A Compressed Sensing Faraday Depth Reconstruction Framework for the MeerKAT MIGHTEE-POL Survey

Miguel Cárcamo, Anna M. M. Scaife, Russ Taylor, M. J. Jarvis, Micah Bowles, Srikrishna Sekhar, Lennart Heino, J. M. Stil

This paper presents a new computational framework called cs-romer, designed to reconstruct Faraday depth signals from radio astronomy data. As we enter the era of "Big Data" radio astronomy with telescopes like MeerKAT and the upcoming Square Kilometre Array (SKA), traditional methods for analyzing cosmic magnetism are becoming computationally prohibitive. This work leverages compressed sensing to provide a scalable, automated solution.

TL;DR

The authors introduce cs-romer, a compressed sensing framework that reconstructs Faraday depth signals from noisy, incomplete radio spectro-polarimetric data. By combining $L_1$ regularization, wavelet transforms, and parallel computing, the framework efficiently processes massive datasets from the MeerKAT MIGHTEE survey to map cosmic magnetic fields.

What Problem Does This Paper Address?

To study cosmic magnetism, astronomers look at Faraday Rotation: the rotation of the plane of polarization of light as it passes through magnetized plasma. By measuring this rotation across different wavelengths (specifically $\lambda^2$), researchers can reconstruct a "Faraday depth" spectrum, which reveals the distribution of magnetic fields and gas along the line of sight.

However, modern radio telescopes produce "dirty" data—noisy, irregularly sampled, and massive in volume. Traditional reconstruction techniques like RM-CLEAN often struggle with "thick" structures (extended regions of magnetized gas) or require significant manual intervention. With the MeerKAT telescope producing data cubes in the tens to hundreds of gigabytes, there is an urgent need for an automated, mathematically robust, and scalable reconstruction pipeline.

Methods

The authors propose the cs-romer framework, which treats Faraday depth reconstruction as an optimization problem. Key components include:

  • Compressed Sensing (CS): Instead of relying on the Nyquist-Shannon sampling theorem, CS assumes the underlying signal is "sparse" (mostly zeros) in some domain (like wavelets). This allows for high-quality reconstruction even from incomplete data.
  • Optimization Algorithm: The framework uses the Fast Iterative Shrinkage-Thresholding Algorithm (FISTA) to solve an unconstrained objective function. This function balances data fidelity (how well the model fits the observations) with regularization terms ($L_1$ norm for sparsity and Total Variation for smoothness).
  • Handling Irregular Sampling: Since radio data is often irregularly spaced in $\lambda^2$, the framework employs the Non-Uniform Fast Fourier Transform (NUFFT). This avoids the artifacts and noise introduced by traditional "gridding" methods.
  • Wavelet Transforms: To handle both "thin" (point-like) and "thick" (extended) Faraday structures, the authors integrated the Undecimated Wavelet Transform (UWT), which is more robust to shifts in the data than standard discrete wavelet transforms.
  • Parallelization: The framework uses the joblib Python library to process independent lines of sight (LOS) in parallel, significantly speeding up the processing of large image cubes.

Datasets

  • MeerKAT MIGHTEE-POL Survey: Specifically, early-release data from the XMM-LSS extragalactic field.
  • Simulated Data: Used to evaluate and select the best wavelet families by measuring Peak Signal-to-Noise Ratio (PSNR), Root Mean Squared Error (RMSE), and information criteria (AIC/BIC).

Key Insights and Ideas

  • Sparsity is Contextual: Faraday signals aren't necessarily sparse in their raw form, but they are highly compressible in the wavelet domain. Choosing the right wavelet is critical for capturing different physical structures in space.
  • Computational Efficiency via Thresholding: The authors found they could reduce the number of lines of sight needing full optimization by 99.75% by only processing pixels where the total intensity and polarization exceeded a 3$\sigma$ noise threshold.
  • NUFFT Superiority: Moving away from standard DFT/FFT to NUFFT allows the framework to handle the inherent irregularities of radio frequency sampling without losing precision.

Key Results

  • Successful Reconstruction: The framework successfully reconstructed the Faraday depth, rotation measure (RM) uncertainty, and polarization fraction for a radio galaxy in the XMM-LSS field.
  • Scalability: A single field reconstruction (requiring ~160 GB of memory) was completed in 81.7 minutes on a high-end server (2x Intel Xeon Gold CPUs, 1.5 TB RAM).
  • Data Volume: The framework proved capable of handling MIGHTEE-POL data cubes of approximately 67 GB (raw) and 105 GB (reconstructed).

Limitations and Future Work

  • Single-Node Limits: While joblib provides parallelization, it is currently limited to a single computing node. The authors plan to integrate big-data libraries like Dask to distribute the workload across multiple nodes in a cluster.
  • Automated Parameter Tuning: While the framework adopts some adaptive regularization methods, further automation in calculating the optimal balance between sparsity and smoothness ($\eta$ parameters) remains an area for improvement.
  • Spectral Index Assumptions: The current demonstration uses a fixed spectral index (0.7) for all lines of sight, though the framework is designed to eventually accept more complex spectral index maps.

Summary updated

Abstract

In this work we present a novel compute framework for reconstructing Faraday depth signals from noisy and incomplete spectro-polarimetric radio datasets. This framework is based on a compressed-sensing approach that addresses a number of outstanding issues in Faraday depth reconstruction in a systematic and scaleable manner. We apply this framework to early-release data from the MeerKAT MIGHTEE polarisation survey.
BibTeX
@inproceedings{carcamo2022compressed,
  author = {Miguel Cárcamo and Anna M. M. Scaife and Russ Taylor and M. J. Jarvis and Micah Bowles and Srikrishna Sekhar and Lennart Heino and J. M. Stil},
  title = {{A Compressed Sensing Faraday Depth Reconstruction Framework for the MeerKAT MIGHTEE-POL Survey}},
  booktitle = {2022},
  year = {2022},
  doi = {10.23919/at-ap-rasc54737.2022.9814329},
}
Figure