20D background sensor noise - DPChallenge Forums

Canon EF 28-135mm f/3.5-5.6 IS USM

03/03/2005 10:30:48 AM · #1

Inspired by a similar thread somewhere, can't remember exactly where, i decided to take a couple of images of the 20D sensor noise (long exposures, no lens, bodycap on tightly).

This photo was shot in raw at iso100, white balance at daylight, no sharpening or hue adjustment. Dynamic range reduced and contrast boosted massively in postprocessing to bring out as much detail in the noise as possible, and scaled down to a more reasonable resolution and to make the patterns more obvious:

//www.tardis.ed.ac.uk/~rain/temp/noise%206%20mins%20100%20small.jpg

I was most interested in how you could see certain patterns in the noise of the example posted in the other thread - so i wondered what would appear on the 20D's sensor.

Looking in the above image, most obvious is the violet semi-circle halo in the bottom right, this is almost certainly noise picked up as infra-red from a warmer element. The example photo in the other thread had five such things down the right side, i was interested to see the 20D only has one. If anyone knows the thread i meant btw, please post a link.

Other interesting things to notice if you look carefully are the broad but very faint check patterns of alternating bluish and reddish noise, which reveal something of the structure of the CMOS. Many finer vertical and horizontal lines are also clearly visible, but i'm not certain as to whether this is electronic noise produced in the sensor or after the capture.

If anyone knows more about the CMOS and can explain the individual patterns, i'd be very interested in finding out more :)

Message edited by author 2005-03-03 10:32:04.

BikeRacer

Canon EOS-300D Rebel

03/03/2005 10:40:45 AM · #2

Wasn't there a banding issue with the early firmware? Do you have the latest firmware?

03/03/2005 10:50:43 AM · #3

Yeah, this is the latest firmware. I don't think it's a software issue, either way the noise here is magnified a massive amount and is totally invisible when not enhanced, even after a six minute exposure - so most of the noise is simply at the one very lowest detectable level and the darker bands i suspect are noticeable because noise there falls below the threshold for that level and comes out as complete black. Either way, the banding is both horizontal and vertical so i suspect it has more to do with the arrangement of signal carrying conductors in the cmos...

03/03/2005 05:40:08 PM · #4

Interesting.....

Did you have the rubber eyepiece cap thingy on as well?

I might repeat the experiment and see what difference there is between cameras.

Cheers, Me.

03/03/2005 06:04:26 PM · #5

Just thinking about your questions for the patterns...

I don't know a lot about the actual design of the sensor itself, but I do know a fair bit about digital electronics...

So, my theory for the day... (Slow day at work, long post...)

I'd suspect that the banding would be caused by differences in the analogue shift-registers used to get the data out of the CMOS array and into the A/D converters in the DIGIC processor. There would also be some type of gain control for the pre-A/D amplifiers which is why there was the high-ISO banding problem, as the DIGIC processot may not have been managing the gain control of these amplifiers correctly in firmware.

I'm making the assumption here that the cmos array itself dosn't contain the A/D circuitry, which seems logical, as the heat generated, and the space taken up on the chip would not lend itself to high quality images.

Basically, each CMOS pixel would gather the accumulated light level as a charge stored in a small capacitor at the pixel site.

It is very unlikely that the sensor has 8.2million connections to the Digic processor, so the reading of the sensor must be multiplexed in some fashion. Most likely by using some sort of 'bucket brigade' analogue shift register, or by using a row/column multiplexing arrangement.

The overall banding would be caused by slight differences in the circuitry associated with the multiplexing.

Stretching it even further.. If we assume that it's a bucket brigade, and the data is shifted horizontally off the sensor...

The horizontal bands would be caused by slight differences in the responce of the circuitry shifting the data off the sensor for each row. The banding could also be caused by amplifier differences in the A/D converters in the DIGIC processsor. As the banding seems to be in large 'clumps' I'd say this effect could also be caused in part by circuit board trace lengths, paths or stray capacitance differences between groups of lines that are grouped on the circuit board.

The vertical banding would come from leakage from the digital addressing circuitry controling the bucket brigade into the analogue data stream. I say that as it appears more regular across the frame, and is at a higher 'frequency' than the 'clumped' banding of the horizontal banding.

The 'hot spot' on the bottom right is, as you say, more than likey caused by another element in the camera, either thermal or radiated interfereance.

The overall random splotchy patterns are just noise, of the pure electrical, alpha-ray, backgound electrons bouncing about kind.

- This is all just guessing obviously, but seems perfectly plausible to me. :-)

Cheers, Chris H.

Canon EF 55-200mm f/4.5-5.6 II USM

03/03/2005 06:28:41 PM · #6

That's from mine... The 'hot spot' is the same, Horizontal and vertical banding similar...

I shot mine in jpeg, parameter 1, daylight WB. (I don't have any raw software here at work) 462 seconds with the remote, in the camera bag with the top zipped closed, and body cap on.

Intresting that my noise is predominantly green, although that is probably as a result of using jpeg rather than raw, as a lot of data is lost in the conversion, particularly below the 'noise floor' where we're looking.

ed: Ooh, and I've got some hot pixels on the frame too that don't show up in this... No suprises there.

Cheers, Chris H.

Message edited by author 2005-03-03 18:30:07.

orussell

Canon EOS Rebel T6s

03/03/2005 08:32:08 PM · #7

I notice that there is an anomalous spot on both sensors in the bottom right.

03/04/2005 06:10:54 AM · #8

Thanks KiwiChris - very interesting stuff! I'd be fascinated to see what difference in patterns you get on other Canon cameras (and other brands too). Anyone with a 300D, 10D, 1D or D60 fancy posting their results? :)

03/04/2005 06:31:14 AM · #9

Originally posted by orussell:

I notice that there is an anomalous spot on both sensors in the bottom right.

I think that's heat from the battery affecting the sensor. Doesn't the 20D use a smaller battery that the other cameras? Or am I getting that fact confused with the XT? I turned my 300D on cold and it wasn't there. I'll post a version later after leaving it on for a while.

Message edited by author 2005-03-04 06:40:36.

03/04/2005 06:47:45 AM · #10

Originally posted by Konador:

Originally posted by orussell:
I notice that there is an anomalous spot on both sensors in the bottom right.

I think that's heat from the battery affecting the sensor.

Urro, it wont be the battery... The 'hot spot' is on the bottom right of the output image, and therefore the top left of the sensor looking from the back of the camera. The wrong side for the battery.. Also the battery is quite a distance from the sensor, so it'd have to be on fire to cause problems, and the battery is a large warm object, not a small point heat source relative to the size of the sensor.

If it is indeed caused by heat (which we're guessing at) it would have to be a high disipation component to have any effect, possibly part of the switch mode power supply used to charge the capacitor for the flash, or a voltage regulator associated with the sensor itself.

Given the 'spot' nature of the pattern I'd say the latter is more likely, as it's obviously something close to the sensor itself, and any electronics engineer worth his salt would isolate the flash circuitry both physically and electrically from the sensitive sensor components.

Anyone got a circuit diagram, and assembly manual for the 20D? I'm sure I could figure it out from that. :-) If not it'd be fascinating to look at from a engineering/geek perspective.

Cheers, Me.

03/04/2005 06:49:29 AM · #11

Originally posted by KiwiChris:

Originally posted by Konador:

Originally posted by orussell:
I notice that there is an anomalous spot on both sensors in the bottom right.

I think that's heat from the battery affecting the sensor.

The 'hot spot' is on the bottom right of the output image, and therefore the top left of the sensor looking from the back of the camera. The wrong side for the battery.

Ahh right, I didnt realise the sensor image was flipped.

03/04/2005 06:51:44 AM · #12

Originally posted by Konador:

Doesn't the 20D use a smaller battery that the other cameras? Or am I getting that fact confused with the XT? I turned my 300D on cold and it wasn't there. I'll post a version later after leaving it on for a while.

Nah, it's the Xt that uses the smaller battery...

Very interesting that the 300D is similar in noise pattern, it shows they've not changed their technology much between generations of the sensor, just packed more pixels in..

Although the marketing folks tell us there is on-chip noise filtering as well, we'd never know from these images as we're deliberately enhancing it.. :-).

Be interesting it someone could do the same for a 1DII or 1Ds II which supposedly have a different sensor again...

edit: More importantly, a Nikon D70.. Anyone?

Cheers, Me.

Message edited by author 2005-03-04 06:54:17.

03/04/2005 06:57:40 AM · #13

I only used a 30 second exposure, not bulb mode as i don't have a remote. I've just placed a chair on my camera though to get a long bulb shot :P

03/04/2005 07:06:26 AM · #14

Originally posted by KiwiChris:

If it is indeed caused by heat (which we're guessing at) it would have to be a high disipation component to have any effect, possibly part of the switch mode power supply used to charge the capacitor for the flash, or a voltage regulator associated with the sensor itself.

Given the 'spot' nature of the pattern I'd say the latter is more likely, as it's obviously something close to the sensor itself, and any electronics engineer worth his salt would isolate the flash circuitry both physically and electrically from the sensitive sensor components.

What i found fascinating was that in the other thread (dammit, still can't find the thread, or remember what camera it was) there were five such spots, evenly spaced, down the right hand side of the image. Same size and distribution. If you look at it closely you realise it's not just a spot, but a semi-circle the inside of which is darker than the halo produced by the outside. This has got me quite curious now, i've really no idea what exactly it could be...

Tamron SP AF 28-75mm f/2.8 XR Di for Canon

03/04/2005 07:07:27 AM · #15

Originally posted by Konador:

I only used a 30 second exposure, not bulb mode as i don't have a remote. I've just placed a chair on my camera though to get a long bulb shot :P

:-).. I hope it's not too heavy...

ganders

Canon EOS-40D

03/04/2005 07:10:54 AM · #16

Originally posted by KiwiChris:

edit: More importantly, a Nikon D70.. Anyone?

Your wish is my command...

ISO1600, 1 minute(ish) exposure. Obviously some fairly heavy-handed level tweaking to make things visible :-)

03/04/2005 07:18:37 AM · #17

Originally posted by ganders:

Originally posted by KiwiChris:
edit: More importantly, a Nikon D70.. Anyone?

Your wish is my command...

ISO1600, 1 minute(ish) exposure. Obviously some fairly heavy-handed level tweaking to make things visible :-)

Cor...

Obvious horizontal banding, but none of the vertical visible on the Canon sensor...

The two spots at top are interesting... Similar thing to the Canon, a nearby component causing thermal or elctrical interferance.

And what's even more interesting, IMHO is the pattern at the bottom, about 1/4 of the way across. That's more of a sensor defect or electronics error looking noise pattern. I'd be very suprised if something that defined got through R&D at Nikon. I can see why the'd be OK with the 'blobs' we're seeing in this and the canon sensors, but the more defined geometric shapes would be a no-no I would have thought. (And the bottom right corner...)

Thanks for that one..

Cheers, Me.

Message edited by author 2005-03-04 07:33:29.

Tamron SP AF 28-75mm f/2.8 XR Di for Canon

03/04/2005 07:32:03 AM · #18

Originally posted by riot:

If you look at it closely you realise it's not just a spot, but a semi-circle the inside of which is darker than the halo produced by the outside. This has got me quite curious now, i've really no idea what exactly it could be...

OK, I'll go for another theory here, put your thinking caps on...

What about ground plane differential error? (What you say?)

I imagine that in the make-up of the sensor there is a 'sheild' or ground plane layer in the device itself, appart from the physcal bit of metal it's attached to.

If the top left (bottom right in the image) at the center of the 'blob' were the electrical connection point for that ground plane it would make sense that you'd get a donut effect around it in the output from each photosite..

My reasoning being that the Canon marketing folks tell us that there is 'noise reduction circuitry on the sensor' from which I deduce that there are small signal amplifiers beside each photosite on the silicon. If they all reference their analogue 'ground' or 0 volt reference to the ground plane, there would be a small amount current flowing in the ground plane.

For the purposes of cancelling noise the amplifier would compare a stable voltage (ground plane, 0 volts) to the light level, and output a 'damped' version of the light level to be shifted out of the sensor via the bucket brigade (Building on my previous theory). This would remove 'common mode' noise such as Magnetic and external electro magnetic interfearance from radios etc.

So anyway, beacuse any amplifier has a 'dead band' where the inputs change, but nothing happens on the output, and you get a voltage drop across any conductor if there is current passing through it, you should get a 'band' around the point of contact for the ground plane layer to the electrical ground of the rest of the camera. Good analogue theory says you have only one point of contact, which almost makes this plausible.

So.. At the senter of the contact (The donut) there is no issue with 'extra' noise as the ground plane voltage is solid and stable. Moving out from it you get a linear voltage drop, radiating out towards all other parts of the sensor surface. Hence the pixel sites and therefore amplifiers in a 'donut' around the point of contact would have to operate with their reference voltage in the 'dead band' range unless the noise or light level was large enough to overcome the voltage drop across the ground plane.

Not sure that makes sense to anyone but me, but heck, it's late...

Of course the Nikon one with two spots sorta points straight back at the thermal theory, but heck, it's all good fun...

By all means theorise amongst yourselves, its 1:31am here, and I'm off to bed.

Cheers, Me.

Message edited by author 2005-03-04 07:35:59.

ganders

Canon EOS-40D

03/04/2005 07:35:38 AM · #19

Originally posted by KiwiChris:

And what's even more interesting, IMHO is the pattern at the bottom, about 1/4 of the way across. That's more of a sensor defect or electronics error looking noise pattern. I'd be very suprised if something that defined got through R&D at Nikon. I can see why the'd be OK with the 'blobs' we're seeing in this and the canon sensors, but the more defined geometric shapes would be a no-no I would have thought. (And the bottom right corner...)

To be fair, some of that might be leakage from the viewfinder or something. As it's a (relatively) short exposure then massively boosted to make the noise visible it's probably not a very fair comparison. I'm just too lazy to hold the button down for 460 seconds like some people :-)

Canon RF 24-70mm f/2.8 L IS

03/04/2005 07:47:06 AM · #20

2767 sec (about 45 min) exposure on 300D. Much less banding, many more hot pixels, more obvious change on the right hand side with 4 semi circles, 2 pairs of 2, one at the top and one at the bottom.

Message edited by author 2005-03-04 07:52:10.

kirbic

Canon EOS R5

03/04/2005 08:32:13 AM · #21

Some of what Ganders posted looks like light leakage. Konador's pattern is classic for the 300D and 10D, all these cams seem to show the same type pattern on the same side of the sensor. The most plausible explanation I have seen is that the sensor is being warmed along that edge by the heat generated from either the electrical current flowing in the sensor power leads, by nearby electronics, or both.