Low-Complexity Adaptive Sonar Imaging

We have studied the low-complexity adaptive (LCA) beamformer in active sonar imaging. LCA can be viewed as either a simplification of the minimum variance distortionless response (MVDR) beamformer, or as an adaptive extension to the delay-and-sum (DAS) beamformer. While both LCA and MVDR attempt to minimize the power of noise and interference in the image, MVDR achieves this by computing optimal array weights from the spatial statistics of the wavefield, while LCA selects the best performing weights out of a predefined set. To build confidence in the LCA method, we show that a robust MVDR implementation typically creates weight sets with shapes spanning between a rectangular and Hamming window function. We let LCA select from a set of Kaiser windows with responses in this span, and add some steered variations of each. We limit the steering to roughly half the -3-dB width of the window's amplitude response. Using experimental data from the Kongsberg Maritime HISAS1030 sonar we find that LCA and MVDR produce nearly identical images of large scenes, both being superior to DAS. On point targets LCA is able to double the resolution compared to DAS, or provide half that of MVDR. This performance is achieved with a total of six windows: the rectangular window and the Kaiser window with β = 5, in an unsteered version, and versions that are left and right steered to the steering limit. Slightly smoother images are produced if the window count is increased to 15, but past this we observe minimal difference. Finally, we show that LCA works just as well if Kaiser windows are substituted with trigonometric ones. All our observations and experiences point to LCA being very easy to understand and manage. It simply works, and is surprisingly insensitive to the exact type of window function, steering amount, or number of windows. It can be efficiently implemented on parallel hardware, and handles any scene without the need for parameter adjustments.