Quantization for decentralized estimation from correlated data

A decentralized estimation scheme with two sensors, where data are remotely encoded by scalar quantizer to fulfill capacity constraints of channels conveying information to a central estimator, has been investigated. The quantizers are to operate separately, but a joint model of observations, allowing for spatially correlated data, may be taken into account at the design stage. A design algorithm has been derived. For a linear Gaussian observation model the limit performance of any encoding-decoding scheme has been obtained in terms of an upper bound to the rate-distortion function. Under various combinations of per-channel signal-to-noise ratios and spatial correlation coefficients, quantizers designed by the proposed algorithm nearly achieve such limit performance.