Algorithms for change detection with unknown number of affected sensors

Abstract

In this paper, we consider change detection in a sensor network where an unknown subset of sensor nodes are affected by the change. We consider two models for the channel between the sensors and the fusion center: (1) parallel non-interfering channels, and (2) physical layer fusion. For both these models, we propose quantized transmission schemes for the sensors and corresponding fusion rules at the fusion center. The proposed fusion rules are based on an adaptive version of CUSUM. The detection delay performance of the proposed schemes is studied as a function of the number of affected sensors for a given false alarm constraint. Simulation results show that the proposed schemes can work well for a wide range of the fraction of affected sensors.