Investigation Of Channel Estimation And Signal Detection In 5G/B5G System

Massive multi-input multi-output(MIMO),as one of the key technologies in the 5th generation(5G)mobile communication system,can significantly improve the spatial resolution and spatial multiplexing capability with tens or even hundreds of antennas at the base station.Due to the huge array gain,Massive MIMO can serve multiple users simultaneously with much lower power consumption and thus achieve a huge boost in both spectral efficiency and energy efficiency.Accurate channel state information(CSI)is essential for reaping the full potential of Massive MIMO systems.It is very challenging to acquire reliable CSI in Massive MIMO systems,i.e.in time division duplex(TDD)system,downlink CSI is directly inferred from the uplink CSI by exploiting the channel reciprocity.However,the pilot contamination caused by the pilot reuse among multi-cells may seriously degrade the performance of the TDD system.In frequency division duplex(FDD)mode,where the channel reciprocity does not hold,downlink CSI is firstly estimated at the user based on the pilot signal and then fed back to the base station.The resulting pilot and feedback overhead is proportional to the number of base station antennas,which would significantly reduce the system spectral efficiency.In a word,it is very necessary to study how to effectively use limited pilot resources to enhance the accuracy of channel estimation massive MIMO system.To achieve this goal,this thesis focuses on the optimization of pilots in channel estimation.Currently,with 5G being deployed around the world,the beyond 5G(B5G)mobile communication system has attracted much attention.To achieve higher spectral efficiency,B5 G extends massive MIMO to very large scale antenna and combines it with Terahertz spectrum to explore new antenna technology.In addition,the ever increasing network complexity and emerging intelligent applications,such as autonomous vehicles,industrial automation and e-health,make it very essential to enhance the network intelligence to realize self-organizing features.Machine learning(ML),especially deep learning,is the key to realize network intelligence in B5 G.To introduce deep learning based artificial intelligence into Medium Access Control(MAC)layer,physical layer etc in the traditional communication system,the artificial neural network is trained offline with a huge amount of data,and then deployed online along with the real-time update.The offline training can effectively mine the nonlinear characteristics of the communication system in complex scenes and improve the system performance of online deployment.The application of deep learning in wireless communication is still in its infancy for now.To promote the development of B5 G intelligent signal processing,this thesis explores the deep learning based signal detection.Main contributions of this paper are described as follows:(1)The channel estimation problem with superimposed pilot pattern in the massive MIMO orthogonal frequency division multiplexing(OFDM)system is transformed into the block sparse signal reconstruction problem by exploiting the spatial common sparsity across different channels.The block sparse signal is reconstructed using the structured compressed sensing technique.According to the error upper bound of the reconstruction algorithm,a new pilot design criterion is proposed to optimize the recovery matrix.The proposed criterion makes full use of the principal angles across the subblocks in the recovery matrix,and can effectively reduce the average correlation level and eliminate the worst correlation cases simultaneously.This can ensure the reliable reconstruction performance.Simulation results show that the proposed optimized pilots outperform the random pilot schemes in terms of mean-squared error(MSE)over 3 d B.Moreover,the proposed criterion is more likely to achieve better measurement matrices than the traditional criteria.(2)For channel estimation in Massive MIMO OFDM systems with superimposed pilot pattern,two pilot design schemes are proposed to jointly optimize the pilot location and symbol.The first scheme decomposes the joint optimization of pilot location and symbol into a sequential optimization problem,in which the pilot location is firstly determined,and then the pilot symbol is assigned.In the second scheme,the interaction between the pilot location and symbol is explored to alternately optimize the pilot location and symbol.Simulation results show that the proposed two schemes outperform the random pilots in terms of MSE and bit error rate(BER).(3)The optimal pilot and feedback lengths are investigated among the three stages of downlink pilot transmission,CSI feedback and data transmission.The analytical mathematical model regarding the pilot and feedback length is established based on the approximation of quantization error covariance matrix in random vector quantization(RVQ)and the deterministic equivalent approximation of received SNR and spectral efficiency.With the established mathematical model,the optimal lengths of the pilot and feedback are obtained.Simulation results show that the deterministic equivalent approximation of receive SNR and spectrum efficiency is very tight,even when the number of transmit antennas is small.The optimal pilot and feedback lengths depend on the power of pilot,feedback and data symbols as well as the channel correlation.For strongly correlated channels,fewer pilot and feedback lengths can obtain a fairly high spectral efficiency.(4)For OFDM system,a reliable signal detection algorithm based on the attention mechanism and Bi-directional Long Short-Term Memory(Bi LSTM)network is proposed,which aims to exploit the powerful feature extraction and nonlinear mapping ability of deep learning.The proposed signal detection algorithm uses the Bi LSTM network to explore the correlation among subcarrier channels,and introduces the attention mechanism to adaptively quantify the interference level in different subcarriers.According to different interference levels,the network parameters associated with different subcarrier are trained in an end-to-end manner.Simulation results show that the proposed signal detection algorithm has better BER performance than the traditional signal detection algorithms,especially in the presence of linear and nonlinear distortion.