Help - need information on HDV

[TingSern Wong]

I understand that HDV is derived from MPEG2 datastream. If I understand MPEG2 compression, it compresses individual frames as groups of pictures (GOP).

Question - can I access individual frames in HDV datastream (like editing today's standard definition DV) without resorting to interpolation?

I understand that today's SD is taking one frame, compressing it using MJPEG, and writing that one frame to tape. Hence, at post editing, I can get one frame of the scene at anytime.

However, HDV is MPEG2 compressed. How does the editing hardware/software be able to access one frame at a time, if the individual frames are grouped by MPEG2 and compressed, and written out as one GOP (say 15 frames per GOP)?

Is my understanding flawed here? Thank you.

[Mark Grant]

Editing HDV is just like editing any other kind of video, except that the computer needs to be fast enough to decode up to 15 frames of video to get to the frame you want to view. If you want an easier life, you can always transcode to a different format like DVCPRO-HD or Avid's DNxHD which compresses each frame individually but takes up a lot more disk space.

For single-stream editing and real-time effects (e.g. color correction) my P4-3GHz is fast enough, but it can't manage multi-stream real-time effects like dissolves (CPU is too slow for dual-stream effects in HDV, and the hard disks are too slow for dual-stream effects in DNxHD-185).

[TingSern Wong]

Does this mean that HDV is "less precise" than standard DV? If at acquisition time, 15 individual frames are merged into one "logical" frame (GOP) by MPEG2 compression, it won't be possible to reverse the process. Even interpolation has its limits. To my mind, this means a coarseness of about 0.5 seconds.

Transcoding means more time on the workflow. What I am trying to find out is - does anybody knows what MPEG2 parameters are used for encoding HDV? Is it 15 frames per GOP or 1 frame per GOP? MPEG2 allows 1 frame per GOP and upwards.

Pure CPU power has its limits as well, like what Mark found out. I will look into hardware assist later on (like one of those Canopus HDV editing cards).

[Barry Green]

Nothing is "merged", and there's no "interpolation".

As a basic explanation, MPEG-2 works by encoding the changes between frames. Each frame is encoded individually, but those individual frames are dependent on what's gone on before.

So you'll have I-frames, which are basically the equivalent of how DV or MJPG compresses their data -- I-frames stand alone. They can be completely reconstructed by using only the data that exists within that frame.

Then you have the rest of the frames in the GOP. Those frames are dependent upon the changes from frame to frame. Typically, very little will change from one frame to the next, and that lets MPEG get its great efficiency. Imagine a talking-head interview -- everything in the background will be exactly the same, from one frame to the next. Why re-compress all that data? MPEG doesn't -- it just points back to the prior frame and says "use that background", and then it adds unique compressed data for the elements of the frame that did change. What if you were shooting a still frame of a resolution chart, for example? Where absolutely nothing changes? MPEG can encode the entire frame in just one flag: a flag that says "repeat the last frame".

When it comes time to uncompress, uncompressing can be very slow by comparison. Because intra-frame-only compression schemes (like DV) contain all the data necessary to reconstruct an entire frame, all you have to do is supply the compressor with a pointer to the compressed data, and away it goes.

But with MPEG-2, it's not so easy. Every frame (except the I-frame) is dependent on every other frame. So if you want to decompress the 14th frame of a 15-frame GOP, the codec will have to decompress the I-frame, and then every frame after the I-frame until it gets to the frame you want. That means that uncompressing a particular frame could take 14x as long, under long-GOP MPEG, as it would with an intra-frame codec like DV (or DigiBeta or DVCPRO-HD or MJPG or most any other compression scheme, even including I-frame-only MPEG-IMX).

Now, as far as interpolation -- here's where things get tricky with MPEG. Sure everything's fine, and quite efficient, as long as not very much changes from frame to frame. But what happens if a *lot* changes from frame to frame? MPEG loses all its efficiency under those circumstances, and the results can get really ugly, really quick. There's a fixed bandwidth of compression available to each GOP -- with Sony HDV, it's about 12.5 megabits per 15-frame GOP. If too much changes from frame to frame, the compressor will not be able to resolve all the changes. The result is excessively-compressed frames; you get blockiness and macroblocking and mosquito noise and low resolved resolution. It affects the entire GOP and can make for some unpredictable, and ugly, footage (and it's one reason why some of us consider low-bitrate MPEG "not trustworthy" for professional applications). Such situations can never arise with intra-frame compression schemes like DV or the rest -- they compress each and every frame individually, using the full bandwidth per frame, and are not affected by what happens from frame to frame. The results are completely predictable and are never influenced by the type of material you're shooting. The same cannot be said of MPEG-2; its results are very much dependent on what you're shooting, and sometimes it resolves wonderfully, and sometimes it's awful.

I put together a page to show the comparison of an HD image compressed by DVCPRO-HD (a high-def intra-frame-only codec) vs. MPEG-2 HDV under "best-case" and "worst-case" scenarios; the "worst-case" shows the results of a completely bit-starved situation where there's excessive change between frames.

Here's the 1080i page:
http://www.icexpo.com/1080vs1080.PNG

Here's the 720p page:
http://www.icexpo.com/720vs720.png

[Stu Holmes]

that's a nice post Barry and sums up pretty well how MPEG2 works and it's advantages and disadvantages.

Can i ask - do you think it's likely that at some point in the next few years, the companies that agreed the HDV spec will come up with HDV2 ? with maybe 50Mbits/second or something ? I think the problem is storing an hour of data on a MiniDV tape. Either you sacrific quality or you sacrifice recording time.

Also, related to the above, do you think that an HD equivalent of DVCAM will be introduced? (or is there one already?!) I'd be very happy only to have 30mins. per tape if it meant less compression and therefore fewer image problems of the sort that MPEG2 currently struggles with.

thanks in advance

[Boyd Ostroff]

Originally Posted by Stu Holmes

I'd be very happy only to have 30mins. per tape if it meant less compression and therefore fewer image problems

Just looking at DVCAM, the compression is exactly the same - the same data is written to tape. But it's done in a way that makes it less prone to dropouts. So, all things being equal, "HDVCAM" might be less susceptible to recording glitches but not less compressed.

Also, I'm wondering if the issue isn't the amount of time they can fit on tape, but the amount of processing needed to handle a bigger data stream while keeping the camera cost down?

[Barry Green]

Originally Posted by Stu Holmes

Can i ask - do you think it's likely that at some point in the next few years, the companies that agreed the HDV spec will come up with HDV2 ? with maybe 50Mbits/second or something ?

Yes and no.

Yes, in that JVC and Sony are already developing something on their new cameras that will record 1080i in a fatter bandwidth.

But no, because the basic premise of your question is flawed. It's flawed from this section: "the companies that agreed the HDV spec ". No companies actually "agreed" on anything (regardless of what the press release said). I think it was more of a "truce" than a common format definition. JVC and Sony are both listed in the HDV coalition, and they both produce HDV equipment that carries the "HDV" logo, yet their formats are almost completely and thoroughly incompatible and share nothing. No camera now, nor announced in the future, shares both JVC and Sony HDV. JVC's implementation actually pre-dates HDV. JVC produced their own high-def MPEG-2 format six months before there was an announcement for an "HDV standard".

I think the way it went down is that JVC developed their own, and Sony developed their own. Sony was going to announce it as "HDV", and JVC begged 'em to include their implementation under the same banner. Whether that was to give JVC's format "legs", or to have Sony's format look like it was part of a wider, all-inclusive "standard", is debatable. What it's done is serve to confuse the issue significantly.

So, to further confuse the issue, Sony is going to be introducing a new HDV-style format for their forthcoming XDCAM HD, which we'll probably hear more about at IBC. It'll likely be 36 megabits per second, and may be backward-compatible with 1080i HDV, but almost certainly will have no compatibility with JVC HDV at all. And further, because it's disc-based, it rejects one of the basic tenets of HDV, which is "hd recording on a DV tape", so it may not carry any sort of "HDV" identification. But from what I've heard teased, it may have three modes of recording, being 18, 25, and 36 megabits; the 25-megabit version may indeed be the same as Sony 1080i HDV.

And JVC is going to develop their own 1080i MPEG-2 format, which will have nothing in common with Sony's HDV format at all. Won't share the name, won't be the same pixel dimensions, won't be the same color sampling, won't have anything at all in common with Sony's, except that it'll also be MPEG-2. JVC seems to be swinging for the fence though -- they're not limited by bandwidth (either by tape or optical disc) because their format will only record straight to hard disk. So they're going to make it 1920x1080 (no 1440x1080 subsampling!) at 4:2:2, with the ability to record 1080i and 1080p. Don't have a clue as to what the bitrate will be, but since it's going direct to hard disk (and only to hard disk) they don't have to be nearly so concerned about data rates as Sony does with their limited-speed XDCAM disc, or the limited-speed tape transports.

So yes there will be better MPEG-2 HD formats. No, it won't likely be under the banner of HDV.

And who knows if Canon or Sharp (the other HDV "coalition" members) will ever get on the bandwagon and produce any high-def cameras under the "HDV" banner...

[Mark Grant]

So, all things being equal, "HDVCAM" might be less susceptible to recording glitches but not less compressed.

I really wish the Z1 could record HDV at DVCAM speeds... I guess that would remove one of the reasons (dropouts) for people to buy more expensive HD cameras instead.

[John McCully]

[QUOTE=Barry Green]results are very much dependent on what you're shooting, and sometimes it resolves wonderfully, and sometimes it's awful.

My experience:)

I have been using the HDR-FX1 shooting HDV since soon after the release of this camera and with the assistance of Cineform’s excellent product Connect HD I have no difficulty whatsoever in ‘resolving’ individual frames most times wonderfully and on the very odd occasion not quite so good. Yes, a little more time in the workflow but the results are hugely worth it for me. Yesterday I accompanied a friend in his Safari convertible, build to go almost anywhere; the suspension does not deserve to be called a suspension; even on a smooth road riding in this ‘thing’ is a bone jarring experience, and to add injury, serious bruises, to insult, people looking at me as if I’m nuts, we drove around the cobblestone streets of a colonial town in central Mexico while I did my best holding the camera so as to minimize the violence of the movement. I shot in auto everything which will no doubt make some of my tecky friends wince. But hey, given the circumstances… Well, when I put the Cineform converted files on the timeline in Vegas 6 most of the individual frames looked amazingly good and some, very few, looked a bit ‘interlaced’, if you know what I mean.

‘Sometimes this’ and ‘sometimes that’ implies fifty fifty, and no doubt Barry did not mean to imply that shooting HDV with this device results in a fifty/fifty wonderful/awful outcome at the individual frame level, or at any level for that matter.

My background is photography and I routinely select individual frames from the Cineform file on the Vegas 6 timeline and save. I also not infrequently bring the rendered m2t file into Vegas 6 and export individual frames.

Notwithstanding the somewhat complex technology behind HDV, for me I should add, in my experience most of the time, I repeat, most of the time, the individual frames look very good indeed. Whether it’s ‘less precise’ than standard DV (my PDX10 in 16:9) or not is, I suggest, a mute point. The exported individual selected frames, absolutely most of the time, look way better to my eyes when viewed on my Dell 20” LCD monitor.

I trust my comments are helpful.

[TingSern Wong]

Hi all,

Thank you to all the useful comments. Especially Barry's link to the png file comparing DVCPRO-HD and HDV .... very, very convincing. These are things "mother don't tell you" at all. Looking at the grossy brochures of Sony, and fhe promotional video of HDV, you can't tell what lies under the hood.

My hat goes off to all the folks out there for making such a complicated subject simple to understand. Once again, thank you.

To John McCully,
Colonial town in Central Mexico with cobbed street - hmmm, that sounds very much like this place I visited when I was climbing volcanoes a few years ago.
Puebla, right?

TS

[John McCully]

Hey TS

Actually no; not Puebla. It’s San Miguel de Allende, but I have visited Puebla. Very nice, but the San Miguel de Allende cobbles are tougher, harder, more challenging…

Climbing volcanoes…wow. I’m a lazy bugger actually. I drove to the top of an almost 10,000 ft high volcano in Costa Rica a few years ago. Good road. Splendid view.

But yes, Mexico; wonderful country; I love it.

Mother…well, she was not that withholding; not that I listened – perhaps you have an opportunity there. Suggestion: don’t be too concerned with what lies, and blatantly lies again, under the hood. It can get greasy, dirty, and quite boring down deep in the mechanics. Forget the grossy brochures, forget Sony, enjoy the promo video as a production in and of itself, but little more even if you like it. Forget the very very convincing png file, other than for curiosity. I’m not questioning the authenticity of the png file but I find the relevance somewhat fleeting.

I like to pay attention to real production stuff, what’s in front of my face, on the screen. . For my miserable hard-won Pesos HDV shot with the HDR-FX1 is way cool to work with.

I’m glad you understand it.

Salud

John