PDA

View Full Version : New Idea: Record analog HiDef!


jmac698
11th March 2006, 18:46
I knew it was possible! The C4All technique, has paved the way for new and more crazy experiments.
Try this now: set your component outputs hidef, move the green plug (Y) to the composite output instead. You have 480i, and repeating patterns of crazy colors. If you can cap that, you can recover fully the hidef signal.

Imagine hidef were twice the speed, now two lines occur in the same space your capcard can record just one:
Hidef:
1111 (37uS per line)
2222
3333
4444

Hidef as recorded by standard def:
11112222 (65uS per line)
33334444

So if you mix them, then two lines of hidef color are now on top of one line of standard luma.
To reverse the process, just extract the color only from the two halves, and turn them into two lines of 720 wide color.
It seems like this is pointless because, the two halves are only 360 resolution when capped, so there is no gain.
There is real gain, but it's a little complicated. Up to 4x gain is possible for very deep reasons.
I don't know how to explain it now, you have to understand video very well.
So I will just say, in this example you can see twice the vertical lines but only half the horizontal resolution. You could fit thousands of lines into the same space, if you put them beside each other to fill the screen, but they would have to be scaled smaller. So in a sense you are trading vertical lines for horizontal resolution. Does this make any sense? Would you rather 1080 vertical lines of 160 pixels, or 480 lines of 720 pixels?

So what if you do both and try to mathematically combine them? You should get 1080 lines of 720 pixels, or half of true hidef. I'm skipping some deep details here, but you will gain a lot of noise from all the computations being done.

To skip ahead quite a bit, imagine a math problem like this:
Original (1440x960):
A B C D
E F G H
Normal cap (720x480):
(A+B+E+F)/4 (C+D+G+H)/4
Hidef4All cap (expands to 360x960):
(A+B+C+D)/4
(E+F+G+H)/4

This forms a set of simultaneous equations, 8 unknowns in 4 equations, that can't be solved.
A+B+E+F=W
C+D+G+H=X
A+B+C+D=Y
E+F+G+H=Z

What about a lesser problem, to achieve true 720x960 rez:
AB+EF=W
CD+GH=X
AB+CD=Y
EF+GH=Z
X-Z+W-Y=0, there's no way to extract info without different scale factors. This might be possible with your capcard, at least it's the only thing you have under your control. If I can solve this problem it's the equivalent of capping in higher horizontal rez anyhow.

Well, I have a million thoughts, it would take weeks to sort through all the possiblities, if anyone is seriously interested let me know, and together we can make hidef possible.

I have made a script to do this, not ready to post pictures yet. One of the many problems is the hsync area forms a gap in the picture.

neuron2
11th March 2006, 19:21
Does this make any sense? Not really, because my high definition capture card avoids any need for all that stuff. It's a fun experiment, though.

jmac698
11th March 2006, 19:48
Aren't you lucky :) Let me know when you're ready to sell...
My technique would work with normal capcards, of course.

I have an obvious improvement right away, if you cap NTSC in PAL-M mode, you can trick your capcard into giving you 768 horizontal rez. In my HiDef4All technique, you are trying to record a very high frequency signal, so the 768 pixels will all be "real" information. It's only a 7% improvement though :(
You can combine a 720 and 480 cap, like this idea:
Original (1440)
A B C D...
Normal cap: (720)
A+B=X
Lowres cap: (480)
A+B+C=Y
So C=Y-X. You can only get half the hidef pixels. This has been invented before, there's a way to sample a repeating pattern at a low sample rate, if you can shift the phase each time. It's how GHz digital scopes work. In this case you need to delay the hidef signal by a delay equivalent to a hidef pixel. The HDELAY register of cards typically don't have that kind of rez, but they could have multiples, ie 18+1 hidef pixel delay, or 9 standard def pixels plus a "phase" delay, which is what I need. This assumes perfect stability in timing and low noise.

tedkunich
12th March 2006, 17:49
Not really, because my high definition capture card avoids any need for all that stuff. It's a fun experiment, though.


I just rip the .TS files off of my HDTivo! ;)