Peter, from page 4 it's also "clear" that color sampling is 4:2:0 pre-compression, yet it's been widely stated that SDI is 4:2:2.

If SDI is in fact 4:2:2, why can't quantization be 10 bit as well? Or the corollary is that neither is chroma 4:2:2. Confusion reigns.

Tom,

But the heading on the top of page 9 say "Recorded H X V Samples, by Channel," so Sony is indicating that color subsampling varies by recording type. They make no such indication when it comes to color bit depth.

Also, Canon supplies 4:2:2 color out of its HDV cameras' HD-SDI and HDMI ports. I would imagine it's much easier to supply full horizontal resolution color samples than it is to requantify all three channels to a new bit depth. So true 4:2:2 color subsampling seems reasonable, while true 10-bit color depth less so, IMHO.

Well I suppose this "debate" continues because of constant confliction of facts and opinions, especially when the company making these cameras releases an official whitepaper stating such.

According to this official Sony Whitepaper, it seems what Peter Moretti has stated is correct: sans HDCAM SR, every other Sony camera quantizes the image data to 8-bit before even going into the color-subsampling stage (which needs to happen before going out the HD-SDI).

People I've talked to who have recorded the EX1 via HD-SDI to a 10-bit codec have concluded that it is outputting 8-bits. This official Sony paper only seems to back that up.

Wow, the PDF whitepaper from Sony is misleading.

They talk about Quantization and "Requantization" in the same step on page 7.

The last two paragraphs on page 7 should be in their own section entitled "Requantization" at the top of page 10. Requantization occurs after signal processing, but before compression/recording.

It even says as much in the last paragraph on Page 7:

"Prior to recording, signals are requantized with fewer bits per sample."

As far as the information throughout, the whitepaper keeps talking about what is RECORDED.

First off, just about every Sony tech rep, technician, engineer and marketing person has said for the last year and a half that the camera outputs full 10 bit data over HD-SDI. I think the manufacturer has been as clear as possible about this matter.

I've already said that I've monitored with 10 bit monitors and worked with EX1 footage captured over SDI. I'm not the only one.

Its 10 bit 4:2:2 data at full raster.

I've analyzed it in Shake and Color. Its 10 bit 4:2:2.

I can also say it looks rather different than the recorded output off SxS cards.

Just zoom in so you can see the pixels on fine lines, especially diagonal or curved ones with saturated reds or blues. The difference is immediately obvious even to people who have no training whatsoever.

10 bit is harder to see, because there are so few displays that support it.

The easiest way I've seen to show the difference is to pull a key on the sky. As you might for sky replacement. See the banding in 8bit... don't see it in 10 bit. Pretty simple really.

I wish I still had some of that footage, I could post and end that topic. I don't shoot EX1 anymore though, I am shooting RED presently. The cheapest way to get 4:4:4 recorded.

As a final point... if you can't tell the difference after all this time, then maybe its a difference that doesn't matter to your productions. Get back to the real work and stop fiddling unimportant digits.

;)

Tom,

But the heading on the top of page 9 say "Recorded H X V Samples, by Channel," so Sony is indicating that color subsampling varies by recording type. They make no such indication when it comes to color bit depth.

Incorrect, they say so on page 7 in the very last paragraph.

The same dumb paragraph, along with the paragraph immediately before, that should be in its own section on page 10.

Apply a little logic: Why would you do quantization in two steps one immediately after the other? You wouldn't. You do it in one step in that case.

They do it in two steps so that signal processing can happen at high bit depth, then they requantize for compression and recording.

Alexander,

The Sony document may be misleading, or it may indeed be accurate. But it clearly states that the color depth is lowered for the first time during the analogue to digital conversion stage. The resulting bit depth is either 14 or 12 bits depending upon the camera. It then states that a second bit depth lowering takes place. So for the EX1/3 to output true 10-bit color, this stage would have to lower the bit depth from either 14 or 12 bits to 10 bits.

But then another/alternate bit depth conversion will also have to take place to convert the signal to 8 bits prior to the recording, because the codec itself does not do bit depth conversions.

Having an intermediary, unmentioned, conversion to ten bits makes little sense to me. It is of course possible.

As for Sony reps, there are quotes from reps who say it's 4:2:0 out of the HD-SDI port (see the first page of the thread I linked). So which rep should be listened to?

You reference David from Cineform, but he seems to believe it's 8-bit not 10-bit but has done no testing.

You reference your own experience with the footage, which mostly supports that it's 4:2:2, not 4:2:0. I never believed it was 4:2:0. But you also say it works like F900 footage, which is probably HDCAM 3:1:1 8-bit, so what does that prove?

Your sky test is the most convincing, but it doesn't give a definitive answer. Does all this matter? For people equipped to do HD-SDI capture, yes it does. For those who aren't, no it does not.

Has anyone done a footage comparison of EX1/3 SDi versus SxS? Is there significant real world image improvement? No need for number slinging, what about the actual image side by side?

Alexander,

The Sony document may be misleading, or it may indeed be accurate. But it clearly states that the color depth is lowered for the first time during the analogue to digital conversion stage. The resulting bit depth is either 14 or 12 bits depending upon the camera. It then states that a second bit depth lowering takes place. So for the EX1/3 to output true 10-bit color, this stage would have to lower the bit depth from either 14 or 12 bits to 10 bits.

Actually, the phrase it uses is "quantization" which is actually only valid at sampling.

Later the paper talks about "requantization" which is poppycock. The data is rounded to the nearest 8 bit value and truncated. That isn't quantization in any sense.

Why do you think that HD-SDI ports are so expensive? It isn't merely some conspiracy to keep the good stuff to the "big boys."

Every SDI port has the "intelligence" to take the "raw" signal from the internal processing chain and convert it to conform to the appropriate SDI format and send it out.

Let me get a bit more technical: every version of the SDI standard includes a set of video signal generators. At this stage the generators are still handling the "raw" internal signal with the full bit depth of the first quantization at or near the sensor. Then the signal passes to SMPTE SDI encoder that takes a raw 74MHz signal- at whatever bit depth you send it- and converts that to the SDI signal specification. HD SDI encoders are made, essentially, of two SD SDI encoders. (Which by the way means that while SD SDI handles 10 bits per clock, HD SDI handles 20 bits per clock.)

Only then is it passed to the transceiver and sent down the wire.

But then another/alternate bit depth conversion will also have to take place to convert the signal to 8 bits prior to the recording, because the codec itself does not do bit depth conversions.

Having an intermediary, unmentioned, conversion to ten bits makes little sense to me. It is of course possible.

Well, if you follow me and Sony.... this isn't an intermediary conversion.

There is a conversion at the HD-SDI port from a 14 bit source to 10 bit.

There is a conversion before the compressor from a 14 bit source to 8 bit.

Both of these are on separate signal paths. How else could you get the HD-SDI signal out in real time and deliver near real time performance on the compression recorder?

Let me put that differently... if you instead send the signal to the compression encoder and then out the SDI port you'll get serious lag on the SDI monitoring.

The Sony paper we are discussing entirely ignores the live video outputs of the cameras. It is focused on, and discusses the recording path. (And does so poorly at that.)

You reference your own experience with the footage, which mostly supports that it's 4:2:2, not 4:2:0. I never believed it was 4:2:0. But you also say it works like F900 footage, which is probably HDCAM 3:1:1 8-bit, so what does that prove?

No, I did not say it worked like F900 footage.

I said I compared it to RED, Genesis and F900 footage. I am sorry I even said that, I had no idea how far it would get taken out of context.

For the record, RED and Genesis are 4:4:4 and 16 bit. They blow the EX out of the water.

The F900 is more complicated.

Yeah, its data kind of sucks these days, and yes its 3:1:1. Its much lower compression though, and that matters.

The camera is also much better than an EX1/EX3. It ends up being about on par for most uses, though the EX series is better for compositing and still scenes, the F900's CCD's still kick a$$ compared to the EX1/EX3- especially on fast motion shots or rapid flashing.

The point of value to THIS conversation is merely that the camera has been compared to other digital cinema systems, and understood relative to them.

I specifically reject your misinterpretation that somehow all those cameras are equivalent. In fact I count it as absurd.

Your sky test is the most convincing, but it doesn't give a definitive answer.

A definitive answer that you can test for yourself?

From the command line on a Mac with Shake 4 or 4.1 installed, enter the following:

shake filename.mov -info

That should get you something like this:


[athena:Psyche/Alex Dynamic Range Tests 1-23-2009/ProRes HQ] aibrahim% shake "DR TEST AGI Red Dynamic Range Tests.mov" -info
info: Node: SFileIn1
info: FileName: //Athena/Volumes/Psyche/Alex Dynamic Range Tests 1-23-2009/ProRes HQ/DR TEST AGI Red Dynamic Range Tests.mov
info: Type: RGBA
info: Size: 2048x1024
info: Depth: 16 bits
info: Z-Buffer: none
info: Format: QuickTime movie
info: Duration: 1-7917


That represents a file with 16 bit per channel available. Its actually a 14 bit log file, but that's another matter.

If you have an 8-bit file you'll see this:

info: Depth: 8 bits

Does all this matter? For people equipped to do HD-SDI capture, yes it does. For those who aren't, no it does not.

Apparently not.

I think it matters to people who are equipped for HD-SDI capture AND monitoring 10 bit video.

I think it matters to the colorist and compositor if you can capture HD-SDI.

If you are either of those people you can already see the difference in your day to day work.

If you aren't one of those folks- then no matter what you do you are actually getting 8 bit results. That isn't intrinsically bad.

If nothing else you do get to see 4:2:2 results over HD-SDI as opposed to 4:2:0 on the card.

Alexander,

Your test, I imagine, cannot tell the difference between ten bits of significant data and eight bits of significant data padded with two zeros.

The bottom line is that neither of us can prove it either way. You bring up some good points, but clearly this ? is far from answered.

But it seems that not all HD-SDI encoders work as you stated, otherwise Canon's XH's and XL-H's would provide ten bit color out of their HD-SDI ports. They don't, they provide eight bit color in a ten bit wrapper. Perhaps the Canon chips quantify the color to eight bits during the first go around, so that's all that can be had off the sensor. IDK.

I DO think it's very fair to say the EX's HD-SDI data is 4:2:2, not 4:2:0. But truly ten bit? The jury is still out.

I don't think it's that hard to test if it's 10bit or 8bit. Underexpose a wedge by 1 stop and see how much of the blacks you can recover.
As to the white paper, it does seem to be referring only to recording and from memory the HD SDI signal was stated as being derived from the component outputs. The again that could raise some questions too.

Has anyone done a footage comparison of EX1/3 SDi versus SxS? Is there significant real world image improvement? No need for number slinging, what about the actual image side by side?

Dear Brian,

Paul Cronin performed that test.

He recorded using an EX1 or EX3 (I can not remember) to the SxS card internally and to a Flash XDR via HD-SDI.

He found the same frame in both recordings displayed them on two identical monitors.

The difference was dramatic. 4:2:2 should always beat 4:2:0.

His wife came into the room and asked: "Why did you color correct one image and not the other?"

He had not corrected either image, the HD-SDI output, 4:2:2, as recorded by the Flash XDR was just much better visually.

Disclaimer: I work for Convergent Design.

Very nice Dan, so it really IS 4:2:2 - I think we've established it already in other threads.

But the questions remains (in never was answered by CD): is it true 10 bit, or just 8 but wrapped in 10 bit for HD-SDI compatibility?

Well the Flash XDR contains an 8-bit codec, so they won't be able to tell from XDR footage ...

Why don't we just test this sometime. Does anyone have an EX1/3 that can do this? If anyone is in NYC, I have a Canon XL H1S and the AJA IO HD (which can record to 10-bit ProRess 422HQ), so I have the infrastructure to capture in the proper codec.

Depending on schedules, I may be free to do this at some point in the future.

Very nice Dan, so it really IS 4:2:2 - I think we've established it already in other threads.

But the questions remains (in never was answered by CD): is it true 10 bit, or just 8 but wrapped in 10 bit for HD-SDI compatibility?

Dear Piotr,

I personally do not know if the EX1/3 is 8 bits in a 10-bit HD-SDI stream. I have not run any tests.

All MPEG2 is 8-bits.

But, full raster MPEG2, at a high-bit rate Long-GOP can produce amazing images.

We have had industry experts project our images on a 25' screen with a high end Christie projector. It works well.

Dear Dan,

I have no doubts whatsoever that FlashXDR (and hopefully NanoFlash, which I have on order) are capable of producing stunning images out of the full raster EX output. What I'd like to know is whether the future upgrade to 10 bit uncompressed is worthwhile with this particular camera.

Dear Piotr,

That is a hard question. We will have to run some tests.

We believe that the EX1/EX3 produces 10-bit HD-SDI output.

We do not keep an EX1/EX3 permemently in our lab.

I believe that we have not actually tested this in our lab, I just checked with our engineers.

We do have the equipment to test this and we will.


Obviously we have a stake in the 8-bit versus 10-bit discussion as the Flash XDR can currently record at only 8-bits. The future Full Uncompressed support will support 8 bit or 10 bits.

Our experience has been and very experienced people with very expensive cameras that can record full uncompressed at 10-bit has been that it just does not make much of a difference.

Richard Welnowski has spend weeks running these tests with his Thompson Viper. He finds it very difficult to see a difference, even at 300% magnification of the images.

So Piotr, we will test this as soon as we can get a EX1/EX3 in our lab.

I hope this helps.

Dan,

Thank you so much for being willing to look into this ?. BTW, I would think a good test would be to open footage in a ten bit project, and select swatches using the eyedroper tool.

If the ten-bit R, G and B values never increment/decrement by one, two or three, then it's eight footage. If they do change in amounts other than +/-4, then it's ten bit footage. No?

Dear Peter,

Yes, that would be an interesting test.

Another would be to look at the values with the eye dropper.

If one is in a 10-bit timeline (or canvas), and finds an odd value, then the data is 10 bit.

By odd, I mean ending in 1, 3, 5, 7, 9.

Right b/c if 8-bit values are just scaled to ten bits, they will equal their original 8-bit values multiplied by 2^2, i.e. 4. And any whole number multiplied by 4 is even.

But if, on the other hand, they are values with ten significant bits, then odd numbers will be present as well. Very cool.


Michael Galvan,

I wonder if you'd be willing to do either of these two tests (Dan's version or mine) with your XL-H1 HD-SDI footage? It would be a nice confirmation of the methodology. ;)


P.S. So in the case of the Canon HDV cameras, which do supply "only" 8-bit color out of their HD-SDI ports, I guess it's a misnomer to say the "eight bit data is padded with two zeros" to fit in a ten bit stream. What really happens is the eight bit values are scaled by a factor of four. Is that correct?

... Michael, a test at your convenience would be nice :).


P.P.S. Actually, I guess a waveform monitor set to a 10-bit scale would show empty horizontal bands if the data were eight-bit scaled to ten-bit. ... Okay, enough of thinking about this for one night, LOL.

P.P.P.S. But, an eight-bit to ten-bit converter could dither the values, in which case, these tests would not work.

I checked with our engineers.

We checked an EX1 last year and found that it is indeed 10-Bit HD-SDI.

I hope this helps.

If they are correct, which I have little reason to doubt, then that is great news, and puts the EX's on par with some very powerful cameras.

To all those who have said it's ten-bit, I tip my hat. You were right.

Thanks very much Dan.

Dear Peter,

We have used the Sony EX1 with the Flash XDR and the images are just stunning.

We produced a small "Flash XDR Promo" with the EX1 + Flash XDR and a Canon XL H1 + Flash XDR. All footage was record 100 Mbps Long-GOP. All product shows are with the Canon XL H1, all others are with the Sony EX1.

No color correction was done to the images.

We shot with tungsten lighting and set both cameras to 3800K.

The Flash XDR performed flawlessly for this multiple day shoot.


Our seasoned editor, who is also an educator in this field, was just amazed at the footage.

He had been working in HD for a while and was truely impressed.

Our "Flash XDR Promo" is on YouTube. We would have liked to put it on Vimeo, but we did not want to violate the rules as it is rightfully considered commercial material.

So far I've yet to see any footage of EX footage through SDI. I wish someone would do a shootout.

Dear Brian,

I attempted to upload some EX1 / Flash XDR footage yesterday.

My sample file was 2.44 GB. After hours of work, I found out that our high-speed hosting site limits the filesize to 2 GB.

I have spoken with our editor, who has the original footage handy, and he will upload some original clips as soon as he can. His equipment is at another location, so this can not be done until next week.

Our office is scheduled to upload some on Monday.

Dear Brian,

Your DVD with sample files was sent on Thursday.