AI-Assisted DoA Estimation for MIMO Radar Applications

Direction-of-arrival (DOA) estimation is an essential technique for precise target detection and localization in autonomous driving and surveillance applications. However, in sparse multiple-input multiple-output (MIMO) radar systems, using fewer antennas reduces accuracy while lowering hardware costs. This study explore the use of machine learning (ML) based interpolation as an AI-powered DOA estimation technique in MIMO sparse arrays, which are made up of two TX antennas and 60 RX antennas in the full array configuration, forming a 20 RX sparse array model. MUltiple Signal Classification (MUSIC) processes signals reconstructed using Natural Neighbors, Random Forest, K-Nearest Neighbors (KNN) and Polynomial interpolation algorithms. Performance is evaluated across 5 scenarios, with 1 to <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{5}$</tex> targets under signal-to-noise ratios (SNR) ranging from −20 <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$d B$</tex> to <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$20 d B$</tex>. Performance measurements consider noise effects alongside target position separations and computing power usage. The study confirms that AI interpolation techniques optimize performance because KNN demonstrates the lowest average MSE of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{0. 0 1 2}$</tex> at <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$10 d B$</tex> while the sparse array maintains an MSE of 1287.96. The MSE performance of Natural Neighbor reaches 0.03 at <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$20 d B$</tex>. We performed additional tests that indicated better resolution capabilities for nearby targets and evaluated computational performance where Random Forest required the most processing time. Research findings demonstrate that AIassisted interpolation through KNN achieves most accurate results for precise DOA estimation within sparse MIMO radar systems.