Forward-Aware Multi-Source Distributed Vector Quantization for Noisy Channels via Information Bottleneck Principle

A generic distributed / multiterminal setup is considered here wherein, through a joint design, several intermediate nodes must locally quantize their noisy observations from various sets of potentially common and uncommon user / source signals, ahead of a forward transmission over multiple error-prone and rate-limited channels to a remote processing unit. The local compressors should be designed in such a fashion that the impacts of the noisy forward channels are taken into account as well. Fully aligned with the principal idea of the Information Bottleneck (IB) method, the Mutual Information is then selected here as the fidelity criterion, and by means of Variational Calculus, the corresponding stationary solutions are derived for two various types of processing flow / strategy. Thereupon, an iterative algorithm, the Forward-Aware GEneralized Multivariate IB (FAGEMIB), is introduced as well to efficiently tackle the challenging (non-convex) design problems. The pertinent convergence proofs to a stationary point of the objective functionals are also provided, together with a couple of numerical results on typical digital data transmission scenarios, corroborating the effectiveness of these joint source-channel coding techniques. The presented compression schemes in this article, with the most flexibility w.r.t. the assignment of users to the serving nodes, highly generalize the State-of-the-Art techniques designed exclusively for a single (common) user / source signal.