Improving Signal Detector by Precoding in Uplink Multiuser MIMO System

The condition number of the channel matrix greatly affects the performance of existing signal detectors in a multiple-input multiple-output (MIMO) system. When the channel matrix is ill-conditioned, existing sub-optimal and optimal detectors exhibit poor performances. In our early work, we have proposed a learning to learn iterative search algorithm (LISA), in which the MIMO detection problem is modeled as a decision problem over tree. The performance of LISA is close to the maximum likelihood detector (MLD) in some scenarios, but degenerates when the MIMO channel is with large condition number. In this paper, we propose to remedy this problem by incorporating a precoding scheme at the transmitter side to reduce the condition number of the channel matrix. Furthermore, we improve LISA by adding a parallel signal recovery module which is named I-LISA. Extensive experimental results show that incorporating the precoding scheme improves not only the performance of LISA, but also existing signal detectors including MLD. Extensive experimental results show that I-LISA with the proposed precoding scheme achieves better performance than MLD for QPSK and 16QAM, and for 64QAM at low signal-to-noise ratio (SNR) regimes.