Research On MmWave Massive MIMO Channel Estimation Algorithm Based On Manifold Learning

With the increasing number of users and terminals and the emergence of new information technologies such as the Internet of Things and artificial intelligence,human requirements for wireless communication system speed,capacity and delay have also greatly improved.Millimeter wave technology can improve the utilization of large spectrum because of its large bandwidth,and is a potential new technology in the next generation of mobile communication systems.At the same time,due to its short wavelength,millimeter wave can reduce the antenna size,realize higher density antenna deployment,and combine massive MIMO technology to increase the utilization of frequency band resources and improve the spectral efficiency of the system.However,in order to take full advantage of massive MIMO systems,channel state information needs to be obtained quickly and accurately at both the receiving and transmitting ends.In future mobile communications,channel estimation faces great challenges as the scale of antennas continues to increase.To this end,this thesis will focus on the channel estimation problem of mm Wave massive MIMO systems.Firstly,the relevant background knowledge and research status of this thesis are explained,and the direction of improvement is proposed.Three traditional channel estimation algorithms,the relevant theoretical knowledge of extreme learning machines and the basics of manifold learning are introduced,and the four main problems encountered in the application of manifold learning algorithms are briefly introduced.Then,a channel estimation algorithm based on single-hidden layer neural network-extreme learning machine(ELM)is proposed to make up for the shortcomings of traditional channel estimation methods.Firstly,the appropriate number of hidden layer nodes and activation functions are selected through simulation experiments to optimize the parameters of the ELM algorithm.Then,the signal received by the user terminal is used as the input of the ELM network for channel estimation,and the simulation results are comprehensively compared and analyzed under different signal-tonoise ratios,pilot costs and path numbers with traditional LS,MMSE and DNN algorithms.Then,the manifold learning algorithm is introduced on the basis of the extreme learning machine network.The algorithm realizes dimensionality reduction from high-dimensional data to lowdimensional data.By utilizing low-dimensional data for channel estimation,ELM networks are no longer limited by activation function selection.In addition,the problems of ISOMAP,LLE and NPE manifold learning methods in reducing the dimensionality of the received signal were studied,including the selection of neighbor size and eigendimension.Compared with the traditional algorithm and the channel estimation algorithm based on ELM network,the ML-ELM algorithm can better retain the essential characteristics of the data and improve the robustness and anti-interference ability of the network.Aiming at the neighbor size selection problem and the eigendimension estimation problem of high-dimensional data encountered in manifold learning applications,an improved ML-ELM algorithm is proposed.The new algorithm adopts a new neighbor distance calculation formula,and uses the adaptive weighting method to calculate the eigendimension to obtain the low-dimensional data after dimensionality reduction.Finally,the improved ML-ELM algorithm compares the performance of the traditional method under different signal-to-noise ratio and pilot overhead sentiment.Simulation results show that the improved ML-ELM algorithm can provide superior estimation accuracy in channel estimation under mm Wave massive MIMO.