Federated Learning-Based Jamming Detection for Distributed Tactical Wireless Networks

In this paper, we propose a federated learning (FL)-based JDWC algorithm for distributed tactical wireless networks (TWNs). Specifically, we consider a distributed TWN with multiple clusters under the presence of a mobile jammer, where various types of waveforms are used over the network. On local servers, we perform frequency domain analysis of the received waveforms to extract the unique features from the spectral correlation function (SCF) of each waveform and use these features for training local convolutional neural networks (CNNs) to detect the jammer attacks and classify waveforms. Moreover, considering a practical distributed TWN where each cluster head (CH) has a partial observation of the TWN with insufficient data samples, the proposed algorithm exploits the distributed learning feature of FL, i.e., global learning aggregation, to detect the existence of jammers and to distinguish the types of received waveforms over the entire TWN. We implement a rigorous TWN simulation using Matlab Toolboxes and our proposed algorithm using TensorFlow Federated (TFF). Numerical results show that the proposed algorithm outperforms the standalone local SCF-CNN algorithm. We further demonstrate that using the SCF feature provides more accuracy than using the In-phase/Quadrature (I/Q) features.