Simple adaptation of vector-quantizers to combat channel errors

Abstract

Vector quantization (VQ) is an effective and widely-used method for low-bit-rate communication of speech and image signals. A common assumption in the analysis of VQ systems is that the compressed digital information is transmitted through a perfect channel. Under this assumption, quantizing distortion is the only factor affecting output signal fidelity. However, in physical channels, errors may be present, degrading overall system performance. In order to reduce performance degradation, previous authors suggested to optimally redesign the VQ for noisy channels ("noisy" VQ). The "noisy" VQ results in smaller distortion as compared to the original ("noiseless") VQ for the specific channel it was designed for. The main drawback of this approach is the need to design and store, both in the transmitter and the receiver, several codebooks for different bit error rates (BER). We show that a simple gain adaptation, which depends on the channel BER, also improves system performance, while using the original ("noiseless") VQ design. The improvement in some cases, as shown by numerical examples, is close to what can be achieved by the optimal approach.<>