HD For Indies

For the zillionth time, somebody asked me about the quality differences between HDV and a traditional "real" HD camera. The person emailing was saying they could get "HD quality" out of the camera, so why pay 30 times as much? After explaining about the differences in the electronics, the dynamic range, the optics, the compression involved with each format, they asked a very obvious question:

How much difference does it make?

...and I didn't have a concrete answer for them.

So I decided to get one.

In short, this is a rough comparison between the two cameras, not a completely thorough analysis covering all situations. It's a bit of an ICBM vs a bottle rocket comparison, but hey, folks want to know...

The Cameras

Sony HDR-FX1

This is the $3500 new 1080i60 camera that has everyone excited. (I've written extensively about it before, use the Google bar at the top of this page and search for "FX1" for lots of coverage.) It uses 3 CCDs that are 960x1080, with the green pixels offset halfway between the red and blue, which buys you some more resolution. Sony claims an effective resolution of 1440x1080, I'm not sure it really sure it really delivers that much.

The lense used on this consumer camera pretty much moots that argument anyway - it's better than most prosumer lenses, but having the optical resolving power for 1440x1080 takes a LOT of precision. It's not here.

It records to the HDV format, which is a highly compressed MPEG-2 transport stream, with a GOP size of 15. GOP stands for Group of Pictures. This means that every 15th frame is a full update of the entire frame, and the next 15 essentially just record changes from the previous frame. So that if you lose a frame due to tape failure, you can lose up to 15 frames of video - 1/2 a second. Eek.

This data is recorded onto inexpensive DV format tapes. Sony makes special HDV tape that supposedly is more robust.

The original image, which if recorded in uncompressed RGB (red/green/blue) would be about 190 MB/sec. Instead you get about 3.125 MB/sec. This is because the video gets compressed using the MPEG-2 codec, which is usually used for delivery, when you want just enough visual data, instead of for acquisition, when you want as much as feasibly possible. This can cause problems later in post when you want to color correct the footage, when that "just enough" approach may be not quite enough.

Audio is compressed, unlike with DV, using MPEG 1, Layer 2 technology. This worries some of the audio guys I know. Even lowly DV records uncompressed audio.

So, in brief:

Camera: Sony HDR-FX1
Cost: $3500ish street
Image acquisition: 3 CCDs, 960x1080, arguably 1440x1080
Color Space: 8 bit 4:2:0 YCrCb
Recording media: small form factor DV tapes - if shooting serious stuff, consider the "special for HDV" tapes
Recording format: HDV - 1440x1080 recorded in 8 bit YCrCb, compressed with MPEG-2 TS using GOP of 15 to 3.125 MB/sec
Ingest: via FireWire, built into all modern Macs and included or available on PCs via PCI cards
and in this corner....

Sony F900/3

This is Sony's first camera that could shoot 24p, in addition to 25p, 30p, 50i and 60i. They revved the camera twice, so it's now the mark 3 camera. Ian Ellis of Texas High Def was kind enough to bring his over for testing of my new uncompressed HD to hard disk product I'm working on (but that's another story altogether).

The camera has 3 2.2 million pixel CCDs, capable of true 1920x1080 image acquisition. It can handle interchangeable lenses of extremely high quality (and cost). It can record 23.976,24, 25, 29.97, or 30 progressive frames per second, or 50 or 59.94 interlaced frames per second, all to the same HDCAM format tape. The HDCAM format drops the 10 bit, 1920x1080 signal down to 8 bit, 1440x1080 before compressing it with a JPEG-like compression algorithm down to about 22 MB/sec from what could have been a 240 MB/sec data stream.

What can I say? It's an extremely high quality professional tool with tons of controls made to be used under demanding professional environments. It also lists for over $100,000.

In short:
Camera: Sony F900/3
Cost: around $100,000 list price
Image acquisition: 3 2.2 million pixel CCDs, delivering 1920x1080 pixels at 10 bits/channel
Color Space: 10 bit 4:2:2 YCrCb (camera, available from HD-SDI tap), 8 bit 4:2:2 (internally 3:1:1) on HDCAM
Recording media: HDCAM tape
Recording format: HDCAM - 1440x1080 recorded in 8 bit YCrCb, compressed with using DCT like algorithms to 22.5 MB/sec
Ingest: via HD-SDI, available as a third party add-in board for Macs and PCs, starting at $600 (DeckLink HD) and up. $2500 or so will get you a nice card for Mac or PC.

OK, to the meat of this thing: In usage.

Comparison of footage

I still don't have an HDR-FX1 to play with, so I downloaded some of the raw .m2t files from the internet (you can get the same clips from here). This is the raw, source, directly from camera stuff you get when you copy the bits over FireWire from the camera. To be clear, this is EXACTLY what is recorded to tape from the camera. There are several outdoor shots - panning shots of a country club and it's lawn and trees, driving down the road in a car, stuff like that. I converted these files using LumiereHD version 1.5b9, which can now capture footage from the HDR-FX1 but can't (yet) get it back out to tape. I converted those source files to uncompressed QuickTime media so that there would be no image degradation and I could work with the footage in After Effects and Final Cut Pro.

I also have some footage from the Sony F900 that was shot from my front porch as we tested my new uncompressed HD to disk device I'm developing (more later on that). To be fair, I used the 8 bit footage captured from tape, not the source 10 bit footage I'd recorded to disk.

Both sets of footage were shot on sunny days outdoors, which somewhat levels the playing field.

HDR-FX1 1080i60 HDV Analysis

Here's my notes on what I discovered about HDV, really getting into the 1080i footage for the first time:

-viewed at a "pixel for pixel" level, it's not as good as good DV. By this I mean that if you were to look at a DV sized (720x480) chunk of image, it looks like consumer grade DV. Not really good DV, just good consumer DV.

-stepping back, looking at the full 1440x1080 image (stretched to 1920x1080), it's clearly miles better than DV in terms of detail and resolution. This makes sense - It's got 4 1/2 times as many pixels recorded to tape as DV does.

-footage of fire causes problems due to 4:2:0 issues. What's 4:2:0 mean? I'll explain 4:4:4 and work from there. It means for every group of 4 pixels, all 4 pixels' worth of information is recorded for the Y (luminance/brightness), Cr (component red), and Cb (component blue). For 4:2:2, all 4 pixels of Y are recorded, but only 2 pixels worth of color (the Cr & Cb). Each group of two pixels is averaged together to make only 2 values recorded. For DV, which is 4:1:1, only 1/4 as much color information is recorded as brightness information. HDV is a 4:2:0 codec, meaning for one frame 4:2:0 is recorded (4 of Y, 2 of Cr, none of Cb), and for the next frame, 4:0:2 is recorded (4 of Y again, but now none of Cr and 2 of Cb). The Cr and Cb take turns essentially every other frame. This trick usually works, except in video footage where the color changes a lot from frame to frame, such as footage of fire. You get streaks of color every other horizontal line, remnants of the color info from the previous frame. HDCAM never does this

-HDV looks very consumer quality, even when scaled down to standard definition. Technically the pixels might be in there, but what's making the pixels is miles apart from pro level gear. Pay 30 times more, it's going to look substantially better. Gee, big surprise...

-pushing the footage in post to color correct will break it pretty quickly. Some of my sample footage survived a push of +45 (using the Hue/Saturation controls in After Effects) and a gamma tweak of 1.15, but some didn't. Sky footage, for example, showed noticeable blocking in the raw footage, and when color corrected to give a pleasing saturation, it had EXTREMELY visible blocking, distractingly so. Think BAD digital cable.

-Mosquito noise was prevalent in any shot with a lot of fine detail. If you don't know what mosquito noise is in video, Google it. It's an artifact of compression, it shows up particularly clearly when you have fine detail next to a flat color, like small tree leaves against a sky. Once color corrected, it was very apparent.

-color correcting HDV makes almost decent looking stills, but you get all kinds of distracting popping in compression block artifacting once the footage is playing. Compression blocks are areas that the codec decides should be perfectly flat color...even when they shouldn't be. These blocks tend to dance around distractingly.

-Again, realize the compression techniques used for HDV are EXACTLY what's used for broadcasting digital cable. So when you're watching commercials on Deep Cable at 3am and see all kinds of wierdness (look at a defocused flat colored wall over a focused speaker's shoulder), that's the kind of stuff you get with HDV.

-lots of motion forces the codec to sacrifice quality in each frame. Travelling car footage, especially out the side window? LOTS of square edged artifacts.

-When compared to the F900 footage, HDV looks cheap and consumer level. There is no way you're going to be able to intercut these two cameras without jarring dissimilarities. If you're going for a verite look, a handheld camera look, OK, then HDV could be used in the same project.

-comparing similar shots - daylight shots zoomed in on distant leaves - the differences were extreme. My eyeball guess is that the HDCAM is capturing 2-3 times the sharpness of HDV. Between the superior lense, electronics, CCDs, and compression technology, HDV doesn't even pretend to be in the same ballgame as HDCAM.

-consider HDV a nice very high res consumer DV with this camera. I'd rank the HDR-FX1 about 1/2 of the way between a good DV camera (Canon XL1S or XL2, Sony VX2100) and the Panasonic Varicam. It's much better resolution than DV, but the clarity and color fidelity of the Varicam blows HDV away. But it's nowhere near the F900 in quality by any stretch. (Which shouldn't surprise anyone, anyway)

-tree & leaves, zoomed in on from 50 feet away, were still very very sharp in the F900 footage. Everything HDV was soft focus, but rough edged due to compression. The F900 had a smoothness that the HDV footage didn't touch.

-don't consider HDV a replacement, a challenge, a match, or a camera B for a Varicam, HDX400 (new Panasonic 1080i DVCPRO HD camera) or Sony F900. Consider it a really, really sharp DV.

-if you want to shoot on it because it's more resolution and detail than DV....well, kinda. Overall yes, but some of the motion artifacts will be shocking to DV folks down at the pixel for pixel level. But overall it's a nice prosumer camera and that's it.

If you've been dreaming that you could get a Sony HDV camera (even the upcoming improved model with XLR adaptors) and go out and "shoot a movie" with it with available light, you're going to be dissapointed. Perhaps future HDV cameras will have better optics and dynamic range, but the format will always be hobbled by the compression technology.

Compare it to Open Water, shot on good PAL DV cameras, and then verrrrry carefully posted. That with more resolution, but more compression artifacts.

I'll be very curious to see Panasonic's new $4000ish DVCPRO HD camera at NAB and see how it compares.

-mike