Distributed detection with a minimalistic signal model: A framework for exploiting correlated sensing

We propose and investigate a minimalistic model for distributed detection using a sensor network. The signal of interest is a priori unknown. When a signal is present, sensors receive scaled, delayed and noisy versions of it, and signal presence is decided solely based on the correlation between sensor observations. We obtain encouraging performance results for both centralized and distributed detection, and subsequent centralized signal estimation. We observe that temporal alignment of sensor observations prior to combining is the bottleneck that determines the SNR threshold after which these methods work well. For ideal temporal alignment, detection performance improves exponentially with the number of sensors.