Semantic Communication System Based on Meta-Learning Framework

With the continuous advancements in deep learning technology, semantic communication, particularly deep joint source-channel coding (DJSCC), has garnered significant attention for its potential to enhance compression efficiency, reduce transmission delays and simplify system complexity. However, deep learning usually relies on large datasets for training and exhibits certain limitations in its generalizability. Therefore, this letter proposes a semantic communication system based on a meta-learning (ML) framework. This system is designed to achieve high transmission reliability and generalizability, even in communication scenarios with limited data. Furthermore, the integration of second-order optimization principles with semantic communication mechanisms significantly enhances the model’s training performance and generalization capability in complex task scenarios. Simulation results demonstrate that the proposed scheme outperforms existing few-shot semantic communication models in terms of transmission accuracy and robustness.