GSM-Based Mobile Handset Identification Using RF Fingerprints and Deep Learning on the Edge Computing Devices

The global system for mobile communications (GSM) is a 2G technology accepted by almost every communication device. However, the received GSM signals from different mobiles differ irrespective of their manufacturer or category. This article proposes a GSM-based mobile handset identification approach using the continuous wavelet transform (CWT) and deep learning. The dataset consists of radio frequency (RF) fingerprints of GSM-based mobile handset signals from 12 different mobile gadgets. In each burst of the signal, the information field is removed. This reduces the redundancy across various gadgets. Here, the dataset includes multiple devices from the same manufacturer to ensure completeness and accuracy. Then, these signals are preprocessed using CWT to generate the spectrograms, which are then fed to a deep convolutional neural network (CNN) model for accurate classification. We show with the results that the proposed model achieves a test accuracy of 94.16%, surpassing all benchmark models considered in this work. We also deploy all these models on edge computing devices such as the Raspberry Pi 5, different central processing units (CPUs), and graphical processing units (GPUs) to evaluate the inference time.