Design Of Resource Optimization Algorithm For D2D System In Unlicensed Spectrum

With the rapid development and implementation of the fifth generation mobile communication technology 5G,Device to Device Communication(D2D)technology has confronted more diverse application scenarios and higher transmission demand.Due to the limited spectrum resources in licensed spectrum,the deployment of D2 D systems may cause serious interference to the performance of existing cellular networks.In order to address this challenge,Device-to-Device Communication on Unlicensed Bands(D2D-U)has been proposed to allow D2 D devices to reuse unlicensed spectrum with Wi Fi system,which could significantly increase system capacity and optimize transmission performance.This thesis mainly studies the distributed resource allocation scheme of D2D-U system.The main purpose is to design a distributed spectrum and power resource allocation algorithm for D2 D users to achieve the maximum transmission performance of D2D-U system with the premise of harmonious coexistence with Wi Fi networks.In order to achieve the harmonious coexistence with Wi Fi system,the thesis first proposes two schemes based on deep reinforcement learning and neural network filtering for Wi Fi traffic load estimation on unlicensed channels.In the model based on deep reinforcement learning method,D2 D users perceive Wi Fi traffic load and adjust the spectrum resource usage strategy based on the transmission collision between D2 D devices and Wi Fi users in channels.In the model based on neural network filtering approach,D2 D users directly estimate the number of active Wi Fi users by monitoring the collision probability of Wi Fi users’ transmission,thereby realizing the estimation of Wi Fi traffic load.Here both two methods are distributed algorithms,where D2 D users use Q function and unsupervised objective function to realize online training of neural networks,respectively.Then based on the obtained unlicensed band traffic load,a joint resource allocation algorithm for distributed D2D-U system based on the price model is proposed.In view of the traditional algorithm maximizing system throughput while ignoring the issue of internal fairness of D2D-U system,this thesis uses the price model to adjust the transmission balance among D2 D users.In addition,unsupervised training neural networks are utilized to realize unlicensed channels pricing for different D2 D users.As neural networks achieve convergence during online training and federated learning,D2 D users can maximize their data rate while meeting the fairness of the Wi Fi system and the fair transmission within D2D-U system.Finally,a large number of numerical simulations and analyses are provided to verify the performance of the mentioned algorithms,which provides references for the follow-up research and deployment related to D2D-U.