Anomaly Detection of Wireless Relays Based on Markov Models Through the Wald-Wolfowitz Runs Test

Abstract

This letter presents a new anomaly detection method adopting the Wald–Wolfowitz runs test for abnormal wireless relays with intermittent packet loss modeled as Markov chains. The proposed method identifies anomalies by leveraging the characteristics of runs in time series of packet loss behaviors without prior knowledge. An asymptotic analysis of the detection performance shows that the false alarm probability rapidly converges to a sufficiently small value, and the missed detection probability decays to zero exponentially. According to theoretical analysis, this method can discern arbitrary Markov-based intermittently abnormal relays from normal ones compared to the general approach based on the overall proportion of dropped packets. Numerical results verify the analysis and corroborate its superiority of wide detection range and short detection time.