Resolution / bit rate questions: Best quality for Hi8 camcorder tapes ?

[myron]

I recently put together a new computer system for my family. One of the objectives was to build a system capable of transferring old Hi8 camcorder tapes of my children to DVD; I've been worried about the tapes going bad.
ASUS P4C800-E Deluxe motherboard with a 2.8 GHz P4 processor (800 MHz bus)
1 GB ram (dual-channel DDR matched)
ATI AIW 9800 PRO video/graphics
Sony DRU 510A DVD/CD
About the only thing I "skimped" on was audio - I'm using the audio from the motherboard (AI audio: 6-channel AD 1985 digital audio CODEC) as opposed to some flavor of SoundBlaster card or such.

The ATI card came with some software components, one of which was Pinnacle's Studio 8. I initially made a couple of DVDs from camcorder tape transfers. While the software was completely buggy, I was able to construct playable DVDs. I did encounter out of sync (OOS) issues with audio not syncing w/ video on most transfers. I found out that Pinnacle was about to introduce version 9 and so waited to obtain that hoping it would yield better results. Studio 9 was a major disappointment in that the OOS issues were worse. This prompted me to start reading the user forum on Pinnacle's site with respect to Studio 8 and 9. There were numerous postings about OOS issues with 8 and 9 (with everyone experiencing greater problems with 9).

Because of the Pinnacle issues I began searching the net for information and encountered the 'digitalFAQ' site (along with 'dvdrhelp' and 'doom9'). The 'digitalFAQ' site is well constructed with a logical layout and seemed to have the most applicable information that I was trying to learn about. I have read many, if not all, of the guides and am beginning to understand the video to DVD process somewhat.

I would like to create DVDs, for playback on TVs (analog) from three sources: my old Hi8 camcorder tapes, my new camcorder (Digital Video), and TV (DirecTV satellite) captures (shows like NOVA and Invader ZIM cartoons).

I would like to start with the cartoons as their content seems easiest to see artifacts and such with – thus helping debug the process of producing quality DVDs. As such I will start asking questions about capturing and we’ll see where we go from there:

Reading the various capture related guides I decided to capture the cartoons in MPEG-2 as I only plan on “editing out” the commercials. I am more interested in achieving quality than fitting more content on the DVD (DVD media is now reasonably priced if you catch sales of good base media).

I have been capturing, using the ATI video recorder, with settings:
FormatMPEG-2
StandardNTSC (525)
Resolution720x480

I used the 720x480 resolution since DirecTV mostly uses 544x480 and it is full D1 video (matching closely my intended final output resolution). Since TVs (NTSC standard) have approximately 324-350 or so of capability, what is happening with DirecTV and MPEG-2 full D1 video when played on a TV? It doesn’t make sense that these resolutions are “more capable” than the TVs capability (I haven’t stated this question very but I’m betting you understand my questioning)?

After reading the “Capturing MPEG with an ATI card” guide I understood most of the suggestions with the exception of “bit rate”. I’ve read other places that “bit rate” is the single most differentiable aspect of video quality. What in particular I do not understand is the suggestions in both the “AVI vs. MPEG” guide and the “Capturing MPEG with an ATI card” guide. The examples in the former are:
352x240 (again NTSC) = 2.0 MB/s or 2000k
352x480 = 4.0 MB/s or 4000k
720x480 = 8.0 MB/s or 8000k
The math doesn’t work out for me. I shouldn’t lead my question but I’m betting that the answer is related to MPEG-2’s frame sequencing (I, P, and B frames) in that the P and B frames are ‘references’ and not a full capture thus requiring less total overall bandwidth. So the question is what specific or general formula is being used to calculate the suggested bit rates?

[admin]

TV is old technology, derived from research and implementation in the 1930s and following decades (up to about the 70s-80s). The resolution for tv is peaked out at 300-500 resolutions. The idea behind DVD and HDTV is that the signal will one day be shown on progressive hi-def displays. In the end, both should be the same, a PC and "tv set" will just be high res monitors for all kinds of content. Interlace will also disappear someday (though many HD devices can play it in interlaced mode or with special filters). DVD is between the two (old and new tech) in terms of resolution. Where does extra res go on tv? Nowhere. It's just not played. But on a true hi-res display, you can easily see flaws.

I HONESTLY would rather stick to tv, or at least put "fuzz" filters on the new HD sets. Even a perfect VHS tape can look like trash on a hi-def display.

Bitrates controls quality, but only to the effect that enough is given. You can give too much or too little. too much does nothing but make big files. Too little creates macroblocks.

704x480 is true Full-D1 resolution, as is the 720x480 sizing. Now ignore 480. 704 is half of 352. The max for 704/720 is about 8.0 Mb/s. The max for the half 352 is half at 4.0 Mb/s. Now, remember x480 .. well half it to x240, and again, half the bitrate. The 2.0 Mb/s is the max. You can go more, but it won't do much good most times. I can make 15.0 Mb/s 160x120 res, but it'll just make a huge file with a small image (as well as be non-compliant for CD/DVD formats).

I-P-B GOP is the nature of the compression of MPEG, not to be confused with bitrate. I-P-B, or I or I-P, gets the percentage of the bitrate no matter what it is. Uncompressed AVI is uncompressed video, with about 15GB/hour file. MPEG is much less. I know that a 352x480 at 2.5 Mb/s is 1GB/hour. Most of the compressions (about 3/4ths) is due to the MPEG having I-P-B (with I as the main image, and then P and B data ONLY having data from what is DIFFERENT between the I frames). The bitrate determines the amount of data given to the IPB, and the encoder makes it's own judgments on rationing it out, using various algorithms (DC8, DC9, or DC10).

More B and P frames will cause smaller files (but not always), and can degrade quality. The standard delta pattern is 1-4-2, but it can be changed for special situations (like I use 1-2-2 and 1-3-2 sometimes for cartoons using live capture on ATI card). Always test on a relatively smooth background and then again on a fast motion scene to see which looks best. GOP can also affect motion fluidity.

Also, a big reason for audio drift (OOS as you said) is onboard sound. It often uses junk for chipsets (ESS or VIA). Even a $20 SoundBlaster from Walmart will do better in the end, assuming you've still got an open PCI slot.

But PINNACLE STUDIO really is terrible software.

You'd be better off using ATI MMC 8.x and then editing in WOMBLE MPEG VCR for direct MPEG captures.

For AVI captures, use ATI MMC 8.x and then import the captured file into Adobe Premiere (or Vegas Video) and then have your way with it. Export using MainConcept (may have another name like Adobe MPEG Encoder) which comes with new version of the NLE softwares I mentioned.

[myron]

With respect to the bit rate and taking just one example from the “AVI vs. MPEG” guide, I do not understand how the suggested bit rate was calculated. If I were to calculate the bit rate required for a 720x480 NTSC capture I would come up with:
(720 * 480 bits of res per frame) * 29.97 frames per second)
This yields 10357632 bits per second which is already above MPEG-2’s 9.8 MB/s (including audio). Both of these are considerably greater than the suggested bit rate of 8 MB/s or 8000k?

Since we are on the bit rate topic (and hopefully closing it out to move on to new topics) would you explain a little about “Max Bit Rate”, “Target Bit Rate”, and “Motion Estimation Quality”. The first two are somewhat intuitive but not the latter. By the way, I don’t mind you pointing me to other sources if I’m asking questions that have been answered already.

Also, you mentioned macroblocks with respect to too low of a bit rate setting. What visual manifestation would macroblocks yield (one thing I remember when I wrote a graphics driver long ago was the visual feedback that was very helpful in debugging my code – I would likewise expect macroblocking to manifest itself visually and thus remind one that they had captured with too low of a bi trate setting.)

Thanks for mentioning that my OOS could well be related to my usage of the on-board audio chip. I was beginning to suspect this which is why I was careful to explicitly state that I was using such.

Yes, Pinnacle Studio is pretty much trash. This is disappointing in two respects. First it was included by ATI with my video/graphics card and was my first introduction to creating video DVDs. Second, it really is well structured and has a nice interface and flow but what good is a nice user interface if the core functionality isn’t there.

[admin]

Math is not my strong point. I'll go ahead and admit this up front. I'm a college-educated journalist, with strong computer and online background and self-taught in the hobbies of video and Web design (and I also offer freelancing in all these topics). Math was never needed in any of these things.

What I can say in this respect, is that I know what the constants and the variables are, in regards to how they affect the final video files. How the math added up actually never quite came to the focus of what I looked at. I just know that under or over a certain bitrates, problems develop, and I know what those problems are. One of my biggests mottos in videos is that experience and actual occurences are more important than the theory or science, which is sort of where this is leading. But I can admire trying to understand "why" so I'll try to answer to the best of my ability.

Know that I mainly experimented to get the numbers I needed for "time x bitrate = final DVD size" .. and then I sought out the pros for agreement and advice. All of them agreed with my findings for the most part.

I think the problem here is the math is larger than you're calculating, and quite honestly more complex of a problem than I know how to deal with, having little more than high school pre-calculus as my highest math, and barely passing at that. Again, I'm a writer and designer, not heavy on the math skills beyond financial usage.

You have to consider that bitrate is per second. Each second is so many fields (not really frames with interlacing like DV-fed source, but we'll stick to frame anyway as it's calculated the same with two field sharing the frame bitrate), and a bitrate is parceled out at an average, as the DC (discreet cosine) is not perfect. It's just a great big calculus equation, pumping out 30 or more answers per second.

What I can say in response to the question is think of bitrate allocation in a bell-curve. You cannot have too much, else you'll get macroblocks.

An MPEG is essentially an image made up of sectors calculated by the motion compensation. It has more blocks inside these larger block sectors. Here's a quick-n-dirty image:
[img]uploads/lordsmurf/grid.jpg[/img]

You have intrablock and innerblock motion to consider. Intrablock is the interaction between surrounding blocks. Innerblock is the stuff changing within a blocks. When something moves, these are affected. The bitrate is pumped up in the areas alone. This is where I-P-B compression takes hold. Areas that do not change are left alone in the IPB compression. Only the areas of movement get the bitrate in the IPB compressed areas.

When there is enough bitrate, motion is smooth, with zero errors. When bitrate is too low, you can see the blocks. It needs adequate data to maintain both the color quality AND the fluidity of movement.

Macroblocks being this:
[img]uploads/lordsmurf/macroblocks.jpg[/img]

While this was a blow-up of the image, on a tv screen, you can see the blocks, as they tend to change shade/quality several times per second. It become noise or what's called an "artifact".

The bell-curves are as follows:

[img]uploads/lordsmurf/mpeg1bell.jpg[/img]

[img]uploads/lordsmurf/352mpeg2bell.jpg[/img]

[img]uploads/lordsmurf/720mpeg2bell.jpg[/img]

You'll notice that 1800, 3500, and 7000 are good quality. For slight better, go for the gold at 2000, 4000-5000, and 8000-9000. Anything more is on the plateau until it hits a sharp spike at the overhead number of 9800 where a DVD player rejects it. I forget the max for MPEG1.

Also notice the poor quality of VCD at 1150k MPEG1. Had MPEG1 been 1850 (the max MPEG1 allowed by DVD is 1856k), then VCD would never be known as "bad quality". Only issue would have been proper deinterlacing (not that deinterlacing is good, but some methods are better than others).

Here's my quick calculation table. This is done in terms of 23-minute episodes since tv is my main hobby. Take 23xep count to get total running time.

On my APEX DRX-900 using 4-hour mode:
11 eps = 370MB = 2520k bitrate

On my ATI AIW card using ATI MMC MPEG2:
9 eps = 477MB = 2610k bitrate
8 = 537 = 2967
7 = 614 = 3426
6 = 716 = 4033
5 = 860 = 4890

Different IPB configurations, different MPEG encoders, different DC, and different hardware can affect these numbers (as you may notice between the ATI AIW and the APEX).



Edited by: lordsmurf