Hybrid Generalized Approximate Message Passing for Active User Detection and Channel Estimation With Correlated Group-Heterogeneous Activity

Lélio Chetot, Malcolm Egan, Jean-Marie GorceIEEE Trans. Communications 2024 · DOI

TL;DR

This paper introduces a novel algorithm, GHetA-HGAMP, for simultaneously detecting active users and estimating their communication channels in grant-free random access (GFRA) wireless networks. It addresses the challenge of correlated and heterogeneous user activity, a common scenario in industrial IoT, by employing a sophisticated copula-based model and a hybrid message-passing approach, significantly outperforming existing methods.

What Problem Does This Paper Address?

Modern wireless networks, particularly for industrial IoT (IIoT) and future 6G applications, aim to support a massive number of devices with ultra-reliable and low-latency communication (uRLLC). A key bottleneck is the random access procedure, which traditionally requires significant control signaling. Grant-free random access (GFRA) offers a solution by allowing devices to transmit without explicit permission, but this necessitates efficient methods for active user detection and channel estimation. Existing algorithms often assume user activity is independent and homogeneous, which is unrealistic in many IIoT scenarios where sensors on the same equipment might exhibit correlated behavior. This paper tackles the problem of active user detection and channel estimation (AUDaCE) in GFRA systems where user activity is not only random and sporadic but also correlated and heterogeneous across groups of devices.

Methods

The core of this work lies in two main innovations:

  • Group-Heterogeneous Activity (GHetA) Model: The authors propose a new statistical model for user activity that captures complex correlation structures. This model uses copula theory to model the dependence between latent variables representing activity probabilities, which in turn influence the actual user activity states. This approach is more flexible than previous models, allowing for a wider range of correlation patterns.
  • Hybrid Generalized Approximate Message Passing (GHetA-HGAMP) Algorithm: To solve the AUDaCE problem under the GHetA model, the authors develop a hybrid message-passing algorithm. This algorithm combines principles of belief propagation (BP) for handling the correlated activity with generalized approximate message passing (GAMP) for efficient estimation. The hybrid approach allows for tractable computation of the joint user detection and channel estimation problem, even with complex correlation structures.

Datasets

Not discussed in the paper. The evaluation is based on extensive Monte-Carlo simulations with synthetically generated data based on the proposed GHetA model and its variations.

Key Insights and Ideas

  • Correlation Matters: The paper strongly emphasizes that ignoring or oversimplifying user activity correlation leads to significant performance degradation in AUDaCE.
  • Copulas for Flexibility: Using copula theory provides a powerful and flexible way to model diverse dependence structures between user activities, going beyond simpler correlation models.
  • Hybrid Approach is Key: Combining BP for correlation modeling with GAMP for efficient estimation is crucial for tackling the complexity of the GHetA model.
  • Robustness to Mismatched Correlation: The GHetA-HGAMP algorithm shows a degree of robustness to imperfect knowledge of the correlation structure, performing better than simpler models even with some mismatch.
  • Antenna Diversity Helps: Increasing the number of antennas at the access point can compensate for a lack of precise correlation information, highlighting the benefit of spatial diversity.

Key Results

  • Significant Performance Gains: The GHetA-HGAMP algorithm consistently outperforms state-of-the-art methods (ISA-HGAMP, GSA-HGAMP, GHomA-HGAMP) in terms of both user detection (lower FAR, MDR, UER) and channel estimation (lower NMSE) across various correlation regimes.
  • Improved Detection Errors: The paper reports gains of up to 10 times fewer detection errors and 10 dB less channel estimation error compared to existing algorithms.
  • Best Performance Across Regimes: GHetA-HGAMP achieves the best or near-best performance across a wide range of correlation scenarios, from low to high.
  • Adaptability: The algorithm demonstrates its ability to leverage the dependence structure of user activity, leading to more accurate AUDaCE.

Limitations and Future Work

  • Estimation of Correlation: The paper acknowledges that estimating the copula dependence structure when it's unknown remains an open problem.
  • State Evolution Derivation: Deriving state evolution for the GHetA-HGAMP algorithm is noted as a difficult challenge due to the non-separable and non-identically distributed nature of channel coefficients in the proposed model.
  • Computational Complexity: While GHetA-HGAMP offers superior performance, its computational complexity, particularly due to the numerical integration for BP updates, can be a bottleneck for very large-scale systems. Future work could focus on more efficient approximations or alternative algorithms.

Summary updated

BibTeX
@article{chetot2024hybrid,
  author = {Lélio Chetot and Malcolm Egan and Jean-Marie Gorce},
  title = {{Hybrid Generalized Approximate Message Passing for Active User Detection and Channel Estimation With Correlated Group-Heterogeneous Activity}},
  journal = {IEEE Trans. Communications},
  year = {2024},
  doi = {10.1109/tcomm.2024.3367760},
}
Figure