High Resolution Capture

[LavaJockey]

With my Maxtrox e450, I can capture video at 704x480, 30 fps with no frame
loss (thanks to Huffyuv). At that resolution, what is really happening for
each horizontal line? Is the A/D converter chip really sampling the input
video 704 times each line? Or does it sample a lesser number of times, say
352, and then do some sort of interpolation to provide the 704 data points?

How about for other capture cards? Do they all sample at the rate you ask
for, or do they sample at a lower rate, and then do processing to provide
the extra pixels? Also, in the future, where would I go to determine this
sort of information for myself?

[norm]

Yes the Video A/D chip really does sample at 720 x 480 they all do with the exception of some very old A/D's.
It only an Mpeg1 encoder that will spacially decimate down to 352 before applying the mpeg algorithm. ( only 704 of the 720 pixels are actually used in the process )
If a product did sample at 352 then interpolate up to 704 then the quality would suffer.
If you checked what video decoder ( the A/D ) a capture card is using and then looked up the manufacturers specs that will tell you what it is capable of.

[Anonymous65]

The A/D converter should sample at the rate you specify. So, if you specify 704 pixels per line, that is how many samples that are taken during the active video line. If you ask for less samples, the A/D converter will sample less. This is the reason why it's easier to capure at lower sample rates, and more difficult at higher rates. The A/D converter has to work harder for higher sample rates, and output more data. I don't know of any A/D converter that works at full capacity all the time and then use software to downconvert. There might be some cards (old) that has a max capacity of 352 and use software to double the pixel rate to 704. However, the quality will suffer.

ATI cards and the PV231 capture cards does a interpolation of two fileds into one when capturing at x240 (vertical resolution). It actually captures at x480, but use software to convert it to x240. You can see a "big" difference between a true x240 capture and a interpolated x240 capture. The interpolated (2:1 vertcal reduction) capture looks better. You can do the same test with horizontal pixels. Capture a file with 352 pixels, and then capture a file with 704 and convert the pixels to 352. The last file will look better, even though both files has 352 pixels.

[norm]

skittelsen - Your assumption that by asking for less samples the A/D will sample less is not necessarily true. It depends whether the A/D has a built in scaler. Not all of them have. Your PV233 has and so does a DVC2. Just because the A/D has a scaler built in doesn't mean that it has to be used. The scaler is very handy when the capture card transfers the video information to the PCI bus for displaying. Theres no need to have full bandwidth when you are only going to display 352 x XXX. Both the DVC2 and the W99200 don't have a PCI bridge built in but many do such as the PV256. The PCI bridge device will scale the video being sent to the PCI bus but the Mpeg encoder on board will accept the full bandwidth then spacially decimate the video signal it receives if it needs to ( for mpeg1 ). The only difference between 720 and 352 pixels to a A/D converter is the time it takes convert it. It doen't work any harder.
Maybe I've got it all wrong, if I have ( my appologies ) I'll go and do the digital video course again.

[Anonymous65]

I agree,
I was thinking about the BT8x8 chip with no hardware mpeg encoder on the board.

[stanwebber]

how can pixels(and square pixels at that) be selectively sampled from an analog signal? what you are communicating is gibberish

[LavaJockey]

>how can pixels(and square pixels at that) be selectively sampled from an
>analog signal? what you are communicating is gibberish

You sample points along the horizontal line. One can choose as many sample
points as one likes; 10, 100, 1000, whatever. Now, it may be that the
bandwidth of the analog signal is such that sampling any more than X number
of points is, well, pointless. But that's a different issue.

[stanwebber]

i still think you are both wrong about the bt8x8. my experience is that the brooktree chip samples the full analog signal before applying an algorithmn to resize for your screen