An Improved Gauss-Seidel Algorithm and Its Efficient Architecture for Massive MIMO Systems

Compared to most of the linear detection algorithms in massive multiple-input multiple-output (MIMO) systems, the Gauss-Seidel algorithm with initialization (GSI) promises a faster convergence rate. We improve the GSI algorithm by employing the first three items of Taylor expansion to approximate the reciprocals of diagonal entries and using the group inversion to obtain the inversion of the lower triangle matrix, which is feasible and efficient to implement. Simulation results show that the bit error rate performance of the improved GSI algorithm is similar to that of the minimum mean square error equalization for a MIMO system with 128 base station (BS) antennas and eight users. The algorithm can be applied to 64 and 32 BS antennas by employing two and three iterations, respectively. Based on the above improved algorithm, we develop a high-throughput and area-efficient architecture, which is further implemented on a Xilinx Virtex-7 XC7VX690T FPGA. The data are interleaved during the pipeline processing so that we can make the best use of the hardware. The presented architecture can achieve a maximum throughput of 127 Mb/s, which is 2.6 times higher than that of the existing architecture based on the GS algorithm. When ten received vectors are considered at a time, the throughput of the proposed architecture can reach up to 607 Mb/s, and it outperforms the optimized coordinate descent method in terms of resource utilization efficiency (throughput per look-up tables) by two times.