⚡ electrical engineering

Hybrid-Field Channel Estimation for XL-MIMO Systems: Dictionary-based Sparse Signal Recovery

This paper proposes a dictionary-based sparse signal recovery method for hybrid-field channel estimation in XL-MIMO systems that adaptively determines the number of propagation paths without prior knowledge of sparsity levels or near-field/far-field ratios, while mitigating grid mismatch through continuous parameter refinement.

Original authors: David William Marques Guerra, Taufik Abrao

Published 2026-02-04

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: David William Marques Guerra, Taufik Abrao

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

Imagine you are trying to listen to a conversation in a massive, crowded stadium (the future 6G network). The stadium is so huge that the sound waves behave differently depending on where the speakers are standing.

Far-Field (The Distant Crowd): If someone is far away, their voice reaches you as a flat, smooth wave, like a gentle breeze hitting a wall.
Near-Field (The Person Next to You): If someone is standing right next to the microphone, their voice reaches you as a spreading, spherical wave, like ripples in a pond expanding outward.

In XL-MIMO (Extremely Large-Scale Multiple-Input Multiple-Output) systems, the antenna array is so big that it captures both types of waves at the same time. This is called a Hybrid-Field. The challenge is that the "sound" (the data signal) gets mixed up, and the receiver needs to figure out exactly who is speaking, how far away they are, and what their voice sounds like, all while ignoring the background noise.

Here is how the paper solves this problem, using simple analogies:

1. The Problem: The "Blind" Search

Traditionally, to find these signals, engineers had to guess two things beforehand:

How many people are speaking? (The number of signal paths).
What is the split between near and far speakers? (The ratio of spherical to flat waves).

If you guessed wrong, your estimate was off. It was like trying to find a needle in a haystack while being told, "There are exactly 5 needles, and 3 are blue," but you didn't actually know if that was true.

2. The Solution: A Smart, Adaptive Detective

The authors propose a new method called $\epsilon$ -OMP-SSIGW. Think of this method as a very smart detective who doesn't need to know the number of needles beforehand.

The Dictionary (The List of Suspects): The system has a giant "dictionary" or a list of every possible direction and distance a signal could come from. It's like having a photo album of every possible angle and distance.
The Hunt (Greedy Selection): The detective looks at the mixed-up signal and picks the "suspect" (the atom from the dictionary) that matches the signal the best.
The "Stop" Sign (Residual Rule): Instead of guessing how many people are talking, the detective keeps picking suspects until the "leftover noise" (the residual) is small enough. Once the leftover noise is just random static, the detective stops. This means the system automatically figures out how many paths exist without needing a human to tell it.

3. The Refinement: Tuning the Radio

Once a suspect is picked, the initial guess is usually a bit "off-grid" (like guessing a person is at 10 meters when they are actually at 10.4 meters).

Old Way: Previous methods tried to fix this by doing massive, complex math calculations (like trying to solve a giant puzzle all at once).
The New Way: This paper uses a "fine-tuning" approach. It treats the adjustment like turning a single knob on a radio. It makes tiny, simple adjustments to the angle and distance one by one.
- Analogy: Imagine you are trying to focus a blurry photo. Old methods tried to rewrite the whole image code. This new method just nudges the focus ring slightly until the image is sharp. It's much faster and requires less computing power.

4. The Results: Faster and Sharper

The paper tested this new detective against eight other top methods.

Accuracy: It found the signals more accurately than the others, whether the speakers were far away, close up, or a mix of both.
Speed: Because it uses simple "knob-turning" math instead of heavy "puzzle-solving" math, it runs much faster.
Robustness: It works well even if the system doesn't know the exact noise level perfectly.

Summary

In short, this paper introduces a smarter, faster way for future 6G networks to "listen" to signals in massive antenna arrays. It stops guessing how many signals there are and instead listens until the noise is gone. It then uses simple, step-by-step adjustments to pinpoint exactly where the signals are coming from, making the whole process more efficient and reliable.

Technical Summary: Hybrid-Field Channel Estimation for XL-MIMO Systems

Problem Statement
Extremely large-scale multiple-input multiple-output (XL-MIMO) systems are pivotal for future 6G networks, offering significant gains in spectral efficiency and spatial resolution. However, the massive antenna apertures in XL-MIMO create a unique propagation environment where the traditional far-field (FF) plane-wave assumption fails for many scatterers. Instead, a hybrid-field (HF) regime emerges, characterized by the coexistence of FF planar waves and near-field (NF) spherical waves. Accurate Channel State Information (CSI) is essential for exploiting XL-MIMO potential, yet estimating HF channels is challenging due to the need to model both angular and distance variations simultaneously.

Existing approaches, such as Hybrid-Field Orthogonal Matching Pursuit (HF-OMP) and Sparse Gradient Projection (SGP) methods, often suffer from practical limitations. Many require prior knowledge of the channel sparsity level ( $L$ ) and the specific ratio of NF to FF paths ( $\gamma$ ). Others rely on computationally expensive operations, such as matrix pseudo-inversions or multi-dimensional gradient steps, to refine parameters. These dependencies and complexities hinder robust deployment in dynamic XL-MIMO environments.

Methodology
The authors propose a unified channel estimation framework called $\epsilon$ -OMP–SSIGW (epsilon-Orthogonal Matching Pursuit with Simplified Sparse Iterative Gridless Weighting). The approach models the HF channel as a superposition of FF and NF components, leveraging the inherent sparsity of the channel in both the angular (DFT) and polar (angle-distance) domains.

The core methodology involves a greedy, iterative process:

Dictionary-Based Selection: A combined dictionary containing both FF (angular) and NF (polar) atoms is used. At each iteration, the atom most correlated with the current residual is selected.
Adaptive Path Counting: Unlike prior methods, the algorithm does not require $L$ or $\gamma$ as inputs. Instead, it employs a residual-based stopping rule ( $\|r\|^2/N \leq \sigma^2$ ) to automatically determine the number of significant paths, making it robust to unknown channel structures.
Simplified Refinement (SSIGW): Once an atom is selected, the method performs a "coarse-to-fine" refinement to mitigate grid mismatch.
- Gain Update: The path gain is updated using a closed-form scalar least-squares (LS) solution derived from a single-column refinement structure, avoiding matrix pseudo-inversions.
- Parameter Refinement: The angular and distance parameters are refined using scalar gradients updated via an Armijo backtracking line search. This contrasts with previous works that utilize complex matrix gradients.
Termination: The process iterates until the mean squared error (MSE) falls below a noise-dependent threshold or a maximum iteration count is reached.

Key Contributions

Blind Estimation: The proposed method operates without prior knowledge of the total sparsity level ( $L$ ) or the NF/FF path split ratio ( $\gamma$ ). It only requires an estimate of the noise variance, which is practically obtainable.
Computational Efficiency: By deriving closed-form expressions for gain updates and utilizing scalar gradients for parameter refinement, the authors significantly reduce the computational complexity per iteration compared to matrix-based approaches like SGP or standard OMP variants.
Unified HF Model: The work introduces a unified estimation framework that seamlessly handles both FF and NF components within a single sparse recovery problem, rather than treating them as separate sequential stages.
Robustness: The algorithm is designed to be robust against noise variance mismatches and grid discretization errors through its iterative refinement mechanism.

Results
Numerical simulations were conducted under both Non-Line-of-Sight (NLoS) and mixed Line-of-Sight/Non-Line-of-Sight (LoS/NLoS) conditions, comparing $\epsilon$ -OMP–SSIGW against eight state-of-the-art benchmarks, including LS, MMSE, HF-OMP variants, and HF-SGP methods.

Accuracy: The proposed estimator achieved the lowest Normalized Mean Squared Error (NMSE) across the entire Signal-to-Noise Ratio (SNR) range in both LoS and NLoS scenarios. It notably outperformed methods that required prior knowledge of $L$ and $\gamma$ .
Complexity: While the refined $\epsilon$ -OMP–SSIGW incurs slightly more overhead than the basic $\epsilon$ -OMP, it remains significantly faster than HF-SGP and exhaustive search-based HF-OMP variants. The runtime was shown to be SNR-independent and dominated by fixed matrix operations.
Adaptivity: The simulations confirmed that the estimator correctly adapts the number of detected paths ( $L_{est}$ ) based on the SNR and residual criteria, validating its ability to function without explicit $L$ input.

Significance
The paper claims that the $\epsilon$ -OMP–SSIGW framework offers a practical and computationally efficient solution for HF channel estimation in XL-MIMO systems. By eliminating the need for prior knowledge of channel sparsity and path ratios, and by simplifying the mathematical operations required for refinement, the method achieves a superior performance-complexity trade-off. The authors posit that this makes the estimator particularly suitable for real-world deployment where channel characteristics are dynamic and unknown. The work lays the groundwork for future extensions to multi-antenna user equipment, general array geometries (UPA), and hardware impairment modeling.