An experimental comparison of universal beamformers

The challenge of rejecting interference and noise to reveal a signal of interest is common in array processing. Buck and Singer introduced the Universal Dominant Mode Rejection (UDMR) beamformer, which calculates filter weights by blending multiple dominant mode rejection (DMR) beamformers with different subspace dimensions [IEEE SAM, 2018]. Tucker and Wage's Performance Weighted Blended (PWB) beamformer adopts a similar strategy, combining classical estimators with different tapers [IEEE SAM, 2024]. While the UDMR beamformer requires a computationally expensive eigen-decomposition, the PWB beamformer can be implemented more efficiently using fast Fourier transforms. UDMR is doubly adaptive, adjusting its blend weights to combine adaptive DMR beamformers. In contrast, PWB is singly adaptive, blending classical fixed-taper beamformers. Since UDMR blends adaptive beamformers, it can provide better interference suppression. However, PWB's computational efficiency may be desirable for resource-limited applications. This talk presents an experimental comparison of the two algorithms using the PhilSea10 dataset [Worcester et al., JASA (2013)], benchmarking both against fixed rank DMR beamformers. [Work supported by ONR.]