基于信任度累积的联邦协作频谱感知  

Federated learning-cooperative spectrum sensing based on belief accumulation

作　　者：梁燕[1,2] 孙永鑫 Liang Yan;Sun Yongxin(School of Communication&Information Engineering,Chongqing University of Posts&Telecommunications,Chongqing 400065,China;Chongqing Key Laboratory of Signal&Information Processing,Chongqing University of Posts&Telecommunications,Chongqing 400065,China)

机构地区：[1]重庆邮电大学通信与信息工程学院,重庆400065 [2]重庆邮电大学信号与信息处理重庆市重点实验室,重庆400065

出　　处：《计算机应用研究》2025年第3期880-887,共8页Application Research of Computers

摘　　要：针对低信噪比下协作频谱感知(CSS)的检测概率低且易受表现较差节点干扰的问题,提出一种基于信任度累积的联邦协作频谱感知方法(FL-CSS)。该方法分为离线训练和在线检测两部分:在离线训练部分设计了模型融合和更新策略,保证了次用户(SU)本地模型的性能,并避免全局模型受到表现较差SU的干扰;在线检测部分设计了数据融合策略,提出信任度累积算法为SU分配权重,综合考虑了SU在模型融合中的权重、在上一次决策中的表现以及决策的历史因素的影响。FL-CSS针对CSS中SU所处环境条件不同和数据融合时易受干扰的情况,保证了检测概率,提高了融合数据决策的准确率。仿真结果表明,该方法提高了低信噪比下的检测概率,并且能有效应对干扰和攻击行为,优于当前已有的协作频谱感知方案。Aiming at the problem that CSS has low detection probability and is susceptible to interference from poorly perfor-ming nodes under low signal-to-noise ratio,this paper proposed a federated learning-cooperative spectrum sensing method(FL-CSS)based on belief accumulation.The FL-CSS was divided into two parts:the offline training part and the online detection part.It designed the model fusion and update strategy in the offline training part,which ensured the local model performance of secondary users(SU),and avoided the interference of the global model by the poor performance of SUs.It designed the data fusion strategy in the online detection part,and proposed the belief accumulation algorithm to assign weights to the SUs,which integrally took into account the weights of the SUs in model fusion,the influence of the SUs in the last decision,and the influence of the historical factors of the decision.The proposed FL-CSS method ensured the detection probability and improved the accuracy of fusion data decisions by addressing the varying environmental conditions of SUs in CSS and the susceptibility to interference during data fusion.Simulation results indicate that this method enhances the detection probability at low signal-to-noise ratios and efficiently addresses interference and aggressive behaviors,which is better than the existing cooperative spectrum sensing scheme.

关 键 词：频谱感知 协作频谱感知 联邦学习 信任度 

分 类 号：TN925[电子电信—通信与信息系统]