[Paper Review] High-Resolution Channel Estimation for Intelligent Reflecting Surface-Assisted MmWave Communications

Chenglu Jia, Julian Cheng+2 more|arXiv (Cornell University)|Jun 21, 2020

Advanced Wireless Communication Technologies26 references4 citations

TL;DR

This paper proposes a two-step cascaded channel estimation protocol for intelligent reflecting surface (IRS)-assisted mmWave MIMO communications, leveraging sparse signal recovery via an adaptive grid matching pursuit (AGMP) algorithm to achieve high-resolution CSI with low complexity. Simulation results show it significantly outperforms beam training-based schemes and approaches perfect CSI performance in spectral efficiency and NMSE.

ABSTRACT

In this paper, we study the high-resolution channel estimation problem for intelligent reflecting surface (IRS)-assisted millimeter wave (mmWave) multiple-input-multiple-output (MIMO) communications, which is a prerequisite to guarantee further high-rate data transmission. Considering the typical sparsity of mmWave channels, we formulate the cascaded channel estimation problem from a sparse signal recovery perspective, and then propose a novel two-step cascaded channel estimation protocol to estimate the cascaded user-IRS-base station channel with high-resolution for IRS-assisted mmWave MIMO communications. More specifically, the first step is to estimate the coarse angular domain information (ADI) and further establish the robust uplink by beam training. In the second step, by exploiting the coarse ADI, an adaptive grid matching pursuit (AGMP) algorithm is proposed to estimate the high-resolution cascaded channel state information (CSI) with low complexity. Simulation results verify that the proposed two-step channel estimation protocol significantly outperforms the state-of-the-art scheme, i.e., beam training based channel estimation, and meanwhile can reap near-optimal system performance achieved by perfect CSI.

Motivation & Objective

• Address the challenge of high-resolution cascaded channel state information (CSI) estimation in IRS-assisted mmWave MIMO systems, where conventional methods fail due to hardware constraints and high-dimensional channel matrices.
• Overcome the limitations of beam training-based schemes, which suffer from coarse angular resolution and poor accuracy due to finite phase shifts in IRS elements.
• Enable accurate CSI acquisition for system optimization and high-rate data transmission by exploiting the sparsity of mmWave channels.
• Develop a low-complexity, high-accuracy channel estimation framework that leverages coarse angular domain information (ADI) as prior knowledge for improved performance.

Proposed method

• Formulate the cascaded channel estimation problem as a sparse signal recovery task, exploiting the inherent sparsity of mmWave channels.
• Use a two-step protocol: first, perform beam training to estimate coarse angular domain information (ADI) for robust uplink training and as prior knowledge.
• In the second step, apply an adaptive grid matching pursuit (AGMP) algorithm that refines the coarse ADI to estimate high-resolution CSI with reduced computational complexity.
• Design the AGMP algorithm to adaptively refine the grid resolution based on the coarse ADI, improving estimation accuracy while minimizing pilot overhead.
• Utilize a dictionary with adjustable resolution (denoted as $ ilde{G} $) to balance accuracy and complexity, with performance evaluated across different resolutions.
• Integrate the estimated high-resolution CSI into system performance evaluation for spectral efficiency (SE) and normalized mean square error (NMSE).

Figure 1: Illustration of the considered IRS-assisted mmWave MIMO system.

Experimental results

Research questions

• RQ1 How can high-resolution cascaded CSI be efficiently estimated in IRS-assisted mmWave MIMO systems under hardware constraints of finite phase resolution?
• RQ2 To what extent does using coarse ADI from beam training improve the accuracy of high-resolution channel estimation?
• RQ3 Can a low-complexity algorithm like AGMP achieve near-optimal performance compared to perfect CSI in terms of NMSE and spectral efficiency?
• RQ4 How does the dictionary resolution $ ilde{G} $ affect the trade-off between estimation accuracy and computational complexity?
• RQ5 What is the performance gap between beam training-based estimation and the proposed two-step protocol in practical SNR regimes?

Key findings

• The proposed two-step channel estimation protocol achieves significantly lower normalized mean square error (NMSE) than the beam training-based benchmark, especially at low SNR, due to improved angular resolution.
• Even with only 4 iterations, the AGMP algorithm achieves near-optimal spectral efficiency (SE), approaching the upper bound of perfect CSI, particularly when $ ilde{G} = 3 $.
• The NMSE performance improves with increasing dictionary resolution $ ilde{G} $, and SE converges to the perfect CSI upper bound when $ ilde{G} = 3 $, indicating this is the optimal resolution under the given system model.
• The beam training-based scheme suffers from poor NMSE performance even at high SNR due to limited phase resolution and beam misalignment, highlighting the necessity of high-resolution estimation.
• The proposed protocol maintains strong performance at low SNR, demonstrating robustness to noise and multipath effects, thanks to the beamforming gain provided by IRS.
• The integration of coarse ADI as prior knowledge enables the AGMP algorithm to achieve high accuracy with low complexity, making it suitable for practical IRS-assisted mmWave systems.

Figure 2: The frame structure of the considered IRS-assisted mmWave system.