The Research And Optimization For Expectation Propagation Algorithm In MIMO Communication Systems

As the requirements of wireless communication services grow,systems with high spectral efficiency,high throughput and high stability become the development direction of wireless communication.MIMO technology can enhance the capacity of communication systems without increasing the frequency bandwidth through multiple antennas at the transmitter and receiver,and is one of the key technologies for the future wireless communication systems.The Expectation Propagation(EP)algorithm,as one type of message passing algorithms,is proven to achieve Bayes-optimal performance when used for MIMO detection.Thus EP algorithm can play an important role in MIMO communication systems.However,the complexity of the EP detection algorithm is high,especially for Turbo receivers of MIMO communication systems.Besides,the EP detection algorithm needs a large system and compression ratio threshold to achieve Bayesian optimality,and the performance will be degraded when the condition is not satisfied.In order to increase the practicality of the EP algorithm in MIMO communication systems,this dissertation focuses on the reduction of the computational complexity of the EP detection algorithm and its performance improvement when the Bayes-optimal condition is not satisfied.In the dissertation,five main works are carried out in the following three areas.Firstly,in order to reduce the computational complexity of the EP detection algorithm,two optimized detection algorithms are proposed in this dissertation.The first one is the high-efficiency EP detection algorithm(HE-EP)based on successive updating.By converting the matrix inversion in the original EP detection algorithm to iterative vector multiplication through the successive updating method,the computational complexity of linear processing can be reduced.Combined with the sorting updating method,HE-EP can improve the accuracy and convergence speed of iterative updating,and the sphere search aided method can significantly reduce the search space of the constellation points.Thus,the HE-EP also reduces the computational complexity of nonlinear processing.The second one is the low-complexity EP detection algorithm(EP-SVD)based on singular value decomposition(SVD).Through variance equalization and the preprocessing of SVD,the matrix inversion in the original EP detection algorithm is converted into the multiplication of matrices and vectors,which reduces the computational complexity order.The simulation analysis shows that both optimized detection methods can effectively reduce the computational complexity of the EP detection algorithm,where HE-EP has the same applicable scenarios as the original EP detection algorithm and EP-SVD can save more computational complexity in specific scenarios.Then,in order to improve the performance of the EP detection algorithm when the Bayes-optimal condition is not satisfied,two improved detection algorithms are proposed in this dissertation.The first one is the improved EP detection algorithm(Improved EP)based on the initial parameter selection.By analyzing the effects of initial parameters on the means and variances of the iterative approximate posterior distributions,more reasonable initial parameters are selected,and the moment matching strategy is changed to achieve the performance optimization of the EP detection algorithm.The second one is the Modified Expectation Propagation Detection(MEPD)algorithm based on neural network learning scheme.On the basis of Improved EP,in order to achieve optimization at system level,the training parameters are selected in three parts: linear processing,nonlinear processing and external information calculation,respectively,and the iterative algorithm is expanded into a neural network.The parameters are trained and optimized in the network to obtain the optimal combination of parameters.The simulation analysis shows that both improved detection algorithms can significantly improve the performance of the EP detection algorithm when the Bayes-optimal condition is not satisfied,and the MEPD has better robustness than traditional methods.Finally,for the LDPC-coded MIMO communication system,a joint detection decoding iterative receiver(JDD)based on the EP algorithm is proposed in this dissertation in order to reduce the computational complexity of the EP detection algorithm and improve the iterative update efficiency.By using a different priori feedback scheme and joint message update method,the update frequencies of the EP detection algorithm and sum-product decoding algorithm can be performed simultaneously,thus optimizing the internal structure of the EP detection algorithm,increasing the information exchange frequency between MIMO detection and LDPC decoding,and accelerating the convergence speed of the Turbo receiver.Simulation analysis shows that JDD has a faster message update frequency compared with the conventional Turbo receiver,which can reduce the number of internal iterations of MIMO detection and LDPC decoding while improving the overall receiver performance.In addition,the internal structure of the EP detection algorithm is optimized to further reduce the computational complexity of the EP detection algorithm in the proposed JDD receiver.Through the above five works,this dissertation reduces the computational complexity of the EP detection algorithm and improves its detection performance when the Bayesoptimality condition is not satisfied,and optimizes its application in Turbo receivers for MIMO communication systems.Therefore,this dissertation increases the utility of the EP algorithm in MIMO communication systems.