🔭 astrophysics

Slow focus sensor for the Keck I laser guide star adaptive optics system using focal plane wavefront sensing

This paper presents the development and on-sky validation of a Gerchberg-Saxton-based focal plane wavefront sensing technique for the Keck I telescope, which successfully tracks slow focus errors induced by sodium layer drift to eliminate the need for natural guide stars and improve adaptive optics sky coverage.

Original authors: Rafael M. Salgueiro, Carlos M. Correia, Benoit Neichel, Antonin Bouchez, Peter Wizinowich, Avinash Surendran, Max Service, Thierry Fusco, Cédric Taïssir, Pierre Jouve

Published 2026-02-18

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Rafael M. Salgueiro, Carlos M. Correia, Benoit Neichel, Antonin Bouchez, Peter Wizinowich, Avinash Surendran, Max Service, Thierry Fusco, Cédric Taïssir, Pierre Jouve

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 you are trying to take a crystal-clear photo of a distant star using a giant telescope. The problem is that the Earth's atmosphere is like a wobbly, shimmering heat haze over a hot road. It distorts the starlight, making the image blurry and shaky.

To fix this, astronomers use a technology called Adaptive Optics (AO). Think of it as a "smart mirror" that bends itself hundreds of times a second to cancel out the atmospheric wobble, sharpening the image instantly.

However, there's a catch. To make the mirror bend correctly, the telescope needs a reference star to look at. Usually, they use a bright natural star nearby. But bright stars are rare; it's like trying to find a specific streetlight in a dark city—you can only see a tiny fraction of the sky.

The Laser Solution (and its Flaw)

To see more of the sky, astronomers invented Laser Guide Stars. They shoot a laser into the sky to excite sodium atoms about 90 kilometers up, creating an artificial "star" right where they need it. This solves the "bright star" problem, allowing them to look at almost anywhere.

But this new laser star has a glitch. The sodium layer in the sky isn't a solid floor; it's a floating cloud that drifts up and down slowly, like a balloon rising in the wind. Because the distance to this "star" keeps changing, the telescope's focus gets slightly blurry over time. This is called Slow Focus Error.

The Old Way vs. The New Way

The Old Way (The 20x20 Grid):
Currently, the Keck telescope fixes this drift using a special sensor that looks at a real, faint natural star nearby. It's like trying to keep a camera focused by looking at a dim lightbulb in the distance. The sensor is a bit clunky (a 20x20 grid of tiny lenses), and because the natural stars are so faint, the system has to wait a long time to get a clear reading. This delay means the focus is always a little bit "behind" the reality, like trying to catch a moving train by looking at it through a foggy window.

The New Way (The "Focal Plane" Trick):
This paper introduces a smarter, faster way to fix the focus. Instead of using a separate sensor with a grid of lenses, the team decided to use the camera itself that takes the final picture.

Imagine you are trying to guess how far away an object is just by looking at how blurry it is on a piece of paper. If the blur is stretched horizontally, the object is too close. If it's stretched vertically, it's too far. The Keck telescope's camera naturally has a tiny bit of "astigmatism" (a built-in lens imperfection) that acts like a ruler. When the focus drifts, the star's image stretches in a specific way.

The team developed three different "mathematical detectives" (algorithms) to look at these stretched images and instantly calculate exactly how much to adjust the focus:

Gerchberg-Saxton (GS): A classic, iterative detective that guesses and checks repeatedly.
LiFT: A linear, fast calculator.
Gaussian Fit (Gf): A simple shape-fitter that measures the stretch.

The Race to the Finish

The researchers tested these three detectives in computer simulations and on a telescope bench. They found that while all three could do the job, Gerchberg-Saxton (GS) was the champion.

Why? Because in the real world, the atmosphere is messy. When there are "high-order residuals" (extra jitters and turbulence), the other two detectives got confused and started making mistakes. The GS detective, however, was like a seasoned sailor in a storm; it stayed calm, stable, and kept giving the right answer even when the conditions were terrible.

The Big Win

By using this new method, the telescope can now:

See fainter stars: It doesn't need a bright natural star nearby anymore. It can use much dimmer ones, effectively increasing the "sky coverage" by a massive amount (imagine going from seeing 1% of the sky to seeing 100%).
React faster: Instead of waiting 30 seconds to check the focus, it can check it every 1 second. This eliminates the "lag," making the images sharper.
Work without new hardware: The best part? They didn't have to build a new machine. They just used the existing camera and wrote new software.

The Bottom Line

The team successfully tested this on the Keck I telescope in Hawaii. They intentionally messed up the focus and watched the new system fix it in real-time, even under bad weather conditions.

In simple terms, they taught the telescope to "feel" its own focus using the camera's own blurry pictures, making it faster, more accurate, and able to explore much more of the universe than ever before. It's a huge step toward clearer views of the cosmos, not just for Keck, but for future giant telescopes too.

1. Problem Statement

Context: Ground-based telescopes use Laser Guide Stars (LGS) to overcome the limited sky coverage of Natural Guide Star (NGS) Adaptive Optics (AO) systems. However, the sodium layer (at ~90 km altitude) used to create LGSs experiences slow altitude drifts, causing a "slow focus error" in the optical system.
Current Limitation: The Keck I telescope currently corrects this error using a Shack-Hartmann (SH) wavefront sensor (WFS) with 20x20 sub-apertures (or a 5x5 mode for fainter stars). This approach has significant drawbacks:

Limited Sky Coverage: It requires relatively bright NGSs, restricting the usable sky area.
High Lag Error: The low bandwidth of the SH sensor (limited by faint star brightness) results in slow correction rates, leading to significant temporal lag errors.
Inefficiency: The current system utilizes only 10% of the visible light for focus sensing, discarding the rest.

Goal: Develop a new slow focus tracking technique using Focal Plane Wavefront Sensing (FPWFS) via the existing Near-Infrared (NIR) Tip-Tilt (TT) sensor (TRICK). This aims to increase sky coverage (by using fainter stars), increase correction frequency (reducing lag error), and require no hardware modifications.

2. Methodology

The authors investigated three distinct FPWFS algorithms to retrieve the focus phase ( $Z_4$ ) from focal plane images:

Gerchberg-Saxton (GS): A classical iterative phase retrieval algorithm that alternates between the pupil and focal planes, applying constraints (pupil mask and amplitude).
Linearized Focal Plane Technique (LiFT): A maximum likelihood estimation method assuming Gaussian noise statistics, utilizing a small phase approximation for low-order mode sensing.
Gaussian Fit (Gf): A simplified approach fitting a 2D Gaussian to the focal plane image. It exploits the fact that astigmatism (present naturally in the TRICK beam) causes the point spread function to stretch differently depending on the focus sign.

Experimental Setup & Conditions:

Hardware: Keck I LGS-AO system using the TRICK NIR sensor (H and K bands).
Challenges: The system operates under "challenging" conditions:
- Sub-Nyquist Sampling: TRICK pixels are ~50 mas, which is <0.5 Nyquist for H and K bands.
- Low Signal-to-Noise Ratio (S/N): Due to the faintness of available NGSs.
- High-Order Residuals: Atmospheric turbulence residuals that increase with angular separation between the LGS and NGS.
Calibration: A critical step involved "diversity calibration" using in-focus images to characterize Non-Common Path Aberrations (NCPAs) and the natural astigmatism (~200 nm rms) to serve as the initial guess for the algorithms.

3. Key Contributions

Algorithmic Comparison: A comprehensive characterization of GS, LiFT, and Gf specifically for the Keck I LGS-AO context, evaluating linearity, computational cost, and robustness against noise and residuals.
Handling Sub-Nyquist Sampling: Demonstrated that GS and LiFT can maintain linearity and accurate focus estimation even when the imaging model assumes 1.0 Nyquist sampling while the actual data is <0.5 Nyquist (requiring a stable gain factor).
Robustness to Residuals: Identified that the presence of high-order residuals significantly degrades LiFT and Gf, whereas GS remains stable.
Operational Validation: Successfully executed closed-loop on-sky tests using the GS algorithm on the Keck I telescope, proving the feasibility of replacing the SH WFS with FPWFS.

4. Results

Numerical Simulations:

Linearity: All three algorithms showed linear behavior within $\pm 200$ nm rms.
Limiting Magnitude (H-band):
- LiFT: $m \approx 20$
- Gf: $m \approx 19.5$
- GS: $m \approx 18$
- Note: While LiFT and Gf could theoretically see fainter stars, they were more sensitive to noise and residuals.
Computational Cost: All algorithms could perform multiple focus estimates per second, allowing for correction rates of 1 Hz or faster.

Bench Tests (Keck AO Bench):

Open Loop: All algorithms tracked focus ramps well.
With High-Order Residuals: GS demonstrated superior stability.
- GS: RMS error $\approx 33$ nm, $R^2 = 0.74$ .
- Gf: RMS error $\approx 66$ nm, $R^2 = 0.40$ .
- LiFT: RMS error $\approx 102$ nm, $R^2 = 0.22$ .
Conclusion: GS was selected as the primary candidate for on-sky deployment due to its robustness against high-order residuals, a critical factor for real-world AO systems.

On-Sky Tests (February 8, 2025):

Conditions: H-band, bright TT star ( $m_V=10$ ), challenging seeing ( $\sim 1''$ in K-band).
Performance: The GS algorithm successfully operated in a closed loop at 1 Hz.
- It correctly converged from an initial focus error to zero.
- It successfully recovered and compensated for intentionally introduced focus errors (step changes).
- The retrieved focus values remained stable within $\pm 100$ nm rms despite poor seeing conditions, confirming the algorithm's robustness.

5. Significance and Impact

Sky Coverage Improvement: By utilizing FPWFS, the system can track focus using much fainter NGSs. The paper estimates a magnitude gain of $\Delta m \approx -8.7$ compared to the current Low-Bandwidth WFS (LBWFS), significantly expanding the accessible sky area.
Lag Error Reduction: The ability to use higher S/N from the full pupil allows for faster integration times.
- Example: For a magnitude 16 star, the current LBWFS requires ~30s integration (lag error $\sim$ 50 nm). FPWFS (GS) could achieve this in ~1s, reducing lag error to $\sim$ 13 nm (a reduction of $\sim 37$ nm).
Scalability: This proof-of-concept is directly applicable to current 8-10m class telescopes and future Extremely Large Telescopes (ELTs), where slow focus tracking is even more critical due to the $D^2$ scaling of focus errors.
Operational Readiness: The successful closed-loop on-sky tests represent a major step toward fully operationalizing FPWFS for slow focus tracking at Keck I, potentially replacing the legacy SH-based sensor in the coming months.