Practical source-network decoding

• Main Authors: Maierbacher, Gerhard (Author), Barros, Joao (Author), Medard, Muriel (Contributor)
• Other Authors: Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science (Contributor), Massachusetts Institute of Technology. Laboratory for Information and Decision Systems (Contributor), Massachusetts Institute of Technology. Research Laboratory of Electronics (Contributor)
• Format: Article
• Language: English
• Published: Institute of Electrical and Electronics Engineers, 2010-11-12T19:02:26Z.
• Subjects: Article
• Online Access: Get fulltext

 When correlated sources are to be communicated over a network to more than one sink, joint source-network coding is, in general, required for information theoretically optimal transmission. Whereas on the encoder side simple randomized schemes based on linear codes suffice, the decoder is required to perform joint source-network decoding which is computationally expensive. Focusing on maximum a-posteriori decoders (or, in the case of continuous sources, conditional mean estimators), we show how to exploit (structural) knowledge about the network topology as well as the source correlations giving rise to an efficient decoder implementation (in some cases even with linear dependency on the number of nodes). In particular, we show how to statistically represent the overall system (including the packets) by a factor-graph on which the sum-product algorithm can be run. A proof-of-concept is provided in the form of a working decoder for the case of three sources and two sinks.