A Deep Learning Based Intelligent Transceiver Structure for Multiuser MIMO.

Precoding and post-processing are necessary technical steps for information recovery of multiple-input multiple-output (MIMO) systems, which can effectively suppress interference between data streams and improve system capacity and resource utilization. However, it is not trivial to design the precoders for multiuser MIMO system and the complexity of the traditional precoding algorithms is usually very high. Deep learning sheds new light to overcome this challenge via data-driven solutions. In this paper, we study the intelligent information transmission technique for a multiuser MIMO broadcast channel network based on deep learning (DL). We propose a DL-based intelligent transceiver structure in this work. The proposed structure is composed of a DL network at the transmitter that played the role of precoder and a post-decoding DL network with a radio transformer network (RTN) at the receiver. Given the channel state information at the transmitter, the proposed intelligent transceiver is trained through the symbols drawn from a discrete constellation by decreasing the mean-squared error (MSE) loss. Simulation results show the proposed intelligent structure is capable of suppressing the inter-stream and inter-user interference adaptively through the training.