🔭 astrophysics

Murriyang cryogenic phased array feed: spectral-line results and noise-reduction methods

This paper presents the spectral-line performance and noise-reduction capabilities of the new cryogenic phased array feed on the Murriyang telescope, demonstrating its superior survey speed and field of view through HI observations of the LMC and NGC 6744, while validating robust 3D SVD techniques for effectively mitigating RFI and continuum contamination to detect faint cosmological signals.

Original authors: L. Staveley-Smith, S. Barker, R. Berangi, A. B. Bolin, S. Broadhurst, J. D. Bunton, N. Carter, S. Castillo, W. Chandler, A. Chippendale, J. R. Dawson, F. Di Dio, A. R. Dunning, S. Gordon, J. A. Green

Published 2026-05-13

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: L. Staveley-Smith, S. Barker, R. Berangi, A. B. Bolin, S. Broadhurst, J. D. Bunton, N. Carter, S. Castillo, W. Chandler, A. Chippendale, J. R. Dawson, F. Di Dio, A. R. Dunning, S. Gordon, J. A. Green, A. Hafner, D. B. Hayman, D. Humphrey, A. Jameson, S. Johnston, J. F. Kaczmarek, J. Ma, G. Perry, M. Pilawa, J. Rhee, L. Toomey, J. van Aardt, N. Wang

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Big Picture: A New "Super-Eye" for the Parkes Telescope

Imagine the Murriyang telescope (formerly the Parkes radio telescope) in Australia as a giant ear listening to the whispers of the universe. For decades, this ear used a single, large "horn" to catch radio signals. It was good, but it could only listen to one tiny spot in the sky at a time.

This paper introduces a new upgrade: the cryoPAF. Think of this as replacing that single horn with a giant, high-tech honeycomb containing 72 tiny ears packed together.

The "Cryo" part: These ears are frozen to extremely cold temperatures (cryogenic). This is like putting the microphone in a soundproof, ice-cold room to stop it from hissing with its own internal noise.
The "PAF" part: This is a "Phased Array Feed." Instead of moving the whole telescope to look at different spots, this device can listen to 72 different spots in the sky simultaneously.

The goal of this paper was to test if this new "super-eye" works as promised and to figure out the best way to clean up the static (noise) so we can hear the faintest cosmic whispers.

Part 1: Testing the New Ear (The Results)

The researchers pointed the telescope at two famous cosmic neighborhoods to see how well the new system performed:

The Large Magellanic Cloud (LMC): A small galaxy right next to our own Milky Way.
NGC 6744: A beautiful spiral galaxy similar to our own.

What they found:

Super Clear: The new system is incredibly quiet. It can detect gas clouds that were previously invisible.
The "Missing" Gas: When they compared their new data with old data from the same telescope (using the old single-horn receiver), they found something surprising. The new system saw a faint, diffuse layer of gas around the LMC that the old system missed.
- The Analogy: Imagine looking at a foggy forest with an old pair of glasses. You see the big trees, but the thin mist is invisible. The new cryoPAF is like a pair of high-definition glasses that reveals the mist. The old system likely "over-cleaned" the image, accidentally wiping out that thin mist while trying to remove the background fog.
Speed: Because it has 72 ears working at once, it can map the sky much faster than the old single-ear system.

Part 2: Cleaning Up the Static (The Noise Problem)

Radio telescopes are like radios in a busy city. They pick up not just cosmic signals, but also:

RFI (Radio Frequency Interference): Signals from cell phones, satellites, and GPS.
Continuum Noise: Constant "hiss" from the sky itself.

The researchers wanted to know: How do we remove this static without accidentally deleting the faint cosmic signals we are trying to find?

They tested several mathematical "cleaning" methods using a technique called SVD (Singular Value Decomposition).

The Analogy of the Cleaning Methods:
Imagine you have a messy room (the data) with:

Dust bunnies (random noise).
A bright red ball (a strong radio signal from a satellite).
A tiny, rare blue marble (the faint cosmic signal you want to keep).

The researchers tested different ways to clean the room:

Method A (2D SVD): This method looks at the room one floor at a time (like looking at a 2D photo of the room). It's good at removing the dust, but it's a bit clumsy. If you tell it to clean "hard enough," it often sweeps the blue marble right out the window along with the dust.
Method B (Tensor SVD / 3D SVD): This method looks at the room as a full 3D volume (like walking through the room). It understands that the blue marble has a specific shape in 3D space. It can remove the dust and the red ball while keeping the blue marble safe.

The Winner:
The paper found that the 3D (Tensor) methods were much better at keeping the faint signals while removing the noise.

When the "noise" was low, both methods worked okay.
When the "noise" was high (like a storm of static), the 2D methods kept deleting the signal they were supposed to find. The 3D methods, however, were robust and kept the signal intact.

Summary of Key Takeaways

The Hardware Works: The new cryoPAF on the Parkes telescope is a massive success. It is sensitive, fast, and can see faint gas clouds that older equipment missed.
The Math Matters: To get the most out of this powerful new tool, you can't just use old cleaning methods. You need 3D mathematical techniques (Tensor SVD) that understand the full shape of the data.
Protecting the Signal: If you use the wrong cleaning method, you might accidentally throw away the very faint signals you are hunting for (like the "missing ordinary matter" of the universe). The new 3D methods act like a fine-toothed comb that removes the tangles without pulling out the hair.

In short, the team has built a better telescope and figured out the best way to polish the lens so we can see the universe more clearly than ever before.

Technical Summary: Murriyang Cryogenic Phased Array Feed: Spectral-Line Results and Noise-Reduction Methods

Problem Statement
Radio astronomy faces significant challenges in detecting weak cosmological signals, such as the Epoch of Reionisation (EoR) and HI intensity mapping, due to contamination from receiver noise, foreground/background continuum sources, and Radio Frequency Interference (RFI). Traditional single-dish receivers often lack the field of view (FoV) and bandwidth required for efficient large-scale surveys. While Phased Array Feeds (PAFs) offer increased FoV and beam flexibility, many existing implementations (e.g., ASKAP) operate at ambient temperatures, resulting in higher system temperatures ( $T_{sys}$ ). Conversely, cryogenic receivers offer low noise but traditionally lack the multi-beam capabilities of PAFs. Furthermore, extracting weak signals from high-dimensional datasets (simultaneous beams, time, and frequency) often requires data reduction techniques that risk signal loss when aggressively removing foregrounds and RFI.

Methodology
This paper presents the commissioning results of the Murriyang cryogenic phased array feed (cryoPAF) on the Parkes telescope. The instrument operates in the 0.7–2 GHz range with up to 72 simultaneous dual-polarisation beams and a system temperature of approximately 17 K at 1.4 GHz.

Observations and Calibration:
- Targets: Observations were conducted on the Large Magellanic Cloud (LMC) and the nearby galaxy NGC 6744.
- Calibration: Beamformer weights were calculated using maximum signal-to-noise ratio (maxSNR) techniques based on covariance matrices measured toward Virgo A. System temperature was derived relative to dish aperture efficiency ( $\eta_d$ ) and main-beam efficiency ( $\eta_{mb}$ ) using flux calibrators (PKS B1934-638) and Galactic HI sources (S9).
- Data Processing: Data were recorded in SDHDF format. Pre-conditioning steps included gain and bandpass calibration, subtraction of median spectral and temporal baselines, spectral smoothing, and normalization.
Signal Injection and Noise Reduction Techniques:
To evaluate noise reduction methods without relying on unknown cosmological signals, artificial signals were injected into the calibrated data:
- Compact Sources: Point sources injected at central positions in frequency, time, and beam space.
- Extended Sources: A mock HI intensity map derived from the GAMA G23 galaxy survey, scaled to simulate low signal-to-noise ratio (S/N) conditions.
- Algorithms: Six Singular Value Decomposition (SVD) based methods were compared:
  - Standard 2D SVD.
  - Robust 2D SVD variants: L1-norm minimization (L1SVD), Clipped SVD (CSVD), and Stacked CSVD (CSVDstack).
  - Higher-order (Tensor) SVD: Tucker decomposition (TuckerSVD) and Canonical Polyadic decomposition (CPSVD).

Key Contributions and Results

Instrument Performance:
- Within 0.3 degrees of the optical axis, the cryoPAF achieved a ratio of system temperature to dish aperture efficiency ( $T_{sys}/\eta_d$ ) of 25.5 K and to main-beam efficiency ( $T_{sys}/\eta_{mb}$ ) of 21.1 K.
- Derived efficiencies are $\eta_d \approx 0.7$ and $\eta_{mb} \approx 0.8$ , consistent with a $T_{sys}$ of 17 K.
- Spectral-line observations of NGC 6744 and the LMC showed excellent agreement with previous multibeam observations in velocity and flux density. The cryoPAF spectra were notably flatter, requiring only linear baseline fits compared to the high-order polynomial fits needed for previous narrow-band data.
Discovery of Low-Column-Density Gas:
- A comparison between the new cryoPAF LMC data and archival multibeam data revealed a systematic offset. The cryoPAF detected extended emission at low brightness temperatures ( $0.1 \sim 0.3$ K) that was absent in the archival data.
- This suggests the previous observations missed an extended component of low-column-density gas ( $\sim 8 \times 10^{18}$ cm $^{-2}$ ), likely due to signal loss from over-subtraction of spectral and spatial baselines in the older processing pipeline. This implies the total HI mass of the LMC may be higher than previously measured.
Evaluation of Noise-Reduction Methods:
- Compact Sources: Traditional 2D SVD techniques (SVD, L1SVD, CSVD) effectively suppressed noise but suffered from severe signal loss; injected compact sources were almost entirely removed when more than 2 singular values were truncated. In contrast, higher-order tensor SVD methods (CPSVD and TuckerSVD) retained the signal while flattening the background, leveraging the source's compactness in the third dimension (beam space).
- Extended Sources: For extended signals (simulating cosmological intensity maps), 2D techniques recovered signals at high S/N but suppressed power at large scales (low wavenumbers). At low S/N ( $\sim 0.1$ ), only the tensor SVD methods (specifically CPSVD) permitted source recovery.
- Robustness: While robust methods (L1, clipping) reduced residuals, they did not significantly improve signal recovery over standard SVD once data were pre-conditioned. However, L1SVD performed well among 2D methods for high-noise regimes.

Significance
The paper demonstrates that the Murriyang cryoPAF is a highly sensitive instrument capable of outperforming previous multibeam receivers in both speed and sensitivity, particularly in detecting faint, extended gas components that were previously lost to baseline subtraction artifacts.

Crucially, the study establishes that higher-order (tensor) SVD techniques offer a significant advantage over traditional 2D methods for data reduction in high-dimensional radio astronomy datasets. By utilizing the full dimensionality of the data (time, frequency, and spatial beam), tensor SVD methods (specifically CPSVD) achieve superior noise reduction while minimizing signal loss for both compact and extended sources. This provides a promising, unsupervised pathway for future cosmological surveys aiming to detect weak HI intensity mapping signals in the presence of strong foregrounds and RFI, without the extensive training requirements of deep learning approaches.

The Big Picture: A New "Super-Eye" for the Parkes Telescope

Part 1: Testing the New Ear (The Results)

Part 2: Cleaning Up the Static (The Noise Problem)

Summary of Key Takeaways

More like this