Low Complexity Machine Learning Based AoA Estimation for mMIMO DFRC Systems

Abstract

This paper proposes a machine learning (ML) based approach for estimating the target angle in a massive multiple-input multiple-output (mMIMO) dual functional radar and communication (DFRC) scenario with multiple clutters. The approach utilizes an artificial neural network (ANN) based ML technique for angle of arrival (AoA) estimation and compares its performance with the conventionally used multiple signal classification (MUSIC) algorithm in the mMIMO DFRC setup. The study includes a numerical comparison based on two metrics: root mean squared error (RMSE) and time complexity for the estimation. The results demonstrate that the proposed ANN technique achieves significantly lower RMSE and requires less time for AoA estimation. Additionally, the paper examines the impact of signal-to-noise ratio and the number of snapshots in pilot signals on RMSE performance. Furthermore, it introduces an mMIMO channel estimation using the estimated angle, which is compared with proven computational methods such as least square (LS) and minimum mean square error (MMSE) techniques. The numerical results reveal that using the angle estimated by the ANN for MIMO channel estimation results in significantly lower RMSE compared to LS, MMSE, and using the angle estimated by the MUSIC algorithm. © 2023 IEEE.