Efficient distributed source coding for multiple receivers via matrix sparsification

Abstract

Consider the problem of source coding with side information in large networks with multiple receivers. In this case, standard coding techniques are either prohibitively complex to decode, or require source-network coding separation, resulting in sub-optimal transmission schemes. To alleviate this problem, we offer a joint network-source coding scheme based on matrix sparsification at the code design phase, which allows the terminals to use an efficient decoding procedure (syndrome decoding using LDPC), despite the network coding throughout the network. Via a novel relation between matrix sparsification and rate-distortion theory, we give lower and upper bounds on the best achievable sparsification performance, and analyze our scheme in the limit of weak side information at the receivers. Simulation results motivate the use of this scheme at non-limiting rates as well.