Spectrogram based Wi-Fi usage activity classification using deep learning

Abstract

Abstract The article presents and characterizes the ability of a neural network to distinguish between different Wi-Fi activities based on their distinct spectral features. To achieve this objective, we have built a database containing spectrogram images retrieved by over-the-air measurements of signals emitted in 802.11ax Wi-Fi networks deployed in both the 2.4 GHz and 5 GHz frequency bands. The dataset consists of spectrograms labeled by six distinct user activities, namely file download, speed test, video streaming, file upload, video call, and voice over IP call (VoIP). The network training parameters such as learning rate and validation frequency were optimized to enhance model performance. The influence of different activation functions, ReLU, leaky ReLU, eluLayer, and swishLayer from Matlab was also evaluated to achieve neural network fine-tunning. Results demonstrate the effectiveness of the proposed neural network architecture in accurately classifying Wi-Fi usage patterns, achieving test accuracies above 98%.