Manufacturers are preparing now for consumer HDV with better CCDs and optics
Ever wonder why over the past year many new camcorders including Canon's Optura 10 and 20, Elura 60, 65, and 70, (and in Europe the PAL-based MVX150i, MVX200, MVX200i, and MVX250i) are using the same 1.33 megapixel CCD? Also Panasonic's popular PVDV73x, JVC's GR-DVP9, GR-D200, and Sharp's VL Z8H. This is not by chance. We have a whole host of new cameras with exactly the same resolution -- 1280x960 pixels. The same as is used in the GR-HD1.

So this offers an easily downsizable 4:3 aspect ratio for NTSC now since when you divide the vertical resolution by 2, you get 480 lines -- NTSC. But in the future these CCDs directly support the resolution used in the HDV spec -- 1280x720. Just don't use the top and bottom 120 pixels on the block.

So manufacturers are moving to utilize solely CCDs with this higher resolution to prepare for HDV. It will be their bread and butter over the next five years, so they had better be ready. I expect that the exact same 1/4" CCD part and optical paths in today's cameras will be used for the initial HDV consumer offerings. Enhanced CCDs will satisfy resolution enhancements for the pro-sumer crowd.

Even the GL2 was designed with HDV in mind since its still image capture is in the same ballpark as HDV. The whole optical path has been optimized for the emergence of HDV. Not that you could upgrade that model to support HDV, just that the same R&D effort doesn't have to be repeated now that they're designing a similar HDV unit.

Will we see 4:2:2 HDV? Not for the consumer, but that doesn't matter!

4:2:2 video has 2 color samples for every 4 luma samples. We won't see that kind of clarity hit the HDV format. The limitation is that recording twice the color information like this takes about twice the bandwidth, which takes us well over DV's 25 Mbps data rate. As well it is a more intense compression process, so to implement 4:2:2 would require almost double the horsepower for the encoding chips. No small feat for battery-conscious consumer gear.

In the long run, HDV is really trying to address the consumer market, not the pros. As such, it samples color with the same frequency as did the DV format. DV did 4:2:0 for PAL and 4:1:1 for NTSC, and these both equate to the same ratio of one color sample for every four luma samples.

Although we're not at 4:2:2, the thing we gain -- and the thing that makes HDV compelling in the near term to users of low-end pro gear -- is total resolution. So it's 4:2:0. If that's your only metric of quality, you're missing the boat. Compare overall just how much more chroma detail we're actually getting in contrast to the 4:2:2 NTSC of today!


For most, there is no real need to get bogged down worrying about 4:2:0 vs 4:2:2. We're still advancing by leaps and bounds both in resolution and color.

I'm sure as Steve said, future pro formats will address 4:2:2 recording. But for the moment, especially those of us on a budget or who are used to standard DV, HDV is pretty compelling indeed!

Consider also that if for some reason you're only interested in color quality, then if you divide the resolution on each axis exactly in half, a 720p HDV camera gives you 4:4:4 sampling, with 640x360 progressive video at 60fps. Amazing. And if 1080p cameras ever go mainstream, you'll have 720x540 at 30fps. You'd still have traces of what would then be 4x4 MPEG2 macroblock artifacts in high-motion sequences, but don't miss the fact that it's tons of color resolution for cheap.

I do long for the better color resolution that we will eventually see out of variants from the HDV trend. At the moment I'm also sold on HDV being an excellent near-term solution, even though nobody is yet extending the spec to offer 4:2:2. That's the real message I was trying to convey in my post. HDV is great as-is for now, and due to get better over time. I wanted to bring to light the fact that MPEG2 can do 4:2:2, so we've got quite a future in this format.

with 720p, basically you can scale down the image to get 640x360 at 4:4:4. In other words, use a good quality HDV camera as a high-quality source, and end up with fabulous widescreen NTSC. So the current JVC cams don't offer the best color... but with the (rumored) 3CCD camera that will show up at NAB this year, we've got a killer format we can use now to release high-quality material on standard DVDs, and capability to release the same material in the future on HD DVD.

in a few years it will end up being primarily a consumer-grade format, just like DV is today. At that point in time, I expect the pros to embrace better variants of HDV that use higher bitrates for 4:2:2 and 1080i recording. (1080i needs it more than 720p since it's more suceptible to macroblock artifacts in high-motion scenes.)

Who knows... Based on Steve Mullen's comments as of late, perhaps higher-end variants of HDV will be announced by Sony sooner rather than later! Would be nice to lay out the roadmap of HDV for people and show that it really has a strong future.

Start with 1280x720 and downsize by some whole number amount to achieve the least loss possible during the resize. I'm advocating dividing by exactly 2 here in each dimension. Now we aren't ending up with a full 720x480 resolution, so there still is a sacrifice in resolution compared with a true high-end NTSC cam. But overall you end up with some danged great 640x360 30 or 60fps progressive video. Better than DVD quality since after the resize it ends up being 4:4:4 sampling. It has still gone through the MPEG2 compression process, so you will have slight macroblock artifacts in high-motion scenes, but it's quite bearable. It could be stretched in width from 640 to 720 pixels to accomodate the NTSC aspect ratio, and put on a DVD with excellent results. When you divide both the vertical and horizontal resolution exactly in half, the result is that every 2x2 block of pixels from the original becomes one pixel in the result. Lots less luma noise since you're averaging four values together. And since color is sampled precisely once in every 2x2 block of pixels, each pixel in the result also ends up with its very own chroma info.

For a fun experiment, try doing this with today's GR-HD1 cam. You're averaging luma down 4:1 in the process, so much less of that camera's grain in those low-lit scenes. As well the camera already has somewhat washed out color, so it won't be outstanding, but still usable. You can start to appreciate where this format can take us with this kind of experiment. I imagine if you do that with a good 3CCD HDV cam then you should be able to get similar results to a high-end NTSC cam, but without spending that much money. Okay, you'd still lack a black balance adjustment, and some depth of field, and you're limited to letterboxed 640x360, but it's 4:4:4, and better than pro DV for the same price. Video you release now in high-quality NTSC can later be re-released in 720p.

For upcoming cams that do 60fps, you can use NTSC's interlacing to your advantage by taking out every even-numbered scan line on odd numbered frames, and vice-versa for even-numbered frames. Weave sequential frames of this "bob'd" video together, and you have true 60 field per second interlaced NTSC with great quality. All this pixel shifting madness is fairly easy to accomplish in the PC world using the freeware AVISynth program. To feed AVISynth, use the version of VirtualDub that reads MPEG2, and set it up as a frame server. The output then gets cranked through the AVISynth script, and finally put through a lossless compression like HuffYUV. Stellar NTSC results on a $10K budget, and HD leftovers for later. Now we just need someone to release a 720/60p 3CCD HDV camera!


Comparing HDV to Varicam
GR-HD1 actually averages 0.66 Mb per frame at 30fps. When 720/60p cameras come out (probably at NAB), they will be recording at 0.33 Mb per frame. And that's not megabyte -- that's megabit! Only 41KBytes per frame on average. Pretty amazing. So there's dramatically less data per frame being recorded with HDV.

These bitrate figures to not consider the hefty ECC that the DVCPROHD format uses. HDV, by comparison, doesn't chew up as much space for error recovery. It's destined to be a consumer format anyway, so although it's OK, it's not great at error recovery. Pro formats have to be rock-solid, and have better ECC. It doesn't affect picture quality at all unless there are problems with the tape, and it can often kick in and recover what would otherwise be lost footage.

"Can the cheap MPEG2 converters like the ones on HD10 really encode MPEG2 most efficiently, or do you normally end up with more or less all the frames having as much info as the I frames? "

They really do capitalize on the difference between frames rather than doing the frame-independent thing of DV. But it's not some ultra-great MPEG2 compressor like Hollywood uses to make DVDs or anything. There's better algorithms that could be used, but the problem is they chew up so danged much more computation that it's not worth it. The compression circuitry would be much larger, run hotter, and chew up way more battery power. As Moore's law makes these things faster and smaller though, we can expect HDV to eventually embrace 720/60p and 1080i formats targeted for the consumer.

"Is it possible that with a good MPEG2 encoder, lenses, etc. the quality of the HDV camcorder can be similar to that of the Varicam? "

That's the exciting part -- although we won't ever have 4:2:2 like Varicam does, so we're limited to half the color resolution right out off the bat, at least we have danged good luma resolution. Perfect for today's pro-sumer. Varicam will still look better, but overall HDV is pretty good.

"Is it true that the Varicam does not record full 1280 horizontal pixels, but just 960? "

For the luma, yes. Some corners had to be cut to fit the signal within 100Mbps for that format. See this analysis: http://jkor.com/peter/riddles.html

"How many pixels does HD10 actually record?"

In the 720p mode:
Luma: 1280x720
Chroma: 640x360
Both at 29.97fps

But bear in mind that the luma in the GR-HD1 is coming off just 1CCD. With 3CCDs we will obviously have the same exact resolution numbers, but picture quality will improve.

"With the HD10, if everything else was equal, could it have compression artifacts similar to Varicam? "

Compression artifacts are a little different between DV-related formats and MPEG2. Okay, so both use DCT, so they're both going to exhibit macroblock artifacts, there's still a difference. In high-motion scenes, MPEG2 falls short since (unless you're only using I-frames), it depends on inter-frame compression to do its job. Certain corners had to be cut when JVC built the small, hand-held, low-cost GR-HD1. The SuperEnc III chip that does the work is operating pretty much at its limits to produce 720p video at 30fps. The Varicam records double the color resolution, so already there's quite a bit more information stored in that format. And DVCPROHD enjoys no inter-frame dependency, so high-motion scenes are just as clear as stationary ones. Just like the cheap DV25 cams of today.

There are many little differences here in Varicam's favor. It's tough to draw a comparison between the comparatively low-end GR-HD1 with its MPEG2 encoding against such a high-end model. Hopefully this analysis helps, though.


Notes about the train sample from http://videosketch.myhome.cx/sl-sanka/victor/8630-ps-5137.wmv.
I think the leaves in the foreground are a more impressive display of resolution.


Some annoying characteristics of the JVC cam that poke through in this sample:


Some great things about the JVC cam that come out in this sample:


We've got a great start into what will be a tremendous format for the next decade.


"Can you compensate for the GR-HD1's washed out colors in post?"

I've only done initial rudimentary testing, and yes, you can bring up the color a bit. But you just can't fully escape that "washed out" look that it has. I don't know how to fully explain it. It definitely falls short of the stunning footage you can get out of today's average 3CCD unit.

But man, it does have some great resolution! Should have a black and white option in my opinion, since it is so danged good. A selection to throw out the color signal entirely and get the additional bandwidth to do flawless black and white filming.

I'm sure within a year we'll be able to "have our cake, and eat it too" as 3CCD units will start trickling into the market.


Lossless Video Compression
If you're on the PC, try the freeware HuffYUV. It does 8-bit, and is also a flawless compression. Only intraframe compression, so you can still do frame-accurate editing. Compresses to about half of the original size depending on source material, and it a VERY fast codec.

If you want a bit more compression and don't mind alpha code, talk to Michael Niedermayer about his FFV1 codec, which is currently under development. It may offer inter-frame compression and use arithmetic coding, which may end up offering up to about 5X compression depending on content.

Note that the other payware option mentioned, SheerVideo, will generally only compress to half the original size and not smaller. But it does have the benefit of handling 10-bit video.

It seems that the SheerVideo offering may have been adapted from HuffYUV. A bold statement for me to make, and one that impossible for me to fully substantiate, but based on the very similar options it provides and the fact that the HuffYUV source code has been freely available in the public domain for some time now, there is that possibility.


MPEG2 Compression:
"The 720p and 1080i HDV data rates exceed the data rate of HDTV broadcasts (19mbs - overhead) There seem to be very little arifacting on fast movement in HDTV broadcasts - just whip pans and the like. There should be even less with HDV. "

You're comparing apples and oranges here. Although broadcast HDTV does take less bandwidth, it's because there's incredibly more computation being done behind the scenes there. It takes exponentially more processing power to make good MPEG2 video as motion gets higher and higher. Most of the efficiency of the codec comes from determining the motion in the scene, and Hollywood doesn't mind taking whatever time it takes to get that part perfect. Professional HDTV programs often take many times longer to compress than their actual duration. For anything professional done in real-time, expensive codecs are used with powerful DSPs. $30,000-$100,000 rack-mounted units are used for this task. So in a lowly camcorder there just isn't the real estate or battery power to achieve that quality of compression. Instead of a more standard 15GOP structure, HDV uses a 6GOP structure which also sacrifices some possible bandwidth gains. The larger I-frames come along twice as frequently in HDV as other MPEG streams.

So with all these differing factors, you just can't make a good comparison between HDTV bandwidth and HDV. In my article on the GR-HD1 a year ago, I had compared them (see the chart about 1/3 of the way down). But I had thrown that out there because back then so many people were doubtful that MPEG2 would be able to effectively carry us into this next generation of product. Now that people generally realize that MPEG2 is viable, it's time to better clarify exactly what it can and can't do with it.


Ways to use HDV footage for computer-only playback
All the resolution choices offer some real possibilities when converting to a computer-only format, which is what I do exclusively with my footage. If I've done my math right, 720p will offer more resolution overall than 1080i:



FormatResolutionFramerateLuma samples per secondColor Samples per second
1080i1440 x 108029.97fps46,609,34411,652,336
720p1280 x 72059.94fps55,240,70413,810,176


The other perks with 720p when going to computer-based playback include no aspect ratio differences or interlacing to worry about. With 1080i, you have to first deinterlace, then stretch the 1440 pixel line to 1920 or squish the 1080 high video to 810 to get square pixels again to play back on a computer.

I just hope that the internal downsizing done in the camera from 1440 x 1080 native down to 1280 x 720 doesn't lose too much in the process. I'm a high-resolution and high-framerate junkie. I like video that looks as much like reality as possible. with an f1.4 maximum aperture lens, a 1/2" CCD camera has depth of field capability almost identical to a 2/3" CCD camera with an f2 MA lens. I got this info from the Panavision/New Zealand website: www.panavision.co.nz/main/kbase/reference/calcFOVform.asp

1440x1080i:
Stretch to 1920x1080 (resolution loss and hard to play back since it's so massive)
Deinterlace, then 1440x810p or 960x540p 60p, or 4:4:4 at 720x540p

Smooth Deinterlace to 1440x1080 60p, then resize to 960x540 60p to regain much of the lost resolution. Offers 194400 color samples per field, or 1 color for every 2.66 luma, or 0.375 luma samples for every color sample. Totally smooth, and very VERY good resolution tied into the native CCD resolution

1280x720p:
More color samples overall, 230400 instead of 194400.
Hosted by www.Geocities.ws

1 1