Hybrid Model-Based RF Fingerprinting and Spiking Neural Networks for IoT Device Classification

Abstract

Radio frequency fingerprinting identification (RFFI) leverages the unique features of communication transmitter signals to classify Internet of Things (IoT) devices, enabling individual recognition through waveform analysis. Traditional RFFI methods face challenges in extracting nonlinear features, which machine learning (ML) techniques help overcome by providing advanced wave characteristic analysis. This study introduces RFFI-SCNN, a hybrid model integrating RFFI with a spiking conventional neural network (SCNN) to enhance IoT device authentication within networks. The model operates in two phases: signal processing, where wave data are collected and preprocessed, and SCNN-based classification, where features are extracted and devices are authenticated. The proposed model's performance is evaluated against three ML-based models—1SNN, 1CNN and DCNN—based on accuracy, execution time and memory usage. Experimental results, conducted using a publicly available dataset from the Institute for the Wireless Internet of Things at Northeastern University, indicate that RFFI-SCNN achieves superior accuracy in classifying communication devices compared to 1CNN and 1SNN while also requiring less memory and shorter execution time than DCNN and 1CNN. These findings highlight the effectiveness of RFFI-SCNN in secure and efficient IoT device identification.