Source-Optimized Clustering for Distributed Source Coding

Motivated by the design of low-complexity distributed quantizers and iterative decoding algorithms that leverage the correlation in the data picked up by a large-scale sensor network, we address the problem of finding correlation preserving clusters. To construct a factor graph describing the statistical dependencies between sensor measurements, we develop a hierarchical clustering algorithm that minimizes the Kullback Leibler Distance between known and approximated source statistics. Finally, we show how the clustering result can be exploited in the design of index assignments for distributed quantization and source-channel decoders of manageable complexity.