Covariance Matrix Estimation for Broadband Underwater Noise

In this paper, a technique is presented for frequency-smoothing the sample covariance matrix observed at the output of a set of elements or beams so as to obtain a frequency-smoothed estimate of the broadband noise covariance matrix due to underwater noise. The smoothing technique is based on a linear model that is derived through analysis of the delay-domain cross covariance of the ambient noise observed by a pair of beams or omnidirectional hydrophones. A simple and robust least squares fitting algorithm allows for extraction of a frequency-smoothed broadband noise covariance estimate even in the presence of narrowband energy. Simulated passive sonar examples demonstrate the validity and utility of the estimate. Possible applications include adaptive beamforming and measuring the vertical/horizontal directionality of diffuse broadband ambient noise sources such as wind and rain.