Mask Learning Enabled Multi-Rate Text Semantic Communications
Mingtong Zhang, Haixia Zhang · 2024
Deep learning (DL)-based semantic communication has been deemed a promising communication paradigm to break through the bottleneck of traditional communications. However, most of the existing research focuses on designing semantic communication systems with a fixed rate, which is inflexible and inefficient for practical dynamic scenarios. To address this problem, this paper proposes a mask learning enabled multi-rate text semantic communication (MMSC) scheme. Specifically, by incorporating the target rate as side information, a lightweight rate-aware attention module is developed to improve the DL model's adaptability. Then, an effective feature masking method with learnable padding is proposed, wherein a padding-related regularization loss is designed to assist the recovery of the masked symbols. Through end-to-end mask learning, arbitrary rates with high semantic recovery performance can be achieved. Simulation results demonstrate that the proposed MMSC scheme outper-forms the other multi-rate semantic communication methods in terms of bilingual evaluation understudy (BLEU) score.