⚡ electrical engineering

PARAFAC-Based Time-Varying Channel Estimation for IRS-Aided Communications

This paper proposes a 3rd-order PARAFAC tensor model solved via an iteratively ALS algorithm to achieve enhanced time-varying channel estimation performance in IRS-aided communication systems without increasing computational complexity.

Original authors: Kenneth B. A. Benício, André L. F. de Almeida, Bruno Sokal, Fazal-E-Asim, Behrooz Makki, Gabor Fodor

Published 2026-05-29

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Kenneth B. A. Benício, André L. F. de Almeida, Bruno Sokal, Fazal-E-Asim, Behrooz Makki, Gabor Fodor

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Big Picture: A Smart Mirror in a Noisy Room

Imagine you are trying to have a conversation with a friend (the User) in a large, echoey room. The person you are talking to (the Base Station) is on the other side of a thick wall, so you can't hear them directly.

To solve this, someone installs a giant, high-tech Smart Mirror (the IRS or Intelligent Reflecting Surface) on the wall. This mirror doesn't just reflect light; it can instantly change the angle of the sound waves hitting it to bounce your voice perfectly toward your friend.

The Problem:
To make the mirror work, the Base Station needs to know exactly how the sound travels from you to the mirror, and then from the mirror to the Base Station. This is called Channel Estimation.

The path from you to the mirror is tricky because you might be moving around (changing the sound).
The path from the mirror to the Base Station is usually steady because both are bolted to the wall (static).

The paper proposes a new, smarter way to figure out these sound paths without needing to shout louder or use more energy.

The Old Way vs. The New Way

The "Old" Way (Least Squares):
Imagine trying to figure out the shape of a puzzle by looking at the pieces one by one and guessing. It works, but it's slow and often misses the big picture, especially if there is background noise.

The "Competitor" Way (Khatri-Rao Factorization):
This is like looking at the puzzle pieces and realizing, "Hey, these pieces fit together in a specific pattern!" It's better than the old way, but it still treats the puzzle as a flat, 2D image.

The New Way (The Paper's Solution):
The authors realized that the data coming from the mirror isn't just a flat list of numbers; it has a hidden 3D structure (like a Rubik's cube or a stack of photos).

They use a mathematical tool called PARAFAC (which sounds fancy, but think of it as a "3D Puzzle Solver").

The Analogy: Imagine you have a stack of photos (Time), a grid of pixels (Space), and different angles (Frequency). Instead of flattening this stack into a long line of pixels (which loses information), the new method keeps it as a 3D cube.
The Trick: Because the mirror-to-Base Station path is steady (static), the "3D Cube" has a very specific, rigid shape. The new algorithm exploits this rigidity. It's like knowing that a specific type of Lego brick always connects in a certain way, allowing you to solve the puzzle much faster and more accurately.

How It Works (The "Magic" Step)

The paper uses an algorithm called ALS (Alternating Least Squares).

Think of it like tuning a radio: You turn the dial for the "Time" setting, then the "Space" setting, then the "Frequency" setting. You keep going back and forth, making tiny adjustments each time.
The Result: With every pass, the picture gets clearer. Eventually, the picture is so sharp that you know exactly where every sound wave went.

What the Results Show

The authors ran computer simulations (virtual experiments) to test their idea against the other methods.

Clearer Picture (Better Accuracy): Their method made the "picture" of the channel about 7 to 10 decibels clearer than the other methods. In real life, this means the connection is much more stable, and you can hear your friend perfectly even if the room is noisy.
No Extra Cost (Efficiency): Usually, getting a clearer picture requires a super-computer. Surprisingly, this new method didn't require more computing power than the others. It just used the existing data more cleverly.
Speed: The "tuning" process (the iterations) usually settles down very quickly (around 10 to 80 steps), unless the environment is extremely complex or very noisy.

The Bottom Line

This paper introduces a smarter way to "listen" to the environment in future 6G networks. By treating the signal data as a 3D object instead of a flat list, and by realizing that the connection between the mirror and the tower is steady, they can figure out the network's layout much more accurately.

In short: They found a way to solve a complex puzzle by noticing a pattern everyone else was ignoring, resulting in a clearer connection without needing a more powerful computer.

Technical Summary: PARAFAC-Based Time-Varying Channel Estimation for IRS-Aided Communications

Problem Statement
The paper addresses the challenge of channel estimation in Intelligent Reflecting Surface (IRS)-assisted Multiple-Input Multiple-Output (MIMO) systems, specifically focusing on scenarios where channel parameters are time-varying. In such systems, the IRS consists of passive elements and cannot perform signal processing; thus, channel estimation must be conducted at the network endpoints (Base Station or User Equipment) using pilot signals. While previous works have addressed static or quasi-static channels, this paper targets a more realistic scenario where the Base Station (BS) and IRS are in fixed positions (resulting in a static BS-IRS channel), while the link between the IRS and the User Equipment (UE) is time-varying due to mobility. The goal is to estimate the concatenated channel parameters efficiently without increasing computational complexity compared to existing methods.

Methodology
The authors propose a tensor-based parametric channel estimation technique that exploits the multidimensional structure of the received signal. The core methodology involves the following steps:

System Modeling: The system is modeled as an uplink IRS-assisted MIMO scenario. The BS-IRS channel ( $G$ ) is treated as quasi-static, while the IRS-UE channel ( $H_k$ ) varies between transmission blocks. The received signal is modeled using a multipath channel model suitable for millimeter-wave (mmWave) scenarios.
Tensor Formulation: The received signal model is transformed into a 3rd-order PARAFAC (Parallel Factor Analysis) tensor model. By utilizing properties of the vectorization operator and the Khatri-Rao product, the signal collected over $T$ symbol periods and $K$ blocks is arranged into a third-order tensor $\mathcal{R}$ . This tensor captures the combined channel geometry and path loss information.
Algorithm Development: To solve for the channel parameters (specifically the steering matrices and the combined path loss vector), the paper employs an Alternating Least Squares (ALS) algorithm. The algorithm iteratively updates the factor matrices ( $A$ , $B$ , and $F$ ) by fixing the others and solving least-squares sub-problems until convergence is reached.
Pilot Design: The pilot signal matrix is designed as a Hadamard matrix, and the IRS phase-shift matrix utilizes a Discrete Fourier Transform (DFT) structure to facilitate the tensor decomposition.

Key Contributions

Novel Tensor Model: The paper introduces a 3rd-order PARAFAC tensor model specifically tailored for IRS-aided systems with a single time-varying channel (IRS-UE) and a static channel (BS-IRS). This differs from the authors' previous work which used Tucker models for quasi-static channels or two-stage frameworks for fully time-varying scenarios.
ALS-Based Estimation: The authors derive an ALS algorithm to decompose the PARAFAC tensor, allowing for the estimation of the concatenated channel parameters.
Complexity Analysis: The paper provides a detailed computational complexity analysis, comparing the proposed ALS method against the classical Least Squares (LS) filter and the Khatri-Rao Factorization (KRF) method.
Performance Validation: Extensive simulations are conducted to evaluate the Normalized Mean Square Error (NMSE) and computational costs under various system parameters (e.g., number of IRS elements, SNR, and channel paths).

Simulation Results

NMSE Performance: The proposed ALS solution demonstrates superior performance compared to both the classical LS filter and the state-of-the-art KRF method. Specifically, the proposed approach achieves approximately a 10 dB improvement over the LS filter and a 7 dB improvement over the KRF method in terms of NMSE for the concatenated channel. This gain is observed to be largely independent of the Signal-to-Noise Ratio (SNR).
Convergence Behavior: The algorithm typically converges within approximately 10 iterations. The number of iterations required increases with higher noise variance (low SNR) and a larger number of channel paths ( $L_1L_2$ ). However, in high SNR regimes, convergence is achieved rapidly.
Computational Complexity: The analysis indicates that the computational complexity of the proposed ALS algorithm is comparable to that of the competing LS and KRF solutions. The complexity is dominated by the initial LS estimation step required to form the tensor, meaning the iterative refinement does not significantly increase the overall computational burden.

Significance and Claims
The paper claims that the proposed tensor-based approach offers a significant enhancement in channel estimation accuracy by capitalizing on the intrinsic tensor structure of the received signal. Crucially, the authors assert that this performance gain is achieved without increasing the computational complexity of the channel estimation process compared to competing solutions. The work highlights the benefit of exploiting the static nature of the BS-IRS link to simplify the model while still accounting for the time-varying nature of the IRS-UE link, providing a practical solution for IRS-assisted communication systems where hardware constraints limit active signal processing at the IRS.