Optimizing V2V Communication in 5G/6G Networks with Enhanced Deep Cooperative Q-Learning

Résumé fourni par la source

Understanding the physical channel characteristics of V2V communications is important, but it is neither cost-effective nor practical to have full information about them. Since there is always delay and data-loss during communications, which are further accelerated when the wireless channel status and device parameters remain uncalibrated frequently, full or even partial information of the environment is not feasible, even for a short period. Moreover, the vehicular mobility can also introduce abrupt changes to communication. These uncertainties dictate that a system should be well-and adaptively-designed to handle changing information. V2V communications that exploit this information naturally serve as candidates for artificial Intelligence (AI) algorithms. Specifically, there are applications of cooperative learning strategies that are suitable, including machine learning, deep reinforcement learning, and deep multi-agent reinforcement learning. This paper applies novel Enhanced Deep Cooperative Q-Learning (DCO-DQN) model for V2V communication, to achieve the best trade-offs (latency/ reliability/safety) under various states, including short and long memory, asymptotic and non-asymptotic, possible and unpredictable outcomes scenarios. The proposed framework incorporates spatial wisdom and historical information into Euclidean spaces, accounting for the above behavior and challenges. For this reason, future V2V communication in 5G/6G networks may benefit from adopting our approach, as previously described. A thorough comparative study has been conducted to illustrate our advantages and shortcomings. Our analysis also offers insights on hybrid techniques and decentralized approaches, as possible means for future development.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.