Direct Localization in Underwater Acoustics Via Convolutional Neural Networks: A Data-Driven Approach

Towards Robust Data-Driven Underwater Acoustic Localization: A Deep CNN Solution with Performance Guarantees for Model Mismatch

“…In a recent work [19], an architecture based on a deep convolutional neural network (CNN) has been proposed as a DD-DLOC solution for the UWA localization problem. This architecture is trained in two phases.…”

“…To this end, in this work we consider a simulated UWA environment as our virtual testbed, and adopt the architecture proposed in [19], to which we introduce a two main modifications. First, we now directly concatenate the outputs of the three sub-models, such that the modified architecture already provides a working solution at the end of the first phase of training (mentioned above).…”

“…Given a source-receiver pair, the relevant rays propagating from the source to the receiver can be described in a plane perpendicular to an ideal flat bottom, in which the source and the receiver lie. As an example, the complexity of the method [22] in its extended form [19] is Ω(N L 2 R 2 ) even before considering that all the LR time-delays of all the rays to all the receivers must be computed as well. 4 It is therefore natural to adopt a learning-based solution approach, that will provide a method to localize the source while: (i) judiciously using those environmental characteristics that can be learned during the training phase; (ii) being robust to reasonably moderate deviations from the learned model; and (iii) efficiently approximating a reliable estimator p = f (x1, .…”

“…i=1 of size J is available. Such a dataset can be obtained in several ways (see [19] for details). 4 For environments such as the on in Fig.…”

“…Due to the success of deep neural networks (DNN) in other domains, data-driven (DD) methods have been recently considered more extensively as potentially viable solutions to the UWA localization problem [8][9][10][11][12][13][14][15][16]. Specifically, a DD variant of the statistically superior direct localization (DLOC) approach [17,18] has been proposed in [19]. Yet, while these DD methods are in principle exceptionally powerful, they can be difficult to analyze, and their robustness is generally not guaranteed.…”