Loading...

DOA Estimation in Communication Systems Using Deep Neural Networks

Alikhani, Morteza | 2023

157 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 56641 (05)
  4. University: Sharif University of Technology
  5. Department: Electrical Engineering
  6. Advisor(s): Behrouzi, Hamid; Karbasi, Mohammad
  7. Abstract:
  8. The purpose of Direction-of-arrival(DOA) estimation in MIMO communication networks is to find the direction of signals that are mixed together and mixed with noise by using an array of multiple receiving antennas. DOA estimation is an important issue in array signal processing. The traditional methods for DOA estimation try to solve this problem with mathematical algorithms. For example, some of these methods are based on parameter estimation; such as DOA estimation methods using maximum likelihood estimation methods. Some of other traditional methods are based on subspace decomposition; Like the famous MUSIC method. The traditional methods of DOA estimation usually have a drawback, and that is, in non-ideal and real conditions such as the existence of various imperfections, the existence of signal coherence, the reduction of the number of snapshots and the reduction of SNR, their performance drops drastically and they cannot meet the needs. Today, with the advancement of machine learning and deep learning and their application in different fields, the discussion of using machine learning and deep learning in the field of DOA estimation has also flourished. Most of the researches that have used deep learning in DOA estimation have used networks based on CNN and DNN, and the gap of using recurrent neural networks(RNNs) and sequence-to-sequence networks in the field of DOA estimation is felt. In this thesis, we intend to solve the DOA estimation problem in MIMO communication networks using deep learning methods, and more precisely, recurrent neural networks(RNNs) as sequence classifiers, as well as sequence-to-sequence networks, and evaluate these methods. In this thesis, some selected models are introduced that according to the results and simulations of this thesis; These models are good. They have high accuracy and in addition, in non-ideal conditions, such as the existence of various imperfections, the existence of signal coherence, the reduction of the number of snapshots and the reduction of SNR, their performance loss is small, and also they perform well both in terms of speed and execution time. In general, the proposed models are better than other methods that we use in this thesis for comparison
  9. Keywords:
  10. Direction of Arrival (DOA)Estimation ; Uniform Linear Array (ULA) ; Deep Learning ; Recurrent Neural Networks ; Sequence-to-Sequence Network ; Phased Array Antenna ; Maximum Likelihood Estimation

 Digital Object List

 Bookmark

...see more