⚡ electrical engineering

Blind Channel Estimation and Data Detection for Near-Field XL-MIMO Systems

This paper proposes a two-stage blind channel estimation and data detection framework for uplink near-field XL-MIMO systems that combines an on-grid polar-domain sparse recovery algorithm with off-grid refinement to significantly improve symbol error rates, particularly in low signal-to-noise ratio conditions and when pilot overhead is limited.

Original authors: Maral Safari, Italo Atzeni

Published 2026-05-27

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Maral Safari, Italo Atzeni

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: The "Noisy Party" Problem

Imagine a massive, high-tech party (the Base Station) with hundreds of microphones (antennas) trying to listen to a few guests (users) shouting across a crowded room.

In the future of wireless technology, these parties are getting huge (XL-MIMO), and the guests are speaking very fast (high frequencies). However, there's a catch: the room is so big and the sound waves are so short that the sound doesn't just come from a specific direction (like "left" or "right"). It also depends on exactly how far away the guest is standing. This is called Near-Field propagation.

The Problem:
Usually, to understand what someone is saying, the host asks them to shout a specific "test word" (a pilot) first. The host listens to the test word, figures out how the sound travels, and then understands the rest of the conversation.

The Issue: In these future systems, the "test word" takes up too much time. The room is so chaotic that by the time the host finishes listening to the test words, the sound has already changed. There isn't enough time to shout the test words before the conversation is over.

The Solution:
The authors propose a new way to listen. Instead of asking for a test word, the host tries to figure out both the path the sound took and what the guests are saying at the same time, just by listening to the mixed-up noise. They call this Blind Channel Estimation and Data Detection.

How It Works: The Two-Stage Detective

The authors created a two-step detective process to solve this puzzle.

Stage 1: The "Rough Sketch" (B-OMP)

Imagine you are trying to find a few specific people in a massive crowd based on the sound of their voices. You don't know exactly where they are, but you have a giant map with a grid of possible locations (angles and distances).

The Trick: The system knows that sound usually comes from a few strong paths (like a direct line of sight or a bounce off a wall), not from everywhere at once. This is called sparsity.
The Method: The algorithm acts like a detective using a "matching pursuit" technique. It looks at the noisy recording and asks, "Which grid square on my map explains the loudest part of this noise?"
The Result: It picks the best grid squares, estimates the direction and distance, and guesses what the data (the message) might be. It's a "rough sketch" because the map grid isn't perfect—the person might be standing between two grid lines.

Stage 2: The "Fine-Tuning" (BCD)

The rough sketch is good, but it's not perfect because the person wasn't exactly on a grid line.

The Trick: Now, the system stops looking at the grid and looks at the continuous space. It uses a method called Block-Coordinate Descent (think of it as a "tweak-and-check" loop).
The Method: It takes the rough guess from Stage 1 and says, "Okay, if I move the angle slightly to the left, does the noise match better? What if I move the distance slightly closer?" It tweaks the angle, then the distance, then the volume, then the message, over and over again.
The Result: It smooths out the rough edges, finding the exact location and message, even if the person wasn't standing on a pre-drawn grid line.

Why This Is a Big Deal (The Results)

The authors tested their method against the old way (asking for test words first).

Saving Time: Their method works great even when there are very few "data symbols" (short messages) compared to the time available. It doesn't waste time shouting test words.
Low Volume: It works surprisingly well even when the signal is very weak (low Signal-to-Noise Ratio), which is common in these high-frequency systems.
The "Grid" Problem: They showed that just using the "Rough Sketch" (Stage 1) is already much better than the old method. But adding the "Fine-Tuning" (Stage 2) makes it even better, especially when the signal is weak.

The Catch:
If there are too many guests shouting at once, or if they are using very complex codes (high-order modulation), the system gets confused. In those specific crowded scenarios, the old method (asking for test words) still wins. But for most normal scenarios, the new "Blind" method is the winner.

Summary Analogy

Think of the old method as a teacher asking a student to read a specific sentence aloud to check their pronunciation before letting them read the story.

The New Method: The teacher listens to the student reading the story and, by analyzing the voice patterns and the words being spoken simultaneously, figures out both the student's accent (the channel) and the story content (the data) without needing a separate practice round.

The paper proves that this "listen-and-figure-out" approach is faster and more efficient for the massive, high-speed wireless networks of the future.

Technical Summary: Blind Channel Estimation and Data Detection for Near-Field XL-MIMO Systems

Problem Statement
Future wireless systems operating at high carrier frequencies (sub-THz) with extremely large-scale multiple-input multiple-output (XL-MIMO) arrays face significant challenges due to near-field propagation. In this regime, the channel depends jointly on the angle and distance of the user, rather than angle alone. This, combined with high carrier frequencies, results in short coherence intervals, making traditional pilot-based channel state information (CSI) acquisition inefficient due to the excessive training overhead required. The paper addresses the problem of Blind Channel Estimation and Data Detection (B-CE-DD) for uplink near-field XL-MIMO systems, aiming to jointly recover user-specific channels and transmitted data from superimposed received signals without dedicated pilot symbols, thereby reducing training overhead.

Methodology
The authors propose a two-stage framework that formulates the B-CE-DD problem as the recovery of user-specific rank-one channel-data products ( $\Xi_k = g_k \bar{d}_k^T$ ) from a superimposed signal. This formulation combines a polar-domain sparse channel model (exploiting the sparsity of angle-distance paths) with a low-dimensional data subspace model.

On-Grid Stage (Blind Orthogonal Matching Pursuit - B-OMP):
- The algorithm operates on a predefined polar-domain dictionary of angle-distance pairs.
- It iteratively identifies the dominant angle-distance components (atoms) corresponding to each user's propagation paths.
- At each iteration, it updates the estimates of the channel-data products by projecting the residual signal onto the subspace spanned by the selected atoms.
- Upon convergence, the estimated rank-one channel-data products are factorized using Singular Value Decomposition (SVD) or power iteration to separate the channel and data vectors. A phase alignment step using a single known pilot symbol per user resolves phase ambiguity.
Off-Grid Refinement Stage (Block-Coordinate Descent - BCD):
- To mitigate the approximation errors caused by the finite resolution of the on-grid dictionary (grid mismatch), a refinement stage is introduced.
- This stage allows angle and distance parameters to take continuous values within the polar domain.
- The algorithm minimizes the residual error between the effective received signal and its reconstruction using a Block-Coordinate Descent approach.
- In each iteration, it updates the angle and distance estimates via Gradient Descent (GD), while updating the small-scale fading coefficients and data vectors in closed form via least-squares solutions.
- The process iterates until convergence to a stationary point of the objective function.

Key Contributions

Problem Formulation: The paper formulates the near-field B-CE-DD problem as a structured blind bilinear inverse problem, recovering rank-one channel-data products by leveraging polar-domain sparsity and low-dimensional data subspaces.
B-OMP Algorithm: Development of a blind orthogonal matching pursuit algorithm that jointly estimates near-field channels and data vectors on a grid, specifically designed to handle the superimposed nature of multi-user signals without inter-user interference under specific load conditions ( $T \geq K(S+1)$ ).
BCD Refinement: Introduction of a block-coordinate descent algorithm to refine the on-grid estimates in the continuous polar domain, jointly optimizing channel parameters and data symbols to mitigate grid mismatch effects.
Complexity Analysis: The authors analyze the computational complexity, showing that B-OMP scales linearly with the number of antennas, coherence interval length, and dictionary size, while the BCD stage scales quadratically with the number of antennas.

Numerical Results
The proposed framework is evaluated against a baseline employing Zero-Forcing (ZF) beamforming with channel estimates obtained via a pilot-based OMP method (from prior work [5]).

Performance Gains: The proposed B-CE-DD framework significantly improves the Symbol Error Rate (SER) compared to the pilot-based baseline, particularly at low Signal-to-Noise Ratios (SNR) and when the number of data symbols is small relative to the coherence interval length.
Refinement Impact: The addition of the off-grid BCD stage provides further performance improvements over the on-grid B-OMP alone, especially at higher SNRs, by correcting grid mismatch errors.
System Load: The framework performs best when the total number of data symbols across all users is smaller than the coherence interval length. Performance degrades as the system load (ratio of total data symbols to coherence interval) increases, eventually underperforming the pilot-based baseline at high loads (e.g., $KS/T > 3/4$).
Modulation and Users: The method supports high-order modulations (up to 64-QAM) under moderate user loads but faces challenges with dense constellations and high multi-user interference (e.g., 8 users with 64-QAM).

Significance and Claims
The paper claims that the proposed two-stage B-CE-DD framework offers a viable solution for near-field XL-MIMO systems where short coherence intervals render traditional pilot-based estimation inefficient. By exploiting the coupled structure of the received signal to jointly recover channels and data, the method achieves substantial performance gains over conventional separated estimation and detection approaches, particularly in low-SNR and low-overhead scenarios. The authors note that the gains are most pronounced when the number of data symbols is small relative to the coherence interval, a regime relevant for short-packet transmissions. The work highlights that near-field B-CE-DD cannot be achieved by simply extending far-field methods or replacing angular dictionaries with polar ones, due to the inherent inter-user coupling and the joint angle-distance dependency of the channel.