Flexible Deep Joint Source-Channel Coding: A Vibrotactile Example

2026-08-15

AI Summary

This study addresses the limitations of fixed-rate, high-storage, and bandwidth-inflexible joint source-channel coding (JSCC) by proposing the fd-JSCC framework. The method achieves end-to-end optimization through hierarchical gain adaptation, rate-switchable residuals, and channel feature processing modules, enabling dynamic bandwidth adaptation and enhanced noise robustness without model switching. Experimental results demonstrate that fd-JSCC matches the reconstruction performance of fixed-rate baselines while supporting four flexible transmission rates. Notably, it reduces storage requirements by 61.1%, thereby facilitating efficient and adaptive transmission of vibrotactile signals under varying channel conditions.

Abstract

The increasing demand for real-time tactile communication in multimedia systems has exposed the limitations of existing Joint Source-Channel Coding (JSCC) techniques. While current JSCC models facilitate end-to-end optimization, they typically operate at fixed coding rates and require separate model instances for different rate settings. This results in significant storage overhead and limited adaptability to dynamic bandwidth conditions. To address these challenges, we propose the Flexible Deep Joint Source-Channel Coding (FD-JSCC) framework for vibrotactile signals, which supports flexible-rate transmission without the need for model switching. The FD-JSCC integrates a flexible-rate encoder-decoder enhanced with Hierarchical Gain Adaptation Module (HGAM) and Rate-Switchable Residual Module (RSRM), enabling bitrate-aware compression by selectively preserving salient vibrotactile features. Additionally, we introduce a Channel Feature Processing Module (CFPM), which leverages real-time SNR information to enhance robustness against channel noise and signal degradation. Trained on the IEEE 1918.1.1 vibrotactile dataset, FD-JSCC achieves reconstruction performance comparable to fixed-rate baselines (e.g., DeepSC-S), while reducing storage requirements by 61.1\% when supporting four rates. These results underscore its potential for scalable, low-latency tactile communication in next-generation networks.

Problem

• Joint Source-Channel Coding
• Vibrotactile Communication
• Flexible Rate Transmission
• Storage Overhead
• Dynamic Bandwidth Adaptation

Innovation

• Flexible Deep JSCC
• Vibrotactile Communication
• Hierarchical Gain Adaptation Module
• Rate-Switchable Residual Module
• Channel Feature Processing Module