Pilot Assignment and Channel Estimation for User-Centric Cell-Free Massive MIMO Systems
User-centric (UC) cell-free (CF) massive multiple-input multiple-output (MIMO) systems have emerged as a promising solution for the beyond fifth generation (B5G) and the sixth generation (6G) wireless communication systems, providing enhanced coverage, capacity, and user fairness. However, the presence of challenges during channel state information (CSI) estimation, such as pilot contamination and channel aging, hinder the full potential of UC CF massive MIMO systems. Thus, accurate CSI acquisition is very essential. In this thesis, pilot assignment, channel estimation, and channel prediction are considered for UC CF massive MIMO systems with fully-digital beamforming and hybrid beamforming. This thesis consists of four main contributions as follows. First, pilot assignment design is considered for UC CF systems. Two pilot assignment schemes based on user clustering, including a low-complexity k-means clustering-based pilot assignment (KCPA) and a high-accuracy k-means clustering-based tabu-search-based pilot assignment (KCTSPA), are proposed to reduce pilot contamination in UC CF networks. The clustering-based method efficiently organizes users into groups and allocate orthogonal pilots, while tabu search optimizes the pilot assignment further. Analysis of scalability and complexity prove the deployment feasibility, and simulation results confirm the effectiveness of the proposed schemes. Second, a subspace-based semi-blind uplink channel estimation method is proposed to mitigate the pilot contamination for UC CF massive MIMO system. With the aid of uplink data signal, interference could be separated and reduced through subspace projection, without the need of channel covariance matrix (CCM) as prior knowledge. A subspace eigenvector selection algorithm based on both pilot assignment result and channel gain is designed accordingly to adapt various pilot assignment schemes. Cramér-Rao lower bound (CRLB) of the proposed estimator is derived. Simulation results validate the superior channel accuracy achieved by the proposed channel estimation method. Third, two downlink channel prediction schemes are proposed to combat the channel aging effect in high-speed mobility scenarios for UC CF networks. With sequential prediction framework, autoregressive (AR) basis expansion models (BEM)-based channel predictor is developed to enhance conventional AR approaches by a weighted Prolate spheroidal projection before prediction. With parallel prediction framework, a gated recurrent unit (GRU) transformer BEM channel predictor is proposed to further improve the accuracy, employing a more flexible data-driven manner with temporal attention mechanism. Simulation results demonstrate that the proposed methods significantly improve the channel prediction performance. Last but not least, the CSI acquisition design for UC CF systems with hybrid beamforming at access points (APs), including pilot assignment, estimation and prediction for effective channel, are developed. For pilot assignment, a low-complexity KCTSPA (LCKCTSPA) scheme is proposed by exploiting the spatial separation by radio-frequency (RF) beamformer. For channel estimation, the subspace-based semi-blind estimator is tailored for effective channel based on effective channel gain. For channel prediction, GRU transformer BEM predictor is designed for effective channel prediction. Simulation results show the improved efficiency and accuracy of the proposed methods in acquiring effective CSI for UC CF hybrid beamforming systems. To summarize, this thesis presents a comprehensive framework for CSI acquisition for UC CF massive MIMO systems, including pilot assignment, channel estimation, as well as channel prediction with/without hybrid beamforming. The proposed techniques effectively reduce pilot contamination, mitigate channel aging, and improve robustness in hybrid beamforming systems. The results and analyses presented herein serve as valuable references for researchers and engineers involved in this topic.
Also available via: Open MIND