Presentation Information
[8a-PB1-31]Coarse-to-Fine 3D Underwater Localization Using a Single Receiver
〇Yunui Hong1, Myeongseong Kim1, Eunyoung Lim1, Hojun Lee1 (1.Hoseo Univ.)
Keywords:
ocean acoustics,underwater localization,deep learning
This paper proposes a coarse-to-fine 3D underwater localization framework designed for reverberant shallow water environments. By using a single receiver without fixed anchors, we achieve millimeter-level accuracy. In the coarse phase, the receiver performs repetitive discrete dips while advancing along a spiral trajectory to form a virtual receiver array. These signals are processed via a 1D convolutional neural network (Model 1) and causal Transformer (Model 2) to estimate the target's 3D coordinates (mean absolute error = 0.569 m). In the fine phase, a time-of-flight (ToF) multipath mitigation filter and RANSAC-based trilateration are applied to near-field measurements (ranging error ≈ 8 mm) collected within 5 m of this coarse phase estimate to isolate direct-path signals, refining the final 3D positioning error to an average of 30.6 mm. This empirical error approaches the theoretical Cramér-Rao Lower Bound (CRLB = 9.29 mm) by 3.3 times. The framework runs in under 10 msec, presenting a highly practical real-time solution for wearable safety devices.