Adaptive Network Selection for Latency-Aware V2X Systems under Varying Network and Vehicle Densities

TL;DR

This paper introduces ANS-V2X, a heuristic framework for real-time network selection in V2X systems that balances latency, computational load, and application needs, outperforming existing methods in speed and near-optimality.

Contribution

The paper proposes a novel heuristic-based network selection framework for V2X systems that achieves near-optimal performance with significantly reduced decision time compared to MILP and reinforcement learning methods.

Findings

ANS-V2X achieves within 5-10% of MILP optimal utility.

It reduces decision time by over 85% compared to MILP.

It consistently delivers lower latency than Q-learning and MILP approaches.

Abstract

This paper presents ANS-V2X, an Adaptive Network Selection framework tailored for latency-aware V2X systems operating under varying vehicle densities and heterogeneous network conditions. Modern vehicular environments demand low-latency and high-throughput communication, yet real-time network selection is hindered by diverse application requirements and the coexistence of multiple Radio Access Technologies (RATs) such as 4G, 5G, and ad hoc links. ANS-V2X employs a heuristic-driven approach to assign vehicles to networks by considering application sensitivity, latency, computational load, and directionality constraints. The framework is benchmarked against a Mixed-Integer Linear Programming (MILP) formulation for optimal solutions and a Q-learning-based method representing reinforcement learning. Simulation results demonstrate that ANS-V2X achieves near-optimal performance, typically within 5 to 10% of the utility achieved by MILP-V2X, while reducing execution time by more than 85%. Although MILP-V2X offers globally optimal results, its computation time often exceeds 100 milliseconds, making it unsuitable for real-time applications. The Q-learning-based method is more adaptable but requires extensive training and converges slowly in dynamic scenarios. In contrast, ANS-V2X completes decisions in under 15 milliseconds and consistently delivers lower latency than both alternatives. This confirms its suitability for real-time, edge-level deployment in latency-critical V2X systems