Near-Field Motion Parameter Estimation: A Variational Bayesian Approach

Chunwei Meng, Zhaolin Wang, Zhiqing Wei, Yuanwei Liu, Zhiyong Feng

Abstract

A near-field motion parameter estimation method is proposed. In contrast to far-field sensing systems, the near-field sensing system leverages spherical-wave characteristics to enable full-vector location and velocity estimation. Despite promising advantages, the near-field sensing system faces a significant challenge, where location and velocity parameters are intricately coupled within the signal. To address this challenge, a novel subarray-based variational message passing (VMP) method is proposed for near-field joint location and velocity estimation. First, a factor graph representation is introduced, employing subarray-level directional and Doppler parameters as intermediate variables to decouple the complex location-velocity dependencies. Based on this, the variational Bayesian inference is employed to obtain closed-form posterior distributions of subarray-level parameters. Subsequently, the message passing technique is employed, enabling tractable computation of location and velocity marginal distributions. Two implementation strategies are proposed: system-level fusion or subarray-level fusion. Numerical results demonstrate that the proposed VMP method achieves centimeter-level location accuracy and sub-m/s velocity accuracy with a magnitude lower complexity than existing approaches.

Introduction

As 6G networks transition toward extremely large-scale MIMO (XL-MIMO) and higher frequency bands (mmWave/THz), the electromagnetic "near-field" region expands significantly. In this region, waves propagate spherically rather than as flat planes. While this spherical wavefront allows a single base station to estimate a target's full 2D velocity vector, it introduces severe non-linear coupling between the target's location $(x, y)$ and velocity $(v_x, v_y)$. Traditional estimation methods require a computationally explosive 4D joint search, making real-time tracking practically impossible.

System Model and Geometry

The proposed system considers a moving target at position $\mathbf{p}_0 = (x_0, y_0)$ with velocity $\mathbf{v}_0$. The massive XL-MIMO array is partitioned into smaller subarrays to manage complexity. Within each subarray, a piecewise-far-field approximation is applied, while the global near-field characteristics are preserved across the entire array.

Geometry of the near-field sensing system showing a moving target and the subarray-based Tx/Rx configuration.

As shown in the figure above, each transmit-receive subarray pair effectively acts as a bistatic radar, characterized by specific angles of departure (DoD) $\theta_i$ and arrival (DoA) $\phi_j$. This geometric setup forms the basis for decoupling the motion parameters.

The VMP-NMPE Framework

To break the 4D computational bottleneck, the Variational Message Passing-based Near-Field Motion Parameter Estimation (VMP-NMPE) algorithm is introduced. The core innovation lies in transforming the complex estimation problem into a probabilistic factor graph.

Factor graph representation of the proposed VMP-NMPE algorithm, illustrating the dependencies between location, velocity, and intermediate subarray parameters.

Decoupling via Factor Graphs

Instead of direct estimation from raw signals, the framework introduces intermediate variables: Direction of Arrival (DoA), Direction of Departure (DoD), and bistatic Doppler shifts for each subarray pair. This factor graph (shown above) allows for:

  1. Variational Bayesian Inference (VBI): Finding closed-form posterior distributions for intermediate variables using continuous von Mises (VM) distributions to avoid grid-mismatch errors.
  2. Message Passing Fusion: Aggregating subarray estimates to find global parameters. This can be done via System-level fusion (centralized optimization) or Subarray-level fusion (distributed Gaussian product rule).

Key Results and Performance

The proposed VMP method demonstrates significant advantages over traditional Maximum Likelihood (ML) and grid-based methods:

  • High Accuracy: It achieves centimeter-level location accuracy and sub-m/s velocity accuracy, closely approaching the theoretical Cramér-Rao Bound (CRB) at high SNRs.
  • Computational Efficiency: The algorithm reduces processing time from hundreds of seconds (for ML) to approximately 0.1–1 second, representing a 2-to-3 order of magnitude speedup.
  • Mobility Robustness: Higher target velocities induce stronger Doppler diversity across spatially distributed subarrays, which the VMP architecture exploits to improve estimation robustness in high-dynamic scenarios.

Conclusion

The VMP-NMPE algorithm provides a low-complexity, high-accuracy solution for real-time near-field sensing in 6G XL-MIMO systems. By leveraging subarray-based factor graphs and variational inference, it successfully overcomes the parameter coupling challenges inherent in spherical-wave propagation. Future work will focus on validating this framework on real-world hardware testbeds to account for synchronization errors and hardware impairments.

Summary updated

Abstract

A near-field motion parameter estimation method is proposed. In contract to far-field sensing systems, the near-field sensing system leverages spherical-wave characteristics to enable full-vector location and velocity estimation. Despite promising advantages, the near-field sensing system faces a significant challenge, where location and velocity parameters are intricately coupled within the signal. To address this challenge, a novel subarray-based variational message passing (VMP) method is proposed for near-field joint location and velocity estimation. First, a factor graph representation is introduced, employing subarray-level directional and Doppler parameters as intermediate variables to decouple the complex location-velocity dependencies. Based on this, the variational Bayesian inference is employed to obtain closed-form posterior distributions of subarray-level parameters. Subsequently, the message passing technique is employed, enabling tractable computation of location and velocity marginal distributions. Two implementation strategies are proposed: 1) System-level fusion that aggregates all subarray posteriors for centralized estimation, or 2) Subarray-level fusion where locally processed estimates from subarrays are fused through Guassian product rule. Cramér-Rao bounds for location and velocity estimation are derived, providing theoretical performance limits. Numerical results demonstrate that the proposed VMP method outperforms existing approaches while achieving a magnitude lower complexity. Specifically, the proposed VMP method achieves centimeter-level location accuracy and sub-m/s velocity accuracy. It also demonstrates robust performance for high-mobility targets, making the proposed VMP method suitable for real-time near-field sensing and communication applications.
BibTeX
@article{meng2025near,
  author = {Chunwei Meng and Zhaolin Wang and Zhiqing Wei and Yuanwei Liu and Zhiyong Feng},
  title = {{Near-Field Motion Parameter Estimation: A Variational Bayesian Approach}},
  journal = {IEEE Trans. Wireless Communications},
  year = {2025},
}
Figure