Jump to content

  •  

CNers have asked about a donation box for Cloudy Nights over the years, so here you go. Donation is not required by any means, so please enjoy your stay.

Photo

Canon 7D MkII Dark Current Test

  • Please log in to reply
19 replies to this topic

#1 sharkmelley

sharkmelley

    Mercury-Atlas

  • *****
  • topic starter
  • Posts: 2510
  • Joined: 19 Feb 2013

Posted 17 September 2017 - 06:45 AM

I had some friends over last night for a night of astronomy.  One of them brought a Canon 7D MkII.  It has been written elsewhere that this camera has a very low dark current compared with other Canon cameras so I was eager to put this to the test.

 

I used my standard test methodology.  The camera was left at room temperature for a few hours and then a series of 5 minute dark  exposures was taken in a darkened room with the lens cap on and the rear LCD switched off.  The ambient temperature of the room was 19C.  Bias frames and flat frames were then taken. 

 

I put the results into my PixInsight DSLR Sensor Parameters script:

 

=============================
DSLR Sensor Parameters v0.0.7
=============================
Camera: 'Canon EOS 7D Mark II'
ISO speed: 1600
Image Size: 5496 x 3670
Bias frames exposure time: 1/8197 sec
Flat frames exposure time: 1/1328 sec
Dark frames exposure time: 294.6 sec
Channel order: []
Gain per channel(e/ADU): [ 0.178, 0.178, 0.18, 0.177 ]
ISO for unit gain: [ 284, 284, 288, 284 ]
Read Noise(e): [ 2.53, 2.5, 2.57, 2.53 ]
Dark Current (e/pixel/sec) [ 0.0847, 0.0803, 0.0873, 0.077 ]
Dark Current (e/pixel/sec) [ 0.1181, 0.1109, 0.1203, 0.107 ]
Dark Current (e/pixel/sec) [ 0.1478, 0.1339, 0.1509, 0.1344 ]
Dark Current (e/pixel/sec) [ 0.1754, 0.1594, 0.1794, 0.1594 ]
Dark Current (e/pixel/sec) [ 0.1995, 0.18, 0.2045, 0.1816 ]
Dark Current (e/pixel/sec) [ 0.2236, 0.2008, 0.2264, 0.2014 ]
Dark Current (e/pixel/sec) [ 0.2454, 0.2214, 0.2503, 0.2227 ]
Dark Current (e/pixel/sec) [ 0.2637, 0.2396, 0.271, 0.2407 ]
Dark Current (e/pixel/sec) [ 0.287, 0.2852, 0.2886, 0.2575 ]

 

Noise estimates for 295sec exposure
Read Noise(e): [ 2.53, 2.5, 2.57, 2.53 ]
Thermal Noise(e): [ 9.2, 9.17, 9.22, 8.71 ]

 

Mean values in ADU
Bias Frame 1: [2048.2, 2047.7, 2048.4, 2047.8 ]
Bias Frame 2: [2048.3, 2047.8, 2048.4, 2047.8 ]
Flat Frame 1: [4346, 3640.5, 3080.1, 4350.9 ]
Flat Frame 2: [4376.3, 3661.3, 3093.4, 4380.8 ]
Dark Frame 1: [ 2052.2, 2052.1, 2051.6, 2052.1 ]
Dark Frame 10: [ 2090.7, 2092.6, 2088.1, 2092 ]

 

DSLR Sensor Parameters v0.0.7 Completed.

 

Here are the results plotted on the same graph as other Canons I have tested:

 

Canon7D2_DarkCurrent.jpg

 

During the course of the experiment the temperature recorded in the EXIF header climbed from 19C (the same as the room temperature) to 34C.

 

Although there wasn't time to perform a full 2 hour test, my conclusion is that under typical deep sky imaging conditions the Canon 7D MkII appears to have a similar dark current profile to other entry level Canon cameras.

 

Mark

 

 


  • whwang and RareBird like this

#2 johnpane

johnpane

    Vostok 1

  • *****
  • Posts: 111
  • Joined: 13 Dec 2014
  • Loc: Wexford, PA, USA

Posted 17 September 2017 - 07:03 AM

Thanks for this information. The result is interesting in comparison to what Roger N. Clark posted in 2014. Roger reports a single measure at 10C and how it changes in relation to temperature. Your first measurement is higher than Roger's formula predicts unless sensor temperature was already elevated above room temperature. But Roger projects a much steeper increase with temperature.

 

Would it be useful to plot all of those results with (sensor) temperature on the x-axis instead of just the index of the measurements?

 

John


Edited by johnpane, 17 September 2017 - 07:06 AM.


#3 sharkmelley

sharkmelley

    Mercury-Atlas

  • *****
  • topic starter
  • Posts: 2510
  • Joined: 19 Feb 2013

Posted 17 September 2017 - 10:01 AM

Thanks for this information. The result is interesting in comparison to what Roger N. Clark posted in 2014. Roger reports a single measure at 10C and how it changes in relation to temperature. Your first measurement is higher than Roger's formula predicts unless sensor temperature was already elevated above room temperature. But Roger projects a much steeper increase with temperature.

 

Would it be useful to plot all of those results with (sensor) temperature on the x-axis instead of just the index of the measurements?

 

John

If you look at graph 3 on Roger's page, you'll see a graph of a whole set of values, not just a single point.

 

My set of values is here:

 

Temp1 Temp2  DarkCurrent

20C   25C      0.0823
25C   27C      0.1141
27C   29C      0.1418
29C   31C      0.1684
31C   32C      0.1914
32C   32C      0.2131
32C   33C      0.2350
33C   34C      0.2538

34C   34C      0.2796

 

Each dark current measurement is calculated from a pair of exposures so the table shows the EXIF temperature of both.  They appear to be broadly in line with Roger's data.

 

Here is a graph of the average of the two temperatures plotted against dark current, on a logarithmic scale.

 

Canon7D2_DarkCurrentVsEXIF.jpg

 

It's more or less a straight line but to be honest I'm always a bit suspicious of the temperature reported in the EXIF header because it's not clear if it is reporting the actual sensor temperature or the temperature of some other component.

 

Mark
 


Edited by sharkmelley, 17 September 2017 - 10:03 AM.


#4 DuncanM

DuncanM

    Mercury-Atlas

  • -----
  • Posts: 2657
  • Joined: 03 Nov 2009
  • Loc: Arizona Sky Village or the rain forest

Posted 17 September 2017 - 04:05 PM

Thanks, that's very interesting.

 

Have you measured the dark current for any Nikon or Pentax DSLRs?



#5 Jim Waters

Jim Waters

    Gemini

  • *****
  • Posts: 3175
  • Joined: 21 Oct 2007
  • Loc: Phoenix, AZ USA

Posted 17 September 2017 - 04:20 PM

So how does low camera Dark Current directly relate to having noise free 'Dark Frames' and pattern background noise in astrophoto's?



#6 pfile

pfile

    Fly Me to the Moon

  • -----
  • Posts: 5291
  • Joined: 14 Jun 2009

Posted 17 September 2017 - 05:45 PM

as far as i can tell the temperature measurement in canon cameras is from a diode in some other component beside the sensor. i think those who have installed cold fingers on their canons show EXIF temperatures much higher than the setpoint of their cold finger.

 

rob



#7 Jon Rista

Jon Rista

    ISS

  • *****
  • Posts: 23857
  • Joined: 10 Jan 2014
  • Loc: Colorado

Posted 17 September 2017 - 08:33 PM

Canon EXIF temps are actually a measure of the DIGIC DSP, rather than the sensor, from what I understand. They are not actually indicative of the sensor temperature at all, really. In my testing of the 5D III, I've often found that while literal dark current measurements from the data may follow one curve, the EXIF temp values can follow an entirely different curve. Sometimes they correlate, sometimes they do not. 



#8 sharkmelley

sharkmelley

    Mercury-Atlas

  • *****
  • topic starter
  • Posts: 2510
  • Joined: 19 Feb 2013

Posted 18 September 2017 - 12:29 AM

So how does low camera Dark Current directly relate to having noise free 'Dark Frames' and pattern background noise in astrophoto's?

The dark current directly relates to thermal noise.  The thermal noise (in electrons) of an exposure is approximately the square root of the accumulated dark current (in electrons).  So a 300sec exposure with a dark current of 0.25e/sec will result in 75 accumulated electrons and a thermal noise of  8.7 (the square root of 75) electrons (RMS).  The thermal noise adds in quadrature to both the read noise and the skyfog background noise.

 

Fixed pattern noise is something else entirely.  It's very difficult to measure objectively and its visual impact tends to be far more annoying because it forms an obvious pattern that the brain immediately detects.

 

Thanks, that's very interesting.

 

Have you measured the dark current for any Nikon or Pentax DSLRs?

I don't have any reliable measurements for Nikon.  In the past, Nikon cameras have clipped their data at the black point.  For instance I have a dark frame from a Nikon D7000 where 80% of the pixels have a value of zero.  This totally upsets the calculation of dark noise making it appear a lot lower than it actually is.  The script that I use now detects clipping and warns the user but it cannot correct for it.  Other folk have published results for the Nikon cameras that don't clip and the dark current tends to be lower than the Canons tested above.

 

I have data for the Sony A7S but the Sony cameras employ spatial filtering in bulb mode exposures - the well-known "star eater" issue.  This reduces the apparent noise in long exposure darks.  With a correction factor I have calculated (star eater reduces standard deviation in the difference of 2 darks by approx. 20%.), it appears that the dark current of the Sony A7S is around 0.18e/pixel/sec at an ambient temperature of 20C, rising to 0.25e/pixel/sec after 2 hours of imaging.   So although it starts off higher than the Canons, the sensor temperature doesn't rise so quickly during typical deep-sky astrophotography.

 

It should be noted that Nikon cameras also employ spatial filtering but not so severe as the Sony version.  It is not clear to me how much effect this has on any calculation of dark current calculation but I don't think the effect is big. In any case, unlike data clipping, the script is unable to detect spatial filtering. 

 

Mark



#9 jforkner

jforkner

    Ranger 4

  • -----
  • Posts: 302
  • Joined: 04 Aug 2015
  • Loc: Copperopolis, CA

Posted 18 September 2017 - 08:26 AM

I’m curious…What is the value in knowing all this?  Does it really matter the kind & source of noise?  If the finished image exhibits noise, that’s not good—I get that.  But what do I gain by knowing my camera produces x-amount of y-noise?  Seems like if I can do something to reduce the noise in my finished product, that should be good enough (i.e., I don’t need to know how much dark noise was produced).

 

What am I missing?

 

Jack



#10 Qkslvr

Qkslvr

    Apollo

  • -----
  • Posts: 1343
  • Joined: 23 Jun 2006
  • Loc: NE Ohio, US

Posted 18 September 2017 - 10:21 AM

Each pixel is a reverse biased diode, Leakage currents are strongly linear with temp. So with a calibration you can use any pixel on a sensor as a thermometer. Each pixel is charged (the junction acts like a capacitor), and an exposure is a measurement of how much charge is left when measured. Leakage currents are the source of warm or hot pixels. This and photons cause a bit of energy to cross the junction depleting some of the charge, that is your signal (and dark currents). So dark's are just an uncalibrated temperature map of your sensor.

 

I calibrated my 300D, but found you can't just average the sensor, I took 10 10 sec(?)subs, as I didn't want any sensor heating during calibration. What I found though was 1 in 10 had cosmic ray hits, which threw off the averages.

 

I just picked a pixel, and used it and had a good calibration.



#11 sharkmelley

sharkmelley

    Mercury-Atlas

  • *****
  • topic starter
  • Posts: 2510
  • Joined: 19 Feb 2013

Posted 18 September 2017 - 01:51 PM

I’m curious…What is the value in knowing all this?  Does it really matter the kind & source of noise?  If the finished image exhibits noise, that’s not good—I get that.  But what do I gain by knowing my camera produces x-amount of y-noise?  Seems like if I can do something to reduce the noise in my finished product, that should be good enough (i.e., I don’t need to know how much dark noise was produced).

 

What am I missing?

 

Good question! 

 

For imaging the faintest objects, noise is always the limiting factor so it is good to know and understand the various sources of noise in an image.  It can be useful in planning an image acquisition strategy to ameliorate those effects e.g. faster F-ratio, longer exposures, dithering etc. and a processing strategy to make the most of the acquired data e.g. calibration frames.

 

More than that, a knowledge of the QE (quantum efficiency), read noise, dark current and other camera quirks (e.g. fixed pattern noise, spatial filtering etc.) can help in making an informed choice of camera for deep sky astrophotography.  For instance, if an entry level Canon has similar characteristics to a Canon 7D MkII then I might choose the less expensive camera.  If a Nikon or Sony has lower read noise and lower dark current than the Canon then I might choose the Nikon or Sony in preference to the Canon.

 

Unfortunately there is currently too little information available on dark current and sensor temperature rise during an imaging session.

 

Mark


Edited by sharkmelley, 18 September 2017 - 01:51 PM.

  • Qkslvr and Jon Rista like this

#12 sharkmelley

sharkmelley

    Mercury-Atlas

  • *****
  • topic starter
  • Posts: 2510
  • Joined: 19 Feb 2013

Posted 18 September 2017 - 01:54 PM

Leakage currents are strongly linear with temp.

No!  Dark current tends to double with every 5-7C rise in temperature.  It is not linear with temperature - it is exponential.

 

Mark


Edited by sharkmelley, 18 September 2017 - 01:55 PM.


#13 Qkslvr

Qkslvr

    Apollo

  • -----
  • Posts: 1343
  • Joined: 23 Jun 2006
  • Loc: NE Ohio, US

Posted 18 September 2017 - 02:23 PM

 

Leakage currents are strongly linear with temp.

No!  Dark current tends to double with every 5-7C rise in temperature.  It is not linear with temperature - it is exponential.

 

Mark

 

You are right Mark!

 

I meant it more as that leakage was a function of temps, and with a calibration they are thermometers. But it is an exp relationship, and not linear like I said.


  • sharkmelley and Herra Kuulapaa like this

#14 DuncanM

DuncanM

    Mercury-Atlas

  • -----
  • Posts: 2657
  • Joined: 03 Nov 2009
  • Loc: Arizona Sky Village or the rain forest

Posted 18 September 2017 - 02:41 PM

 

 

 

Thanks, that's very interesting.

 

Have you measured the dark current for any Nikon or Pentax DSLRs?

I don't have any reliable measurements for Nikon.  In the past, Nikon cameras have clipped their data at the black point.  For instance I have a dark frame from a Nikon D7000 where 80% of the pixels have a value of zero.  This totally upsets the calculation of dark noise making it appear a lot lower than it actually is.  The script that I use now detects clipping and warns the user but it cannot correct for it.  Other folk have published results for the Nikon cameras that don't clip and the dark current tends to be lower than the Canons tested above.

 

I have data for the Sony A7S but the Sony cameras employ spatial filtering in bulb mode exposures - the well-known "star eater" issue.  This reduces the apparent noise in long exposure darks.  With a correction factor I have calculated (star eater reduces standard deviation in the difference of 2 darks by approx. 20%.), it appears that the dark current of the Sony A7S is around 0.18e/pixel/sec at an ambient temperature of 20C, rising to 0.25e/pixel/sec after 2 hours of imaging.   So although it starts off higher than the Canons, the sensor temperature doesn't rise so quickly during typical deep-sky astrophotography.

 

It should be noted that Nikon cameras also employ spatial filtering but not so severe as the Sony version.  It is not clear to me how much effect this has on any calculation of dark current calculation but I don't think the effect is big. In any case, unlike data clipping, the script is unable to detect spatial filtering. 

 

Mark

 

The Nikonhack firmware mod and the Dark Current Enable Tool (USB command) allow for accurate measurement of Nikons that are compatible.

 

Most Nikons with Sony CMOS sensors are compatible with the Dark Current Enable Tool which is just a simple program fed to the camera via USB. It changes some CPU registers in the camera and is retained only in volatile memory, so that when the camera is rebooted the commands are returned to factory defaults.


  • sharkmelley likes this

#15 Lorien

Lorien

    Sputnik

  • -----
  • Posts: 41
  • Joined: 01 Dec 2013

Posted 18 September 2017 - 07:00 PM

 

Leakage currents are strongly linear with temp.

No!  Dark current tends to double with every 5-7C rise in temperature.  It is not linear with temperature - it is exponential.

 

Mark

 

Thank you very much for doing this test and publishing the data, since it contradicts previous claims that the 7Dii was much better.  I have a (perhaps dumb) question:  I thought that Canon subtracted dark current when making the RAW file, so that all you can measure is the residual dark noise.  At least, I think that is the case on my 5Diii.  If this is indeed Canon's approach, how are you measuring dark current on the 7Dii?  Are you inferring it from the residual noise?


  • DuncanM likes this

#16 sharkmelley

sharkmelley

    Mercury-Atlas

  • *****
  • topic starter
  • Posts: 2510
  • Joined: 19 Feb 2013

Posted 19 September 2017 - 12:32 AM

 

Thank you very much for doing this test and publishing the data, since it contradicts previous claims that the 7Dii was much better.  I have a (perhaps dumb) question:  I thought that Canon subtracted dark current when making the RAW file, so that all you can measure is the residual dark noise.  At least, I think that is the case on my 5Diii.  If this is indeed Canon's approach, how are you measuring dark current on the 7Dii?  Are you inferring it from the residual noise?

 

The 7Dii still remains much better than the selection of noisy cameras Roger was comparing it to.  My 7Dii dark current estimates are broadly in line with Roger's but they are still no better than Canon's entry level camera of 10 years ago, at least not for typical multiple long exposure deep sky astrophotography.  So I see no evidence of a paradigm shift in sensor technology in the 7Dii.

 

Yes, you're right that an estimate of the accumulated dark current is first subtracted before the raw file is written - this is necessary for a consumer camera.  I'm inferring dark current from the residual noise in the difference between two successive dark exposures.  It's necessary to take the difference of 2 frames, in order to remove any thermal fixed pattern noise.

 

Mark


Edited by sharkmelley, 19 September 2017 - 12:33 AM.

  • Jon Rista likes this

#17 Herra Kuulapaa

Herra Kuulapaa

    Viking 1

  • -----
  • Posts: 511
  • Joined: 10 Dec 2014
  • Loc: Vantaa, Finland

Posted 19 September 2017 - 04:53 AM

The Nikonhack firmware mod and the Dark Current Enable Tool (USB command) allow for accurate measurement of Nikons that are compatible.

 

 

Most Nikons with Sony CMOS sensors are compatible with the Dark Current Enable Tool which is just a simple program fed to the camera via USB. It changes some CPU registers in the camera and is retained only in volatile memory, so that when the camera is rebooted the commands are returned to factory defaults.

 

I measured a cooled and true dark current hacked Nikon D5100 some years ago. Unfortunately only with pixinsight noise tool:

D5100AC_Chart1.jpg

Shen measured it in eps though:

curve95.png

 

https://landingfield...g/dark-current/



#18 Qkslvr

Qkslvr

    Apollo

  • -----
  • Posts: 1343
  • Joined: 23 Jun 2006
  • Loc: NE Ohio, US

Posted 19 September 2017 - 08:47 AM

 

The Nikonhack firmware mod and the Dark Current Enable Tool (USB command) allow for accurate measurement of Nikons that are compatible.

 

 

Most Nikons with Sony CMOS sensors are compatible with the Dark Current Enable Tool which is just a simple program fed to the camera via USB. It changes some CPU registers in the camera and is retained only in volatile memory, so that when the camera is rebooted the commands are returned to factory defaults.

 

I measured a cooled and true dark current hacked Nikon D5100 some years ago. Unfortunately only with pixinsight noise tool:

D5100AC_Chart1.jpg

Shen measured it in eps though:

curve95.png

 

https://landingfield...g/dark-current/

 

The first is values plotted linearly, second plotted exp.



#19 sharkmelley

sharkmelley

    Mercury-Atlas

  • *****
  • topic starter
  • Posts: 2510
  • Joined: 19 Feb 2013

Posted 20 September 2017 - 06:41 PM

As a reminder of my testing methodology, I take continuous 5min exposures with the lens cap on in a dark room at an ambient temperature of as close to 20C as possible.  Dark current is estimated by subtracting successive pairs of frames, to reduce the effect of any fixed pattern noise.

 

The following graph includes the Nikon D7000 and the Sony A7S:

 

DarkCurrentGraphs1.jpg

 

To get around the problem of black clipping in the D7000 I had to estimate the shape of the distribution from the unclipped tail of the distribution.  To get around the problem of spatial filtering (star eater) on the Sony A7S reducing the standard deviation I had to apply a carefully calculated correction factor to increase the standard deviation.

 

Points of interest:

  • The Nikon D7000 dark current appears to be the lowest but it still goes up by a factor of 3 during the course of the test
  • The Sony A7S appears to start at a very high value but it only climbs by a factor of 1.5 during the course of the test.  The Sony does an excellent job of conducting heat away from the sensor, probably because it was designed for video.

Why does the Sony A7S have such a high initial dark current?  Could it be the big pixels?  Maybe dark current depends on the area of the pixel?  To test this idea I divided the dark current by the pixel area for each camera and produced a new graph showing dark current per square micron of sensor:

 

DarkCurrentGraphs2.jpg

 

When I do that, something extraordinary happens - the results from all the sensors bunch up very closely at the start.  Now remember that to generate the first point on the graph, two 5 minute frames were taken i.e. the first point is 10 minutes into the test and the sensor temperature would have been climbing during this period.  So it makes sense to extrapolate those curves back in time.  When this is done there is incredible agreement across all the sensors - a figure of around 0.0025 electrons/sec per square micron of sensor at a temperature of 20C.  It might be interpreted to mean that dark current has not changed at all in more than 10 years of CMOS sensor development.  Maybe there is a fundamental law of Physics at work here?

 

Certainly some recent cameras e.g. 600D and 7D MkII seem to really suffer from a rapid increase in sensor temperature during successive long exposures.  This must affect their suitability for multiple long exposure deep sky astrophotography.

 

Mark


Edited by sharkmelley, 20 September 2017 - 07:01 PM.

  • Jon Rista and tkottary like this

#20 Qkslvr

Qkslvr

    Apollo

  • -----
  • Posts: 1343
  • Joined: 23 Jun 2006
  • Loc: NE Ohio, US

Posted 21 September 2017 - 02:02 PM


  It might be interpreted to mean that dark current has not changed at all in more than 10 years of CMOS sensor development.  Maybe there is a fundamental law of Physics at work here?

 

It is, it's the physic's of the photodiode. The pixel size has stayed fairly constant iirc, so it's bulk silicon, and doping profiles. Which even if they changed, Not sure it would impact this much. Most the rest is Silicon carrier physics, and thermal properties.




CNers have asked about a donation box for Cloudy Nights over the years, so here you go. Donation is not required by any means, so please enjoy your stay.


Recent Topics






Cloudy Nights LLC
Cloudy Nights Sponsor: Astronomics