352 vs 720 resolution at 2700 Kb/s

[MrMoody]

Here's an explanation of standard "lines of resolution" as in "VHS has 240 lines of resolution," which is also what this test pattern measures,
-VS- Pixel resolution:
http://videoexpert.home.att.net/artic1/201res.htm

According to that article, 352 pixels is less resolution than VHS, about 185 "lines of resolution." Of course in the test shots it looks like it gets up to about 220-230.

[Wilbert]

According to that article, 352 pixels is less resolution than VHS, about 185 "lines of resolution.

You must be confused. I don't know how you concluded this from your link.

VHS has 240 lines of (horizontal) resolution, which "equates" (in an ideal world) 312 pixels.

[SatStorm]

Just to remind (to the non NTSC users), that for us, the PAL users, 4:3 is 768 x 576 (not 640 x 480)

BTW: For 2700 Kb/s I would use only 352 x 576/480.

[MrMoody]

According to that article, 352 pixels is less resolution than VHS, about 185 "lines of resolution."

You must be confused. I don't know how you concluded this from your link.

VHS has 240 lines of (horizontal) resolution, which "equates" (in an ideal world) 312 pixels.

I don't see how you get to 312 either.
By the math they use:
352 pixels x .7 (Kell factor) x .75 (aspect ratio) = 185 lines.

240 lines / .75 (aspect) / .7 (Kell factor) = 457 pixels.

Granted I think this is somewhat off because in the test posted 352 is very close to VHS, and when I watch 480 it looks a lot better than VHS.

I think the way they measure favors the analog signal more than the naked eye does, and the Kell factor appears to be around .85 before the lines disappear.

[lordsmurf]

Kell factor is a display phenomena, it is outside the scope of a source argument. I wish people would quit using that theory, as it gives off totally bogus information for configuring analog->digital equivalents.

[Wilbert]

In addition to the remark of LS,

"Take for example a standard PAL VHS tape. The bandwidth typically will be 3.0 MHz for consumer grade equipment. So you need at least 6.0 MHz sample rate to capture all possible detail from the tape. 6.0 MHz sample rate with the 52 µs line duration of PAL equals a minimum of 6.0 x 52 = 312 pixels per line." (see chapter 3 of capture guide at doom9)

eq.2 of http://dbserv.maxim-ic.com/appnotes.cfm?appnote_number=750
relates bandwidth and lines of horizontal resolution.

[MrMoody]

Thanks! Then for NTSC, the line duration is 53.5 µs, so VHS would be equiv. to 321 pixels. This makes sense. Broadcast/composite is 3.58MHz (discounting comb filters) x 2 x 53.5=383 pixels. S-VHS I believe is about 4.5 MHz (assuming the source was that good, i.e. not broadcast, and you aren't ruining it using a composite connection anywhere), which would be about 480 pixels. This matches more closely what I'm seeing.

[Wilbert]

You are confused between ntsc line 63.5 and 52.65 being its active part. Furthermore, the bandwidth of ntsc broadcast is 5 MHz. Getting 2*5*52.65 = 526 pixels.

The bandwidth of SVHS is also 5 MHz.

http://www.doom9.org/capture/introduction.html

3.58 MHz is the color subcarrier frequency (of a ntsc signal), and has nothing to do with bandwidth.

[MrMoody]

That page quotes NTSC at 4.2 MHz, not 5.0, which is 442 pixels.

[Wilbert]

Yes, I switched pal and ntsc. My mistake.

[tedkunich]

In addition to the remark of LS,

"Take for example a standard PAL VHS tape. The bandwidth typically will be 3.0 MHz for consumer grade equipment. So you need at least 6.0 MHz sample rate to capture all possible detail from the tape. 6.0 MHz sample rate with the 52 µs line duration of PAL equals a minimum of 6.0 x 52 = 312 pixels per line." (see chapter 3 of capture guide at doom9)

eq.2 of http://dbserv.maxim-ic.com/appnotes.cfm?appnote_number=750
relates bandwidth and lines of horizontal resolution.

<seaches for time domain <-> descreet domain transform text book..>
.
.
.
<dammit, cannot find it.... warning rant/incoherent rambling follows>


Wibert, appologies in advance, as I really respect you experience and advice here and on Doom9, but I really wish people would stop regurgitating this "at least 2x" nonsense. A 2x over sample, i.e sampling 2x faster than the source, i.e. Nyquist frequency, ONLY guarantees that you will not get aliasing - it says nothing about getting anything close to an equivalent approximation of the source. Consult ANY textbook on sampling theory and they all say the same thing, the more samples you use, the more accurate the time->descreet transform becomes. They will all point out that sampling at the Nyquist frequency is less than optimum.

If you read a little further in the posted Maxim app note, you will find this:

Using equation 1, we calculate a maximum signal bandwidth (BWS) of about 4.2MHz. This is the highest frequency in the signal. Now let's assume that we need less than 0.1dB attenuation. Using equation 3, we calculate the minimum signal bandwidth necessary to be 27.5MHz. Using equation 5, to account for variations, gives 41.3MHz. This is the circuit -3dB bandwidth required to achieve our desired resolution and maintain the signal quality.

Although this quote does not specifically mention digital sampling, the same holds true - to properly capture a 4.2MHz signal requires a circuit bandwidth of ~41MHz, and simlarly a A/D sampling rate of ~41MHz, a ~10x over sample. Most capture cards come out of the box sampling at 27.5 or 28MHz, and are already at a disadvantage. I cannot speak for all capture cards out there, but the Brooktree/Conexant cards all sample at 27.5/28MHz regardless of your "capture" resolution, and then rescale the sampled data into the requested resolution.

....ok, enough of my rambling/rant.... I need my morning coffee.... and I have a full day today, so no forum time for me today....


T

[Wilbert]

, but I really wish people would stop regurgitating this "at least 2x" nonsense. A 2x over sample, i.e sampling 2x faster than the source, i.e. Nyquist frequency, ONLY guarantees that you will not get aliasing - it says nothing about getting anything close to an equivalent approximation of the source.

If you read the books on sampling theory you will see there exists an ideal reconstruction filter which enables you to get the original signal back when sampling at 2x times Nyquist. Of course this reconstruction filter exists only on paper. In practice you (the chip/driver or you during postprocessing) use less-then-ideal resizers, which implies that 2x Nyquist is not enough.

This may imply or may not imply, that a higher sampling rate is always better (for some other folks which are reading this we are talking about lossless capping here). However, also this is a theoretical issue. In practice there will be a limit when passing that you won't (or hardly won't) see any difference any more. (btw, limit doesn't mean "sharp" limit here ..., and you means you by visual inspection)

If you read the analog capture guide (at doom9; chapter 3 and especially 5) you will see we mention these issues, and advice people to capture at 3xNyquist when possible (which will be enough in practice).

But, it's a bit tiresome to mention this over and over again.

btw, if you have remarks about the capture guide, feel free to discuss them with us.

Although this quote does not specifically mention digital sampling, the same holds true - to properly capture a 4.2MHz signal requires a circuit bandwidth of ~41MHz, and simlarly a A/D sampling rate of ~41MHz, a ~10x over sample. Most capture cards come out of the box sampling at 27.5 or 28MHz, and are already at a disadvantage.

10x or 5x times, I don't think you will see the difference in practice.

....ok, enough of my rambling/rant.... I need my morning coffee....

It's evening for me

[Nelson37]

Ted and Wilbert - Do you have any additional info on the "fixed sampling rate of most capture cards" and the "internal rescaling to desired res"?
This is exactly what I have been talking about.

Also, what resolution does 28 Mhz correspond to?

[tedkunich]

Ted and Wilbert - Do you have any additional info on the "fixed sampling rate of most capture cards" and the "internal rescaling to desired res"?
This is exactly what I have been talking about.

Also, what resolution does 28 Mhz correspond to?

In my previous caffeine deprived rant, I was referencing the CX2388x data sheet... you can find it here http://yanyan.dtdns.net/capture/pdf/cx2388x.pdf. Have a look at section 2.6. This is technical data sheet and is targeted towards engineers.

In short, the chip can sample at several different frequencies depending on the standards. My NTSC Leadtek is fixed at 8xfsc in the driver which equates to a 28.6MHz sampling rate - this is the rate at which the incoming video is sampled regardless of the selected "resolution". The chip has a 6 TAP interpolation filter that resizes the captured line to the desired resolution.

As to your question about resolution.... from the data sheet:

2.6.8.5 The Horizontal Scaling Ratio (HSCALE) Register
The 16-bit HSCALE register is programmed with the horizontal scaling ratio. When
outputting unscaled video (in NTSC), the CX2388x produces 910 pixels per line.

That 910 pixels includes blanking and sync pulses which get cropped by other register settings, so your effective capture resolution is either 780 pixels for NTSC Square pixels, or 858 pixels for NTSC CCIR601/656. The output resolution that your capture app sees is scaled from these values with the devices' internal interpolation filter.

Hope that makes it clear as mud


T

[tedkunich]

10x or 5x times, I don't think you will see the difference in practice.

Oh, I agree with you whole heartedly. I believe that that Maxim app note was intended for broadcast quality data.

I just noticed that there are new sections in the Doom capture guides! after the holidays are over and I settle into the new job, I'll have to check them out... looks like a lot of new info has been added in the past couple of months!


T