Abstract
Loud moving interferers challenge adaptive beamformers (ABFs) attempting attenuate these interferers while detecting quiet sources. Specifically, moving interferers create a dynamic tension in choosing the best window length for estimating the sample covariance matrix (SCM) from received snapshots. If the window is too short, the ABF does not accurately steer its nulls toward the interferers, diminishing its effectiveness in attenuating the interferers. If the window is too long, the interferers transit multiple beamwidths, splitting their power across multiple eigenvectors, also diminishing the ABF's effectiveness (Baggeroer and Cox, 1999). Previously proposed approaches to address this challenge include Multirate Adaptive Beamforming (Cox, 2000) and Null-Broadening (Song et al., 2003). This talk proposes a universal adaptive beamformer (UABF) exploiting online learning techniques to adapt the length of the SCM averaging window. The UABF's array weights blend the array weights of several competing ABFs. The recent performance of each fixed-window ABF determines its contribution to the UABF array weights. The UABF's performance provably converges to rival the best fixed-length window ABF for every finite power input sequence, without assuming a specific probability distribution. In nonstationary environments where interferers change speed, the UABF may outperform all of the competing fixed-window ABFs. [Supported by ONR 321US.]