Lightweight Semantic-Aware Communication With Packet Transmission

This work is concerned with a lightweight semanticaware communication framework named DeepJSCC-L, aiming at maximizing rate-distortion (RD) performance under the resource-constrained scenario. Firstly, a lightweight coding is designed, which employs parameter-efficient operators including a lightweight convolution and an efficient self-attention mechanism. Next, a novel packet transmission strategy is proposed, which aims to further improve RD performance under varying channel environments by allocating crucial packets with a optimal channel. Experimental results prove that DeepJSCC-L is competitive compared with benchmark in terms of RD performance and model complexity.