Abstract
Distributed Source Coding (DSC) has gained wide popularity in applications such as video coding and distributed sensor networks. However, DSC has not been widely explored for use in low delay, low bit rate applications such as quantization of speech parameters. This is due to the difficulty of designing quantizers with imperfect side information resulting from decoding errors, quantization noise and packet losses. We address this fundamental problem by modeling the decoder as a three state system based on the packet losses/receptions and correct/incorrect decodings. Then we derive the expressions for the error variance of each of the states and solve them under stationary conditions. This enables one to model, analyze and design scalar uniform coset quantizers for imperfect side information. Simulation results verify the improved performance of the designed WZ quantizers over predictive and non-predictive quantization schemes in packet loss.