Signal processing for underwater real time 3D acoustical imaging

Abstract

Underwater 3D acoustical imaging is a cutting-edge technology used in ocean exploration newlineand navigation, enabling explorers, operators, or researchers to capture the newlinelocation, orientation, shape, size, reflectivity, and shadowing effects of underwater objects newlinewithout requiring motion or position adjustments. This technology has a variety of newlineapplications, including real-time tasks such as obstacle avoidance and high-resolution newlineapplications like bathymetric mapping. A trade-off exists between image quality and newlinecomputational speed, with real-time uses prioritizing speed over image detail, while newlinehigh-resolution applications can tolerate longer processing time. This thesis aims newlineto develop signal processing techniques for real-time 3D imaging sonar systems to newlinesupport autonomous underwater vehicles (AUVs) in navigation and environmental newlinesensing in deep-sea environments. The requirement for large arrays with thousands newlineof elements and complex beamforming algorithms presents a significant challenge for newlineachieving real-time performance. To enable real time 3D acoustical imaging, a fast newlinebeamforming algorithm is needed in place of conventional delay and sum (DAS) beamforming. newlineAdditionally, replacing a uniform planar array (UPA) with a sparse planar newlinearray can reduce both hardware cost and computational complexity while maintaining newlineimage resolution. While most fast beamforming algorithms in the literature accelerate newlineDAS without compromising image quality, they typically operate in the frequency newlinedomain. Since frequency domain methods are computationally complex for wideband newlinesignal processing, time domain beamforming is preferred. However, the complexity newlineof time domain DAS arises from the need for interpolation filters. By using efficient newlinesoftware-based interpolation, such as linear or spline interpolation, accurate sensor data newlineestimation at high sampling rates becomes feasible, facilitating real-time processing.