Flat Fields- What signal do you use?

[narrowbandpaul]

Hi All,

I'm curious as to how bright your flats are and how you came to this decision. I see lots of different numbers being quoted and I'm interested as to how you came across these.

The quality of a flat field can be expressed as Quality=Signal of single flat x Number of Flats. This quality term actually makes an appearance in the mathematics of flat fielding

By using a low value for signal you only end up hurting yourself. You need a high quality flat to minimise the injection of random noise. Using 1/3rd of saturation and 10 flats just won't cut it really. You do want to keep your flats in the linear portion of the CCD but even with antiblooming CCDs are still linear over most of the dynamic range.

So please post how you take your flats, the signal level and number of images. Perhaps we can all reach an informed consensus and cut through some of the drivel and misinformation you can find on the internet

Cheers

Paul

[carastro]

I use an EL panel made into a lightbox.  I normally aim for 22,000 ADU (approx) and normally take around 15 flats.

I used to take 30,000 ADU but find 22,000 works better.

Carole

[Stephen]

Gerd Neumann flat panel, 21K ADU, 20

That was the config with my previous camera (Atik 460ex) ... nothing will change with my new camera bar the signal level which will be determined by experimentation

[narrowbandpaul]

Thanks Carole, Stephen.

How do you determine what works best and what doesn't. How do you quantify it?

Paul

[carastro]

In my case whether it does the job well, i.e. removing dust and vignetting.

Carole

[Stephen]

Thanks Carole, Stephen.

How do you determine what works best and what doesn't. How do you quantify it?

Paul

In my case whether it does the job well, i.e. removing dust and vignetting.

Carole

What Carole said. I spent a night taking varying ADU flats and applying them to an image with a standard stretch. Under the target value and above the target value I noticed strange artifacts - at the value the result was nice and even across the FOV

The number of flats (20) is my opinion only (from reading lots of threads like this) of the optimal number before you start to hit the law of diminishing returns. This may be garbage, but it works for me and has done so across a few cameras

[Stephen]

I also just found this post from a prior thread which sort of concurs with my visual estimation

http://stargazerslounge.com/topic/159317-atik-460ex-what-adu-level-for-flats/#entry1609800

[ollypenrice]

With refractors I found anything between a third and two thirds seemed to work fine. The fun started when we tried to make flats for Yves' 14 inch. We could never get them to work. They over corrected, producing inverse vignetting and inverse bunnies. In the end I brought the count down to 13000 and these seem to work much better.

As for why, alas I have not the faintest idea.

Olly

[davew]

I had to look at my flats to double check but I did 34700 x 20+ last time out with a KAF 8300 chip.

The rational for this was a paper written by Richard Crisp on noise, ADUs and number of flats. I believe he has adequately demonstrated that Kodak's chips require up to around 80% full well so long as it doesn't saturate.. Kevin Nelson of QSI seems to agree and has published the following list for their chips - http://store.qsimaging.com/kb_results.asp?ID=28  Of note is that he errs towards 70% with ABG and 80% with NABG. I can't now find the link where he explained this. Also note that full well in this case means the chip full well divided by gain.

( I believe ) you can take Flats at a lower level but you'd need many more to reduce the noise levels. I wimped out of taking higher flats and just took more ! I really should up the target ADU and probably do more anyway.

Dave.

I imagine many people could be swayed by QSI   Shame they haven't listed their Sony chips yet.

[RobH]

I use an ADU of about 20,000 and use 25 minimum, and more if I can. I use sky flats at dusk or dawn but with a diffuser screen in front of the scope to avoid catching any stars.

I calibrate the flats using a bias master comprised of 100 subs.

The ADU value seems to be the best for getting a well corrected light with the 460ex, but with the H18 I used about 13000 ADU . More than this caused over correction for some reason. Like Olly, I have no idea why.

The number of flats is based on getting a decent amount to get a good S/N ratio before losing the light to the point that I'd have to shoot flat darks. I stop when the exposures get to 18 seconds.

I don't use an EL panel as they're rather expensive in A2 size, I find that sky flats seem to work better with my RC, and the sky is a very good price....free :-)

[ollypenrice]

I use an ADU of about 20,000 and use 25 minimum, and more if I can. I use sky flats at dusk or dawn but with a diffuser screen in front of the scope to avoid catching any stars.

I calibrate the flats using a bias master comprised of 100 subs.

The ADU value seems to be the best for getting a well corrected light with the 460ex, but with the H18 I used about 13000 ADU . More than this caused over correction for some reason. Like Olly, I have no idea why.

The number of flats is based on getting a decent amount to get a good S/N ratio before losing the light to the point that I'd have to shoot flat darks. I stop when the exposures get to 18 seconds.

I don't use an EL panel as they're rather expensive in A2 size, I find that sky flats seem to work better with my RC, and the sky is a very good price....free :-)

Rob, the common factor here is the SX camera, with regard to needing low ADU. Yve's 14 inch has an SX H36 in the back. You were using an H18. Could the camera be the defining factor? I'd assumed it was the telescope. But suppose there were some non-linearity? The H36, like a friend's H18, won't work in Bin 2. SX say to adjust the gain via a potentiometer but we tried that without success. I'm not competent to do anything more than pose the question.

Olly

[russellhq]

I'm thinking out loud here. If you use a flat with a high ADU and divide a light with a small ADU then could you end up with a very small ADU and run into quantization problems?

[davew]

I'm thinking out loud here. If you use a flat with a high ADU and divide a light with a small ADU then could you end up with a very small ADU and run into quantization problems?

I didn't think that's how flats worked. I thought each pixel in the flat was divided by an average value to make each one appear the same value and that ratio was either multiplied or divided into each light frame ( can't remember which now ! ). So if a light frame had less light on a few pixels ( Bunnies etc ) the flat would normalise the light.

Well, I know what I mean

Dave.

[RikM]

I use 23kADU and take 31 for each filter using a cold-cathode lightbox that used to be an MD's x-ray film viewer. This is with an SXV-H9. I haven't really messed about to see what happens at different values. I read on here that 30-50% is a good range and take plenty of them to keep the noise down. I tried it, vignetting and dust bunnies went away. I didn't get any funny colour gradients or anything, so I stuck with it.

[narrowbandpaul]

Hi all. Thanks for the responses,

Dave is right with regards to how flats work. Each pixel in the flat is divided by the average. This creates a correction factor that then normalises the response of all pixels in the raw image. A high signal flat doesn't lead to large numbers for the correction factor, it leads to a high accuracy correction factor.

As I said before the quality of a flat is signal per flat x no of flats.

The issue of low signals used in the SX cameras is interesting. There could be a link there, although correlation doesn't always imply causation.

I think it's possible for camera manufacturers to influence the linearity. Only thing is I'm not entirely sure how. Perhaps the gain changes with amount of signal. That effect does happen with CMOS sensors so it's possible. Thinking about I suspect that the gain is changing with signal.

I don't know why the higher signal flats over correct things. The theory says nothing about that. If the gain is changing then using lower signals still might improve things. Just use lots of flats though!

Good find on the QSI site, I would agree with Kevin. I normally aim for 40000DN, which is about 2/3rds.

Cheers

Paul

[martin_h]

I must admit that use flats of around 22K ADU for no other reason than having read it somewhere!.After reading this thread I hot footed it out to the obs and shot off 40 at around 48K ADU, and to be honest, visualy I can see no difference in the final stack of images.

[opticalpath]

I use a GN flat panel and aim for 30000 ADU with my QSI583; that's about 60% of full well capacity.  Have occasionally gone lower, and somewhat higher, but didn't notice any difference.

 

I assume we are talking about frame-average ADU value, when setting a target?  I aim for a whole frame average of 30000, but keep a watch on maximum pixel values too so they don't go above 35000 or so.  Interestingly, I noticed that the gap between average and max. values is quite a bit wider for some filters compared to others.

 

Adrian

[russellhq]

I must admit that use flats of around 22K ADU for no other reason than having read it somewhere!.After reading this thread I hot footed it out to the obs and shot off 40 at around 48K ADU, and to be honest, visualy I can see no difference in the final stack of images.

Could it be you reached your maximum SNR with the 22K flats, that there was no benefit of going to 48K?