A MDP-based Method For Physical Security In Massive MIMO Systems

With the rapid development of wireless communication technology,the 5G will complete the standard formulation and be fully commercialized in 2020.As one of the key physical layer technologies of 5G,Massive MIMO has great research potential.The information transmission of wireless network face the threat of eaves-dropper because of the openness of wireless media.Traditional wireless security relies heavily on complex high-level encryption,while physical layer security,as the underlying security technology,can obtain unconditional security using the inherent differences in the physical characteristics of wireless channels.Therefore,it is necessary to study the physical layer security of wiretap channels in Massive MIMO system.In addition,wireless communication and artificial intelligence are advancing together,giving birth to an interdisciplinary subject in recent years.Machine learning represented by reinforcement learning and deep learning has become an important research method of wireless communication technology.Massive MIMO signal processing based on reinforcement learning has a large number of theoretical literature,but they are not general and don't form a deep and complete theoretical system.Due to the intelligent disciplinary characteristics of large scale,high rate and numerous information of the wireless signals in the physical layer of Massive MIMO exhibit,this paper adopts reinforcement learning algorithm to solve the security problems of the physical layer of Massive MIMO and puts forward a new research idea for wireless signal processing.Therefore,this paper proposed a value iterative algorithm based on Q-value update based on the analysis of the physical layer characteristics of the wiretap channel and the construction of the MDP model,which aims at maximizing the secrecy capacity and obtains the optimal power allocation policy.At last,the parameters of the algorithm are evaluated and analyzed by the simulation platform,and also the performance is verified,which indicates that the algorithm can obtain the optimal policy of power allocation stably and effectively under the premise of the safe transmission of information in Massive MIMO wiretap channel.Therefore,the main research work is as follows:(1)It studied the physical layer characteristics and theoretical channel model of Massive MIMO system and established a downlink channel for the wiretap channel by considering the mutual coupling impedance and load impedance of large-scale antenna array.Then,it deduced and analyzed the channel capacity of MIMO channel,the optimal scheme of MIMO channel capacity,and the secrecy capacity of different channel in the wiretap channel,especially the wiretap channel with large antennas,which makes a model foundation for the decision and optimization of the reinforcement learning.(2)It studied the characteristics of Markov decision process and designed the channel environment model of MDP according to the wiretap channel.And it derived the transition probability expression of the MDP model by using the interval transfer probability model of finite state Markov channel,which realized the sub-channel transfer of the base station when transmitting power is allocated.Then the MDP model over the discounted reward has been established with maximizing the secrecy capacity of Massive MIMO wiretap channel,which constructed the basic framework of reinforcement learning algorithm in the physical layer of Massive MIMO.(3)It studied different solving methods of MDP,and calculated the optimal value function according to Bellman's optimal equation,and proposed a value iterative algorithm based on Q-value update,which enables the base station to obtain an optimal transmitting power allocation policy and can allocate transmitting power optimally according to the changes of channel environment.Then,the learning process of the system model is simulated on the simulation platform.Firstly,it evaluated the performance of the algorithm by analyzing the influence of different parameters.Secondly,the optimal policy of the system model is found by combining the optimal parameters.The simulation results show that,when the base station increases the transmitting power within a certain range,it can improve the secrecy capacity of the physical layer of Massive MIMO wiretap channel under the premise of ensuring the safe transmission of information,which further verify the performance of the algorithm.