SVD Processor to Minimize the Number of Rotations in JACOBI Algorithm

Singular Value Decomposition (SVD) Processor for 8X8 MIMO precoding systems is designed. The 2-sided Cyclic Jacobi algorithm is used in our work. Here, the entire matrix is divided into matrix of 2x2 and are processed independently, which at each iteration exchange values between them. The adaptable precision controller designed, has effectively decreased the matrix size for processing and expedite the diagonalization to increase the throughput. To cover the higher precision of Eigen vector elements and Eigen values from 8x8 MIMO channel matrix, 32-bit Fixed data path representation is used. The thresholds for precision can be modified and thus the accuracy of the SVD processor is changeable according to the user requirements. The SVD processor provides throughput of 675k matrix/s in 40nm CMOS technology.