Executive Summary : | Underwater Acoustic (UWA) Source Localization is an ill posed problem in comparison to plane-wave or aerial acoustic based Direction of Arrival (DOA) estimation. This problem is often solved by deploying matched-field processing (MFP) techniques. The well known existing non-parametric approaches, such as Bartlett and Capon processors, and subspace-based methods such as MUSIC have been studied extensively in the context of UWA source localization. Beamforming suffers from Rayleigh resolution limits. In contrast, MUSIC and MVDR, which are super-resolution techniques, work well at moderate to high SNR (Signal to Noise Ratio), provided there is a low correlation between sources and a sufficient number of snapshots. Maximum Likelihood (ML) Estimator, highly computation intensive, is asymptotically efficient particularly in an non-Gaussian noise environment. Accurate Sparse Signal recovery (SR), based on convex optimization regime, has been an active research interest in UWA Source Localization. Convex-optimization based SR algorithms have proved to be robust under limited data cases, but they are computationally costly. These techniques also show high-resolution capabilities for few snapshots and work even for coherent sources contrary to Eigenvalue based algorithms. The main objective is to apply a robust and weighted SR technique i.e. l1 -SVD and Weighted l1 -SVD for source localization in UWA. Classic l1 based SR techniques estimate the sparse signal with high computation time complexity. This complexity further increases with the number of snapshots L. To alleviate this, l1 -SVD takes singular vectors (fixed with number of sources) instead of snapshots thus reducing computation time. The l1 -SVD algorithm is robust to SNR and snapshot variations (provided if the SNR is not too low). Further, the sensitivity to the assumed number of sources is also reduced. Therefore, the re-weighted l1 -norm minimization has proved effective to enhance sparsity and bring robustness to noise. The method works by penalizing smaller coefficients with larger weights whose indices are more likely to be outside of the signal support (enforcing sparsity). This method improves not only sparsity but also recovery accuracy to the noisy case. The idea of re-weighted l1 has been expanded to the l1 -SVD case by utilizing the orthogonality between signal and noise subspace to form the objective function of the re-weighted l1 -SVD. Another important aspect of the proposal lies in the selection criterion for choosing an optimal number of sensors for UWA source localization that will yield the best resolution performance. Hence, the aim of this project is:1)To detect and localize underwater targets in noisy environment like rain, wind, mammal noise, ship noise, etc. 2) Ability to localize contiguous targets with fewer sensors in comparison to the state-of-the-art methods in non-Gaussian and unknown noise fields. 3) To test the algorithms with real-time data to achieve TRL of 4. |