Channel Estimation And Performance In A Multi-cell Massive MIMO System

A wireless communication system with high transmission rate is urgently required to satisfy various applications requirements.A massive Multiple-Input,Multiple-Output(MIMO)system is one of promising candidates because of its high spectral efficiency.It has been shown that a massive MIMO system can provide high total throughout with simple signal processing if the accurate Channel State Information(CSI)is available,while the performance of the system is largely damaged if the CSI is not available.Thus,the accuracy of channel estimation has great impact on the communication quality of the system.Conventionally,acquisition of the CSI is based on the training-based channel estimation method.It has been proven that the accuracy of the method depends on the training-sequences.Usually,the orthogonal training-sequences can provide high quality estimation.However,for a multi-cell,massive MIMO system,this orthogonality will consume a lot of resources.Thus,ones have to consider non-orthogonal training sequences.First,this thesis will deal with design of the non-orthogonal training sequences.Assume that the system has L cells,each cell has a base station(BS)with M antennas and K users with single antenna.The analytic formula of calculating the error variance of the MMSE estimator and its lower bound are derived conditioning on that the energy of the training sequences are limited to a constant.Moreover,a design criterion of non-orthogonal training sequences which can improve the performance of the MMSE estimator is given.Results show that,if the energy of training sequence is fixed,the orthogonal training sequence is optimal.Furthermore,increasing training times can not improve performance of the MMSE estimation.When the training times are less than the number of cells,no matter how the training sequence is designed or the signal-to-noise ratio(SNR)is increased,the MMSE estimation error is bounded away from zero,error floor exists.According to the design criterion above,the correlations of these training sequences should be as weak as possible.Thus,design of training sequences can be implemented by using line packing on a complex Grassmannian Manifold.Finally,these conclusions above are confirmed by simulations.Secondly,for the considered system,this thesis analyze the effects of pilot contamination on error performance.The formula of the pair-wise error probability is given.This formula shows that the main reason which causes the error floor is due to the transmitted signal's “center-overturned”.That is,after operations of the Mean Minimal-Square Error(MMSE)channel estimator and the MMSE decoder,the center of received signals is far away from the transmitted signal and located on the region of making wrong decision.Moreover,the larger the SNR is,the farther away from the transmitted signal the center is.Meanwhile,it is proved that the probability of “centeroverturned” is just the error floor.Thirdly,this thesis introduce an equivalent transmit-receive equation for the system.Based on this equation,we define the “indicated numbers of the system”,and show that the performance of the system depends on these parameters.In fact,the larger the parameters are,the better the performance is.Furthermore,if the indicated numbers of the system are all larger than zero,the error floor will disappear if M goes to infinity,even though the pilot signals are fully reused.These conclusions above are also confirmed by simulations.