HDMI vs. SDI Analog Tape Video Capture

[dellsam34]

I was asked numerous times to compare analog video capture such as VHS between two methods, SDI and HDMI, Although I have the JVC HM-DH5U a NTSC VHS/S-VHS/D-VHS machine that has both S-Video and HDMI outputs for quite some time now but I've never had a proper HDMI capture device or card until now, I recently acquired the Black Magic HyperDeck Shuttle 2 a SDI and HDMI recorder and I thought this is the perfect time to do such a test, So here it is:

How the test was done?
I played back this VHS tape "The adventure of the young Indiana Jones" and captured the exact same segment with both methods, The HD-DH5U outputs the signal on both S-Video and HDMI at the same time which makes the process quick and easy, The tape by the way has Macrovision so this gave me an opportunity to test the HyperDeck if it ignores it, and it seems that it does.

The workflows are as follows:
HM-DH5U --> S-Video to Y-C BNC cable --> S&W TBS800 --> SDI cable --> HyperDeck Shuttle 2 --> Internal SSD recording lossless SD 4:2:2
HM-DH5U --> HDMI cable --> HyperDeck Shuttle 2 --> Internal SSD recording lossless SD 4:2:2

Notes:
- VCR settings for both workflows are the same, Edit, Video stabilizer off, line TBC cannot be turned off in this VCR.
- TBS800's proc amp setting are all at neutral or mid point, Audio gain is slightly higher in the TBS800 because I forgot to reset it from a previous capture job, but it shouldn't matter, The TBC is set to "free run" to lock the audio to the video.
- There is no setting for the HyperDeck, it just saves whatever comes in to the SSD.

Took each of the two files into vdub2 to output HuffYUV, Make accurate trimming at the exact same frames and remove the 14 blank audio channels.

The files have been removed due to a mistake, they are in the next posts.

[themaster1]

So your conclusion/observation are ?

[lollo]

Thanks for the excellent comparison, much appreciated! A lot of incentives to work on it!

I am in rush, so I just can do a quick check on the 2 segments.

The SDI capture has clipped whites:



Initially I had a very small preference to the SDI captures because the overall look, but once the levels are equalized they looks very similar to me, with a small preference for the HDMI capture in term of details (just speculating).

https://imgsli.com/MzEyNzAy

https://imgsli.com/MzEyNzAz

Basically, we are comparing here the quality of the A/D conversion of the S&W TBS800 versus the HM-DH5U (and both look nice) together with any degradation in the signal path because the standard (no evidence of that, obviously).

I will perform some more test on my side in the next days, thanks again for the material you provided!

v1=AviSource("VCR-TBS800-SDI.avi").convertToRGB().crop(0,0,-1,0).addborders(1,0,0,0).crop(8,10,-10,-10).convertToYUY2()\
#.convertToYV16().histogram("levels")
v2=AviSource("VCR-HDMI.avi").crop(0,0,0,-4).addborders(0,4,0,0).crop(8,10,-10,-10)\
.colorYUV(off_y=-27, gain_y=20)\
#.convertToYV16().histogram("levels")

#return(v1)
#return(v2)

#interleave(\
#stackhorizontal(\
#stackvertical(\
interleave(\
subtitle(v1,"VCR-TBS800-SDI.avi",size=20,align=2),\
subtitle(v2,"VCR-HDMI.avi",size=20,align=2)\
)

[dellsam34]

So your conclusion/observation are ?

It's up to you or anyone how to interpret the differences, I just made the files.

[dellsam34]

I forgot to mention that the TBS800 has some sort of digital DNR built in that cannot be turned off, It does however have an option for analog NR and chroma NR which are turned off, It also has a check box for legalizer, It doesn't appear to make a difference when checked.

[Sharc]

Both extend the brights beyond luma 235, the HDMI just a bit less than the SDI. Neither is severely clipped though, so that's no problem as long as one stays in YUV.
However, upon importing these YUV captures into certain NLEs (including Vdub) or applying certain filters the YUV may be converted to RGB which get clipped for the Rec.601 matrix, means loosing fine details irreversibly in those brights.
The screenshot shows the "hot" pixels (in good approximation) for the HDMI in Cyan. (Much the same for the SDI.)
Apart from this the 2 methods are pretty much equivalent IMO.

[dellsam34]

Initially I had a very small preference to the SDI captures because the overall look, but once the levels are equalized they looks very similar to me, with a small preference for the HDMI capture in term of details (just speculating).

I agree, But that's possible the S-Video output is to blame due to the fact that converting back from digital to analog after the TBC incur some loss but HDMI is not because it's fed from the digital out of the digital processing unit, I don't know how signal handling is done inside the HM-DH5U to be honest, But one thing for sure, for this machine the HDMI is much better than S-Video, That's probably how I'm going to use this VCR from now on.

Edit: It turns out I had wrong TBS800 setting, samples re-uploaded.

[Sharc]

I agree, But that's possible the S-Video output is to blame due to the fact that converting back from digital to analog after the TBC incur some loss but HDMI is not because it's fed from the digital out of the digital processing unit, I don't know how signal handling is done inside the HM-DH5U to be honest,

I don't think that the extra D/A conversion is to blame for lower sharpness. You did mention that the TBS800 has some DNR which cannot be disabled. I don't know if and how it is inserted into the workflow, possibly affecting the signal. Also, the TBS800 default settings (neutral or midpoint) may compromize the sharpness slightly. Just some thoughts, no proof.
So take the best and leave the rest.....

[dellsam34]

I did compare the TBS800 against other devices before, it does not seem to took duller than the rest, So I highly doubt its DNR is the cause.

The way proc amp works is if you move the slider it will activate the procamp and reset button becomes active, you click on reset it becomes grayed out and all sliders move to their default position which means the procamp is off.

[Sharc]

Taking a second look it appears that the HDMI has elevated darks resulting in lower contrast ("milky" look in comparison with the SDI). This can be easily corrected in post processing though.

Edit: Interleaved comparison attached.
Edit2: Original attachment removed, needs to be IVTCed. Here we go:

[dellsam34]

I've introduced some new workflows for comparison and another VCR, I will include the first two tests here for easy access, order is rearranged:

1- HM-DH5U --> HDMI cable --> HyperDeck Shuttle 2 --> Internal SSD recording lossless SD 4:2:2
2- HM-DH5U --> S-Video to Y-C BNC cable --> BE75 --> SDI cable --> HyperDeck Shuttle 2 --> Internal SSD recording lossless SD 4:2:2
3- HM-DH5U --> S-Video cable --> Miranda ASD-272P --> SDI cable --> HyperDeck Shuttle 2 --> Internal SSD recording lossless SD 4:2:2
4- HM-DH5U --> S-Video cable --> Pinnacle 500-USB --> USB cable --> CaptureFlex recording lossless SD 4:2:2
5- HM-DH5U --> S-Video to Y-C BNC cable --> S&W TBS800 --> SDI cable --> HyperDeck Shuttle 2 --> Internal SSD recording lossless SD 4:2:2

Notes:
Miranda is a video only S-Video to SDI stick, no audio, it was sent to me by one of the members long time ago for this purpose.

Edit: Samples redone and reuploaded due to a mistake, check next posts.

[dellsam34]

I made a huge mistake, I left the chroma slider a little bit above the center from a previous capture for the TBS800, So I removed the files from the original post and posted a corrected recapture in the above post, While at it I've added two more captures for the HM-DH5U-TBS800 combo, one with legalizer checked and one with all the DNR options turned on (see the menu attached), It turns out the legalizer does change something, I noticed it in the sky section of the frame.

[pandy]

One important remark - SDI is unidirectional interface - there is no bidirectional communication there - HDMI is also unidirectional (from Source toward Sink a.k.a receiver) but connection between Sink and Source is established trough EDID communication line and mutual negotiation - based on EDID content stored in HDMI Sink a HDMI Source select best link parameters (what is best is up to Source capabilities and software design choices).

In other words SDI need to be manually configured but usually there is no need to do this - perhaps except selecting some few basic options where HDMI may require more tweaking (like color space selection, quantization range etc) - also there is possibility to "cheat" HDMI Source by modifying EDID communication on the fly (or create EDID different than reported by Sink).

So HDMI may offer theoretically higher flexibility but also may be more sensitive to configuration errors.

[dellsam34]

I'm not sure I'm following you there, Analog to SDI device are FPGA based, they have their own control software as seen in the above screenshot, the whole reason I use them is because they don't rely on computer CPU for converting analog to digital, why would you want to tie them to the computer?

The HDMI from VCR can be used with a HDMI capture card for procamp tweaking but I'm not a big fan of capturing from HDMI.

[pandy]

My point was that HDMI have more options to "broke video" and SDI has less features to "broke video".

[Sharc]

Interesting. I had a quick look at your "legalizer" clip only without really analyzing the video. The "legalizer" appears to have done what it is supposed to do, namely legalizing the captured YUV such that upon conversion to RGB the transformation results in valid and unclipped RGB. Therefore, the sky turns into bluish instad of grey-white, and you see more details and structures in the bright areas like the former mainly flat white parasol.
When I find some time I want to take a closer look to see whether this speculation is right .....
The analog NR has usually 2 effects: On the positive it reduces the noise. On the negative it softens the picture. Typically a compromise.
Also, I am not sure whether the analog pedestal should be enabled. It may just raise the dark level (?). You may just try.

Edit:
Yup, look at this: sigificantly less hot (illegal = marked cyan) pixels than before (post#6), and no luma excursion >235. The legalizer apparently did its job. I would assume that the user manual should describe its function and purpose.

[dellsam34]

Good point about the analog Pedestal, To be honest I have no idea what's its purpose exactly, but when unchecking it, it turns the frame very bright and the blacks grey, Having it on does darken the picture but it doesn't crash the detail in blacks, that's why I like having it on, plus I can always brighten the image using the procamp controls on the left side if I needed to, I could be wrong.

The legalizer should be on then from now on.

The manual doesn't say much, here it is:

[Sharc]

My guessing is that the analog Pedestal refers to the 7.5 IRE setup for North America's NTSC "analog reference black". This article for Final Cut Pro explains the effect pretty detailed.
http://www.kenstone.net/fcp_homepage/video_levels_nattress.html
or see here:
https://en.wikipedia.org/wiki/IRE_(unit)

For the 8bit realm, if at the end the luma waveform of a captured video fluctuates between 16 (black, dark) and 235 (white, bright) one is usually pretty much on the safe side even for avoiding "hot" pixels causing clipping in RGB.
The "Legalization" setting introduces some processing to warrant this even when the original luma exceeds the 16.....235 range and/or would produce out of gamut RGB, see chapter 5.2.3.5 on page 29 in the manual. Perfectly useful.

[aramkolt]

I've been messing with my CVR250 that has a similar feature set to the TBS800 - and was made by the same company around the same time.

What I found was that automatic gain control setting (not sure if the TBS800 has that setting - it would be under input, then decoder) doesn't scale the full range of the waveform down if it detects illegal luma levels, it only scales the upper portion of the waveform down that right next to the 100IRE level. I wouldn't exactly call it "clipping" because clipping to me would suggest that AGC would just assign all values higher than 100IRE to 100IRE and not affect the non-100+IRE parts of the waveform which AGC clearly does - I didn't take a video of this, but maybe it will make sense to Sharc and Lollo.

I did take some videos of the RGB legalizer on versus off and a separate example of just lowering the luma level to what your playback device peaks out at

My test chain here is Hi8 playback via S-Video -> CVR250 via SDI output -> iKan LCD monitor with false color enabled (Red onscreen equals values higher than 100IRE) and passthrough to VideoTek TVM waveform monitor with filter set to make the SDI (which is really digital component) look like the equivalent composite waveform.

What I found was that without adjusting the incoming luma (leaving it at zero), the playback device (or the tape) allows values up to 110 IRE to be output. Part of this could be that the Sony camera was probably expecting you to use a 6ft or more cable which may have dropped the signal a bit lower, and here I'm using about 2ft of 75ohm BNC cables off of an Extron S-Video splitter similar to what delisam34 has (though not sure how long his cables are).

At least with this batch of tapes, I can get IRE values to never exceed 100 by setting luma to -0.8dB within the CVR250's proc amp. This probably also pushes black levels a little lower than they should be as a side effect though. There is a separate setting in the proc amp for black level where you could manually compensate for the black level in mV in increments of 5mV at a time. 1IRE is 7.14mV, so it is relatively fine control, but I haven't messed with it yet.

I've attached two photos showing what the waveform looks like with the luma at -0.8dB versus keeping it at zero and you can see that the entire waveform is linearly scaled if you go that route. There is what appears to be prior peaking/clipping that I think either occurred at the time of the recording onto the tape, or it could be the playback camcorder is doing its own sort of internal clipping above the original 110IRE.

Alternatively, I've attached a couple of videos showing what the RGB legalizer does - you can kind of hear me clicking it on and off with a physical button and you can see some "Red" areas show up on the false color monitor on the video that shows it. One is on a static image, and the other is with live video. You can see that the RGB legalizer starts to kind of kick in and will start to sort of logarithmically cut down IRE levels once they get past 80, but will still allow peak whites to get very close to 100IRE if the area is bright enough relative to the rest of the screen. It leaves the middle section in the 7.5-80IRE section unchanged. It also does bring black levels up as well so that it is not possible to crush blacks.

To further complicate things, the AJA KiPro that I've been capturing the SDI signal with into ProRes 422 for testing actually will record the levels as they are, so it keeps values that are below black and also those above 100IRE, so technically they aren't being clipped at the time of capture, but they might clip later if you were to straight encode them to something else without adjusting the levels in something like a nonlinear editor first.

So my question is, what's the lesser of the different evils here for capturing tapes that might have different video levels to begin with? If I determine that the camcorder here will output luma levels up to 110IRE, is there any problem just leaving that -0.8dB luma in place, or perhaps compromise to -0.5dB and keep the RGB legalizer on as well so that the legalizer doesn't have to kick in as often? Or just use the RGB legalizer and be ok with either extreme of the video waveform's dynamic range being reduced a bit automatically regardless of the input source characteristics? With VHS in particular, there can be quite a variance of peak IRE from tape to tape. I've seen commercially produced VHS tapes produce outputs of 130IRE or more, and I'm not sure if AGC is just supposed to kick in on the capture card. Is traditional capture card AGC more like the RGB legalizer, or is it more like a varying luma reduction?

[aramkolt]

Here are the two videos showing RGB legalizer turning on and off.

[dellsam34]

My guessing is that the analog Pedestal refers to the 7.5 IRE setup for North America's NTSC "analog reference black". This article for Final Cut Pro explains the effect pretty detailed.

Oh, so it's for the 7.5IRE / 0IRE? From what I've read before about 7.5IRE and 0IRE, this should be used for professional decks and camcorders because they have a manual IRE setting, For consumer stuff like VCRs and camcorders everything should be compensated for for a flat 0IRE, At least that's how I interpreted it for years. I also did a lot of experimenting between PAL and NTSC sources and haven't seen any difference in brightness at least not a big of a difference that checking or unchecking the pedestal would bring them close to each other, I'll have to dig deeper in this subject to see if it applies to consumer gear, But thanks for the info and links.

[aramkolt]

With my legalizer on/off demo there, I think you'll agree that using that implementation of the legalizer changes the original waveform in such a way that you can't really call it "lossless" anymore - at least on the CVR250. I suppose automatic gain control that different capture cards have may have been doing something similar all along though too. That's kind of why I'm wondering if it is better to reduce the luminance setting to make it so that all values fall within the legal range rather than to use the legalizer?

Looking at the description for the TBS800 legalizer makes me think it actually works differently than the CVR250's. Kind of seems like the legalizer on the TBS800 is how I assumed an automatic gain control would work - basically looking at the brightest value, setting that to 100IRE, and then scaling down the rest by the same fraction (sounds like a linear multiplication from the description). Downside here is that it does clip blacks first and then probably loses some more black after downscaling occurs on the bottom end. The CVR250 doesn't clip blacks or scale the mid range of the wave at all in my demo - it seems to only scale IRE values above 85 and below 15 and leaves the mid range untouched.

This is from the TBS800 Manual:
The Legalizer ensures that output video is kept within the RGB gamut limit. When the Legalizer control is selected:
1. Source video is matrixed from YPbPr to RGB.
2. Any RGB video transitions below black are clipped to black.
3. Black-clipped RGB is matrixed back to YPbPr.
4. Any RGB transitions above peak white are measured relative to peak white to create a scaling factor.
5. The YPbPr video (from stage 3) is multiplied by the scaling factor.
This process ensures minimal disturbance to the perceived black level of the pre-corrected video while maintaining approximately the correct hue in parts of the image above peak white.




CVR600 (which is the same form factor as the TBS800) doesn't have a legalizer. It does describe it's automatic gain control and automatic color gain control functions. Makes AGC sound like you'd use it for weak signals, not for hot ones:

Decoder AGC:
The automatic gain control (AGC) function examines the input sync level and adjusts the luminance and chrominance gain to compensate for overall signal level attenuation. Use this control when the input video signal levels are low and automatic correction is desired.

Decoder ACC:
The automatic color gain control (ACC) function examines the input burst amplitude and adjusts chrominance gain to compensate for chroma level attenuation. Use this control when chroma levels are low compared to luminance. This may occur when using YC installations or composite from RF links.



Lastly, the CVR250:

RGB Legaliser:
When selected this will enable the RGB Gamut legalizer. This will prevent the unit from producing RGB signals greater than 101%. (funny how it doesn't mention anything about bringing up sub-black levels, but it does. Also doesn't lie about not scaling the entire waveform though)

Auto Gain Control
When this item is enabled the luminance gain will vary relative to the input peak white amplitude.
This will maintain the output signal at a normalized level even though the input signal level may be above or below standard level. The control will be effective over an input level range of +3 dB to -6 dB.

Auto Color Control
When this item is enabled the chrominance gain will vary relative to the input burst amplitude. This will maintain correct color saturation regardless of changes in subcarrier amplitude.

[dellsam34]

These are pro settings, sounds confusing to most of us but a professional will definitely know when and how to use them, For me I will just keep experimenting and see which function works best.

Overall all devices perform on par with each other, with the exception of the HDMI version that has more brightness, more details and more noise as well.

[Sharc]

Lastly, remember the inner RGB block within the outer YUV (aka Y'CbCr) cube. Only a fraction of the possible Y'CbCr triplets lie within the inner block and represent valid colors. How to handle the mapping of YUV->RGB properly to prevent out-of-gamut RGB is the art of Legalizing. This process is not lossless but legalizing should ensure a decent looking picture preventing unnecessary losses due to out-of-gamut (hence usually clipped) RGB. How it is done is manufacturer specific. Keeping and scaling the luma in the 16....235 range is a precondition but it is not sufficient to warrant valid (legal) colors. That's at least how I understand it from past discussions.

[Sharc]

Overall all devices perform on par with each other, with the exception of the HDMI version that has more brightness, more details and more noise as well.

Post#11: The HR-S7600AM-TBS800-SDI.avi looses all details in the brights. Clipped at luma 255. That's by far the worst variant IMO.
FWIW interleaved comparison attached.

Did you always have the analog NR enabled for the TBS800 clips? That may explain the softer picture.

[dellsam34]

No, just for this test.

I still feel like something wrong with the setting in the TBS800, The two JVC decks are pretty close in brightness, I may have to do two final tests for both VCRs with the TBS800 and use proper setting, but I have to go over the user manual to make sure I understand everything.

So far the HDMI capture is the most detailed, but has some noise though.

[Sharc]

Agree, the HDMI looks fine as it goes only little into Y>235 and is not clipped. I would try to tweak it for lowering the darks and brights to have it better accommodated within 16...235. You then obtain even more details out of the brights (sky color, parasol details)
Perhaps something like

AVISource("HM-DH5U-HDMI.avi")
assumeTFF()
TFM().TDecimate()
coloryuv(0,-20,0,10) #levels adjustment
#Temporaldegrain2() #enable for denoising
histogram()

[aramkolt]

I'll have to do some more tests on how these different AGCs work since that's really something that we should know about common capture cards by now, but often the discussion gets reduced to saying that a certain card "has AGC issues" without really much guidance on to test for those issues to see if your cards have the issue.

I assume the best way to do it would be with a ramp pattern plus a proc amp to purposely drive the luma too high (say to 120IRE) and see what happens to the output as it gets near and above 100IRE. What I thought AGC was supposed to do was assign the brightest value to 100IRE then linearly scale the rest of the waveform down by the same factor, similar to what you could do if you manually moved a luma adjustment on a proc amp for each (bright) scene

I figured a legalizer might essentially clip the values above 100IRE and set them to 100IRE, which is still clipping, but the clipped values are then also made legal, but that is apparently not the case and can vary not only from manufacturer to manufacturer, but from model to model in the Snell and Wilcox case anyway. On that CVR250, it says that it "prevents outputting of RGB values of greater than 101%", which sounds like clipping, but that's not what it actually does per the crude testing above.

[dellsam34]

I don't thing the legalizer clips the higher values, It just lowers them, Instead of having a single linear curve, you would have two curves, one linear to a certain level and then it changes angle to another linear curve or even non linear.

[Sharc]

"Legalizing" preserving the original intended visual perception (hue, luminance) is generally more complex than AGC or some IRE scaling. When "Legalizing" is offered by some tools or devices it is manufacturer specific how it is implemented IMO (approximate .....with which compromises ....).

Here a BBC report from 1987 dealing with the subject (is/was available on the web, I don't have the link anymore):