A novel spectrum occupancy anomaly detection method based on time series analysis theory

Efficient spectrum management and dynamic spectrum access networks heavily rely on accurate statistics of spectrum utilization and temporal behaviour modelling of spectrum occupancy. In this paper, we propose a novel method for spectrum occupancy time-varying characteristics analysis, which includes modelling and anomaly detection of dynamic spectrum occupancy data. First, through the procedure of preprocessing and statistical test for measured spectrum data, we demonstrate the conditional heteroskedasticity existed in spectrum occupancy time-varying series. Furthermore, we present an EGARCH (exponential generalized auto regressive conditional heteroskedasticity) model to fit the variance of spectrum occupancy. Finally, we present an iteration algorithm to detect spectrum occupancy anomaly, and the empirical results show that the proposed method can identify the outliers of spectrum occupancy series without the need for a prior knowledge.