Research On Machine Learning Based Power Control Algorithm For D2D Communication On Unlicensed Spectrum

The rapid development of the 5th Generation Mobile Communication Technology(5 G)has prompted tremendous enhancement to the performance of wireless communication network.As one of the key technologies of 5G,Device-to-Device communication(D2D)can effectively cope with challenges caused by the large demand for mobile data traffic and the increase of D2D users in 5G era.In addition,the increase of D2D users leads to the scarcity of cellular network capacity,D2D Communication on Unlicensed Spectrum(D2D-U)is capable of improving system throughput and capacity performance,which is considered as a method that adapts to the evolution trend of communication standards.Meanwhile,due to the fast development of machine learning in recent years,the combination of machine learning and wireless communication has become a hot topic,therefore,the theme of the paper is exploiting machine learning to deal with power control issue for D2D communication on unlicensed spectrum.The paper first focuses on real-time operation of D2D-U power control issue,leveraging the advantage of Convolutional Neural Network(CNN)in extracting spatial features from multi-dimensional data,proposes a CNN-based power control for D2D-U.Specifically,we input a large number of input feature of channel state information(CSI)to CNN and train it under the supervision of theoretical label.Simulation results show that CNN can reach the upper limit of theoretical throughput performance while greatly reducing the time consumption,which is conducive to the real-time operation in unlicensed spectrum.The paper then analyzes the correlation between the number of D2D pairs in unlicensed spectrum and the structure of power control network,and proposes Model-Agnostic Meta-Learning(MAML)based D2D power control algorithm in user-variable unlicensed spectrum scenario.Experiment results indicate that MAML can provide D2D communication system with a "value" equipped with prior experience,when the number of D2D pairs in scenario changes,the "value" would assist power control network to quickly adapt to the new scene under a small number of data samples,which avoids the redundancy of rebuilding and retraining new power control network.Finally,the paper comprehensively evaluates the throughput and fairness of D2D and the wireless fidelity(Wi-Fi)system in unlicensed spectrum.According to the scheme of D2D accessing to unlicensed spectrum,the paper analyzes factors that affect the overall performance of unlicensed spectrum,including D2D accessing time proportion and D2D transmit power allocation policy,and we accordingly propose hierarchicalDeep Q Network(h-DQN)based joint D2D accessing time proportion and power allocation algorithm.Experiment results show that the proposed algorithm can ensure the fairness of the system while fulfilling quality of service(QoS)for users in unlicensed spectrum,besides,the proposed algorithm would highlight user experience better than unlicensed spectrum accessing schemes in original standards.