3D Direction of Arrival Estimation: An Innovative Deep Neural Network Approach
Constantinos M. Mylonakis, Pantelis Velanas, Pavlos I. Lazaridis, Panagiotis G. Sarigiannidis, Sotirios K. Goudos, Zaharias D. Zaharis · 2024
The recent integration of neural networks into the domain of direction of arrival estimation marks a promising fron-tier in the landscape of next-generation wireless communications. Our paper meticulously delves into the architecture of the proposed deep convolutional neural network (DCNN), presenting a novel framework designed to streamline the classification process within the output layer. Operating on correlation matrices created by signals received by a 4 × 4 planar antenna array, our DCNN predicts angles of arrival in 3D space. We assess the model's performance in scenarios involving the simultaneous reception of signals, employing the mean absolute error as a metric to gauge prediction errors in the angle domain. The simulation results affirm the superior performance of the proposed deep learning-based scheme. The model's robustness is rigorously examined across various validation cases, providing conclusive evidence of its potential in real-world applications.