Efficient matrix inversion architecture for linear detection in massive MIMO systems

Resulted by the hundreds of antennas at the base-station (BS) side, the dimension of matrices involved in linear detection for massive multiple-input multiple-output (MIMO) uplink systems increases drastically. Being an indispensable part of linear detection, the matrix inversion suffers a lot from the huge matrix size of massive MIMO, and therefore becomes inefficient for realization. By achieving good tradeoff between complexity and performance, approximate matrix inversion via Neumann series now boosts one promising solution for linear detection of massive MIMO systems. In this paper, an efficient hardware architecture for approximate matrix inversion is proposed. This architecture is hardware efficient and suitable for various applications with different approximation precisions. FPGA implementation results of matrix inversion for 4 × 32 massive MIMO system have shown that the proposed architecture can achieve 56.7% higher frequency with only 58.9% hardware resources of its existing counterparts. Look-ahead transformations which can make the proposed architecture more suitable for high speed applications are also mentioned.