Research On Beamspace Channel Estimation And Beam Selection For Mm Wave MIMO Systems

Millimeter wave(mm Wave)massive multiple-input multiple-output(MIMO)has become one of the key technologies for future wireless communication systems.Generally,each antenna is required to connect to one dedicated radio-frequency(RF)chain,resulting in high hardware complexity and energy consumption.To reduce the number of RF chains,the mm Wave MIMO system based on lens antenna array(LAA)has become a research hotspot.The traditional spatial channel can be transformed into the sparse beamspace channel,and the dominant beams can be selected for data transmission through beam selection to reduce the number of RF links.Effective beam selection usually depends on perfect channel state information,which can be obtained through channel estimation.The existing beamspace channel estimation methods usually need to pre-estimate the support,however,the inaccurate estimation of support will lead to the degradation of system performance.Especially affected by the beam squint effect in wideband systems,the support is correlated with the carrier frequency,which makes it difficult to accurately estimate the support.In addition,most of the existing beam selection methods prefer to employ the same beam selection criteria for all users without fully considering the difference of inter-user interference(IUI),which is detrimental to maintaining an attractive trade-off between computational complexity and system performance.Besides,the traditional beam search space usually contains more redundant beams with low power and a low probability of being selected,resulting in low beam search efficiency.To improve the performance of the mm Wave MIMO system based on LAA,beamspace channel estimation and beam selection will be studied in this thesis from the above perspectives.This thesis aims to estimate the beamspace channel without pre-estimating the support,design dedicated beam selection criteria for different users according to the different of IUI levels,and optimize the beam search space for the higher efficiency.The main work is as follows:(1)Considering the difficulty of accurately estimating support in beamspace channel estimation,a two-stage estimation network without estimating support is proposed.In the first stage,the estimation model based on the approximate message passing(AMP)is constructed,arising from the AMP does not need to pre-estimate the support.Then,the prior information that the beamspace channel follows the Gaussian mixture(GM)distribution is introduced into the AMP model,and the distribution parameters of the Gaussian variables are trained using the neural network.In the second stage,a residual network is designed to improve the estimation performance,to relieve the impact of channel noise and the coarse estimation error in the first stage.For wideband mm Wave MIMO systems,a multi-carrier GM threshold shrinkage function is proposed to estimate the user’s channel in all sub-carrier frequencies without considering the correlation between support and carrier frequencies,thus avoiding the estimation error caused by the inaccurate estimation of the support.The experimental results show that the proposed network reduces the normalized mean square error by 7～8 d B compared with the algorithms that require pre-estimating the support.(2)To improve the beam selection performance in narrowband mm Wave MIMO systems,a candidate beam set(CBS)containing all users’ dominant beams is constructed,considering the sparsity and power distribution of the beamspace channel.On the basis,all users can be classified into non-interference users(NIUs)and interference users(IUs)by judging whether the strongest beams in CBS overlap.For NIUs,the beams with large power will be directly assigned;for IUs,a beam selection optimization model is built based on the cuckoo search(CS)algorithm,here the beams assigned to all IUs are regarded as an optimized individual,and the dominant beams of the IUs in the CBS are used as elements of the search space to improve the search efficiency.Furthermore,the ant colony optimization(ACO)algorithm is introduced to optimize the random solution generated by Levy flight in the CS algorithm.The experimental results shows that the proposed method can improve the sum-rate performance by 3.45% compared with the existing ACO beam selection.(3)Considering the discrepancy of IUI in wideband systems,a beam selection method based on interference user grouping is proposed to reduce computational complexity.A CBS is first constructed,considering the sparsity and power distribution of the beamspace channel.Then,users can be classified into NIUs,low-interference users(LIUs)and high-interference users(HIUs)by judging the overlapping degree of dominant beams in CBS.For NIUs,the beams with large power will be directly assigned;for LIUs,a novel magnitude maximization(MM)criterion with a tabu list is designed,which can effectively tackle the drawback of assigning shared beams in the traditional MM;for HIUs,the incremental algorithm based on the sum-rate maximization criterion is employed to search for the optimal beams from CBS.The experimental results shows that the running time and sum-rate of the proposed method are 2.15% and 98.9% of the interference-aware(IA)beam selection,respectively.