Some Problems in Sensing and Communication in Wireless Systems
Files
TR Number
Date
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Sensing and communication constitute two fundamental pillars of modern wireless systems, enabling intelligent interaction between the physical environment and communication networks. Wireless sensing empowers systems to perceive environmental states, object locations, and motion dynamics, forming the basis for situational awareness in applications such as autonomous systems, smart cities, healthcare monitoring, and industrial automation. Meanwhile, wireless communication provides the essential infrastructure for reliable and efficient information exchange across distributed devices, supporting massive connectivity, ultra-low latency transmission, and high data rate services in emerging networks. Beyond these individual functionalities, emerging applications such as intelligent transportation systems, smart manufacturing, and unmanned aerial networks increasingly require both environmental perception and real-time data transmission within the same operational network. In these systems, sensing and communication jointly contribute to system integration and coordination, motivating an integrated sensing and communication (ISAC) paradigm within next-generation wireless systems.
Across the three scenarios—sensing, communication, and ISAC systems—numerous challenges emerge in the pursuit of overall performance optimization. This dissertation systematically addresses the critical issues inherent in each of these domains. The work is organized into three main parts, each dedicated to one of the three scenarios.
In the first part, we investigate two key problems in automotive radar sensing.
We study real-time super-resolution direction-of-arrival (DoA) estimation for automotive radar systems, targeting a 1-degree angular resolution within 1 ms processing time. To achieve this stringent requirement, we propose a novel DoA estimation algorithm that first exploits the intrinsic correlation between radar measurements in adjacent time slots to significantly reduce the search space of the maximum likelihood estimation (MLE) problem. Instead of performing an exhaustive search over the entire angular domain, we adopt the compressed sensing algorithm Orthogonal Matching Pursuit (OMP) to efficiently obtain an optimal or near-optimal solution within the reduced search space. Both simulation results and hardware experiments demonstrate that the proposed method outperforms state-of-the-art DoA estimation techniques in terms of accuracy, while being the only known approach capable of satisfying the demanding real-time constraint.
We further investigate real-time interference mitigation for automotive radar systems. The design objective is to enhance target estimation accuracy under strong interference while satisfying stringent real-time computational constraints. The proposed algorithm first exploits the sparsity of radar signals in the joint time–frequency domain to identify interference components. It then employs OMP algorithm to effectively separate the desired target signal from the interference. In addition, the intrinsic correlation between measurements in neighboring time frames is leveraged to reduce the search space of the OMP procedure, thereby significantly lowering computational complexity. Simulation results demonstrate that the proposed method achieves approximately 1 ms processing time and outperforms state-of-the-art approaches in terms of target detection accuracy under severe interference conditions.
In the second part, we investigate two key scheduling problems in Open Radio Access Networks (O-RAN).
We first study real-time robust scheduling in O-RAN. The design objective is to optimize network performance under imperfect channel state information (CSI). We develop a comprehensive scheduling framework that jointly performs resource block (RB) allocation, modulation and coding scheme (MCS) selection, and beamforming computation. Notably, the proposed scheduler leverages a limited number of CSI samples to provide probabilistic quality-of-service (QoS) guarantees. To satisfy stringent real-time requirements, each subproblem is carefully structured for parallel execution, operates over a reduced search space, and is implemented on a GPU platform to accelerate computation. Experimental results demonstrate that the proposed algorithm achieves throughput performance competitive with state-of-the-art methods while simultaneously satisfying QoS guarantees.
We further investigate a model-guided learning scheduler for O-RAN. The design goal is to combine the strengths of model-based and learning-based approaches while mitigating their respective limitations. The proposed framework first formulates a model-based optimization problem to derive a reduced yet high-quality scheduling space. This reduced space is then used as the action space for an online deep multi-agent reinforcement learning (DMARL) algorithm, which determines the final scheduling decisions. The scheduler is specifically tailored to the O-RAN architecture, enabling seamless integration into different control loops while meeting strict timing constraints. Experimental results show that the proposed approach outperforms both state-of-the-art model-based and learning-based schedulers in overall network performance.
In the third part, we study joint scheduling in ISAC networks. The objective is to efficiently allocate shared resources between sensing and communication in 5G networks to maximize overall performance. We formulate a unified scheduling problem that captures both cooperative and competitive interactions between sensing and communication, thereby accounting for coordination and integration gains. To solve this problem, we first derive a reduced yet promising scheduling space, which is then used as the action space of a multi-agent deep deterministic policy gradient (MADDPG) framework to determine the final scheduling decisions. Experimental results demonstrate that the proposed algorithm outperforms state-of-the-art methods in terms of both QoS guarantees and throughput, while satisfying stringent real-time requirements.