Fast Time-Varying mmWave Channel Estimation: A Rank-Aware Matrix Completion Approach
Tianyu Jiang 1 , Yan Yang 1 , Hongjin Liu 2 , Runyu Han 1 , Bo Ai 1 , and Mohsen Guizani 3 · 2025
Read Fast Time-Varying mmWave Channel Estimation: A Rank-Aware Matrix Completion Approach by Tianyu Jiang 1 , Yan Yang 1 , Hongjin Liu 2 , Runyu Han 1 , Bo Ai 1 , and Mohsen Guizani 3 (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2511-00607.txt.
Introduction
Tianyu Jiang, Yan Yang, Hongjin Liu, Runyu Han, Bo Ai, and Mohsen Guizani attack channel estimation for fast-moving millimeter-wave systems. A hybrid MIMO receiver must learn a high-dimensional channel that changes quickly as users move. The channel matrix is both low-rank and sparse, and the authors exploit each property in a separate phase of one compressed sensing framework.
First, an observation matrix is completed with a robust rank-one completion algorithm that updates the matrix and its rank together, no prior rank knowledge needed. Second, a rank-aware matching pursuit recovers the sparse channel with low latency. An autoregressive model tracks rank changes across time, exploiting their correlation between nearby instants. Simulations confirm state-of-the-art recovery accuracy with theoretical guarantees, a practical step for high-mobility millimeter-wave links.
Catalog record
| CREATOR | Tianyu Jiang 1 , Yan Yang 1 , Hongjin Liu 2 , Runyu Han 1 , Bo Ai 1 , and Mohsen Guizani 3 |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 31 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2511-00607
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Fast Time-Varying mmWave Channel Estimation: A Rank-Aware Matrix Completion Approach Thanks: This work was funded by the National Nature Science Foundation of China under Grant 62271041 and in part by the Hongguoyuan Science and Technology Project under Grants WX-2025-0803, WX-2024-0718, and WX-2024-0805. Tianyu Jiang 1 , Yan Yang 1 , Hongjin Liu 2 , Runyu Han 1 , Bo Ai 1 , and Mohsen Guizani 3 Abstract We consider the problem of high-dimensional channel estimation in fast time-varying millimeter-wave MIMO systems with a hybrid architecture. By exploiting the low-rank and sparsity properties of the channel matrix, we propose a two-phase compressed sensing framework consisting of observation matrix completion and channel matrix sparse recovery, respectively. First, we formulate the observation matrix completion problem as a low-rank matrix completion (LRMC) problem and develop a robust rank-one matrix completion (R1MC) algorithm that enables the matrix and its rank to iteratively update. This approach achieves high-precision completion of the observation matrix and explicit rank estimation without prior knowledge. Second, we devise a rank-aware batch orthogonal matching pursuit (OMP) method for achieving low-latency sparse channel recovery. To handle abrupt rank changes caused by user mobility, we establish a discrete-time autoregressive (AR) model that leverages the temporal rank correlation between continuous-time instances to obtain a complete observation matrix capable of perceiving rank changes for more accurate channel estimates. Simulation results confirm the effectiveness of the proposed channel estimation frame and demonstrate that our algorithms achieve state-of-the-art performance in low-rank matrix recovery with theoretical guarantees. I Introduction By leveraging abundant spectrum resources, millimeter wave (mmWave) communications emerge as a key enabler for future wireless networks to meet ever-increasing capacity demands [ 1 ] . To combat the high propagation losses experienced in mmWave bands, a hybrid analog-digital (HAD) beamforming architecture is a low-complexity solution that can offer a high-gain directional beam, substantially addressing this radio propagation limitation. In mobile mmWave communication scenarios, the transmitter needs to form a beam that adapts to the receiver’s movement so that the beam can maintain continuous beam alignment in the receiver’s direction to ensure reliable transmission. To do this, the acquisition of precise channel estimates is essential. However, despite notable theoretical advancements, practical mmWave MIMO communication systems inherently involve high-dimensional channel matrices. In the case of high mobility, a complete measurement of channel state information (CSI) is often not practical; real-time measurement data is often outdated, even corrupted with large errors. Thus, accurate channel estimation remains difficult, especially in rapid channel variations under high mobility, such as the sixth-generation (6G) communication systems for vehicle-to-everything (V2X) communications, where terminals move at a speed of 120 km/h or even higher. An alternative solution is to exploit the inherent sparsity of mmWave MIMO channels. So far, this property has been exploited in the literature and identified as low-rank matrix recovery/completion methods in compressed sensing [2-5]. The basic idea is to formulate the problem as a matrix rank minimization problem and solve it efficiently by nuclear-norm minimization. Theoretically, a low-rank channel matrix can be stably recovered from incomplete, inaccurate, and noisy observations by solving a rank minimization problem, if the observation matrix satisfies a restricted isometry property (RIP) [2]. Although classical orthogonal matching pursuit (OMP) is effective against the absence of small Gaussian noise in the data, it suffers from a high computational cost in sparse recovery that may render it impractical for high-dimensional mmWave channel matrices. Various adaptive techniques have been proposed to address the challenges posed by time-varying mmWave channel estimation. Specifically, sparse Bayesian learning (SBL) based channel estimation in mmWave hybrid MIMO OFDM systems was proposed in [ 4 ] and employed to track the propagation paths. It is shown in [ 5 ] that the authors proposed a novel SBL-based group-sparse paradigm for mmWave MIMO OFDM channel estimation, and developed a low-complexity version to reduce the computational cost of the high-dimensional channel matrix. Most recently, this method could be viewed as a way to sense and recover a low-rank channel matrix [7]. Nevertheless, existing methods predominantly focus on static sparsity patterns, leaving out the effect of rank uncertainty in high-mobility scenarios [ 8 , 9 ] . In brief, the contributions of this paper are as follows: • By incorporating an \ell_{1} -norm regularization, we devise a robust rank-one matrix completion (R1MC) algorithm to achieve high-precision observation matrix completion and explicit rank estimation from incomplete or corrupted observations. • We propose a novel compressed sensing (CS) framework for high-dimensional mmWave MIMO channel estimation. The channel reconstruction is accordingly formulated as a sparse matrix recovery problem. • To achieve accurate channel estimation, we develop an adaptive rank estimation framework integrated with a discrete-time autoregressive (AR) model to capture the channel dynamics, enabling temporal tracking of rank variations without prior knowledge. The rest of the paper is organized as follows: Section II introduces the system model, the problem formulation, and a sparse representation of the channel in the angular domain. In Section III, a two-phase CS channel estimator is proposed. Simulation results are presented in Section IV, and the conclusion is provided in Section V. II System Model Fig. 1: mmWave signal transmission model and time-varying channel estimation from incomplete observations. II-A Channel Model We consider a single-user mmWave MIMO system comprising a base station (BS) and a mobile station (MS). The BS employs {{\mathcal{N}}_{\text{BS}}} antennas with {{\mathcal{M}}_{\text{BS}}} RF chains, while the MS is equipped with {{\mathcal{N}}_{\text{MS}}} antennas and {{\mathcal{M}}_{\text{MS}}} RF chains. Both BS and MS adopt a hybrid architecture. Fig. 1 illustrates the end-to-end transmission structure, where \widetilde{\textbf{Y}}_{t} and \widehat{\textbf{Y}}_{t} represent incomplete- and complete observation matrix at time instance t , corresponding to coarse \widetilde{\textbf{H}}_{t} and refined \widehat{\textbf{H}}_{t} , respectively. With a geometric wideband mmWave channel model, the beam space MIMO representation of the delay- d MIMO channel matrix, {{\textbf{H}}}(d) , can be written as [6] \displaystyle{{\textbf{H}}}[d]=\, \displaystyle\sqrt{\frac{{\mathcal{N}}_{{\text{BS}}}{\mathcal{N}}_{\text{MS}}}{{\rm L}_{\rm P}}}\sum_{\ell=1}^{\mathcal{L}}\sum_{\kappa_{\ell}=1}^{{\mathcal{K}}_{\ell}}\alpha_{\kappa_{\ell}}\varrho\left(d{\mathcal{T}}_{\rm s}-\tau_{\ell}-\tau_{\kappa_{\ell}}\right) \displaystyle\times{\textbf{a}}_{\text{MS}}\left(\theta_{\ell}-\mathcal{\vartheta}_{\kappa_{\ell}}\right){\textbf{a}}^{\dagger}_{\text{BS}}\left(\phi_{\ell}-\varphi_{\kappa_{\ell}}\right), (1) where \mathcal{L} is the number of paths, \alpha_{\kappa_{\ell}} is the complex gain associated with the \ell th cluster, and each cluster is assumed to contribute \kappa_{\ell} paths between the BS and the MS; \varrho(\cdot) is a pulse-shaping filter for \mathcal{T}_{\text{s}} -spaced signaling generated at d{\mathcal{T}}_{\rm s}-\tau_{\ell}-\tau_{\kappa_{\ell}} seconds; {\theta}_{\ell}\in[0,\pi] and {\phi}_{\ell}\in[0,\pi] are the associated azimuth AoAs and azimuth AoDs respectively; \vartheta_{{\kappa_{\ell}}} denotes the offset of the \kappa th path compared to the mean AoDs with the \ell th cluster; \varphi_{\kappa_{\ell}} denotes the offset of the \kappa th path compared to the mean AoAs with the \ell th cluster; Superscripts (\cdot)^{\dagger} stands for the Hermitian transpose. Suppose a uniform linear array (ULA) antenna is used, the beamforming gain at the BS and MS is respectively given by \displaystyle{{\displaystyle\textbf{a}}_{\text{MS}}}({{\theta_{\ell}}})=\frac{1}{\sqrt{{{\mathcal{N}}_{\text{MS}}}}}{[1,{{e}^{j\frac{2\pi}{\lambda_{c}}\delta\sin{{\theta_{\ell}}}}},\ldots,{{e}^{j({{\mathcal{N}}_{\text{MS}}-1)\frac{2\pi}{{\lambda_{c}}}\delta\sin{{\theta_{\ell}}}}}]}^{{T}}} (2) \displaystyle{{\displaystyle\textbf{a}}_{\text{BS}}}({{\phi_{\ell}}})=\frac{1}{\sqrt{{{\mathcal{N}}_{\text{BS}}}}}{{[1,{{e}^{j\frac{2\pi}{\lambda_{c}}\delta\sin{{\phi_{\ell}}}}},\ldots,{{e}^{j({{\mathcal{N}}_{\text{BS}}}-1)\frac{2\pi}{\lambda_{c}}\delta\sin{{\phi_{\ell}}}}}]}^{{T}}}, where {{\textbf{a}}_{\text{MS}}} and {{\textbf{a}}_{\text{BS}}} denote the antenna array response vectors of the MS and BS, respectively. Here, Superscripts (\cdot)^{T} stands for the transpose; \lambda_{c} is the signal wavelength, and \delta is the distance between neighboring antenna elements, which is usually set to \delta=\lambda_{c}/2 . As a result, the delay- d channel model in (1) can be represented as {\textbf{H}}=\sum_{d=0}^{D-1}\textbf{{H}}[d]e^{-j2\pi d}. (3) This clustered channel model exploits the fact that each cluster comprises multiple paths. It leaves out the low-rank structure inherently induced by the clustering of paths. It verifies that the considered time-varying channel matrix \textbf{H}_{t} has a rank equivalent to