Measuring CCD Linearity

mjc wrote: As you have S.B. Howell can I ask if you feel that you got value out of it?

Its a good book if you have no idea, or very little idea, of how a CCD works. I've found it very useful as a background read. I paid just over 25 sterling for it (incl P&P) so I think it was worth it.

I think the Janesick is, as you say, a much more technical book - definately not for the beginner. Its certainly recognised as a key engineering/reference book for those who know what they're talking about...

A gain of 0.2 -e - wow - that's tiny. And 5 e- per ADU... What's the full-well? Surely it can't be 13k?

I'm not sure if the maths are correct, but if my readout noise is 15 e- and the full-well is ~150k, then read noise is 0.01% at full well. That doesn't seem too bad. At the lower end of the linearity range (about 15K e-) that still only amounts to 0.1%.

A journey of discovery - more like a nightmare journey for me

Thanks for the comment re S.B Howell. I'll think about that - but I am tempted to go for the Janesick book. I do expect it to be a challenge.

Others have measured the gain of the Atik 16 IC to be more like 0.25 e - that's why I'm thinking that I may not have measured it right.

Full well - in my estimation is probably around 11,000 e but if g is really 0.25 then its more like 13,000 e.

[Edit: also full-well is generally correlated with pixel size - my pixels are 7.4 microns square - yours are probably larger than that.]

I could be doing this all wrong so I'm going to revisit everything tomorrow and also do Janesicks method to determine g.
The method I used is a quick method.

Basically get a flat and g = the mean ADU's of a region of interest divided by the square of the std dev of that region. If signal is high enough then this is supposed to work. I didn't bias correct first and I suspect that I should have and this could throw my measurement - The HAIP isn't particularly explicit here (or I've missed it).

That paper I posted reference to - despite being only a five pager it still requires the brain to be in gear. I think its relevant to your two amp implementation.

From what I read thus far is that you effectively use one amp on low gain so that this amp is optimised for low signal pixels and the other amp has higher gain which is optimised for pixels requiring a large ADU range.

But I haven't completed reading it yet.

Mark C.

You're welcome to borrow the Howell book any time - no need to buy it.

The KAF1101E has 24um pixels, so yes, much larger full-well of course.

On the amp thing, you choose one or the other at time of purchase, as far as I can tell. There is one manufacterer that offers the choice. I wasn't offered the choice, though if I was, I wouldn't have had a clue what they were talking about. One choice gives you full-well of ~600k, the other (mine, I think) gives ~200K.

It is possible that someone *could* mount the CCD on a board such that you could choose prior to imaging. Possibly...

For the ROI, you could take a long dark - say 10 mins (preferably as long as you can - I did 4 hour darks) and just make sure your ROI doesn't include obviously hot pixels.
If you can overscan the CCD (I doubt it) you can use the overscan region of the flat as the bias correction. Without an overscan region, I'm not sure what the standard wisdom is - presumably bias subtract. But taking a bias is easy, right?

As far as I know, the method I followed for the PTC and DTC is Janesick's (or possibly with minor mods). You can actually get the gain etc etc just from darks. Less variables when taking the images (non-uniform illumiantion etc) but they take much longer to take (though you can take them at room temp).

Whatever you do, its well worth doing it. I've learned a lot from this exercise and know a bot more about my CCD. I'm half-way through another run of data which I'm hoping is cleaner than the last lot. I'll finish it tomorrow and let you know.

Cheers and good luck!
Dave

Dave

I just ordered the Janesick book. Slightly more than €100. I had a look at the preface where the contents are described and feel that it's going to be my cup of tea. I appreciate your loan offer. Maybe we should consider a mutual loan a bit later (I'm sure there will be some stuff in the Janesick book that you'd appreciate). I'd like to have S.B Howell's work on my book shelf but the Janesick book appears to address detail that can be difficult to find unless one was initially half-way savey. At the moment I'm reluctant to buy both. I suspect that I'll get about ten percent genuinely understandable and applicable knowledge from it - about 30 percent historical and educational about matters that I won't have the ability to control and about 60 percent semi-conductor physics that at best will teach me a few new buzz words, phrases, and acronyms.

The preface does lead me to believe that a serious amateur (an aspirational statement on my part) is not excluded from the audience.

It covers PTCs and some concepts that I've encountered but have little knowledge off. It appears to repeat the Nyquist critical sampling story - which I remain suspicious off - despite overwhelming acceptance by people more educated and experienced than I (hey, I'm a rebel at heart).

I revisited my data and had a great time - its doing stuff like CCD characterization that educates me to understand that I don't really have a grasp of this stuff. I'm walking in the water at the edge of the beach - but I am not swimming or surfing.

I tried doing a PTC and failed miserably.

My initial calculation of gain - using a simple method (but mathematically valid - and I don't yet know why it was so way off - maybe small full-well capacity - really don't know). Would like to know why though.

Janesick's method - taking two light frames and two bias frames and

(F1 + F2) - (B1 - B2)
g =

---

sigma^2(F1-F2) - sigma^2(B1-B2)

Where F1 + F2 is the sum of two flat-field frames of same exposure times
F1 - F2 is the difference between the same two flat frames

ditto with bias frames B1 and B2.

Gives (for me) g = 0.29 e per ADU (my previous estimate was 0.2 which I believe was way off and others have got about 0.25 (so I could still be off - but I did my test twice) - I don't know if individual CCDs can vary by that much but 0.29 is what I currently measure).

Readout noise

Take two bias frames - subtract one from the other so that

sigma(B1 - B2)
RN = g

---

sqrt(2)

I got 4.6 e RMS compared with 7 in spec. Suspiciously good.

Dark Current
I took a 1hr dark frame. Dark current is very good with the Atik16 IC and I really needed a long exposure.

D - B
D(electrons) = g

---

t(secs)

Where D is the mean pixel value (ADUs) in the dark frame and B is the mean pixel value in the bias frame, and t is the exposure time in seconds.

I got 0.0014 e / pixel / sec. (if I'm doing this right - that's very good).

Linearity.
I plotted this several times and in several ways. When I had an estimate of where linearity broke off I did a higher resolution exposure series about that rough estimate. There's a subjective decision here but I feel comfortable that full-well capacity (break with linearity - not the maximum value a pixel can hold) is no less than 54,500 ADUs. With a gain of 0.29 e / px that equates to a FWC of 16,000 e (to 2 significant digits). That's small - its a budget device and pixel size is small at 7.4 microns square (comparable to a webcam).

Dynamic Range
Just because I stumbled on the formula - I did this.

DR(dB) = 20 log(FWC / RN) - FWC and RN in electrons and log - being base ten (the stuff we learnt at school)

DR = 82dB (and that means diddly squat to me).

There's other stuff that can be measured but I'm not comfortable enough with what I have measured to try and go into further detail at present.

Mark C.

After initial posting:
In all equations you have to imagine white space on the stuff above and below the divide bar. It looks ugly as presented but isn't that horrible if white space is correctly envisioned. What happened to preview button?

Chaps, I certainly need to read this thread fully when I have an hour and a pot of coffee. In the meantime I came across this cript on the Maxim DL site which might be useful.

www.cyanogen.com/down/runSequence.zip
from the maxim dl site;
"runSequence.vbs is a script by Hilary Jones that can be used to evaluate the linearity of your CCD. The script takes a sequence of pictures using different exposure times. It measures each exposure and saves the average ADU level in a file. The pictures themselves are (normally) not saved. Each exposure is bracketed with reference exposures that allow you to compensate for drift in your light source's intensity. The data in this file is formatted in a way that lets you evaluate your CCD using an Excel spreadsheet. The spreadsheet is not included in this distribution; but the author will be glad to help you set one up. You must edit the script to specify which exposure times are to be used, what filter to use, where to save the data file, etc."

Dave - that's a neat script for figuring out your linearity.

However:
1. It assumes you have MaximDL...
2. It assumes you have a sense of what variables to change and what to change them to...

To be honest, the process/method I used was probably a lot quicker, software independent, verified and possibly more "analytical". But the biggest plus was being able to get a bunch more parameters from the analysis than just linearity.

I had hoped to write up a very simple/short process of what I did and how, but work commitments etc are keeping from doing it (and a load of other stuff too). If you, or anyone else, needs any help etc, PM me and we'll arrange a call.

The process is actually simple, the preparation can be quite time consuming and tricky alright. I'd recommend anyone who is big into imaging, especially CCD-based, should characterise their imager to get the basic performance data. This will then show you what you can/can;t do with your imager and give you better results.



Mark - I'd love to take a look at the Janesick book to see if its worth me getting a copy. Let me know (in a few months time) when you've assimilated it! It looks like it would certainly help me understand the maths behind what I already did...

I've already forgotten what the Dynamic Range (in dB) means - Howell covers it. The fact I've forgotten means its probably a "nice-to-know" number but not terribly critical.

You've some interesting results there. How did you take your flats? And is the CCD cooled? (I suspect not - but I'd imagine as long as temperature conditions vary by only a small amount over the imaging time that the data is still good).

Also, do you know if you can overscan the CCD? If so, that will eliminate the need for bias frames - I think...

The formulas are difficult to read but following your postcript, they can be deciphered. No idea where the preview button went (presumably it is keeping the email notification facility company )

Thanks for posting your data though. Its rasied a few more qquestions for me that I shoudl go back and revisit.