Learning-Based One-Bit Direction of Arrival Estimation with Mutual Coupling

This study investigates a novel, data-driven methodology for Direction of Arrival (DOA) estimation using neural networks, designed for resource-constrained hardware environments and under mutual coupling, a common impairment in real-world applications. Focusing on one-bit quantization, this approach addresses the critical need for reduced system complexity and power consumption, offering significant advantages in power-efficient, cost-effective, and hardware-limited applications. One-bit DOA estimation is particularly relevant to wireless communication systems, including massive multiple-input multiple-output (MIMO) and satellite communications, as well as automotive radar, underwater acoustics and sonar, and biomedical applications. This work examines several learning-based structures, including a fully connected (FC) network, a convolutional neural network (CNN), an encoder/decoder-based architecture, and introduces a proposed neural network architecture specifically tailored for this task. The performance of this learned DOA estimation technique is evaluated against established learning-based and conventional algorithms, demonstrating its improved performance.