Favorable-Propagation-Exploited Variational Inference for Massive MIMO Detection

In this paper, we investigate the massive MIMO detection under the framework of mean-field variational inference (VI), which leads to a better detection trade-off between performance and complexity. First of all, by fully taking advantages of the favorable propagation characteristic of massive MIMO, the favorable-propagation-exploited variational inference (FPE-VI) algorithm is proposed for the low-complexity detection. Secondly, with respect to the system with
$K$
transmitting and
$N$
receiving antennas, the linear version of the FPE-VI detection is studied in detail, where its convergence is ensured when
$N / K > {1}/ {\left(\sqrt{2} - 1\right)^{2}}$
. Thirdly, by examining the evidence lower bound (ELBO) of the proposed FPE-VI, further optimization via the application of discrete Gaussian distribution is presented for extra performance gain. Finally, all the related theoretical analysis and the improved performance-complexity trade-off of FPE-VI are demonstrated by numerical results.