Joint Channel Estimation And Data Detection For Cell-Free Massive MIMO Systems

The popularity of intelligent terminals and the emergence of new mobile service requirements have posed new challenges to wireless transmission rates.The development of massive Multiple Input Multiple Output(MIMO)technology has dramatically improved transmission efficiency by providing services for multiple users simultaneously at the same time-frequency resources under increasingly restricted frequency resources.In recent years,cell-free massive multi-user MIMO(MU-MIMO)systems have further improved the spectral efficiency of MIMO systems in the form of distributed deployments and are one of the strong contenders for next-generation wireless communication technologies.However,we also see that cell-free MU-MIMO systems are not yet perfect.Most studies still focus on the theoretical analysis of maximizing the signal-tointerference-noise ratio.Only a small amount of literature considers the more realistic metrics,like bit error rate.Due to the distributed deployment of the system,the performance of distributed processing schemes in densely-populated cell-free massive MU-MIMO systems still falls far short of the expected performance.Therefore,in this dissertation,we propose the Joint channel Estimation and data Detection(JED)algorithm for cell-free massive MU-MIMO systems,which provides performance far beyond linear schemes which separate channel estimation from data detection.In Chap.Ⅲ,we propose a JED algorithm for densely-populated cell-free massive multi-user MU-MIMO systems,which reduces the channel training overhead caused by the presence of hundreds of simultaneously transmitting user equipments(UEs).Our algorithm iteratively solves a relaxed version of a maximum aposteriori JED problem and simultaneously exploits the sparsity of cell-free massive MU-MIMO channels and the boundedness of QAM constellations.In order to improve the performance and convergence of the algorithm,we propose methods that permute the access point and UE indices to form so-called virtual cells,which leads to better initial solutions.We assess the performance of our algorithm in terms of root-meansquared-symbol error,bit error rate,and mutual information.We demonstrate that JED significantly reduces the pilot overhead compared to orthogonal training,enabling reliable communication with short packets to a large number of UEs.In Chap.IV,we focus on cell-free massive MU-MIMO systems with multi-antenna access points(APs)with correlated channels.Such correlation information will be exploited for channel estimation and data detection to improve the achievable data rates.However,accurately estimating the necessary covariance information results in additional pilot overhead.We study the trade-offs between error-rate performance and covariance pilot overhead by developing various JED algorithms with distinct correlation models.The proposed JED algorithms utilize Gaussian and block-sparse priors with different requirements on the number of model parameters.Our results demonstrate that the accuracy of the estimated correlation parameters and the correlation model significantly impact the achievable data rates and strongly depend on the channel’s coherence time.We also show that JED enables more than 70%to 90%of UEs to achieve 1.5 × to 7×higher data rates,respectively,compared to the commonly used centralized linear minimum-mean-squareerror(MMSE)equalizer that separates channel estimation from data detection.In Chap.V,we propose a novel soft-output JED algorithm for MU-MIMO wireless communication systems which could be applied to any channel without the need of the channel prior.Our algorithm approximately solves a maximum a-posteriori JED optimization problem using deep unfolding and generates soft-output information for the transmitted bits in every iteration.The parameters of the unfolded algorithm are computed by a hyper-network that is trained with a binary cross entropy(BCE)loss.We evaluate the performance of our algorithm in a coded MU-MIMO system with 8 basestation antennas and 4 UE antennas and compare it to state-of-the-art algorithms separate channel estimation from soft-output data detection.Our results demonstrate that our JED algorithm outperforms such data detectors with as few as ten iterations.More importantly,the trained hyper-network could be used in any given signal-to-noise ratio(SNR)range after training instead of sticking to one single SNR.