Detection and Classification of Smart Jamming in Wi-Fi Networks Using Machine Learning

Smart adversaries can exploit the publicly known frame structure of OFDM-based Wi-Fi protocols to disrupt communications by strategically jamming specific time samples or specific subcarriers. Such attacks are very difficult to detect by traditional techniques like spectral analysis and signal strength indicators. Machine learning (ML) based methods have been proposed to tackle this problem. However, existing ML methods are computationally intensive and perform well only at low signal-to-jamming power ratios (SJRs). In this paper, we propose a computationally efficient deep convolutional neural network (DCNN) consisting of only four convolution layers to detect and classify several smart jamming attacks in Wi-Fi networks. To deal with the time-frequency selectivity of smart jamming, we apply the continuous wavelet transform (CWT) to partially overlapped segments of the received I/Q samples to extract features. The scalogram of the CWT is used as input to the DCNN. We focus on three smart jamming attacks: preamble jamming, pilot jamming, and interleaving jamming. These attacks share similar characteristics, making their differentiation particularly challenging. Our proposed classifier achieves high accuracy in detecting and classifying these jamming attacks across a range of SJRs, from -6 dB to 15 dB, with an overall classification accuracy of 98%. Even at high SJR levels, the accuracy remains high at around 90%. We also train the classifier to be robust against partial preamble jamming and pilot jamming, The resulting classification accuracy is over 90% at SJRs up to 12 dB. Additionally, we compare our classifier with one that uses the spectrogram (short-time Fourier transform) as input to the DCNN, and demonstrate the superior performance of the proposed scalogram-based classifier, particularly in the high SJR regime.