• Blogs
• Forums
• Tradeshows
• Partners
• White Papers
• Events

 

The HDV Sweet Spot

Sep 1, 2004 12:00 PM, By Steve Mullen

The second in a pair of articles about the future of HDV examines the inherent advantages of the 720p30 format.

Charts
Chart 1
Chart 2

Sidebars
HD on a CD-ROM

Last month we examined situations that may arise when the second wave of HDV equipment arrives. The coming 1080i camcorders, in particular those from Sony, have been the subject of intense speculation — especially their price and arrival date. The big impact of 1080i will, surprisingly, be felt when we edit 1080i.

NLE playback performance is based on the number of pixels in an MPEG-2 frame. Table 1 shows these values for 720p30 and 1080i30 video — and also for 720p60, were it to become available.

Simply put, with 1080i30 video, every MPEG-2 decode will require either nearly twice the computing power or nearly twice the time — or some increase in both. Do we have this much computer power or time?

Non-native 1080i HDV editing

Let's look first at the capture and demux phase of editing when an intermediate format is employed. According to CineForm, Aspect HD requires a 3.2GHz Pentium 4 to transcode 720p30 to Wavelet in realtime during capture. In theory, to transcode 1080i30 will require a computer almost twice as powerful. Thankfully, CineForm has implemented a double-buffer technology, so the transcode process can continue after the capture ends. Therefore, the total input process will be longer than it is now with 720p30.

Three points about non-native editing on a Mac. First, Apple's DVHScap has no timecode-based logging ability, so one typically captures an hour of HDV to disk. Second, transcoding to DVCPRO HD — as may be done with Lumière HD (www.lumierehd.com) or my HDpartner Pro software — will take twice as long. Third, transcoding to a proxy will also take twice as long. It's obvious, then, that any approach that requires you to transcode your raw 1080i tapes to another format is going to impact editing efficiency severely.

One alternative for Mac users is made possible by the new MPEG Streamclip utility (www.alfanet.it/squared5/mpegstreamclip.html). After capture by Apple's DVHScap, MPEG Streamclip quickly accomplishes demuxing. Demuxing is fast with either 720p or 1080i because no MPEG-2 decoding is involved. Then, using my new HDVideoSplicer software, you can quickly splice together an HD “selects reel” from your demuxed — but still MPEG-2 — source files. After all your source material has been winnowed down to the 720p or 1080i HDV clips you are likely to use, only these selections need to be transcoded to DVCPRO HD by HDVideo-Splicer. This “native pre-edit” approach can save time, especially if you do not have a powerful Mac.

Nevertheless, once you start editing 1080i DVCPRO HD, you'll still encounter the 1080 performance penalty. Because any intermediate format must have the same resolution as the original, it too will carry twice the pixels. So are we back to the days of non-realtime editing? Thankfully, no. I found that a dual 2GHz G5 can play four streams of 1080i30 DVCPRO HD simultaneously. (Likewise, according to CineForm, a 3.2GHz Pentium 4 can edit two to three steams of Wavelet in realtime.)

Table 1: Frame Sizes for 720p30 and 1080i30

	720p30	1080i30	720p60
Row Size	1,280	1,440	1,280
Column Size	720	1,080	720
Number of Pixels per Frame	921,600	1,555,200	921,600
Decode per 1/30 Second	921,600	1,555,200	1,843,200
Load vs. 720p30		1.7	2.0

Native 1080i HDV Editing

The other alternative is to edit MPEG-2 without the need to demux and transcode. This is possible with native HDV editors running under XP. (No native Mac NLE yet exists.) In the last few weeks, I've been running experiments using synthetic 1080i MPEG-2 with an NLE that edits native HDV. I compared 720p30 vs. 1080i30 performance using Vegas 5.0.

I used Vegas because it has a frames-per-second display that provides a digital readout while a timeline plays. I performed my tests on a 3.2GHz Pentium 4 with 1GB of RAM. I conducted nine simple MPEG-2 tests, the results summarized in Table 2. (The exported movies had a few short transitions and a few short video filters.)

HDV at 720p30 requires all the CPU power that's currently available, and it's obvious that the moment we begin to edit 1080i, we will need a computer two times more powerful. Without that much power, smooth playback and realtime effect previews may not be possible. At NAB 2004, the president of a company highly regarded for its codec development and NLE systems told me that editing 1080i30 would likely require a dual Xeon PC. Is there another way?

One potential solution is to use graphics processing unit (GPU) power rather than CPU power. Modern GPUs offer a hardware-based inverse discrete transform (iDCT) function that can offload much of the decoding computation from the CPU(s). This is accessed via the DxVA function in DirectX 9. Future NLEs might obtain the necessary computing power by using a GPU. (See “Home-Built HDV Editing,” p. 39.) Apple is following the same path by implementing Core Image/Video in the 10.4 Tiger version of OS X. At WWDC 2004, Steve Jobs explained, “…although CPU speed increases are slowing down, GPU speeds are increasing dramatically. Therefore, it makes sense to take advantage of that processing power to manipulate images and video.”

Over time computer systems will become more powerful, and native HDV editing software will become more sophisticated. So eventually we will be editing HDV just as we now edit DV.

The Progressive Advantage

There is another way to look at the problems of 1080i editing. That is to appreciate the “sweet spot” that 720p30 represents. Discounting has given the two shipping 720p30 camcorders a lower price than is expected for future 1080i HDV camcorders. However, the sweet spot involves more than price and availability.

The 1280×720 HDV format has the inherent advantage of progressive scanning. Progressive scanning yields an image with no interline flicker, no motion artifacts, and more efficient encoding that minimizes other types of artifacts. Not only does progressive scanning deliver high-quality motion video, but it is also equally superb whenever stills are obtained from video (for example, the picture on p. 19).

In fact, the only debate about progressive scanning is what frame rate it should be shot at. If you are going to film, 24fps or 25fps is the obvious choice. Although 25fps is not available from the current JVC HDV camcorders, one can hope the next generation of JVC prosumer camcorders will offer 720p25. By using PAL-to-film procedures, 25fps HD can be made to work even for those of us in an NTSC region.

Table 2: Performance Comparison

	720p29.97	1080i29.97
Timeline playback	29.97fps	4fps
Timeline with color correction filter	29.90fps	3.6fps
3-second cross-dissolve	7.6fps	2.9fps
15-second cross-dissolve	3.8fps	1.5fps
Export to HDV	5.25 minutes	16 minutes
Size of 3-minute HDV movie	412MB (~8.3GB/hour)	542MB (~12GB/hour)
Export to WM9 HD	45 minutes	132 minutes
Playback of WM9 on a PC	29.970fps	29.400fps
Playback of WM9 on an iMac	1/3fps	1/3fps
Playback of WM9 on a dual G5	29fps	3fps

Of course, many of us prefer the high temporal resolution of 60fps. Nevertheless, for the reasons presented in Part 1 of this story, I don't expect we'll get this option anytime soon. That means we'll need to continue to shoot 720p30.

Others like that the temporal resolution of 720p30 is far closer to 24fps than is 1080 interlaced at 60 fields per second. For these producers, today's 720p30 HDV offers a free film look.

Progressive scanning offers more real-world resolution than is measured by static resolution charts. A study by William E. Glenn showed that apparent progressive image resolution is almost twice that of interlaced image resolution for moving objects (“Understanding Camera Resolution,” Broadcast Engineering, August 1999). Nevertheless, subjectively higher resolution may not be convincing compared to the numerical superiority of 1080i. Let's look at 1080i in greater detail.

First we will look at vertical resolution. Because interlaced cameras use CCD row summation in order to suppress interline flicker, vertical resolution is effectively reduced by 25 percent — with NTSC, from 480 to 360 lines. Therefore, the 1080-line interlaced format will carry an image that has about 800 lines of vertical resolution — not much more than the 720 lines provided by 720p. (JVC single-CCD HDV camcorders provide a measured vertical resolution of only 650 lines.)

Turning to horizontal resolution, we can expect effective horizontal resolution to measure about 75 percent of the number of CCD columns. It is customary to employ CCDs that have at least the number of columns as the maximum number of columns supported by the video format. A 1440×1080 format (1080i HDV) should use chips with 1440 elements per row. Therefore, we can expect the effective horizontal resolution of three-chip camcorders to be about 1080 lines for 1080i and 960 lines for 720p. (JVC single-CCD HDV camcorders deliver a measured horizontal resolution of only 700 lines.) Thus, if we compare total resolution per frame, there is not the huge difference we may expect. Total effective resolution for 1080i is 864,000 pixels and 691,200 for 720p.

There is one more issue to examine. Will a data rate of 25Mbps be too low for 1080i MPEG-2? This, of course, recalls another, familiar question: “How can 19Mbps carry HD when we need 25Mbps just for NTSC DV?” Many of us answered the question about 19Mbps carrying HD by responding that ATSC HD broadcasts use only 19Mbps — so why shouldn't the same rate be adequate for HDV? The correct answer is far more complicated.

Very expensive, microwave-sized encoders encode broadcast HDTV. Moreover, broadcast MPEG-2 uses a long, 15-frame GOP. Today's HDV camcorders utilize a far less efficient six-frame GOP. (A short GOP does make editing easier, however.) Even with these advantages, we see MPEG-2 “blocking” artifacts (i.e., pixelization on rapidly moving objects) far too often during 1080i sports broadcasts.

Surprisingly, we do not see artifacts from 720p HDV. The reason is that 720p at 30fps places half the input load on an encoder as does ATSC 60fps 720p. Thus, 720p30 video encoded at 19Mbps looks as good as if we allowed broadcast 720p60 HDTV to have a 38Mbps bandwidth. No wonder 720p30 HDV looks so good! Thus, our flip answer about HDV was wrong.

When 1080i is fed into an HDV camcorder's encoder, two times more pixels are inputted, yet the encoder bandwidth increases by only 32 percent — from 19Mbps to 25Mbps. Will this increased data rate be adequate?

If the 1080i GOP is defined to be 15 frames, the extra efficiency from a long GOP, plus the increase in encoding bandwidth, might enable 1080i to look as good as 1080i ATSC broadcasts. Unfortunately, we may need to choose between two types of motion-induced artifacts when we work with either 720p30 (at 19Mbps) or 1080i30 (at 25Mbps) MPEG-2 video. With 1080i, rapidly moving objects may break up into small blocks — an MPEG-2 artifact. With 720p30, rapidly moving objects will show strobing — an eye-tracking artifact. Although strobing can be reduced by locking shutter speed to 1/60 second — likely to be easy to do with future 720p HDV camcorders — it can't be eliminated.

Interestingly, were 720p60 to arrive, at only 19Mbps, it too would potentially be burdened with MPEG-2 blocking on rapidly moving objects. The only solution to MPEG-2 pixelization is a significant increase in the encoding data rate.

You should now have a greater understanding of future HDV technology. And hopefully, you've also come to realize that the 720p30 HDV technology available right now represents a sweet spot for those wanting to work with HD at DV prices.

---

Sidebar

HD on a CD-ROM

Working with Vegas 5.0 gave me an opportunity to encode a 5.1-channel surround-sound HD timeline using Micro-soft's Windows Media 9 encoder. After creating a timeline, I issued File > Render As… then, under Save As, I set the Type to “Windows Media Video V9 (*.wmv).”

To check playback of the WM9 movie on a Mac, I had to transfer it to my two Macs from the PC. My first thought was to move it via WiFi. I decided against moving an HD movie via wireless because I expected it would take far too long. My next thought was to burn a DVD, but I didn't have a blank DVD+R. So, I popped in a CD and burned the 4.25-minute movie. After the burn, I checked how much space was free. It was only then I realized how tiny the HD movie was — only 222MB. A quick calculation shows 12 minutes of 720p30 HD with 5.1-channel sound can be stored on a CD-ROM. That's long enough for many purposes.

With a powerful computer and a fast CD-ROM, you can play a movie from the optical disk. The better option, however, is to copy the movie to the internal hard disk. My appreciation of WM9 encoding increased when discovered I could play my movie on the 1280×768-pixel screen of a 3lb. Sony VAIO PCG-TR3A notebook.

The 5.1-channel audio track of a WM9 movie is not encoded to AC-3 (Dolby Digital), so it requires a sound card or USB sound box with six channels of analog output. To output via a SP/DIF, the audio track must be encoded to AC-3. This can be done — in realtime during playback — using the “AC3filter” plug-in for the Windows Media 9 Player (http://ac3filter.sourceforge.net/do…r_1_01a_rc1.exe).

---

feedback

To comment on this article, email the Video Systems editorial staff at vsfeedback@primediabusiness.com.

© 2008 Penton Media, Inc.