Power Allocation In Massive MIMO Systems Based On Deep-Learning

Multiple input multiple output(MIMO)systems can improve communication quality and reliability by transmitting and receiving data simultaneously on multiple antennas.The centralized massive MIMO further increases the number of antennas,allowing the system to achieve higher system capacity with more users.With the rapid development of wireless communication system,Cell-Free massive MIMO technology has been proposed.It has advantages of better data transmission rate and quality,and more flexible network deployment.It is expected to become one of the key technology of 6G network communication system in the future.In order to achieve better performance of the system,power allocation is a key problem and has become a hot research topic.In this paper,the power allocation problem of massive MIMO and Cell-Free massive MIMO is discussed,and a power allocation method based on deep learning is proposed respectively.For a centralized massive MIMO system,this paper uses deep learning method,based on convolitional neural network(CNN)and long short term memory(LSTM)network,designed two kinds of neural network models.Under the multi-cell minimum mean squared error precoding,the calculation results of iteration-based Max-Min Fairness power allocation method and user’s geographical location data were used as the training data to build a deep neural network model to learn the mapping relationship.Specifically,the neural network model is used to learn the features of two kinds of data to find the relationship between them,so as to predict the optimal power distribution under certain conditions.Finally,the performance of the models with different training rounds is compared and analyzed.The results show that,compared with the simple fully connected neural network,the two proposed deep neural network models converge faster in the training process,and the obtained power allocation coefficients significantly improve the total spectral efficiency.For Cell-Free massive MIMO system,this paper uses conventional power allocation schemes based on Large-Scale Fading coefficients and iterative optimization as the initial data set under regularized zero-forcing and maximum ratio precoding,respectively,with the help of deep learning method.Three neural network models are designed based on CNN,LSTM and attention mechanism,and a unique neural network model is trained for each distributed Access Point(AP)to predict the power distribution coefficient according to the local large-scale fading coefficient of AP.To minimize interference and improve spectral efficiency,each AP can allocate different amounts of power to nearby user equipment depending on its geographical location.Finally,based on the above work,the simulation experiment and comparative analysis are carried out.The results show that the power allocation strategy based on deep learning can significantly reduce the time complexity,and has good feasibility in practical application.