Dimension Increased Random Matrix Method for Anomaly Detection in Wireless Networks

The rapidly growing spatio-temporal correlated data in wireless networks provide a natural platform for Integrated Sensing, Computation and Communication (ISCC). Random Matrix Theory (RMT) is an effective tool to analyze anomaly network behaviors in multi-dimensional datasets. But real-time anomaly detection methods based on RMT spectral analyses may fail to analyze low-dimensional datasets such as Internet of Things (IoT), thus yield unsatisfactory detection accuracies. In this paper, we propose a dimension increasing RMT (DI-RMT) anomaly detection method to analyze low-dimensional random matrices. A random matrix is formulated using the signal plus noise model, with preserved key performance indicators as the augmented matrix and the status data as the rest part of the matrix. On the basis of the tensor product, we put forward a dimension increasing method, which can detect and localize anomalies in real time, and is robust enough to cope with random disturbances and measurement errors. A case study with real-world low-dimensional datasets indicates that our proposed method can achieve a 4.45 times higher accuracy than the traditional RMT approach, which validates the feasibility to apply RMT to the anomaly detection of low-dimensional datasets.