Determining the capture area of a capture card

[Arachnotron]

To split off the discussion about determining the capture area of capture cards from the discussion about the Capture Guide, I summarize the results so far here:

Erratic, Ati Radeon 8500 DV, Theater chip
PAL, 705 DVD pixels, 52.2 µs
NTSC, 715 DVD pixels, 53.0 µs

North2Polaris, ATI Radeon AIW, Theater chip
NTSC, 715, 53.0 µs

North2Polaris, ATI Radeon AIW 9000 pro, Theater 200 chip
NTSC, 705, 52.2 µs

North2Polaris, Canopus ADVC
NTSC, 720, 53.3333, 13.5 MHz sampling rate

The DVD images on which this data is based can be found here

A new guide how to do this yourself involving a much simpler calculating method and new DVD and VCD images will shortly appear
here

[Arachnotron]

edit dec 6th 2003
I found some inaccuracies in these numbers. Some are 1 or 2 pixels off.
for example:
25, 9-712, 704, 10.2, 52.15, BT878, BTWincap v5.3.6.1
this should be 702 ITU pixels, 52 µs.

I will re-check them and be back later with corrections if needed.





For convenience I repeat my own earlier measurements

PAL:

25, 9-712, 702, 10.2, 52.0, BT878, BTWincap v5.3.6.1
23, 15-710, 696, 10.7, 51.56, BT878, Hauppauge WDM v3.35 b 21125
23, 15-710, 696, 10.7, 51.56, BT878, Iulabs universal WDM v3.1.28.36
23, 13-708, 696, 10.5, 51.56,CX23881, Hauppauge WDM v2.75.21070
23, 3-722, 720, 9.8, 53.33, SAA7113, Terratec Cameo Grabster 200 USB v3.05
22, 10-713, 704, 10.3, 52.15, SAA7134, Terratec Cinergy TV 400 WDM v1.2.0.5

NTSC:

23, 2-713, 713, 9.3, 52.80, BT878, BTWincap v5.3.6.1
23, 17-704, 688, 10.4, 50.96, BT878, Hauppauge WDM v3.35 b 21125
23, 17-704, 688, 10.4, 50.96, BT878, Iulabs universal WDM v3.1.28.36
24, 16-703, 688, 10.3, 50.96, CX23881, Hauppauge WDM v2.75.21070
22, 0-719, 720, 9.1, 53.33, SAA7113, Terratec Cameo Grabster 200 USB v3.05
22, 8-711, 704, 9.7, 52.15, SAA7134, Terratec Cinergy TV 400 WDM v1.2.0.5

PAL-60:

290?, 16-714, 699, 10.3, 51.78, BT878, BTWincap v5.3.6.1
22, 8-711, 704, 9.71, 52.15, SAA7134, Terratec Cinergy TV 400 WDM v1.2.0.5

[Wilbert]

I did my measurements:

http://www.geocities.com/wilbertdijkhof/SAA7108.rar

PAL: 1-720, 720 (720x576 and 704x576 are scalings, hence not ITU compliant)
NTSC: 1-720, 720 (all three are scalings)

1) I captured with vdub using the nvidia WDM 30.82 drivers.

2) I couldn't capture at 768x576.

3) Why aren't the vertical lines 1-4 captured?

[Arachnotron]

1:
I think this card might be 'ITU compliant with scaler' at 720x480/576, but the sampling frame is slighly shifted compared to the output of your DVD player. Which one of the two is or isn't compliant is difficult to say.

Note that the gray pattern in the 'unscaled' 720x480 cap is the same for all vertical lines, but changes if you do 704x480, which is scaled from the 13.5 MHz base. In fact, I think that it should be possible to calculate back the base sampling rate from the interference patterns

Did you use a composite lead by the way?

This is the first DVD player I saw which actually outputs all pixels without cropping.

2: this is probably a driver issue. I remember reading a driver .inf file in which the revision comments mentioned limiting the allowed horizontal resolution to a maximum of 720.(can't find it back though)

3. They are captured, but your DVD player has cropped them, so they are black. It seems all DVD players crop lines, mostly 4 but which ones vary.
Your DVD player only does this in PAL mode.

The end result is that there is a 4 line variation in which lines a DVD player puts the center of the image.

edit

Notice by the way your card starts capturing below line 23, since the WSS signal is not visible in your PAL cap. A WSS line looks like this in the top of a capture.

[Wilbert]

I think this card might be 'ITU compliant with scaler' at 720x480/576, but the sampling frame is slighly shifted compared to the output of your DVD player. Which one of the two is or isn't compliant is difficult to say.

Hmm. I always thought that 'ITU compliant' meant "the 4:3 picture is given by the 702x576 part of the image (thus active area is 52 µs)". How can the one at 720x576 be ITU compliant?

Note that the gray pattern in the 'unscaled' 720x480 cap is the same for all vertical lines, but changes if you do 704x480, which is scaled from the 13.5 MHz base. In fact, I think that it should be possible to calculate back the base sampling rate from the interference patterns

How do you suggest to do that?

Did you use a composite lead by the way?

Yes. I couldn't get s-video to work a while ago (image remains grey). I guess both scarts of my tv are composite.

This is the first DVD player I saw which actually outputs all pixels without cropping.

Pioneer DV77.

2: this is probably a driver issue. I remember reading a driver .inf file in which the revision comments mentioned limiting the allowed horizontal resolution to a maximum of 720.(can't find it back though)

I could you can change this in your registry. But I don't know what to change ...

[erratic]

@Arachnotron, regarding that WSS line:
I read in this PALplus document (section 5.0) that a WSS line can also flag the format as Letterbox 16:9 Center. I was wondering if such a line can be inserted into a letterboxed MPEG-2 file to make a TV zoom in automatically when you play a letterboxed (not anamorphic) DVD.

This question doesn't really belong in this thread but I think the answer will be short: No, it's impossible.

[Arachnotron]

@ Wilbert,

I always thought that 'ITU compliant' meant "the 4:3 picture is given by the 702x576 part of the image

Your card puts 720 DVD pixels in 720 grabbed pixels when you grab at 720x576;
so, 720x576 grabs 53.333 µs, which means 702x576 does 52.0 µs after cropping the 720.

How do you suggest to do that?

I don't know the math for it, but the resulting patterns in the vertical lines is the result (harmonic) of the interference between the requested sampling frequency and the base sampling frequency of the card. Theoretically if you know the one you should be able to deduce the other, but I don't know how myself. Just a thought.

Yes. I couldn't get s-video to work a while ago (image remains grey). I guess both scarts of my tv are composite.

I thought so because of the extra blurring and rainbow artefacts caused by the comb filter of the card.

could you can change this in your registry. But I don't know what to change

I'll check. I just remembered which driver it was: the new drivers for an Asus SAA7134 based TV card. But many options are compiled into the driver and cannot be changed from the registry.


@ Erratic:

You might fake the pattern, but the resulting line would have a colorburst in it, which the original signal doesn't have. Also, you would have to be certain which line, if any, from the MPEG2 your DVD player puts out in TV line 23. Last, most DVD players crop a few lines, and since this would be the first line it is a sure thing to go. And of course, any DVD player would output it's own WSS signal in line 23, overwriting your WSS.

But I think you can set AR info when you author a DVD. I remember ripping a DVD with a wrong AR and burning it back with an adapted menu once. Unfortunatly, that was a while back and I don't know how I did it or where the option is. Something for a DVD forum I guess.

[trevlac]

Originally posted by Arachnotron
I don't know the math for it, but the resulting patterns in the vertical lines is the result (harmonic) of the interference between the requested sampling frequency and the base sampling frequency of the card.

Hey,

So many interesting things ....


Does this mean that the card's resize causes an interference pattern in the picture? Would it be better to cap at full size and then resize? Maybe I should learn more about this, instead of looking at post cap processing.

You can never really know enough ...

[Arachnotron]

@Trevlac,

This remark is based upon something I noticed when I was making my series of captures. When I captured using the Terratec grabster at 720x576 or the Terratec Cinergy at 704x576 I got really beautifull captures that looked almost identical to the original DVD. A 1 pixel vertical line in the DVD would result in a 1 pixel wilde line in the capture.

When I did the same with the BT878 or the CX23881 the resulting lines were blurred. I was very surprised by this; the sample rate of those cards is higher than those of phillips.

Untill I realised that the phillips card were working at the same base sample rate as the DVD player, 13.5 MHz AND that the horizontal sync happened at the same time. So every pixel in my cap was positioned exactly under the analogue peak which in turn resulted from a matching pixel on the DVD.

Basically, the system was genlocked and as a result the Kell factor no longer applied.

This was a happy coincidence; the other 13.5 MHz devices in this thread had a slight offset. As a result, the peak resulting from a white pixel on the DVD is 'hit' by at least two pixels in the capture. But because the sample rates still match, the resulting gray pattern is the same for each horizontal line. This is how I know Wilberts cap was unscaled and 13.5 MHz. (this picture illustrates some of this effect)

The moment your card has another sampling rate than 13.5 (as I suspect is the case for the ATI theater range and certainly the case for the conexant cards), you will get an interference pattern. The gray pattern changes going from left to right, but after a certain interval will start to repeat itself.

From this interval you should - in theory - be able to calculate back what the sampling rate was; the pattern is caused by interference between the 13.5 MHz of the DVD peaks (or, to be more presize, the cycle time of the repeating white pixel/19 black pixel sequence) and the sample rate of the card.

Dropping a scaler over this adds a third sample rate to the mix.

And of course, if you use composite (as wilbert did) you blur the original signal throught the comb filter. It migth turn out that Wilberts card WAS genlocked, and that with a svhs leas the cap would be sharp.

In analogue caps this may not be important, unless the source came from a digital master. If that is the case, there might be a capturing 'sweet spot' where the sampling matrix matches the source.

Edit: explanation of the pic

# A. line 130 of the 720x480 mpeg file. Pixels 1-7 are white, the following ones are 50% gray
# B. On playback, the DVD player crops the first 3 pixles and replaces them with horizontal blanking.
# C. The analogue signal produced by the DVD player. The signal rises from blanking level to white level, stays at white level for about 0.2 µs and drops down again to 50% grey level. Note the rise time going from blanking level to white level (rise time is the time it takes to go from 10% to 90% of maximum signal). This is almost 0.06 µs which equals 0.8 DVD pixel! This is why a vertical change from black to white on a DVD will always result in a grey pixel in the capture. To increase the precision, the test DVD uses white lines on a 50% grey background, since going from white level to 50% grey takes less time.
# D. Theoretical capture card with the capture window coinciding with the signal resulting from DVD pixel 7.
# E. Theoretical capture card with the sampling grid 'out of phase' with the pixelgrid on the DVD. The first pixel captured contains signal from DVD pixels 6 and 7, and is white, but capture pixel 2 contains signal from the flank and results in a grey color between white and 50% grey.
# F. Theoretical capture card with the capturing window beginning before the start of the active part of the DVD signal. Because of this, two black pixels are captured on the left. The third pixel is grey, since most of the analogue signal in this area is on a slope between black and white

The above is a bit of a rambling story, but I hope you understand what I am getting at? It's just a bit of speculation on the side, but interesting. It might be complete nonsense of course.

edit2 Looking at it, the pic may not be the best illustration if what I am talking about, but it is the only illustration of a shifted sampling frame I have

[trevlac]

@Arachnotron

For some strange reason I love this stuff.

Some observations form a guy who has no formal education in electrical engineering.

1) I don't see why a 13.5MHz card could not represent either the perfectly in phase line and (D), and the out of phase line (E). Depends on how the card locks on the sync, and considering it may not be working with a full lenght signal, I'm suspecious of an insync capture. *** Forget this. I re-read your post. You got 1 of these not all the time like I thought you said. ***

2) A higher sample rate card should have more 'pixels' in it's line. 17.74MHz is about 25% more for PAL. *** Hence the pattern you mention *** I may try 754 NTSC and look for the pattern. No scaler here.

3) I have never seen kell used when refering to horizontal. At least in a reliable source. Here is an example of many articles. No horizontal kell that I see. http://broadcastengineering.com/micr...49&PageID=5126

Also, It is my understanding that you should not be able to 'see' a single DVD pixel. The best you can do is see 1 pixel for every 1.333 pixels using a 6.75MHz bandwidth s-video connector. This assumes everything thru the DVD player supports this bandwidth. Composite is most likely less.

4) I am following your tests and information. I think it is great. I plan on doing a test on the drivers for my MSI CX card, when I get a chance.

Thanks

[Arachnotron]

For some strange reason I love this stuff.

Yep, me too (including the no formal training part

I will re-do some caps using my new test DVD's. See if I can reproduce it again. But the "in phase" pattern for 13.5 MHz caps without scaling is clearly visible and not only in my caps.

@3:
As to Kell: I am talking about verical lines here. The pattern can be seen in a horizontal band consisting of repeating single pixel vertical lines in my test DVD's.

But I took Kell as an illustration of the principal behind this, not as something that applies in the strict sense.
My understanding of Kell and Nyquist is that both are based on the same thing: if you are trying to sample /display a signal that has random lines in it, how narrow should your grid be to be sure to 'catch' an event which happens at a certain speed (line-width). The key here is random. A DVD player is a synthetic signal, produced from a perfect 13.5 MHz clock. Not random. So you loose the Nyquist/Kell like fudge factor if you cap it at a fixed sampling rate.

@4
As far as I have seen, composite is less.
But under optimal conditions, you can see a single DVD pixel. Only those conditions don't apply to a normal video signal.

For example: If you try alternating black and white single pixel vertical lines from a DVD, In your cap you will probably see an alternating light gray/dark gray pattern. This is not a capture artefact; the peaks resulting from the white pixels are fusing, resulting in a signal that oscillates somewhat between white and blacklevel. edit but alternating double pixel lines you can separate, which sounds about right for the limit you mentioned. /edit

This is I think the bandwidth limitation you are talking about. You cannot have an ordinary picture, cap it from DVD and then reproduce it. Because of bandwidth limitations, the finer details will fuse an come out blurred.

But I was very carefull to avoind this on my test DVD. I have single pixel lines on a background that is already on about 25-30% gray level. The resulting peaks can be picked up in a single sampling period of a 13.5 MHz card. But once again, not anywhere near standard.

But I have made some Oscilloscope caps of what the signal looks like. It will take me a few days to process (they are CSV files, I have to plot them first) and I think this will explain a bit why this works.

[Arachnotron]

@ Trevlac
I did some tests. The results are in this pic (26k):

http://www.arachnotron.nl/videocap/download/strips.rar

it shows a cutout of the center 7xx x 16 pixels of a cap, aligned under each other for easy comparison.
the strips are:

1. original BMP source of the DVD
2. SAA7113 cap 720x576 (ITU device, does not have a scaler)
3. SAA7134 cap 704x576 (13.5 MHz device, does have a scaler)
4. BT878, IUlabs, 696x576 (17.2 , scaler)
5. BT878, Iulabs, 704x576

All show a pattern around a white center line except the last cap.

So, as long as the samplerate is 13.5 all vertical lines look the same. wether you get 13.5 directly or through scaling does not matter.

The moment the sample rate is not exactly 13.5 you do get interference patterns.

I found out the pattern you see is caused by the first and second harmonic on the sides of the main peak. This again is caused by the high 'offset' caused by the light gray background. An effect I hadn't considered when I made this testDVD. . When the oscilloscope pictures are finished I can show this more clearly.

What this does make me suspect though is that these devices sample at 27 resp. 35.44 MHz, 9bit resp 8 bit and from these raw samples (27/35 MHz IRE values) directly interpolate the desired number of pixels. If there was (for the BT878) a 921 pixel intermediate that was resized to 696 there WOULD be a different pattern around the white lines between the unscaled 13.5 and scaled 17.2 devices.

[trevlac]

@Arachnotron

I don't really have anything to add. I just wanted to thank you for the test pic and all of the work you have been doing on this. I havn't been able to run the test myself (busy time of the year), but I'd like to. Have you ever seen the AVIA pixel cropping test pattern. This might give you some ideas on yours. They stagger the lines on a grey background. I'll try to post a link in a few hours.

Cheers.

Edit:

Here are the pics. The first is the pattern they use. The second is some instructions they provide. I thought this might be helpful as a contrast to your approach. As to the merrits, you'd know better than me. It appears they avoid the issue of high frequency events by staggering the vertical lines. Also, they include markers in the corners that can be used on a scope (they say).

http://pics.trevlac.us/img.htm?crop.jpg

http://pics.trevlac.us/img.htm?cropinfo.gif

[Arachnotron]

@Trevlac

Hi Trev,

Thanks for the info. reading back my last two posts in this thread I realize they are a bit messy, partly because my understanding of these effects changes/increases while writing and discussing them.

Once I have a firmer grip on these things and have some scope pictures to go along with this I'll do a short webpage on the sort of artefacts/phenomena you see in captures.

It will take some time since I have two other pages to write first. :-)

Thanks for the test pattern you mentioned. I'll look into it. Your links didn't work by the way, but from the descriptions on the web I get the inpression it is a tespattern you can use to find out which pixels are cropped by the DVD player on the edges.

[trevlac]

Originally posted by Arachnotron
Your links didn't work by the way, ....

They work for me on different machines. They use javascript to show the image because my provider does not allow direct links to images.

The direct links are below. To use them you have to paste them into a new browser window.

pics.trevlac.us/crop.jpg
pics.trevlac.us/cropinfo.gif

[Arachnotron]

Thanks, I got them. I am using Mozilla, perhaps that is the problem. I get a red smily that appears to give me the middle finger (or is it thumbs up ? )

By the way, do you own this test DVD? Is it any good?

[trevlac]

By the way, do you own this test DVD? Is it any good?

I do own it. It's a bit old. There is a newer/cheaper one at Amazon.co.uk.

I have found it to be helpful. It probably has 100 test patterns. However, it doesn't really provide great instruction on their uses. The more I learn, the more I go back and say 'oh, thats what you use that for.'

Red smilies? How did they get there?



NOTE:The link I gave was not to the product I have. The one I have is older and costs more. You should read reviews of the newer one before you buy.

[Arachnotron]

Red smilies? How did they get there?

Yep, that is the one I am getting. In IE6 the pics display ok, in mozilla I get the finger

I have found it to be helpful. It probably has 100 test patterns. However, it doesn't really provide great instruction on their uses. The more I learn, the more I go back and say 'oh, thats what you use that for.'

OK, thanks for the info. I have been thinking about ordering it, but wasn't sure about it's usefullness.

[Arachnotron]

@trevlac

Look here

Small pic showing DVD pixels, resulting oscilloscope trail and resulting pixels in BT878/btwincap 720x576 cap. (gray values taken from actual cap).

Kinda explains the 'interference' pattern, and the bandwidth limitations you mentioned.

[trevlac]

@Arachnotron

I've been reading a bunch and here is what I am guessing on this stuff today.

1) Sampling Theorem says "A signal can be reconstructed from its samples without loss of information, if the original signal has no frequencies above 1/2 the sampling frequency. (Nyquist)"

This tells me there is no such thing as Kell in captures. I think the 'missing link' is that a signal is (and can be) broken down into sin or cosine waves. Even though the samples may not line up with the signal peaks, they can be reconsturcted because the interpolation between samples is not a straight line, it is a sin curve that matches the orignial peak. Hense, "without loss".

This makes the "perfect sync" capture a mystery. I believe you had it, I'm just not sure what it was.

2) The less than exact samples are not due to the cards, but due to the dvd player. A single pixel is higher in frequency than is supported by DVD. The player filters this and the result is the curve you posted. It looks like a "sinc function". Here is where I really can't connect the dots. A sinc function is a representation of a low pass filter.

This has been great fun. However, some of the reading does make my head hurt. I found I lacked knowledge of things like "periodic trigonomic functions". But I think I am beginning to muddel thru.

Here is a paper I read about 20 times before I started to 'get it'. I might still not get it.