LLM-MHPSC framework slashes distortion in multi-hop image transmission

A team of researchers from multiple institutions has introduced LLM-MHPSC, a large language model-enhanced multi-hop parallel image semantic communication framework designed to tackle the persistent problem of distortion accumulation in multi-hop wireless image transmission. Unlike conventional single-hop semantic communication systems, LLM-MHPSC deploys an extra residual compensation link at each intermediate hop to counteract errors that build up across the transmission chain. To keep bandwidth overhead minimal, the framework employs a coarse-to-fine residual compression scheme that integrates a deep learning-based compressor with adaptive arithmetic coding (AAC), ensuring efficient use of radio resources.

The framework's core innovation is the LLM-based Residual Transmission Optimizer (LLM-RTO), which accurately estimates the distribution of residual errors and enables channel state- and hop-aware rate adjustment. This allows the system to dynamically optimize compression efficiency under varying channel conditions and network depths. Additionally, an adaptive hop selection strategy activates the residual compensation link only when needed, striking a balance between transmission quality and computational cost. Experimental results demonstrate that LLM-MHPSC outperforms both state-of-the-art semantic communication systems and traditional transmission methods, achieving robust image quality with only a marginal bandwidth increase. This work paves the way for practical deployment of semantic communication in real-world multi-hop scenarios like drone swarms or mesh networks.