Faster Convergence of Q-Learning in Cognitive Radio-VANET Scenario

摘要

Abstract Cognitive Radio (CR) based Vehicular Ad hoc Network (VANET) or CR-VANET has become a very promising research domain. VANET is used to reduce road accidents, traffic congestion, and to provide other user experiences such as uninterrupted entertainment services. CR, on the other hand, solves bandwidth scarcity issue of VANET. For the high-speed mobility of the vehicles, the cognitive process of CR faces several challenges. Machine Learning (ML) has arrived as an integral tool to handle such challenges. Q-learning algorithm, a member of Reinforcement Learning (RL), which is a type of ML, is the most suitable for CR-VANET as it does not need any prior environment model and training dataset. But the problem is that it takes a longer time for learning purposes. In this paper, a dynamic ML framework is proposed. Case-based reasoning learning, cooperative spectrum sensing, teacher-student transfer learning approach will be aligned with the Q-learning for the faster convergence regarding the spectrum sensing issues in CR-VANET. The framework will accelerate the learning of the vehicles, and that is very important for the energy-efficient and real-life VANET implementation.