pp. 945-954, 2003.

[9] N. P. Vitalari and G. W. Dickson: "Problem solving for effective systems analysis: An experimental exploration", Communications of the ACM, Vol. 26 No. 11, pp. 948-956, 1983.

[10] J.-N. Kim and J. E. Grunig: "Problem solving and communicative action: A situational theory of problem solving", Journal of Communication, Vol. 61 No. 1, pp. 120-149, 2011.

[11] R. Fuentes-Fernández, J. Gómez-Sanz, and J. Pavón: "Understanding the human context in requirements elicitation", Requirements Engineering, Vol. 15 No. 3, pp. 267-283, 2010.

[12] R. Dawson: Developing Knowledge-Based Client Relationships: The Future of Professional Services, Butterworth-Heinemann, Boston. 2000.

[13] A. Marnewick, J.-H. Pretorius, and L. Pretorius: "A perspective on human factors contributing to quality requirements: A cross-case analysis",IEEE International Conference on Industrial Engineering and Engineering Management (IEEM), pp. 389-393, 2011.

[14] C. Rupp: "Requirements and psychology", IEEE Software, Vol. 19 No. 3, pp. 16-18, May/Jun 2002.

[15] C. Rupp: "Are systems engineers complete losers when it comes to communication? ", Cutter IT Journal, Vol. 19 No. 7, 2006.

[16] S. Ramachandran, S. Dodda, and L. Santapoor: "Overcoming Social Issues in Requirements Engineering", in Advanced Computing, Vol. 133, N. Meghanathan, et al., Eds., Heidelberg: Springer-Verlag, 2011, pp. 310-324.

[17] L. Karlsson, A. G. Dahlstedt, B. Regnell, J. Natt och Dag, and A. Persson: "Requirements engineering challenges in market-driven software development - An interview study with

practitioners", Information and Software Technology, Vol. 49 No. 6, pp. 588-604, 2007.

[18] D. Damian and J. Chisan: "An empirical study of the complex relationships between requirements engineering processes and other processes that lead to payoffs in productivity, quality, and risk management", IEEE Transactions on Software Engineering, Vol. 32 No. 7, pp. 433-453, 2006.

[19] A. G. Sutcliff, A. Economou, and P. Markis: "Tracing requirements errors to problems in the requirements engineering process", Requirements Engineering, Vol. 4 No. 3, pp. 134-51, 1999.

[20] B. Curtis, H. Krasner, and N. Iscoe: "A field study of the software design process for large systems", Communications of the ACM, Vol. 31 No. 11, pp. 1268-1287, 1988.

[21] E. Bjarnason, K. Wnuk, and B. Regnell: "Requirements are slipping through the gaps - A case study on causes & effects of communication gaps in large-scale software development", 19th IEEE International Requirements Engineering Conference (RE), 2011, Trento, Italy, pp. 37-46, 2011.

[22] J. Siddiqi: "Requirement engineering: The emerging wisdom", IEEE Software, Vol. 13 No. 2, p. 15, 1996.

[23] B. Penzenstadler, G. Haller, T. Schlosser, and G. Frenzel: "Soft Skills Required: A Practical Approach for Empowering Soft Skills in the Engineering World", Collaboration and Intercultural Issues on Requirements: Communication, Understanding and Softskills, pp. 31-36, 2009.

[24] K. Siau and X. Tan: "Technical communication in information systems development: The use of cognitive mapping", IEEE Transactions on Professional Communication, Vol. 48 No. 3, pp. 269-284, 2005.

[25] D. Damian and D. Zowghi: "RE challenges in multi-site software development organisations", Requirements Engineering, Vol. 8 No. 3, pp. 149-160, 2003.

[26] J. Hartwick and H. Barki: "Communication as a dimension of user participation", IEEE Transactions on Professional Communication, Vol. 44 No. 1, pp. 21-36, 2001.

[27] J. Fisher: "The value of the technical communicator's role in the development of information systems", IEEE Transactions on Professional Communication, Vol. 42 No. 3, pp. 145-155, 1999.

[28] A. Al-Rawas and S. Easterbrook: "Communication problems in requirements engineering: A field study", Professional Awareness in Software Engineering, pp. 47- 60, 1996.


[29] K. Holtzblatt and H. R. Beyer: "Requirements gathering: The human factor", Communications of the ACM, Vol. 38 No. 5, pp. 31-32, 1995.

[30] R. P. Bostrom: "Successful application of communication techniques to improve the systems development process", Information & Management, Vol. 16 No. 5, pp. 279-295, 1989.

[31] J. Coughlan, M. Lycett, and R. D. Macredie: "Communication issues in requirements elicitation: A content analysis of stakeholder experiences", Information and Software Technology, Vol. 45 No. 8, pp. 525-537, 2003.

[32] B. Paech, T. Koenig, L. Borner, and A. Aurum: "An Analysis of Empirical Requirements Engineering Survey Data", in Engineering and Managing Software Requirements, A. Aurum and C. Wohlin, Eds., Berlin Heidelberg: Springer, 2005, pp. 436-487.

[33] A. Hoffmann and C. Lescher: "Collaboration and Intercultural Issues on Requirements: Communication, Understanding and Softskills (CIRCUS)", Collaboration and Intercultural Issues on Requirements: Communication, Understanding and Softskills, pp. 1-4, 2009.

[34] R. Zeffane, S. A. Tipu, and J. C. Ryan: "Communication, commitment & trust: Exploring the triad", International Journal of Business & Management, Vol. 6 No. 6, pp. 77-87, 2011.

[35] C. Tenopir and D. W. King: Communication Patterns of Engineers, Wiley, Hoboken. 2004.

[36] K. Wiegers. (2003, 23 February 2012). So You Want to Be a Requirements Analyst. Software Development 11(7). Available: http://www.processimpact.com/pubs.shtml#requirements

[37] G. Dietz and D. N. Den Hartog: "Measuring trust inside organisations", Personnel Review, Vol. 35 No. 5, pp. 557-588, 2006.

[38] N. A. Gillespie: "Measuring trust in organizational contexts: An overview of survey-based measures", in Handbook of Research Methods on Trust, F. Lyon, et al., Eds., Surrey: Edward Elgar, 2012, pp. 175-188.

[39] B. McEvily and M. Tortoriello: "Measuring trust in organisational research: Review and recommendations", Journal of Trust Research, Vol. 1 No. 1, pp. 23-63, 2011/04/01 2011.

[40] D. Veshosky: "Managing innovation information in engineering and construction firms", Journal of Management in Engineering, Vol. 14 No. 1, pp. 58-66, 1998.

[41] L. M. Rea and R. A. Parker: Designing and Conducting Survey Research - A Comprehensive Guide, Jossey-Bass, San Francisco. 2005.

[42] F. J. Fowler: Survey Research Methods, Sage, Thousand Oaks, Fourth Edition. 2009.

[43] B. Carkenord, C. Cartwright, R. Grace, L. Goldsmith, E. Larson, J. Skrabak, and C. Vonk: "Partnering for Project Success Project Manager

and Business Analyst Collaboration", Proceedings of the PMI Global Congress, Washington, DC, 2010.

[44] International Institute of Business Analysis: AGuide to the Business Analysis Body of Knowledge® (BABOK® Guide), International Institute of Business Analysis, Toronto Vol. 2. 2009.

[45] IEEE, "IEEE Guide for Developing System Requirements Specifications," IEEE Std 1233, 1998 Edition, p. i 1998.

[46] IEEE, "IEEE Recommended Practice for Software Requirements Specifications," IEEE Std 830-1998, p. i 1998.

[47] U. Narula: Communication Models, Atlantic, New Delhi. 2006.

[48] I. Sommerville, et al.: "Large-scale complex IT systems", Communications of the ACM, Vol. 55 No. 7, pp. 71-77, 2012.

[49] I. Sommerville: Software Engineering, Pearson Education, Harlow, Seventh Edition. 2004.

[50] C. Stohl and K. Walker: "A Bona Fide Perspective for the Future of Groups Understanding Collaborating Groups", in NewDirections in Group Communication, L. R. Frey, Ed., Thousand Oaks: Sage, 2002, pp. 237-252.

[51] S. Weese and T. Wagner: CBAP/CCBA Certified Business Analysis Study Guide, Wiley. 2011.

[52] N. Gillespie: "Measuring trust in work relationships: The Behavioural Trust Inventory," presented at the Academy of Management Seattle, 2003.

[53] ECSA, Engineering Council of South Africa, "Competency Standard for Registration as a Professional Engineer", R-02-PE: ECSA 2011.

[54] IIBA, "Business Analysis Competency Model Version 3.0", 2011.


CARRIER FREQUENCY SYNCHRONIZATION FOR OFDM-IDMA SYSTEMS M.B Balogun*, O.O Oyerinde** and S.H Mneney* *Centre for Radio Access and Rural Technologies (CRART), University of KwaZulu-Natal, Durban 4041, South Africa E-mail: [email protected], [email protected] **School of Electrical and Information Engineering, University of the Witwatersrand, Johannesburg, 2050, South Africa E-mail: [email protected] Abstract: The recently proposed Orthogonal Frequency Division Multiplexing-Interleave Division Multiple Access (OFDM-IDMA) scheme has been at the forefront of wireless communication systems research due to its promising features which include high efficiency, flexibility and high data-rate wireless transmission. Earlier, major research on this new scheme had only focused on the perfect-condition scenarios, with the assumption that the system is free from synchronization errors. However, recent studies show that synchronization errors, especially the carrier frequency offsets, have significant impact on the OFDM-IDMA systems. This work therefore examines comprehensively, analyses and verifies the impact of synchronization errors on the performance of the OFDM-IDMA hybrid multicarrier scheme. Furthermore, simple but effective synchronization algorithms, namely; the linear minimum mean-squared error (MMSE) based algorithm, and the Kernel Least Mean Square (KLMS) algorithm as well as its normalized counterpart called the Normalized-KLMS are adopted and adequately exploited to combat the degrading impacts of synchronization errors on the OFDM-IDMA scheme. The linear minimum mean-squared error (MMSE) based algorithm, is a non-data aided method, which focuses on the mitigation of the inter-channel interference induced by the effect of synchronization errors on the system. The Kernel Least Mean Square (KLMS) algorithm, as well as its normalized counterpart, presents an efficient approach for estimation and effective correction methods to combat the effect of carrier frequency offset errors on the performance of the OFDM-IDMA system. The bit error rate (BER) performance of these algorithms is presented, compared, and analyzed. Also, the achievable performance of these synchronization algorithms in a Rayleigh multipath channel scenario with varying mobile speed is analyzed and documented. Keywords: IDMA, OFDM-IDMA, KLMS, MMSE, CFO, synchronization

1. INTRODUCTION

The demand for a reliable and high-data-rate mobile communication service has been phenomenal in recent decades. This has necessitated several research works in advanced communication techniques to serve this unprecedented demand, while also tackling the challenges associated with mobile communications, which include multipath fading, co-channel interference, and Doppler Effect among others. Various techniques have been studied to proffer solutions to these wireless communication challenges. Prominent among these techniques are Orthogonal Frequency Division Multiplexing (OFDM) and the Code Division Multiple Access (CDMA) multiuser schemes. Both schemes offer a number of advantages but it is noteworthy that the combined scheme of OFDM-CDMA appears most attractive due to its high spectral efficiency and radio resource management flexibility. However, the multicarrier CDMA scheme [1] faces a major challenge in Multiple Access Interference (MAI) at the receiver. This causes degradation in performance of the system. This becomes even more severe as the number of active users increases. Multiuser detection (MUD) techniques, which are usually deployed at the receiver end of the

system, involving the removal of interfering signals from the signal of interest, have been widely employed to tackle MAI in OFDM-CDMA systems [2, 3]. Several MUD techniques have been studied with the aim of achieving a reasonable level of complexity with some compromise in the system performance. However, the complexity of these detection schemes tends to grow exponentially as the number of users increases. A new multiuser scheme called Interleave Division Multiple Access (IDMA), with associated low complexity in comparison with the OFDM-CDMA scheme was proposed in 2002 by Li Ping et al [4]. A combined scheme of OFDM-IDMA was later proposed in 2006 by Mahafeno et al. [5] to enhance the performance of the conventional IDMA scheme, although a related technique had been proposed earlier by Xibin Xu et al [6] but with major attention on the downlink scenario. The study by Mahafeno et al. evaluated the performance of the OFDM-IDMA scheme in a perfectly synchronous scenario and presented a comparison between the conventional IDMA and the OFDM-IDMA scheme. In Fig. 1, the block diagram of a conventional IDMA transceiver is shown. The OFDM-IDMA transceiver, as shown in Fig. 2, is similar to Fig. 1 but for the insertion of the OFDM component in the system. A simple chip-by-chip iterative


multiuser detection strategy is employed for the implementation of the multiuser IDMA scheme [7]. The major priority of the OFDM-IDMA scheme is to mitigate MAI and Inter-symbol Interference (ISI) with low complexity. Comparisons are made between the multiple access schemes IDMA and CDMA in [8, 9]. The OFDM-IDMA scheme offers a better system performance and higher spectral efficiency than the conventional multicarrier CDMA scheme, combining all the inherent

advantages of the OFDM and the IDMA schemes. The OFDM-IDMA enables high flexibility of resource allocation among users and the MUD in this scheme can be implemented efficiently with complexity per user independent of the channel length and the number of users [10, 11].

Fig. 1 Block diagram of a typical IDMA Transceiver. Fig. 2 The block representation of the OFDM-IDMA Transceiver.

Elementary Signal

Estimator

Multiple Access

Channel

FEC Interleaver K

FEC Interleaver 1 x1 n

User K

User 1

xK n

.

.

.

.

.

.

DEC Interleaver 1

De-interleaver 1

De-interleaver K

Interleaver K

DEC

User 1

User K

.

.

.

eESE xk n

eDEC xk n

eESE xk n

pESE xk n

pDEC xk n eDEC xk n

pESE xk n

pDEC xk n

IDMA Transmitter Structure

IDMA Receiver Structure

OFDM BLOCK

ESE

DECs

User K

User 1

r n

FEC

FEC Interleaver 1

Interleaver K

User 1

User K

x1 n

xK n

.

.

.

.

.

.


However, recent studies [12, 13, 14] show that the presence of the OFDM component makes the OFDM-IDMA scheme vulnerable to synchronization errors. The OFDM technique is highly sensitive to carrier frequency offsets (CFOs) especially when users are transmitting asynchronously with diverse delays and fading at the uplink. Doppler Shifts and instabilities in the local oscillators are the major cause of carrier frequency offsets [15], which cause Inter-channel Interference (ICI) and loss of orthogonality among the sub-carriers, leading to an overall performance degradation of the system. The contribution of this paper therefore, is to examine, analyze, and verify the impact of synchronization errors on the performance of the recently proposed multicarrier IDMA scheme. The impact of synchronization errors on the multicarrier scheme is investigated in both slow fading and fast fading multipath channel. Also, much has not been done hitherto to address the degrading effect of synchronization errors, especially the carrier frequency offset errors on the performance of this noble scheme. Hence, we present and implement a linear Minimum Mean-squared Error (MMSE) based algorithm as well as the Kernel Least Mean Square (KLMS) synchronization algorithms to combat the undesirable impact of synchronization errors on the overall performance of the OFDM-IDMA scheme. To the best of our knowledge, the proposed algorithms have not been exploited and jointly studied in literature, to tackle synchronization errors in multicarrier IDMA systems. Furthermore, computer simulations are presented and the algorithms are applied in a Rayleigh fading multipath channel with varying mobile speed to verify their effectiveness and to clearly demonstrate their influence on the performance of the OFDM-IDMA system in a practical scenario. Also, the algorithms are comprehensively compared to ascertain which of the algorithms offer the most efficient system performance. The layout of the rest of the paper is as follows: Section 2 describes the conventional system model of the OFDM-IDMA scheme as well as in the presence of CFOs. Section 3 presents the proposed carrier frequency synchronization algorithms; the linear MMSE-based algorithm and the KLMS synchronization algorithms. Section 4 discusses and analyses the simulation results and lastly in Section 5, the conclusion is given. 2. THE MULTICARRIER IDMA SYSTEM MODEL

The multicarrier IDMA scheme is a combination of two different mobile communication techniques. These are the Orthogonal Frequency Division Multiplexing (OFDM) and the Interleave Division Multiple Access (IDMA) techniques. The OFDM-IDMA scheme was proposed in a bid to improve the performance of the conventional IDMA, proffering a host of advantages, which include high spectral efficiency, low receiver

complexity, robustness against fading caused by multipath, among others. 2.1 The conventional OFDM-IDMA transceiver The OFDM technique [16], essentially involves the division of high data rate streams into parallel streams of a lower data rate which are modulated by different narrowband sub-carriers and then transmitted. This technique takes care of the problems of Inter-symbol Interference and Inter-channel Interference, which are encountered while transmitting streams of high data rate. However, orthogonality must be maintained between sub-carriers to avoid interference and enable good signal reception at the receiver. The IDMA multiuser technique essentially involves the assignment of distinct interleavers, which are randomly generated, to distinguish signals from different users unlike the CDMA technique where spreading codes are used as the main identifiers in the system. Considering the transceiver structure of the OFDM-IDMA system, as shown in Fig. 2 with K users communicating concurrently, the input data of a particular user k, is first encoded using a Forward Error Correction (FEC) code [17, 18] then interleaved by a randomly generated interleaver to obtain xk(n). The Forward Error Correction codes replace the spreading codes needed for reliable data transmission and reception in the multicarrier CDMA scheme. The FEC technique increases the coding gain of the OFDM-IDMA scheme and controls errors in the information transmitted over the fading channel. The signal, which results from the interleaving process, is transmitted over the multipath channel to the receiver. The received signal is given as: ∑

1 , (1) (2) where hk(n) is the multipath channel coefficient, (n) is the additive white Gaussian noise (AWGN) and the combination of the interference due to other simultaneous users, with respect to a particular user k and the Gaussian noise (n) is represented by , and can be written as: ∑ (3) Also present at the receiver end of the OFDM-IDMA system are the Elementary Signal Estimator (ESE), which is of low complexity, and a posteriori probability (APP) decoder [19] for each of the users. The ESE function operates in a chip-by-chip sequence, as detailed in [19, 20].The ESE mainly carries out a low computational chip-by-chip detection to coarsely subtract interference among concurrent users in the system. The mean and variance for each chip, as computed by the APP decoders, are iteratively fed back to the ESE unit. This is an important process which enables the ESE unit to provide a better output as well as to obtain an improved APP estimation. Now considering a system in which the BPSK


signaling is assumed and chip-by-chip detection takes place in a quasi-static multipath channel, the interference mean and the variance are therefore estimated as [21]: (4) | | (5) From the interference mean and variance expressed above, the output of the ESE for an active user k is derived, based on the extrinsic log-likelihood ratios (LLRs) generation, as follows [21]:

(6) All derivations above represent an ideal situation where it is assumed that the OFDM-IDMA system is perfectly synchronized, with no errors. However, this is not feasible in practice where active users transmit asynchronously with diverse delays and fading at the uplink. These lead to the loss of orthogonality and subsequently, degrades the overall output of the system. 2.2 The OFDM-IDMA System analysis with carrier frequency offsets Major studies on the OFDM-IDMA scheme have been considered under perfect conditions, without considering the impact of carrier frequency offsets on the overall performance of the system. However, as studied recently by some papers [12, 13, 14], the OFDM-IDMA scheme suffers from carrier frequency offset (CFO) errors, a problem inherent in the OFDM component, causing high performance degradation in the system. Synchronization errors cause Inter-channel interference and loss of orthogonality among the sub-carriers, which in turn causes performance degradation. Therefore, considering the scenario where carrier frequency offsets (CFOs) are introduced into the system, equation (1) then becomes: ∑

1 (7) where n represents the sub-carrier index, N is the number of sub-carriers and , [22], represents the normalized CFO. After the discrete Fourier transforms operation, equation (7) becomes: ∑ 1

(8) The mathematical expression in equation (8) can therefore be represented, with respect to other simultaneous users as: ∑ (9) and the overall interference in the OFDM-IDMA system, symbolized by , is written as: ∑ (10)

where is a Gaussian random variable expressed as: ∑ 1

(11) while the interference due to the carrier frequency offset between a particular user denoted as and another active user indicated as is represented by . The mathematical expression of the interference is therefore given as [12]: ∑ 1

(12) From equations (4) and (5), having in mind that carrier frequency offsets are now introduced into the system also modifying equation (6), the output of the elementary signal operator, in a multipath channel, based on the extrinsic log-likelihood ratios (LLRs) generation [21], therefore becomes:

. (13) Hence, the equations above establish the presence and the effect of carrier frequency offset in the OFDM-IDMA system model. Addressing the degrading impact of synchronization errors is therefore crucial for optimum performance and system efficiency.

3. CARRIER FREQUENCY SYNCHRONIZATION Carrier frequency offsets have critical influence on the overall efficiency of the multicarrier IDMA scheme as described above. A linear MMSE-based synchronization algorithm, which is a non-data aided technique, as well as the KLMS algorithm are therefore presented and implemented to combat the deteriorating effect of carrier frequency offset errors on the performance of the OFDM-IDMA system. 3.1 The MMSE-Based Synchronization Algorithm The CFO of a particular user k can be compensated coarsely in the time domain of the multicarrier IDMA system. The received signal can be multiplied by the complex exponent of the offset estimate, in the presence of carrier frequency offsets, as seen in equation (7), to achieve a coarse error compensation for an active user. However, the main challenge is the Inter-channel interference which results from the residual CFOs due to other simultaneous active users in the system. The focus of this MMSE-based synchronization algorithm is therefore to reduce the effect of the ICI originating from the residual CFOs of concurrent users in the OFDM-IDMA system. Hence, the proposed linear MMSE-based algorithm is obtained by computing the second order statistics from equations (8), (11), and (12) as [23]: | | 1 (|∑

1

| )


| | (14)

1 ∑ ∑ ( ) 1

1

⁄

1 ∑ ∑ [∑ | | | | 1

] 1

1

⁄

| | ⁄ (15) where is the data symbol energy with sub-carrier length , and ⁄ is the power spectral density. From equations (9) and (15), the power of the ICI term is therefore derived as stated below [23]: | | | | | | | | (16)

| | (17) In equation (17), the presence of CFO is visible and the ICI power will have its minimum value at . Also, the power of the ICI in the OFDM-IDMA system will increase as the value of the CFO increases. Therefore, the impact of carrier frequency offset errors can be mitigated adequately by reducing the spectral power of the ICI in the OFDM-IDMA system. To achieve this, the following cost function stated below can be employed [23, 24]:

1 ∑ | | 1 (18)

where is the expectation of | | taken to ensure the cost function is minimized for carrier frequency value .Further expansion of equation (18) gives:

1 ∑ | | 1

1 ∑ | | 1

(19) As the second term of equation (19) is independent of the CFO as obtained in equation (9), we can therefore compute the gradient of with respect to the CFO , so that the stochastic update signal may be expressed as [23]: 1

∑ | | 1

( ) (20)

and

is given as:

1√ ∑ (

) 1

(21)

To achieve a lower system complexity, the update error signal in equation (20) is simplified and the approximated stochastic update signal is therefore obtained as follows [23]: 1

∑ | | 1

Therefore, the expression in equation (22) is used to update the carrier frequency-tracking loop and effectively combats the influence of the ICI induced in the OFDM-IDMA system by multiple simultaneous users. 3.2 The Kernel Least Mean Square Algorithm The Kernel least mean square algorithm is simple to implement and highly efficient for prediction, estimation and correction. The modulated input data is utilized in the implementation of the Kernel function, which enhances the efficiency of the algorithm and its overall influence on the output of the system. The KLMS synchronization algorithm is achieved by executing the popular Least Mean Square (LMS) algorithm in the kernel space [25]. The update of the KLMS coefficient vector, like the conventional LMS, is obtained as follows: (23) where is the coefficient vector, and the mapping process is achieved by the function while is the step-size. The error signal, which updates the frequency-tracking loop is represented by , and can be expressed mathematically as: (24) where is the desired signal and is the received baseband signal in the presence of synchronization errors. The expected output of the OFDM-IDMA system is therefore given as [25]: ⦋ ⦌ (25) Also, expressing equation (23) in its non-recursive form gives [26]: ∑ 1

(26) Initializing , we therefore obtain as derived from equations (25) and (26) as: ∑ ⦋ 1

⦌ (27) Now introducing the Kernel function, the output in terms of the error signal is therefore written as: ∑ xk n x k n 1

(28)


where denotes the Kernel function, which is given as [26]: xk n x k n exp ||xk n x k n || (29) and is the kernel width parameter. However, for an enhanced system tracking and amplitude stability, the normalized KLMS (NKLMS) is presented. The NKLMS helps to stabilize the step-size parameter so that controls better and minimizes error in the system. Thus, from (23),

Similar to (26),

∑ ( )

1

Initializing with no a priori information available [27],

∑

1

Thus, from (25) and (32), is derived as

∑

1

The derivation above therefore gives the expression for the normalized KLMS. Hence, with the sufficient knowledge of the modulated signal at the receiver end of the system and the update error signal as derived, the degrading influence of the synchronization errors is effectively corrected.

4. SIMULATIONS AND DISCUSSION Computer simulations are carried out first to demonstrate and verify the performance of the OFDM-IDMA system in the presence of carrier frequency offset errors. Then the impact of the proposed algorithms on the overall performance of the multicarrier IDMA system is validated. The computer simulations are documented based on the bit error rate performance of the OFDM-IDMA system model used. The OFDM-IDMA system model operates at a carrier frequency of 2GHz in a fading multipath channel of M= 16 paths. The QPSK modulation technique is employed, and the number of sub-carriers is set at N=128 with input data of length 32 bits. The number of sub-carriers N should be varied in practice, to provide each active user with variable data rates. However, we assume the same number of sub-carriers for all users in the computer simulation for convenience. Spreading length is also set to 4, with randomly generated

interleavers and sampling period of 0.5µs. The performance of the OFDM-IDMA system model is demonstrated, applying the proposed synchronization algorithms, in a Rayleigh multipath channel with normalized Doppler frequencies of fD = 0.0136, and fD = 0.1808 corresponding to mobile speeds of v= 15 km/h, and v = 200 km/h respectively. The computer simulations are carried out with varying mobile speed to validate the performance of the proposed algorithms in both slow and fast fading multipath channels. As stated earlier, the introduction of the OFDM process makes the OFDM-IDMA system susceptible to carrier frequency offset errors. Carrier frequency offsets attenuate the useful signal, increase the system bit error rate and degrade system performance. In Fig. 3, the bit error rate performance of the OFDM-IDMA scheme in the presence of CFOs is shown. For this simulation, CFOs values are assumed to be known based on the study in [28], where different ways to obtain carrier frequency offsets in OFDM systems are presented. CFO values are therefore varied from zero, which represents the perfect synchronization scenario, up to 0.18. The degrading effect of carrier frequency offset errors on the system model, as studied in [12, 13, 14], is observed as the value of the CFO increases. The trend of the plot depicts that further increase in carrier frequency offset values will only lead to greater degradation in the overall output of the OFDM-IDMA system. The ideal OFDM-IDMA scenario, where there are no synchronization errors in the system at CFO=0, is included as a benchmark in the computer simulations to clearly demonstrate the impact of carrier frequency offsets on the system, as well as the subsequent influence of the proposed synchronization algorithms on the overall performance of the system model. Fig. 4 shows the impact of the linear MMSE-based synchronization algorithm on the OFDM-IDMA system model. The algorithm is applied in the presence of high carrier frequency offsets of 0.1 and 0.18 for clear demonstration of its effectiveness. An appreciable improvement in the overall output of the system is observed, which signifies an effective reduction in the power of the ICI caused by the residual CFOs of the concurrent users in the system. In Fig. 5, the bit error rate performance of the OFDM-IDMA system upon the application of the KLMS algorithm is shown. The proposed algorithm works in accordance with the kernel matrix, which is achieved by the implicit products of the information data samples, using the kernel function [29]. The step-size of the algorithm is set at for quick convergence [30]. At the lower CFO of 0.05, an absolute synchronization is obtained. Also, a significant system performance is observed as well, even at a higher CFO value of 0.1. The impact of this algorithm on the system model is therefore, conspicuous and highly effective.


For better comparison, the MMSE and the KLMS algorithms are shown on the same plot in Fig.6. The KLMS algorithm attains a comprehensive synchronization, better than what is obtained in the MMSE-based algorithm. The efficacy of both algorithms is also visible at a higher CFO value of 0.1, with the KLMS algorithm achieving a better system output. Also, the impact of the KLMS and the linear MMSE algorithms on the OFDM-IDMA system model is investigated in a Rayleigh fading channel of varying mobile speeds of 15km/h and 200km/h. The efficiency of the proposed algorithms is significant and obvious in both cases, as shown in Fig. 7. Fig. 8 shows the comparison of the KLMS algorithm shown in Fig. 6 and that of the improved normalized KLMS algorithm. The NKLMS brings stability into the system as well as an improved overall performance. As seen from the plot, the NKLMS even outperforms the regular KLMS in both cases (i.e. CFO=0.05 and CFO=0.1) although this is more visible at a higher CFO value of 0.1. Also, both algorithms are implemented at a varying mobile speed of 15km/h and 200 km/h with CFO =0.1, as shown in Fig. 9 for better comparison. From the plot, the NKLMS offers a better and improved output. Finally, the MMSE-based algorithm, the KLMS algorithm and its normalized counterpart, are implemented and compared in Fig. 10. As discussed earlier, the plot ascertains the trend, which shows that the MMSE-based algorithm has the least impact on the system while the NKLMS exceeds the other algorithms in performance. In terms of computational complexity, the discrete-time derivative [23, 31] is employed for the MMSE-based algorithm to obtain the stochastic update signal in (22) with computational complexity of O(N2). Also, the KLMS computational complexity is of O(N) [32], which ensures significant overhead is not added to the overall system complexity. The implementation of the KLMS algorithm therefore offers a better and improved BER output with lower computational complexity.

5. CONCLUSION In light of the proven degrading impact of synchronization errors on the performance of the OFDM-IDMA system, three effective algorithms have been proposed and presented to combat this undesirable effect. The linear MMSE-based synchronization algorithm, which focuses on the reduction of the ICI power due to the residual CFOs emanating from simultaneous users, is first employed to minimize the impact of synchronization errors in the system. Furthermore, the KLMS algorithm, together with its normalized counterpart, was implemented for effective carrier frequency synchronization in the OFDM-IDMA systems. The KLMS algorithms, like the MMSE-based algorithm, was implemented in practical conditions of high carrier frequency offsets, with varying mobile speeds depicting

both slow and fast fading multipath channel scenarios. Simulation results, which show the effectiveness and the significant influence of the KLMS algorithm, have been documented. It is noteworthy, however, that the NKLMS algorithm achieves the most efficient, as well as improved system performance and stability even in a varying, fast fading multipath scenario.

Fig. 3 The general performance of the OFDM-IDMA system model with increasing CFOs from 0 to 0.18.

Fig. 4 The impact of the linear MMSE-based synchronization algorithm on the OFDM-IDMA system model with carrier frequency offsets 0.1 and 0.18.

0 2 4 6 8 10 12 14 16 18 2010

-4

10-3

10-2

10-1

100

SNR[dB]

BE

R

CFO = 0CFO = 0.02CFO = 0.05CFO = 0.1CFO = 0.18

0 2 4 6 8 10 12 14 16 18 2010

-4

10-3

10-2

10-1

100

SNR[dB]

BE

R

CFO = 0 (Ideal)CFO = 0.1 (no correction)CFO = 0.18 (no correction)CFO = 0.1 (mmse correction)CFO = 0.18 (mmse correctino)


Fig. 5 The impact of the KLMS synchronization algorithm on the OFDM-IDMA system model with carrier frequency offsets 0.05 and 0.1.

Fig. 6 The impact of the MMSE and the KLMS synchronization algorithms on the OFDM-IDMA system model with carrier frequency offsets 0.05 and 0.1.

Fig. 7 BER performance of the OFDM-IDMA system model with two of the proposed algorithms in both slow and fast fading multipath channel of mobile speeds 15 km/h and 200 km/h.

Fig. 8 Comparison of the KLMS and the NKLMS synchronization algorithms with carrier frequency offsets 0.05 and 0.1.

Fig. 9 Speed varying comparison of the KLMS synchronization algorithm with its normalized counterpart.

Fig. 10 BER performance of all the implemented synchronization algorithms in both slow and fast fading multipath scenario.

0 2 4 6 8 10 12 14 16 18 2010

-4

10-3

10-2

10-1

100

SNR[dB]

BE

R

CFO = 0CFO = 0.05CFO = 0.1CFO = 0.05 (klms correction)CFO = 0.1 (klms correction)

0 2 4 6 8 10 12 14 16 18 2010

-4

10-3

10-2

10-1

100

SNR[dB]

BE

R

CFO = 0CFO = 0.05CFO = 0.1CFO = 0.05 (klms correction)CFO = 0.1 (klms correction)CFO = 0.05 (mmse correction)CFO = 0.1 (mmse correction)

0 2 4 6 8 10 12 14 16 18 2010

-4

10-3

10-2

10-1

100

SNR[dB]

BE

R

CFO = 0.1 (15km/h)CFO = 0.1 (200km/h)CFO = 0.1 (15km/h) -klms correctionCFO = 0.1 (200km/h) -klms correctionCFO = 0.1 (15km/h) -mmse correctionCFO = 0.1 (200km/h) -mmse correction

0 2 4 6 8 10 12 14 16 18 2010

-4

10-3

10-2

10-1

100

SNR[dB]

BE

R

CFO = 0CFO = 0.05CFO = 0.1CFO = 0.05 (klms correction)CFO = 0.1 (klms correction)CFO = 0.05 (nklms correction)CFO = 0.1 (nklms correction)

0 2 4 6 8 10 12 14 16 18 2010

-4

10-3

10-2

10-1

100

SNR[dB]

BE

R

CFO = 0.1 (15km/h)CFO = 0.1 (200km/h)CFO = 0.1 (15km/h) -klms correctionCFO = 0.1 (200km/h) -klms correctionCFO = 0.1 (15km/h) -nklms correctionCFO = 0.1 (200km/h) -nklms correction

0 2 4 6 8 10 12 14 16 18 2010

-4

10-3

10-2

10-1

100

SNR[dB]

BE

R

CFO = 0.1 (15km/h)CFO = 0.1 (200km/h)CFO = 0.1 (15km/h) -klms correctionCFO = 0.1 (200km/h) -klms correctionCFO = 0.1 (15km/h) -nklms correctionCFO = 0.1 (200km/h) -nklms correctionCFO = 0.1 (15km/h) -mmse correctionCFO = 0.1 (200km/h) -mmse correction


6. REFERENCES [1] D. Schilling, L. Milstein, R. Pickholtz, M. Kullback,

and F. Miller: “Spread spectrum for commercial communications,’’ IEEE Comm. Mag., vol. 29, no. 4, pp. 66-79, April 1991.

[2] Z. Xie, R.T. Shortand and C.T. Rushforth: “A family

of suboptimum detectors for coherent multiuser communications,” IEEE ISAC, SAC-8, pp. 683-690, May 1990.

[3] S. Moshavi: “Multiuser Detection for DS-CDMA

Communications,” IEEE Comm. Mag., vol. 34, pp.124–36, October 1996.

[4] L. Ping, K.Y. Wu, L.H. Liu and W. K. Leung: “A

simple unified approach to nearly optimal multiuser detection and space-time coding,” Information Theory Workshop, ITW'2002, pp. 53-56, October 2002.

[5] I. Mahafeno, C. Langlais and C. Jego: “OFDM-

IDMA versus IDMA with ISI cancellation for quasi-static Rayleigh fading multipath channels,” in Proc. 4th Int. Symp. on Turbo Codes & Related Topics, Munich, Germany, April 3-7, 2006.

[6] X. Xu, J. Wang and Y. Yao: “Novel Techniques to

Improve Downlink Multiple Access Capacity for Beyond 3G,” in IEEE Comm. Mag., Special Issue in Wireless Communication in China, pp. 61-67, January 2005.

[7] L. Ping: “Interleave-division multiple access and

chip-by-chip iterative multiuser detection,” IEEE Comm. Mag., vol. 43, no. 6, pp. S19–S23, June 2005.

[8] K. Kusume, G. Bauch and W. Utschick: “IDMA vs.

CDMA: analysis and comparison of two multiple access schemes,” IEEE Trans. Wireless Comm., vol. 11, pp. 78-87, Jan. 2012.

[9] L. Ping, Q. Guo and J. Tong: “The OFDM-IDMA

Approach to Wireless Communication System,” IEEE Wireless Communication, pp.18-24, June 2007.

[10] L. Ping, L. Liu and W. Leung: “A simple approach

to near-optimal multiuser detection: interleave-division multiple-access,” in Proc. IEEE Wireless Comm. Networking (WCNC 2003), vol. 1, pp. 391-396, March 2003.

[11] L. Kai Li, W. Xiaodong and L. Ping: “Analysis and

Optimization of Interleave-Division Multiple-Access Communication Systems,” IEEE Trans. on Wireless Communications 2005, pp. 917-920, 2005.

[12] L. Yong, X. Xiong and Z. Luo: “Effect of Carrier

Frequency Offsets on OFDM-IDMA Systems,”

2012 IEEE 2nd International Conference, pp. 209-302, 2012.

[13] J. Dang, F. Qu, Z. Zhang and L. Yang:

“Experimental results on OFDM-IDMA communications with carrier frequency offsets,” OCEANS 2012 – Yeosu, pp. 1-5, 2012.

[14] M.B. Balogun, O.O. Oyerinde, and S.H. Mneney:

“Performance Analysis of the OFDM-IDMA System with Carrier Frequency Offset in a Fast Fading Multipath Channel,” in IEEE 3rd Wireless Vitae Conference, USA, June 24- 27, 2013.

[15] M. Morelli, A. D'Andrea and U. Mengali:

“Feedback frequency synchronization for OFDM applications,” IEEE Communication Letter, vol. 5, pp. 134-136, January 2001.

[16] A. Molisch: “Orthogonal Frequency Division

Multiplexing (OFDM),’’ Wiley-IEEE Press eBook Chapters, second edition, pp. 417–443, 2011.

[17] P. Frenger, P. Orten and T. Ottosson: “Code-spread

CDMA using maximum free distance low-rate convolutional codes,” IEEE Trans. Comm., vol. 48, pp. 135–144, January 2000.

[18] L. Liu, W.K. Leung and L. Ping: “Simple chip-by

chip multi-user detection for CDMA systems,” in Proc. IEEE VTC-Spring, Korea, Apr. 2003, pp. 2157-2161.

[19] Q. Huang, K. King-Kim, P. Wang, L. Ping and S.

Chan: “Interleave-division multiple-access based broadband wireless networks,” Information theory workshop, pp. 502-506, 2006.

[20] L. Ping, L. Liu, K. Wu and W. Leung: “On

interleave-division Multiple-Access,” in IEEE International Conference on Communications, vol. 5, pp. 2869–2873, June 2004.

[21] L. Ping, L. Liu, K. Wu and W.K. Leung:

“Interleaved-Division Multiple-Access.” IEEE Trans. Wireless Communication, vol. 4, pp.938-947, April 2006.

[22] B. Dongming, Y. Xinying: “A new approach for

carrier frequency offset estimation in OFDM communication system,” IEEE communication Tech. Proc., ICCT 2003, vol. 2, pp. 1922-1925, 2003.

[23] J.P.H. Roh and K. Cheun: “An MMSE fine carrier

frequency synchronization algorithm for OFDM systems,” IEEE Trans. Consumer Electronics, vol. 43, no. 3, pp. 761–766, August 1997.


[24] D.N. Godard: “Self-recovering equalization and carrier tracking in two dimensional data communication systems,” IEEE Transactions on Communications, vol. 28, pp. 1867-1875, November 1980.

[25] H. Modaghegh, R.H. Khosravi, S.A. Manesh and

H.S. Yazdi: “A new modeling algorithm – Normalized Kernel Least Mean Square,” IEEE International conference on Innovations in Information technology, IIT 2009, pp. 120-124, 2009.

[26] Y. Xu, B.Sun, C. Zhang, Z. Jin, C. Liu and J. Yang:

“An implementation framework for Kernel methods with high-dimensional pattern,” IEEE Fifth International Conference on Machine Learning and Cybernetics, Dalian, pp. 3271-3276, August 2006.

[27] N. Benvenuto and G. Cherubini: “Algorithms for

communications systems and their applications,” John Wiley and Sons Ltd, The Atrium, Southern Gate, Chichester, West Sussex PO19 8SQ, England, 2002.

[28] M. Morelli, C. Kuo and M. Pun: “Synchronization

techniques for orthogonal frequency division multiple access (OFDMA): a tutorial review,” Proc. IEEE, vol. 95, no. 7, pp. 1394–1427, July 2007.

[29] F. Perez-Cruz and O. Bousquet: “Kernel methods

and their potential use in signal processing,” IEEE Signal Processing Mag., vol. 21, no. 3, pp. 57–65, May 2004.

[30] E. Alameda-Hernandez, D. Blanco, D.P. Ruiz and

M.C. Carrion: “The Averaged, Overdetermined, and Generalized LMS Algorithm,” IEEE Transactions on signal processing, vol. 55, no. 12, pp. 5593-5603, December 2007.

[31] E.A. Lee and D.G. Messersschmitt: Digital Communication, Kluwer Academic Publishers,

Norwell, 1994. [32] W. Liu, P. Pokharel and J. Principe: “The kernel

least-mean-square algorithm,” IEEE Trans. Signal Process., vol. 56, no. 2, pp. 543–554, Feb. 2008.


A FRAMEWORK FOR THE TRANSMISSION OF MULTIMEDIA TRAFFIC USING HM AND RS CODING OVER NAKAGAMI-M CHANNELS T. Quazi*, H. Xu** and A. Saeed*** School of Engineering, Electrical, Electronic and Computer Engineering Building, Howard College Campus, University of KwaZulu-Natal, Durban, 4041, South Africa. *E-mail: [email protected] **E-mail: [email protected] ***E-mail: [email protected] Abstract: Hierarchical modulation (HM) provides different levels of error performance for the high and low priority bits modulated. This differentiation can be aligned to various classes of multimedia traffic demanding different levels of error rate performance. In a previous cross-layer design scheme, which combined conventional modulation at physical layer with RS coding at the application layer, an unequal error protection mechanism was used at the application layer for guaranteeing quality of service for various classes of multimedia traffic. Instead of using only a single layer, in this paper we present a more flexible two layer cross-layer design framework for unequal error protection, combining RS coding at the application layer with HM at the physical layer. Furthermore, in order to improve error performance of the system, a signal space diversity (SSD) scheme is also incorporated into the physical layer of the framework. Without consuming additional transmit power and bandwidth, the use of the SSD HM in the cross-layer design system resulted in significant gains when compared to the non-SSD, standard HM system. It is shown that the cross-layer design system with SSD HM has a 10dB gain over the non-SSD HM system in a Nakagami m=1 channel with a frame error rate target of 10-4. Keywords: Cross-layer design, hierarchical modulation, multimedia traffic, unequal error protection, signal space diversity.

1. INTRODUCTION One of the fundamental requirements of future-generation mobile networks is the support for multimedia traffic such as voice, video and data. Previously these classes of traffic would be transmitted separately but as demand for efficiency increases, so does the need to transmit these various classes of traffic simultaneously in mixed traffic transmission frames [1]. One of the methods considered recently for the simultaneous transmission of various classes of traffic with differentiated levels of quality of service is hierarchical modulation [2]. In HM, a single physical layer bit stream can be separated into several multiplexed sub-streams with different levels of priority and receive different or unequal error protection (UEP), which conventional modulation schemes cannot achieve. Forward error correction (FEC) codes, such as rate compatible punctured convolutional codes (RCPC), turbo codes, low-density parity check (LDPC) codes, have been used for providing UEP to bits of unequal importance in methodologies labelled as joint source channel coding (JSCC). This has primarily been in the application area of image or video transmission [3]. However, unlike such coding schemes HM can support UEP without adding (varying) redundancy to the transmitted data. As an example, if two classes of multimedia traffic are being transmitted using HM: data and voice, the data stream (with a lower error rate requirement) can be inserted to the high priority (HP) sub-stream of the hierarchical modulation scheme while the voice stream (with a higher error rate requirement) can be multiplexed onto the low

priority (LP) sub-stream. HM can also be used to enhance audio/video broadcasting services using its ability to transmit multiple data streams. The HP sub-stream can be used for the base layer of the media transmission while the LP sub-stream can be used for the enhancement layer. All terminals, whether in good or bad channel conditions can receive the base layer, as the only the HP sub-stream can be decoded. However those terminals in good channel conditions can receive both the base and the enhancement layers, as they will be able to decode both the HP and LP sub-streams. This flexible transmission system can be used in media broadcasting systems which allow different resolutions of video display or terminals with variable screen sizes. The cross-layer design (CLD) methodology has been widely used for the transmission of multimedia traffic over the challenging wireless environment. In previous work [4] studied a CLD used an FEC at the application layer to allow for the transmission of mixed classes of traffic in a single physical layer frame. The scheme in [4] used an adaptive modulation scheme to ensure a baseline bit error rate (BER) performance at the physical layer, while the unequal error protection for the three classes of traffic, namely voice, video and data was provided by a Reed-Solomon (RS) coding mechanism at the application layer. Although functional, this scheme, like many proposed UEP enabling schemes, provided UEP using a single layer, namely the application layer. In this paper, we propose a more flexible two layered unequal error protection framework in which the RS coding UEP mechanism at the application layer is supplemented with


a second UEP mechanism at the physical layer where the efficient UEP capabilities of HM is applied. The reason for the use RS coding in the CLD mechanism over other FEC coding schemes such as LDPC codes is that it is more suited for use in the applications layer as compared to LDPC codes which is more practically applied in the physical layer. The LDPC decoder requires the use of soft information regarding the channel and it is better placed to retrieve this at the physical layer. In addition, LDPC codes, unlike RS codes, are not maximum distance separable (MDS) codes and require a long code length to be effective. RCPC is another FEC scheme that could be considered, but it also requires channel information in order to achieve good performance. RS coding does not have this requirement and is thus used for the proposed UEP CLD scheme. Without consuming any additional bandwidth, transmit power or space, signal space diversity is a technique which takes advantage of the intrinsic diversity of multi-dimensional constellations to improve BER performance [5]. Thus the system error performance of the proposed CLD scheme is further enhanced by the incorporation of SSD into the physical layer. It will be shown that this significantly improves the error rate performance at the physical layer and consequently the performance of the overall CLD scheme.

2. RELATED WORK The CLD schemes for the handling of multimedia traffic have traditionally focused on the transmission of scalable video or progressive image data over varying channel wireless links [3]. The concept is to encode the source video/image data such that there is a decreasing level of importance and then to provide UEP depending on the unequal importance. Various types of coding schemes have been used for the purposes of designing joint source channel coding video/image multimedia traffic transmission [3]. There have been some schemes proposed that use modulation assisted JSCC such as the one in [6] which uses a three level LDPC code and a quadrature phase-shift keying (QPSK)/16 quadrature amplitude modulation (16QAM) combination for providing UEP. The limitation of using coding schemes for providing the wide range of UEP required for multimedia traffic is that, consequently, a wide range of redundant data has to be added to the transmitted data. This limitation can be overcome by the use of hierarchical modulation which can achieve UEP without adding redundancy [1]. HM is a popular strategy for providing UEP and has been included in various standards such as Digital Video Broadcast-Terrestrial (DVB-T). HM has been combined with channel coding for UEP in the transmission of progressive image in [3] and data partitioned video in [7]. HM has also been included in the DVB-H standard, and [8] presents a study of HM combined with scalable video coding (SVC) for Mobile TV. These schemes restrict the

definition of multimedia traffic to video and image data. Although the multimedia traffic is more generalized in [1] and [2] which use HM for UEP, these schemes only focus on the physical layer, while the CLD framework studied in this work looks at combining two layers where an application layer FEC scheme is combined with HM at the physical layer for providing UEP for various classes of multimedia traffic. Furthermore in a recent study, [9] has shown that incorporating an SSD mechanism to HM significantly improves the system BER performance in a fading channel. Thus the same SSD is combined with HM at the physical layer of the CLD scheme to improve its overall error rate performance. A study of a LDPC coded HM system has been presented in [10-11]. There is a strong coupling between the HM modulator/demodulator and the LCPC encoder/decoder of the scheme in the studies, and since the LDPC decoder requires soft information from the channel, the entire system is practically implemented in the physical layer where such information is readily available. An application and physical layer cross-layer UEP scheme is presented in [12]. The FEC code used at the application layer is based on RS codes, and HM is used at the physical layer, however the system restricts the traffic to video data only while in this work multimedia traffic is considered more generally. The system proposed in this paper also incorporates the SSD HM mechanism to the physical layer for improved error rate performance. SSD is considered an important enhancement technology for the next generation DVB standard, namely DVB-T2 [13]. Constellation rotation has been combined with HM in [14-15], however in both these systems, the objective is to increase the error performance of the LP sub-stream transmission, whereas, the SSD HM scheme in this paper considers both sub-streams. In summary, none of the schemes listed here considers a system that combines an application layer FEC scheme with SSD enhanced HM scheme as it is done in the proposed system. The remainder of the paper is organized as follows. Section 3 describes the system model and the two configurations of the proposed CLD framework. In the first configuration the application layer FEC coding scheme is combined with a typical HM scheme while in the second it is combined with a SSD HM scheme. Section 4 presents the theoretical system performance which is then verified using simulation results. Section 5 concludes the paper.

3. SYSTEM MODEL The architecture for the two layer UEP CLD mechanism is shown in Figure 1.


Nakagami-mFading Channel

HP-Stream LP-Stream HP-Stream LP-Stream

RSEncoder

RSEncoder

RSDecoder

RSDecoder

H Modulator H Demodulator

Figure 1 HM and RS CLD system model. The multimedia traffic is generated by different applications and each stream will have its own error performance requirement. The structure in Figure 1 shows a high and low priority stream (labeled HP and LP respectively), but this could be generalized to accept more streams and the system would then be appropriately scaled. Each stream is then sent to the application layer FEC module where each stream is encoded by an RS encoder to produced RS code blocks/frames. The frames are sent next to the physical layer where the proposed system has two possible configurations. In the first configuration, the data from the RS encoder is modulated using the standard hierarchical quadrature amplitude modulation (HQAM), while in the second a SSD HQAM mechanism is used. The modulated signal is then transmitted over a Nakagami-m fading channel. At the receiver, the (generalized) received signal is given by 𝑟𝑟 = ℎ𝑢𝑢 + 𝑛𝑛 where 𝑢𝑢 is the transmitted signal, 𝑛𝑛 the additive white Gaussian noise which is modeled as a zero-mean complex Gaussian random variable with variance 𝑁𝑁0/2 per dimension, where 𝑁𝑁0 is the single-sided power spectrum density of the noise; and ℎ is the fading coefficient. The amplitude of the fading is modeled as the Nakagami-m distributed random variable according to

𝑓𝑓(𝛽𝛽) = 2Γ(𝑚𝑚) (

𝑚𝑚𝛺𝛺)

𝑚𝑚𝛽𝛽2𝑚𝑚−1𝑒𝑒𝑒𝑒𝑒𝑒 (−𝑚𝑚

𝛺𝛺 𝛽𝛽2) (1)

where 𝑚𝑚 is the Nakagami fading parameter and Γ(𝑚𝑚) is the Gamma function defined by Γ(𝑚𝑚) = ∫ 𝑦𝑦𝑚𝑚−1𝑒𝑒𝑒𝑒𝑒𝑒(−𝑦𝑦)𝑑𝑑𝑦𝑦∞0 and 𝛺𝛺 = 𝐸𝐸[|ℎ|2] is the average

fading power, 𝐸𝐸 being the expectation. The received signal is demodulated at the physical layer which leads to the extraction of the HP and LP symbol streams which are then passed to the respective RS decoders for the decoding of the particular traffic data frames. In the next two subsections the two physical layer mechanisms used in the proposed CLD scheme, namely HM and the HM with SSD, are overviewed. 3.1 Hierarchical Modulation The system in HQAM assumes two separate input bit streams carrying information of unequal importance. In each HQAM symbol, 2 HP bits and 𝑙𝑙𝑙𝑙𝑙𝑙2(𝑀𝑀) − 2 LP bits

(𝑀𝑀 is the QAM modulation order) are arranged as follows. The HP bits are assigned to the most significant bit position in the in-phase and quadrature phase of the symbol. The remaining bits positions in the in-phase are assigned 12 𝑙𝑙𝑙𝑙𝑙𝑙2(𝑀𝑀) − 1 LP bits while the other half of the LP bits are assigned to the quadrature-phase. Figure 2 shows Gray encoded HQAM constellation using a 4/16-HQAM system. The HP sub-stream will be transmitted using the 2 MSBs while the LP sub-stream will be transmitted with the 2 low significant bits (LSBs). The level of relative unequal protection between the two streams is controlled by varying two distance parameters: the distance between two symbols in adjacent quadrants and the distance between two neighbouring symbols in the same quadrant (2𝑎𝑎 and 2𝑏𝑏 in Figure 2 respectively). The ratio 𝛼𝛼 = 𝑎𝑎

𝑏𝑏 determines the hierarchy of bit priority and thus the level of UEP, and is commonly referred to as the constellation or hierarchical parameter.

Figure 2 4/16-QAM Hierarchical constellations. The demodulation process is done by treating each quadrature of the received 4/16-HQAM as a PAM signal and using the decision boundaries as shown in Figure 2, with the solid line for the HP bits and the dashed line for the LP bits [16]. 3.2 SSD using Rotated HQAM In a recent work, [9] has studied the performance of HM with signal space diversity (SSD) for single and multiple antenna configurations. SSD exploits the intrinsic diversity of multidimensional constellation such as HQAM to further improve their error rate performance, and this is done without the cost of additional bandwidth, power or antennas. The only cost of the performance improvement is the increase in complexity at the receiver where maximum-likelihood (ML) detection is required. The procedure for the SSD scheme is shown in Figure 3.

0001

0010

0011

1010 1000

1011 1001

0101

0100

0111

0110

1111 1101

1110 1100

0000

2𝑎𝑎

2𝑏𝑏


Nakagami-mFading Channel

HQAM symbols

Rotate

Interleave

HQAM symbols

ML Detect andun-rotate

De-interleave

Figure 3 SSD Procedure.

This procedure follows that which is shown in Figure 1. The HQAM symbols 𝑥𝑥𝑖𝑖 = [𝑥𝑥𝑖𝑖𝐼𝐼 𝑥𝑥𝑖𝑖

𝑄𝑄], which have the RS encoded bits, are first rotated to produce a rotated version of the input symbol stream: 𝑥𝑥��𝑖 = 𝑥𝑥𝑖𝑖𝑅𝑅𝜃𝜃. 𝑥𝑥𝑖𝑖𝐼𝐼 and 𝑥𝑥𝑖𝑖

𝑄𝑄 represent the in-phase and quadrature-components of the HQAM signal and 𝑅𝑅𝜃𝜃 is given by [9]

𝑅𝑅𝜃𝜃 = [ cos 𝜃𝜃 sin𝜃𝜃−sin𝜃𝜃 cos 𝜃𝜃] (2)

The optimum rotation angle 𝜃𝜃𝑜𝑜𝑜𝑜𝑜𝑜 was derived in [9] by maximizing the minimum distance between the unique components of the rotated constellation points. The final expression of the optimization, as shown in (3), is a function of the distance parameters 𝑎𝑎 and 𝑏𝑏 via the hierarchical parameter 𝛼𝛼.

tan𝜃𝜃𝑜𝑜𝑜𝑜𝑜𝑜 = 𝛼𝛼𝛼𝛼 + 3 (3)

The rotated symbol stream is then arbitrarily grouped into pairs and passed through a component interleaver where the in-phase and the quadrature-phase components in each pair are interleaved to produce new symbol pairs 𝑢𝑢1and 𝑢𝑢2 as shown in (4).

𝑢𝑢1 = ��𝑥1𝐼𝐼 + 𝑗𝑗��𝑥2𝑄𝑄 𝑢𝑢2 = ��𝑥2𝐼𝐼 + 𝑗𝑗��𝑥1𝑄𝑄

(4)

The symbols 𝑢𝑢1and 𝑢𝑢2 are transmitted on two separate time slots ensuring that each is sent over an orthogonal channel where the transmission is affected by fading and is perturbed by noise. The corresponding signal at the receiver is given by

𝑟𝑟𝑖𝑖 = ℎ𝑖𝑖𝑢𝑢𝑖𝑖 + 𝑛𝑛𝑖𝑖 𝑖𝑖 ∈ {1,2} (5)

where 𝑖𝑖 ∈ {1,2} is the orthogonal channel time slot index, and ℎ𝑖𝑖 and 𝑛𝑛𝑖𝑖 are the fading coefficient and the noise parameter respectively for the corresponding channel index. De-interleaving is performed on a symbol pair to reassemble the in-phase and quadrature-phase components of the original symbols. The symbols are then passed to the ML detector where the availability of full channel state information (CSI) is assumed. The detected signal is thereafter un-rotated using the transpose of the rotation matrix (2) to produce un-rotated HQAM symbols.

4. PERFORMANCE ANALYSIS

The block diagram of the proposed system in Figure 1 shows that the beginning and end process of the CLD system at the transmitter and receiver is the RS encoder and decoder respectively, thus the system performance is quantified as a function of the HP and LP stream frame error rates (FER) and the first system parameter that affects this is the level protection by the RS encoder at the application layer. The level of error protection provided is determined by number of parity symbols inserted in the RS code frame. The frame error rate (FER) at the receiver is given by

𝑃𝑃𝑓𝑓 = ∑ (𝑁𝑁𝑘𝑘) (1 − 𝑃𝑃𝑠𝑠)𝑁𝑁−𝑘𝑘𝑃𝑃𝑠𝑠𝑘𝑘𝑁𝑁

𝑘𝑘=𝑜𝑜+1 (6)

where each RS code frame has 𝑘𝑘 message symbols to which are appended 𝑁𝑁 − 𝑘𝑘 parity symbols. The correcting potential is determined by the parameter 𝑡𝑡 (𝑡𝑡 = 𝐹𝐹 [𝑁𝑁−𝑘𝑘2 ], 𝐹𝐹[∙] is the floor function). Since each RS decoder at the receiver only produces one type/stream of bits i.e. either HP or LP bits, the RS symbol error calculation is only dependent on the BER of one of the streams of data. Thus the symbol error rate (𝑃𝑃𝑠𝑠) is given by

𝑃𝑃𝑠𝑠 = 1 − (1 − 𝑃𝑃𝑏𝑏)𝐿𝐿 (7)

where 𝑃𝑃𝑠𝑠 represents the symbol error rate for either a HP symbol or LP symbol as shown in Figure 1, 𝑃𝑃𝑏𝑏 the bit error rate of the corresponding HP or LP bits and 𝐿𝐿 the number of bits in the RS symbol. The bits in the RS symbol are assumed to be independent. The physical layer BER achieved is the second parameter that affects the system FER performance and derivations of the two configurations of the proposed system is presented in the following two subsections. Note that FEC is implemented only at the application layer and there is no error correcting coding at the physical layer. 4.1 HQAM BER The BER of the HP and LP bits of a 4/16-QAM has been analysed in [16]. The BER of HP bits is given by

𝑃𝑃𝐻𝐻𝐻𝐻(𝑀𝑀) = 12 [𝑃𝑃𝑏𝑏(𝑖𝑖1,𝑀𝑀) + 𝑃𝑃𝑏𝑏(𝑞𝑞1,𝑀𝑀)] = 𝑃𝑃𝑏𝑏(𝑖𝑖1,𝑀𝑀) (8)

where 𝑃𝑃𝑏𝑏(𝑖𝑖𝑘𝑘 ,𝑀𝑀) and 𝑃𝑃𝑏𝑏(𝑞𝑞𝑘𝑘 ,𝑀𝑀) give the BER of the 𝑘𝑘th in-phase and quadrature phase bits of a 4/16-QAM constellation respectively. The BER of the LP bits is derived by averaging over all the remaining in-phase and quadrate bit as shown in (9).

𝑃𝑃𝐿𝐿𝐻𝐻(𝑀𝑀) =2∑ 𝑃𝑃𝑏𝑏(𝑖𝑖𝑘𝑘 ,𝑀𝑀)(12) log2 𝑀𝑀

𝑘𝑘=2(log2 𝑀𝑀 − 2) (9)

The general BER expressions, 𝑃𝑃𝑏𝑏(𝑖𝑖𝑘𝑘 ,𝑀𝑀), for square 4/M-QAM constellations were derived in [9] and are reported in the Appendix for convenience. The above method from [16] for determining the BER of HP and LP bits in HQAM used the probability that a transmitted symbol’s component exceeds a set boundary. This approach, which is an adaptation of the approach used for standard pulse amplitude modulation (PAM) schemes, is easily applied for standard HQAM where the


components for the symbols have regular distances from the boundaries. However when HQAM with SSD is considered, the HQAM constellations are rotated these distances are no longer regular and thus applying such an approach becomes complicated. In order to derive closed BER expressions for HQAM with SSD, [9] presented accurate approximations for the respective BERs using a simplified nearest neighbourhood (NN) union bound approach. The derivations refer to Figure 4 and for convenience we include the final expressions here.

xAxB

xC (a+2b, a+2b)(a, a+2b)(-a, a+2b)

(a, a)

(-a, -a)

(a, a+2b)(-a, a)

10101000

10011011

00000010

00110001

11111101

11001110

01010111

01100100

(-a-2b, a)

(-a-2b, a+2b)

xDxE

Figure 4 Rotated hierarchical 16-QAM Nearest

Neighbour error analysis. The BER for the HP bits are given by

𝑃𝑃𝐻𝐻𝐻𝐻 = 12 𝑃𝑃(𝑋𝑋𝐵𝐵 → 𝑋𝑋𝐶𝐶) + 1

4 𝑃𝑃(𝑋𝑋𝐵𝐵 → 𝑋𝑋𝐷𝐷) + 14 𝑃𝑃(𝑋𝑋𝐸𝐸 → 𝑋𝑋𝐶𝐶). (10)

𝑃𝑃(𝑋𝑋𝐵𝐵 → 𝑋𝑋𝐶𝐶) represents the pairwise error probability (PEP) of perpendicular neighbhouring symbols of the constellation set and is given by (B1) in the Appendix. 𝑃𝑃(𝑋𝑋𝐵𝐵 → 𝑋𝑋𝐷𝐷) and 𝑃𝑃(𝑋𝑋𝐸𝐸 → 𝑋𝑋𝐶𝐶) represent the PEP of diagonal neighbouring symbols is given by (B2) and (B3). The BER for the LP bits are given by

𝑃𝑃𝐿𝐿𝐻𝐻 = 𝑃𝑃(𝑋𝑋𝐴𝐴 → 𝑋𝑋𝐵𝐵), (11)

where 𝑃𝑃(𝑋𝑋𝐴𝐴 → 𝑋𝑋𝐵𝐵) represents the PEP that a transmitted symbol is detected as a perpendicular neighbour within the same quadrant of the received signal and is given by (B4) in the Appendix. In order to verify the approximate BER expressions derived using the NN approach, we compared the BERs for HP and LP bits using (10) and (11) for standard, un-rotated HQAM with those derived in [16], namely (8) and (9). The Nakagami m parameter used for the comparison is 1 and the hierarchical parameter 𝛼𝛼 used is 2. It can be seen from the curves in Figure 5 that there is a close match between the approximate expressions and those derived in [16]. Thus (10) and (11) are used for determining the theoretical BERs for the HP and LP bits for the standard and the rotated HQAM systems in the following section.

Figure 5 Comparison of two approaches for BER.

5. SIMULATIONS AND DISCUSSION

The proposed system was implemented in Matlab and was simulated using an accurate Nakagami-m channel model [17]. In this section, the simulation results are compared to the theoretical expressions for the two system configurations, namely RS coding combined with standard HQAM and with rotated HQAM, and it is shown that there is a close match between the two sets of results for both configurations. The parameters varied in the simulation setups are the Nakagami m value, the 𝛼𝛼 value determining the level of hierarchy in HQAM and the 𝑡𝑡 value for varying the level of protection provided by the RS coder. 5.1 RS coded HQAM without SSD The frame error rate theoretical performance of the two layer UEP CLD system with RS and standard HQAM is shown in Figure 7-8. However, in order to demonstrate the physical layer UEP contribution of the system, the BER performance of the HM layer of the CLD is shown in Figure 6.

Figure 6 BER of the UEP modulation layer of the CLD

with Nakagami-m, m=2, 𝛼𝛼 =3.

5 10 15 20 25 30 35 40 4510

-5

10-4

10-3

10-2

10-1

100

Eb/No, dB

Bit

Err

or R

ate

LP - Decision Boundary ApproachLP - NN ApproachHP - Decision Boundary ApproachHP - NN Approach

5 10 15 20 25 3010

-6

10-5

10-4

10-3

10-2

10-1

100

Eb/No, dB

Bit

Err

or R

ate

Theoretical BER LPSimulation BER LPTheoretical BER HPSimulation BER HP


In this setup, as well as that for Figure 7, the hierarchical parameter 𝛼𝛼, which represents the level of UEP between the HP and LP streams is 3. The Nakagami m parameter used for the experiments for Figure 6-8 is 2. As the curves in Figure 6 show, the BER performance for the HP bit stream is much lower than that of the LP bit stream throughout the entire SNR range.

Figure 7 FER of HP and LP streams for two layer UEP

CLD in Nakagami-m, m=2, 𝛼𝛼 =3. The theoretical FER graphs shown in Figures 7 now show the FERs of both the HP and LP streams which are now unequally protected both via the hierarchical modulation as well the RS forward error correcting mechanism. The values 𝑁𝑁 and 𝑘𝑘 for the RS encoder is set to 15 and 13 respectively, and thus the FER curves represent the frame error rate for the system with and without adding 2 parity symbols (𝑡𝑡 = 1) to the RS code block. In order to highlight the flexibility of the proposed two layer UEP mechanism, the RS and HM parameters are varied while keeping the channel parameter the same and the results are shown in Figure 8. The hierarchical parameter 𝛼𝛼 used in this setup is 2, so this layer provides a lower level of UEP at the physical layer, but the RS parameter 𝑡𝑡 for the HP sub-stream is increased from 1 to 2, thus 𝑘𝑘 is changed from 13 to 11 while 𝑁𝑁 equals 15. This significantly reduces the FER for the HP sub-stream as can be seen from gradient of the ‘HP – RS, t = 2’ theory and simulation curves which are much steeper as a result of the greater gain relative to the ‘LP – RS, t = 1’ curves. The combination of the RS coding to the already unequal error protection provided by the HM layer ensures that the HP protected further, thereby increasing the level of unequal error protection achieved. Thus the variable mechanism of the proposed two layer CLD can be used to provide further unequal protection to certain streams of data over others than previous single layered mechanisms.

Figure 8 FER of HP and LP streams for two layer UEP

CLD in Nakagami-m, m=2, 𝛼𝛼 =2. 5.2 RS coded HQAM with SSD The effect of incorporating the SSD scheme to the physical layer of the two layer CLD mechanism is presented in this subsection. The graphs in Figure 9 compares the BER of rotated HQAM with the standard HQAM modulation and shows that there are significant gains (approximately 7 dB at BER target of 10-3 ) as a consequence of adding the SSD scheme. The hierarchical parameter 𝛼𝛼 is set to 3 and the Nakagami m fading parameter is 1.

Figure 9 BER of the UEP modulation layer of the CLD

with Nakagami-m, m=1, 𝛼𝛼 =3. The FER performance improvement for the two layer CLD system, with RS coding at the application layer and standard HQAM at the physical layer, was demonstrated in Figures 7 and 8 in the previous subsection. In this subsection, Figures 10 and 11 show the FER curves for the CLD system, with and without the SSD mechanism incorporated at the physical layer. In each case 𝑡𝑡 is equal to 1 and 𝛼𝛼 = 3 but the Nakagami m value is 1 and 2 for Figures 10 and 11 respectively. As can be seen in both figures, the incorporation of the HM SSD scheme to the CLD system results in gains throughout the SNR range when compared to the system without the SSD

5 10 15 20 25 30 35 40 4510

-6

10-5

10-4

10-3

10-2

10-1

100

Eb/No, dB

Fram

e E

rror

Rat

e

Theoretical FER HP - RS,t=0Theoretical FER LP - RS,t=0Theoretical FER HP - RS,t=1Theoretical FER LP - RS,t=1Simulation FER - RS,t=1

5 10 15 20 25 30 35 40 4510

-6

10-5

10-4

10-3

10-2

10-1

100

Eb/No, dB

Fram

e E

rror

Rat

e

Theoretical FER HP - RS,t=0Theoretical FER LP - RS,t=0Theoretical FER HP - RS,t=2Theoretical FER LP - RS,t=1Simulation FER - RS,t=1Simulation FER - RS,t=2

5 10 15 20 25 30 35 40 4510

-6

10-5

10-4

10-3

10-2

10-1

100

Eb/No, dB

Bit

Err

or R

ate

Standard HQAM LPRotated HQAM LPStandard HQAM HPRotated HQAM HPSimulation BER


mechanism implemented. To get a quantified value for the improvement, if a FER target of 10-4 for the HP sub-stream is taken, the FER curves in Figure 10 (which is for a Nakagami m=1 channel with HM hierarchy 𝛼𝛼 = 3) show that there is a gain of approximately 10dB for the CLD system with SSD HQAM system at the physical layer as compared to the system with the standard HQAM system at the physical layer, and this gain is achieved without the system using any additional transmit power or bandwidth. Another observation that can be made by comparing the graphs in Figures 10 and 11 is that the gain produced by the SSD HQAM system reduces as the channel condition improves. If an FER target of 10-4 for the HP sub-stream is once again taken, the graphs for the HP sub-stream in Figure 11 (which is for a Nakagami m=2 channel with HM hierarchy 𝛼𝛼 = 3) shows that the gain of the SSD HQAM system over the standard HQAM system is now approximately 3dB. This result is to be expected as the diversity benefits of the SSD scheme reduce as the channel conditions improve, thus conversely, the two layer UEP CLD system with SSD HQAM performs better than the standard HQAM system under deep fading channel conditions.

Figure 10: FER of HP and LP streams for two layer UEP CLD in Nakagami-m, m=1, 𝛼𝛼 =3.

Figure 11: FER of HP and LP streams for two layer UEP

CLD in Nakagami-m, m=2, 𝛼𝛼 =3.

6. CONCLUSION

In this study a framework for a two layer UEP CLD system using RS coding at the application layer coupled with HQAM at the physical layer was presented. It was shown how such a mechanism can provide flexible variable UEP for the transmission of multimedia traffic in a single frame by varying parameters that affect UEP both at the application and physical layers. Furthermore, the inclusion of the SSD HQAM mechanism at the physical layer of the CLD system led to noticeable gains as it was shown that for a FER target of 10-4 over a Nakagami m=1 channel, the HM SSD system had a 10dB gain over the standard HM system. The results also showed that these gains were reduced when the channel conditions improved and this leads to the conclusion that the SSD HQAM incorporated CLD system is more effective than the standard HQAM system under deep fading channel conditions. In future work, the proposed system’s various adaptable parameters, such as the level of RS error correcting capability; the level of hierarchy using HQAM, will be adapted to channel conditions to design an energy efficient, QoS aware, adaptive cross-layer design system for the transmission of multimedia traffic over fading channels.

APPENDIX

A. BER expressions for the Decision Boundary Approach The general BER expressions, 𝑃𝑃𝑏𝑏(𝑖𝑖𝑘𝑘 ,𝑀𝑀), used in (8) and (9) as derived in [16] is shown in (A1).

𝑃𝑃𝑏𝑏(𝑖𝑖𝑘𝑘 ,𝑀𝑀) = 𝑃𝑃𝑏𝑏(𝑞𝑞𝑘𝑘 ,𝑀𝑀) = 1√𝑀𝑀

∑ 𝐼𝐼(1, 2𝑗𝑗)(√𝑀𝑀2 )−1𝑗𝑗=0 , (A1)

𝐹𝐹𝐹𝐹𝐹𝐹 𝑘𝑘 > 1: 𝑃𝑃𝑏𝑏(𝑖𝑖𝑘𝑘 ,𝑀𝑀) = 𝑃𝑃𝑏𝑏(𝑞𝑞𝑘𝑘 ,𝑀𝑀)

= 1√𝑀𝑀

[ ∑ (−1)𝐹𝐹[𝑗𝑗𝑗𝑗(𝑘𝑘,𝑀𝑀)]𝐽𝐽(𝑘𝑘,𝑀𝑀)−1

𝑗𝑗=0

× [2𝑘𝑘−1 − 2𝐹𝐹 [𝑗𝑗 ∗ 𝑍𝑍(𝑘𝑘,𝑀𝑀) + 12]]

× 𝐼𝐼(0, 2𝑗𝑗 + 1) + ∑ (−1)𝐹𝐹[𝑗𝑗𝑗𝑗(𝑘𝑘,𝑀𝑀)]𝐽𝐽(𝑘𝑘,𝑀𝑀)−1

𝑗𝑗=0

× 𝐹𝐹 [𝑗𝑗 × 𝑍𝑍(𝑘𝑘,𝑀𝑀) + 12] × 𝐼𝐼(2, 2𝑗𝑗 − 1)

+ ∑ (−1)𝐹𝐹[𝑗𝑗𝑗𝑗(𝑘𝑘,𝑀𝑀)]𝐾𝐾(𝑘𝑘,𝑀𝑀)

𝑗𝑗=𝐽𝐽(𝑘𝑘,𝑀𝑀)

× [2𝑘𝑘−1 − 𝐹𝐹 [𝑗𝑗 × 𝑍𝑍(𝑘𝑘,𝑀𝑀) + 12]]

× 𝐼𝐼(2, 2𝑗𝑗 − 1)]

(A2)

where 𝐹𝐹[∙] is the floor function, 𝑍𝑍(𝑘𝑘,𝑀𝑀) = 2𝑘𝑘−1

√𝑀𝑀 ,

𝐾𝐾(𝑘𝑘,𝑀𝑀) = √𝑀𝑀(1 − 2−𝑘𝑘) − 1, 𝐽𝐽(𝑘𝑘,𝑀𝑀) = √𝑀𝑀 (12 −12𝑘𝑘),

𝐼𝐼(𝑎𝑎, 𝑏𝑏) = 𝑒𝑒𝐹𝐹𝑒𝑒𝑒𝑒(√𝐺𝐺(𝑎𝑎, 𝑏𝑏;𝛼𝛼,𝑀𝑀)𝛾𝛾) and

5 10 15 20 25 30 35 40 4510

-6

10-5

10-4

10-3

10-2

10-1

100

Eb/No, dB

Fram

e E

rror

Rat

e

Theoretical FER LP - Standard HQAM+RS,t=1Theoretical FER LP - Rotated HQAM+RS,t=1Theoretical FER HP - Standard HQAM+RS,t=1Theoretical FER HP - Rotated HQAM+RS,t=1Simulation FER - RS, t=1

5 10 15 20 25 30 35 40 4510

-6

10-5

10-4

10-3

10-2

10-1

100

Eb/No, dB

Fram

e E

rror

Rat

e

Theoretical FER LP - Standard HQAM+RS,t=1Theoretical FER LP - Rotated HQAM+RS,t=1Theoretical FER HP - Standard HQAM+RS,t=1Theoretical FER HP - Rotated HQAM+RS,t=1Simulation FER - RS, t=1


𝐺𝐺(𝑎𝑎, 𝑏𝑏;𝛼𝛼,𝑀𝑀) =(𝑎𝑎 + 𝛼𝛼𝑏𝑏)2

[2 [1 + 𝛼𝛼 [√𝑀𝑀2 − 1]]2

+ 23 [𝑀𝑀4 − 1] 𝛼𝛼2]

(A3)

The above expressions, which are for additive white Gaussian noise (AWGN) channels, were extended to Nakagami-m fading channels in [16]. These expressions (10) and (11) are valid for Nakagami-m fading channels with the modification to the 𝐼𝐼(𝑎𝑎, 𝑏𝑏) function shown in (13), where ��𝛾 is the average signal-to-noise ratio (SNR).

𝐼𝐼(𝑎𝑎, 𝑏𝑏) = √𝐺𝐺(𝑎𝑎,𝑏𝑏;𝛼𝛼,𝑀𝑀)��𝛾𝜋𝜋 ∗ 𝑚𝑚𝑚𝑚

(𝑚𝑚 + 𝐺𝐺(𝑎𝑎, 𝑏𝑏;𝛼𝛼,𝑀𝑀)��𝛾)𝑚𝑚+12

∗Γ (𝑚𝑚 + 1

2)Γ(𝑚𝑚 + 1) ∗ 2𝐹𝐹1 (1,𝑚𝑚 + 1

2 ;𝑚𝑚

+ 1; 𝑚𝑚𝑚𝑚 + 𝐺𝐺(𝑎𝑎,𝑏𝑏;𝛼𝛼,𝑀𝑀)��𝛾)

(A4)

where .2 𝐹𝐹1(. , . ; . ; ) is the Gauss hyper-geometric function, 𝑚𝑚 is the Nakagami fading parameter and Γ(∙) is the gamma function. B. PEP expressions for the Nearest Neighbourhood Approach The pairwise error probability (PEP) of perpendicular neighbhouring symbols in adjacent quadrants, 𝑃𝑃(𝑋𝑋𝐵𝐵 →𝑋𝑋𝐶𝐶), is given by 𝑃𝑃(𝑋𝑋𝐵𝐵 → 𝑋𝑋𝐶𝐶)

= 14𝑛𝑛 (

2𝑚𝑚2𝑚𝑚 + ��𝛾𝛽𝛽1 cos2 𝜃𝜃)

𝑚𝑚( 2𝑚𝑚

2𝑚𝑚 + ��𝛾𝛽𝛽1 sin2 𝜃𝜃)𝑚𝑚

+ 12𝑛𝑛 ∑ ( 𝑚𝑚𝑆𝑆𝑘𝑘

𝑚𝑚𝑆𝑆𝑘𝑘+��𝛾𝛽𝛽1 cos2 𝜃𝜃)𝑚𝑚( 𝑚𝑚𝑆𝑆𝑘𝑘𝑚𝑚𝑆𝑆𝑘𝑘+��𝛾𝛽𝛽1 sin2 𝜃𝜃

)𝑚𝑚𝑛𝑛−1

𝑘𝑘=1 ,

(B1)

where ��𝛾 is the average SNR per hierarchical 16-QAM symbol, 𝜃𝜃 is the optimum rotation angle from (3), 𝑆𝑆𝑘𝑘 =2 sin2(𝑘𝑘𝜋𝜋/2𝑛𝑛), 𝛽𝛽1 = 𝛼𝛼2

𝛼𝛼2+2𝛼𝛼+2 and 𝑛𝑛 is the number of summations. 𝑛𝑛 greater than 6 will result in an accurate approximation. The PEP of diagonal neighbouring symbols, 𝑃𝑃(𝑋𝑋𝐵𝐵 → 𝑋𝑋𝐷𝐷) and 𝑃𝑃(𝑋𝑋𝐸𝐸 → 𝑋𝑋𝐶𝐶), is given by (B2) and (B3) respectively. 𝑃𝑃(𝑋𝑋𝐵𝐵 → 𝑋𝑋𝐷𝐷) = 1

4𝑛𝑛 (2𝑚𝑚

2𝑚𝑚+��𝛾𝛽𝛽2(𝛼𝛼cos𝜃𝜃−sin𝜃𝜃)2)𝑚𝑚( 2𝑚𝑚2𝑚𝑚+��𝛾𝛽𝛽2(𝛼𝛼sin+cos𝜃𝜃)2)

𝑚𝑚+

12𝑛𝑛∑ ( 𝑚𝑚𝑆𝑆𝑘𝑘

𝑚𝑚𝑆𝑆𝑘𝑘+��𝛾𝛽𝛽2(𝛼𝛼cos𝜃𝜃−sin𝜃𝜃)2)𝑚𝑚( 𝑚𝑚𝑆𝑆𝑘𝑘𝑚𝑚𝑆𝑆𝑘𝑘+��𝛾𝛽𝛽2(𝛼𝛼sin+cos𝜃𝜃)2)

𝑚𝑚𝑛𝑛−1𝑘𝑘=1 ,

(B2)

𝑃𝑃(𝑋𝑋𝐸𝐸 → 𝑋𝑋𝐶𝐶)= 1


2𝑚𝑚 + ��𝛾𝛽𝛽2(𝛼𝛼cos𝜃𝜃 + sin𝜃𝜃)2)𝑚𝑚( 2𝑚𝑚

2𝑚𝑚 + ��𝛾𝛽𝛽2(𝛼𝛼sin − cos𝜃𝜃)2)𝑚𝑚

+ 12𝑛𝑛∑ ( 𝑚𝑚𝑆𝑆𝑘𝑘

𝑚𝑚𝑆𝑆𝑘𝑘 + ��𝛾𝛽𝛽2(𝛼𝛼cos𝜃𝜃 + sin𝜃𝜃)2)𝑚𝑚( 𝑚𝑚𝑆𝑆𝑘𝑘𝑚𝑚𝑆𝑆𝑘𝑘 + ��𝛾𝛽𝛽2(𝛼𝛼sin − cos𝜃𝜃)2)

𝑚𝑚,

𝑛𝑛−1

𝑘𝑘=1

(B3)

where 𝛽𝛽2 = 1𝛼𝛼2+2𝛼𝛼+2.

The PEP of a perpendicular neighbour within the same quadrant, (𝑋𝑋𝐴𝐴 → 𝑋𝑋𝐵𝐵), is given by (B4). 𝑃𝑃(𝑋𝑋𝐴𝐴 → 𝑋𝑋𝐵𝐵)= 1


2𝑚𝑚 + ��𝛾𝛽𝛽2 cos2 𝜃𝜃)𝑚𝑚( 2𝑚𝑚

2𝑚𝑚 + ��𝛾𝛽𝛽2 sin2 𝜃𝜃)𝑚𝑚

+ 12𝑛𝑛∑( 𝑚𝑚𝑆𝑆𝑘𝑘

𝑚𝑚𝑆𝑆𝑘𝑘 + ��𝛾𝛽𝛽2 cos2 𝜃𝜃)𝑚𝑚( 𝑚𝑚𝑆𝑆𝑘𝑘𝑚𝑚𝑆𝑆𝑘𝑘 + ��𝛾𝛽𝛽2 sin2 𝜃𝜃)

𝑚𝑚𝑛𝑛−1

𝑘𝑘=1

(B4)

REFERENCES

[1] J. Hossain, P.K. Vitthaladevuni, M.-S. Alouini, V.K. Bhargava, A.J. Goldsmith, “Adaptive hierarchical modulation for simultaneous voice and multiclass data transmission over fading channels,” IEEE Transactions on Vehicular Technology, vol. 55, issue 4, pp. 1181 – 1194, July, 2006.

[2] H. Mukhtar, M. El-Tarhuni, “An Adaptive Hierarchical QAM Scheme for Enhanced Bandwidth and Power Utilization,” IEEE Transactions on Communications, vol. 60, issue 8, pp. 2275 – 2284, August, 2012.

[3] S.S. Arslan, P.C. Cosman, L.B. Milstein, “Coded Hierarchical Modulation for Wireless Progressive Image Transmission” IEEE Transactions on Vehicular Technology, vol. 60, issue 9, pp. 4299 – 4313, November, 2011.

[4] T. Quazi, H. Xu, F. Takawira, “Quality of Service for Multimedia Traffic using Cross-Layer Design,” IET Communications, vol. 3, issue 1, pp. 83 – 90, January, 2009.

[5] J. Boutros, E. Viterbo, "Signal Space Diversity: A Power- and Bandwidth-Efficient Diversity Technique for the Rayleigh Fading Channel," IEEE Transactions on Information Theory, vol. 44, no. 4, pp. 1453 – 1467, July, 1998.

[6] P. Ma, K.S. Kwak, “Modulation-assisted UEP-LDPC codes in image transmission,” in Proc. IEEE ISCIT, 2009, pp. 230 – 233.

[7] M. Ghandi, B. Barmada, E. Jones, M.Ghanbari, “Unequally error protected data partitioned video with combined hierarchical modulation and channel coding,” in Proc. IEEE International Conference Acoustics, Speech Signal Process., 2006, vol. 2, pp. II-529 – II-532.

[8] C. Hellge, T. Schierl, T. Wiegand, “Mobile TV with SVC and hierarchical modulation for DVB-H broadcast services,” in Proc. of the IEEE International Symposium on Broadband Multimedia Systems and Broadcasting, 2009, pp.1 – 5.

[9] A. Saeed, T. Quazi, H. Xu, “Hierarchical Modulated QAM with Signal Space Diversity and MRC Reception in Nakagami-m fading channels,” IET Communications, vol. 7, issue 12, pp. 1296 – 1303 , August, 2013.

[10] Z. Hu and H. Liu, “A Low-Complexity LDPC Decoding Algorithm for Hierarchical Broadcasting: Design and Implementation,” IEEE Transactions on Vehicular Technology, vol. 62, no. 4, pp. 1843 – 1849, May, 2013.

[11] Z. Hu and H. Liu, “Structure-based Decoding for Hierarchically Modulated, LDPC coded Signals,” in Proc. Global Communications Conference 2012, pp.4036 – 4042.

[12] A. Bouabdallah, D. Pradas, J. Lacan, M. Vazquez Castro, M. Bousquet, “Cross-layer optimization of unequal protected layered video over hierarchical


modulation,” in Proc. IEEE Global Telecomunications Conference, 2009, pp.1 – 6.

[13] L. Vangelista, N. Benvenuto, S. Tomasin, C. Nokes, J. Stott, A. Filippi, M. Vlot, V. Mignone, A. Morello, “Key Technologies for Next-Generation Terrestrial Digital Television Standrad DVB-T2,” IEEE Communications Magazine, vol. 47, no. 10, pp. 146 – 153, October, 2009.

[14] S. Wang, S. Kwon, and B. K. Yi, “On enhancing hierarchical modulation,” in Proc. of the IEEE International Symposium on Broadband Multimedia Systems and Broadcasting, 2008, pp.1 – 6.

[15] H. Zhao,X. Zhou,Y. Yang, W. Wang, “Hierarchical Modulation with Vector Rotation for E-MBMS Transmission in LTE Systems, ” Journal of Electrical & Computer Engineering, Hindawi Publishing Corporation, vol. 2010, pp. 1 – 9, August, 2010.

[16] P.K.Vitthaladevuni, M.S. Alouini, “BER Computation of 4/M-QAM Hierarchial Constellations,” IEEE Transactions on Broadcasting, vol. 47, issue 3, pp. 228 – 239, September, 2001.

[17] G. Abreu, “Accurate Simulation of Piecewise Continuous Arbitrary Nakagami-m Phasor Processes,” in Proc. IEEE Global Telecommunications Conference, 2006, pp. 1 – 6.


LANDMINE DETECTION BY MEANS OF GROUND PENETRAT-ING RADAR: A RULE-BASED APPROACH

P.A. van Vuuren ∗

∗ School of Electrical, Electronic and Computer Engineering, North-West University, Private BagX6001, Potchefstroom, 2520, South-Africa. E-mail: [email protected]

Abstract: For a humanitarian landmine detection system to be of practical value it has tomeet the following criteria: detection must be performed in real-time from a moving platform;plastic mines must be reliably detected; and the false alarm rate should be kept to a minimum.Ground penetrating radar is a promising technology for the detection of landmines withlow metal content. The improved algorithms presented in this paper can meet the abovementioned criteria to a large extent. Clutter removal is performed by an adaptive filterwhich removes virtually all background clutter in real time and can adapt to changing soilcharacteristics. Classification is performed by a new rule-based classifier capable of detectingboth metal and plastic anti-tank mines. Despite the fact that the algorithms are currentlyimplemented in Matlab�, mine detection can be performed on a vehicle moving at 9 km/h.

Key words: Landmine detection, ground penetrating radar, clutter removal

1. INTRODUCTION

Each year undetected landmines destroy manyinnocent lives. Although most nations agree thatlandmine removal is an important humanitarianproblem, the reliable and efficient detection ofburied landmines remains a challenging problem.Ground penetrating radar is a sensing technologythat has delivered promising results in the detectionof landmines with a low metallic content.

Traditionally, pattern recognition systems consistof three successive steps namely: preprocessing,feature extraction and final identification. Someresearchers dispense with the first two steps andattempt to solve the pattern recognition problemin a single identification stage operating on theraw measured data. Inevitably, the lack of aconcise set of features exposes the pattern recognizer(e.g. a neural network) with a sparsely populated,high-dimensional input space resulting in mediocreperformance at best. Without a preprocessing stepthe data is often too noisy and/or dominated byunimportant phenomena to allow the extraction ofusable features with high discriminatory abilities.Our practical experience tells us that a chain is onlyas strong as its weakest link. This advice also ringstrue in pattern recognition problems. Consequentlyall of the stages in a pattern recognition system haveto be performed as optimally as possible.

In landmine detection via GPR the preprocessingstep is primarily occupied with the removal ofas much clutter as possible to facillitate successfullandmine detection. To a large extent the GPRdemining literature hasn’t shown any significant newdevelopments on the topic of clutter removal duringthe past few years. Although the clutter removal

algorithms used at present are all quite good, theyare far from perfect.

Section 2 of this paper presents an improved adaptiveclutter removal algorithm which can be applied inreal time. Clutter removal is typically followed byfeature extraction and classification. In this paper,both of these steps are condensed into one. Theclassification approach presented in this paper ispresented in section 3. The performance of boththe clutter removal algorithm and the mine detectionalgorithm are separately evaluated in section 4, whilefinal recommendations for future work are given insection 5.

This paper is the second paper in a two-part serieson novel techniques that have been developed forthe detection of landmines by means of groundpenetrating radar (GPR). More details on the specificGPR array that has been used in this researchcan be found in the companion paper (Landminedetection by means of ground penetrating radar: amodel-based approach). In the previous paper clutterremoval was performed on the original frequencydomain data delivered by the stepped-frequencycontinuous-wave GPR antenna array. The clutterremoval procedure described in the previous paperhowever isn’t amenable to an elegant real-timeimplementation. In the current paper the focusis on accurate landmine detection in real-time. Atime-domain approach is followed throughout thispaper.

2. CLUTTER REMOVAL

To be of practical worth, clutter removal procedureshave to exhibit the following properties.


1. Real-time processing. In practice, mine clearanceoperations entail that large areas of land beprocessed. Time is therefore of the essence,which implies that landmine detection mustbe performed from a moving platform. Itis therefore imperative that clutter removal beperformed in real-time.

2. Adaptive behaviour. It is important that anyclutter removal technique can adapt to changingsoil conditions [1]. This calls for an adaptivemodel of the background clutter; one thatcan easily adapt to gradual variations in soilproperties, but is too slow to track the abruptchanges caused by mine reflections.

3. Reduced correlation between A-scans of emptysoil (i.e. without any mines), but improvedcorrelation between A-scans containing mineresponses.

As mentioned in the previous paper, the basicidea behind a number of existing clutter removalprocedures is to construct an adaptive model ofthe background present in each A-scan and then tosubtract this model from the observed data. Thisprocess then emphasizes all rapid variations fromthe background clutter. Fundamentally, this processamounts to adaptive high-pass filtering.

Probably one of the simplest, yet effective solutionsto this problem is to model the background clutterin the current A-scan as the weighted average ofthe previous N A-scans produced by the sameantenna-element (i.e. the previous N A-scans in thedown-track direction). Mathematically, this can beexpressed as follows:

xij =i−1

∑k=i−N

(αkxkj

), (1)

where xij is the estimated A-scan at position (i, j),αk is the weight reflecting the contribution of the kth

previous A-scan and xkj is the measured A-scan atposition (k, j). All A-scans are represented by Nz × 1vectors, where Nz refers to the number of samples inthe vertical (depth) direction.

In terms of vector notation, the above model can bewritten as follows:

xij = XΘ, (2)

where X is a Nz × N matrix of the previous A-scansand Θ is the N × 1 vector of coefficients αk.

If the coefficients of this weighted average arere-calculated for every new cross-track B-scan, themodel can adapt to changing conditions. The numberof previous A-scans (N) determines the ability of

the model to adapt to changing conditions: thelarger N becomes, the more inertia the model showsagainst change. Optimal weights can be obtained byminimizing the difference between the backgroundmodel and the current A-scan.

A least-squares solution for the coefficient vectorcan be found by means of the Moore-Penrosepseudo-inverse [2] as follows:

Θ = X+xij, (3)

where X+ = (XTX)−1XT .

Finally, the decluttered version of any A-scan(xij

)can then be obtained by subtracting the model ofthe background clutter

(xij

)in the A-scan from the

A-scan(xij

)as follows:

xij = xij − xij (4)

The above mentioned clutter removal procedure isundeniably very simple. It is however very effectiveas the results presented in section 4.1 show. The nextstep in mine detection entails feature extraction andclassification. In this paper both of these steps arecombined in a single rule-based detection algorithm.

3. RULE-BASED MINE DETECTION

Two of the tell-tale signs of the presence of a mine inGPR data are the so-called hyperbolas and concentriccircles. As figure 1 shows, these patterns canbe visible even in raw GPR data (in the case ofmetal anti-tank mines). It is no small wonder thathyperbolas and circular patterns are used by humanoperators to detect mines.

Hyperbolas are in fact produced in GPR data byany distinct subsurface object due to the imperfectdirectivity of the radar antennae [3]. Morespecifically, radar signals are transmitted in relativelywide angle beams which allow the detection ofobjects some distance from the central axis of thebeam. As a GPR antenna array moves along thesurface, the accumulated reflections of these objectsresult in hyperbolic reflection patterns visible invertical B-scans [4]. Concentric circles are visiblein horizontal B-scans. These circular patterns arecaused by the same mechanism as hyperbolas, butare limited to circular objects. (Buried pipes can beseen by hyperbolic patterns in vertical B-scans, butform linear patterns in horizontal B-scans.)

Other researchers have capitalized on the distinctiveshapes of hyperbolic and circular reflections todesign mine detection systems. One innovativeapproach was to perform clutter removal bymeans of a digital two-dimensional high-pass filterwhose pass-band was determined by the frequency


(a) Concentric circles in a horizontal B-scan (b) Hyperbolas in a vertical B-scan

Figure 1: Circles and hyperbolas present in raw GPR data

response of hyperbolic patterns [5]. Their approachdid improve the signal to clutter ratio of the datawithout significantly reducing the quality of minereflections. However, as mentioned previously,hyperbolic patterns aren’t unique to mines - in fact,pebbles also frequently cause such patterns.

Similarly, the idea to use only circular patterns inGPR data as an indicator of the presence of minesisn’t a new one. Some authors followed an imageprocessing approach in which each horizontal B-scanis regarded as an image [6] and [7]. Features are thenextracted from each image to measure the properties(such as the eccentricity and compactness) of anycircular objects in the image. The resultant featuresare then used by either rule-based classifiers [6] orneural networks [7] to come to a final decision onthe identity of the buried object. Classifiers basedon features that measure some aspect of the circularpatterns have obtained some success on practicalGPR data [7]. Unfortunately the computational costincurred by the required feature extraction process isquite high, since the circular nature of the patternsare inferred indirectly from e.g. eigenvalue ratiosobtained from moment matrices extracted from theB-scan image.

An alternative approach is to use the (circular)Hough transform to directly detect circular patternsin horizontal B-scans [8]. A major drawback of theHough transform however is that it isn’t accuratefor small circles. (Even though the GPR antennaarray used in this paper has an effective scan widthof 1.575 m, the resultant image only has 21 pixelsin the cross-track direction, which corresponds toa very small image.). Furthermore, the Houghtransform is a computationally intensive procedurewhich precludes any real-time applications.

It is clear that a mine detection approach based oncircular patterns shows promise if the computationalcost can be decreased without a commensurate lossin detection accuracy.

3.1 Algorithm overview

Observation of horizontal B-scans show that mineresponses exhibit the following characteristics:

• Circles consisting of a mixture of light and darkpatterns.

• The radius of the outer ring increases as theobservation depth is increased. The intensity ofthe outer ring however also quickly diminishes.

• The size of the rings correspond to the reflectiveproperties of the mine. Metal mines are morereflective and are therefore also more visible.

• Sequences of concentric circles sometimes be-come visible. These are due to resonantscattering of the radar waves induced by the topand bottom edges of the mine [9].

The success of a rule-based mine detection algorithmdepends heavily on the quality of features that areextracted or images that are used within the ruleantecedents. If the presence of circular patterns in ahorizontal B-scan are to be used to detect the presenceof mines, it is therefore important to enhance theimage sufficiently that the eventual classificationbecomes a mere formality. The entire classificationalgorithm will now be reviewed. Details of some ofthe intermediate steps in this algorithm are discussedin sections 3.2 to 3.5.

1. Extract a segment from the GPR C-scan. In thepresent version of the algorithm a “segment”is defined as a rectangular cylinder whose twohorizontal dimensions are equal to the numberof antennas in the cross-track direction. Theheight (depth) of the cylinder is determined bythe desired maximum depth of the GPR scan.For the GPR data at our disposal, the segmentencompasses a 21 × 21 × 160 three-dimensional


matrix of data. The motivation for the largesize of this segment is twofold. First is therelative rare occurence of landmines within anyparticular C-scan. Second is the relatively largesize of the circular patterns within a horizontalB-scan. (In practical data the concentric ringsassociated with a metal anti-tank (AT) mine havea radius of anything between three and eightpixels, where each “pixel” refers to the reflectionamplitude detected by a specific antenna withinthe entire GPR antenna array. In physicalterms, these circular patterns therefore haveradii stretching between 22.5 cm and 60 cm, fora cross-track channel spacing of 7.5 cm [10].)

2. For each depth index within the extracted seg-ment, the following steps are then performed.

(a) Emphasize both the darker and lighterpixels in each horizontal B-scan image.What is interpreted by the human visualsystem as a circular ring often actuallyconsists of semi-circular arcs of quitedifferent grayscale values, as can be seenin figure 2. This figure shows the responseof a low metal content anti-tank (AT) mineafter clutter removal has been performed.The image is shown purposefully small sothat the circular patterns are more clearlyvisible. To facillitate automatic detectionof circular patterns both lighter and darkergrayscale values in the image have to beemphasized above the general background.The result of this nonlinear amplitude filteris shown in figure 4. A detailed discussionon this filter can be found in section 3.2.

(b) Convert the filtered grayscale image intoa binary image consisting of backgroundpixels (black) and pixels in the vicinity ofa semi-circular arc (white). An example ofthe thresholding process is given in figure5. More details on this step can be found insection 3.3.

(c) Fit a circle to the pixels identified by thethresholding process. The fitted circlecan then be regarded as a model for theobserved circular pattern.

(d) Determine the validity of the aforemen-tioned model (namely the fitted circle). Thiscan be accomplished by calculating theaccuracy of the fitted model with respectto the observed data. For more detail onthe procedure to fit a circle to the data, aswell as to validate the fitted circle, consultsection 3.4.

(e) If the fitted circle passes the test, itsparameters (namely its centre position andradius) are used in a rule-based classifierto determine whether a mine is actuallypresent or not.

3. Increment the down-track dimension index bya preset interval (e.g. half of the width ofthe extracted segment) and repeat the entireprocedure from step 1.

Figure 2: Circular patterns caused by a plastic anti-tankmine

In the following paragraphs some detail aspects ofthe algorithm outlined in section 3.1 are presented.

3.2 Contrast enhancement

In image processing the subjective quality of animage can be improved by means of contrastenhancement [11]. Various techniques exist toperform contrast enhancement ranging from heuris-tic piecewise linear mappings of pixel grayscalevalues to histogram equalization. The disadvantageof piecewise linear mappings is that the resultantmapping from input to output pixel intensities isnon-smooth, which might result in unexpected edgesbeing artificially introduced in the image. Histogramequalization on the other hand strives to obtaina more uniform distribution of grayscale valuesthroughout the image. This powerful and eleganttechnique does however tend to amplify noise in theimage.

A useful alternative to piecewise linear contrastenhancement is the logistic function. Also known asthe sigmoid function in the neural network literature,the logistic function represents a smooth nonlinearmapping between an arbitrary input space and theunit interval. As can be seen in figure 3 the logisticfunction can be easily modified to accomplish anydesired mapping between the input and outputspace. Mathematically, this function can be expressedas follows:

y =1

1 + e−a(x−b), (5)

where x represents the original grayscale value ofa pixel, y is the transformed grayscale value and aand b are two parameters that control the particularnature of the mapping. The crossover point of thelogistic function (defined as the input value for whichthe function value is equal to 0.5.) is determined byb, while a controls the gradient of the function at itscrossover point.

As is evident in figure 2 circular patterns oftenconsist of a combination of light and dark pixelsarranged in semi-circular arcs. The objective of


Figure 3: The logistic function

image improvement in this context therefore entailsenhancing the contrast of both light and dark pixelsin comparison with the general background. Thiscan be accomplished by the sum of two separatelogistic functions, one responsible for enhancing darkpixels (y1) and another responsible for enhancinglight pixels (y2). The resultant amplitude filter issummarised in (6).

y = y1 + y2

=1

1 + eb(x−a)+

11 + e−c(x−d)

(6)

It is clear from (6) that the performance of theresultant filter is determined by the values of itsfour parameters: a, b, c and d. Optimal values forthese parameters were obtained through a process oftrial and error. In this context, “optimal” parametersare defined as those parameter values that result inthe isolation of only those pixels that form part ofa circular arc. The final filter function (obtainedthrough a process of visual evaluation of the resultsobtained by various sets of filter coefficients) is givenby

y =1

1 + e50(x−0.5)+

11 + e−50(x−0.6)

. (7)

Example results of the contrast enhancement processare given in figure 4. The input image for thenonlinear amplitude filter is a decluttered horizontalB-scan. Figure 4(a) shows a B-scan of a metal ATmine. The dark and light arcs that comprise thecircular patterns are clearly visible in this image.After the nonlinear amplitude filter of (7) is appliedto the image in figure 4(a) only a subset those pixelsthat form part of the circular pattern are isolated, ascan be seen in figure 4(b). It is possible to include alarger subset of the pixels on the circular patterns bymerely adjusting the parameters in (6). Such a largersubset does however come at the cost of more noise

in the filtered image. Increased noise levels in turnleads to difficulty in fitting an accurate circle to theisolated pixels.

3.3 Thresholding

At this stage of the process the original B-scan imageis reduced to a predominantly dark image containinga small set of lighter pixels that are located in thevicinity of a circular pattern (see figure 4(b)). Theisolated pixels however still have differing grayscalevalues, as is also evident in figure 4(b). In order to beable to fit a circle to the observed circular pattern, itis necessary to distinguish unambiguously betweenthe pixels that are definitely on the circular patternand those who aren’t. This choice can be made byconverting the grayscale image to a black and whiteimage by means of a thresholding function.

More specifically, if a pixel’s grayscale value is lessthan 0.5 it is converted to zero (black). If however apixel’s grayscale value is 0.5 or greater, it is convertedto one (white). An example of the result of thethresholding process is presented in figure 5.

The choice of the threshold grayscale value isdetermined by the distribution of grayscale values inthe enhanced image. As an example, figure 6 showsa histogram of the grayscale values in the enhancedimage in figure 4(b). Clearly, the vast majority of thepixels in the image are in fact part of the background,with only a very small number with grayscale valuesexceeding 0.5. If the threshold is chosen too low,spurious pixels are included in the dataset that isto be used for the circle fitting algorithm. If thethreshold is chosen too high, some of the pixels whichshould have been on the circle are omitted. Eithererror will result in inaccurate circle fitting. Typically,the threshold should be chosen to be somewhere atthe bottom of the valley in the histogram in figure 6.A threshold value of 0.6 was used to obtain the resultsin presented in this paper.

3.4 Circle fitting

Algorithms that fit circles to observed data pointcan be divided in two main categories, namely:geometric and algebraic algorithms [12]. Geometricalgorithms minize the geometric distance betweenthe model (fitted circle) and the data and are oftenimplemented by means of orthogonal least squares.The geometric distance is in essence the differencebetween the radius of the fitted circle and the“true” radius of the circle which gave rise to theobserved data. Algebraic algorithms on the otherhand minimize the algebraic distance (which is thedifference between the squared radius of the fittedcircle and the squared radius of the true underlyingcircle) between the observed data and the fitted circle.

Geometric algorithms are often regarded as being


(a) Decluttered horizontal B-scan of a metal AT mine (b) Result of nonlinear amplitude filtering

Figure 4: Contrast enhancement by means on logistic filters

(a) Image after contrast enhancement (b) Binary image after thresholding

Figure 5: Isolation of circle pixels by means of thresholding

more accurate than algebraic algorithms, but arealso iterative and therefore more computationallyintensive. The relative inaccuracy of algebraicalgorithms should however be seen in the contextthat the differences in the accuracy between algebraicand geometric algorithms can only be revealed byhigher order error analysis [12]. The slight lossin accuracy incurred by an algebraic circle fittingalgorithm is therefore handsomely compensated forby its speed.

Obtaining a least-squares fit for a circle is a nonlinearproblem. In algebraic algorithms parameter substi-tutions are used to convert this nonlinear probleminto a linear problem. Probably the fastest algebraiccircle fitting algorithm is the Kasa algorithm [13].Compared to other algebraic algorithms it does havea bias towards small circles (i.e. the radius of thefitted circles are consistently slightly too small). Thisshortcoming is however irrelevant for the application

in mind, as we’ll see when the classification rule isdiscussed in section 3.5.

Figure 7 shows the circle fitted on the data obtainedfrom the binary image in figure 5(b). The circle infigure 7 is superimposed on the enhanced image offigure 5(a) to show that the fitted circle is indeed anaccurate model for the observed circular pattern.

The acceptance of any mathematical model ishowever subject to its validation. This principle alsoholds true in this landmine detection algorithm, sincea lot depends on the accuracy of the fitted circle. Thenext step of the mine detection algorithm is thereforeto ensure the quality of the fitted circle.

The quality of the fitted circle can be assessed by thefollowing three self-evident rules:

(a) The centre of the fitted circle must be within the


Figure 6: Histogram of grayscale values in the enhancedimage

Figure 7: Circle fitted to a circular pattern

boundaries of the image.

(b) The diameter of the fitted circle must fit insidethe boundaries of the image.

(c) And most importantly, the fitted circle must bea good description of the underlying pattern inthe data.

The last consideration is in essence the accuracy ofthe fitted model and can be conveniently measuredwith the so-called coefficient of determination [14].This measure is often used in system identificationand can be interpreted as the “. . . fraction of theraw variation in y accounted for using the fittedequation.” [14]. Mathematically, the coefficient ofdetermination can be expressed as follows:

R2 =∑i (yi − y)2 − ∑i (yi − yi)

2

∑i (yi − y)2 , (8)

where yi refers to the ith measured datapoint (the rownumber of a pixel), y represents the average valueof the measured datapoints and yi signifies the valueof the fitted circle corresponding to the ith measured

datapoint.

If the R2-value of a fitted circle is less than aprescribed value, the model is rejected. In thatcase it can be concluded that there is no circularpattern present in the data, which in turn leads tothe conclusion that no mine is present at least at thedepth of the current B-scan. The threshold againstwhich the R2-value of the fitted circle is compareddetermines the stringency of the detection algorithm.Low values for this threshold result in numerousfalse alarms. If the threshold value for the fittedcircle’s quality is however too high, some minesmight escape detection. In the present version of thealgorithm only circles with an accuracy exceeding 95percent were retained.

3.5 Classification rule

As mentioned in section 3.1, the reflections causedby landmines have a number of characteristicproperties. In short, as the depth index increases anumber of concentric rings fan out from a commoncentrepoint. This observation forms the basis of a rulewith which landmine detection can be performed. Itis important to note that the precise size of the fittedcircles is not as important as the reliable detectionand placement of the circles in the first place. Sincethe rings are arranged in concentric patterns and theKasa algorithm is biased towards smaller circles itinvariably works out in practice that the radius ofthe fitted circle remains roughly constant as the depthincreases.

The following rule can therefore be used to detectmines from a sequence of B-scans:

IF two or more circles occur in consecutive B-scans ANDIF their centrepoints remain approximately constant ANDIF their radii also remain approximately constant THEN alandmine occurs at the horizontal position and depth of thefitted circles.

From the above mentioned rule it is evident that oneadvantage of this approach to mine detection is thatit is quite easy to pinpoint the position of the mine inthree dimensions.

4. RESULTS

The algorithms discussed in sections two and threewere implemented on the same dataset that was usedin the companion paper. Similar measures are alsoused to assess the results as in the previous paper.

4.1 Clutter removal results

The clutter removal algorithm presented in this paperonly has a single adjustable parameter, namely thenumber (N) of previous A-scans that are included inthe weighted average. It was found experimentally


that the best balance between the quality of clutterremoval and computational cost is struck for N = 2.

Figure 8 shows a horizontal B-scan taken at a specificdepth both before and after clutter removal. Smalldifferences between the different antenna-elementscause interference patterns in the raw data. The resultare the linear patterns present in figure 8(a) (wherethey resemble car tracks). From figure 8(b) it is clearthat all antenna-related noise has been removed bythe clutter removal procedure of (4).

The quality of clutter removal algorithms can also bemeasured more objectively by the average level ofcross-correlation present in the data after it has beenprocessed. Figure 9 shows the average degree of peakcross-correlation for every 3 × 3 neighbourhood ofA-scans in the data. Figure 9(a) shows the averagedegree of correlation of neighbouring A-scans afterapplication of a time-domain principal componentbased clutter removal algorithm [15]. This figureshould be interpreted as a top-down view of anarea of ground which has been scanned with aground-penetrating radar. Areas showing high de-grees of correlation may be indicative of interestingsubsurface objects or soil layers. Superimposedon this figure are the positions of known mines(indicated with black squares). Clearly, there is roomfor improvement in the clutter removal algorithmsince large areas of ground are shown to have highlycorrelated A-scans.

The average degree of cross-correlation was alsocalculated for the new clutter removal algorithm(figure 9(b)). Clearly, the new clutter removalalgorithm removes almost all of the backgroundreflections from the data with the result that theA-scans are almost entirely decorrelated (with theexception of the immediate vicinity of the two metalAT mines).

Probably the most attractive feature of the improvedclutter removal algorithm is its speed. Thenew algorithm requires 3.7 ms to process anentire cross-track B-scan, which makes real-timeimplementation possible. If the reader bears in mindthat all of the results presented in this paper wereobtained by means of Matlab�, one could easilyexpect another twofold improvement if the algorithmis implemented in C.

4.2 Adaptability of the background model

The new clutter removal algorithm can adapt tochanging soil conditions. This property is due tothe fact that the background clutter present in eachA-scan is modelled by the weighted average ofthe previous A-scans in the down-track direction.One could however wonder whether the model’scoefficients are in fact adapted in response to thechanging soil conditions or whether the algorithm

merely gravitates to a fixed filter after a few rows ofA-scans.

In section 4.1 it is reported that only two priorA-scans are necessary to obtain a useful weightedaverage. The background clutter in any A-scancan therefore be modelled as in (9). The modelin (9) is in effect a moving average low-pass filterwhose filter coefficients are adaptively calculated.The bandwidth of a moving average low-pass filteris inversely related to the window width. Ashort window (as in the filter in (9)) thereforecorresponds to a low-pass filter with a largebandwidth. Consequently, the decluttered A-scanstherefore only contain very high frequencies.

xij = θ1x(i−1)j + θ2x(i−2)j. (9)

The distributions of the coefficients θ1 and θ2 in (9) aregiven in the histograms presented in figure 10. Thesehistograms were obtained from the filter coefficientscalculated for each and every A-scan in the entireavailable C-scan of data. Clearly, the values for bothcoefficients are distributed continuously over quitesmall ranges. Furthermore, the coefficients seem tobe normally distributed.

The distribution of the filter coefficients calculatedfor each A-scan are given as a function of thephysical location of the corresponding A-scan infigure 11(a) and (b). Although it seems that thefilter coefficients were chosen at random, a carefulstudy of both figures reveals that small regions ofsimilar coefficient values are present in the images.This observation supports the conclusion that themodel in (9) does indeed adapt to changing local soilconditions.

4.3 Landmine detection results

Although the rule-based classifier seems quite sim-plistic compared to other classifiers, its performanceis nothing to be scoffed at. As figure 12 shows, therule-based classifier can correctly identify metal ATmines as well as a plastic AT mine. In this figure,the positions of known mines are indicated by cyansquares, while red stars indicate the positions wherethe algorithm thought there were mines.

Although this classifier can detect metal andnon-metal AT mines, its false alarm rate is inneed of attention, as can be clearly seen in figure12. Furthermore, the rule-based algorithm couldn’tdetect plastic anti-personnel (AP) mines. An analysisof the GPR data at the positions of the false alarmsshow that there are indeed circular patterns presentat the locations of the false alarms. As figure12 shows, false alarms frequently occur near theknown locations of meerkat tunnels as well as onthe boundary between the sandpit and normal soil.


(a) Before clutter removal

(b) After clutter removal

Figure 8: Subjective evaluation of the quality of clutter removal

(a) Time domain principal component based filtering

b) New clutter removal algorithm

Figure 9: Comparison of the cross-correlation present in the data after clutter removal

Indeed, there are a number of subsoil objects withGPR reflections quite similar to those of mines.

Another (more serious) problem is that therule-based algorithm couldn’t detect any AP mine.The available GPR data is partly to blame for thisstate of affairs. In this specific dataset the spacingbetween consecutive cross-track B-scans is 10 cm. Ifthe small dimensions of e.g. an M-14 mine are takeninto consideration, it is no small wonder that thepresence such mines are well nigh invisible in theGPR data

One advantage of the rule-based classifier is its

speed. On average it requires 417 µs to processan entire cross-track B-scan. If this is added to the3.7 ms per cross-track B-scan required by the clutterremoval algorithm, the entire duration is 4.1 ms.This is almost neglegible compared to the 35.7 msrequired to obtain the GPR cross-track B-scan data inthe first place. If successive cross-track B-scans areseparated by a distance of 10 cm (as is the case withthe available data), the landmine detection (includingradar measurement) can be performed at a maximumspeed of 2.51 m/s or 9.04 km/h.


Figure 10: Histograms of the values of θ1 and θ2 in (9)

(a) Values of θ1

b) Values of θ2

Figure 11: Filter coefficient values as a function of physical location

Figure 12: Detection results

5. CONCLUSION

Compared to the existing GPR literature, theperformance of the rule-based classifier of section 3 isquite promising. In fact, one can state with a certaindegree of confidence that real time GPR detection ofplastic AT mines is feasible. The improved clutterremoval procedure described in section 2 not only

removes almost all of the background clutter, but alsoin a time frame that allows real-time implementation.However, the following two concerns remain:

1. The false alarm ratio which is still too high.Improving the circle fitting algorithm willprobably address this issue satisfactorily.


2. The inability of the classifier to detect plasticAP mines. This deficiency is however tobe expected, since the particular mines arepractically invisible even in the decluttered data.The resolution of the original raw GPR data canbe improved by sampling the ground at finerintervals in the down-track direction. However,in certain soil conditions the difference in thepermittivity of an AP mine and the surroundingsoil will still be insufficient to make the minevisible in GPR data.

REFERENCES

[1] A. Van der Merwe and I. Gupta, “A novelsignal processing technique for clutter reductionin GPR measurements of small, shallow landmines,” IEEE transactions on geoscience and remotesensing, vol. 38, no. 6, pp. 2627–2637, November2000.

[2] A. Ben-Israel and T. Greville, Generalizedinverses: theory and applications. New-York: JohnWiley & Sons, 1974.

[3] S. Perrin, E. Duflos, P. Vanheeghe, andA. Bibaut, “Multisensor fusion in the frame ofevidence theory for landmines detection,” IEEETRANSACTIONS ON SYSTEMS, MAN, ANDCYBERNETICSPART C: APPLICATIONS ANDREVIEWS, vol. 34, no. 4, pp. 485–498, November2004.

[4] O. Lopera, E. C. Slob, N. Milisavljevic, andS. Lambot, “Filtering soil surface and antennaeffects from GPR data to enhance landminedetection,” IEEE Transactions on Geoscienceand Remote Sensing, vol. 45, no. 3, pp.707 – 717, March 2007. [Online]. Available:http://dx.doi.org/10.1109/TGRS.2006.888136

[5] D. Potin, E. Duflos, and P. Vanheeghe,“Landmines ground-penetrating radar signalenhancement by digital filtering,” IEEETransactions on Geoscience and RemoteSensing, vol. 44, no. 9, pp. 2393– 2406, 2006. [Online]. Available:http://dx.doi.org/10.1109/TGRS.2006.875356

[6] P. Gader, W.-H. Lee, and J. N. Wilson,“Detecting landmines with ground-penetratingradar using feature-based rules, order statistics,and adaptive whitening,” IEEE Transactions onGeoscience and Remote Sensing, vol. 42, no. 11,pp. 2522 – 2534, 2004. [Online]. Available:http://dx.doi.org/10.1109/TGRS.2004.837333

[7] J. N. Wilson, P. Gader, W.-H. Lee, H. Frigui,and K. Ho, “A large-scale systematic evaluationof algorithms using ground-penetrating radarfor landmine detection and discrimination,”IEEE Transactions on Geoscience and Remote

Sensing, vol. 45, no. 8, pp. 2560 –2572, August 2007. [Online]. Available:http://dx.doi.org/10.1109/TGRS.2007.900993

[8] D. Daniels, Ground penetrating radar, 2nd ed.London: The Institution of Engineering andTechnology, 2004.

[9] Daniels, “Ground penetrating radar fundamen-tals,” Appendix to a report to the U.S. EPAregion V, Tech. Rep., 2000.

[10] “V-series multi-channel antenna arrays for GPR:Mkt-ds-v-series-040810v1,” 2010.

[11] R. Gonzalez and R. Woods, Digital image process-ing. Reading, Massachusetts: Addison-Wesley,1992.

[12] A. Al-Sharadqah and N. Chernov, “Erroranalysis for circle fitting algorithms,” ElectronicJournal of Statistics, vol. 3, pp. 886–911, 2009.

[13] I. Kasa, “A curve fitting procedure and its erroranalysis,” IEEE transactions on instrumentationand measurement, vol. 25, pp. 8–14, 1976.

[14] S. Vardeman, Statistics for engineering problemsolving. Boston: PWS, 1994.

[15] S. Tjora and E. Eide, “Evaluation of methodsfor ground bounce removal in GPR utilitymapping,” in Tenth International Conference onGround Penetrating Radar, 2004.


NOTES


This journal publishes research, survey and expository contributions in the field of electrical, electronics, computer, information and communications engineering. Articles may be of a theoretical or applied nature, must be novel and

must not have been published elsewhere.

Nature of ArticlesTwo types of articles may be submitted:

• Papers: Presentation of significant research and development and/or novel applications in electrical, electronic, computer, information or communications engineering.

• Research and Development Notes: Brief technical contributions, technical comments on published papers or on electrical engineering topics.

All contributions are reviewed with the aid of appropriate reviewers. A slightly simplified review procedure is used in the case of Research and Development Notes, to minimize publication delays. No maximum length for a paper

is prescribed. However, authors should keep in mind that a significant factor in the review of the manuscript will be its length relative to its content and clarity of writing. Membership of the SAIEE is not required.

Process for initial submission of manuscriptPreferred submission is by e-mail in electronic MS Word and PDF formats. PDF format files should be ‘press

optimised’ and include all embedded fonts, diagrams etc. All diagrams to be in black and white (not colour). For printed submissions contact the Managing Editor. Submissions should be made to:

The Managing Editor, SAIEE Africa Research Journal, PO Box 751253, Gardenview 2047, South Africa.

E-mail: [email protected]

These submissions will be used in the review process. Receipt will be acknowledged by the Editor-in-Chief and subsequently by the assigned Specialist Editor, who will further handle the paper and all correspondence pertaining

to it. Once accepted for publication, you will be notified of acceptance and of any alterations necessary. You will then be requested to prepare and submit the final script. The initial paper should be structured as follows:

• TITLE in capitals, not underlined.• Author name(s): First name(s) or initials, surname (without academic title or preposition ‘by’)• Abstract, in single spacing, not exceeding 20 lines.• List of references (references to published literature should be cited in the text using Arabic numerals in

square brackets and arranged in numerical order in the List of References).• Author(s) affiliation and postal address(es), and email address(es).• Footnotes, if unavoidable, should be typed in single spacing.• Authors must refer to the website: http: //www.saiee.org.za/arj where detailed guidelines, including

templates, are provided.

Format of the final manuscriptThe final manuscript will be produced in a ‘direct to plate’ process. The assigned Specialist Editor will provide you

with instructions for preparation of the final manuscript and required format, to be submitted directly to: The Managing Editor, SAIEE Africa Research Journal, PO Box 751253, Gardenview 2047, South Africa.

E-mail: [email protected]

Page chargesA page charge of R200 per page will be charged to offset some of the expenses incurred in publishing the work.

Detailed instructions will be sent to you once your manuscript has been accepted for publication.

Additional copiesAn additional copy of the issue in which articles appear, will be provided free of charge to authors.

If the page charge is honoured the authors will also receive 10 free reprints without covers.

CopyrightUnless otherwise stated on the first page of a published paper, copyright in all contributions accepted for publication is vested in the SAIEE, from whom permission should be obtained for the publication of whole or part of such material.

SAIEE AFRICA RESEARCH JOURNAL – NOTES FOR AUTHORS


South African Institute for Electrical Engineers (SAIEE)PO Box 751253, Gardenview, 2047, South Africa

Tel: 27 11 487 3003 | Fax: 27 11 487 3002E-mail: [email protected] | Website: www.saiee.org.za

Vol.105 (4) December 2014 SOUTH AFRICAN … (4) December 2014 SOUTH AFRICAN INSTITUTE OF ELECTRICAL...

Documents

Transcript of Vol.105 (4) December 2014 SOUTH AFRICAN … (4) December 2014 SOUTH AFRICAN INSTITUTE OF ELECTRICAL...