Applying Event-Triggered Q-Learning Algorithms

2 ||| 1556-6072/19©2019IEEE IEEE VEHICULAR TECHNOLOGY MAGAZINE | MONTH 2020 Ultradense networks (UDNs) have emerged as a promising architecture that can support the extremely high demand for data traffic in the future. Through the dense deployment of massive small base stations (SBSs), the system can be well promoted in terms of network capacity and spectrum efficiency, but this deployment scheme will also bring huge challenges to wireless resource management. Artificial intelligence (AI) can be applied to UDN scenarios, enabling intelligent communication devices to learn and complete resource allocation. This article discusses resource allocation schemes based on AI algorithms in UDNs and proposes an event-triggered, reinforcement-learning-based subchannel and power allocation algorithm. We consider nonorthogonal multiple access (NOMA) in UDNs, which allows a subchannel to be used by multiple users at the same time. Each user is regarded as an agent, and each agent obtains observational information from the environment during the learning process. We design event-triggered conditions based on current and previous moments of observational information, and the agent uses those conditions to decide whether to update policies and perform actions. Simulation results show the effectiveness of the proposed eventtriggered, reinforcement-learning-based resource allocation algorithm.