Bit-estimate based decoding for vector quantization over noisy channels with intersymbol interference

We introduce new techniques for quantization over noisy channels with intersymbol interference. We focus on the decoding problem, and present a decoder structure that allows the decoding to be based on soft minimum mean square-error estimates of the transmitted bits. The new bit-estimate based decoder provides a structured lower-complexity approximation of optimal decoding for general codebooks, and for so-called linear mapping codebooks, it is shown that its implementation becomes particularly simple. We investigate decoding based on optimal bit-estimates, and on suboptimal estimates of lower computational complexity. We also consider encoder optimization and combined source-channel code design. Numerical simulations demonstrate that bit-estimate based decoding is able to outperform a two-stage decision-based approach implemented using Viterbi sequence detection plus table look-up source decoding. The simulations also show that decoding based on suboptimal bit-estimates performs well, at a considerably lowered complexity.