Adaptive Feature Representation to Improve, Interpret and Accelerate Channel Estimation and Prediction for Shallow Water Acoustic Environments
Author | : Ryan A. McCarthy (PhD) |
Publisher | : |
Total Pages | : 0 |
Release | : 2021 |
ISBN-10 | : OCLC:1373262432 |
ISBN-13 | : |
Rating | : 4/5 ( Downloads) |
Download or read book Adaptive Feature Representation to Improve, Interpret and Accelerate Channel Estimation and Prediction for Shallow Water Acoustic Environments written by Ryan A. McCarthy (PhD) and published by . This book was released on 2021 with total page 0 pages. Available in PDF, EPUB and Kindle. Book excerpt: In my doctoral dissertation I investigate new approaches to real-time channel estimation of underwater acoustic communications that complement existing estimation techniques. Modified sparse optimization algorithms have been used to improve channel estimation with some success. This work aims to improve these algorithms by applying pattern recognition through adaptive signal processing and machine learning to accelerate estimation time. Specifically, it investigates a model-agnostic geometric feature morphology based on braid theory to interpret diverse channel phenomena. The computational goal is to detect, separate and interpret multipath features in the channel delay spread across time, frequency, and varying degrees of channel sparsity. The main contribution of the thesis is development of braids feature representations and related channel tracking and learning algorithms to track salient bands of multipath activity. We develop robust signal processing and braided feature engineering approaches that evolve dynamically to the fluctuating channel multipath activity. To test the hypothesis that braids can track and adapt to diverse activity developing within the channel, simulated shallow water environments created through the well-known BELLHOP model and data from the SPACE08 field experiment are examined. Several simulated shallow water environments are examined with additive white Gaussian noise and varying degrees of activity to evaluate the performance of braiding and machine learning for shallow water acoustic channel estimation and interpretation. Performance is evaluated through visual confirmation and ground truths are provided by BELLHOP's outputs (e.g. eigenrays, arrivals, etc.). Results show that braids can evolve to capture dynamically changing multipath scattering activity in the shallow water acoustic channel. Furthermore, we demonstrate that leveraging braid feature representations with acoustic physics propagation models can successfully predict the number of reflectors in active channel multipath. We also demonstrate the significance of braid manifold representation in improving the computational speed for channel estimation. On average, this technique has improved estimation speed by ~.02 seconds as compared to the existing estimation techniques. These results suggest that braids can be used for useful pattern recognition to bridge the gap between purely statistical data analysis and physics-driven interpretation of the ocean acoustics that create the multipath channel delay spread. Beyond underwater acoustics, these feature learning techniques are broadly applicable to any paradigms where spectral features may evolve and intersect.