On Complexity Reduction of Trellis-Based Scalar-Vector Quantizers by Signaling of Regions

This memo presents a method to reduce the complexity of trellis-based scalar-vector quantizers (TB-SVQ). Based on Calderbank and Ozarow's classical idea, the reproduction alphabet is partitioned into equal area regions and the codebook is shaped subject to a rate-reduced signaling (selection of sequences) of these regions. As the shaping bit rate decreases, the coder complexity is significantly reduced, which, in principle, does not trade off any rate-distortion performance. In addition, the complexity is approximately rate-independent, making the TB-SVQ more appealing for high rate coding applications.