| Summary: | 碩士 === 國立宜蘭大學 === 電子工程學系碩士班 === 107 === Device-to-device (D2D) technology enables neighboring devices to communicate directly without going through an E-UTRAN NodeB (eNB), thereby reducing the load of eNB, reducing transmission delay, increasing spectrum usage efficiency, and extending network coverage. D2D enables point-to-point communication technology in the LTE-A system by sharing the resources of Cellular User Equipment (CUE). However, devices that reuse the same resources will interfere with each other and may reduce the efficiency of spectrum usage.
In order to save the very limited spectrum, it is an important issue to make D2D devices efficiently reuse the radio resources of CUEs. In recent years, there have been a lot of literature that attempts to use reinforcement learning technique to deal with resource allocation issues. The Multi-Armed Bandit (MAB), also known as the K-Armed Bandit, is a classic model in decision science. It can be used for resource allocation under uncertain environment.
Allocate As Granted (AAG) is a scheme for dynamic resource allocation for CUEs. It can perform resource scheduling for CUEs, while considering transmission requirement and QoS, thereby improving user satisfaction and spectrum usage efficiency. We apply the concept to D2D resource allocation so as to consider the transmission requirements and QoS of D2D at the same time.
In this thesis, we model the D2D resource allocation problems in Multi-Player Multi-Armed Bandit (MP-MAB) framework. By employing reinforcement learning to obtain the preference order of D2D devices. The simulation results show that under the same Greedy Resource Allocation scheme (GRA), The proposed approach can allocate more capacity increase of 12% compares with the Smallest Degree First (SDF) algorithm, while slightly reducing the number of D2D starvation. After switching to the AAG scheme to allocate RBs, although the overall D2D capacity will decrease, it can greatly reduce the number of D2D starvation and improve the satisfaction of D2D, which is more suitable for users' needs.
|
|---|
