dual shutter speed

[larrycafe]

instead of dual ISO for different exposure in one single picture.

would it be possible to have dual shutter speed in one single shot? certainly it is not possible with mechanical shutter curtain. But with the electronic first curtain, would it be possible?

with this, in very bright light condition, we can shoot at ISO 100, half of the lines are at 1/4000, half of the lines are at 1/125.

I am not sure how much benefit to the picture quantity comparing to use dual ISO in this situation.

[ItsMeLenny]

Firstly, me personally am not sure if it's possible.

But you should check out http://www.clarkvision.com/articles/iso/
it gives a good explanation of digital ISO,
and which may show that higher iso isn't as bad as what you possibly think?

[larrycafe]

in the case of bright sunlight
ISO 100 f4 1/1600s

If Dual ISO use ISO 1600, which is +4 stops

It might yield the same exposure for ISO 100 at 1/100s, which is +4 stops also.

so it is comparing ISO 1600 at 1/1600s, to ISO 100 at 1/100s.

ISO 100 at 1/100s should be having better tonal range and color sensitivity, and might give better DR also when combining.

It also hold true if you are doing exposure bracketing, if object motion is not an issue, and you are using a tripod, most of the people will do Tv bracketing instead of ISO bracketing.

[Marsu42]

and you are using a tripod, most of the people will do Tv bracketing instead of ISO bracketing.

Btw ML also has the option to combine Av/Tv with iso bracketing, the latter taking half the ev load to prevent too long shutter times (for example for landscape cloud movement) or too small apertures (diffraction).

[larrycafe]

thanks Marsu42. yes, the combine Av/Tv with ISO bracketing is good.

however, I am not sure how much better it can be comparing to dual ISO, could be better shadow, color and DR.

[Audionut]

so it is comparing ISO 1600 at 1/1600s, to ISO 100 at 1/100s.

ISO 100 at 1/100s should be having better tonal range and color sensitivity, and might give better DR also when combining.

Exposure is controlled by shutter/aperture.  In the scenario you describe, the ISO 100 at 1/100s exposure will without doubt enable a better reproduction of the scene.  Not necessarily because you are using a lower ISO, but because you are allowing 16x the number of photons to hit the sensor.

ISO is post exposure gain.

ISO is useful (in ISOful cameras (Canon)) where you can no longer adjust exposure settings to allow more photons to hit the sensor.  Here it becomes useful to apply gain (ISO) to the voltage from the sensor before it hits the noisy Analog to Digital Converter.  You should always attempt to dump as much voltage into the ADC as possible (before overexposure).  Voltage from photons hitting the sensor are best, voltage gain is second.

[kgv5]

I believe REDs HDRx is based on different shutter speeds but not in single frame of course but one frame after another like our HDR. This results clean picture but with some motion artifacts (different motion blur in dark and light frame). It needs to be recorded with double framerate.

[ItsMeLenny]

I believe REDs HDRx is based on different shutter speeds but not in single frame of course but one frame after another like our HDR. This results clean picture but with some motion artifacts (different motion blur in dark and light frame). It needs to be recorded with double framerate.

I could be wrong but I think it has two sensors. The 2nd sensor records at a faster shutter speed, which can cause motion artifacts like you say.
But also, the reason they claim to have that feature is to save highlights in case one accidentally blows them out. I think it's more recommended to expose properly in the first place

[larrycafe]

correct me if I am wrong, there is electronic first curtain in live view mode.

if we can make the first curtain running in two different timing, like the odd lines are initialized earlier for longer shutter speed, the even lines are initialized later for shorter shutter speed. and then the physical shutter curtain end the exposure, we will be able to have dual shutter speed in one single picture.

[ItsMeLenny]

I think that would go against how a CMOS sensor works.

Dual ISO still dumps sensor data line by line, it just alternates the range of the data it is collecting.

[dmilligan]

correct me if I am wrong, there is electronic first curtain in live view mode.

if we can make the first curtain running in two different timing, like the odd lines are initialized earlier for longer shutter speed, the even lines are initialized later for shorter shutter speed. and then the physical shutter curtain end the exposure, we will be able to have dual shutter speed in one single picture.

Yes this is theoretically possible, but you'd have to use electronic first and second curtain (or you'd get a gradient from top to bottom due to the way electronic shutter works), and you'd potentially get rolling shutter type artifacts. You also can't do even/odd rows because of the bayer pattern or you'd end up with all blue pixels at one shutter speed and all red pixels at another (read the dual iso pdf by a1ex). So you'd need to do every two rows.

I think that would go against how a CMOS sensor works.

The fact rolling shutter exists, means that this is 'theoretically' possible. Rather than going down through the sensor and turning the pixel rows on line by line (which is what video mode is doing, and is the cause of rolling shutter), you could skip lines and come back and turn them on later, then do the same thing when turning them off and reading out the data. Resulting in some lines being 'on' for less time, resulting in a shorter shutter speed. (With rolling shutter, all the lines are 'on' for the same amount of time, but they are not all on at the same time, they are turned on, then turned off sequentially line by line).

The problem is, we don't know if the hardware can even actually be programmed to power up/read out lines non-sequentially (the hardware itself, might only have a 'rolling shutter' mode and only be capable of sequntially turning on rows) and then even if it can, can somebody actually figure out how to do it (would require extensive reverse engineering).

Also, ML doesn't know how to use or control an electronic shutter in photo mode, so something like this might only be able to work for video or for silent pictures captured via LV (which would probably mean a lower resolution).

[ItsMeLenny]

Rather than going down through the sensor and turning the pixel rows on line by line (which is what video mode is doing, and is the cause of rolling shutter), you could skip lines and come back and turn them on later, then do the same thing when turning them off and reading out the data.

This would result in a sort of interlaced rolling shutter.

[l_d_allan]

Would it be possible to have dual shutter speed in one single shot? certainly it is not possible with mechanical shutter curtain. But with the electronic first curtain, would it be possible?

IANACMOEOSE ... I am not a Canon Mechanical or Electrical or Software Engineer ... but

I would speculate that the shutter (mechanical or electrical) would have little or nothing to do with "dual shutter". My uninformed guess is that:

• Camera in LiveView at lowest "Real" ISO
• Press shutter
• ML scans sensor values and "tucks them away somewhere"
• Adjust ISO by several stops
• Milliseconds later (no ghosting!), rescan sensor values (or with DIGIC-5+, perhaps microseconds)
• Rescan sensor and tuck values away
• Repeat until top "real" ISO reached, or allowed shutter duration elapsed
• Maybe "only" two scans at 1/100th second or faster?
• Maybe ten rescans at 1 to 30+ seconds?

Potentially: no ghosting, and DR as good or better than the human eye? And retain excellent resolution?

CR2's could get Really Big. Memory speed an issue? Available quick-write memory?  But isn't this what Moore's Law is for?

In effect, you would be doing the equivalent of 5k x 4k video, PLUS dual/triple/quadruple/etc ISO. That's too much memory movement for video (at least in late 2013), but for landscape and 1+ second exposures, seems theoretically feasible.

Or not?

[dmilligan]

You are proposing to basically read out the pixels more than once?

This has been asked and proposed before. What I said about gradients and rolling shutter effects would still apply b/c it is only possible to readout the sensor line by line, and this takes some time (or you wouldn't see rolling shutter in video, I imagine the time it takes to read the whole sensor is somewhere around a little faster the the maximum fps of video mode, i.e. 1/60 second ~17ms, not "microseconds" as you seem to think, and the fact that's DIGIC 5+ has little to do with it, we're talking about the sensor's electronics here, not the image processor). AFAIK the main limiting factor to faster FPS is the readout speed of the sensor (Why would Canon bother to make the sensor readout faster than the maximum FPS? This would only add cost). Also in video mode you're only reading part of the sensor (you're either line skipping or cropping out of the center), not the whole thing, so it's likely that it takes even longer to sample the whole sensor (like in photo mode).

I really doubt this is possible. Hardware electronics control the readout circuitry, they are very unlikely to able to be reprogrammed like this, and even if it is possible, it would be extremely difficult to reverse engineer it, without any specs on the hardware. The photosites themselves would also have to be capable of being readout without loosing their accumulated voltage, which AFAIK is only possible if specific transistor configurations are used (basically you need a buffer, which would require at least 1 extra transistor per pixel, potentially increasing the required transistor density of the sensor by 20-30%, and thus increasing cost by an even larger amount. Since it seems as though Canon is not making use of such a 'multiple-sampling' technique, I doubt they would have bothered adding all the extra electronics and associated cost to make it possible)

http://en.wikipedia.org/wiki/Active_pixel_sensor

[l_d_allan]

I really doubt this is possible.

Perhaps, but "Moore's Law" is on our side, at least in the longer term. My understanding is that Canon uses relatively ancient circuitry in some parts of their electronics. Sensor? That might be part of why Sony has gotten ahead?

Like 300 nm features instead of under 50 nm? I infer from that ... lots of room for improvement.

And the ML team have accomplished some things that were thought to be not possible.

Hardware electronics control the readout circuitry

Agree ... but I am the first to acknowledge I'm not a EE ... I suppose somewhat like the difference between DRAM and SRAM? Refresh after read?