Time-varying techniques for multisensor signal detection

In source detection and localization, the presence of a common but unknown signal can be detected from several noisy sensor measurements using the generalized coherence (GC) estimate. We propose to improve the performance of the GC estimate for multisensor detection using noise-suppressed signal estimates obtained from time-varying techniques. If one of the sensors has a significantly higher signal-to-noise ratio (SNR) than the other sensors, then it could be preprocessed prior to the GC estimate to improve detection performance for the remaining, lower SNR sensors. We perform this processing by estimating time-varying signals of interest with nonlinear phase functions using two methods: (a) a modified matching pursuit decomposition (MMPD) algorithm whose dictionary is similar, in time-frequency structure, to the signal and (b) an instantaneous frequency (IF) estimation method using highly localized time-frequency representations. The MMPD can yield signal estimates with lower mean square errors than the IF estimation technique but at the expense of higher computational cost and memory requirements. Using simulations, we compare the performance of the GC estimate with the significantly improved performance of the GC estimate that employs the signal estimate from the high SNR sensor. For the two-sensor detection, the estimated signal is also used with a generalized likelihood ratio test statistic to further improve performance.