This is the authors' abstract. We don't add key points for this paper.
Adaptive array processing is a key enabler of interference-resilient underwater acoustic (UWA) links, yet the noise seen by UWA receivers is spatially anisotropic, impulsive, and non-stationary, conditions under which low-complexity stochastic-gradient beamformers are poorly characterized. This paper analyzes the Griffiths-type least-mean-squares beamformer with diagonal loading and a power-normalized step size (LMS-DL) on a ten-element half-wavelength line array, and benchmarks it against sample-matrix-inversion (SMI-DL), normalized LMS (NLMS-DL), and exponentially weighted recursive-least-squares (EW-RLS-DL) beamformers. A closed-form approximation of the steady-state output signal-to-interference-plus-noise ratio (SINR) of LMS-DL is derived. It shows that the misadjustment is driven by the total output power, so the SINR saturates at 10log₁₀(2/γ) dB for step-size parameter γ; Monte Carlo simulations show the prediction is conservative and accurate to within 0.75 dB at SNR ≤ 10 dB. For a 10 dB signal and interferers at 20 and 40 dB, LMS-DL settles 3.3 dB below the optimum after about 10⁴ snapshots, whereas SMI-DL comes within 1 dB in 681 snapshots. In Cron–Sherman surface-generated noise the optimum gains 7.4 dB over white noise; SMI-DL recovers 6.4 dB of this gain but LMS-DL only 2.5 dB. Symmetric α-stable noise with α = 1.2 costs LMS-DL 5.6 dB, a loss that per-snapshot normalization (NLMS-DL) reduces to 0.3 dB. LMS-DL tracks a moving 40 dB interferer (worst-case loss 4.1 dB) but not a moving 20 dB interferer (13.3 dB). A link budget with Thorp absorption and Wenz noise shows that converged LMS-DL extends the QPSK range at a bit-error rate (BER) of 10⁻³ from 3.4 km to 15.2 km (optimum 16.0 km), but leaves an error floor of 2–5×10⁻⁵ at short range. The analysis yields design rules for the step size and for SNR-dependent loading in UWA array receivers.
Answers are grounded in this paper and cite their sources.
No takes yet. Share an insight, caveat, or question.
Sehrish (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: