分布式协作频谱感知网络中恶意节点检测和定位方法研究

Detection and Localization of Malicious Nodes in Distributed Cooperative Spectrum Sensing Network

认知无线电是解决无线通信能量有效性问题的关键技术,其中频谱感知对于提高频谱的利用效率有着重要意义.针对基于共识的分布式协作频谱感知算法易受到恶意节点数据注入攻击,影响认知网络性能的问题,本文提出了两种基于神经网络的恶意节点检测和定位方法抵制网络内的恶意攻击行为,并采用基于Gossip Learning的联合学习策略进一步增强训练邻域检测和定位模型的鲁棒性.本文在9个认知节点的曼哈顿网络上模拟了分布式频谱感知的过程,并验证所提出方法的有效性.结果表明,所提方法具有良好的恶意节点检测和定位性能,联合学习策略能够使神经网络在样本局部有限的情况下学习到更多的攻击特征,提高本地检测和定位模型的可靠性.

Cognitive radio is a key technology to solve the problem of energy efficiency in wireless communication,and spectrum sensing is of great significance for improving the efficiency of spectrum utilization.To solve the problem that the consensus-based distributed cooperative spectrum sensing algorithm is vulnerable to malicious node data injection attacks,we propose two approaches for detecting and localizing malicious nodes based on neural networks.And a collaborative peer-to-peer machine learning protocol(Gossip Learning)is adopted to facilitate training these neural network models.We simulate the process of distributed cooperative spectrum sensing on a 9-node Manhattan network,and verify the effectiveness of the proposed approaches.Numerical results illustrate that the proposed neural network-based approaches can effectively improve the performance of detecting and localizing malicious nodes.The collaborative learning strategy can enable nodes to learn more attack characteristics,and thus make the network more robust to attacks.