Tensor-based modeling/estimation of static channels in IRS-assisted MIMO systems

Technical Summary: Tensor-based Modeling and Estimation of Static Channels in IRS-assisted MIMO Systems

Problem Statement
The paper addresses the challenge of channel estimation in Intelligent Reflecting Surface (IRS)-assisted Multiple Input Multiple Output (MIMO) systems. In such networks, the receiver must estimate the cascaded channel (User Equipment to IRS to Base Station) using pilot signals reflected by the IRS. Existing methods are often categorized into structured techniques (exploiting parametric geometric models) and unstructured techniques. While previous works have utilized tensor approaches for specific scenarios (e.g., mmWave systems or semi-blind data detection), there is a need for robust frameworks that exploit the specific geometric structure of static IRS-assisted channels to improve estimation accuracy without incurring prohibitive computational costs.

Methodology
The authors propose a new signal modeling framework that formulates the channel estimation problem as a third-order Tucker tensor model. The core of the methodology involves the following steps:

1. System Model: The system considers an uplink scenario with a Base Station (BS) equipped with $M$ antennas, a User Equipment (UE) with $Q$ antennas, and an IRS with $N$ passive elements. The channels (BS-IRS and IRS-UE) are modeled using a multipath geometric model that includes both Line-of-Sight (LOS) and Non-Line-of-Sight (NLOS) components, characterized by steering vectors and path loss/fading coefficients.
2. Tensor Formulation: By vectorizing the received pilot signals and utilizing properties of the Kronecker and Khatri-Rao products, the authors derive a linear relationship between the received signal and the concatenated channel parameters. This relationship is restructured into a third-order Tucker tensor model:
   $$\mathcal{R} \approx \mathcal{F} \times_1 \mathbf{A}_{rx} \times_2 \mathbf{A}^*_{tx} \times_3 \mathbf{P}_B$$
   Here, $\mathcal{R}$ represents the received data tensor, $\mathbf{A}_{rx}$ and $\mathbf{A}_{tx}$ are the steering matrices for the BS and UE, $\mathbf{P}_B$ contains the IRS geometry information, and $\mathcal{F}$ is the core tensor representing the path coefficients.
3. Proposed Algorithms: Two algorithms are derived to solve the parameter estimation problem (recovering the factor matrices and the core tensor):
   • Tucker-ALS (Alternating Least Squares): An iterative algorithm that alternately updates the estimates of $\mathbf{A}_{rx}$, $\mathbf{A}_{tx}$, $\mathbf{P}_B$, and the core tensor vector $\mathbf{f}$ by solving least-squares sub-problems until convergence.
   • Tucker-HOSVD (High-Order Singular Value Decomposition): A closed-form solution that computes the factor matrices by performing truncated SVDs on the mode-unfoldings of the received tensor, followed by a pseudo-inverse step to estimate the core tensor.

Key Contributions

• Novel Signal Model: The paper introduces a 3rd-order Tucker tensor model specifically tailored for static IRS-assisted MIMO channels, explicitly exploiting the geometric structure of the cascaded channel.
• Dual Algorithmic Approach: The authors provide two distinct solutions: an iterative ALS method and a non-iterative HOSVD method, offering flexibility in implementation based on convergence requirements versus computational speed.
• Complexity Analysis: The study includes a detailed computational complexity analysis comparing the proposed methods against the classic Least Squares (LS) and the state-of-the-art Khatri-Rao Factorization (KRF) algorithm.

Simulation Results
Numerical simulations evaluate the performance using the Normalized Mean Square Error (NMSE) of the estimated concatenated channel:

• Performance Gain: The proposed tensor-based methods (both ALS and HOSVD) significantly outperform the classic LS and the KRF benchmark. Specifically, the proposed methods achieve approximately 10 dB improvement over LS and 5 dB improvement over KRF.
• Algorithm Comparison: The iterative ALS algorithm demonstrates slightly better performance (approx. 1 dB gain) compared to the closed-form HOSVD, though both are superior to benchmarks.
• Convergence Behavior: The ALS algorithm requires more iterations in low Signal-to-Noise Ratio (SNR) regions or when the number of propagation paths increases. Conversely, increasing the number of IRS elements ($N$) reduces the number of iterations required for convergence.
• Computational Cost: Despite the performance gains, the computational complexity of the proposed ALS and HOSVD schemes is comparable to that of the benchmark LS and KRF algorithms. The authors note that the dominant cost in all schemes remains the initial LS step or the SVD/pseudo-inverse operations, which scale similarly across methods.

Significance and Claims
The paper claims that by exploiting the algebraic tensor structure of the combined channel, the proposed methods achieve superior estimation accuracy without increasing computational complexity relative to existing state-of-the-art techniques. The authors position this work as a step toward efficient channel estimation for beyond 5G (B5G) and 6G networks, where IRS deployment is expected to be critical. The work does not claim to solve time-varying channel issues (referencing their previous work for that) but focuses on optimizing the estimation of static channels to enable effective IRS phase-shift configuration.