Design, performance, and complexity analysis of residual trellis-coded vector quantizers

In this paper multistage trellis-coded vector quantization (MS-TCVQ) is developed as a constrained trellis sourcecoding technique. The performance of the two-stage TCVQ is studied for Gaussian sources. Issues of stage-by-stage design, output alphabet selection, and complexity are addressed with emphasis on selecting and partitioning the stage codebooks. For a given rate, MS-TCVQ achieves low encoding and storage complexity compared to TCVQ, and comparisons with samedimensional multistage vector quantization indicate a 0.5–3-dB improvement in signal-to-quantization-noise ratio.