CMOS flat frame exposure duration puzzle

This post is an initial assessment of a number of different CMOS cameras in which flat fielding is a particular issue.

Some months ago, a friend returned a Kepler 4040 camera, which stubbornly refused to flat-frame effectively. After some head scratching, the proposed solution was to ensure the flat frames had the same duration as the light frames…which was not met with enthusiasm.
Rolling on to 2021 - I have a new CMOS 4/3-inch camera with an IMX492. It is a much lesser sensor than the 4040 but it too refuses to flat frame effectively.
To my astonishment, I found a 1-second exposure with an EL display had a different flat-frame pattern to a 15-second one with a ND filter over the lens. There were dark vertical bands on either side and the width of these changed with the exposure duration. Both had median values about 2/3rds along the histogram. Using the 15-second flats to calibrate my 900-second NB exposures did a better job than the 1-second. Still not perfect, so I’m going to rig up a dimmer light source and see how far I can go. (I use darkflats in all cases, rather than bias and all traces of amp glow were removed… and everything at -15C)

So, - the dark frames themselves look evenly ‘illuminated’, so this is not an offset thing, it has to be a gain thing. From my electronics background, I can only think that the sensor amplifier elements are temperature senstive and the localized heating changes with exposure duration (as the readout speed in all cases would be the same). If true, and I cannot be sure if is a design thing, or due to manufacturing variation, - this is a real PITA for astrophotographers.

Has anyone had similar experiences and thoughts?

1 Like

I had an issue with my ASI294MC-Pro. My first attempts at flats had fairly low exposure times - 0.5 secs or so - and they did not correct the lights at all well even though they looked fairly sensible on the face of it. I remember reading somewhere that there was an issue with the electronics - the gain circuitry seems to ring a bell - whereby the exposure for flats needed to be a couple of seconds. I now do my flats at the same gain and offset as the lights (always have in fact) and adjust the panel brightness to hit 30-50% with 3 second exposures and all seems to be well. I also have an ASI183MC-Pro and I do the same 3 second flats, but I have tried shorter flats and that camera seems to be quite happy at 0.5 seconds or less. I just find it easier to stick to the 3 second rule (odd, but does that not also apply when you drop food on the floor :slight_smile: ). Your problem sounds much more severe.

Terry - yes, all my exposures use exactly the same camera settings … in your case, what were your exposure times for the lights? Mine were 900 seconds and I’m hearing vague noises that long exposures with CMOS cameras simply are a bad idea. Nothing more specific… it might be referring to amp glow - which I can remove entirely with matched darks.

Chris, the longest I’ve ever taken with a CMOS sensor is 600 secs, but that was a poor choice. I now try to do a calculation by taking a 60 second light frame, computing the background signal (in e/s) and figuring out the time required to keep the read noise at 5% of that. That often comes out at 60 seconds or less unless I’m using narrowband filters when it can be 3-4 minutes. I them look at the bright stars and pick a time higher than the minimum but not so long as to saturate too many stars. That is, in part, to avoid having hundreds of subs that I would then have to process. So often I’m shooting in the 1 to 3 minute range. The starburst noise on the 294 is not so bad, the 183 is much worse, or so it appears to me. Dark do remove it well as long as they are matched to the lights, no dark frame optimisation, etc.

Yes, I was always doing 15 or 20-min subs with the QSI683wsg for NB. Maybe 5-mins is the practical limit for CMOS. I few experiments will confirm. I have the same approach to exposure, just sufficient to clip one or two pixels in the brightest stars.

Terry and buzz:

On my QHY183M I made a photon transfer curve and was surprised to discover that, with set gain and offset, the raw signal level actually goes down from (near) 0s exposure to about 0.5s exposure and then rises as expected as the exposure time increases. That funny dip around 0.5s concerned me that something was going on with this sensor and/or electronics that I did not understand. As a result of this, I decided not to take flat exposures at anything less than 0.5s, and I usually shoot for exposure of 2s or more with my flat panel.

I’d be curious how the other sensors you mention behave with such short exposures.

I have not had any issues flat fielding with this sensor nor with calibrating out the impressive star burst amp glow. However, I am familiar with your comments about star saturation.

I have a QHY268M on order and hope to see its larger full well and wider dynamic range perform better.


This is not unusual. Unlike CCDs, many CMOS sensors have in built primitive dark subtraction that kicks in for ‘long’ exposures. It basically subtracts the mean value of shaded pixels in the margins from the image pixels. It is not exact and quite a few sensors’ mean dark value reduce with exposure duration. It is for that reason you do not use bias frames in CMOS calibration, but dark flats and darks. This, however, is not my issue. In my case, the flat ‘pattern’ changes with exposure. QHY has offered a replacement.

1 Like

I think I a might make a novel flatbox. It will be lit by a range of LEDs and rather than control their intensity through current, I’m going to pulse them, with a monostable (or more likely an Arduino).
It occurs to me that if I need long-duration flat exposures, the sensor does not care when it gets the photons, so I can give a 5 second burst of LED light and then turn off for the remainder of the flat exposure.

Chris, I bought an LED lightbox from Amazon and found a simple mod that could be done so that I can control it through an Arduino Nano fixed to the back of the panel. The Arduino is programmed to accept commands from SGPro (or NINA and others) and the PWM output is connected to a MOSFET (taking 5V from the Nano) to drive the panel. I’ve checked the uniformity of the brightness by doing flats with various panel orientations, and the variation from one side to another on my panel was less than a percent. I’ve just put together a circuit to allow me to do something similar with an EL panel, but I’ve not quite finished that yet so I don’t know if I can get a sufficient range of brightness for the various scopes I’ve got (F/2.8 to F/5.1) with Luminance through to Narrowband filters while allowing me to expose for the 3 secs that I prefer. A work in progress still, but the EL panel is very uniform brightness.

I have also used an LED lightbox and yes there’s a difficulty handling the range of exposures and brightnesses required. It’s straightforward to reduce the brightness for Lum filter with a resistor and/or opal panels but then the exposure required for Ha is quite long.
I am experimenting with adding an additional mosfet that will allow or bypass the extra resistor to the supply line depending on the brightness selected. That is if SGP calls for a brightness more than (say) 128 I would bypass the resistor boosting the brightness. An alternative is to use two mosfets connected to different resistors and different PWM lines and select the PWM by required brightness.

I might do the same - but substitute some white LEDs for deep red ones, covering Ha and S2.

I have the KL4040 camera and use Alnitak flat panels (both the FlipFlat and the larger Flat-Man L. I get banding when taking sort flat exposures which I believe is caused by the PWM technique of varying the brightness. When the exposures are short, there appears to be some kind of interaction between the electronic shutter and the timing of the PWM which is causing the banding. When I lengthen the exposures to longer than about .5 seconds, the pattern goes away. I use ND gells to reduce the brightness enough to take the longer flat exposures, especially for the UVIR R G and B filters. It’s not an issue for the NB filters.

The KL4040 has a fairly high QE at around 82%. In terms of exposure times, on my Officina Stellare RIFast f3.8 scope I saturate the UVIR filter at ~60 seconds. For the RGB filters ~90 seconds, and for narrowband filters around ~500 seconds.

Also… the Alnitak drivers allow you to control brightness between 1-255 and I have noticed that below a setting of ~10, the brightness is not consistant. I usually use an exposure time that allows the brightness setting of the flat pannel that is above 30. This allows for repeatable results.

I think that is generally the case. I have seen the same with a tungsten bulb and short exposures on a 4/3 CMOS sensor.

I use an Arduino controller and you can set the PWM frequency higher or lower. I forget which I needed to do (I think lower it), but I had to adjust that frequency to avoid banding. But it’s been a number of years since I did this and I forget how exactly.

I upped the frequency to 31k, the default of 490Hz is too low generally

I have a QHY 174 MM, the queen in banding effect camera. However, I take 5 secs exposure flats and darkflats with the peak in the histogram to be at 40-50%. Issue of banding solved when if lights are calibrarted in Pix with WBPP. I love this camera!

Yes, not all cameras have the issue. It is a bit of a lottery. :confused: