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

HDV in the Real World

Jul 1, 2005 12:00 PM, By Steve Mullen

A guide to the ways that current NLEs handle the extra pixels of HDV.

As you made your away around the nonlinear editor booths at NAB, it's very likely the manufacturers presented you with marketing messages that promoted their ideal way of editing HDV. It's also likely that you had at least a gut notion of what these sales pitches were about. But gut-level “feelings” are not a very good way to arrive at a purchasing decision.

Of course, I'm willing to bet that despite these impassioned pitches, very few editors will switch from one NLE to another simply to gain access to a couple of neat features. I strongly suspect that if you currently use Brand X, you'll stick with Brand X unless the new version for HDV editing is simply disastrous.

So if you're not looking at HDV editing technology strictly from the point of view of “which is best,” why do you care? Two reasons. You need to understand the strengths and limitations of the approach you're buying into. First, you need to know if your computer system will adequately support the approach. That is not just CPU gigahertz — it is also motherboard chipset and disk storage systems. For example, you may need to buy a RAID with several terabytes of storage depending on the NLE you choose.

Second, your NLE may provide you with several ways of editing, so you need to know which best fits your needs. For example, I'll show you how the newly released FCP 5 HD can use three of four HDV approaches that I'll discuss.

More pixels, more problems

HDV's transportability over FireWire and the fact that it records to MiniDV tape make it easy for folks to think of HDV as simply “high-definition DV.” There are, however, three significant differences between HDV editing and DV editing. First, HDV uses inter-frame compression, while DV does not. Second, HDV audio is MPEG-2, not PCM. And third, the number of pixels per HDV frame is either 2.7X (1280×720) or 4.5X (1440×1080) the number of pixels employed by DV.

All these differences create the need for a huge amount of computation to accomplish editing. This need can be satisfied in several ways. The four techniques for editing HDV are draft, proxy, native, and intermediate. Let's work our way through all these options.

As a general note, native HDV performs all the computations while you are working with the timeline. As you might expect, to obtain good timeline performance with a native HDV editor, a very powerful computer is required. Alternatively, HDV editing can break the task into less computationally intensive steps, which is the approach used by the draft, proxy, and intermediate solutions.

Draft editing

We are all likely familiar with using draft editing techniques. We batch-capture a tape using a codec different from the one used by our camcorder. Naturally, this codec is chosen because it decodes faster than our tape's codec. Apple's PhotoJPEG can be decoded faster than DV, so it has long been used as non-NTSC/PAL draft codec for DV.

Of course, our draft codec should require less storage space and disk bandwidth. One way to accomplish this is to simply lower compression quality. Alternately, we can degrade image resolution. It is also popular to drop every other field, or every other column, or both. The latter option cuts storage by a factor of four.

HDV certainly can be transcoded to Apple PhotoJPEG — as long as your Mac has enough power to decode HD-sized MPEG-2 frames and recompress them, while downscaling them.

Because computers can now process DV so efficiently, it is possible for DV25 itself to be used as an NTSC/PAL draft codec for HDV. Now all the effects that support DV can be used on our draft HD material. All that is needed is a computer that's fast enough to transcode HDV to anamorphic DV. The advantage of working with DV is that the timeline is ready for playback on an NTSC/PAL display when connected via IEEE 1394 to the monitor. In fact, you can directly export a DV production.

After the production has been edited, the timecode of the draft clips is used to generate a batch-recapture of the HDV material. Of course, this requires HDV equipment with frame-accurate capture ability.

Sony provides a superb way of supporting DV draft. Its HDV camcorders provide a hardware downconversion of HDV. Now you can edit with any machine that can edit DV. FCP HD and FCP 5 both support draft mode editing.

Proxy editing

Adobe Premiere 4.2 supplied the first implementation of proxy editing that I remember. We would capture all our taped material to motion-JPEG and save the file on our $10,000 4GB disks. Next we initiated a batch conversion of all these clips to a draft codec. This often was an all-night job — and in those days it was rare that a computer could run more than a few hours without crashing. Nevertheless, if we were lucky, it worked. The proxy codec typically supported 15fps playback with a resolution of 240×180 pixels.

Today we can use the same approach. In fact, my HDpartner Plus software used this approach and was used by dozens of editors to edit 720p30 HDV with FCP 3 and FCP HD. HDpartner Plus used 16:9 PhotoJPEG as the proxy codec. When the edit was complete, proxy clips were replaced by 720p MPEG-2 clips.

Currently there is now only one HDV proxy solution — DVgate Plus from Sony. It is available only with Sony VAIO computers. DVgate captures the entire HDV tape into clips. Because HDV is just MPEG-2, it is a simple task to play and trim HDV source clips. However, once you drop a trimmed clip into the timeline, the software performs a background transcode from HDV to proxy. The transcoded clip is called a temporary proxy clip. The proxy is a temporary format that can easily be trimmed, assembled, and re-arranged.

Once video has been edited, VAIO PCs perform smart rendering. This process only re-encodes the portions that have been edited. This greatly reduces the total amount of rendering time. It also allows for minimal degradation of video quality because HDV is only decoded once and recoded once.

HDV video can be recorded back to an HDV camcorder. Moreover, the edited version of the original HDV video can, in a data file format, be stored onto a data DVD at the original HD quality for PC playback — or for preservation.

Intermediate editing

By splitting the pre-editing process into two phases, the amount of required computing power is significantly reduced — to the point that a fast computer (3.2GHz P4 or a dual 2.5GHz G5) can handle the task.

Surprisingly, a 720p HDV file takes less space than does a DV25 file. The captured MPEG-2 transport stream requires only about 9GB of disk space per hour. Naturally, 1080i at 25Mbps requires the same disk space as does DV25 — 13GB per hour.

The key to splitting the computational tasks is first to capture and convert HDV to an intermediate codec, and then to edit this more efficient intermediate codec.

iMovie HD and Final Cut Express HD capture to the Apple Intermediate Codec (AIC). AIC is not at the same quality level as the CineForm and Canopus HQ intermediate codecs, but it does enable HDV editing. AIC, an I-frame-only codec, has the specifications displayed in Table 2 above.

After capture to AIC, you can edit as usual with iMovie HD and FCE HD. Of course, you need to select the appropriate HDV Project Type when opening a new project.

In the Windows XP world there two manufacturers of digital intermediate codecs: CineForm and Canopus. There are four versions of CineForm software: a high-powered version called Prospect HD for Adobe Premiere Pro, Connect HD sold by Sony for Vegas 5 and incorporated into Vegas 6, Aspect HD sold by CineForm for Adobe Premiere Pro, and a lite version of Aspect HD supplied by Adobe as a plug-in for Premiere Pro.

CineForm employs an intermediate codec it calls Carlsberg (CFHD). The CFHD codec is based on wavelet compression. CineForm emphasizes that CFHD was designed to be a very high-quality intermediate codec operating in a 4:2:2 color space. Quality is high because low-data-rate MPEG-2 HDV is decoded and then recompressed into a high-data-rate (80Mbps) codec.

To output your HD production back to an HDV camcorder (or to a D-VHS deck) for distribution, you must first render out to the MPEG-2-TS format. Your NLE likely includes HVD presets for the included MPEG-2 encoder. Then use the HVD I/O function to output the TS file over IEEE 1394.

Canopus has its own intermediate codec called HQ that is used by Edius Pro 3, Edius NX for HDV, and Edius SP for HDV. Canopus recognized that many editors use uncompressed video as an intermediate codec. Naturally, using uncompressed HD video for editing keeps video at its original quality. However, because of the massive data bandwidth, sustaining stable performance is difficult for the current workstations and RAID devices. Also, non-realtime rendering of uncompressed images for effects wastes time, which makes uncompressed editing systems unappealing.

Canopus believes that by using an intermediate codec that does not degrade either HDCAM or DVCPRO HD — yet requires vastly less bandwidth than using 8-bit uncompressed — it has developed the optimal solution for editing HD and HDV. (Canopus also supports native HDV editing for those with dual 3.4GHz Xeon computers.)

The HQ codec was developed to maintain the image quality of Sony HDCAM and Panasonic DVCPRO HD video data, while allowing compression for realtime editing.

The HQ Codec supports variable bit rate (VBR), to sustain high quality while keeping total data bandwidth low. VBR adjusts compression rates to suit either simple images or complex images within a video stream. (Editing systems using hard drives allow VBR codecs to save disk space and avoid system bottlenecks, thereby increasing system performance.) In addition, the HQ codec's original calculation method allows optimal-speed performance during decompression and compression. Tape formats such as HDCAM and DVCPRO HD require constant bit rate compression (not VBR) to be able to record to tape.

All the capture solutions presented so far transfer HDV via IEEE 1394. An alternative is to allow an HDV device to decode the 720p or 1080i video to analog component HD video (YPbPr). In this scenario you use a tiny HD10A converter ($2,800) from AJA Video Systems. The HD10A digitizes the analog input and converts it to an HD-SDI bit-stream.

HD-SDI (SMPTE 292M) is a technology of sending uncompressed 4:2:2 digital video and, optionally, digital audio, ancillary, and metadata, at either 1.485Gbps (60Hz) or 1.001Gbps (59.94Hz). (HDCAM and DVCPRO HD VTRs output HD-SDI.) Next, you connect the HD10A's BNC output to a Kona HD, Kona HD 2, or a Blackmagic DeckLink Pro HD PCI board's input BNC using a high-quality 75ohm coaxial cable.

Upon HD input, the uncompressed digital video can be stored to disk using Apple's uncompressed 8-bit and uncompressed 10-bit. A RAID is required because, for example, 1080i video requires a data transfer data rate of 124MB per second. In fact, a very large RAID is required because an hour of 1080i requires 448GB. Moreover, this is using uncompressed 8-bit video. These values increase — for 1080i, to 166MB per second and a whopping 597GB per hour of footage — were you to digitize to 10 bits.

Another alternative: the Blackmagic Design and AJA Kona 2 HD boards have the option of compressing using the DVCPRO HD codec. This compressed HD video can be stored to disk. Thus, you have the option of transcoding uncompressed HD-SDI to a compressed HD codec. The latter allows multi-stream editing without need for a RAID.

Native editing

NLEs that can directly capture and edit an MPEG-2 Transport Stream include Apple FCP 5, Canopus Edius Pro 3, Sony Vegas 5 and 6, Pinnacle Liquid Edition 6, and MPEG Edit Studio Pro. Support for native HDV editing in Avid's Xpress Pro HD is forthcoming (see “Avid & HDV: Worth the Wait?” here.)

Native DV editing means that with a cuts-only production, exactly the same bits are output to a tape as were input from a tape. No quality is lost. For manufacturers of HDV editing systems, however, “native” means any NLE in which MPEG-2 clips are placed into a timeline.

This blurring of the definition of native has come about because legacy MPEG-2 NLEs were rapidly modified to handle HDV, but not in the same way as second-generation NLEs expressly designed to edit HDV (i.e., Final Cut Pro 5). So in fact “native” HDV editing can be accomplished three ways, based on the ways that manufacturers use the term. For the sake of discussion, I've dubbed these approaches quasi-native, pseudo-native, and true-native.

A quasi-native HDV editing system repacks the transport stream to a program stream in realtime during capture. (Audio may also be decoded to PCM to facilitate audio editing.) The program stream is edited in the timeline.

A timeline can be recorded to an HDV camcorder only after the program stream — along with any encoded audio — is repacked to a transport stream. Although some time is wasted during export repacking a program stream to a transport stream, no MPEG-2 frames are decoded and encoded — except for a few GOPs at each cut point and where effects have been applied. The ability to handle cut points by encoding only a few new GOPs is often called Smart GOP Encoding.

However, some NLEs, like MPEG Edit Studio Pro, force a complete decode of the program stream followed by an encode back to a transport stream. I'm calling this technique pseudo-native editing.

A pseudo-native NLE places transport stream clips into the timeline. (Audio may, during capture, be decoded to PCM to facilitate audio editing.) The transport stream is edited in the timeline. A timeline can be recorded to an HDV camcorder only after every MPEG-2 frame is decoded and encoded. The new video elementary stream — along with any encoded audio — is packed into a new transport stream.

Under this system, hours may be wasted to export even a cuts-only timeline. Moreover, quality is obviously lost. (It is possible to use both Vegas and Edius Pro in this manner.)

A true-native HDV editor also works directly with transport streams. (Again, audio may be decoded to PCM to facilitate audio editing.) And, again, the transport stream is edited in the timeline. During export, however, no MPEG-2 frames are decoded and encoded, except at cut points and during effects. This type of technology essentially eliminates export delays. It is a perfect analog of native DV editing.

Pinnacle Liquid Edition 6 and Apple's FCP 5 HD support true-native editing. True-native editing naturally requires a powerful computer — but software engineers can do a great deal to improve editing performace.

Bottom line, when you read marketing claims about “native” editing, you've got to ask serious questions about exactly what type of technology is employed. Likewise, when you read about the virtues of employing an intermediate codec, do not simply buy into all the negative claims made about native editing.

In future issues we will further explore these concepts and the NLE systems that use them. On p. 26 of this issue we begin with Pinnacle Liquid Pro 6. Next month: Canopus Edius NX for HDV.

---

feedback

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

© 2008 Penton Media, Inc.