Date of Award

3-2-2021

Publication Type

Master Thesis

Degree Name

M.A.Sc.

Department

Electrical and Computer Engineering

First Advisor

Mehrdad M.S. Saif

Second Advisor

Shahpour S.A. Alirezaee

Rights

info:eu-repo/semantics/openAccess

Abstract

Localization is a critical step in any navigation system. Through localization, the vehicle can estimate its position in the surrounding environment and plan how to reach its goal without any collision. This thesis focuses on underwater source localization, using sound signals for position estimation. We propose a novel underwater localization method based on machine learning techniques in which source position is directly estimated from collected acoustic data. The position of the sound source is estimated by training Random Forest (RF), Support Vector Machine (SVM), Feedforward Neural Network (FNN), and Convolutional Neural Network (CNN). To train these data-driven methods, data are collected inside a confined test tank with dimensions of 6m x 4.5m x 1.7m. The transmission unit, which includes Xilinx LX45 FPGA and transducer, generates acoustic signal. The receiver unit collects and prepares propagated sound signals and transmit them to a computer. It consists of 4 hydrophones, Red Pitay analog front-end board, and NI 9234 data acquisition board. We used MATLAB 2018 to extract pitch, Mel-Frequency Cepstrum Coefficients (MFCC), and spectrogram from the sound signals. These features are used by MATLAB Toolboxes to train RF, SVM, FNN, and CNN. Experimental results show that CNN archives 4% of Mean Absolute Percentage Error (MAPE) in the test tank. The finding of this research can pave the way for Autonomous Underwater Vehicle (AUV) and Remotely Operated Vehicle (ROV) navigation in underwater open spaces.

Share

COinS