Abstract
Massive multiple-input-multiple-output (m-MIMO) systems promise substantial spectral efficiency gains but are limited by the high multiplication complexity of signal detection. With a huge number of antennas in m-MIMO, the optimum maximum likelihood (ML) detection or near-ML techniques are infeasible due to prohibitively high complexity. Conventional linear detectors, such as zero-forcing (ZF) and minimum mean-square error (MMSE), are generally among the simplest detection schemes, yet they incur significant computational costs in m-MIMO, particularly due to matched filtering, Gram matrix computation, and matrix inversion. Although prior works have proposed iterative methods to approximate matrix inversion, the complexity of Gram matrix computation remains a major limitation as the number of antennas increases. In this paper, we propose SIGN8, a novel detection algorithm based on dimensionality reduction that eliminates the need for matched filtering and Gram matrix computation. A newly designed reduction matrix is introduced, which transforms the high-dimensional channel matrix and received signal vector into reduced dimensions without involving multiplication operations. The resulting reduced matrix is inverted using a circulant approximation based on the discrete Fourier transform. SIGN8 achieves a detection complexity that depends only on the number of user ends (UEs) and remains independent of the number of base station (BS) antennas, in terms of multiplications. In contrast, the number of additions remains the same as the conventional ZF and MMSE detectors. Simulation results demonstrate that SIGN8 achieves bit error rate (BER) performance with a 1dB performance difference compared to ZF detection. The proposed method is scalable and well-suited for large-array networks, where conventional detectors face scalability bottlenecks.