Joint Visibility Region and Channel Estimation for Extremely Large-Scale MIMO Systems

Anzheng Tang, Jun-Bo Wang, Yijin Pan, Wence Zhang, Xiaodan Zhang, Yijian Chen, Hongkang Yu, Rodrigo C. de LamareIEEE Trans. Communications 2024 · DOI

TL;DR

This paper tackles the challenge of accurately estimating wireless channels in Extremely Large-Scale MIMO (XL-MIMO) systems, which are crucial for future 6G networks. They introduce a novel two-stage approach that jointly detects the "visibility region" (VR) of users and estimates the channel, outperforming existing methods, especially in low signal-to-noise ratio (SNR) conditions.

What Problem Does This Paper Address?

Future wireless communication systems, particularly those operating in millimeter-wave (mmWave) and sub-terahertz bands, are moving towards Extremely Large-Scale MIMO (XL-MIMO). These systems use a massive number of antennas, offering significant performance gains. However, this scale introduces new challenges:

  • Near-Field Effects: With large antenna arrays, users are often in the "near-field" region, meaning the wavefront is curved, invalidating traditional far-field assumptions.
  • Spatial Non-Stationarity (SnS): Obstacles and scattering can cause the channel characteristics to vary across different antenna elements, making the channel "spatially non-stationary." Existing channel estimation methods often assume spatial stationarity, leading to performance degradation.
  • Visibility Region (VR): The SnS property means that only a subset of antennas might "see" a particular user. Accurately identifying this "visibility region" is crucial for efficient communication.
  • Joint Detection and Estimation: Effectively estimating the channel requires knowing the VR, and accurately determining the VR can benefit from channel information. This interdependence makes joint detection and estimation a complex problem.

Existing methods struggle to simultaneously address these issues, particularly in low SNR scenarios where accurate VR detection is difficult.

Methods

The authors propose a novel two-stage VR detection and Channel Estimation (TS-VDCE) scheme:

  1. Stage I: VR Detection-Oriented Message Passing (VRDO-MP):
    • This stage focuses on detecting the user's visibility region (VR).
  2. It formulates VR detection as a Bayesian inference problem.
  3. A key innovation is the VRDO-MP algorithm, which leverages the spatial correlation of the VR in the antenna domain. This is a departure from previous methods that relied solely on statistical channel characteristics.
  4. The algorithm uses a structured message-passing framework, decoupling the problem into two modules: one for estimating the antenna-domain channel (LMMSE estimator) and another for detecting the VR (VR detector). This iterative approach refines both estimates.
  5. Stage II: Belief-Based Orthogonal Matching Pursuit (BB-OMP):
    • This stage uses the VR information obtained from Stage I to estimate the channel.
  6. It exploits the wavenumber-domain sparsity of the channel, which is a characteristic of XL-MIMO systems.
  7. The BB-OMP algorithm is proposed, which uses the "soft" visibility probabilities from Stage I to construct a more accurate measurement matrix. This is an improvement over "hard" decisions (binary visibility) or ignoring VR information altogether.
  8. This stage effectively transforms the channel estimation problem into a sparse signal recovery problem.

Datasets

Not discussed in the paper. The authors use simulated data based on system parameters and channel models.

Key Insights and Ideas

  • VR is Key: The paper strongly emphasizes that accurate VR detection is a prerequisite for effective channel estimation in SnS XL-MIMO systems, especially at low SNRs.
  • Exploiting Multiple Domains: The core novelty lies in simultaneously leveraging both the antenna-domain spatial correlation (for VR detection) and the wavenumber-domain sparsity (for channel estimation). This multi-domain approach is more powerful than methods focusing on a single domain.
  • Two-Stage Advantage: Separating VR detection and channel estimation into two stages, with feedback between them (implicitly through the VR information used in the second stage), simplifies the problem and improves performance.
  • Soft VR Information: Using "soft" visibility probabilities (rather than binary decisions) in the BB-OMP algorithm leads to more robust and accurate channel estimation.
  • Robustness to VR Size: The proposed TS-VDCE scheme demonstrates robustness to variations in the size of the visibility region, making it applicable to both stationary and non-stationary scenarios.

Key Results

  • Improved VR Detection: The VRDO-MP algorithm significantly outperforms existing methods (like RFEB) in VR detection accuracy, achieving over 90% correct detection at 0 dB SNR for certain parameters, compared to around 70% for the baseline.
  • Enhanced Channel Estimation: The proposed BB-OMP algorithm, leveraging the VR information, achieves significantly better Normalized Mean Square Error (NMSE) compared to algorithms that ignore VR or SnS properties (e.g., P-OMP, W-OMP, AD-AMP).
  • Near-Optimal Performance: The TS-VDCE scheme approaches the performance of a "Genie-aided" W-OMP algorithm (which assumes perfect knowledge of VR), demonstrating its effectiveness.
  • Low SNR Performance: The proposed methods show substantial gains in low SNR regimes, where traditional methods struggle.
  • Spectral Efficiency: The improved channel estimation translates to higher achievable spectral efficiency, closely matching that of perfect CSI.

Limitations and Future Work

  • Sequential Nature: The current two-stage approach is sequential, meaning errors in VR detection can propagate to channel estimation. The authors suggest that concurrent VR detection and CE is an interesting avenue for future research.
  • Parameter Overhead: The VR is represented by a vector whose dimension scales with the number of antennas. For extremely large antenna counts, this can lead to significant estimation overhead. Future work could focus on more efficient representations of SnS channels using fewer parameters.
  • Approximation in VR Detector: The paper notes that the Gaussian approximation for the VR detector's input messages might not always be accurate. Future work could explore more sophisticated methods for obtaining these messages.
  • Markov Chain Assumption: The VR is modeled as a first-order Markov chain. Future work could consider Markov random fields for more accurate modeling of spatial correlations.

Summary updated

BibTeX
@article{tang2024joint,
  author = {Anzheng Tang and Jun-Bo Wang and Yijin Pan and Wence Zhang and Xiaodan Zhang and Yijian Chen and Hongkang Yu and Rodrigo C. de Lamare},
  title = {{Joint Visibility Region and Channel Estimation for Extremely Large-Scale MIMO Systems}},
  journal = {IEEE Trans. Communications},
  year = {2024},
  doi = {10.1109/tcomm.2024.3394757},
}
Figure