Low-rank 2D DoA Estimation and Adaptive Tracking for Reconfigurable Intelligent Surface Enabled Vehicular Networks Major in Information and Communication Engineering Farheen Naaz = 재구성 가능한 지능형 표면 기반 차량 네트워크를 위한 저랭크 2차원 도래각 추정 및 적응 추적

다국어 초록 (Multilingual Abstract)

Reconfigurable intelligent surfaces have emerged as a promising technology for future vehicular networks by enabling controllable wireless propagation and providing large spatial apertures for sensing applications. In intelligent transportation systems, accurate two-dimensional direction-of-arrival estima- tion and tracking are essential for traffic monitoring, cooperative perception, collision avoidance, and autonomous driving support. However, practical RIS- enabled vehicular sensing systems may operate under sparse and incomplete observations because full access to all RIS element responses can require high hardware complexity, large numbers of RF chains, increased power consump- tion, and significant acquisition overhead. These sparse observations degrade spatial covariance estimation and reduce the reliability of conventional DoA estimation and tracking methods. This thesis presents a low-rank 2D DoA estimation and adaptive track- ing framework for RIS-enabled vehicular networks under sparse observation conditions. The proposed framework exploits the low-rank structure of the RIS response matrix generated by a limited number of dominant vehicle tar- gets. A robust low-rank regularization stage is applied to the incomplete RIS measurement matrix to preserve the dominant spatial subspace and sta- bilize covariance estimation. To improve robustness against impulsive and non-Gaussian disturbances, a Huber-loss-based ADMM formulation is used. The regularized covariance matrix is then processed using a beamspace-based 2D DoA estimator to extract instantaneous azimuth and elevation angles of moving vehicles. To improve temporal tracking performance, the estimated angular mea- surements are further processed using a variable-forgetting and noise-adaptive recursive least squares tracker. The proposed VFN–RLS tracker adaptively ad- justs its forgetting factor and process-noise level based on innovation statistics and motion-aware information. This enables stable tracking during smooth vehicle motion and improved responsiveness during abrupt maneuvers such as U-turns. Simulation results demonstrate that the proposed framework improves 2D DoA estimation and tracking accuracy under sparse RIS observations, varying signal-to-noise ratios, impulsive-noise conditions, and dynamic vehicle trajec- tories. The ablation study confirms that low-rank regularization significantly improves covariance stability, while adaptive tracking further enhances tem- poral robustness. Overall, the thesis shows that combining low-rank spatial regularization with adaptive temporal tracking provides an effective solution for RIS-enabled vehicular sensing when full RIS observation is unavailable.

목차 (Table of Contents)

• 1 Introduction 1
• 1.1 Contributions 5
• 2 Literature Review and Research Context 7
• 2.1 Integrated Sensing and Communication for Vehicular Networks 7
• 2.2 Reconfigurable Intelligent Surfaces 9
• 2.3 Direction-of-Arrival Estimation 10
• 2.4 Sparse RIS Observation and Switching Constraints 11
• 2.5 Robustness under Non-Ideal Noise 12
• 2.6 Computational Practicality 13
• 2.7 Target Association and Dynamic Vehicle Motion 14
• 2.8 Position of This Thesis 15
• 2.9 Low-Rank Recovery and Robust Matrix Completion 16
• 2.10 Adaptive Tracking of Dynamic DoA States 17
• 2.11 Research Gap 18
• 3 System Model for RIS-Enabled Vehicular Sensing 19
• 3.1 RIS Geometry and Received Signal Model 22
• 4 Proposed Low-Rank DoA Estimation and Adaptive Tracking Framework 28
• 4.1 Algorithm Overview 29
• 4.1.1 Sparse RIS Observation Model 29
• 4.1.2 Low-Rank Subspace Stabilization 29
• 4.1.3 Beamspace-Based DoA Estimation 30
• 4.1.4 Adaptive DoA Tracking 30
• 4.2 Robust Low-Rank Regularization for Sparse RIS Measurements . 30
• 4.3 Beamspace-Based 2D DoA Estimation 36
• 4.4 Bootstrap Initialization and Target Association 38
• 4.5 Adaptive VFN–RLS Tracking Framework 39
• 4.5.1 Innovation and Variance Tracking 42
• 4.5.2 Motion-Aware Adaptation 44
• 4.5.3 Variable Forgetting Factor and Process Noise Inflation 45
• 4.5.4 Robust Measurement Update 46
• 5 Simulation Results and Discussion 50
• 5.1 Ablation Study of the Proposed Pipeline 57
• 5.2 Computational Complexity Analysis 59
• 5.2.1 Beamspace DoA Estimation 59
• 5.2.2 VFN-RLS Tracking 60
• 5.2.3 Robust Low-Rank Preprocessing 60
• 5.3 Convergence Behavior of ADMM under Sparse Observations 60
• 5.4 Impact of Impulsive Noise and Adaptive Tracking Behavior 62
• 6 Conclusion and Future Work 64
• Bibliography 66
• 요약 73