Multi-level Reliability Interface for Semantic Communications over Wireless Networks

TL;DR

The paper addresses semantic communications over wireless networks under practical TCP/IP constraints by decoupling source and channel design through a binary, multi-level reliability interface. It introduces Split DeepJSCC, a two-stage scheme where the source mapper/demapper are trained first on a fixed multi-level medium, followed by training of the channel mapper/demapper to align with the source’s reliability profile, all via variational learning. The key contributions include a formal multi-level interface, a two-phase training procedure with probabilistic encoders that are later determinized, and empirical results on CIFAR-10 showing lower end-to-end distortion and graceful degradation relative to digital baselines and fully joint DLJSCC. This work demonstrates that semantic communication gains can be realized within existing network architectures by separating design responsibilities while preserving effectiveness at the application layer, enabling practical deployment in wireless networks.

Abstract

Semantic communication, when examined through the lens of joint source-channel coding (JSCC), maps source messages directly into channel input symbols, where the measure of success is defined by end-to-end distortion rather than traditional metrics such as block error rate. Previous studies have shown significant improvements achieved through deep learning (DL)-driven JSCC compared to traditional separate source and channel coding. However, JSCC is impractical in existing communication networks, where application and network providers are typically different entities connected over general-purpose TCP/IP links. In this paper, we propose designing the source and channel mappings separately and sequentially via a novel multi-level reliability interface. This conceptual interface enables semi-JSCC at both the learned source and channel mappers and achieves many of the gains observed in existing DL-based JSCC work (which would require a fully joint design between the application and the network), such as lower end-to-end distortion and graceful degradation of distortion with channel quality. We believe this work represents an important step towards realizing semantic communications in wireless networks.

Figures (9)

• Figure 1: Diagram illustrating how application and network providers can interoperate by adhering to the proposed binary multilevel reliability interface, eliminating the need for direct communication or synchronization, and thus realizing the benefits of in practice.
• Figure 2: Diagram depicting training process of the source mapper and demapper. The blue boxes are trainable components of the architecture.
• Figure 3: Diagram illustrating the training process of channel mapper and demapper. The green boxes are trainable parts of the architecture.
• Figure 4: architectures used for source mapper ($f_{\boldsymbol{\theta}}$) and demapper ($g_{\boldsymbol{\eta}}$).
• Figure 5: architectures for channel mapper ($h_{\boldsymbol{\psi}}$) and demapper ($q_{\boldsymbol{\phi}}$). The two arrows that merge into one represent the concatenation operation.

Theorems & Definitions (2)

• Lemma 1
• Lemma 2