Deep Reinforcement Learning-Based Dynamic Scheduling for Real-Time Applications in LTE and RAN Slicing for eMBB in 5G

While LTE (4G) networks offer high-speed data transmission, 5G is necessary to meet the growing demands of modern communication. The main goals of 5G are to make communications more agile by significantly increasing network capacity, lowering latency, and speeding up the networks to handle a massive number of connected devices. To materialize this vision, a new concept has emerged in 5G called network slicing; hence, network providers can offer tailored quality of service based on every type of user according to their specific requirements. However, efficient resource allocation in LTE and across slices or even within a slice in 5G systems to meet QoS is still one of the most critical issues for network designers while resources change over the complexity, dynamic nature, and resource constraints of the network. Classic resource allocation methods cannot cope with the dynamics of the changing network environment; hence, there is an increasing need for faster and smarter resource allocation schemes. This paper proposes an efficient dynamic scheduling algorithm based on machine learning methods for real-time applications in LTE networks, extending to 5G using Radio Access Network (RAN) slicing with a focus on enhanced Mobile Broadband (eMBB) slices. Carrying out extensive simulations and evaluations by using the ns-3 simulator, we compare and analyze various downlink scheduling rules in terms of system throughput, spectral efficiency, average delay, and packet loss ratio. The simulation results show that the proposed algorithm always outperforms the QoS performance, especially for real-time applications.