Multiterminal source coding: sum-rate loss, code designs, and applications to video sensor networks

摘要

Driven by a host of emerging applications (e.g., sensor networks and wireless video),distributed source coding (i.e., Slepian-Wolf coding, Wyner-Ziv coding and various otherforms of multiterminal source coding), has recently become a very active research area.This dissertation focuses on multiterminal (MT) source coding problem, and consistsof three parts. The first part studies the sum-rate loss of an important special caseof quadratic Gaussian multi-terminal source coding, where all sources are positively symmetricand all target distortions are equal. We first give the minimum sum-rate for jointencoding of Gaussian sources in the symmetric case, and then show that the supremum ofthe sum-rate loss due to distributed encoding in this case is 12 log254 = 0:161 b/s when L = 2and increases in the order of?L2 log2 e b/s as the number of terminals L goes to infinity.The supremum sum-rate loss of 0:161 b/s in the symmetric case equals to that in generalquadratic Gaussian two-terminal source coding without the symmetric assumption. It isconjectured that this equality holds for any number of terminals.In the second part, we present two practical MT coding schemes under the frameworkof Slepian-Wolf coded quantization (SWCQ) for both direct and indirect MT problems.The first, asymmetric SWCQ scheme relies on quantization and Wyner-Ziv coding, and itis implemented via source splitting to achieve any point on the sum-rate bound. In the second,conceptually simpler scheme, symmetric SWCQ, the two quantized sources are compressedusing symmetric Slepian-Wolf coding via a channel code partitioning technique that is capable of achieving any point on the Slepian-Wolf sum-rate bound. Our practicaldesigns employ trellis-coded quantization and turbo/LDPC codes for both asymmetric andsymmetric Slepian-Wolf coding. Simulation results show a gap of only 0.139-0.194 bit persample away from the sum-rate bound for both direct and indirect MT coding problems.The third part applies the above two MT coding schemes to two practical sources, i.e.,stereo video sequences to save the sum rate over independent coding of both sequences.Experiments with both schemes on stereo video sequences using H.264, LDPC codes forSlepian-Wolf coding of the motion vectors, and scalar quantization in conjunction withLDPC codes for Wyner-Ziv coding of the residual coefficients give slightly smaller sumrate than separate H.264 coding of both sequences at the same video quality.