The potential performance gain in using spectral information in passive detection/localization of wideband sources

We address the problem of detecting the presence of wideband point sources and/or estimating their location using data collected by an array of sensors. This passive localization problem is typical to applications where the radiated signals do not carry information and are of no interest. However, knowledge of their waveshape, or their cross spectral matrix, is essential for applying optimal/sub-optimal detection/localization algorithms. In this paper, we present results of a study in which we compare the optimal procedures for detection/localization and their performance to suboptimal procedures that use no spectral information or only partial spectral information. We show that if the number of sensors in the array is larger than the number of sources to be localized (or detected), then the performance gain of the optimal procedures is potentially significant mainly in conditions of small separation and/or high correlation between sources. The performance gain is also a function of the spectrum: it is potentially higher if the sources are known to be correlated and if their spectra is not smooth. If, however, the number of sources is smaller than the number of sensors, then the role of the spectral information in improving detection/localization performance becomes major: only the use of spectral prior enables reliable localization of more sources than sensors.