A New Randomized Iterative Detection Algorithm for Uplink Large-Scale MIMO Systems

Abstract

In this paper, a new randomized iterative detection algorithm (NRIDA) is proposed for uplink large-scale MIMO systems, where the random iterations in it are designed to work for the detection model (denoted by y = Hx + n) directly. Different from those traditional iterations designed for the linear system (denoted by Ax = b with A = H <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><i>H</i></sup> H and b = H <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><i>H</i></sup> y), we show that besides the complexity reduction about the matrix inversion, in the proposed NRIDA the computational complexity of matrix multiplication for the linear detection is also greatly reduced without any performance loss, thus leading to a much lower detection complexity. Meanwhile, according to convergence analysis, we demonstrate that the proposed NRIDA enjoys a globally exponential convergence performance, enabling it well suited to the various detection cases of interest. Besides, further complexity reduction and the choices of the sampling distribution in NRIDA are studied as well in full details. Moreover, in order to achieve a better detection trade-off between performance and complexity, we introduce the concept of the conditional sampling into NRIDA, which brings significant gains in both iteration convergence and efficiency. Finally, simulations with respect to the uplink large-scale MIMO detection are presented to illustrate the remarkable gains of the proposed NRIDA in both performance and complexity.