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

 

The ProHD Approach

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

Though not a format unto itself, JVC’s ProHD pushes HDV forward by capturing PCM audio and true 24p.

Web Expanded

Sidebar
ProHD Motion Smoothing

At NAB 2005, JVC introduced its new HDV camcorder, the GY-HD100. This camcorder records to videotape in the HDV format, yet it is described as a ProHD product. ProHD is not a format unto itself, nor is it a sub-format of HDV. According to Dave Walton, JVC's general manager for corporate communications, ProHD is “JVC's approach for delivering affordable HD products. It is also used to describe a family of bandwidth-efficient professional HD models.”

ProHD differs from the types of HDV that have been implemented for JVC's HD10 and Sony's Z1/FX1 camcorders in that it includes two important new features. First, PCM audio on the tape — the other implementations of HDV capture only MP2 audio. Second, ProHD allows true 24p capture.

HDV sub-formats

The HDV format includes three sub-formats: 720 horizontal pixels by 480 scanning lines SD progressive at 25p, 50p, and 60p; 1280×720 HD progressive at 24p, 25p, 30p, 50p, and 60p; and 1440×1080 HD interlace format (50i and 60i). Current shipping HD HDV products employ 720p29.97 (e.g., JVC GR-HD1/JY-HD10) or 1080i50 and 1080i59.94 (e.g., Sony HDR-FX1/HVR-Z1).

The HDV format specifies the data recording of MPEG-2-compressed high-definition signals. Because the format employs the same cassette case, tape speed, and track pitch as the DV format, it can utilize mechanical parts based on the DV format. In fact, the label HDV can be applied only to products that record to DV tape. (Although after NAB, is it quite clear that recording to a hard drive is also allowed. Moreover, JVC has stated its forthcoming GY-7000 may support data rates higher than 25Mbps.)

HDV employs MainProfile@High1440 (MP@H-14) MPEG-2. The sampling space is 4:2:0. The HD 720p sub-format is called “HD1,” and the 1080i variant is called “HD2.”

HD1 has a 19.4Mbps data rate that is composed of three types of data: MPEG-2 video (17.8Mbps), MPEG-1 Layer 2 audio (0.384Mbps), and supporting data (1.2Mbps). A closed six-frame GOP (IBBPBB) is employed in Region 60 products.

HD1 is recorded to DV tape almost exactly as is DV. A DV track has three segments. The first and largest segment of a DV track holds “Video.” A small “Subcode” segment holds data such as timecode. In addition, PCM is recorded into an “Audio” segment. All three DV track segments are shown in Diagram 1.

When HD1 is recorded, both MPEG-2 video and MP2 audio are recorded into the “Video” segment. In fact, HD1 is recorded just as is DV, where 10 tracks are used to record one video frame.

Interestingly, HDV is more robust than DV and so is less subject to dropout damage. That's because with Region 60 HD, a six-frame GOP requires 60 tracks. Moreover, adding to an error-correction scheme, the IBP frames are interleaved in small blocks over all 60 tracks, so a dropout cannot damage a complete I, B, or P frame.

ProHD

What, you might ask, is written into the “Audio” segment of an HDV recording? Nothing, as is shown in Diagram 2.

With ProHD, two-channel PCM audio is written into this segment. The audio is completely independent of the MPEG-1 Layer 2 audio. The PCM audio is sampled at 48kHz using 16-bit samples.

The PCM tracks receive exactly the same audio as does the MPEG-1 encoder. ProHD does not, at this point, support four---channel audio. See Diagram 3.

How are the PCM tracks obtained from ProHD products? It is not fully clear, but one possibility is to switch the camcorder into HDV-to-DV Playback mode. Now as a pair of flying heads crosses each pair of tape tracks, one frame of PCM data is read out and is output via IEEE 1394.

After audio capture, the DV video (from this HDV-to-DV Playback mode) is discarded. Now you simply replace the MPEG-1 audio tracks with the PCM audio tracks. Although this seems complicated, it is almost identical to the double-system editing used in moviemaking. The difference is that most moviemakers record PCM sound on a separate digital audio recorder.

There is one small quirk. Before six frames of video (one GOP) are ready for output via IEEE 1394, 60 tracks must be read. Therefore, PCM audio will lead MPEG-2 video and MPEG-1 audio by six frames. That means your PCM tracks will be six frames out of sync with video. Therefore, when you replace MPEG-1 audio with PCM audio, you'll have to nudge the PCM track earlier by six frames.

It also means that PCM audio — really DV audio — may be able to be played on existing HDV equipment.

To record widescreen SD at 480p60, a ProHD camera, such as the new GY-HD100, captures 59.94 progressive frames per second from a 720×480 16:9 window from the 1280×720-element CCDs. (To record 480p50, the camcorder is clocked at 50Hz.) As each frame is captured, it is encoded. After each six-frame GOP has been encoded, it is written to tape over 60 tracks.

When a ProHD camcorder records 720p30 (720p29.97), the camera captures 59.94 progressive frames per second from the 1280×720 CCDs. As each frame is captured, it is sent to the encoder. Therefore, the encoder inputs two frames for every frame that will be written to tape. Diagram 4 shows how six frames are processed. For one second of 30p video, this pattern is repeated five times.

Red boxes denote the incoming frames from the CCD. The blue boxes indicate the frames that are generated from the incoming frames by the Motion Smoothing filter. Green boxes describe the MPEG-2 frames — making up one GOP — recorded to tape. After each six-frame GOP has been encoded, it is written to tape over 60 tracks.

When a ProHD camcorder records 720p24 (720p23.976), the camera captures 47.952 progressive frames per second from the 1280×720 CCDs. As each frame is captured, it is sent to the encoder. Therefore, the encoder inputs two frames for every frame that will be written to tape. For 24p, the pattern illustrated in Diagram 4 is repeated four times per second instead of five.

JVC has not described how 24fps is recorded. One option is to record each I-frame twice. In this way, 24fps is naturally converted to 30fps and can be encoded as 720p30. Alternately, ProHD camcorders such as the GY-HD100 could record to tape at a rate of 24fps.

Such discussion of 24p implementation begs the question: Will ProHD be able to be played back in legacy HDV equipment? (The JVC BR-HD40U recorder, available in June, will support signals recorded in 24p HDV.) JVC notes that “24p record/playback capability have been part of [the] HDV format since it was established.” However, if 24p is part of the origial HDV spec, is it included in both HD1 and HD2? If it is part of HD2, will Sony implement 24p in the Z1? These are all aspects of HDV implementation that will become clearer over time, as JVC rolls out the ProHD camcorder and the market continues to use HDV in general.

ProHD in the future

Neither 720p50 nor 720p60 recording are offered by the GY-HD100 camcorder. However, both could be included in a future extension of ProHD since it is part of the HDV specification. It is easy to see how both frame rates could be handled. Each incoming frame from the CCDs would be encoded and recorded to tape. No Motion Smoothing is required.

The frame rate for 50p/60p requires the ability to process twice as much data per second. There are several ways this greater data could be recorded. Perhaps the GOP length could be increased. Or, perhaps, the recorded MPEG-2 data rate could be increased from 18.2Mbps to 36Mbps.

The most obvious way to record 36Mbps to tape is to use either double-density recording on each track (heads rotate twice as fast) or record pairs of tracks. In the latter case, a six-frame GOP would require 120 tracks.

To record double the number of tracks per second, the typical technique is to roll the tape twice as fast — thereby cutting recording time in half. Of course, a standard-sized DV cassette is a simple solution to this problem.

At this point, all that can be said is that although 720p60 is part of the HDV specification, it is not clear how this capability will be implemented.

Analog Playback

In order to avoid video flicker upon playback, 720p25 or 720p30 video must be frame-doubled during output, yielding 720p50 or 720p59.94 analog component video. (Also possible: 720p25 and 720p30 can be upconverted to 1080i50 and 1080i59.94 — or downconverted to 480i50 or 480p50 and 480i59.94 or 480p59.94.). The possibilities are determined by the circuitry in a ProHD camcorder or deck.

To output 24p as an analog signal, a different but familiar process is used. Pulldown is performed to convert every four progressive video frames to five interlaced video frames, as shown in Diagram 5. This pattern is repeated six times per second.

24p IEEE 1394 playback

When 24p is processed during playback, the actual 24 frames are output, or 36 “repeat” frames are added to yield 60 frames. With repeat frames, capture can be accomplished in two ways. The 720p60 file can be directly recorded to disk — and then converted to 24fps video. Alternatively, the capture driver can strip out the extra frames automatically. Once 24p video has been obtained, you can edit it in a 23.98 timeline exactly as you now do telecined film.

---

ProHD Motion Smoothing

Our brain does not see motion from a series of pictures until they are presented at a rate over 12 frames per second. Moreover, according to Douglas Trumbull, inventor of the Showscan film system, humans do not see motion as perfectly smooth until the frame rate reaches about 60fps. (As the frame rate increases people report a greater sense of "reality" to a motion picture.)

Whenever continuous information—including motion over time—is sampled at regular intervals, unless the sampling rate is at least 60fps, the result motion pictures will exhibit "temporal aliasing." You may not think you know what this means, but we have all seen aliasing. Whenever you see a covered wagon wheel begin to spin in reverse, the backward spin is aliasing of repetitive motion. Aliasing is inherent with low sampling rates such as 24fps or 30fps.

Low-temporal-rate motion pictures have another visual problem. When objects move rapidly, or when one pans too fast, objects in the scene may appear as "double objects." For example, when a car drives past, two images of the car may be visible—one image slightly displaced from the other.

This is called "foreground strobin"g—and is an "eye tracking" artifact. It is also called "judder", although that term is also applied to the look of film after 2:3:2:3 pulldown has been applied in order to transfer it to video.

To understand foreground strobing, imagine a horizontally moving square shot on film at 24fps. In a theater when the film is projected at 24fps with a double-bladed shutter, each frame is shown twice. This raises the "flicker frequency" from 24Hz to 48Hz. If this were not done, we would see intolerable flicker—especially in bright areas of the picture. As we watch the projected image, our eyes automatically track the moving square. Our brain quickly determines the square's "movement vector." Therefore, when the shutter opens on a new frame, the square is where our brain has directed our eyes to look. When, however, the projector shutter opens a second time on the same frame, our gaze point has already advanced halfway to the anticipated position of the square at the next frame time. Thus, the square is imaged onto our retina a second time—at a position displaced along the axis on which the object is moving. This is the origin of the "double" image. The faster the object moves, the faster our eyes move, and the greater the displacement of the "double image."

When we connect a 24p or 30p camcorder to a monitor, the camcorder creates a 60Hz image—either 60 progressive frames per second (720p60 output) or 60 interlaced fields per second (480i or 1080i output). Thus, the camcorder's electronics work exactly like a two-bladed projector shutter, causing each frame to be displayed twice. Interestingly, stationary objects also strobe. This phenomenon is called "background strobing." Naturally, neither problem is found with either 720p60 or 1080i60 video. Foreground and background strobing can be ameliorated by controlling shutter speed. When shutter speed is kept slow, each image is blurred. At 1/60 second there is some motion blur, while at 1/30 second there is far more blur. The image motion blur is recorded to tape.

When we watch such video, the blur of the moving object helps cover the gap between the real image and the eye-tracking artifact. Therefore, the two tend to blur into one larger object. Because we are so used to seeing motion blur, our eyes tend to see one object rather than two—and the film, or video, looks more acceptable. According to JVC, ProHD provides proprietary Motion Smoothing filter technology. Every 1/48, 1/50, or 1/60 second a new frame is captured. This video can be processed prior to the MPEG-2 encoding—or sent into the encoder. Therefore, motion video that is sampled at 48fps, 50fps, and 60fps is available. These rates are all high enough to avoid foreground and background strobing.

Compared to the 24fps, 25fps, and 30fps that will be recorded to tape, the input is "oversampled." Oversampling is often employed by analog-to-digital converters to improve the quality of audio and video. Oversampling provides samples that are more accurate.

How are the two samples converted to one sample? There are two possibilities that would be obvious—frame blending and applying blur to specific objects that move in ways that would yield judder—but JVC employs neither of them. JVC has not disclosed its method of converting two samples into one.

The company simply claims the Motion Smoothing filter helps remove judder. Since judder is always a negative—filmmakers devote enormous effort to prevent judder—the filter offers a labor-saving function. (Of course, the Motion Smoothing filter may be turned off.) Those of us who want 720p60 will use the filter to enable 720p30 video to emulate closely the higher frame rate.

---

feedback

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

© 2009 Penton Media, Inc.