HDV and the Sony FX1
Posted Feb 15, 2005

After testing Sony's HDR-FX1, I'm an HDV believer. It's not the second coming of DV, but it can be extremely useful in a number of circumstances, including when you need to down-sample the results to SD formats.

Conversation between Stephen Nathans, editor and myself, December 20, 2004:
Nathans:
So, now that you've spent more time testing with HDV, how do you like the Sony HDR-FX1?
Me: It's an awesome camcorder. The camcorder to buy in 2005.
Nathans: How does HDV look?
Me: Great, surprisingly good. Makes me rethink all the negatives I've said in the past about MPEG-2 as an acquisition format.
Nathans: So, did you shoot any real events in HDV?
Me: Are you crazy?

It was right at that moment I knew that the nice, pleasant, straightforward review of the FX1 was about to morph into a complicated, frustrating, and potentially embarrassing first look at the HDV format itself. In my view, before the conversation, I could recommend the FX1 wholeheartedly simply for price and DV performance, with HDV a nice bonus if HD-DVD or Blu-ray ever came around.

From Steve's perspective, given the claims that video shot in HDV could ultimately be output in standard definition (SD)-resolution formats at higher quality than original DV footage shot at those resolutions--with zoomed and re-framed shots, for example--we needed to dig a bit deeper. Humming to myself the immortal Mellencamp lyrics "I fight authority, authority always wins," I said, "Sure Steve, I'll find my shovel."

Let me explain my reluctance to shoot in HDV. My first experience with technology was producing PC Fax boards back in 1995 when sending a fax from your PC was pretty compelling. My company's engineers were brilliant, ramblin' wreck Georgia Tech grads, each and every one--as the fight song goes--"a hell of an engineer."

We purchased the facsimile specification and conformed our technology precisely to the spec. In-house testing on our single facsimile machine was flawless, and we expected a problem-free release, since, of course, all other facsimile machines out there were similarly synched to the standard.

Long story short, our "JT Fax" worked with about 65% of the installed base of facsimile machines upon its release, and left us deluged with support calls. The lesson learned was that there's a long way between technology that works in the lab and technology you trust to ship to your customers and family (can you say "DVD+R DL," boys and girls?).

So, while I'm an unabashed technology cheerleader, I'm from Missouri when it comes to actually deploying new technology on a shoot. After all, during my time with the FX1, exactly whom did I want to test HDV acquisition on? The tae kwon do instructor when shooting his seven-year old daughter's demo? My wife for her ballet training DVD or class exhibition? Taimo Toomast, international opera star, for his concert? The VFW troop for their Veteran's Day show?

In the end, I copped out and performed these HDV acquisition tests on experimental, testing-only footage, not footage I was hoping to ship to customers.

My conclusion? I'm an HDV believer. It's not the second coming of DV, but it can be extremely useful in a number of circumstances, including when you need to down-sample the results to SD formats. I spent most of my time looking at 16:9 output, since that mirrors the 16:9 footage you get with the HDV. But it should also be possible to get good results even at 4:3 ratios, at least from a technology perspective.

Technology Background
Let's start with some background. First, though HDV video has a pixel resolution of 1440x1080, during display, the codec zooms the horizontal pixels out to 1920, providing an output resolution of 1920x1080--which, if you do the math, is an aspect ratio of 16:9. HDV is not unique in this regard; most HD formats don't store the entire HD frame, pixel for pixel.

Instead, they capture a subset of the image in pixels, and usually subdivide the information stored even further by dropping chrominance or luminance information. For more on this, see http://www.broadcastpapers.com/dcinema/PinnacleCinewave06.htm and http://www.hdforindies.com/2004/11/video-cameras-vs-videotape-image.

Note, however, one critical difference between HDV and other HD formats. To reduce the capture bandwidth so the stream can be captured onto DV tapes, HDV camcorders store video in MPEG-2 format. In contrast, most other HD formats, which use much larger tapes, store the video either uncompressed, or using a Motion JPEG-like compression scheme (akin to DV's 5:1 Motion-JPEG compression). While DV/HDV share a bitrate of 25Mbps, Panasonic's DVCPro-HD has a bitrate of 100Mbps.

It's also worth noting that downconversion from HD to SD is a fact of life in the broadcast world, and far from unique to the HDV-to-DVD issue. Many of today's television shows are shot in HD and distributed in SD.

Sure, the networks are using much more expensive cameras with better glass, less compressed high-bandwidth formats, and usually (but not always) more sophisticated editing equipment. But in essence, the workflow is the same. Specifically, they're acquiring a 16:9 signal, usually either 1280x720 (720p) or 1920x1080 (1080i) and then outputting in both HD and 720x480 SD with an aspect ratio of 4:3.

Interestingly, when it comes to shooting in HD and outputting in SD, the technology part is easy, and will become easier, while the artistic part is complicated, and will remain so or get worse. For example, graphics optimized for 16:9 typically don't work at 4:3, and the optimal shot framing for 16:9 output may be totally inappropriate if the outer edges are chopped off for 4:3 distribution. These issues and more are discussed at http://www.digitaltelevision.com/publish/dtvbook/ch4.shtml. If you're thinking about shooting in HDV for 4:3 SD output, you should start your research here.

All of this is a long way of saying that what we're trying to do--shoot in HD and output in SD--is being done every day, by all the networks, using cameras that are similar to the FX1 in terms of storage resolutions, if not storage formats. Once you get your brain around the fact that SD output from HD input is a daily occurrence, the issue subtly shifts from "Can I do it?" to "How do I do it?" and "How good will it look?"

Let's take those questions in reverse order.

Test 1: Conservative
The first question I wanted to answer is how the quality of HDV video downsampled to DV 16:9 output compares to video originally shot in 16:9 DV mode. To test this, I shot footage with the FX1 in HDV mode and captured the video using CineForm's Connect HD utility, which converts HD to CineForm's AVI codec for importing into Premiere Pro.

I shot the same footage with the Canon XL2, a comparably priced and positioned 3CCD DV camcorder (see "Who Will Wear the New 3-CCD Camcorder Crown?," January 2005, Who Will Wear the New 3-CCD Camcorder Crown?) in 16:9 DV mode, and captured it directly from within Premiere Pro. To level the format playing field, I output both test streams in MPEG-2 format at 6Mbps using a Premiere Pro DVD preset, and compared that output.

In the most conservative test, the results are very close. While the FX1 down-sampled image looks a bit smoother, the DV camcorder image looks a bit sharper and clearer. In my view, however, the difference is so small that viewers wouldn't notice the lower quality of the FX1 unless they looked at the images side by side.

I included the original FX1 image to show the clarity of the captured video, which is simply outstanding. Clearly, the MPEG-2 compression used for HDV is not introducing significant artifacts.

Test 2: Zooming In
My first test is a worst-case for HDV, since it does nothing to leverage the higher resolution of the captured image. For example, many HDV proponents claim that its higher resolution pays off when you zoom into the image for additional detail.

This capability is very attractive in one-person, two-camera shoots, where one camera is set to capture the entire stage to ensure that no critical action is missed. For example, when shooting the opera concert with two cameras, I framed one mostly static camera to capture the entire stage, while using the other to zoom in for close-ups of the singers and piano player.

As it turned out, I framed the static camera too tightly on the pianist and the singer, and paid the price several times when the singer moved off camera. Since HDV captures at 1920x1080 resolution, and I was outputting to DV at 720x480, as the theory goes, I could have framed the stage much more loosely and then zoomed in without losing image quality because the extra resolution is there, waiting to be harvested. You can't do this with video shot in DV without losing quality, because the acquisition and output resolutions are the same.

To test this theory, I zoomed into the test videos with Premiere Pro's motion controls, essentially reframing the video from the HDV and DV camcorders around identical regions in the scenes. This would be analogous to a situation where you framed a stage loosely during shooting, then zoomed in during editing.

Here the results clearly favor the FX1, with the image on the right noticeably clearer. How did this translate in real world tests? Figure 6 shows some results with live footage that confirm the resolution chart tests: the FX1 images are noticeably sharper and produced significantly fewer artifacts.

For example, note the frizziness of the hair in the DV camcorder shot, and the macro blocks in the coat and yellow paper on the right. We also noted many fewer interlacing artifacts in the FX1 footage. Overall, when zooming into the video, the DV camcorder produced artifacts that would likely be noticed by viewers, even if there were no side-by-side comparisons with higher native-resolution source material.

What's interesting about these findings is the realization that when shooting in HDV for SD output, your image gets clearer as you zoom into the image during editing. This means there is no penalty for framing loosely during shooting, and no premium on zooming in to get close-ups, which sounds easy but can be hard to do, especially when your subject is moving around.

When shooting the opera singer, for example, many of my tightly framed close-ups were ruined when the singer stepped out of the framed shot (and we all know how fast opera singers move). If I were shooting in HDV, I could frame more loosely, and then zoom in later in post, actually improving the quality of the video as I zoomed in.

Traditionalists no doubt will argue that you should frame the shot you want, and pay attention to what you're doing. To a certain degree, I can't argue with that. "I can always fix it in post" is textbook videographer hubris. In real-world shoots, however--multiple-camera productions especially--you simply can't be everywhere at once, and shots can get lost or under-maximized. In these instances, the FX1 and HDV can fix many common shooting problems with no loss in visual quality.

Test 3: Chromakey
We tested chromakey for two reasons. First, we wondered if MPEG-2 artifacts in the stream would make a clean chromakey difficult. Second, since fine detail is important to successful chromakeying, we suspected that performing a chromakey at higher resolution might produce better results.

In the center is the HDV clip in CineForm AVI format at 1920x1040 resolution, which produced almost a perfect chromakey. On the right, the FX1 image output in MPEG-2 shows slight jaggies on the face, but nowhere near as much as the DV camcorder footage on the left. Thus it appears that HDV's extra resolution can boost chromakey quality as well.

Interestingly, this is one area where we had to experiment to produce optimal results. For example, you might notice that the shirt on the right looks gray, while the shirt in the middle, from the same source footage, looks blue. That's because we color-corrected the shirt before chromakeying for the full-resolution shot in the middle.

However, color-correcting before chromakeying and outputting at our 16:9 NTSC resolution made the entire image a bit blurry; the blurriness went away when we dropped the color correction. There probably is a workaround to get good color correction at 16:9, but since this wasn't the main goal of our story, we moved on. I'm guessing that the first time I do a real shoot in HDV I'll run into dozens of these idiosyncrasies, and so should you.

Note also that we performed these tests in Premiere Pro, which appears to perform the chromakey on the high-resolution input before scaling and compressing for SD output. While this seems logical, this workflow may vary by editor, so it's impossible to say that all editors will produce better results with HDV footage than with original DV footage.

Test 4: Low Light
In low-light tests, the DV camcorder produced a brighter image than the FX1 in both HDV and DV modes, though it does look a bit faded. Note that we tested the FX1 in a number of low-light situations, including a night shoot of a Christmas parade here in politically incorrect Galax, Virginia. For the test footage compared in this article we used the traditional Ozer Video Lab Lance-Armstrong-Picture-Under-the-Desk diagnostic exercise, with interesting results.

Overall, both the FX1 and the HDV video it produced performed remarkably well, degrading gracefully and exhibiting neither gain noise nor MPEG macro blocks. Thus it appears that there is nothing inherent to HDV (as manifested in Sony's FX1) that produces poor-quality images in low light.

Workflow
The state of HDV workflow remains primitive. None of the editors we tested could control the HDV camcorder, bucking bronco that it is, which left us to capture live like the good old analog days. Since our testing was mostly for SD output, we did not test writing back to the HDV camcorder, which is also performed in let 'er rip mode without machine control.

Beyond capture, we experienced aspect ratio issues with all SD video produced during our initial tests, which turned circles into ovals and squares into rectangles. With Pinnacle Liquid Edition, the solution was choosing the right presets for both the timeline sequences and the video clips; you can get it working if you make the right selections.

Premiere Pro was equally frustrating until we downloaded Aspect HD from CineForm, which provided both a capture utility and Premiere Pro project preset. Without the preset, you probably won't be able to get your projects working; you can probably get a free download at www.cineform.com. In addition, Adobe has announced a licensing deal for CineForm technology that should lead to a more seamless working environment in the short term.

By the time you read this story, editing in both Edition and Premiere Pro should be better-documented, and more of the workflow pieces securely in place. However, don't assume your editor can capture and/or edit HDV properly until specific support for the format is announced by the vendor.

We tested extensively on Edition, which edits natively in Long GOP MPEG-2, and in Premiere Pro using the CineForm AVI codec, which converts to a more easily edited format (see Stephen Nathans' "From HDV to DVD," http://www.eventdv.net/Articles/ReadArticle.aspx?ArticleID=9057). We'll withhold specific quality and performance results for later stories involving more direct comparison testing, when we'll be more confident that the working conditions we report at press time will look like the ones you'll find when the magazine comes out. That said, I found little difference in either quality or workflow with either system, though our tests were primitive, involving only trimming and chromakey, and differences may emerge under more comprehensive testing.

Once we had figured out the aspect ratio issues, both products worked pretty much as expected. Significantly, we edited both HDV and DV on both Edition and Premiere, and at least from a timeline responsiveness standpoint, found little difference between editing HDV and DV in either program.

Just for the record, we tested Premiere Pro/CineForm on a dual 3.06GHz Xeon Dell Precision 650 workstation while our Edition test station was a dual 3.6GHz Xeon Dell Precision 670 workstation. Both computers were running Windows XP SP2 with 2GB RAM.

Conclusion
Overall, the net/net on the FX1 is that it's an absolute darling of a camcorder. It performed wonderfully in the DV tests we reported in January, and the HDV output created in the testing for this article was remarkable. Once we got over our initial production hurdles with HDV, our tests confirmed many of the marketing claims made by camera and tools vendors.

Though our initial conclusion was to recommend the FX1 primarily as a DV camera, our recent experience with HDV production indicates that the time to move to HDV is now. While the difference won't be as dramatic as the sea change from Hi-8 to DV, the benefits in the short term are very real. When HD-DVD or Blu-ray comes along (which one doesn't matter so much as when), the advantages of HDV will be even greater. Even now, in the SD delivery era, if you're shooting for 16:9 output with the FX1, there really is no reason not to start shooting in HDV mode.

Companies Mentioned in this Article
Adobe Systems
www.adobe.com

Canon USA
www.usa.canon.com

CineForm
www.cineform.com

Pinnacle Systems
www.pinnaclesys.com

Sony Electronics
www.sonystyle.com