Deep-Q Reinforcement Learning for Fairness in Multiple-Access Cognitive Radio Networks
Deep-Q Reinforcement Learning for Fairness in Multiple-Access Cognitive Radio Networks
复制标题
DOI：
10.1109/wcnc51071.2022.9771661
复制
发表时间：
2022-04
期刊：
影响因子：
--
通讯作者：
Zain Ali;Z. Rezki;H. Sadjadpour
中科院分区：
文献类型：
--
作者：
Zain Ali;Z. Rezki;H. Sadjadpour
本文提出了一种深度q强化学习（DQ-RL）框架，以实现多访问认知无线电（CR）系统的公平性。该框架提供了快速的解决方案，并且对通道动态具有鲁棒性。此外，为了消除从辅助接收器（SR）反馈数千个权重的计算开销和负担，我们提出了一种在辅助发射器（STs）上进行学习过程的解决方案。仿真结果表明，采用该方法可以实现较好的公平性，主系统的中断概率小于0.04。我们还将所提出的技术与暴力优化方法进行了比较，并展示了与速率最大化模型相比，所提出框架的公平性增益。
This work presents a deep-Q reinforcement learning (DQ-RL) framework to achieve fairness in multi-access cognitive radio (CR) systems. The proposed framework provides fast solution and is robust to channel dynamics. Further, to remove the computational overhead and the burden to feedback thousands of weights from the secondary receiver (SR), we propose a solution where the process of learning is carried out at the secondary transmitters (STs). The simulations show that by using the proposed technique, a good level of fairness is achievable with an outage probability of the primary system less than 0.04. We also provide the comparison of the proposed technique with a brute-forcing optimization method, and show the fairness gain of the proposed framework compared to the rate maximization model.