A Task-Oriented and Lightweight Semantic Communication System With Secure Federated Aggregation in Distributed Wireless Networks

Semantic communication (SemCom) has recently emerged as a promising paradigm for enhancing the efficiency and intelligence of wireless networks. Nevertheless, device heterogeneity, resource constraints, and the vulnerability of deep neural networks in open environments pose significant challenges to its practical deployment. In this paper, we propose a task-oriented and lightweight SemCom system with secure aggregation for ensuring efficient and privacy-preserving interactions in distributed networks. First, we design a multi-task SemCom framework that unifies semantic feature extraction from sample-based datasets. To accommodate resource-constrained devices, we further introduce a feature distillation mechanism that derives lightweight local models without sacrificing inference accuracy. To preserve the privacy of local datasets while leveraging the generalization capability of distributed devices, we develop a secure model aggregation algorithm based on multiparty homomorphic encryption. Simulation results and comparative experiments validate the effectiveness of our system, which fully utilizes the knowledge embedded in existing high-performance models. Our results demonstrate that the proposed local semantic models outperform the baseline models under limited datasets and reduced parameters. We also analyze the trade-off between computational complexity and security in the proposed aggregation scheme, highlighting its applicability to distributed SemCom scenarios.