mike@BRL.MIL (Michael John Muuss <mike>) (11/22/90)
(I wrote this rather lengthy response, to INFO-IRIS. It seemed worthwhile to share it with this somewhat larger community as well. Please note that many of my assessments are subjective; if you have achieved different results, by all means report them. Mileage varries. -Mike) In a recent note to INFO-IRIS, Stuart Levy wrote: >> From what I hear, super-VHS is supplanting 3/4" tape. (In fact, JVC and >> Panasonic are discontinuing their 3/4" lines.) SVHS uses Y/C component video, >> with two channels on the (1/2") tape, one each for luminance (Y) and chroma (C) >> plus the usual audio stuff. SVHS is supposedly higher resolution than 3/4" >> but somewhat worse noise. The difference is apparently not large, but >> SVHS equipment is decidedly cheaper. ... This is somewhat misleading. 3/4" U-matic tape is also recorded as separate luminance (Y) and chroma (C), as is regular VHS, Super-VHS, and many other formats. The differences between these formats are in terms of bandwidth available for each signal, and signal/noise ratio. While some very low-budget operations have begun mastering on S-VHS, there is no question that S-VHS is a lower quality format than 3/4". A 5th or 6th generation dub on 3/4" is still of acceptable quality (although not prizewinning), while the same can not be said of even a 3rd generation S-VHS dub. Certainly not with any S-VHS equipment that I have worked with. It is important to note that the Y/C bandwidths on 3/4" come in two flavors as well. Regular 3/4" has about 260 lines of resolution (e.g. about 520 pixels/scanline), while SP-format 3/4" has about 360 lines, or about 720 pixels/scanline. The NTSC format is bandwidth limited by law to have no more than 720 pixels/scanline, so this is as good as it gets in real NTSC. Wider bandwidth signals can be easily created in your studio (the "multiburst" signal from most test sets is an example), but it isn't real NTSC. Thus, the SP version of 3/4" fills roughly the same need for 3/4" users that S-VHS serves for VHS users, but with the 50 dB video S/N and tape speed stability of the 3/4" format. A QUALITY COMPARISON: Subjective, and Measured. Just last week, I had the opportunity to make several dubs of my recent video, and used this chance to compare S-VHS to regular (not SP) 3/4". I tried to give the S-VHS dub every advantage. I used my NEC 1000 S-VHS machine ($1000), which I have measured as having about 375 lines of resolution in S-VHS mode (i.e., full NTSC bandwidth) on the composite NTSC input. Note that the manufacturer claims this machine to have "425 lines of resolution" when fed a non-band-limited S-Video (Y/C) input signal (which is hard to come by). I connected the output of the 1" mastering machine directly to the composite NTSC input of my S-VHS deck, and used TDK's best S-VHS tape. Hi-Fi sound was also taken straight from the 1" machine. (BTW, 1" machines have rather mediocre sound, ~45 dB S/N, compared to better than 80 dB S/N on a Hi-Fi VHS or S-VHS deck). Next, I also made a copy on 3/4", through the regular distribution amps, i.e. I had additional gunky circuitry in the signal path over the S-VHS dub. Comparing my S-VHS dub to the 3/4" dub requires no special talent. The S-VHS copy is *much* noisier than the 3/4" copy. And recall, this is conventional 3/4", *not* the wider bandwidth SP format. The S-VHS is still quite good to look at, but not nearly as nice as the 3/4". If you are accustomed to viewing NTSC only on VHS tape, you would not be upset, but if you are accustomed to viewing NTSC as it exists in the mastering process, you would mourn the loss. However, it is true that 3/4" is no longer the best way to capture video. In order of increasing quality, the competition comes from Hi-8, Betacam SP, D-2 digital, and D-1 digital. Betacam SP is a "component" video system, recording Y, R-Y, and B-Y signals, rather than Y and C. D-2 is a "composite" (Y/C) digital format, and D-1 is the "component" (Y, R-Y, B-Y) digital format. D-2 decks are now available for less than $50k, and the prices are comming down. Since a good 3/4" machine costs from $10k to $30k, this isn't such a large price difference (reference: the Sony BVU-850 and BVU-870). Note, however, that while the digital formats offer zero generation loss, the digital format has inherrent quantization, which can be quite noticable in the D-2 format. 8 bit samples are just not wide enough; we need 10 or 12 bits to satisfy the discerning eye, even with dither used to prepare the original signal. Mind you, I'm no big fan of NTSC -- it is a very limiting, low quality video format. But, it has supplanted 16mm film (a superior quality format) as the standard way of communicating scientific results, so one might as well understand it, and master it. NTSC, when done well, can be a very striking and effective way of communicating. AN EXAMPLE OF MAKING A VIDEO Just to give you a quick idea of how I created my recent video: Images were created in RGB at 1280x960, and decimated using a 5x5 filter kernel down to 640x480. These RGB images were converted to Y/U/V format, the U and V data was low-pass filtered to meet NTSC bandwidth limitations, and the YUV data was transmitted over Ethernet to an Abekas A60 digital video disk ($60k), where they were stored in D-1 digital video format. When an entire segment of video had been loaded onto the Abekas, it was played back in real time, under control of an FCC-approved broadcast quality sync generator. The RGB outputs of the Abekas were sent to a Faroudja CTE-1 RGB->NTSC encoder ($7k), and the composite NTSC was recorded on a Sony BVU-870 3/4" VTR ($30k) with SMPTE time code generator. Real-time sequences were captured from an SGI-4D/240 GTX ($100k) with CG3 board ($7k?). The sync pulses were fed into the CG3. With the SGI in NTSC mode, genlocked to "house sync", it provided RS-170 RGB, which was fed into the Faroudja encoder, and the composite NTSC was recorded on the BVU-870. I note in passing that the CG3 still does not succeed in producing broadcast quality video, although it is far better than earlier attempts. Unlike earlier SGI CG boards, the output of the CG3 is sufficiently good that it can actually be used, if you don't mind "tweeking" the Faroudja into forgiving SGI's sins. Each segment went onto a separate 3/4" cartridge, and all were indexed event-by-event according to the time codes. This allowed the script to call for segments to be assembled with individual frame accuracy. (Allow me to make a BIG PLUG for the use of SMPTE time code on your video tapes. It makes the editing process vastly easier, and more accurate). All the 3/4" original tapes went into a big box, and I took it all to our 1" edit suite. The original tapes were read on a pair of Sony 3/4" machines (VO series), stablized by a time base corrector (TBC), routed through an edit controller with Ampex effects box (for fades and wipes) and Ampex ADO (for inset screens, flip-away effects, and "MTV" style rotating and tumbling images), after which each finished sequence was recorded on an Ampex 1" machine ($50k). After the video was complete, I added the music to Channel 1, going from Compact Disc to the 1" machine via a sound board, where the mix to mono was accomplished. Then, I recorded the narration in a sound booth, onto a 3/4" tape. Using the same edit system, the narration track was copied onto the 1" machine (although I'm proud to say that in 8.5 minutes of narration, I only made one "flub" that we had to splice in a replacement for. This was done using the edit system, just like editing video). At this point, the master tape was finished. Dubs for the final distribution were made from the 1" master onto 3/4" and VHS, through a 1->12 video and audio distribution amplifiers. Thus, the distribution copies delivered to the clients are only 3rd generation. Since the trip from 3/4" to 1" is "nearly lossless", the distribution copies have excellent quality. Given that, in my instance, distribution is done on 3/4" and VHS, increasing the quality of the intermediate steps is not likely to make a noticable improvement in the quality of the product. The 1" master would be completely suitable for broadcast. Even if you don't have the financial resources to engage the services of a 1" edit suite for your video productions (not that it is very expensive), and you do all your production and editing in 3/4", you will get a high quality result. Certainly better than anything you could produce on Super-VHS. FRAME AT A TIME In the past, I have also done quite a bit of frame-at-a-time recording of video, using the Sony 3/4" machines and a Lyon-Lamb VAS-4 VTR controller. This produces results of equal quality (all other elements being equal), but takes longer. It is also more of a bother to find and fix botched frames when recording this way. It also requires a much smaller investment in equipment! ($5k for Lyon-Lamb Mini-VAS, plus a VTR). LASER DISCS On a related topic that I won't bore you with this evening, I recently investigated the quality of LaserDisc players. They are much better than consumer videotape (even Beta), but 3/4" tape does even better. The machines evaluated were the Pioneer CLD-91 ($2000) and Pioneer CLS-S2 ($3500) [the worlds finest Laser Disc player at the moment]. SUMMARY *) 3/4" videotape is no longer the best thing around. *) 3/4" can be used to produce excellent results. *) 3/4" beats S-VHS every time (video, not audio). *) 3/4" is mature, and not too expensive. Many alternatives exist, and lots of good stuff is happening. Keep your eyes on Betacam SP and D-2. If you have (or are) a good video engineer, there are lots of alternatives. If you don't have access to a good video engineer, the business of video recording still has a lot of pitfalls, and it will pay to be very conservative. Be suspicious. Also, most large universities and companies have a TV studio, and there are many commercial firms in the business. Strongly consider using them to assist with your post-production needs. The aid of a professional video editor (person) can greatly increase the quality of your result; knowing when and how to use fades, wipes, etc, is more of an art than a science. Rates are generally in the $100/hour to $200/hour range, and often less. Best, -Mike Muuss PS: In case you are interested, all the image generation software, image filtering software, Lyon-Lamb controller software, etc are all included as a small, but significant, part of the BRL-CAD Package, which we make available for free. See SGI's software partners catalog for details, or send E-mail.
dave@imax.com (Dave Martindale) (11/23/90)
In article <14556@smoke.brl.mil> mike@brl.mil writes: [ lots of useful and interesting stuff about video recording. However, there were a few errors I'd like to correct.] >Regular 3/4" has about 260 lines of resolution >(e.g. about 520 pixels/scanline), while SP-format 3/4" has about >360 lines, or about 720 pixels/scanline. Video resolution is given in lines, not line pairs. 2 lines == 1 line pair == 2 pixels. So a vertical resolution of 485 lines is really just 485 pixels. Horizontal resolution is a bit odd, since it is usually specified as "lines per picture height" rather than "lines per picture width". Because of the 4:3 aspect ratio, a horizontal resolution figure of "330 lines per picture height" is really "440 lines per picture width", or 440 pixels, or 220 line pairs. Even when you see resolution quoted as "330 lines" with no other qualifiers, they really mean 440 pixels H resolution. This is because vertical resolution should always be the same with a properly-interlaced TV, so nobody quotes it, while horizontal resolution depends on bandwidth which does vary. And because H resolution is quoted in terms of picture height, not picture width, even when only H resolution is being given. This is all a bit weird for those of us used to thinking in pixels. >The NTSC format is >bandwidth limited by law to have no more than 720 pixels/scanline, >so this is as good as it gets in real NTSC. An NTSC signal is limited to 4.2 MHz bandwidth when broadcast. This is equivalent to just about "330 lines" or 440 pixels (horizontal). More pixels than that doesn't gain you any resolution if it's going to be broadcast. Some frame buffers have 512 because it's a convenient number (power of 2), some use 640-650 because it's a convenient number (square pixels), but they won't give any more resolution when broadcast. 720 pixels is what is used by D-1 digital recorders; that's just what the width comes out to with a 13.5 MHz sampling rate. This probably allows them to have about a 6 MHz bandwidth, but anything beyond 4.2 will be lost on broadcast (but is useful for multi-generation processing). All of the above figures are for luminance bandwidth only. Colour bandwidth is always less. D-1 digital recorders give half as much bandwidth for the two colour components as the luminance gets, and everything else is worse. The NTSC standard itself specifies about 500 kHz bandwidth for the Q colour channel, which is equivalent to about 53 pixels horizontally. Yes, yechh. That's why colours smear well beyond the boundaries of the pixels that are supposed to be coloured. The I channel gets somewhat more than twice that, but most (or all) consumer TV sets discard it, limiting both I and Q to 500 kHz. Dave Martindale