A new trellis-coded residual scalar quantizer

Trellis coded quantization (TCQ) is generalized to allow the use of trellis coding with residual quantization. Design techniques for trellis coded residual quantization are discussed and their optimality under various conditions is analyzed. Several quantizer optimization algorithms are discussed. A trellis-into-trellis structure is introduced as a means to obtain additional gain in the residual quantizer. The trellis structure provides a mechanism to capture the cell shape gain, while the multistage structure provides a memory efficient representation. This trellis-based residual structure is shown to yield significant improvement over conventional residual (or multistage) quantizers. The trellis-into-trellis structure is then coupled with a variable length coder in an entropy constrained design procedure. The entropy constrained coder is shown to have performance advantage over entropy constrained TCQ for standard sources.