From Partial Calibration to Full Potential: A Two-Stage Sparse DOA Estimation for Incoherently-Distributed Sources with Gain-Phase Uncertainty

Xu He, Tuo Wu, Wei Liu, Maged Elkashlan, Naofal Al‐Dhahir, Mérouane Debbah, Chau Yuen, Hing Cheung SoarXiv 2025 · DOI · arXiv

This paper, published in the IEEE Internet of Things Journal, addresses a critical challenge in modern wireless sensing: how to accurately locate signal sources in "messy" real-world environments where signals are scattered and hardware is imperfect.

TL;DR

The authors propose a two-stage sparse reconstruction framework that estimates the Direction-of-Arrival (DOA) for scattered signal sources while simultaneously correcting for hardware calibration errors. By "bootstrapping" from a small set of calibrated sensors to the full array, the method achieves high-resolution tracking in low-signal environments where traditional methods fail.

What Problem Does This Paper Address?

In applications like the Internet of Vehicles (IoV) or autonomous positioning, we need to know exactly where a signal is coming from. However, two major hurdles usually get in the way:

  1. Incoherently Distributed (ID) Sources: In "rich scattering" environments (like a city street), a signal doesn't arrive as a single clean point. Instead, it arrives as a "cloud" of reflections, known as angular spread.
  2. Gain-Phase Uncertainty: Real-world antenna arrays are rarely perfect. Small hardware variations in the radio frequency (RF) chains introduce "gain-phase" errors that distort the data.

Most existing solutions tackle these problems in isolation. Subspace-based algorithms (like MUSIC or ESPRIT) often require you to know the number of signal sources beforehand and struggle when the signal-to-noise ratio (SNR) is low. This paper seeks a unified, "sparse" approach that handles scattering and hardware errors simultaneously without needing prior knowledge of the source count.

Methods

The researchers utilize the Generalized Array Manifold (GAM) framework, which mathematically models scattered signals as a central direction plus small deviations. Their "Two-Stage" approach works as follows:

  • Stage 1: Coarse Estimation via Partly-Calibrated Arrays (PCAs). The authors exploit a clever mathematical property: the first element of an antenna array is relatively "immune" to the effects of angular spread. By using a small subset of sensors that are already calibrated, they extract a "clean" covariance vector and use sparse reconstruction ($\ell_2-\ell_1$ norm minimization) to get a rough idea of where the signals are.
  • Stage 2: Full Potential Refinement. Using the coarse estimates from Stage 1, the algorithm calculates the gain and phase errors for the rest of the uncalibrated sensors. Once the whole array is "virtually calibrated," they formulate a Sparse Total Least-Squares (STLS) optimization problem. This is solved using an alternating descent algorithm that iteratively refines the DOA estimates while accounting for model biases and noise.

Datasets

The paper relies on rigorous numerical simulations rather than static datasets, which is standard for algorithmic research in signal processing.

  • Setup: A 16-element Uniform Linear Array (ULA).
  • Evaluation: The performance is measured using Root Mean Square Error (RMSE) across 300 Monte-Carlo trials.
  • Benchmarks: The proposed method (designated as "Pro-SS") is compared against the RARE (Rank-Reduction) estimator, ESPRIT, and the theoretical Cramér-Rao Bound (CRB), which represents the absolute physical limit of estimation accuracy.

Key Insights and Ideas

  • The "Immunity" Trick: The realization that the first row of the steering vector's derivative is zero is the "aha moment." It allows the authors to bypass the complex interference of angular spread during the initial calibration phase.
  • Sparsity is Robust: Unlike subspace methods that break down if you guess the number of sources wrong, the sparse reconstruction approach naturally identifies the number of sources by looking for "peaks" in a sparse vector.
  • Total Least Squares for Model Bias: Most methods assume the mathematical model (GAM) is perfect. By using Total Least Squares in the second stage, the authors explicitly account for the fact that the GAM is just an approximation of real-world scattering.

Key Results

  • High Resolution: In simulations with two sources separated by only $6^\circ$, the proposed method clearly distinguished both, while the competing RARE method failed to resolve them.
  • Low SNR Performance: At a very noisy -9 dB SNR, the proposed method maintained significantly lower error (RMSE) than ESPRIT or RARE.
  • Resilience to Scattering: As the "angular spread" (the width of the signal cloud) increased from $0.5^\circ$ to $2.5^\circ$, the proposed method's accuracy degraded much more slowly than existing algorithms, proving its robustness in multipath environments.

Limitations and Future Work

  • Computational Cost: The second stage involves an iterative alternating descent algorithm. While effective, this is computationally heavier than "one-shot" subspace methods, which might be a concern for real-time edge processing in high-speed vehicles.
  • Small Angle Assumption: The GAM model assumes angular deviations are relatively small. The performance in environments with extremely wide scattering (e.g., very close-range indoor reflections) was not explored.
  • Array Geometry: The study focused on linear arrays. Future work could extend this framework to 2D planar arrays or circular arrays commonly used in 5G/6G base stations.

Summary updated

Abstract

Direction-of-arrival (DOA) estimation for incoherently distributed (ID) sources is essential in multipath wireless communication scenarios, yet it remains challenging due to the combined effects of angular spread and gain-phase uncertainties in antenna arrays. This paper presents a two-stage sparse DOA estimation framework, transitioning from partial calibration to full potential, under the generalized array manifold (GAM) framework. In the first stage, coarse DOA estimates are obtained by exploiting the output from a subset of partly-calibrated arrays (PCAs). In the second stage, these estimates are utilized to determine and compensate for gain-phase uncertainties across all array elements. Then a sparse total least-squares optimization problem is formulated and solved via alternating descent to refine the DOA estimates. Simulation results demonstrate that the proposed method attained improved estimation accuracy compared to existing approaches, while maintaining robustness against both noise and angular spread effects in practical multipath environments.
BibTeX
@misc{he2025from,
  author = {Xu He and Tuo Wu and Wei Liu and Maged Elkashlan and Naofal Al‐Dhahir and Mérouane Debbah and Chau Yuen and Hing Cheung So},
  title = {{From Partial Calibration to Full Potential: A Two-Stage Sparse DOA Estimation for Incoherently-Distributed Sources with Gain-Phase Uncertainty}},
  year = {2025},
  eprint = {2501.16854},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2501.16854},
  howpublished = {arXiv},
}

Loading related content...

Figure