Temporal-Spectral Data Mining in Anomaly Detection for Spectrum Monitoring

As the wireless services developed rapidly in the recent years, a diversity of wireless services emerge such that radio environment becomes more and more complicated. Radio Spectrum security is now attached with great importance. Real time spectrum anomalies detection is vital for increasing demand on security to ensure that wireless services function on the rails. Malicious radio events, such as illegal channel occupation, happened frequently in the recent years, which result in severe interference to the normal radio spectrum usage. There were anomalies detection approaches in different areas proposed to conquer such malicious events. However, those malicious events usually happen in a short interval, this increases the demand on instantaneous responds for real-time events, and the complexity of previous approaches makes them insufficient to handle the real time task. In this paper, a new approach for anomalies detection in spectrum monitoring is proposed. Distinct from previous anomalies detection methods, both temporal and spectral information are taken into account and utilized to find out the potential anomalies. Meanwhile, an adaptive learning ability is proposed along to respond to the real-time change of radio environment. To analyze spectrum measurement data with high dimension, Mahalanobis distance is applied to disclose potential anomalies according to the historical pattern of radio spectrum. Methodology analysis and real case study have been performed to validate the detection effectiveness in practice.