Www.wust.edu.cn Distributed State-Estimation Using Quantized Measurement Data from Wireless Sensor...

28
www.wust.edu.cn Distributed State-Estimation Using Quantized Measurement Data from Wireless Sensor Networks Li Chai with Bocheng Hu Professor College of Information Science and Engineering Wuhan University of Science and Technology Wuhan, 430081, China Email: [email protected]

Transcript of Www.wust.edu.cn Distributed State-Estimation Using Quantized Measurement Data from Wireless Sensor...

Page 1: Www.wust.edu.cn Distributed State-Estimation Using Quantized Measurement Data from Wireless Sensor Networks Li Chai with Bocheng Hu Professor College of.

www.wust.edu.cn

Distributed State-Estimation Using Quantized Measurement Data from

Wireless Sensor Networks

Li Chai with Bocheng Hu

Professor College of Information Science and EngineeringWuhan University of Science and TechnologyWuhan, 430081, ChinaEmail: [email protected]

Page 2: Www.wust.edu.cn Distributed State-Estimation Using Quantized Measurement Data from Wireless Sensor Networks Li Chai with Bocheng Hu Professor College of.

www.info.wust.edu.cn

2

Outline

Introduction of WUST and College of ISE Motivation and related works Problem statements State estimator design Simulation Conclusion

Page 3: Www.wust.edu.cn Distributed State-Estimation Using Quantized Measurement Data from Wireless Sensor Networks Li Chai with Bocheng Hu Professor College of.

www.info.wust.edu.cn

3

Introduction of WUST and College of ISE

LocationWuhan, besides the Yangtze river and very near to Three Gorges Dam

20 colleges, about 1,500s academic staff Feature: tight link with metallurgical company

(Wuhan Iron & Steel Co., Ltd, Panzhihua Iron & Steel Co., Ltd, Handan Iron & Steel Co., Ltd, Baoshan Iron & Steel Co., Ltd)

Page 4: Www.wust.edu.cn Distributed State-Estimation Using Quantized Measurement Data from Wireless Sensor Networks Li Chai with Bocheng Hu Professor College of.

www.info.wust.edu.cn

4

Introduction of WUST and College of ISE

College of Information Science and Engineering 75 Academic staff including 16 professors, 15 AP and 8

professional engineer Two Departments:

Dept. of Automatic Control, Dept. of Electrical Engineering About 200 PG students and 1,200 UG

Feature: metallurgical automationEngineering Research Center for Metallurgical Automation and Measurement Technology, Ministry of Education, China

Page 5: Www.wust.edu.cn Distributed State-Estimation Using Quantized Measurement Data from Wireless Sensor Networks Li Chai with Bocheng Hu Professor College of.

www.info.wust.edu.cn

5

Page 6: Www.wust.edu.cn Distributed State-Estimation Using Quantized Measurement Data from Wireless Sensor Networks Li Chai with Bocheng Hu Professor College of.

www.info.wust.edu.cn

6

Page 7: Www.wust.edu.cn Distributed State-Estimation Using Quantized Measurement Data from Wireless Sensor Networks Li Chai with Bocheng Hu Professor College of.

www.info.wust.edu.cn

7

Page 8: Www.wust.edu.cn Distributed State-Estimation Using Quantized Measurement Data from Wireless Sensor Networks Li Chai with Bocheng Hu Professor College of.

www.info.wust.edu.cn

8

Motivation and related works

A typical sensor network consists of a large number of nodes deployed in an environment being sensed and/or controlled.

The sensors collaborate to perform certain high level task: detection, estimation …

The sensors’ dynamic range, resolution, power and wireless communication capability can be severely limited.

Local data quantization/compression is not only a necessity, but also an integral part of the design of sensor networks.

Page 9: Www.wust.edu.cn Distributed State-Estimation Using Quantized Measurement Data from Wireless Sensor Networks Li Chai with Bocheng Hu Professor College of.

www.info.wust.edu.cn

9

Motivation and related works

Sensor network applications– Environmental monitoring – Habitat monitoring – Acoustic detection – Seismic Detection – Military surveillance – Inventory tracking – Medical monitoring – Smart spaces – Process Monitoring

Page 10: Www.wust.edu.cn Distributed State-Estimation Using Quantized Measurement Data from Wireless Sensor Networks Li Chai with Bocheng Hu Professor College of.

www.info.wust.edu.cn

10

Motivation and related works

The highly decentralized network architecture and severely limited communication constraints presents significant challenges in the design of signal processing algorithms.

In this talk, we will focus on a general state estimation problem

Will not consider Details of communication protocol / network topology Channel fading and uncertainty Location and routing issues

Page 11: Www.wust.edu.cn Distributed State-Estimation Using Quantized Measurement Data from Wireless Sensor Networks Li Chai with Bocheng Hu Professor College of.

www.info.wust.edu.cn

11

Motivation and related works

Static decentralized estimation problemXiao and Luo (2005, 2006) and Riberiro and Giannakis (2006)

Page 12: Www.wust.edu.cn Distributed State-Estimation Using Quantized Measurement Data from Wireless Sensor Networks Li Chai with Bocheng Hu Professor College of.

www.info.wust.edu.cn

12

Motivation and related works

Static decentralized estimation problem

Methods to design local message functions and final fusion function

Methods of estimation if one-bit sensor is assumed.

Analysis of the MSE.

Tradeoff between network size K and MSE under bandwidth constraint.

)( kk xm).,,( 1 Kmmf

Page 13: Www.wust.edu.cn Distributed State-Estimation Using Quantized Measurement Data from Wireless Sensor Networks Li Chai with Bocheng Hu Professor College of.

www.info.wust.edu.cn

13

Problem statements

Dynamic decentralized estimation

1S 2S NS

Fusion Center

1( )y k 2 ( )y k ( )Ny k

1m 2m Nm

)()(

)()()1(

kCxky

kkAxkx

Page 14: Www.wust.edu.cn Distributed State-Estimation Using Quantized Measurement Data from Wireless Sensor Networks Li Chai with Bocheng Hu Professor College of.

www.info.wust.edu.cn

14

Problem statements

In the figure

Page 15: Www.wust.edu.cn Distributed State-Estimation Using Quantized Measurement Data from Wireless Sensor Networks Li Chai with Bocheng Hu Professor College of.

www.info.wust.edu.cn

15

Problem statements

To design the state estimator

such that is “close” to x(k).

Here, “close” means is small, where

0

1

)0(ˆ

))(,),(()(ˆ

)(ˆ)(ˆ)1(ˆ

xx

kmkmgky

kyBkxAkx

N

ff

)(ˆ kx

)()(ˆ)( kxkxke

pke )(

Page 16: Www.wust.edu.cn Distributed State-Estimation Using Quantized Measurement Data from Wireless Sensor Networks Li Chai with Bocheng Hu Professor College of.

www.info.wust.edu.cn

16

Problem statements

Power spectral density

where

Power norm of the error is defined as

)]()([)( * kekeERe

1

0

*2

)}({)}0({

)]()([)(

dffStraceRtrace

kekeEke

ee

p

fjee eRfS 2)()(

Page 17: Www.wust.edu.cn Distributed State-Estimation Using Quantized Measurement Data from Wireless Sensor Networks Li Chai with Bocheng Hu Professor College of.

www.info.wust.edu.cn

17

State estimator design

The augment system

)1(ˆ

)1()(

)(

)(

0

0

)(ˆ

)(0

)1(ˆ

)1(

kx

kxIIke

k

k

B

I

kx

kx

ACB

A

kx

kx

nn

f

n

ff

G

e

Page 18: Www.wust.edu.cn Distributed State-Estimation Using Quantized Measurement Data from Wireless Sensor Networks Li Chai with Bocheng Hu Professor College of.

www.info.wust.edu.cn

18

State estimator design

The power norm of error

An upper bound

The above bound is tight in the sense that it can be achieved if is arbitrary.

1

0

2*2

1

0

2*22

)()()(

)()()()(

dfeGfSeGtrace

dfeGfSeGtraceke

fje

fje

fjeww

fjewp

222

2

2)(

peewpGGke

Page 19: Www.wust.edu.cn Distributed State-Estimation Using Quantized Measurement Data from Wireless Sensor Networks Li Chai with Bocheng Hu Professor College of.

www.info.wust.edu.cn

19

State estimator design

To design the state estimator

such that is minimized.

0

1

)0(ˆ

))(,),(()(ˆ

)(ˆ)(ˆ)1(ˆ

xx

kmkmgky

kyBkxAkx

N

ff

222

2 peew GG

Page 20: Www.wust.edu.cn Distributed State-Estimation Using Quantized Measurement Data from Wireless Sensor Networks Li Chai with Bocheng Hu Professor College of.

www.info.wust.edu.cn

20

State estimator design

Step 1, find g, and upper bound of

Step 2, find such that is

minimized.

Remark: Step 2 is a typical mixed optimization filtering problem, for which various efficient algorithms exist.

ff BA ,

21

2))(),(( kmkmgyE Np

22

2 eew GG

Page 21: Www.wust.edu.cn Distributed State-Estimation Using Quantized Measurement Data from Wireless Sensor Networks Li Chai with Bocheng Hu Professor College of.

www.info.wust.edu.cn

21

Numerical example

Consider the following LTI system

Let

0 0.5( 1) ( ) ( )

1 1

( ) 100 10 ( ) ( ), 1,...,i i

x k x k k

y k x k v k i N

0.4233 0.4457 0.0044,

0.9394 0.9851 0.0003f fA B

TxN 32,100,2 0

Page 22: Www.wust.edu.cn Distributed State-Estimation Using Quantized Measurement Data from Wireless Sensor Networks Li Chai with Bocheng Hu Professor College of.

www.info.wust.edu.cn

22

Conclusion

Distributed state estimator is designed.

The power norm of the error is minimal in worst-case.

The idea applies to other cases, such as different types of sensors are used.

Page 23: Www.wust.edu.cn Distributed State-Estimation Using Quantized Measurement Data from Wireless Sensor Networks Li Chai with Bocheng Hu Professor College of.

www.info.wust.edu.cn

23

Basic multirate elements in digital signal processing

M-fold decimator

M yD[n]x[n] ][][ MnxnyD

-2

-1 0 1 2 3 n

0 1 n

x[n]

yD[n]-1

M=2

Multirate signal processing

Page 24: Www.wust.edu.cn Distributed State-Estimation Using Quantized Measurement Data from Wireless Sensor Networks Li Chai with Bocheng Hu Professor College of.

www.info.wust.edu.cn

24

L-fold expander

n40 1 2 3 865 7

x[n]

n0 1 2

yE[n]

3 4 5 6 7 8

Vaidyanathan 93

Multirate signal processing

Page 25: Www.wust.edu.cn Distributed State-Estimation Using Quantized Measurement Data from Wireless Sensor Networks Li Chai with Bocheng Hu Professor College of.

www.info.wust.edu.cn

25

Multirate Signal Processing in WSNs

(a) Direct high sampling rate measurement x(n)

(b) Low sampling rate measurements vi(n)

(c) Relation between x(n) and vi(n)

Page 26: Www.wust.edu.cn Distributed State-Estimation Using Quantized Measurement Data from Wireless Sensor Networks Li Chai with Bocheng Hu Professor College of.

www.info.wust.edu.cn

26

Multirate Signal Processing in WSNs

To estimate the power spectral density of x(n) using statistics

of the low-rate observable signals vi(n).

O. S. Jahromi, B. A. Francis, and R. H. Kwong, Relative information of multi-rate sensors,

Information Fusion, 5, pp. 119-129, 2004.

Page 27: Www.wust.edu.cn Distributed State-Estimation Using Quantized Measurement Data from Wireless Sensor Networks Li Chai with Bocheng Hu Professor College of.

www.info.wust.edu.cn

27

Multirate Signal Processing in WSNs

Our research: Is it possible to achieve other goals using low-rate sampling

data? If yes, how to design suitable algorithms and how to evaluate those algorithms?

How to deal with quantization and channel uncertainty? Does the dual-rate assumption make sense? For arbitrary

sampling-rate data, what shall we do? Key

Distributed (multirate) signal processing

Page 28: Www.wust.edu.cn Distributed State-Estimation Using Quantized Measurement Data from Wireless Sensor Networks Li Chai with Bocheng Hu Professor College of.

www.info.wust.edu.cn

28

Thank you!