5G Smart Antenna Optimization Algorithm Based On Deep Learning

Massive MIMO antennas in 5G systems need to adjust the direction angle,down-tilt and beam combination frequently according to the user distribution in each cell.Due to the complexity of the joint antenna parameters optimization of all cellular networks,traditional static configuration methods like expert experience can't meet the requirements of flexible and fast adaptive antenna configuration.This thesis proposes a novel and practical real-time 3D massive MIMO antenna parameter optimization algorithm based on deep reinforcement learning,which is used to jointly optimize the antenna parameters of each cell in cellular network.The smart antenna optimization algorithm is trained and implemented on the 5G simulation platform.The preparatory work includes user distribution simulation and generation,pretreatment of relevant antenna parameter data,as well as the construction of user Reference Signal Receiving Power(RSRP)database.The specific research of the smart antenna optimization algorithm is divided into two stages:In the first stage,an algorithm based on the fixed user distribution is proposed,which adopts the Policy Gradient framework to conduct multiple training rounds on the single user distribution.This model can output an optimal antenna beam combination under the input user distribution.In the second stage,an algorithm is proposed for multi-scene user distribution.With the Asynchronous Advantage Actor-Critic framework(A3C),different user distributions are input into the neural network with different threads for asynchronous training.Finally the multi-user distributed reinforcement learning model can output an optimal antenna beam combination under different user distribution.The average RSRP of smart antenna optimization algorithm is 2-4.3%better than other traditional static configuration methods,which shows that using machine learning and deep learning methods of artificial intelligence is feasible and practical to solve complex problems in network planning and massive MIMO antenna parameters optimization.