Intelligent Priority-Aware Spectrum Access in 5G Vehicular IoT: A Reinforcement Learning Approach

Area of Science:

• Wireless communication networks
• Vehicular Internet of Things (V-IoT)
• Next-generation cellular networks

Background:

• Efficient spectrum access is critical for Vehicular Internet of Things (V-IoT) systems.
• Next-generation cellular networks require dynamic Quality of Service (QoS) management.
• Existing spectrum management solutions struggle with the dynamic nature of vehicular environments.

Purpose of the Study:

• To propose a novel reinforcement learning (RL)-based priority-aware spectrum management (RL-PASM) framework.
• To dynamically allocate spectrum resources for high-priority (HP), low-priority (LP), and best-effort (BE) traffic classes.
• To evaluate the performance of different RL algorithms in a centralized RSU-based control system.

Main Methods:

• Modeling the spectrum management environment as a discrete-time Markov Decision Process (MDP).
• Utilizing a context-sensitive reward function for fairness-preserving decisions (access, preemption, coexistence, hand-off).
• Comparing four RL algorithms: Q-Learning, Double Q-Learning, Deep Q-Network (DQN), and Actor-Critic (AC).

Main Results:

• RL-PASM effectively balances throughput, latency, fairness, and energy efficiency.
• DQN achieved the highest average throughput, while Q-Learning offered the lowest average delay and highest energy efficiency.
• Double Q-Learning and Actor-Critic maintained high fairness and low interruption probability.

Conclusions:

• RL-PASM provides a robust and adaptable solution for intelligent, priority-aware spectrum access in vehicular networks.
• The framework is suitable for scalable and resource-constrained deployments, particularly edge-constrained vehicular environments.
• The choice of RL algorithm allows for tailored optimization based on specific network priorities (e.g., throughput vs. energy efficiency).