Deep Predictive Coding Neural Network for RF Anomaly Detection in Wireless Networks

Abstract

Intrusion detection has become one of the most critical tasks in a wireless network to prevent service outages that can take long to fix. The sheer variety of anomalous events necessitates adopting cognitive anomaly detection methods instead of the traditional signature-based detection techniques. This paper proposes an anomaly detection methodology for wireless systems that is based on monitoring and analyzing radio frequency (RF) spectrum activities. Our detection technique leverages an existing solution for the video prediction problem, and uses it on image sequences generated from monitoring the wireless spectrum. The deep predictive coding network is trained with images corresponding to the normal behavior of the system, and whenever there is an anomaly, its detection is triggered by the deviation between the actual and predicted behavior. For our analysis, we use the images generated from the time-frequency spectrograms and spectral correlation functions of the received RF signal. We test our technique on a dataset which contains anomalies such as jamming, chirping of transmitters, spectrum hijacking, and node failure, and evaluate its performance using standard classifier metrics: detection ratio, and false alarm rate. Simulation results demonstrate that the proposed methodology effectively detects many unforeseen anomalous events in real time. We discuss the applications, which encompass industrial IoT, autonomous vehicle control and mission-critical communications services.