In wireless communication networks, the network provider serves certain licensed primary users who pay for a dedicated use of the frequency channels. However, not all the channels are occupied by the primary users at all times. For efficient spectrum utilization, in centralized cognitive radio networks (CRNs), a cognitive base station (CBS) dynamically identifies the spectrum holes and allocates the frequency channels to the on-demand unlicensed secondary users known as cognitive radios (CRs). Although existing literature has developed various dynamic spectrum access mechanisms for CBS, there is still a dearth of studies due to the wide range of assumptions made in the solutions. Most of the existing works study the CBS scheduling problem scheme by adopting optimization-based methods and rely on the prior knowledge of the network parameters such as primary users’ activity. Moreover, the impact of channel switching costs on the network throughput has not been well studied. In this paper, we aim to maximize the CRNs total throughput, and we formulate the CBS scheduling problem as a non-stochastic (i.e., adversarial) combinatorial multi-armed bandit problem with semi-bandit feedback and arm switching costs. We propose two novel online learning algorithms for CBS scheduling with and without channel switching costs, where their regret performances are proved sublinear order-optimal in time as T1/2T^{1/2} and T2/3T^{2/3} , respectively, offering throughput-optimal scheduling for CRNs. Experiments on the synthetic and real-world spectrum measurement data complement and validate our theoretical findings.

中文翻译：

---

集中式认知无线电网络中的多用户调度：多臂强盗方法


在无线通信网络中，网络提供商为特定的许可主要用户提供服务，这些用户为频道的专用使用付费。然而，并非所有信道始终都被主要用户占用。为了有效利用频谱，在集中式认知无线电网络 (CRN) 中，认知基站 (CBS) 动态识别频谱空洞，并将频率信道分配给按需未经许可的二级用户，称为认知无线电 (CR)。尽管现有文献已经开发了各种针对 CBS 的动态频谱接入机制，但由于解决方案中做出的假设范围广泛，因此仍然缺乏研究。现有的大多数工作采用基于优化的方法来研究CBS调度问题方案，并依赖于网络参数（例如主要用户的活动）的先验知识。此外，信道切换成本对网络吞吐量的影响尚未得到很好的研究。在本文中，我们的目标是最大化 CRN 的总吞吐量，并将 CBS 调度问题表述为具有半强盗反馈和臂切换成本的非随机（即对抗性）组合多臂强盗问题。我们提出了两种新颖的在线学习算法，用于有和没有信道切换成本的 CBS 调度，其中它们的遗憾性能被证明是时间上次线性顺序最优的，为 T1/2T^{1/2} 和 T2/3T^{2/3} ，分别为 CRN 提供吞吐量最优调度。对合成和真实世界频谱测量数据的实验补充并验证了我们的理论发现。