Optimization of 5G multi-RAT network traffic scheduling based on DDPG

As the advancement of mobile wireless network technology progresses, an increasingly diverse array of applications with specific Quality of Service (QoS) requirements emerges, presenting heightened challenges for network throughput, latency, and reliability. The diversity and complexity of these applications necessitate more flexible and efficient handling of various data streams by the network, ensuring user experience while meeting stringent performance standards. In the context of 5G, the introduction of Heterogeneous Networks (HetNet) and Multi-Connectivity (MC) technologies significantly enhances network performance. This enhancement is achieved by activating multiple links across various Radio Access Technologies (RAT) to offload traffic, thereby improving throughput and reducing latency. However, this approach of multi-connectivity also introduces greater challenges in terms of interference and power consumption. This paper proposes a traffic scheduling strategy based on Deep Deterministic Policy Gradient (DDPG). This strategy takes into account the diverse QoS requirements, the characteristics of different RATs, and the current network load, aiming to develop an efficient traffic scheduling plan for User Equipment (UE). The primary goal is to maximize QoS satisfaction for UE while minimizing the costs associated with multi-connectivity. To demonstrate the effectiveness of the proposed strategy, We proposed three types of QoS traffic and compared our strategy with those based on QoS and Deep Q-Network (DQN). The experimental results indicate that our strategy significantly outperforms the DQN-based strategy in meeting QoS requirements. Compared to QoS-based strategies, our strategy achieved a maximum reduction of 23.1% in throughput performance and 6.06% in latency performance, while achieving up to 61.04% reduction in connection costs. These results confirm that our strategy not only improves network performance but also significantly reduces resource consumption.