PDA

View Full Version : CMOS Waveform Distortion



Larry Rutledge
10-27-2008, 10:02 PM
Click here to read the full article (http://www.dvxuser.com/articles/article.php/16)

PerroneFord
10-27-2008, 11:45 PM
Fascinating test! I have a question. Given the results would it be fair to say that the faster the rolling shutter of a CMOS camera, the less noticeable the skew? And thus the more it would replicate the performance of a global shutter?

If this assumption is true, then it would seem to me that the way to increase performance over time with CMOS cameras is to improve the rolling shutter rate. Thus as time to read the sensors from top to bottom decreased the improvement in the image would improve geometrically.

Joe Lawry
10-28-2008, 01:53 AM
A great report, thank you very much for posting it.

Barry_Green
10-28-2008, 09:03 AM
Perrone, your understanding is correct; shortening the read/reset time helps reduce the skew effect (and all other effects, wobble etc). Thomson uses this technique in their Infinity camera; they overcrank their sensor to 2x speed so when shooting 60p it's actually running at 120fps and discarding every other frame. Red has been working on shortening up their read/reset time to improve the rolling shutter characteristics of their cameras too.

Buck Forester
10-28-2008, 11:29 AM
So what I'm gathering is that since the earth spins counter-clockwise, with CMOS to limit skew it's best to pan clockwise. :

Interesting tests, thanks for posting.

Anhar Miah
10-29-2008, 04:09 PM
Very interesting article! well written Mikko.



So what I'm gathering is that since the earth spins counter-clockwise, with CMOS to limit skew it's best to pan clockwise. :

Interesting tests, thanks for posting.



Well :P sir, slight mis-calculation there, since the gravitational force keeps the observer moving along with the earth, the relative velocity of the observer to the earth is zero. Thus panning clockwise would induce skew :D (I know, I know you was kidding :) )

Anhar

Buck Forester
11-03-2008, 07:57 PM
I'm curious, is this test done bypassing Long GOP and going straight from the sensors via HD-SDI? And if so, would a very slow, creeping pan on a dolly be enough movement to induce this loss of resolution on a CMOS camera using NanoFlash, even with the high bit i-frame option?

mikkowilson
11-04-2008, 12:24 PM
This test used live (composite) video outputs from the cameras.
The final output was recorded to regular MiniDV.

The test demonstrates that there is no loss of resolution (from the sensor) while panning, but rather the 2D skew from rolling shutter (from the CMOS camera) distorts the 1D display of a waveform monitor.


- Mikko

Buck Forester
11-04-2008, 03:07 PM
Thanks, Mikko. I'm not 100% sure what you said though because I'm technically incompetent. I'm assuming live composite video out bypasses the internal camera codec, so what you were seeing had nothing to do with Long GOP? Also you say there is no 'technical' loss of resolution from the sensor while panning, but is there apparent visual loss due to this 2D/1D (not even sure what that means) distortion? Essentially I just care what my eyes see and what I can do about it, not what a machine reads.

Maybe my question is how fast were your pans and do you think there is any skew at all during a very slow, creeping pan that would distort the display of the waveform monitor? I mean VERY slow, just enough to show the camera is not static. I wouldn't think a CMOS chip would skew with such little movement, but I don't know.

Reading Barry's original comments about apparent loss in sharpness during a pan, he mentioned he didn't think it was the codec but some type of 'agressive noise reduction'. I'm on a quest (until I get my hands on a NanoFlash and/or get my SD-HDI set up for my own tests) to figure out what is causing what my eyes are seeing as a drop in sharpness (but not everyone else sees it apparently). If it's 2D skew distorting a 1D display (I have no idea what that means but I wanted to say it to sound cool), is there a work-a-round? If it's 'agressive noise reduction', is there a work-a-round? (I don't know if that's chip or codec related). If it's codec there's a work-a-round (yay NanoFlash!), if it's skew it's something I'll have to work-a-round it myself with technique until the 2nd generation of CMOS comes out in three 2/3" chips for $999. Or maybe all this is made up in my head. I'm beginning to lose sharpness myself. Somebody please slap a NanoFlash on my brain.

PerroneFord
11-04-2008, 03:25 PM
Buck,

Essentially, what they are saying is that the tests were performed independent of the codec. And though the resolution and actual sharpness of the image didn't change, the image DID skew, and WILL skew of the camera is panned (and apparently dollied). This is a problem at the chip level. As is any noise reduction.

I did a test for myself last week in looking at resolution with a wide open aperature and the lens fully zoomed out. What was interesting to me was watching the peaking as I moved the camera. On very slow movement, there was some minor loss of peaking. The faster the movement, the more peaking was lost. A quick movement, and all peaking was lost.

This led me to believe that with quite slow movement, excellent sharpness could be maintained even with the skew. The question then becomes, how slow would a person move the camera.

Buck Forester
11-04-2008, 04:06 PM
Interesting, Perrone. I guess where I'm concerned is that I am doing very very very slow pans and I see can visually see the drop in sharpness. I'm looking at text on book spines as I s-l-o-w-l-y pan. So if this is in no way codec related but a function of the sensors, then NanoFlash would do me no good. Dang. But hey, I love everything else about this camera, so I guess it's something I'll have to live with. I hope Scarlet solves this issue for me. Again, I must always note, this may just be my ignorance on how video is capture and may be normal motion effects, I'm new to videography. I wish I had a CCD camera to see how it performs on horizontal slow pans against a static foreground subject. I wish so bad there was a CCD based camera shooting EX1 resolution at a EX1 price. It sounds like Scarlet may have a better rendention of motion for CMOS, if not I'll have to throw some $$ and get a Sony PDW700. I eventually want one of those anyway.

MitchLewis
11-04-2008, 04:57 PM
When the EX1 first came out, I remember the review mentioning that the resolution would look great when the camera was totally still. But the minute you start moving it, the resolution would drop. Originally this was a concern, but after seeing the great video it produces I'm not worried anymore.

I find these tests fascinating and much appreciate the people who take the time to preform them, but real world results are also a good "test". Some day we'll have the best of both worlds.....great resolution AND no rolling shutter issues. :)