Rate-Constrained Trellis-Coded Quantization for Large-Scale Noisy Graph Signals

Abstract

We consider the issue of compressing large-scale noise-corrupted graph signals under a rate constraint, to tackle the communication resource limitations, from rate-distortion perspective. To guarantee the fidelity of the overall compression system for noisy graph signals, we adopt the technique of kernel ridge regression on graphs for preprocessing. We show that, as a compression component for the output of the kernel ridge regression, trellis-coded quantization has superior distortion performance in comparison to other feasible quantization methods for long source blocks, and hence is suitable for large-scale graph signals. Targeting at decreasing further distortion in the multiple trellis-coded quantizer system under rate restriction, we design a novel rate allocation scheme based on the intrinsic topology of graph signals and the rate-distortion characteristics of trellis-coded quantization. We perform sufficient simulation with in-depth analysis, which demonstrates both the effectiveness of the proposed trellis-coded quantization for large-scale noisy graph signals and the advantage of the proposed rate allocation scheme over the existing rate allocation schemes in the common distortion measure under communication resource constraint.