Anomaly Detection and Array Diagnosis in Wireless Networks with Multiple Antennas: Framework, Challenges and Tools

Abstract

Anomaly detection and array diagnosis in wireless networks are both important technologies and have many applications ranging from discovering malicious traffic and identifying abnormal nodes, to detecting faulty antennas and so on. In general, anomaly detection mainly depends on relational data, which denotes the links between nodes of the networks, to decide whether abnormal networks caused by intentional attack or array failure are embedded in large wireless networks. Additionally, the typical scheme of array diagnosis is to measure signals radiating from the array antennas under test to detect the faulty elements by using a centralized method. However, in largescale wireless networks, a centralized strategy results in a communication bottleneck because of transmitting all signals to a center node. Moreover, since faulty elements are only a tiny proportion for the whole networks, the method that all antennas are under test is unnecessary and also causes huge computational complexity to identify the failure of elements. Aiming to mitigate these problems, this article provides a novel framework to monitor networks and detect faulty antennas by fusing relational data and measured signals. In this article, we first review the algorithms related to anomaly detection and survey the array diagnosis problem. In particular, we discuss the relationship between anomaly detection and array diagnosis in the new framework and highlight the importance of data fusion. Finally, the main challenges are presented and mathematical tools are introduced to solve the corresponding problems.