Hubble Space Telescope one of 16 suffering data-scrambling sensor error

Flawed analogue-to-digital converter can turn a whole bunch of 1s into a cosmic joke

Users of sixteen of the world's most prestigious optical telescopes - including the Hubble Space Telescope - are revisiting old data in case an analogue-to-digital converter design has polluted the instruments' measurements.

Analogue-to-digital converters (ADCs) are where the real world meets the digital, and in the case of the optical telescopes, they convert pixel data from the charge-coupled devices (CCDs) in the instruments' cameras.

An international group of researchers trying to identify what looked like a systematic error in an instrument in Hawaii discovered the ADC's reference voltage could be affected by the ADC's output - in other words a feedback error.

That's bad news for astronomers using the sixteen instruments, which the researchers found are affected by the bug. The sketchy scopes include LRIS and DEIMOS on the Keck telescopes, WFC3-UVIS and STIS on the Hubble Space Telescope, MegaCam on the Canada France Hawaii Telescope, SNIFS on the University of Hawai'i 88-inch telescope telescope, GMOS on the Gemini telescopes, HSC on Subaru, and FORS on Europe's Very Large Telescope.

In this paper at arXiv, Kyle Boone and seven collaborators from the US, Canada and France needed to identify a strange error in data from the SuperNova Integral Field Spectrograph (SNIFS) instrument in Hawaii. The clue they followed was that the error seemed to be proportional to the number of "1" bits in a measurement - in other words, 1111 would produce a bigger error than 0001.

The error seemed to be proportional to the number of "1" bits in a measurement

They found an artefact in the electronics chain of commonly-used CCD systems which causes crosstalk between pixels (as the paper says: "the binary-coded output of the analog-to-digital converter (ADC)" affects subsequently read-out pixels).

In their paper, the boffins say the errors could be in the order of 4.5 analog-digital units (ADU) per pixel. The ADU is the unit for measuring CCD pixel values - it lets an astronomer work out how many photons struck that CCD pixel.

In the SNIFS data where the effect was first noticed, "a single ADU difference in the driver pixel value can correspond to a difference of up to 2.5 ADU in the mean of the residuals of the pixel read out 2 pixels afterwards".

The paper drily calls a 2.5 ADU error in an instrument designed to have measurement uncertainties of 0.05 ADU "highly statistically significant".

Their proposed cause is feedback into the analogue-digital converter's reference voltage. ADCs are "highly sensitive to small changes" in the reference voltage, they explain:

"One ADU for a 16-bit ADC corresponds to a change of only 0.0015 percent. When earlier pixels are read out, the ADC outputs their binary representations and stores them temporarily. We propose that these charges (representing the ones in the binary code) then introduce a slight offset into the reference voltage when they are released, and the next pixel is read out."

Astronomers who've used the affected instruments should take a look at their data if they work with low background noise or low signal level images, or working with data where there's "a background level just below a large binary transition".

Check your data

The researchers have published software to detect and correct errors in existing datasets, and in future, they write, changes to hardware designs can easily eliminate the problem. For example, giving the ADC a differential input rather than a single reference voltage should reduce the chance of it being affected by a stray signal.

Vulture South asked ANU astronomer Dr Brad Tucker at Mt Stromlo Observatory how significant the effect is likely to be. He told us in an e-mail that effects will vary depending on the instrument and the type of measurements taken, but that the effect is most important for astronomers making precision measurements of faint sources.

"If you have a bright star, this will have less effect than a faint galaxy," he said. "For those working on areas like supernova cosmology (the field a number of the authors work in), getting a precision measurement is key to then measuring a distance to the supernova," he continued.

The ADC design the paper investigated is very common, he said, and in Australia "people are actively looking" at whether Anglo-Australian Telescope data could be affected.

Tucker added that it's feasible some Kepler observations might also be impacted.

"One thing to note is Kepler uses very large, under sampled pixels, with the aim that all of the starlight falls on only a few pixels. Therefore, this issue would not be as large as other cameras. But as I mentioned above, the effect would be worse for fainter (or … low signal-to-noise) sources," he wrote.

Don't worry, though, the universe isn't going to lose a bunch of exoplanets. Rather, Tucker wrote, fainter stars and planets "might have a bias in the properties people have measured". ®




Biting the hand that feeds IT © 1998–2018