Low-complexity semantic communication system based on lite transformer

Semantic communication has gained increasing attention with traditional communication technologies approaching Shannon's physical capacity limit. However, most existing semantic communication systems focus on improving system performance while neglecting model simplification. Therefore, we propose an efficient semantic communication system based on Lite Transformer for text transmission. Unlike the Transformer model used in most studies, Lite Transformer introduces a dynamic convolution module to capture local features, combined with a self-attention module that focuses on global information to jointly accomplish semantic extraction. Simulation results demonstrate that the proposed system exhibits lower complexity and superior transmission performance, and robustness compared with traditional communication systems and typical semantic communication systems based on standard transformers.