Aussies de-blur Webb

The Australian hardware and software de-blurring images on the James Webb Space Telescope

By Dirk Goës
Note: A replica of the Aperture Masking Interferometer (AMI) is currently on display at the Powerhouse Museum, Parramatta, as part of the Task Eternal exhibition.
Unknown to most Australians, the now famous James Webb Space Telescope (JWST) contains an Australian designed hardware component which allows the telescope to function as an interferometer. It is called the Aperture Masking Interferometer (AMI), and it allows the telescope to pick out very fine details on or around bright objects. For example, an exoplanet orbiting a distant star.
When the AMI was activated, it worked as designed. However, the underlying infrared camera which turns light into high resolution images was leaking pixels and creating blur. While this is a known limit of current imaging technology the blurring was worse than expected.
Australian researchers set out to fix this by designing a software system and machine learning model that simulates the physics of the AMI and the underlying camera. After training the machine learning model, they were able to successfully remove the blur.
For validation they compared their results with well-known objects that have been imaged from the ground using larger telescopes than the JWST. This included the correct position of two exoplanets in star system HD 206893, Jupiter’s Moon Io and the black hole jets emerging from galaxy NGC 1068.
Currently the technique works with a subset of the light and data being captured by just one of the instruments aboard JWST. But now that it has been proven to work, researchers are working to apply it to the full telescope and future space telescopes.
The Aperture Masking Interferometer
The James Webb Space Telescope (JWST) launched on Christmas day 2021 and became operational on the 12th of July 2022. Since them, it has amazed us all with stunning images and important science discoveries. It has observed distant galaxies, star systems and black holes much closer to the time of the Big Bang than previously thought possible. It can do this because it is tuned to look at infrared light, the old light from the early Universe that has stretched to the red end of the electromagnetic spectrum due to the expansion of the Universe.
Less well known is that JWST is the first space telescope capable of the technique of optical interferometry. In ground-based astronomy the technique of interferometry normally involves combining two or more telescopes into one larger virtual telescope. For example, the two Keck 10 metre telescopes on summit of Mauna Kea, Hawaii have operated as an interferometer in the past. The two telescopes are 85 metres apart (called the baseline) and when operating as an interferometer they create a virtual telescope equivalent to a mirror diameter of 85 metres. Of course, it does not collect 85 metres of light as the two telescopes only have a mirror of 10 metres each, however the large virtual diameter increases the angular resolution which determines how well fine details can be seen.
The Aperture Masking Interferometer (AMI), designed by Peter Tuhill from Sydney University, works a little differently. The AMI is a small metal disc, with seven hexagonal openings, that resides in the Near-Infrared Imager and Slitless Spectrograph (NIRISS) instrument aboard the James Webb Space Telescope (JWST). When it is selected for use it is placed in between the light coming into the telescope and the NIRISS instrument.

It only allows light through from seven of the eighteen hexagonal mirror segments on the JWST’s main mirror. The seven openings have been carefully selected to allow JWST to work as an interferometer. By combining the light from the seven openings in an interference pattern it is possible to create a higher resolution image of the observed object than would normally be possible.
This may seem counterintuitive as a large part of the light from JWST’s main mirror is being masked (11 out of 18 mirror segments) and normally more light means more detail. In this case the AMI interferometer is being used to reveal finer detail on brighter objects. In fact, interferometry is a technique ideally suited to picking out small details on bright celestial objects.
Fixing the blur
The component of an astronomical camera that takes photons of light and turns them into an electronic signal to be stored as pixels is typically called a detector. Even so the detectors aboard the JWST are some of the most advanced ever built, they are not perfect. At very high resolutions bright pixels can leak into their darker neighbours, known as the brighter-fatter effect. When AMI was first tested this effect was worse than expected, impacting research projects that required very high resolution.
When an image is captured through the AMI it does not look anything like the star, galaxy or celestial object it is aiming at. It is an interference pattern that looks like a patterned disk and is called an interferogram. There is a known algorithm for turning an interferogram into an image of the celestial object that we would recognise. However, in this case the problem was that the interferogram was blurred and needed to be corrected before converting into an image.

Australian researchers, led by Ben Pope, Louis Desdoigts, and Max Charles set out to fix this blurring effect via software and machine learning. They designed a software system that simulates the optical physics of the AMI and a machine learning model that represents the detector (in this case NIRISS) and can be trained to recognise the difference between valid pixels and blurred pixels. After training the system using test stars, they were able to successfully remove the blur from the interferogram before it is converted into an image.
To validate their results the researchers used a variety of celestial objects that had been well studied from ground-based telescopes. For example, the star HD 206893 has two companion planets whose position is known. The team pointed JWST at HD 206893 and imaged it through the AMI. The resulting interferogram was de-blurred and then converted into an image which correctly identified the location of the two exoplanets.

In two other examples the team were able to correctly identify the location of active volcanoes on Jupiter’s Moon Io and the black hole jet emerging from galaxy Messier 77 (M77), also known as NGC 1068.

The future
Currently the AMI observing mode and detector model only works with a small 80x80 pixel subsection of the NIRISS detector. But now that the Australian developed software technique has been successful, work is underway to apply it to the full NIRISS detector as well as other detectors aboard JWST. Work is also underway to apply the technique to future space telescopes that are yet to launch. The technique holds great promise for the direct imaging of exoplanets, studying exoplanet atmospheres and protoplanetary discs as well as imaging the cores of galaxies.
Notes
Thank you to Associate Professor Benjamin Pope (Macquarie University) and PhD candidate Max Charles (University of Sydney) for answering my questions regarding this article. The papers describing the research can be found below:
Desdoigts L, Pope B, Charles M, et al. Amigo: A data-driven calibration of the JWST interferometer. Publications of the Astronomical Society of Australia. 2026;43:e075. doi:10.1017/pasa.2026.10194
Charles M, Desdoigts L, Pope B, et al. Image reconstruction with the JWST interferometer. Publications of the Astronomical Society of Australia. 2026;43:e048. doi:10.1017/pasa.2026.10179

Comments