부산대학교 도서관

The conventional resource distribution methodologies rely on numerical methodologies to enhance diverse performance metrics. Most of these endeavors can be classified as immediate, given that the optimization determinations stem from the present network condition without regard for historical network states. Although utility theory has the capacity to integrate long-term optimization consequences into these optimization actions, the escalating diversity and intricacy of network settings have made the resource allocation challenges insurmountable. The optimization of resources at an optimum level stand as a foundational hurdle for densely populated and mixed wireless environments with an extensive array of wireless connections. Owing to the intricate and nonlinear nature of the optimization conundrum, the quest for the best resource allocation is a resource-intensive undertaking. Among the prospective solutions, reinforcement learning (RL) emerges as a viable candidate to resolve resource allocation dilemmas optimally across fluctuating network scenarios. This paper presents an innovative, centralized RL-based resource allocation method tailored for a multi-cell network, aiming to optimize connection stability and data rate by improving the quality of experience (QoE). Specifically, a deep Q-network (DQN) approach is employed to realize this objective. Empirical findings underscore that the proposed deep reinforcement learning (DRL) based resource allocation strategy delivers better performance within a multi-cell scenario.