Entropy-Constrained Trellis Coded Quantization: Implementation and Adaptation

Abstract

Entropy-constrained trellis coded quantization (ECTCQ) of memoryless sources is known to be an efficient source coding technique in the rate-distortion sense. We develop an ECTCQ scheme that employs a symmetric reproduction codebook. The symmetry of the reproduction codebook, while essentially costs no performance loss, is exploited to reduce the memory requirement in entropy coding the ECTCQ output. In practice, a buffer of finite, and preferably small, size is needed to interface the variable-length codewords to the fixed-rate channel. An adaptive ECTCQ (A-ECTCQ) scheme, which uses a buffer-state feedback to control the quantizer characteristics to avoid buffer overflow/underflow, is studied in this work. The choice of encoding delay is an important issue in A- ECTCQ, as too long a delay will adversely impact the performance of the feedback control. We propose a pathwise-adaptive ECTCQ (PA-ECTCQ) that solves the encoding delay problem. Simulation results indicate that, while the buffer overflow/underflow problems of the PA- ECTCQ can be practically eliminated, the overall quantization distortion is increased only negligibly over theoretical performance predictions. Our experiments also suggests that PA- ECTCQ is robust with respect to source mismatch.