Research On Key Technologies Of Distributed Source Coding Based On Low-Density Parity-Check Code

This thesis mainly researches key technologies of Distributed Source Coding (DSC) based on Low-Density Parity-Check (LDPC) code, and the background of this thesis is the data transmission in Wireless Sensor Networks (WSNs) for seismic exploration for mine. One problem in the application of WSNs is the energy usage of sensors, especially when sensors are distributed in vast fields. Since sensors are placed in the same detection zone, there are usually some correlations between the data collected by sensors. If sources are compressed, the transmission energy can be reduced efficiently. Distributed source coding is a way to compress data with correlation. Different from traditional ways of source compression, sensors do not have to communicate with others to exchange information. Source codes are encoded dependently and decoded jointly, so the coding complexity and energy usage of sensors are reduced. The Slepian-Wolf (SW) theorem shows that the dependent encoding and jointed decoding also can achieve the compressed rate achieved by the jointed lossless coding.This thesis discusses the key technologies of DSC mainly from two perspectives:theory and implementation.From the perspective of theoretic research, this thesis starts with the simulation of schemes in other published paper. There are mainly two kinds of DSC methods:the syndrome-based method and the parity-based method. However, the former is extremely sensitive to channel noise, and unlikely effective in WSNs. The latter is more robust in WSNs, the distributed jointed source-channel (DJSC) coding in particular, which transmits some additional parity check bits to deal with noises. Based on the published results, we have done some research a step further, combining with the background of our WSN. In total, we mainly have two innovation points.The first innovation point is that we propose a parallel structure of jointed decoding, and decoders of multiple sources can decode simultaneously to reduce the latency of jointed decoding. We make use of the soft information of a posteriori probability (APP) returned by decoders of other sources and let decoders re-initialize once during iterations. To make decoders with parallel structure work more efficiently, layered logarithm belief propagation (Log-BP) algorithm is employed. Simulation results show that, decoders with the proposed parallel structure can achieve a performance as good as, or even better than that of decoders with serial structures, while the latency of them can be reduced nearly a half.The second innovation point is that we modify the regular coding scheme since the data collected by sensors in our WSN has its own features. Simulation results show that the modification of encoding scheme has a performance gain compared with that of the regular one.Additionally, from the perspective of implementation, we implement the encoder and decoder in FPGA chips. Since the H matrix we adopt has the same form of the H matrix in802.16e standard, we implement encoder with H matrix. During the implementation of decoder, we optimize the RTL codes to reduce latency and hardware resources. Finally, we test the performance of the encoder and decoder on FPGA board, and provide key modules for wireless transmission in our WSN.