Adaptive focusing for wideband beamforming in multipath environments.

This paper addresses achieving the high time-bandwidth product necessary for low signal-to-noise ratio (SNR) target detection and localization in complex multipath environments. Time bandwidth product is often limited by dynamic environments and platform maneuvers. This paper introduces data-driven wideband focusing methods for passive sonar that optimize parameterized unitary matrices to align signal subspaces across the frequency band without relying on wave propagation models which are subject to mismatch in complex multipath environments. The methods minimize the log-determinant of the wideband covariance, a measure indicative of matrix rank, ensuring the coherence of wideband data and preserving SNR. We propose two approaches: a fully adaptive method with parameters scaling directly with the number of frequency bins, and a partially adaptive method that shares parameters across frequencies to improve robustness to noise. Simulations are conducted in a shallow-water waveguide scenario to demonstrate the flexibility of data-driven focusing over traditional model-based approaches. Results from the SWellEx-96 S59 event validate our methods, showing improvement in tonal target detection and localization in the presence of strong wideband interference.