A Q-learning-based downlink scheduling in 5G systems

Nowadays, due to the rapid growth of network service requests and popularity of IoT device deployment, wireless networks currently suffer from huge traffic produced by these requests and devices. On the other hand, stand-alone 5G networks will soon be available in the near future. People expect to have high quality streaming mechanisms to enrich and color their everyday lives. In a 5G network, data transmission between base station (BS) and user equipment (UE) is one of the biggest challenges for high-quality streaming since the bandwidth that a BS can provide is limited. Besides, the bandwidth efficiency of a BS can be further enhanced and an ideal DL:UL ratio for a residential user is 10:1 to 20:1. Also, most users of a BS are residentials. They frequently download data without generating uplink traffic or generating less uplink traffic. Therefore, in this study, we propose a 5G downlink scheduling mechanism, named the Q-learning based Scheduling and Resource Allocation Scheme (QSRAS), which deploys Q-learning techniques to effectively improve the quality of wireless transmission, and efficiently manage radio resources of a BS. This scheme dynamically adjusts radio resource allocation by referring to QoS parameters, including throughputs, delays and fairness for UEs. According to our experimental results and analyses, it can effectively trade-off the throughput and fairness of overall system in the multiple traffic. The QSRAS outperforms available state-of-the-art schemes on the summation of Fairness and Normalized spectral efficiencies. The improvements range between 2.84 and 10.51%, especially when many more users are served by a base station.