Design And Implementation Of Matrix Inversion Based On USRP RIO In MIMO Systems

The explosive growth of wireless devices puts forward high capacity demands in future wireless communication systems.MIMO systems have a great appeal to both the academy and industry since it can significantly improve spectrum efficiency and can support more connected users by leveraging the space division multiple address technology.However,MIMO signal processing contain large-scale matrix calculations,which seriously decreases real-timelines,thereby degrading the performance of MIMO systems.Matrix inversion is one of the most frequent matrix calculations in MIMO system.Due to the huge complexity of matrix inversion,accelerating matrix inversion is of the great significance to practical MIMO systems.The conventional method of matrix inversion is direct to compute matrix invision However,the method will need huge latency to accomplish the matrix inversion.To overcome above problems,this paper first study channel characteristics in MIMO systems and explore the method of approximate matrix inversion based on Neumann series.This method extends the inverse of a matrix to a series of matrix-vector multiplications by leveraging the characteristics of diagonal dominance of the MIMO channel matrix.With the condition of the Neumann series convergence,the inverse matrix can be obtained by limited number of series to calculating.However,the speed of the series of convergence is very slow with using the initial matrix diagonal matrix as Neumann series.To overcome this problem,this paper proposes a modified Neumann series inversion architecture,and the architectures can use less number of iterations to obtain inverse matrix by optimizing the selection of initial matrix method.Based on USRP platform,we realize the approximate matrix inversion and describes the design method of each module in detail.Then,this paper verifies the feasibility of the matrix inversion in the practical MIMO system,and the results show that the module can work.Finally,this paper evaluates the performance of accelerating matrix inversion.The results show that the latency of our method is less than direct matrix inversion.In addition,our method can get 3.5 dB improvement in SINR.