| With the development of wireless communication technology,the data and energy consumption generated by mobile communication network increase rapidly,and users have higher requirements to improve the transmission rate and reduce the system energy consumption.Beamforming adopts multi-antenna technology,which has the advantages of improving signal quality,expanding system capacity and improving energy efficiency,etc.This paper mainly studies the beamforming that maximizes spectrum efficiency and energy efficiency in the downlink multi-base station MISO interference channel.In this paper,we study the system spectrum efficiency maximization and energy efficiency maximization under the restriction of base station transmission power.These two problems are both non-convex optimization problems,so it is difficult to get the global optimal solution.In traditional methods,non-convex optimization problems are generally transformed into convex optimization problems,and iterative algorithms are used to approach the optimal solution.However,iterative algorithms have the disadvantage of high computational complexity,so it is of great significance to seek low-complexity beamforming methods.In recent years,deep learning,as an important branch of artificial intelligence,has developed rapidly.In deep learning,the neural network is trained offline,and then the trained neural network is used to predict the target online,which reduces the online computing complexity.In this paper,the deep learning algorithm is mainly used to design the beamforming which maximizes the spectral efficiency and energy efficiency.First,this paper proposes a beamforming algorithm based on deep learning to maximize spectral efficiency.Generally,the convolutional neural network is directly used to estimate the beamforming vector,but the dimensionality of the network output will increase significantly as the number of transmitting antennas increases,which makes the training of the neural network becoming difficult.Therefore,this paper proposes a low-complexity deep learning algorithm.The algorithm is mainly divided into two steps.The first step uses deep learning to obtain the parameters of the beamforming vector structure,and the second step uses the network's output to obtain the optimal beamforming through dichotomy algorithm.The algorithm considers both performance and calculation delay,and can learn the optimal beamforming in real time.Finally,by comparing with the benchmark algorithm,the simulation results prove that the algorithm can not only guarantee the effectiveness of capacity performance,but also reduce the computational complexity.Then,this paper considers the power consumption problem on the basis of the system model of spectrum efficiency maximization,and proposes a beamforming algorithm based on deep learning to maximize energy efficiency.The network structure in the algorithm uses a convolutional neural network.In order to reduce the output dimension of the network and reduce the training complexity,some scalar parameters in the optimal solution of the beamforming vector are selected as the output data set.The optimal beamforming is obtained by network prediction's online calculation.Finally,by comparing with the benchmark algorithm,the simulation results prove that the algorithm can not only guarantee the effectiveness of energy performance,but also reduce the computational complexity. |
