Channel Estimation in IRS-Enhanced mmWave System With Super-Resolution Network

To solve the blockage effect in millimeter wave (mmWave) communication, intelligent reflecting surface (IRS) is introduced to create additional links and enhance the system performance, by properly optimizing the IRS phase shifts based on the channel state information (CSI). However, channel estimation in IRS-enhanced mmWave system is challenging, since IRS is unable to perform signal processing and the large number of reflecting elements of IRS leads to high complexity. To reduce the overhead and obtain accurate CSI, we propose a channel estimation scheme based on least square (LS) estimation with partial on-off and super-resolution (SR) network. Specifically, we switch on part of the reflecting elements and estimate the cascaded channel matrix, which can be considered as a low-resolution (LR) image with low-precision. Then it is expanded to a high-resolution (HR) image with low-precision by linear interpolation. Furthermore, we feed this HR image into an SR network to improve the estimation accuracy. Numerical results demonstrate the advantages of our proposed SR channel estimation compared with benchmark schemes.

I. Introduction

Millimeter wave (mmWave) communication, with frequency ranging from 30~300GHz, is considered as a promising method to meet the increasing capacity demand in the fifth-generation system [1]. However, such high carrier frequency leads to severe path loss and makes the mmWave signal vulnerable to blocking obstacles. Meanwhile, mmWave communication inevitably leads to huge energy consumption and hardware complexity. Intelligent reflecting surface (IRS), which smartly transforms the stochastic wireless channels into a software-defined environment, has been proposed as a promising new paradigm to solve these problems efficiently and achieve smart wireless system [2]–[17]. Generally speaking, IRS can be utilized to create additional links and enhance the system performance without incurring huge energy consumption [6]–[17]. Specifically, IRS is a planar array consisting of a large number of reconfigurable low-cost passive elements, which are capable of independently changing the phase shifts of the incident signal with low power consumption.