Feature Partitioning and Semantic Equalization for Intrinsic Robustness in Semantic Communication under Packet Loss
Xiao Yang, Shuai Ma, Yong Liang, Guangming Shi
TL;DR
This paper tackles intrinsic robustness in semantic communication under packet loss by examining how different feature partitioning schemes affect performance across Transformer and CNN backbones. It finds that partitioning along the channel dimension yields better resilience for both architectures, while CNNs suffer from imbalanced channel importance. To address this, the authors introduce Semantic Equalization Mechanism (SEM), comprising a dynamic scale module and a neighboring semantic broadcast module, which balance channel contributions and enable graceful degradation with minimal overhead. Experimental results on COCO/Kodak/DIV2K show CNNs with SEM achieve lossless-to-lossy PSNR degradation comparable to Transformer baselines (retaining substantial quality even at higher loss), suggesting that balanced semantic representations are key to robustness and that SEM can extend to other modalities and task-oriented settings.
Abstract
Semantic communication can improve transmission efficiency by focusing on task-relevant information. However, under packet-based communication protocols, any error typically results in the loss of an entire packet, making semantic communication particularly vulnerable to packet loss. Since high-dimensional semantic features must be partitioned into one-dimensional transmission units during packetization. A critical open question is how to partition semantic features to maximize robustness. To address this, we systematically investigate the performance of two mainstream architectures, Transformer and Convolutional neural networks (CNN), under various feature partitioning schemes. The results show that the Transformer architecture exhibits inherent robustness to packet loss when partitioned along the channel dimension. In contrast, the CNN-based baseline exhibits imbalanced channel utilization, causing severe degradation once dominant channels are lost. To enhance the CNN resilience, we propose a lightweight Semantic Equalization Mechanism (SEM) that balances channel contributions and prevents a few channels from dominating. SEM consists of two parallel approaches: a Dynamic Scale module that adaptively adjusts channel importance, and a Broadcast module that facilitates information interaction among channels. Experimental results demonstrate that CNN equipped with SEM achieve graceful degradation under packet loss (retaining about 85% of lossless PSNR at 40% packet loss), comparable to that of Transformer models. Our findings indicate that, under an appropriate partitioning strategy, maintaining a balanced semantic representation is a fundamental condition for achieving intrinsic robustness against packet loss. These insights may also extend to other modalities such as video and support practical semantic communication design.