AF-100 and rolling shutter?

M

maarek

Veteran
Why do people constantly keep saying that rolling shutter has been fixed? Everytime a new CMOS-camera comes out the rolling shutter has been "fixed"! I remember the threads about the Sony V1 with it's improved rolling shutter. The EX1 and EX3 with their improved rolling shutter. ALL OF THEM have the EXACT rolling shutter as the old Sony A1.

There has been no improvements whatsoever in CMOS-land. Except for RED where the readout speed is slightly faster (though still noticeable).

Now the AF-100 is the newest hype victim. Does it really have reduced rolling shutter compared to other cmos cameras? The previews are always the same. "It has reduced rolling shutter" and when I test it, it's exactly the same.

Same thing when the canon 7d came. Everyone was like "it has less rolling shutter than the 5d". Well it did not. Except in 720p mode but that has horrible aliasing and is almost useless.

If the AF-100 has seriously improved it would be absolutely awesome as I consider rolling shutter to be way worse than 4:2:0 or other slight detriments.
 
I'll take first stab here

It has rolling shutter. Its reduced. How much.. we never tested it fully nor did we set out to. That is all I can say really. We presumed we would see it while shooting, whatever we saw was marginal. We thought we'd find it in post and we did find something in a whip pan, again it was marginal.
Beyond that, i will differ to the experts who will be testing this fully. CMOS is CMOS I expect full rolling shutter. So we know its there but in our opinion we rated it as acceptable. And before anyone bites my head off for not testing it, I can tell you that was not our purpose.

If it was we would have brought it to our techie and he would have needed a few hours. This was not the case. I expect tons of tech reports in the weeks to come, what I wanted to achieve was a good sample video with as much indications I could provide with a subject that was interesting to watch.

Rolling Shutter? Yes.. Reduced? Yes How much? Don't know.
 
Last edited:
you must shoot a lot of hand held stuff :P
cmos will always have rolling shutter
if you cant handle it, too bad. there are no ccd sensors right now under 10k that will match the advantages of cmos.
 
Fact of the matter, at least as I understand it, is the rolling shutter is chiefly (if not solely?) a product of the sensors "read-reset time". This is how long it takes the sensor to read out all of its lines and then get back to square one for image capture again.

The longer the read-reset time, the greater the appearance of rolling shutter. The shorter the read-reset time, the closer the sensor works to a global shutter and the less disturbing rolling shutter artifacts appear. For a hypothetical example, a camera with a 20ms read-reset time moving past a stationary object 10 ft away with a velocity of 15mph may show it with 15° degrees of skew from vertical. A camera with a 5ms read-reset time moving past the same object in the same situation may only show 4° of skew from vertical. I don't know if those numbers are even close to right, but I believe the principle stands.

I'm pretty sure it's possible to put a number on this phenomenon to provide at least a basis of comparison. I'd love to see read-reset time being listed in specification sheets.

All that being said, I'm not an engineer...I'd love to know if this is correct, and if not where I'm missing the point.
 
Think of the CMOS chip as RAM memory chips, except in the case each pixel is an analog gathering of light. So the idea of a reset to zero is correct. Like a RAM memory chip, any row or column can be read at any time. However, by reading it in order, left to right, top to bottom, we get the best results.

Now being CMOS, the faster we read the chip, the more heat is generated. CMOS consumes more power the faster you clock the chips. IF it was just a single pass, we could clock it very fast and reset, read (left to right, top to bottom) very quickly. After that, the chip would have time to cool off.

However, in video we have to keep reading the array over and over. Now, if heat weren't a problem, we could rip through the array in record time, but doing this over and over with each frame would overheat the chip.

IF we compare a large chip to a small chip, the larger chip can dissipate heat more easily (more area to do it). So a really small chip might be able to cover 1920x1080 pixels in 1/30 of a second, but a larger chip might be able to do it in 1/120 of a second. In the case of the slower smaller chip, the jello effect is more pronounced compared to the larger chip.

Another area that presents a problem is the A/D converter. A total of 1920x1080 pixels must be converted in 1/30 of a second or less. If you intend to read the array 4 times faster, you have to convert 4 times faster.



Bob Diaz
 
That's very interesting, Bob. I appreciate the discussion!

Quick question - does photosite size affect things?

Assuming two theoretical 1cm x 1cm CMOS chips. One is 100x100, the other is 400x400. If they have the same read-reset speed, will one get hotter more quickly? And, if so, what can be done to counteract this? How are CMOS chips cooled...heatpipes?
 
This is going to get into some heavy stuff real fast, but I'll try to keep it simple. First of all, I'll have to assume that the circuits that drive the chips are equal as far as design, conversion, and construction.

The size of the image sensor is the same, but 100x100 (10,000 total) has less photo elements than the 400x400 (160,000 total) sensor. Thus the area of each single sensor is 4x4 (16 times) larger on the 100x100 sensor. This makes the 100x100 more sensitive to light.

A single pass on the 100x100 has to read 10,000 elements, but on the 400x400, it reads 160,000 elements. More work = more heat. Now if we only read 10,000 elements on both at the same speed, it should be the same heat generated.

With computers, where speed is everything, it's common to put a heat sink with a fan or even a heat sink with a special cooling system. However, that's not so easy on an image sensor; we need the light to reach the face of the sensor.

It's possible to cool from the back side of the image sensor, but I don't know of any camera doing that. Please understand that there could be some really high end camera that does that and because I'm not into super expensive cameras, I won't know about it.

Heat posed a real problem for JVC, their HD100 camera had to split the CCD chip into two parts to get the full 1280x720 pixels. Many here remember the SSE (Split Screen Effect) caused by having dual A/D Converters that didn't always match exactly.

Going to CMOS allows for higher pixel counts without the excessive heat problem, but as one problem is solved (excessive heat on small chips), another is created (rolling shutter). It is possible to design a CMOS chip with a global shutter, BUT the change requires more area of the chip and that gets into the way the image sensor elements. The result is smaller image sensor elements and even worse low light performance.


Side note:
I was really impressed when Sony developed the IMX017CQE 6.4M-Effective Pixel High-Speed High-Resolution CMOS Sensor for Consumer Products. It was able to do 1080/60p output at 10 bit resolution by using parallel A/D converters. I had high hope that some day Sony would make a high end version of the chip using faster Successive Approximation A/D Converters, but to date, it hasn't happened.....

Maybe someday someone will develop a CMOS chip with a faster A/D conversion without the excessive heat. In that case, the reading of the chip would be faster and the impact of the rolling shutter would be less.


Bob Diaz
 
BobDiaz said:
This is going to get into some heavy stuff real fast, but I'll try to keep it simple. First of all, I'll have to assume that the circuits that drive the chips are equal as far as design, conversion, and construction.

The size of the image sensor is the same, but 100x100 (10,000 total) has less photo elements than the 400x400 (160,000 total) sensor. Thus the area of each single sensor is 4x4 (16 times) larger on the 100x100 sensor. This makes the 100x100 more sensitive to light.

A single pass on the 100x100 has to read 10,000 elements, but on the 400x400, it reads 160,000 elements. More work = more heat. Now if we only read 10,000 elements on both at the same speed, it should be the same heat generated.

With computers, where speed is everything, it's common to put a heat sink with a fan or even a heat sink with a special cooling system. However, that's not so easy on an image sensor; we need the light to reach the face of the sensor.

It's possible to cool from the back side of the image sensor, but I don't know of any camera doing that. Please understand that there could be some really high end camera that does that and because I'm not into super expensive cameras, I won't know about it.

Heat posed a real problem for JVC, their HD100 camera had to split the CCD chip into two parts to get the full 1280x720 pixels. Many here remember the SSE (Split Screen Effect) caused by having dual A/D Converters that didn't always match exactly.

Going to CMOS allows for higher pixel counts without the excessive heat problem, but as one problem is solved (excessive heat on small chips), another is created (rolling shutter). It is possible to design a CMOS chip with a global shutter, BUT the change requires more area of the chip and that gets into the way the image sensor elements. The result is smaller image sensor elements and even worse low light performance.


Side note:
I was really impressed when Sony developed the IMX017CQE 6.4M-Effective Pixel High-Speed High-Resolution CMOS Sensor for Consumer Products. It was able to do 1080/60p output at 10 bit resolution by using parallel A/D converters. I had high hope that some day Sony would make a high end version of the chip using faster Successive Approximation A/D Converters, but to date, it hasn't happened.....

Maybe someday someone will develop a CMOS chip with a faster A/D conversion without the excessive heat. In that case, the reading of the chip would be faster and the impact of the rolling shutter would be less.


Bob Diaz
Click to expand...

Bob, great post and breakdown. it is likely that the first global shutter CMOS chips (that will be 'affordable' to the prosumer/intro professional market) will be low (relative to current HDSLR's or prosumer camcorders) pixel count, not extremely low-light sensitive (due to heat generation), and will probably generate a LOT of complaints among the majority who do not understand the intricacies involved. i highly doubt they will offer high framerates (heat and processing power) or high ISO's (again, heat), but still will be a phenomenal step forwards. skew is an issue for a lot of what i shoot (track events) as is rolling shutter (flash at events/weddings); global shutter, even at 1280x720P 30P/25P/24P with a decent 100-1600 ISO range would MORE than suffice (i shoot at 200 or 400 ISO, ABSOLUTE MAX 800 as we employ a LOT of lighting), and i think a LOT of professionals would agree. 1920x1080P 30/25/24P would be icing on the cake. Even if it was a native 1920x1080 or 1280x720 sensor, global shutter would be a HUGE step forwards (no more half-exposed frames, no more CCD wide-open iris vertical streaking, no more skew in the quick pans some people insist on using, etc).

I am referring to global shutter solutions under 4-5k USD or 2-3k Euro, body. By my estimate, 2-3 years off.
 
You must log in or register to reply here.
Back
Top