ABSTRACT In the deep ocean, where light cannot penetrate and GPS coverage is unavailable, sonar remains the principal modality for underwater perception, navigation, and threat detection. In these acoustically complex environments, traditional sonar signal processing methods face critical limitations characterized by multipath propagation, Doppler shifts, ambient noise, and adversarial stealth. Reverberant littoral zones, low‐observable platforms, and time‐varying interference reduce the effectiveness of classical beamformers, matched filters, and deterministic classifiers. This paper presents a systematic review of recent advances in underwater sonar systems and artificial intelligence (AI)‐driven signal processing for naval and autonomous applications. We trace the evolution from model‐based frameworks to data‐driven architectures, highlighting the growing role of convolutional and recurrent neural networks, deep Kalman filters, transformer‐based classifiers, and multi‐sensor fusion methods. These approaches are assessed in the context of GPS‐denied navigation, constrained bandwidth, and dynamic acoustic conditions. Particular emphasis is placed on AI‐driven motion estimation, where modern models increasingly surpass traditional methods in mitigating inertial drift, enhancing trajectory prediction, and improving operational resilience. This review synthesizes current capabilities and identifies unresolved challenges in model explainability, real‐time adaptability, adversarial resilience, and energy‐aware computation. Beyond summarizing recent developments, the paper offers a forward‐looking perspective on intelligent sonar systems that seamlessly integrate sensing, inference, and decision‐making positioning them as pivotal enablers in the future architecture of autonomous maritime operations.
Das et al. (2025) studied this question.