Low Complexity Distributed Encoding Scheme for Low DensityLow-Complexity Distributed Encoding Scheme for Low-Density P it Ch k C d i Wi l S N t kParity-Check Codes in Wireless Sensor Networks

Shibaura Institute of T h l J Phat Nguyen Huu Vinh Tran Quang and Takumi MiyoshiTechnology, Japan Phat Nguyen Huu, Vinh Tran-Quang, and Takumi Miyoshi

Shibaura Institute of Technology JapanShibaura Institute of Technology, JapanHanoi University of Science and Technology VietnamHanoi University of Science and Technology, Vietnam

Aim SimulationThis paper proposes a distributed image encoding scheme that uses low-density parity-check (LDPC)codes over an additive white Gaussian noise (AWGN) channel in wireless sensor networks (WSNs). In

Simulation Setupcodes over an additive white Gaussian noise (AWGN) channel in wireless sensor networks (WSNs). Inthe scheme, the encoding task is divided into several small processing components, which are thendistributed to multiple nodes in a cluster while considering their residual energy We conduct extensive

• Network area: 400 m × 400 m• Initial energy of sensor: 1 Jouledistributed to multiple nodes in a cluster while considering their residual energy. We conduct extensive

computational simulations to verify our methods and find that the proposed scheme not only solvesthe energy balance problem by sharing the processing tasks but also increases the quality of data by

gy• Base station: The closest node to the center of the field• Input data: Lena image (64 pixels × 64 pixels) with 8 bps (bits per pixel)the energy balance problem by sharing the processing tasks but also increases the quality of data by

using LDPC codes.• Input data: Lena image (64 pixels × 64 pixels) with 8 bps (bits per pixel)

The following five schemes are compared:

Introduction • Huffman and RS coding scheme• Huffman and centralized LDPC coding scheme

• Wireless sensor networks (WSNs) consist of the number of nodes. Each node has two functions: sensor andRF transceiver. These nodes use an ad-hoc protocol to communicate with each other and transfer data to the

Huffman and centralized LDPC coding scheme• Huffman and distributed LDPC coding scheme

GZIP d di t ib t d LDPC di hRF transceiver. These nodes use an ad hoc protocol to communicate with each other and transfer data to thebase station using multi-hop technology.• Two methods to recover the lost data on WSNs are the automatic repeat request (ARQ) and forward error

• GZIP and distributed LDPC coding scheme• miniLZO and distributed LDPC coding scheme

• Two methods to recover the lost data on WSNs are the automatic repeat request (ARQ) and forward errorcorrection (FEC) methods.

Si th d d f i l lti di li ti h i d th d i f ffi i tThe parameters of the computation energy model

• Since the demands for wireless multimedia applications have increased, the design of efficientcommunication system for progressive image transmission has recently attracted attention in many literatures Parameter Value

p p gy

[1], [3].

• Challenges in WSNs are: limited resource (i e power battery bandwidth processing capacities and memory)Energy transceiver electron (εelec) 50 nJ / bit

• Challenges in WSNs are: limited resource (i.e., power battery, bandwidth, processing capacities and memory)and low fidelity.

Energy transmission in free space model (εfs) 10 pJ / bit / m2

Energy transmission in multi-path model (εmp) 0.013 pJ / bit / m4

FEC Coding on WSNsI th FEC ti l d t i d t f FEC di d d t th i k d

Energy transmission in multi path model (εmp) 0.013 pJ / bit / m

Energy for preprocessing (epre) 15 nJ / bitIn the FEC conventional way, source nodes capture image data, perform FEC coding and send to the sink node.Since the data size are often large, the source nodes will exhaust their energy in the short time. Therefore, we

Energy for transforming DCT (eDCT) 20 nJ / bit

Energy for source coding (e ) 90 nJ / bitneed to distribute FEC coding task before sending data to the sink to balance energy consumption for nodes inWSNs.

Energy for source coding (ecod) 90 nJ / bit

Energy for LDPC coding (ecod) 0.115 nJ / bit/ iter.

Related Work Threshold distance (d0) 100 m

• The authors in [2] implemented FEC codes and found that the frequency of bit error was almost zero when thedistance between a transceiver and a receiver was less than 10 m. They also show that Reed Solomon (RS)d sta ce bet ee a t a sce e a d a ece e as ess t a 0 ey a so s o t at eed So o o ( S)codes are difficult to implement on WSNs since the sensors have their limited resources (i.e., battery power,computational capacity, and memory).

Simulation Resultscomputational capacity, and memory).

• In [3], Sartipi et al. used low-density parity-check (LDPC) codes for FEC codes, and they showed that LDPC Comparison of Network QoScodes improved not only energy efficiency but also the data compression rate compared to Bose and Ray-Chaudhuri (BCH) codes and convolutional codes.

• Simulation will be stopped when all source nodes deplete their energy.(

• Based on these results, therefore, we focus attention on LDPC codes and propose a distributed LDPCdi h f WSN

5000

ges Huffman and RS coding

Huffman and centralized LDPC coding 700

800

) Huffman and RS codingH ff d li d LDPC diencoding scheme for WSNs

A h d M th d4000

d im

ag Huffman and centralized LDPC codingHuffman and distributed LDPC codingGZIP and distributed LDPC coding

i iLZO d di t ib t d LDPC di600

700

ergy

(J Huffman and centralized LDPC codingHuffman and distributed LDPC codingGZIP and distributed LDPC coding

Approach and MethodsConventional FEC Method and Problem Statement

3000

ecei

ved miniLZO and distributed LDPC coding

400

500

ual e

ne miniLZO and distributed LDPC coding

Conventional FEC Method and Problem Statement• We can divide basic FEC coding techniques into two categories: block code and convolutional code. In these

2000

er o

f re

200

300

resi

du

categories, we focus on the first type because of its low complexity. The category of block codes includesHamming codes, LDPC codes, BCH codes, and RS codes.

1000

Num

be

100

200

Tota

l

• RS codes are difficult to implement on WSNs since the sensors have their limited resources (i.e., batterypower, computational capacity, and memory). Therefore, we focus on explaining LPDC and BCH codes which

100 200 300 4000

Number of nodes

N

100 200 300 4000

Number of nodesrequire neither complex processing nor large memory [1].• The authors in [2] also show that WSNs using LDPC codes gain 45 percent more energy efficient than those

(a) Number of images (b) Total residual energy

Th bt i d lt b t i l ti ith (64 i l 64 i l ) L iusing BCH codes and 60 percent more energy efficient than those performing convolutional codes. Therefore, we focus on only LDPC codes which are the most suitable to perform on WSNs.

The obtained results by computer simulation with (64 pixels × 64 pixels) Lena image

The Conventional Scheme using LDPC coding• The numbers of received images in the proposed schemes, which use Huffman coding, GZIP which combinesLempel-Ziv (LZ77) and Huffman coding [4], or Lempel-Ziv-Oberhumer (miniLZO) with distributed LDPC

Image data LDPC encoder Modulation

p ( ) g [ ], p ( )encoding, are larger than those in the schemes compressing data using LDPC and RS encoding at the sourcenode while the energy consumptions of all schemes are almost the same

Source node

Bin

node, while the energy consumptions of all schemes are almost the same.• The results also show that the proposed scheme using Huffman and distributed LDPC coding not onlynary s

cha

achieves to send more images but also reduces the energy consumption in the network.

Comparison of Energy Consumption

symm

eannel Comparison of Energy Consumption

To evaluate energy balance we conduct the

etric

21.8

2.0

To evaluate energy balance, we conduct thesimulation with 500 sensor nodes. We measuredImage data LDPC decoder Demodulation 1.5

2

ergy

(J)

1 2

1.4

1.6

the residual energy of nodes after 2000 imageswere received at the base station. Clearly, theTh ti l h i LDPC di

Sink node 0.5

1

dual

Ene

0.8

1.0

1.2

In the scheme, encoding LDPC is performed by LDPC encoder at a source node. The encoding data then aret t th i k d di tl At th i k d th d t d d d b LDPC d d

were received at the base station. Clearly, thedistributions of residual energy of nodes in the

d h l t b l d

The conventional scheme using LDPC coding

400300400

0Res

id

0.4

0.6

0.8

The Proposed Scheme using LDPC on WSNssent to the sink node directly. At the sink node, the data are decoded by LDPC decoder. proposed scheme are almost balanced.

0100

200300

400

0100

200300

N t k Di i ( )Network Dimension (m)0.0

0.2

The Proposed Scheme using LDPC on WSNsPerforming LDPC (a) Huffman and RS coding scheme

Network Dimension (m)Network Dimension (m)

encoding at a source node2.0 2.0

Conventional LDPC encoding scheme on WSNs SinkCluster

H

Source

S

1 5

2

y (J

)

1 4

1.6

1.8

2

y (J

) 1.6

1.8

Conventional LDPC encoding scheme on WSNs SinkC ustehead

Source node

1

1.5

al E

nerg

y

1.0

1.2

1.4

1

1.5

l Ene

rgy

1 0

1.2

1.4

0

0.5

Res

idua

0.6

0.8

0

0.5

Res

idua

l

0.6

0.8

1.0Node S: capture imagesNodes Ti and Tij: perform LDPC diT11d11

100200

300400

100200

300400

0.2

0.4

200300

400

100200

300400

0R

0.2

0.4LDPC codingNodes T: gather syndromes

T

T11

T12

d11

d12d

0100

0100

Network Dimension (m)Network Dimension (m)0.0

0100

200

0100

Network Dimension (m)Network Dimension (m)0.0T1 T13

T

d13d14

(b) Huffman and centralized LDPC coding scheme (c) Huffman and distributed LDPC coding schemeT14

Td1 d21 The distributions of residual energy of sensor nodes after 2000 images received at the base stationT2

T21

T22

d1 21

d22d23 Conclusions

Th d h i th t k Q S b i i th b f i d i t th b

T2T23

T

d2

d23d24

• The proposed scheme improves the network QoS by increasing the number of received images at the basestation.HS

T24

TT

d • The network lifetime is improved by balancing energy of nodes.• Since the threshold value is set to be small in this paper we have to perform the proposed scheme with the

SinkSource node

Cluster head

T31

T32

d3 d31

d32

R f

Since the threshold value is set to be small in this paper, we have to perform the proposed scheme with thenodes in a small cluster in all of our simulations.T3

32

T33d4

32d33d34

References[1] H Y i P Qi d S H i “Th li i f hi h ld d i i i i

T11 T12 T21 T22

T34

Td [1] H. Yongqing, P. Qicong, and S. Huaizong, “The application of high-rate ldpc codes in image transmission over wireless channel,” International Conf. on Commun., Circuits and Syst. Proceedings, pp. 62–65, June

11 12 21 22T13 T14 T23 T24

T T T T

T41

T42

d41

d422006.[2] W. J. Jeong and C. T. Ee, “Forward error correction in sensor networks,” in Int’l Workshop Wireless

T31 T32 T41 T42

T33 T34 T43 T44

T442

T43d43d44

Proposed scheme for LDPC encoding on WSNs with the number of nodes N = 16

Sensor Networks (WWSN 2007), June 2007.[3] M. Sartipi and F. Fekri, “Source and channel coding in wireless sensor networks using ldpc codes,” 1st

T44Check matrix HT

Proposed scheme for LDPC encoding on WSNs with the number of nodes Nn = 16.For the check matrix HT, we divide it into 16 parts at a maximum. In this case, the input data are first

Annual IEEE Commun. Society Conf. Sensor and Ad Hoc Commun. and Netw. (IEEE SECON 2004), pp.309–316, Oct. 2004.

divided into four equal parts and transferred to four nodes T1, T2, T3, and T4. Then, the data at each node Ti

(with i = 1, 2, 3, 4) are divided again into four equal parts and sent to four other nodes Tij .[4] J. Ziv and A. Lempel, “A universal algorithm for sequential data compression,” IEEE Trans. on Infor.Theory, vol. 23, no. 3, pp. 337–343, May 1977.( ) g q p ij