Jump to content


Photo

Read Noise and Narrowband Imaging

  • Please log in to reply
31 replies to this topic

#1 saadabbasi

saadabbasi

    Apollo

  • -----
  • Posts: 1136
  • Joined: 23 Aug 2009
  • Loc: 29N

Posted 03 April 2013 - 05:55 PM

I understand that the common advice is to expose in Narrowband long enough to overcome read noise. But how do I go about determining this? I understand it will vary from target to target but I'm hoping there's a way where I can take a set of exposures, look at the ADU values or something and figure out the best exposure time to go with.

I usually do 30 minute subs of Ha and that gets decent me results.

#2 mikeschuster

mikeschuster

    Vostok 1

  • -----
  • Posts: 150
  • Joined: 25 Aug 2011
  • Loc: SF Bay area

Posted 03 April 2013 - 06:31 PM

Saad,

Ideally you need accurate noise measuring software to determine noise on both darks and lights.

You can also use some rules of thumb. I also image narrowband with a KAF-8300. Bottom line: for narrowband, expose as long as you can because it is usually not possible to overcome read noise with any "reasonable" length exposure.

Take the binned 1x1 case: Read noise plus dark current noise for a 30 minute dark will be roughly 10 e- rms. (Fixed pattern noise and cosmic ray noise is not included in this value.)

Three times this value is 30 e- rms. If the background areas of your lights have this much noise then read noise and dark current noise has been buried enough (roughly).

30 e- rms requires a signal of 900 e- (30 squared). Assuming gain of 0.5 then you need to see 1800 ADU in the darkest areas of your subs. This ADU measurement is of lights after they have been master dark subtracted of course.

For the binning 2x2 case, read plus dark noise is maybe 15 e- rms and gain is about 1. Do the calculation and you will find that 2000 ADU's are needed.

For my usual 40 minute 3nm Ha subs when binning 2x2 with my FSQ-106 at dark sites, I get about 150 ADU not 2000. 40 minutes is way to short to be read noise limited, but I really can't go longer due to temperature dependent focus shift.

IMO expose as long as you can and use lots of subs. I usually go a minimum of 12 hours on a target.

Mike

#3 Alex McConahay

Alex McConahay

    Vanguard

  • -----
  • Posts: 2484
  • Joined: 11 Aug 2008
  • Loc: Moreno Valley, CA

Posted 03 April 2013 - 06:42 PM

Yup, just go as long as you can get out of your mount, satellite traffic, etc......thirty minutes is good if you can get it.

I usually go twenty minutes from my suburban backyard. But that is just because I don't want to go much longer and have something goofy happen to waste a subexposure of that length.

Alex


Alex

#4 vpcirc

vpcirc

    Skylab

  • ***--
  • Posts: 4101
  • Joined: 09 Dec 2009
  • Loc: Merced CA

Posted 03 April 2013 - 09:11 PM

From my understanding with good NB filters you will never reach the sky limit, so longer is always better if your equipment can support it. I recently started shooting 60 min HA frames

#5 SL63 AMG

SL63 AMG

    Viking 1

  • *****
  • Posts: 864
  • Joined: 21 Dec 2009
  • Loc: Williamson, Arizona

Posted 03 April 2013 - 10:50 PM

I recently started shooting 60 min HA frames


Seriously? I just started shooting 1 hour darks for my PL16803 and it's painful getting 8-10 per night when I am trying to reach 50!

I can't imagine trying to get a stack of 20-25 1 hour Ha frames.

I am sure you're right, it's best, but what good are all those subs if your wife discovers you dead while the camera is still collecting 656nm photons?

Posted Image

#6 vpcirc

vpcirc

    Skylab

  • ***--
  • Posts: 4101
  • Joined: 09 Dec 2009
  • Loc: Merced CA

Posted 03 April 2013 - 11:13 PM

LOL That's the beauty of remote and automation. I sleep...

#7 mikeschuster

mikeschuster

    Vostok 1

  • -----
  • Posts: 150
  • Joined: 25 Aug 2011
  • Loc: SF Bay area

Posted 04 April 2013 - 01:00 AM

20 to 30 40 min subs is normal. 128 40 min darks also. TEC is good, runs day and night once a month or so.
Mike

#8 blueman

blueman

    Photon Catcher

  • *****
  • Posts: 5487
  • Joined: 20 Jul 2007
  • Loc: California

Posted 04 April 2013 - 01:36 AM

I don't take too many darks. Normally 5 to 10 is all I need. I have looked at 5 and 50 darks combined and there is little difference. I also use the same dark master for 6 to 12 months and again I see no problems. I dither so any stray hot pixels are removed in stacking anyway.
Blueman

#9 mikeschuster

mikeschuster

    Vostok 1

  • -----
  • Posts: 150
  • Joined: 25 Aug 2011
  • Loc: SF Bay area

Posted 04 April 2013 - 12:35 PM

IMO 5 darks is not enough to get reliable sigma rejection.

10 darks reduces noise by 3x on the master, and when subtracted from lights raises noise by 5 to 10 percent. Effectively you then need 10 to 20 percent longer subs to overcome the additional noise.

Mike

#10 Alph

Alph

    Surveyor 1

  • -----
  • Posts: 1769
  • Joined: 23 Nov 2006
  • Loc: Melmac

Posted 04 April 2013 - 02:26 PM

so longer is always better if your equipment can support it. I recently started shooting 60 min HA frames


Longer is not always better. When dark current shot noise level approaches readout noise level then you have approached the maximum exposure time limit.

#11 Hilmi

Hilmi

    Soyuz

  • *****
  • Posts: 3819
  • Joined: 07 Mar 2010
  • Loc: Muscat, Sultanate of Oman

Posted 04 April 2013 - 02:51 PM

Talking of the number of darks, I saw this statistical analysis somewhere on the net, with pretty pictures for dummies like me, and it mathematically proved that there is little to be gained from more than 20 darks

#12 mikeschuster

mikeschuster

    Vostok 1

  • -----
  • Posts: 150
  • Joined: 25 Aug 2011
  • Loc: SF Bay area

Posted 04 April 2013 - 03:46 PM

With the KAF-8300, dark current is low, a binned 1x1 60 min sub at -20C has maybe 4e- rms of dark current noise, which is well less than read noise. It would take at least 4 hours for this to be as large as read noise. And even then it is not clear that this is a real limit IMO. Background sky flux is usually larger than dark current, so exposures longer than 4 hours will further bury this noise (even as it grows larger).

Mike


#13 Alph

Alph

    Surveyor 1

  • -----
  • Posts: 1769
  • Joined: 23 Nov 2006
  • Loc: Melmac

Posted 04 April 2013 - 04:13 PM

The KAF-8300 might be old but it has one of the lowest dark current of all Kodak CCDs. E.g. the KAF-3200 at -25C will run up 180e- of dark current in an hour. This results in 13e- of dark current shot noise vs. 7e- of readout noise.

#14 vpcirc

vpcirc

    Skylab

  • ***--
  • Posts: 4101
  • Joined: 09 Dec 2009
  • Loc: Merced CA

Posted 04 April 2013 - 07:24 PM

so longer is always better if your equipment can support it. I recently started shooting 60 min HA frames


Longer is not always better. When dark current shot noise level approaches readout noise level then you have approached the maximum exposure time limit.


Which will never happen with narrowband according to Don Goldman.

#15 freestar8n

freestar8n

    Skylab

  • *****
  • Posts: 4057
  • Joined: 12 Oct 2007

Posted 05 April 2013 - 02:49 AM

The time it takes for dark current noise to equal read noise is just r^2/d - where d is the dark current at the given temperature (e/s/pixel) and r is the read noise (e).

If you have high read noise then you can tolerate larger dark current because read noise will dominate - but it all depends.

I see different numbers for r and d from different sellers. SBIG says the 3200ME has read noise of 10e and dark current 0.06 e/s at -20c. That means it would take 100/0.06 = 1700s for dark current noise to equal read noise, or about 28 minutes.

QSI says its 3200ME has read noise of only 7e and dark current of 0.05 e/s at -25c. This means it would only take 16 minutes for dark current to equal read noise. Note that it may sound bad that dark current becomes a problem here in a shorter time - but it's actually *good* because it's due to read noise being smaller. It just means that for that camera, going much beyond 15 minutes has less benefit because dark noise has exceeded read noise.

I have no idea of those specs are accurate - I'm just going by the product description. But read noise and dark current vary a great deal. Sony ccd's tend to have extremely low dark current so it's less of a factor. They also have tend to have less read noise. The SX version of the 694 has 7e read noise and dark current of 0.002 e/s. For those specs it would take 7 hours for dark current noise to equal read noise.

As usual in these discussions - it all depends on the actual parameters of the camera, optics, and sky. Even with 3nm narrowband, with fast optics and a bright sky there may be little benefit in long exposures due to sky background - independent of dark current.

Frank

#16 Kevin Nelson

Kevin Nelson

    Vendor (QSImaging)

  • -----
  • Vendors
  • Posts: 23
  • Joined: 06 May 2009
  • Loc: Southern California

Posted 08 April 2013 - 08:16 PM

Frank,

Our dark current specs tend to be conservative.

Sony sensors have a reputation for having lower dark current, but what they really have is a lower population of bright pixels, which make uncalibrated images look cleaner. Read noise and mean dark current are comparable. The larger population of bright pixels in a Kodak sensor will calibrate out with standard calibration techniques.

There's some additional info in this QSI blog article that dives into parts of this topic in greater detail:

http://www.qsimaging.com/blog/?p=52

Regards,
Kevin
www.QSImaging.com

#17 freestar8n

freestar8n

    Skylab

  • *****
  • Posts: 4057
  • Joined: 12 Oct 2007

Posted 09 April 2013 - 01:50 AM

Read noise and mean dark current are comparable. The larger population of bright pixels in a Kodak sensor will calibrate out with standard calibration techniques.



I have an sxvf-h9 with a sony ccd and it has a very small dark current. The contribution to noise in a long exposure from its dark current would be negligible - especially with its relatively high read noise. A long exposure dark frame is not much different from a bias frame - just containing a number of hot pixels that can be excluded and dithered away.

If a Kodak chip does have higher dark current of around 0.02e/s vs. perhaps 0.002, then the Poisson noise from that current will not calibrate away at all and there would be reduced benefit from long exposures because the dark current Poisson noise will contribute in the same way as sky background noise - proportional to sqrt total exposure time, independent of number of exposures.

For estimating the benefit of long exposure imaging, it's important to have accurate values for read noise and dark current - independent of "system noise" that includes the pattern noise contribution - which does calibrate away (largely). My claim is that the crossover point is given simply by r^2/d - assuming the values for r and d are accurate. As I said - I was just calculating based on the stated specs.

I'm not sure what you are referring to in that blog that would suggest sony chips don't have very low dark current.

My main point in this thread is that there are many parameters that limit the benefit of long subexposure duration - and there may be little benefit in long subexposures if other noise terms dominate read noise in a subexposure. These can be high dark current noise that dominates read noise, or sky background noise in a fast system that dominates read noise.

Frank

#18 vpcirc

vpcirc

    Skylab

  • ***--
  • Posts: 4101
  • Joined: 09 Dec 2009
  • Loc: Merced CA

Posted 09 April 2013 - 05:33 AM

Now Frank you're going to disagree with the guy who makes the cameras for a living, does all the testing and owns the most extensive library of bias and dark frame chips examples that I've ever heard of?

#19 Kevin Nelson

Kevin Nelson

    Vendor (QSImaging)

  • -----
  • Vendors
  • Posts: 23
  • Joined: 06 May 2009
  • Loc: Southern California

Posted 09 April 2013 - 09:38 AM

Frank,

I'm not disagreeing with you. Sony sensors do have very low dark current, but so do Kodak sensors. The main difference between a dark frame from a Sony sensor is that Sony dark has a lower population of bright pixels so uncalibrated images look cleaner. Those bright pixels are removed by standard dark subtraction.

Take a look at the table in the blog article labeled, "KAF-8300 Dark Current vs Temperature," about half way down the page:

http://www.qsimaging.com/blog/?p=52

Dark current in the KAF-8300 on a 5-minute image at -20C is 1.72e-, which works out to 0.0057e-/sec on this specific sensor. Noise in that (unwanted) signal is SQRT(1.72) = 1.31e-. Read noise from the sensor still dominates the noise contribution from the thermal current.

As you noted there are other factors which may limit the benefits of very long exposure imaging.

Kevin
www.QSImaging.com

#20 freestar8n

freestar8n

    Skylab

  • *****
  • Posts: 4057
  • Joined: 12 Oct 2007

Posted 09 April 2013 - 12:03 PM

Hi Kevin-

Yes, I didn't think we disagreed on the principles, but I thought you were saying sony chips had higher dark current than commonly thought. Instead I think you are saying Kodak have much *lower* than commonly thought. I just said I was going by the specs - and for the QSI I pointed out that its read noise was relatively low - which was a good thing - but it meant that it would benefit less from longer exposures than one with higher read noise - particularly since the variance goes as r^2.

With regard to the QSI values for the 8300 dark current - in one of your ads it says it is 0.05 e/s at -20c. If instead it is more like 0.005 e/s at -20c, then that is quite a difference - and it does stand out from typical values for the 8300, which I think are around 0.02 e/s at -20C. Meanwhile, another prominent vendor that sells both kodak and sony chips has a 694 model with a stated dark current of 0.002 e/s at -10C. So the numbers are all over the place.

With regard to this thread and the benefit of long exposures for "narrowband" - my main point is that - after calibration and removal of pattern noise and hot pixels - the relevant trade off exposure time is given by:

r^2 = t(d + ASW/F^2)

Here r is read noise, t is subexposure time, d is dark current, S sky background flux, W is the narrowband width in nm, F is the f/number, and A is a constant that includes the pixel area. The equation indicates the exposure time when the read noise variance equals the combined dark current and sky noise variance.

Each of the terms will set the time at which read noise equals the other noise terms - and longer sub exposures won't help much. Some terms vary tremendously, like sky flux, and terms that are squared, such as read noise and f-number, also play a strong role. F number may range from 2 to 15, so F^2 ranges from 4 to 225. A slow system in a dark sky with a narrow filter will have a completely different time to equal variance than a fast one in a bright sky with a wide filter.

This is all to say that there are many factors that determine whether long exposures have a significant benefit - or not.

I don't have a QSI camera, but I mainly give them credit for incorporating OAG, which helped increase its popularity and acceptance in the last few years.

Frank

#21 Madratter

Madratter

    Fly Me to the Moon

  • *****
  • Posts: 7150
  • Joined: 14 Jan 2013

Posted 09 April 2013 - 01:18 PM

Kevin. Thanks for that article. I'm trying to wrap my head around what it means. I have a 8300 based camera. I think what it means is that obsessing over how cold the camera gets is probably wasted effort in real world terms as long as the temperature regulation is good, exposures aren't extremely long (certainly more than 30 minutes), and I am running below zero (which would be all the time for me). Plus I need to be sure I use darks and flats of the same temperature and not scale the darks.

#22 Kevin Nelson

Kevin Nelson

    Vendor (QSImaging)

  • -----
  • Vendors
  • Posts: 23
  • Joined: 06 May 2009
  • Loc: Southern California

Posted 09 April 2013 - 03:09 PM

Frank,

Each sensor will have the dark current that it's going to have. Short of making a mistake in how the sensor is managed by the camera electronics, there's really nothing a camera manufacturer can do to affect dark current one way or the other.

I suspect the problem with widely varying dark current specs is because the sensor manufacturers don't spec dark current at temperatures anywhere near where cooled CCD camera users run their sensors.

For instance Truesense (formerly Kodak) specs dark current on the KAF-8300 at <200e-/sec at 60C with a dark current doubling temp of 5.8C. Calculating down from there to the specified dark current at -20C yields a number of about 0.02e-/sec.

So, why the big difference between the specified value and the measured dark current? Kodak's spec is conservative to begin with, and so are the specs from most camera manufacturers. The most conservative approach is to match what Kodak specs the sensor to. Trying to gauge quality by comparing the specs of cameras from different manufacturers using the same sensor likely won't yield any useful data. The biggest differences in areas like dark current are likely to be the result of one manufacturer being more conservative than another.

Kevin
www.QSImaging.com

#23 Kevin Nelson

Kevin Nelson

    Vendor (QSImaging)

  • -----
  • Vendors
  • Posts: 23
  • Joined: 06 May 2009
  • Loc: Southern California

Posted 09 April 2013 - 03:47 PM

The best thing to do is set up a test and measure the results with your own camera. It's not hard to generate all the data required to create the graphs shown in the blog article for your own camera.

Let your camera stabilize at your desired set point, perhaps -20C. Take 10 bias frames and 5 each darks at 5 and 10 minutes (and 20 minutes if you plan to shoot that long). After you collect the data, increase the temp to -10C and then 0C. Repeat the acquisition of bias and dark frames for each temperature.

Combine the images in each dataset. Subtract the mean bias value from the mean value of each matching set of darks. Drop the numbers into a table and plot the data on a line chart.

To calculate noise, you'll need to know the gain of your camera. That value may be stored in the "EGAIN" tag in the FITS header. Multiply the ADU counts by the gain to convert to electrons.

But ultimately I believe you're correct. Once you get to -10 or -15C with a Kodak sensor (or most front-illuminated CCDs), other factors besides thermal current and its associated noise dominate the total noise in the image for most amateur imagers.

Regards,
Kevin

#24 wolfman_4_ever

wolfman_4_ever

    Apollo

  • -----
  • Posts: 1245
  • Joined: 15 Jul 2011
  • Loc: El Segundo, Ca, So. Cal

Posted 09 April 2013 - 05:51 PM

Now Frank you're going to disagree with the guy who makes the cameras for a living, does all the testing and owns the most extensive library of bias and dark frame chips examples that I've ever heard of?


:foreheadslap: isn't that how progress is made? The world was known flat until someone disagreed.. String theory would be another example.. heck any scientific theory is torn apart..

Just because one side says global warming is a hoax doesn't mean it's not true.. religion anyone?

Why do you always name drop anyways? As if said theory was fact forever..

#25 Kevin Nelson

Kevin Nelson

    Vendor (QSImaging)

  • -----
  • Vendors
  • Posts: 23
  • Joined: 06 May 2009
  • Loc: Southern California

Posted 09 April 2013 - 06:19 PM

Questions are good. No way should you give me or anyone slack. I've learned a few things about CCDs over the last several years, which I'm happy to share, but that doesn't make me the definitive source for CCD knowledge by any means.

As I always recommend, setup a test and gather results with your own camera. You'll learn more about your camera and CCDs in general, and likely have some fun along the way.

Cheers,
Kevin






Cloudy Nights LLC
Cloudy Nights Sponsor: Astronomics