Low latency communication is one of the fundamental requirements for 5G wireless networks and beyond. In this paper, a novel approach for joint caching, user scheduling and resource allocation is proposed for minimizing the queuing latency in serving user's requests in cloud-aided wireless networks. Due to the slow temporal variations in user requests, a time-scale separation technique is used to decouple the joint caching problem from user scheduling and radio resource allocation problems. To serve the spatio-temporal user requests under storage limitations, a Reinforcement Learning (RL) approach is used to optimize the caching strategy at the small cell base stations by minimizing the content fetching cost. A spectral clustering algorithm is proposed to speed-up the convergence of the RL algorithm for a large content caching problem by clustering contents based on user requests. Meanwhile, a dynamic mechanism is proposed to locally group coupled base stations based on user requests to collaboratively optimize the caching strategies. To further improve the latency in fetching and serving user requests, a dynamic matching algorithm is proposed to schedule users and to allocate users to radio resources based on user requests and queue lengths under probabilistic latency constraints. Simulation results show the proposed approach significantly reduces the average delay from 21% to 90% compared to random caching strategy, random resource allocation and random scheduling baselines.

翻译：低时延通信是5G及未来无线网络的基本需求之一。本文提出了一种联合缓存、用户调度与资源分配的新方法，旨在最小化云辅助无线网络中服务用户请求时的排队时延。由于用户请求具有缓慢的时间变化特性，采用时间尺度分离技术将联合缓存问题与用户调度及无线资源分配问题进行解耦。为了在存储限制下服务时空变化的用户请求，采用强化学习方法优化小小区基站的缓存策略，通过最小化内容获取成本。针对大规模内容缓存问题，提出一种谱聚类算法，基于用户请求对内容进行聚类，从而加速强化学习算法的收敛速度。同时，提出一种动态机制，根据用户请求对耦合基站进行本地分组，以协作优化缓存策略。为进一步降低获取与服务用户请求的时延，提出一种动态匹配算法，基于用户请求和队列长度，在概率时延约束下调度用户并将其分配至无线资源。仿真结果表明，与随机缓存策略、随机资源分配及随机调度基线相比，所提方法将平均时延显著降低了21%至90%。