On the smoothing of trellis coded quantization

In this paper we examine the possibility of applying fixed-lag smoothing techniques to trellis coded quantization (TCQ). For this application, we exploit the tree structure of the predictive TCQ (P-TCQ). We apply the (M,L) algorithm for searching the tree and Kalman smoother to find the smoothed estimates of the data. The performance of the resulting smoothed TCQ (S-TCQ) is compared to that of some other predictive quantization techniques such as predictive TCQ (P-TCQ), smoothed differential pulse code modulation (S-DPCM) and scalar DPCM codes. We find that the proposed S-TCQ outperforms all the other schemes mentioned above by more than 1 dB when applied to several synthetic sources.