Bayesian Based Non-stationary Statistical Channel State Estimation For Massive MIMO System

With the popularization of the fifth generation(5G)communication technology,massive MIMO draws great attention as the key technology of 5G.In order to fully exploit the superiority of massive MIMO system,statistical channel state information(S-CSI)estimation has become an essential part.However,the larger aperture of the antenna array in massive MIMO system also brings unique non-stationary characteristics to the channel,which poses a great challenge to the traditional channel estimation method based on pilot training.As a result,Bayesian approaches are in favor.Fisrt,consider a simple where the statistical channel state information of the users is assumed to be a finite set of known information.This paper poses a probabilistic structured channel model,which combins Hidden Markov Model and variational inference and considers noise.As the number of users and the number of elements in the set of statistical channel states increase,degradation,or statistical ambiguity,may occur.The problem is modeled as a semidefinite programming problem with nonconvex rank constraints and solved using a low-rank decomposition-based approach in this paper.Then,consider a more practical and complex scenario,i.e.,a channel system with dynamic birth and death evolution in the propagation path,neither has any priori information about statistical channel state.This paper proposes a infinite state statistical channel model based on explicit duration hidden semi-Markov model and hierarchical Dirichlet process.As well as a Weak-Limit Gibbs sampling inference method based full Bayesian nonparametric inference is proposed to achieve statistical channel state estimation without any priori information in non-stationary massive MIMO system.The two proposed models are compared with the existing Hidden Markov Structured Channel model by computer simulation.Both models proposed in this paper are more advantageous.The hidden Markov structured channel model only assumes that the state duration only obeys the geometric distribution,which is not applicable when facing complex duration distribution.The proposed model and method in this paper can perfectly estimate the number of channel statistical states and their durations,which is more widely applicable.