Multi-antenna millimeter-wave (mmWave) communication systems have become a promising approach to improve throughput. However, mmWave narrow beams are susceptible to blockages, making it difficult to guarantee continuous services for ultra-reliable and low-latency communications (URLLC). To address this issue, we propose a novel dual-path beam tracking framework and develop a Recurrent Neural Network-based Constrained Deep Reinforcement Learning (RCDRL) algorithm to optimize the beam search sets of the beam tracking algorithm. The objective is to minimize the total time and frequency resources allocated for beam sweeping, beam tracking, and data transmission, subject to the constraint on the service interruption probability of URLLC. A pre-training method is developed to improve the initial performance and stability of the RCDRL algorithm. Comprehensive evaluation results indicate that the proposed approach outperforms other baseline methods in terms of the tradeoff between service interruption probability and resource utilization efficiency. Specifically, the RCDRL algorithm reduces the service interruption probability by three orders of magnitude compared with a standardized single-beam tracking method at the cost of sacrificing the resource utilization efficiency by 14.6%. In addition, our policy achieves lower service interruption probability and higher resource utilization efficiency compared with an existing dual-beam tracking algorithm.

Machine learning has been considered a promising approach for indoor localization. Nevertheless, the sample efficiency, scalability, and generalization ability remain open issues of implementing learning-based algorithms in practical systems. In this paper, we establish a zero-shot learning framework that does not need real-world measurements in a new communication environment. Specifically, a graph neural network that is scalable to the number of access points (APs) and mobile devices (MDs) is used for obtaining coarse locations of MDs. Based on the coarse locations, the floor-plan image between an MD and an AP is exploited to improve localization accuracy in a floor-plan-aided deep neural network. To further improve the generalization ability, we develop a synthetic data generator that provides synthetic data samples in different scenarios, where real-world samples are not available. We implement the framework in a prototype that estimates the locations of MDs. Experimental results show that our zero-shot learning method can reduce localization errors by around 30% to 55% compared with three baselines from the existing literature.

Beam management is one of the most challenging issues in low-earth orbit (LEO) satellites, where the antenna direction is dynamic, and the storage, computing, and communication resources are limited. In this work, we develop a zero-shot learning approach to maximize the average data rate of a user by optimizing beam tracking policy in both spatial and temporal domains. In the spatial domain, we develop a graph recurrent neural network (GRNN) with only 60 training parameters to predict the next beam direction. Compared with an existing recurrent neural network with more than 30, 000 parameters, the GRNN can reduce the storage requirement remarkably. In the temporal domain, we design a Twin deep Q-network (DQN) to determine the time to trigger beam tracking. To improve the generalization ability of GRNN and Twin DQN, we apply meta-learning to train them with different antenna directions and test them in unseen scenarios. Simulation results show that our zero-shot learning approach does not need new data samples in unseen scenarios. Thus, it does not introduce extra computing and communication overheads. Additionally, the achievable average data rate is around 10% to 20% higher than different benchmarks.

In this paper, we consider a point-to-point wireless communication system. The source monitors a physical process and generates status update packets according to a Poisson process. The packets are transmitted to the destination by using finite blocklength coding to update the status (e.g., temperature, speed, position) of the monitored process. In some applications, such as real-time monitoring and tracking, the timeliness of the status updates is critical since the users are interested in the latest condition of the process. The timeliness of the status updates can be reflected by a recently proposed metric, termed the age of information (AoI). We focus on the packet management policies for the considered system. Specifically, the preemption and discard of status updates are important to decrease the AoI. For example, it is meaningless to transmit stale status updates when a new status update is generated. We propose three packet management schemes in the transmission between the source and the destination, namely non-preemption (NP), preemption (PR) and retransmission (RT) schemes. We derive closed-form expressions of the average AoI for the proposed three schemes. Based on the derived analytical expressions of the average AoI, we further minimize the average AoI by optimizing the packet blocklength for each status update. Simulation results are provided to validate our theoretical analysis, which further show that the proposed schemes can outperform each other for different system setups, and the proposed schemes considerably outperform the existing ones without packet management at medium to high generation rate.