Analysis of signal detection SNR limits in snapshot-deficient scenarios with colored noise

Abstract

Most source enumeration algorithms assume a white noise background; however, in many underwater applications, the background noise is colored. Various researchers have investigated the problem of source enumeration when the background noise is colored. The standard approach is to apply a whitening filter before the signal enumeration algorithm. Most of these algorithms perform poorly when the number of noise snapshots used to estimate the whitening filter is small. Nadakuditi and Silverstein [IEEE J. Sel. Topics Signal Process., 2010] developed an algorithm for source enumeration using random matrix theory that provides the fundamental SNR limits for snapshot-deficient scenarios with arbitrary noise. Nadakuditi and Silverstein's analysis focuses on the performance when the number of signal plus noise snapshots varies while the number of snapshots used for the whitening filter stays constant. This talk analyzes the performance of this algorithm when number of snapshots available to estimate the whitening filter varies from N to 10 N. Simulations using two different colored noise models for a large vertical linear array located in the deep ocean will be presented.