Improving whitening filter design using broadband snapshots

Abstract

Traditional methods of source enumeration are derived assuming a white noise background and large amounts of data to estimate a sample covariance matrix. Recently source enumeration methods based on random matrix theory (RMT) have shown better performance than traditional methods when the data sets are small. In the presence of colored noise, a common approach is to whiten the data prior to applying the source estimator. The design of a whitening filter typically requires large amounts of training data, which can be difficult to obtain in practice. This paper presents a method to increase the available training data by using snapshots from neighboring frequency bins. Simulations demonstrate the performance improvement for two RMT-based source enumeration algorithms when broadband snapshots are used. The simulation environment models the colored noise due to distant shipping received by vertical arrays operating in the deep ocean.