QoS Provisioning for 5G Mobile Network

Fifth Generation Wireless technology 5G was created by many different organizations through the 3rd Generation Partnership Project 3GPP and, it is a continuation of the previous wireless technology generations that allows for various general improvements over its predecessors, like better latency and, faster download speeds. These improvements offered by Fifth Generation networks, allow for a variety of new applications that generally fall under one of three classes of services these being, enhanced Mobile Broad Band (eMBB), Ultra Reliable Low Latency Communications (URLLC) and massive Machine Type Communications (mMTC). Autonomous vehicle applications are a good example of URLLC, where reliability and low latency, are key factors in maintaining the Quality-of-Service (QoS). Virtual Reality and, ultra-high quality video streaming are good examples of eMBB services, where the throughput and, peak data rates are key factors in maintaining QoS. High-density Internet of Things IoT deployments is a good example of mMTC applications, where the QoS is dependent on the network being able to support the high density of low-power devices present. Due to the diverse QoS requirements of these applications, resource scheduling has become a more challenging task, that requires scheduling techniques that can maintain QoS for all three classes of service on the network simultaneously. Many approaches have been proposed in the literature to tackle this issue, such as implementing network schedulers that can leverage Machine Learning (ML). ML is a branch of artificial intelligence that focuses on the use of sophisticated algorithms to imitate the way humans learn. ML algorithms enable a more intelligent scheduler to manage the limited wireless mobile resources on a network. ML can be split into different categories, i.e. supervised, unsupervised and reinforcement learning algorithms. Supervised learning is taught by example, where the operator provides a known dataset and the algorithm identifies patterns in the data, with a key factor being that the operator knows the correct answers to the training dataset. Unsupervised learning in which the algorithm studies the given dataset without an operator or a known answer. Reinforcement learning is provided with a set of rules, actions and parameters. It is also referred to as a reward function. The algorithm then is left to explore the best course of action to achieve the best possible results, offering a great deal of automation in the network. This has motivated us to propose a novel network resource allocation algorithm, that utilizes Reinforcement Learning, to improve the QoS of the 5G network. Extensive simulation work will be carried out to create a comprehensive reward function that covers various network conditions to implement the algorithm