A thesis submitted in partial fulfillment of the requirements for the degree of Master of Engineering in Telecommunications

Thesis (M. Eng.) - Asian Institute of Technology, 2022

Direction of Arrival (DOA) Estimation is amongst the pivotal technologies in the field of smart antenna systems. Smart antenna systems are capable of locating and tracking signals by the use of various DOA algorithms. Hence, these systems can dynamically adapt to enhance the reception of the signals in the necessary directions and minimize the effect of the interfering signals. Two-dimensional (2D) DOA estimation provides ample spatial statistics of the signals impinging in the antenna array, and has more practical significance in smart antenna applications for source localization. Since regular planar antenna arrays are unable to meet the requirements for the quick and precise localization of mobile sources, the concept of volumetric antenna arrays has been introduced. However, there have been limited research done on DOA estimation of these volumetric antenna arrays. A novel approach to 2D DOA estimation of a volumetric antenna array using Convolutional Neural Network (CNN) is proposed in this study. The 2D DOA prediction network takes the covariance matrices of the various angle pairs as its input. These matrices are computed from the volumetric extension of the planar array which is preprocessed for better feature extraction. In addition, the signals simulated undergo multipath fading during its generation, to closely resemble a real-world signal. Finally, the CNN outputs the predicted angle pairs for a given number of test samples of angle pair covariance matrices.