An efficient MD-GRao algorithm for quickest change detection in sensor networks with unknown post-change parameters

Consider a quickest change detection (QCD) problem in sensor networks for monitoring a sudden parameters change in signals with unknown post-change parameters, where the change may be subtle. Traditional approaches, such as the generalized likelihood ratio test-based QCD (GLRT-QCD), are often computationally prohibitive for real-time applications. The Rao test-based QCD (Rao-QCD), though simpler, results in higher false alarm rates and longer delays. To address these limitations, we propose a modified drift-oriented generalized Rao (MD-GRao) algorithm, which strikes a well-balanced tradeoff between computational complexity and detection effectiveness, and can be applied to general cases. For the first time, we analyze the drift property of Rao-QCD and reveal the monotonically increasing nature of its statistic. Based on this insight, we propose a dynamic window update strategy to efficiently estimate unknown parameters and develop a recursive update approach that incorporates a negative drift mechanism, enabling rapid identification of the potential change. Theoretical analysis establishes the asymptotic performance of the proposed algorithm, while comprehensive numerical evaluations of heart rate change detection in a radar-based sensor network demonstrate its superior computational efficiency over traditional methods.