🔭 astrophysics

Analysis of Angular-Differential Post-Processing Algorithms for Exoplanet Direct Detection with a Photonic Lantern Nuller

This paper reformulates Angular Differential Imaging (ADI) using principal component analysis to adapt the technique for the Photonic Lantern Nuller's non-rotationally invariant data, demonstrating that injecting an antiplanet signal before stellar estimation effectively mitigates self-subtraction and improves exoplanet localization at close separations.

Original authors: Suvinay Goyal, Yinzi Xin, Nemanja Jovanovic, Dimitri Mawet, Michael P. Fitzgerald

Published 2026-07-03

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Suvinay Goyal, Yinzi Xin, Nemanja Jovanovic, Dimitri Mawet, Michael P. Fitzgerald

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

Imagine trying to take a photo of a tiny, glowing firefly sitting right next to a blindingly bright spotlight. In the world of astronomy, the "firefly" is an exoplanet, and the "spotlight" is its parent star. The star is so bright that its glare usually drowns out the planet completely.

For decades, astronomers have used special sunglasses called coronagraphs to block the star's light. However, these sunglasses have a blind spot: they can't see planets that are too close to the star.

Enter the Photonic Lantern Nuller (PLN). Think of the PLN not as a pair of sunglasses, but as a high-tech "noise-canceling" device for light. It uses a special fiber-optic bundle to split the star's light into different channels. Inside these channels, the star's light waves are made to crash into each other and cancel out (like noise-canceling headphones), while the planet's light slips through. This allows astronomers to see planets much closer to their stars than ever before.

The New Problem: A Moving Target
The catch with the PLN is that it doesn't take a normal 2D picture like a camera. Instead, it gives a very short list of numbers (a 1D signal) representing how much light came through four specific channels.

When astronomers use traditional methods to clean up the data, they rely on the fact that the planet stays in the same shape while the telescope rotates. But with the PLN, as the telescope rotates, the planet's signal changes shape in a weird, unpredictable way. It's like trying to recognize a friend in a crowd where their face keeps morphing every time you turn your head. Traditional cleaning tools get confused and accidentally erase the planet along with the star's glare. This is called self-subtraction.

The Solution: The "Anti-Planet" Trick
The authors of this paper developed a new way to clean the data, which they call Antiplanet KLIP Subtraction. Here is how it works, using a simple analogy:

Imagine you are trying to hear a whisper (the planet) in a room full of loud, shifting chatter (the star's glare).

The Old Way: You try to guess what the chatter sounds like and subtract it. But because the whisper changes shape as you move, your guess is wrong, and you end up subtracting the whisper too.
The New Way (Antiplanet): Before you try to subtract the chatter, you pretend to inject a "fake whisper" (an anti-planet) that is the exact opposite of what you think the real whisper looks like.
You then let the computer figure out what the chatter sounds like while this fake whisper is there. Because the fake whisper is there, the computer is forced to be very careful not to erase it.
Once the computer learns the chatter, it subtracts it. Since the computer was careful not to erase the fake whisper, it also didn't erase the real whisper.

What They Found
The team tested this new method using computer simulations (fake data) to see if it could find faint planets better than the old methods.

Better Detection: The "Anti-Planet" method was the clear winner. It could detect much fainter planets (lower "detection limits") than the other methods, even when the telescope only rotated a small amount.
Better Location: It was also much better at pinpointing exactly where the planet was, how bright it was, and its angle.
The Catch: The method is so sensitive that if the telescope doesn't rotate enough (less than 25 degrees), the computer might get slightly confused about the planet's exact location. However, even in this tricky scenario, it still performed better than the other methods for finding brighter planets. If the telescope rotates more (100+ degrees), the method becomes perfect.

The Bottom Line
This paper proves that by changing how we process the data—specifically by using a "fake planet" trick to protect the real one—we can use the Photonic Lantern Nuller to find and study planets that are incredibly close to their stars. This opens up a new window for discovering worlds that were previously hidden in the glare.

Technical Summary: Analysis of Angular-Differential Post-Processing Algorithms for Exoplanet Direct Detection with a Photonic Lantern Nuller

Problem Statement
Direct imaging of exoplanets, particularly those orbiting close to their host stars (within the inner working angle of standard coronagraphs), relies heavily on post-processing to suppress stellar speckles caused by wavefront errors. While Angular Differential Imaging (ADI) is a standard technique for conventional coronagraphs, it relies on two critical assumptions that fail when applied to the Photonic Lantern Nuller (PLN):

Rotational Invariance: Conventional ADI assumes the planet's point-spread function (PSF) remains rotationally invariant as the sky rotates relative to the instrument. However, the PLN produces a one-dimensional (1D) vector of modal intensities rather than a 2D image. The signal's variation with rotation is governed by the interferometric response of the lantern, not a simple geometric rotation.
Minimal Overlap: Standard ADI assumes the planet signal overlaps minimally with residual starlight at high separations. The PLN operates at very small separations ( $\sim 1\lambda/D$ ), where the planet signal significantly overlaps with residual stellar light, leading to severe "self-subtraction" effects where the planet signal is partially subtracted along with the starlight during standard processing.

Consequently, existing ADI algorithms cannot be directly applied to PLN data without reformulation to account for non-rotationally invariant signals and strong self-subtraction.

Methodology
The authors reformulate the ADI framework using Principal Component Analysis (PCA) to process simulated 1D PLN data. The study utilizes the hcipy package to generate synthetic datasets containing correlated stellar noise (simulated via evolving phase aberrations) and rotating companion signals.

Three specific algorithms were developed and compared:

Direct Fitting: A baseline method that fits a guessed planet signal to the input intensity and calculates the cost as the mean squared stellar residuals, without explicit stellar subtraction.
KLIP Subtraction: A modified application of the Karhunen-Loève Image Projection (KLIP) algorithm. It uses Singular Value Decomposition (SVD) on reference frames to estimate and subtract the stellar signal before fitting the planet.
Antiplanet KLIP Subtraction: A novel variation designed to mitigate self-subtraction. Before estimating the stellar signal, a "fake anti-planet" signal (equal and opposite to the guessed planet) is injected into the data. The stellar signal is then estimated from these residuals. This prevents the aggressive subtraction of the planet signal itself during the stellar modeling phase.

The performance of these algorithms was evaluated across three sky rotation ranges ( $\Delta\theta = 25^\circ, 100^\circ, 150^\circ$ ) using Monte Carlo-based detection testing. Metrics included True Positive Rate (TPR), False Positive Rate (FPR), Receiver Operating Characteristic (ROC) curves, and the accuracy of spatial parameter localization (separation $r$ , flux ratio $\epsilon$ , and initial angle $\Theta$ ).

Key Contributions

Algorithmic Reformulation: The paper provides the first reformulation of ADI for non-rotationally invariant, 1D interferometric data, moving away from geometric derotation to fitting signals in measurement space.
Self-Subtraction Mitigation: The introduction of the "Antiplanet" injection technique specifically addresses the high self-subtraction inherent to PLN data at small separations.
Statistical Benchmarking: The study establishes detection limits and localization accuracies for PLN data under varying sky rotation conditions, providing a quantitative basis for future instrument operation.

Results

Detection Sensitivity: The Antiplanet KLIP Subtraction method consistently outperformed both the Direct Fit and standard KLIP Subtraction methods. It achieved the lowest detection limits (faintest detectable flux ratios) across all tested sky rotation ranges ( $25^\circ, 100^\circ, 150^\circ$ ). For instance, at $\Delta\theta = 150^\circ$ , the Antiplanet method achieved a detection limit of $\epsilon \approx 0.009$ , compared to $0.01$ for standard KLIP and $0.025$ for Direct Fit.
Localization Accuracy: The Antiplanet method demonstrated superior spatial localization. At $\Delta\theta = 100^\circ$ and $150^\circ$ , it achieved perfect localization (recovering exact input parameters) across all tested datasets. Even at the more challenging $25^\circ$ coverage, it correctly localized 5 out of 10 datasets at its own detection limit, outperforming the other methods on the same difficult datasets.
Robustness to Localization Errors: When detection tests were repeated using imperfectly localized parameters (simulating real-world uncertainty), the Antiplanet method showed sensitivity to errors at low sky coverage ( $25^\circ$ ). Because it probes fainter signals, slight deviations in localization led to "bad fits" that degraded detection performance more significantly than for the other methods. However, for brighter planets and higher sky coverages ( $100^\circ, 150^\circ$ ), the method remained the most robust and sensitive.

Significance
The paper claims that the Antiplanet KLIP Subtraction method is the most effective approach for constraining the on-sky position and detecting faint signals from a PLN. By adapting ADI to the unique 1D, non-rotationally invariant nature of photonic lantern data and explicitly addressing self-subtraction, this method enables the detection of exoplanets at separations ( $0.5\text{--}1.5 \lambda/D$ ) previously inaccessible to standard coronagraphic post-processing. The authors note that while the method is highly sensitive, maximizing sky coverage is recommended to mitigate localization uncertainties. The framework presented is also applicable to other instruments producing non-rotationally invariant off-axis PSFs. Future work by the authors aims to extend this angular differential framework to multi-wavelength (spectral) data.

More like this