Research And Implementation Of Heterogeneous Cellular Network Coverage Analysis And Power Control Algorithm Based On Deep Learning

With the development of communication technology,network planning and optimization are facing higher requirements and challenges.Self-intelligence network uses the digital intelligence brain platform to provide massive data management services and endogenous intelligent technology to achieve intelligent sensing,realtime analysis and rapid decision-making and other service capabilities,providing strong support for network planning.Heterogeneous cellular networks are widely used to solve the congestion problem of dense hotspot cells in mobile communication networks.Accurate and realtime coverage analysis can accurately reflect the coverage capability of the network,quickly analyze future network changes,and help operators optimize network planning and deployment solutions.At the same time,according to the coverage effect,the power of different types of base stations can be quickly adjusted,which can improve user satisfaction and spectrum efficiency.Therefore,it is of great significance to accurately analyze the coverage of heterogeneous cellular networks in real time and to control the power quickly and effectively.This thesis aims to build a coverage analysis and power control model for heterogeneous cellular networks.A deep learning algorithm is used to accurately estimate the network coverage effect,and a deep reinforcement learning algorithm without central coordination is used to dynamically adjust the base station power to improve the total network rate and user satisfaction.The research contents are as follows:(1)Accurate and real-time analysis of coverage based on deep learning algorithm.Convolutional neural network combined with fully connected network is used to train the dynamically sampled measurement report data(MR data).Mean absolute error(MAE)and root Mean square error(RMSE)are used as evaluation indexes to compare the accuracy of the algorithm with other algorithms,and the trained model is used for coverage analysis of different prediction scenarioes.Experiments show that the model is more accurate,and can accurately analyze the coverage performance and quickly reflect the coverage effect of the future network.(2)Fast and effective power control based on deep reinforcement learning algorithm.User reference signal receiving power data predicted by a neural network in a coverage model is taken as a sample,and a distributed coordination mechanism multi-agent deep reinforcement learning network is adopted to perform power control on a downlink.The experimental results show that the proposed algorithm has a significant improvement in total data rate and user satisfaction,and has significant advantages over other power control schemes.