Real-Time Adaptive Routing in Vehicular Ad Hoc Networks Using SDN and Graph Neural Networks

Abstract

Vahicular Ad Hoc Networks (VANETs) also introduce interesting issues to provide reliable communication as a result of immense node mobility, regular changing topologies as well as strict latency requirements. The dynamical adaption to such conditions is frequently failing of the traditional routing protocols. The given work proposes a new real-time adaptive routing framework to optimize Software-Defined Networking (SDN) with Graph Neural Networks (GNNs) in VANETs to increase the routing intelligence and scale. The SDN controller is able to see the global, real-time topology of the network through receipt in distributed roadside units (RSUs) of vehicular state and link measurements. This information is then dynamically encoded to a spatiotemporal graph structure, feed into a GNN architectural model that is trained in predicting the optimal routing paths using current, and historical mobility patterns. The potential GNN-SDN system is tested by co-simulations based on the SUMO (to model the traffic) and Mininet-WiFi (to emulate the network). The performance measures are studied under different traffic rates and mobility conditions such as Packet Delivery Ratio (PDR), end-to-end latency, and network throughput. Experimental study shows that the GNN-assisted routing is significantly better than AODV and DSR protocols as they can achieve an up to 27 percent higher PDR and up to 35 percent lower delay in high-mobility urban networks. This article shows the effectiveness of Artificial intelligence based SDN control planes to solve VANETs routing complexities and leads to the development of context-aware, highly scalable and tenacious communication infrastructure essential to the delivery of autonomous and connected transportation systems in 6G-supported smart cities and V2X environments.