使用賽局理論及學習演算法在裝置對裝置通訊系統中的中央節點選擇

Abstract

Device-to-Device (D2D) communications provides a proximity service, consuming less energy and having higher spectrum reuse. It has become more and more popular in recent years. In our work, we consider that the devices in a cell request the same data from a base station (BS). The devices will form some clusters to receive data. Every cluster will have one device be central entity. The central entity in a cluster receives the data from the BS, and then broadcasts the data to all other devices in the same cluster. The central entity suffers from the cost of transmit power consumption, which discourages the devices from being the central entity. As the devices are selfish in maximizing their own utility, game theory serve as a powerful technique for analyzing the behavior of the devices. We formulate the selfish and non-cooperative interaction of the devices under the system as a game problem. To solve this problem, we propose a central-entity-election mechanism that motivates the devices to report the true transmission costs, and elects the most appropriate devices as the central entities to reach the maximum system utility (social welfare). On the other way, we prove that the multiple-cluster central entity election is a NP hard problem. To avoid the NP hard problem, we propose the distributed central entity election learning (DCEE) algorithm to form clusters. We prove the DCEE algorithm can always converge and have many desirable properties as budget balance and individual rationality. In the simulation part, we verify the theoretical analysis in a real LTE system setting. With the proposed mechanism and the simulation results, D2D communications is shown to have the potential to improve the performance of wireless networks.