Articles in this series:
On-sky Photometry (this page)
Having used the CMOS Slug for several months now for exoplanet observations it is time to review its actual on-sky performance.
Actual precision achieved
Here’s a lightcurve for the exoplanet candidate TIC209707800-01, observed on 2021-06-26, together with an image of the field below it.
The image is clean, as I’ve come to expect of the camera. Flat fielding and dark subtraction have worked as expected. The binning strategy I explained here is now standard for PEST so all images, including the one above, is binned 2×2 off-camera. Individual exposures were 120s, through an Rc filter. Total observation duration was nearly 7 hrs and airmass varied between 1.1 to 1.8 towards the end of the observation. Guiding performance was good, for my mount, with excursions of 35 pixels (25″) in RA and 12 pixels (18″) in Dec. The target star is Vmag = 11.9.
The lightcurve shows a nice ~10ppt depth event which looks like an exoplanet transit. There are also no strong systematics or breaks as would happen if guiding and flat-fielding was poor.
To quantify the precision achieved in this observation I should model the transit then calculate the standard deviation (SD) from the residuals. I don’t yet have a software tool to model transits so instead I removed the in-transit points. The achieved precision is then the SD of the remaining points.
The per 120s point precision achieved was 2.5ppt, or 2.7mmag (0.0027 mag). The red points are averages over 1 hour. The SD of these was just 0.4ppt (0.0004 mag).
These figures are very good. To show just how good they are, it is necessary to compare actual and theoretical performance. Are there sources of error or noise that are not accounted for?
Theoretical precision
I use a spreadsheet model for photometric precision that takes into account the following parameters:
- Telescope aperture and focal ratio
- Sensor QE, pixel size, well depth, gain, read noise
- Exposure and image download times
- Star and sky fluxes calibrated to my sky conditions, typical FWHM
- Photometry apertures, number of raw frames stacked for master flats and darks
The model works out the noise added in each step of the photometric reduction, starting with star photon shot noise (i.e. the best theoretical precision for a given star and optical system), then including:
- Read noise, dark subtraction and sky level subtraction (for this sensor, read and dark noise are insignificant)
- Flats. I use 200 raw flats in each master to get high S/N (~2800)
- Differential photometry with an ensemble of stars. I assume the ensemble delivers ~10 times the flux of the target. The actual ensemble I used has 18 stars with a range of brightnesses.
- Atmospheric scintillation. Reference the AAVSO CCD Observing Manual, Sect 4.4.
Below is the model result in terms of cumulative error (i.e. achievable precision), as noise sources are included from left to right. The rightmost bar is the result with all noise sources included.
The flux from the target star was ~90,000 ADU (native 12-bit). Multiplying by gain gives flux of ~340,000 -e. The shot noise is the square root of this = 583 -e, so the theoretical limit on precision is 583/340000 = 1.7 ppt, or 1.9 mmag.
The main contributors to noise on top of this are sky subtraction, and scintillation. Sky noise is the result of shot noise from background sky photons. This observation was done close to a full moon, so the sky was quite bright. Scintillation depends on observatory altitude and telescope aperture. The AAVSO formula does not take altitude into account, however.
So the calculated precision is 2.8 mmag compared to 2.7 mmag actually achieved. I am not surprised that there is a small difference, because many of the parameters in the calculation are estimated and averaged. For example both airmass and flux varied through the observation.
The point to take away, though is that the actual outcome is not worse than the theoretical. This suggests that there are not other sources of noise that have crept in, e.g. RTS pixels, or non-linearity.
Comparison with ST8-XME
How does this compare with my previous camera, the ST8-XME? Below is the same calculation, but using that other camera’s parameters.
Photometric precision would have been better by about 0.3 mmag. This is almost totally driven by the lower shot noise, only 1.5 mmag vs 1.9 mmag for the QHY183M. In turn, this is because a higher flux would have been captured by the ST8-XME, ~525,000 -e vs only ~340,000 -e.
Flux and precision in other filterbands
Why is this? After all the IMX183 sensor is back side illuminated and has a high quoted QE of 84% which is comparable to that of the KAF-1603ME sensor in the ST8-XME.
The passband of the Rc filter is centered on 647 nm. At this wavelength the QE of the IMX183 sensor is just 53%. For the KAF-1603ME at the same wavelength it is 75%. Perhaps the way to get better precision from the QHY183M is to observe in the V band, centered on 551 nm (QE = 80%) or B band, 445 nm (QE = 81%)?
To test this, I observed several stars with different colours and measured the flux detected on the IMX183 in B, V, Rc and Ic.
In no case does the flux in any other band exceed that in Rc. Not even with quite a blue star. The B band flux is much lower than Rc in all cases. So with this sensor the highest precision will be in the Rc band (or equivalent). In addition sky brightness is usually lower towards the red, increasing the advantage of observing in Rc.
Conclusions
The QHY183M performs as expected for high precision photometry, with precision limited mainly by star and sky photon shot noise, and scintillation. There is no detected impact from possible ‘CMOS effects’, such as RTS pixels, amp glow and inconsistent darks.
On sky though, the photometric performance is not as good as may be expected from the high quoted peak QE. This is because this peak is towards the bluer part of the spectrum whereas most stars produce more energy in the red, where the IMX183 is less sensitive.