Distributed source coding in sensor networks: A practical implementation
Abstract
In this thesis we take a closer look at wireless sensor networks and source coding. A necessary condition for this work to have any meaning is that the sensors in the network are spatially co-located and that there is correlation between the data the sensors observe. When there is correlation, there is redundancy in the information communicated that can be removed by source coding techniques. This can be done by emph{distributed source coding}. Slepian and Wolf showed theoretically that there is no rate loss no matter if the sensors are communicating. cite{slepian73} Wyner and Ziv expanded this from the lossless case of Slepian and Wolf to apply to lossy source coding. cite{wyner76} Pradhan and Ramchandran found a practical implementation for the theory of Slepian-Wolf and Wyner-Ziv based on channel coding principles. cite{pradhan03} This can be done because the correlation between any two sources can be modelled as a channel with an error probability. We build our work on their ideas. The channel coding technique we have found most advantageous for this scheme is emph{Low Density Parity-Check} coding. LDPC coding is the most advanced form of linear block coding up to date. It is represented by a sparse parity-check matrix. While LDPC coding in the traditional sense is used for bandwidth expansion of the source to protect it from channel errors, it is used for bandwidth compression, or rate reduction, in the distributed sense. The distributed LDPC scheme is used on medical ECG data as an example. Due to lack of time and the comprehensive task, the adpapted message-passing decoding algorithm needed to fulfill the implementation could not be finished. We have illustrated the distributed encoder system with a $(7,4)$-Hamming code to give an example. The performance of this system is not good enough for any practical use, but will function as a guideline for possible future work in the area.