Ambiguity Resolution in Direction of Arrival Estimation using Mixed Integer Optimization and Deep Learning

In this paper we present two novel approaches to unambiguously estimate the direction of arrival (DOA) of an RF source by an array of receive antennas whose positions can take-on any arbitrary geometry. The first approach employs a simple constrained integer optimization, while the second approach employs deep learning. In both approaches the impact of imperfect array calibration on the performance of DOA estimation is quantified. We demonstrate in Monte Carlo simulations that both approaches are capable of achieving super-resolution performance under imperfect array calibration conditions. It was found that the constrained integer optimization outperforms deep learning when one has an accurate physics model of the receiver imperfections, however deep learning was more robust to significant calibration errors.