Knowledge Distillation-Aided End-to-End Learning for Linear Precoding in Multiuser MIMO Downlink Systems With Finite-Rate Feedback

We propose a deep learning-based channel estimation, quantization, feedback, and precoding method for downlink multiuser multiple-input and multiple-output systems. In the proposed system, channel estimation and quantization for limited feedback are handled by a receiver deep neural network (DNN). Precoder selection is handled by a transmitter DNN. To emulate the traditional channel quantization, a binarization layer is adopted at each receiver DNN, and the binarization layer is also used to enable end-to-end learning. However, this can lead to inaccurate gradients, which can trap the receiver DNNs at a poor local minimum during training. To address this, we consider knowledge distillation, in which the existing DNNs are jointly trained with an auxiliary transmitter DNN. The use of an auxiliary DNN as a teacher network allows the receiver DNNs to additionally exploit lossless gradients, which is useful in avoiding a poor local minimum. For the same number of feedback bits, our DNN-based precoding scheme can achieve a higher downlink rate compared to conventional linear precoding with codebook-based limited feedback.

I. Introduction

In multiuser multiple-input and multiple-output (MU-MIMO) downlink systems with frequency-division duplexing, the transmitter cannot track the downlink channel directly. Thus, finite-rate feedback (or limited feedback) is widely used to achieve partial channel state information (CSI) at the transmitter (CSIT) [1]–[3]. The performance of such systems is significantly affected by the CSIT accuracy, which is determined by the number of feedback bits. Therefore, state-of-the-art commercial mobile devices for cellular networks do not actively use MU-MIMO downlink applications, despite the 4G long-term evolution (LTE)-advanced standards defining up to four layers, and the recently released 5G new radio (NR) standards defining up to 12 layers for these applications [4]. This is mainly because the achievable rate is not sufficiently high to compensate for the cost of the uplink resources that convey the feedback bits.