Learning Methods for MIMO Blind Detection with Low-Resolution ADCs

This paper examines the problem of blind detection in multiple-input-multiple-output (MIMO) systems with low-resolution analog-to-digital converters (ADCs) using learning approaches. Recently, the use of low-resolution ADCs has been considered an effective technique to mitigate the issue of power consumption in millimeter-wave transceivers. One serious problem caused by the low-resolution ADCs is the significant distortion of received signals, resulting in difficulty of obtaining Channel State Information (CSI) at both transmitter and receiver sides. The primary motivation of our work is that learning the input-output relation can help mitigate the impact of CSI unavailability. In both supervised and semi- supervised methods that we propose, a sequence of pilot symbol vectors is used as the initial training data for the learning task. The idea of the supervised learning method is in typical communications systems, cyclic redundancy check (CRC) is used, and thus correctly decoded symbols confirmed by CRC can be exploited as supplementary training data to improve the detection accuracy. In the semi-supervised learning method, the to-be- decoded data is exploited to help the learning process, and so no further training information is required except the pilot symbol vectors. Simulation results show that the two proposed learning methods outperform existing detection techniques.