Traffic Optimization and Optimal Routing in 5G SDN Networks Using Deep Learning

5G networks use a network architecture based on SDN due to the efficiency, low cost, ease of management, and scalability provided by SDN. However, SDN has the issue of determining the routes and traffic optimization centrally. The problem of traffic optimization and routing in 5G networks has been solved through the use of deep learning algorithms such as Deep Deterministic Policy Gradient (DDPG). DDPG provides good results but suffers from the problem of overestimation bias and runs the risk of becoming unstable. These issues have been solved by an alternative deep learning algorithm called Twin-Delayed Deep Deterministic Policy Gradient (TD3). One of the changes in TD3 is training the agent with two Q value functions instead of a single Q value function and taking the minimum of two values. TD3 also uses delayed policy/target updates and smoothing of target policy. There is no mention of TD3 in literature for solving the problem of SDN routing, so this paper analyzes and compares DDPG (existing approach) and TD3 (proposed approach). A simulation environment consisting of the Omnet++ discrete event simulator was used to simulate a 5G network with SDN routing. Two different simulation runs were used—with DDPG and TD3. It was demonstrated that the TD3 approach provided a much better performance with lower latency.