A structured fixed-rate vector quantizer derived from a variable-length scalar quantizer. I. Memoryless sources

A low-complexity, fixed-rate structured vector quantizer for memoryless sources is described. This quantizer is referred to as the scalar-vector quantizer (SVQ), and the structure of its codebook is derived from a variable-length scalar quantizer. Design and implementation algorithms for this quantizer are developed and bounds on its performance are provided. Simulation results show that performance close to that of the optimal entropy-constrained scalar quantizer is possible with the fixed-rate quantizer. The SVQ is also robust against channel errors and outperforms both Lloyd-Max and entropy-constrained scalar quantizers for a wide range of channel error probabilities.<
>