Detection on Abnormal Usage of Spectrum by Electromagnetic Data Mining

With the rapid development of radio services and monitoring facilities, the application of spectrum monitoring steps into the big data era. As a limited resource, the electromagnetic spectrum needs to be authorized for use. According to the spatiotemporal periodic characteristic of radio frequency usage, the new method is put forward for quickly detecting abnormal spectrum usage and abnormal electromagnetic targets (appearing in abnormal time or space). To deal with frequency spectrum data, the piecewise modeling algorithm based on Mahalanobis distance is proposed to find the abnormities in them, which can effectively detect the differences between two spectrums in real time. To deal with the electromagnetic target positioning data, the outlier detection and non-outlier classification algorithm based on Euclidean distance is proposed. Compared with the spatiotemporal data of historical target points, the unknown target points which appear in abnormal time or space can be found in real time. Their validity and applicability are verified by the real monitoring data collected by CS-S05F. It is of great significance for the detection of targets which use spectrum resources illegally.