Research On Distributed Information Compression Algorithms

The booming development of wireless networks and multimedia services, e.g. the data-intensive wireless sensor networks and rich mobile video applications, challenges the traditional information compression technologies. Large quantities of data need to be processed, transmitted and exchanged inside the wireless sensor networks which consist of numerous tiny sensor devices. However, the traditional information compression algorithm is usually with high-complexity and needs inter-sensor communication, so it is difficult to be applied on the sensors with limited energy and limited computational capabilities. There are similar problems in video compression on the mobile visual devices where the complicated inter-frame prediction of traditional video coding algorithms is a heavy burden. In addition, high resolution video applications also require low-complex compression technologies.Considering these emerging requirements, the distributed information compression theories and technologies with low-complexity are widely investigated. In the distributed information compression, the correlated sources are separately encoded and jointly decoded such that the complexity is shifted from the encoder to the decoder, and the problems mentioned above are effectively solved. Thus, the investigation of the distributed information compression algorithms has very important academic and practical significance.In this thesis, the research work on distributed information compression algorithms is unfolded mainly from two aspects:the distributed source coding and the distributed compressed sensing. The bridge between these two aspects is the compressed sensing which is also investigated. The main contributions of this thesis are as follows:In distributed source coding, a new scheme based on the multilevel codes is proposed. The multi-stage decoding algorithm is designed using low density parity check codes, so the side information and the decoded bits can improve the decoding performance. The proposed scheme provides flexible coding rates and unequal error protection to cover the shortage of the traditional methods. By comparison with the existing distributed source coding scheme using single channel codes, the bit error rate and relative distortion are obviously improved in the proposed scheme.Considering the noise of practical transmission, a distributed joint source-channel coding scheme with noisy side information is proposed. The achievable rate region of this scheme is derived. By using the rate-compatible low density parity check codes, the proposed scheme can provide arbitrary rate allocation between sources. Moreover, the correlation among more than two sources is considered. The noisy side information is involved in algorithms design and the theoretical analysis, which is more reasonable in practice. The simulation results verify that the proposed scheme outperforms the scheme using turbo codes, and the bit error rates will be further decreased by considering more side information in the joint decoding algorithm.The sparse signals with non-uniform distribution are found in abundance, so a novel compressed sensing algorithm with unequal protection capabilities is proposed in this thesis by introducing the windowing strategy, called expanding window compressed sensing. Comparing with the ordinary compressed sensing method, the proposed scheme provides enhanced protection for the more important parts of the signal, so the better overall recovery quality is obtained. This advantage is theoretically analyzed and experimentally confirmed. Moreover, the algorithm is applied to the compressed acquisition of image signals, and is verified to have better recovery quality than the ordinary compressed sensing algorithm and the existing unequal protection algorithm. In addition, the proposed algorithm is utilized to provide distributed compression of the networked data in wireless sensor networks where it also has superior performance.Another major content of this thesis is the distributed compressed sensing which is originated from the combination of the compressed sensing and the distributed source coding. A distributed compressed sensing algorithm based on the belief propagation is proposed, and the achievable sensing rates are analyzed. The joint recovery algorithm for the correlated sparse signals is designed by modifying the compressed sensing algorithm via belief propagation with the side information. The proposed algorithm has lower recovery distortion as well as better noise-resilient performance than both of the independent compressed sensing algorithm and the distributed compressed sensing algorithms based on l1-norm optimization.Then, the proposed distributed compressed sensing algorithm is applied to video compression, so a robust distributed compressed video sensing scheme is proposed. According to the specific features of the video signal, the sensing matrix and recovery method are both modified. From the simulation results on the testing video sequences, the proposed scheme can guarantee better recovery quality at low signal to noise ratios.Based on the robust distributed compressed video sensing scheme, another video sensing algorithm is proposed to resist the packet loss. By exploiting the multi-resolution feature of the wavelet transform coefficients, a layered sensing matrix is designed so that the more significant coefficients are preserved with higher probability during the transmission over the packet loss channel. Thus, the overall recovery quality of the video sequence is improved. The proposed scheme is dramatically superior to the layered measurement scheme.Considering the event detection applications in the wireless sensor networks, a multiple events detection scheme via distributed compressed sensing is proposed. By comparison with the traditional event detection methods, the proposed scheme can not only detect the events' locations, but also estimate the actual values of the events without additional cost. Furthermore, the temporal correlation between the events at adjacent time instants is exploited to get higher detection accuracy.Finally, the conclusion is given, and some future research directions in distributed information compression domain are discussed.