More on Signal-to-noise

[Steven Thomas]

There's been some recent discussion on under exposing verses exposing as hot as possible with the EX1.

A couple days ago, I had the EX1 set up with the RED ONE indoors under low light.
First of all, let me say the RED ONE is quite the amazing camera. It's built like a tank and weighed in near 25lbs with the battery and hard drive attached to the rails.

We really want to get a feel on if we could get the EX1 to cut with the RED.
This was not a quality comparison test. It's obvious that the RED ONE would win this, especially with a 4K image with creative control of DOF. It's quite amazing.

Using the RED ONE at 2K, the DOF was close to the EX1. At 3K and especially 4K, it was very tight, as expected.

The RED ONE at 320 ASA looks very clean. The EX1 under the same light was two stops faster than the RED. I realize that the ASA numbers many not add to your calculations, I'm just saying I had to close the EX1 down at least by two stops to come close to matching the RED levels.
The EX1 gain was at -3dB.
The RED only offers a 720 preview, so it was tough comparing with the EX1 which was pumping out 1920x1080 out via SDI.

The EX1 did appear to show more noise (not much, just a bit), but it also was rendering a brighter image. While playing with the different picture profiles with the EX1, I noticed cine2 seemed to offer the lowest noise, with the expense of overall light level at the bottom of the curve. Actually, it was this profile that probably came close to the flat uncorrected look the RED ONE was displaying in these lighting conditions. Even the noise levels were very similar (almost unseen).

Here's Adam Wilt's breakdown on the EX1's Cine gammas:
"
• CINE1 – “deep cine”. Compression starts around 80%; midgray at 37%.
• CINE2 – “broadcast safe cine”. CINE1 rescaled with whites limited to 100%.
• CINE3 – “brighter cine”. Compression starts around 65%; midgray at 45%.
• CINE4 – “video-bright cine”. Compression starts around 65%; midgray at 0%.
"

I realize under controlled lighting, it is wise to over light to improve signal-to-noise, and reduce levels in post. Anymore thoughts on this? I realize when were dealing with lighting conditions out of our control, we need the best settings for both end of the spectrums, the brightest and darkest elements of the image.

I'm really trying to determine optimum settings for both ends of the spectrum.
I'm still learning what gamma curves work best under certain conditions.

Using CINE4 in bright light, I see how this gamma curve can increase light levels in the shadows to offer more detail, but it also increases noise. IMO, to noisy, but it may of been some other corrective setting. Gain was at -3dB.

It's going to take some time to determine which particular gamma curves work the best under certain conditions. I thought I'd have this in the bag by now, but it's easy to find a setiing and leave it alone. This is not always a good approach..LOL. OTOH, you have to becareful that you have not changed your overall look from shot to shot.

Also, after using the cine gammas, I'm wondering if I'll ever use the std gammas. I just do not see a need. Well, not yet.

[Piotr Wozniacki]

Steven,

Your observations are very consistent with mine. I'm leaving the STD curves for indoors, controlled light shooting. Outdoors, I "almost" know by now how to use CINE1-2 and CINE4; the CINE3 I still need to investigate. One thing I didn't observe, though; using CINE4 for backlit scenery in order to stretch the blacks, I didn't experience the increased noise level you're mentioning. To the contrary; moving up from the noise floor level actually helped in getting rid of noise in areas that would otherwise be darkish, chroma-oversaturated and hence noisy!

However, my experience with them is so far limited to only one colour matrix (hisat); I have yet to see how they perform with the Cine matrix.

[Steven Thomas]

Thanks Piotr,
Well we had the EX1 hooked up via SDI to a 1920x1080 24" LCD monitor indoors under low light.

Under those low light conditions, it was easy to see which cine gammas were the noisest. It was confirmd by both viewers that cine2 showed the least noise. Of course, this does not always make it the right choice. This was an observation based on noise itself.

[Mike Testin]

For me, Cine 4 stretches those blacks too much. I haven't been able to use the camera in enough situations to compare the different gammas but Cine 1 looked good. I will try Cine 2 next.

[Steven Thomas]

Originally Posted by Mike Testin

For me, Cine 4 stretches those blacks too much.

Agreed....

[Paul Curtis]

I've just been testing an EX1 against a custom HD solution with a larger (1") kodak sensor (much bigger pixels sizes - 7 micron).

The EX1 was actually a bit more sensitive, or at least produced a better visual image for a given light level/gain setting (candlelight at one point). Now i know the resulting image is heavily processed compared to the basic output of the other solution and i've not reached any conclusions yet but I was pretty impressed as i expected the opposite.

>Using the RED ONE at 2K, the DOF was close to the EX1.

Are you sure? that doesn't make sense, even using half the size of the reds sensor the DOF should still be much shallower than the EX1 (unless you're zoomed in, in which case it's more difficult to tell as the FOV is narrower).

The sensor in the EX is probably 6.4mm across, the red at half size might still be around 17mm. The solution i've been playing with is 14mm across and DOF characteristics are in a different world.

With regard to the cine gamma settings on the EX. Does the overall dynamic range captured change? I've always assumed that the gammas just allow for some control over how the full range is compressed into 8bit for capture. However, I've noticed that there's some highlight recovery going on to alleviate the highlights clipping in strange ways (when some of the channels clip early) although not always sucessfull perhaps these artificial highlight ranges are 'extending' the overall range a bit?

cheers
paul

[Bill Ravens]

Looking at a 12-step black and white test image, with the EX1 on a vectorscope shows quite a large reduction in dynamic range when using the CINE presets. This can be confirmed by looking at the published cine gamma curves for the EX1. I see a reduction of at least 75% or greater in dynamic range.

[Paul Curtis]

>reduction in dynamic range

when i say range i mean the blackest black to the brightest highlight, so the maximum values the sensor is able to capture. Somewhere there is a finite quantity of photons that the pixels can record ontop of a given noise floor.

Are you saying in your tests that these extremes change (so you're might be altering exposure/gain as well as gamma curves for your tests) or that the number or spacing of gradations between black and white change (which i consider bit depth).

I believe that some of the cines force the range into certain video limits. Im unsure whether this means that the signal is compressed into these limits or the highlights are clipped into these limits. After all, at the end of the day you're not getting any more than 8bit out of the mpeg2 no matter how you set things up.

What kind of visual effect are you describing?

many thanks
paul

[Bill Ravens]

Paul...

Take a look at page 9 of this brochure:
http://bssc.sel.sony.com/Broadcastan...mhd_family.pdf
These are the published gamma curves for the XDCAM HD series. They show the STD gamma curve compared to the CINE gamma curves. I wish an identical presentation was available for the EX1, however, on the EX1 brochure they only show Cine gamma curves.

So, with that caveat, the noteable thing about the XDCAM HD curves is the way the endpoint of the STD curve is about twice as high on the ordinate as the Cine curves. There is a rolloff of the gamma curves for the cine mode that doesn't use the entire available bandwidth of the sensor, at least that's the implication of the curves shown.

When looking at the B&W step wedge with the EX1, the vectorscope shows a stretch of the black steps(the spacing gradations get further apart below the knee), a shifting of the "middle gray" consistent with black stretch, as you would expect. However, the endpoints also change, the extent of change being dependent on which Cine gamma you select.

To my understanding of things, this suggests a total reduction in dynamic range to accomodate a knee that will produce "safer" results by providing more headroom during capture. However, since nothing's free, the compromise you make for more headroom is a significant reduction in dynamic range. In my context, dynamic range is the total number of f/stops you can capture between super black and super white.

[Piotr Wozniacki]

Originally Posted by Bill Ravens

There is a rolloff of the gamma curves for the cine mode that doesn't use the entire available bandwidth of the sensor, at least that's the implication of the curves shown...

My understanding is the same, Bill - while safe, the cine curves seem to cause a lot of bandwith waste.

[Bill Ravens]

In thinking, more, about this, it occurs to me that the total dynamic range that the EX1 can capture, from black to white, does not change when you change gamma. However, the IRE value, that is to say the digitized voltage range that is recorded, is compressed to avoid overloading the voltage circuit. Sorry, I wish I could relate the analog equivalent to digital terms, but, I'm not an EE. The net effect is to cram the same data into a smaller space, reducing the range of discreet values written to the media. Something has to give, and I'm not sure what that is. Fewer data bits means less info, so it's something like lower quality data bits.

[Piotr Wozniacki]

Originally Posted by Bill Ravens

In thinking, more, about this, it occurs to me that the total dynamic range that the EX1 can capture, from black to white, does not change when you change gamma. However, the IRE value, that is to say the digitized voltage range that is recorded, is compressed to avoid overloading the voltage circuit. Sorry, I wish I could relate the analog equivalent to digital terms, but, I'm not an EE. The net effect is to cram the same data into a smaller space, reducing the range of discreet values written to the media. Something has to give, and I'm not sure what that is. Fewer data bits means less info.

With the difference that it's not done to "avoid overloading the voltage circuit", but to ensure legal IRE values for broadcast (or film-out) purposes, where digital super-whites are not welcome.

[Bill Ravens]

Originally Posted by Piotr Wozniacki

With the difference that it's not done to "avoid overloading the voltage circuit", but to ensure legal IRE values for broadcast (or film-out) purposes, where digital super-whites are not welcome.

I would agree for Cine2, but, what about the other cine gammas? Looking at a WFM confirms that for cine2, everything is between ire0 and ire100. But for the other cine gammas, there are values recorded above and below broadcast legal IRE.

[Paul Curtis]

Bill,

Those graphs don't have any scales on the left and i wonder if they might be marketingspeak. Sony sensor datasheets also don't feature absolute scales on the Y. Judging by those published curves sony would be insane to cut the range by that much. I think they illustrate (badly) what gamma is doing instead.

I think you're right with your next post, the curves dictate how the range is squeezed into IRE signals. Or another way: there is an 8 bit range of values possible. Or 16 to 235 if you're being legal.

So it could be that some cine curves compress everything into 16-235 (for legal IRE levels) and the others use the full 0-255 (which is how you can get greater levels)

The issue is understanding how the curves are placing important points like mid grey and skintones.

The AD converters in the sensor will produce a possible range (maybe 12 bit). The electronics in the camera use the curves as lookups to sqeeze the 12bit into 8bit. It's also worth bearing in mind that the 12 bit off the sensor is nearly linear, so out of those 12 bits an awful lot of them will represent highlights.

As i said above i think it's possible that the camera is generating more highlight information when the channels clip but this really just adds a bit to the range.

cheers
paul

[Bill Ravens]

Originally Posted by Paul Curtis

Bill,

...the curves dictate how the range is squeezed into IRE signals. Or another way: there is an 8 bit range of values possible. Or 16 to 235 if you're being legal.

So it could be that some cine curves compress everything into 16-235 (for legal IRE levels) and the others use the full 0-255 (which is how you can get greater levels)

The issue is understanding how the curves are placing important points like mid grey and skintones.
paul

yup!

BTW, users are shooting skin tones at 60-65%, reporting that 70% still leaves blown hilights in specular reflections on a face.