移动通信2025，Vol.49Issue(7)：12-20,9.DOI:10.3969/j.issn.1006-1010.20250524-0001
面向带宽与内存双受限的高效语义通信方法
An Efficient Semantic Communication Method for Bandwidth and Memory Constraints
摘要
Abstract
To address the challenges of improving semantic communication performance and deploying models in the Internet of Things(IoT)scenarios with simultaneous bandwidth and memory constraints,we propose a lightweight semantic communication method based on dynamic knowledge distillation(L-SCDKD).The technique consists of two stages,namely knowledge transfer based on dynamic knowledge distillation and network pruning.First,the method builds a teacher model with a complex network structure and large parameter size,as well as a student model with a smaller scale,based on an end-to-end semantic communication method grounded in the information bottleneck principle.Knowledge transfer is performed using dynamic knowledge distillation,where the teacher model assists in training the student model to enhance its performance.Next,network pruning is applied to compress the student model further,ultimately yielding the lightweight semantic communication model.Experimental results show that,compared to existing semantic communication methods,the proposed L-SCDKD method improves semantic communication performance under bandwidth constraints,while significantly reducing model parameters and memory requirements.This makes it more feasible to deploy on memory-constrained intelligent devices,making it suitable for IoT scenarios.关键词
语义通信/深度学习/轻量化/知识蒸馏/带宽受限/内存受限Key words
semantic communication/deep learning/lightweight/knowledge distillation/bandwidth-constrained/memory-constrained分类
信息技术与安全科学引用本文复制引用
刘伟,王孟洋,牛文洁..面向带宽与内存双受限的高效语义通信方法[J].移动通信,2025,49(7):12-20,9.基金项目
国家重点研发计划项目(2021YFA1000500) （2021YFA1000500）
陕西省重点研发计划项目(2022ZDLGY05-05) （2022ZDLGY05-05）
