صادق حقيقي

Direction of arrival (DoA) of signals at an antenna array is a critical parameter in array signal processing. DoA plays a significant role in various applications such as wireless communications, the Internet of Things (IoT), telecommunication networks, an‎d autonomous vehicles. DoA is widely used in wireless systems an‎d networks for source localization an‎d tracking. In 5G an‎d next-generation wireless networks, DoA estimation is crucial for interference mitigation, enhancing the signal-to-interference-plus-noise ratio (SINR), an‎d improving network capacity. Classical DoA estimation methods typically rely on techniques such as beamforming, subspace-based algorithms, an‎d other parametric an‎d non-parametric approaches. While these methods have demonstrated their effectiveness in specific scenarios, they often suffer from limitations such as oversimplifying assumptions, dependency on precise statistical knowledge, an‎d sensitivity to noise an‎d multipath. These dependencies significantly hinder their performance in real-world conditions. In recent years, the growing use of distributed, wideban‎d, an‎d massive multi-antenna systems has driven interest in low-resolution analog-to-digital converters (ADCs), particularly one-bit ADCs, due to their hardware simplicity, low power consumption, an‎d reduced cost. Such converters are especially suitable for large-scale antenna arrays an‎d resource-constrained systems. However, using one-bit ADCs results in significant loss of statistical information an‎d severely degraded signal resolution, posing major challenges for traditional DoA estimation algorithms. This highlights the need for developing new approaches that are compatible with low-resolution quantized data. In this context, machine learning techniques, particularly deep learning, have emerged as promising tools for compensating the loss of precision caused by one-bit quantization. Their ability to learn complex patterns an‎d nonlinear relationships from noisy an‎d imprecise data makes them suitable for such tasks. Although these methods are often considered "black boxes" an‎d lack interpretability, integrating them with model-based approaches can lead to hybrid algorithms that not only maintain accuracy an‎d robustness but also offer interpretability. This thesis investigates such a hybrid approach an‎d proposes an accurate model for DoA estimation in systems employing one-bit ADCs.

زاويه ورود , پردازش سيگنال آرايه‌اي , يادگيري عميق , مبدل آنالوگ به ديجيتال يك‌بيتي , يادگيري تركيبي

Direction of arrival , array signal processing , deep learning , One-bit ADCs , hybrid learning

Sadegh Haghighi

Dr. Shahrokh Farahmand