面向协作频谱感知的个性化差分隐私联邦学习方法OA

To address the degradation of model performance caused by data non-independent and identically distributed(Non-IID)characteristics in collaborative spectrum sensing,a federated learning scheme RebalFL was proposed,which integrated personalized differential privacy with a rebalancing clustering strategy.First,a personalized differential pri-vacy mechanism that allowed heterogeneous privacy budgets for different data sources was introduced,thereby reducing noise injection while preserving privacy.Then,a rebalancing clustering strategy was designed to form client clusters with more balanced data distributions and mitigate model drift.Experimental results show that RebalFL outperforms existing differential privacy methods in Non-IID scenarios,substantially improving the classification accuracy and robustness of spectrum sensing models under privacy protection.