New 383l+ readout noise

[rfdesigner]

I've been testing my new and shiny 383l+ which in most respects seems to be a good piece of kit.

However I'm getting rather high readout noise numbers.  I've been using the method of taking multiple lights/bias with the camera without a lens/scope in dim light to establish the gain (I'm getting just over 0.48) and then using two bias frames, taking the difference to remove artifacts and then running the maths and getting a figure of ~12e.  which is noticably more than the typical 7e quoted for the camera.

I've tried running the test many many times, working the maths differently, using a proper bench top power supply, making bias and flat frames from multiple subs and redoing the maths, just selecting clean sections of raw light and bias frames etc,  but every time I get around 12e.

I've run the camera at -20C and let it settle.  I've also readout and thrown away the first image in each sequence to remove artifacts.

I've thrown two lights and two bias frames up on dropbox if someone would care to take a look.

http://dl.dropbox.com/u/44641599/383bias_13v8_54.fit
http://dl.dropbox.com/u/44641599/383bias_13v8_55.fit
http://dl.dropbox.com/u/44641599/383flat_001_014.fit
http://dl.dropbox.com/u/44641599/383flat_001_015.fit

I make the raw noise ~24 counts for bias and ~240 for the light with a light level of around 27000 counts.  This gives me the gain of .48 and noise of 11.5e.

is this to be expected?

[Konihlav]

I have sent you a PM.

Atik 383L+ from my measurement (couple of samples) all ended with RN of about 10.30 to 10.60e- more or less. I have seen only few KAF-8300 cameras having noise really under 8e- (measured 4 different cameras of the same manufacturer).

I have also brief-tested new 428EX and one CCD had noise over 6e- (like 6.25e-) and the other one around 5e- as in specs (will re-test some day when I get the time to make it 100% precise).

that's all folks. I advice everyone with a new CCD to check bias frames :-)

[rfdesigner]

Thanks for the reply..   was it your thread on cloudy nights?...

I am beginning to think that maybe the KAF8300 sensors aren't performing as advertised, but does mean that spec sheets should be rewritten.

I'm wondering if the new Kodak sensor business owners have cut a corner or two, or given preferential treament to one customer, leaving the rest without the best performing units.

[Konihlav]

yes its mine thread.

My speculation based on what I have heard here or there is that I would suspect Kodak to provide worse chips to Europe and better ones to the USA. That's my personal point of view that I don't want to explain more :-)

Anyway there is some statistical difference in chips like say more or less by 0.5e- . The rest of the difference between say Apogee and say FLI (could be Atik too or whatever) is manufacturer's electronics.

My former research had two goals - to check if what they tell us is really what it is, to compare the same cameras (same chip) from more manufacturers against with a reasonably "smart" approach (consistent) in order to identify who is excelent and who only good and to put another pressure on QC of CCD manufacturers from users perspective. We want the best

[MicroAstro]

I've tested mine as well by 4 different methods and I get a range of values from 8.6 to 10.4e-. So a lot depends upon your method of calculation. Haven't managed to get the reported 7e- however.
...Keith

[Kaidan]

I found close to 8.5-9.5e- for my 383L+ as well.

The specifications are wrong : there's blooming in bin2 (the KAF should be ABG) and the readout noise is not as good as it should be, i'm a bit disappointed really.

[rfdesigner]

I've tested mine as well by 4 different methods and I get a range of values from 8.6 to 10.4e-. So a lot depends upon your method of calculation. Haven't managed to get the reported 7e- however.
...Keith

I'd be interested to know what the four methods you used were, I've only got two in my toolbox.

Derek

[rfdesigner]

Additional

after further testing I can confirm a feature that Christian Buil discovered in the QSI583 camera.  Slight non-linearity.  The gain from a typical mid point is not representative of gain at the bias level...  Re-measuing gain with low light levels I'm now measuring 10.5e noise..  it's not 7e, but then neither is it 12e.

Derek

[MicroAstro]

1. Just plotting variance against ADU, which is the simplest and seems to be the one used by Atik.
2. Craig Stark's method (http://www.cloudynights.com/item.php?item_id=2001) estimated 2 ways: average of 2-point calculations, and by taking a series of flats at different intensities (I also took flats with a red filter because I read that shorter wavelengths pick up pixel variation in addition to the RN). 
3. Apogee CCD University method (http://www.ccd.com/ccd105.html)
4. Photometrics method (http://www.photometrics.com/resources/learningzone/)

Problem is, there is no standard way of measuring RN so it's up to the manufacturer to chose which one to report. The method Atik used seems to be the simplest and gives the lowest estimate. Otherwise, a range of 8.6-10.4 can depend totally upon how you estimated the RN and how saturated you allow your flats to be (generally, lower exposure levels of your flats seem to give lower RN because at higher saturation levels you run into non-linearity).

I got quite interested in this for a while for the same reason others are reporting -- I wanted to confirm my camera had the 7e- reported by Atik. In the end, although I couldn't replicate the exceptional 7e- reported by Atik, it wasn't that far outside of my range (I actually found looking more closely at my spreadsheet that I had a range of 7.9-10.6). So is Atik's number wrong or deceptive? It is one legitimate way to estimate RN, but what's the true RN -- I gave up at that point? That is the value of Konihlav's standardized method that he ran on Cloudy Nights requesting frames from different cameras, it's the same for all cameras and so is actually comparable from manufacturer to manufacturer. However, even then, other things can affect RN such as voltage of your power source -- I find with my 383L that it has a different measured RN depending upon whether I'm running it on a 12v or a 13.8v power source, I think because higher voltage increases the gain slightly.

Another thing I will say in the 383L's favour, the noise is very clean. I also did FFT analysis as per Craig's article and the read noise frame FFT and histogram were near-perfect, meaning that it is very predictable and correctable by calibration frames.

I ended up deciding that finding the "true" RN was futile, and getting worked up about a 10 vs 7e- RN wasn't worth the concern given that there is quite a range just due to the method of estimation. And the textbook FFT of the read frame made me decide that it probably doesn't matter anyway and I'd be better off spending my time collecting photons and processing my images than sitting in front of a spreadsheet.
...Keith

I've tested mine as well by 4 different methods and I get a range of values from 8.6 to 10.4e-. So a lot depends upon your method of calculation. Haven't managed to get the reported 7e- however.
...Keith

I'd be interested to know what the four methods you used were, I've only got two in my toolbox.

Derek

[Konihlav]

Keith: absolutely right!

my research showed who is good manufacturer and who is bad one. Good thing is that vast majority is good. Sometimes I discover a "faulty" CCD with weird noise or patterns that can have roots in e.g. power voltage or wi-fi or whatever...

regarding KAF-8300 there are cameras with 7.7e- RN (I have measured four samples of MII with all of them having 7.5e- to 8.0 e-), some around 9e- (Apogee), lot of cameras having around 10e- (FLI, Atik), and some worse (SX, QHY) with 12e- and more (random ).

So if you have 8e- or 10e- that in matters only slightly and of course only in narrow band imaging. In LRGB it's all the same.

So the best you can do is to rather collect some photons than to dig into playing with statistics... but, if you have new CCD I recommend to take some sample biases just to inspect the camera behavior when you have it new (check that everything is OK).

The real use of low RN cameras (3.7e- is very much different from 9e-) is for lucky imaging and for narrow band imaging where with low RN cameras you can get pretty clean images!

[rfdesigner]

Thanks Kieth.

It's easy to get hung up on the numbers, when I first measured 12e I thought.. hang on!, at 10.5e I care somewhat less, and it looks like the dark current is very low on my unit, about half that advertised.  So you win some you lose some.  I'll have one last go plugging the numbers through the other methods and I need to pack up my data for Konihlav to have a look at.

It's interesting that MII seem to be lowest for noise, because they claim to hand pick their chips.. which does suggest the reason for the variation in performance is the KAF chips, and not the cameras.  It's nice to see Atik matching the performance of cameras from FLI and beating SX and QHY, especially when it comes in such a handsome package.