A Novel Lightweight Joint Source-Channel Coding Design in Semantic Communications

Semantic communication has emerged as a promising solution to meet the growing demand for efficient data transmission in the information age. Unlike traditional communication methods that focus on transmitting raw data, semantic communication prioritizes preserving the meaning of transmitted information, which significantly reduces the data volume. However, implementing semantic communication systems in resource-constrained environments, such as Internet of Things (IoT) devices, remains challenging due to limited computational resources. In this letter, we propose a novel lightweight deep learning (DL) model, termed the lightweight image compression and reconstruction network (LICRnet). LICRnet leverages depthwise separable convolution (DSC) and a local and nonlocal mixture (LNLM) block to significantly reduce computational costs. Additionally, the LNLM incorporates a variable window size-based multiscale attention mechanism (VW-MSA), enabling it to effectively learn from both local detailed features and global high-level meaningful features. Extensive simulations demonstrate that LICRnet significantly reduces computational complexity while maintaining satisfactory image compression and reconstruction performance, making it highly suitable for deployment in resource-constrained environments.