Editing
HDV
by
Adam wilt
In the
last episode of Technical Difficulties, we looked at editing long-GOP
MPEG-2: video compressed not one frame at a time, but in groups of
pictures. Long-GOP compression requires an NLE to fetch and decode
multiple frames to display or edit only one-a considerable complication.
I finished
by saying, "I hope you now see why long-GOP formats are less than
ideal for editing, and can understand why the Pinnacle Systems DC1000,
brave as it was, didn't start a revolution. Still, HDV is long GOP,
and folks want to edit it."
Guess
what? The revolution is here. At NAB 2004, Pinnacle Systems (www.pinnaclesys.com)
showed long-GOP MPEG-2 editing, including native HDV, incorporated
in the Liquid NLEs, including Edition. When I experimented with it,
I could scrub the timeline smoothly, almost effortlessly. As far as
system responsiveness goes, I might as well have been editing DV,
but the pictures displayed were long-GOP, high-definition HDV.
Several
aspects of technology and the marketplace have converged to make Pinnacle
Systems offer a long-GOP NLE again:
- Modern
PCs are fast. They're so fast that they've got plenty of power to
handle the user-interface aspects of nonlinear editing, and have
enough left over to handle the hassle of fetching and decompressing
long-GOP video in the background, with no perceptible impact on
system responsiveness.
- Memory
is plentiful. Today's NLEs often ship with half a gigabyte of memory
or more. Reserving tens of megabytes of main memory for frame buffers
makes a dent, but not a serious one.
- The
Liquid editors have always been champs at multitasking. They've
handled on-the-fly background rendering from the beginning, so the
basic architecture for launching and coordinating background tasks
is already present. Adding the programmatic equivalent of a hunchbacked
assistant to run about, gathering GOPs from disk and decompressing
pictures in anticipation of their display, is a relatively simple
matter.
- The
advent of the HDV format, with its potential to recapitulate DV's
evolutionary success, provides the perfect impetus to develop long-GOP
editing and bring it to market.
And,
of course, once the work is done for HDV, it can be exploited for
higher bitrates, too. Pinnacle Systems calls this PracticalHD-the
ability to handle HD at a variety of bitrates with enough quality
to make it worthwhile, yet enough compression to make it easy to deal
with.
To be
fair, Pinnacle Systems isn't the first to edit long-GOP HDV in its
native format. KDDI's MPEG Edit Studio Pro LE (www.kddlabs.co.jp/eng/index.html),
shipped with the current JVC HDV camcorders, and Ulead Media Studio
Pro 7 with the Ulead HD plug-in (www.ulead.com/msp/plugin.htm)
are both able to edit long-GOP HDV.
Pinnacle
Systems won't be the last company, either. The A Team is hard at work:
Adobe is working to bring HDV-native capability to Premiere Pro; Avid
announced at NAB that future versions of Xpress Pro, NewsCutter XP,
Media Composer Adrenaline, and NewsCutter Adrenaline FX systems will
handle HDV in its native form; and after its NAB press conference
introducing Final Cut Pro HD with DVCPROHD over FireWire, Apple's
name quietly appeared on the supporters' list on the HDV Format Web
site (www.hdv-info.org).
Sony Pictures Vegas also expects to be HDV native in a few months.
These are the obvious suspects; undoubtedly others are working on
native HDV NLEs.
Editing
HDV in its native form has all of the benefits (and drawbacks) of
editing DV in its native form. I should probably define what I mean
by "native HDV editing" before muddying the waters further.
HDV
at a glance
HDV
is a specification for recording 16:9 HDTV on DV cassettes. The format's
main specifications call for both 720p and 1080i recording, with a
4:2:0 sampling structure: Chroma has half the resolution horizontally
and vertically as luma does, just as it does in DVD, digital television
transmission, and 625/50 DV and DVCAM. Long-GOP MPEG-2 is used to
squash HD down to a bitrate suitable for DV tape recording.
In 720p
(720 line, progressive scan), 25, 30, 50, and 60 frames per second
can be recorded. The luma information is sampled at 1280 pixels across
by 720 down-which gives it a higher raw sampling resolution than 720p
DVCPRO HD-and the video is recorded to tape as an MPEG-2 transport
stream (TS) at around 19 Mbps.
In 1080i
(1080 line, interlaced), 50 and 60 field-per-second video is supported,
sampled at 1440 x 1080-the same luma subsampling used in HDCAM. It's
recorded as a 25 Mbps MPEG-2 packetized elementary stream (PES).
Audio
in either case is two channels of 16-bit 48 kHz sound, compressed
to 384 Kbps, using the MPEG-1 Layer 2 specification so that it takes
up one-quarter of the space of uncompressed audio.
Going
native
Both
720p and 1080i recordings are available as transport streams over
IEEE 1394, aka FireWire. As with DV, FireWire allows a bit-for-bit
copy of audio and video between tape and NLE, a lossless, low-cost
means of transfer.
HDV-native
NLEs may choose to store transport streams as is, or to demultiplex
them into elementary audio and video streams for editing. I consider
both methods to be "HDV native" because the audio and video data remain
in their original, compressed formats in either case, with no generation
loss in the capture process.
|
| Figure 1
- Single- and multigeneration HDV-native editing is almost the
same as working with DV. |
With
FireWire transfer and HDV-native editing, your workflow is almost
identical to working with DV over FireWire. The data rate is low enough
to make capture on single disks, including laptop drives, feasible.
Plug in the camera, capture material from it, edit, and print back
to tape (to HDV or to D-VHS). With most NLEs, the material is decompressed
once (actually, each time a frame is displayed or processed) and recompressed
once (after all of the effects on the timeline have been applied),
so your finished program is only one generation removed from the camera
original.
If your
target is Windows Media 9, H.264 (the MPEG-4 Advanced Video Codec),
SD DVD, Fisher-Price Pixelvision, or any other format, the same scenario
applies. The video is captured in its native format over FireWire,
decompressed once in the timeline, and recompressed to your target
format, as shown in the upper half of Figure 1.
This
workflow has been proven with DV25. At 19 or 25 Mbps, HDV is as meek
and tractable as DV for desktop-or laptop-editing.
It also
has the drawbacks of DV-native editing. As with DV, the MPEG-2 compression
used in HDV is lossy; you want to minimize the number of compression
cycles or generations your pictures go through. If you simply capture
the video, edit it in a single application, and output to your final
mastering format, it's fine-but there are times when it's not good
enough.
Better
than native?
Some
NLEs allow "prerenders" or "precomputes" of portions
of the timeline, speeding up future operations. Instead of rendering
a complex effect from scratch every time, these NLEs keep intermediate
render files around and simply composite new information atop them.
Similarly,
some projects require video to be bounced between an editing program
and a compositing app like Adobe After Effects or Discreet combustion.
In both cases, any given frame may pass through several generations
of decompression and recompression. Keeping your material in HDV's
highly compressed MPEG-2 can lead to a nasty buildup of compression
artifacts and generational losses (Figure 1, lower half).
In such
cases, it makes more sense to move video around in a less-compressed
state. Although you can go out today and buy NLEs that capture and
store HD in its full, uncompressed state, the high data rate and the
stack of RAIDed hard disks required to record it make such NLEs both
costly and intimidating.
|
| Figure 2
- Using a low-loss editing format instead of editing native HDV
reduces multigeneration compression losses. |
What
if you had a system that transcoded HDV as it came in over FireWire
into a higher-bitrate, but still compressed, format designed to be
virtually lossless in the editing process? CineForm (www.cineform.com)
offers a "Visually Perfect" CFHD codec in bundles for both Adobe Premiere
and Sony Pictures Vegas (and it's compatible with any AVI-based Windows
program). The CFHD codec-and, in Premiere Pro, the Carlsbad rendering
engine built around it-is designed for optimal HD editing on Pentium
4 CPUs. Its data rate is low enough to be tractable on single drives
and small RAIDs, yet high enough to avoid noticeable generational
losses. The codec exploits the P4's SSE instruction set for rapid
processing; CineForm says that a 2.8 GHz P4 can handle four streams
of 720p CFHD video in realtime.
Once
your editing and effects are finished, you can recompress to HDV's
MPEG-2 or to WM9, H.264, etc., just as in the HDV-native case, but
the CFHD codec allows you to bounce among applications without incurring
the compression losses that native HDV would. CFHD isn't a final or
distribution codec; it's strictly designed to be used during the editing
process (Figure 2).
Proxy
editing
You
can also edit HDV (at least on Final Cut Pro) by proxy. In proxy editing,
you capture HDV, then convert the material into proxy clips using
a codec more suitable for editing, such as SD DV, or SD or HD OfflineRT
(Photo-JPEG based). You edit your show with the lower-resolution proxy
clips, taking advantage of Final Cut Pro's realtime capabilities with
the more friendly format, then conform or online your finished program
from the original material.
Proxy
editing tools include Heuris Pro Indie HD Toolkit (www.heuris.com),
Lumiere HD (www.lumierehd.com),
and Steve Mullen's HDpartner Pro (www.mindspring.com/~d-v-c).
They use separate programs running outside of the NLE to capture HDV
material, demultiplex the MPEG-2 transport streams into separate audio
and video files, convert the compressed audio into uncompressed AIFF,
and transcode the MPEG-2 video to the proxy format.
After
you edit your show, you relink your clips to the original media and
render, typically to an uncompressed file. You then feed that fat
file into a separate compression utility to squeeze it back down to
a long-GOP MPEG-2 transport stream ready for export to HDV or D-VHS,
or export to a variety of formats using Apple Compressor, Discreet
Cleaner, Sorenson Squeeze, or other compression utilities.
Of course,
you can also render an SD timeline, taking advantage of high-resolution
HDV acquisition to give you a richly detailed picture thanks to the
benefits of supersampling. Many HDV users today are making SD DV and
DVD masters; in such workflows, simply choosing DV or uncompressed
SD as your proxy format lets you benefit from HD imaging while keeping
the rest of your editing workflow and final delivery in the same SD
formats you're already used to.
An embarrassment
of riches? Perhaps, and there's more in the works. Most HDV applications
don't have batch-recapture capability yet, for example. But the breadth
of choices hints at the tsunami to come. Poke around the Web sites;
maybe edit an exploratory project or two. That way, when the wave
hits, you'll be ready to ride it.
Streams?
Transport
stream? Packetized elementary stream? To make a long story less long,
a transport stream (TS) is analogous to a QuickTime or AVI wrapper
containing video and audio data.
A packetized
elementary stream (PES) is a "raw" stream of compressed video or audio
data (actually, a TS is more like an MXF wrapper than QuickTime or
AVI, though it isn't that much like any of 'em, but let's split one
hair at a time). There's also the program stream (PS), which is much
like a TS, but designed for comparatively error-free environments
instead of the hurly-burly of long-distance transmission, and better
optimized for editing purposes; however, HDV doesn't use program streams.
Confused yet? No worries-regardless of the stream type, the data is
the same: long-GOP MPEG-2 video.
Adam
Wilt (www.adamwilt.com)
still resides in an SD world, but looks forward to affordable HD.
Copyright 2003, CMP Media LLC
HDV
Introduction
By
Steve Mullen
A
few days before the opening of IBC 2004 in Amsterdam, Sony announced
its new HDV camcorder, the HDR-FX1 (MSRP of $3700). Then in mid-November,
the Sony Business Solutions & Systems group announced the HVR-Z1
with an MSRP of $5,946. The FX1 began shipping in November, while
the Z1 will ship in February 2005.
I
will focus here on the eight features that have the most relevance
for professional shooters.
1.
There will be two versions of the FX1. One for "Region
60" (FX1) that will support 1080i60 (1080i59.94) plus 4:3 and
16:9 NTSC DV. The other for "Region 50" that will support
1080i50 (1080i50.00) plus 4:3 and 16:9 PAL DV. While I do not typically
concern myself with Region 50 equipment, the Sony HDV camcorder
will be an exception because it can easily be used to shoot video
for transfer to film.
Stereo
audio and 1080i video will be encoded using a long GOP (12 GOP in
1080i50 and 15 frames in 1080i60) MPEG-2 and MPEG-1 Layer 2, respectively.
The 25Mbps encoder output will be recorded as PES (Packetized Elementary
Stream) data. The PES data structure is different from the Transport
Stream structure recorded by the JVC HDV camcorders and D-VHS decks.
However, upon i.LINK outputÑPES is converted to TS. Likewise, TS is accepted by the i.LINK
port for recording. ItÕs likely a Copy Protect flag is required
within the TS bit-stream to enable an FX1 to recording HDV. Without
the flag, no recording can occur.
My
HDVideoSplicer and HDpartner Prime utilities have already been enhanced
to support both 1080i50 and 1080i60. Please click to:
www.mindspring.com/~d-v-c
It
is not yet clear if the FX1Õs i.LINK output will be able to be directly
recorded to D-VHS. Sony has already acknowledged the FX1 is will
not be compatible with Blu-ray DVD recorders. SonyÕs Masashi Imamura
stated, ÒI, personally, definitely want to realize the ability to
record HDV on Blu-ray. We are working hard towards this but there
are many copyright issues and many technical issues, but I think
we can come to a resolution at some point in the near future.Ó
2.
The HDR-FX1 has an F1.6 to 2.8,
Carl Zeiss Vario-Sonnar T* (anti-reflecting coating) lens
that accepts 72mm filters. The 12X zoom covers a range of f = 4.5
to 54.0mm (35mm equivalent: f = 32.5 to 390mm.) You can zoom by
variable servo control, a lever, and a non-perpetual zoom ring.
A built-in ND has two settings: 1/6 and 1/32.
3.
Sony developed a new 1/3-inch, Super HAD, CCD for their HDV
camcorders. Each CCD has 1,012 (horizontal)
by 1,111 (vertical) elements (1,120,000
pixels) that provide an effective pixel count of 1,070,000
pixels (972 (horizontal) by 1,100 (vertical). Vertical smear level
is rated at a very low Ð107dB. Each element has a 2:1 rectangular,
rather than square, aspect ratio.
This
aspect ratio explains how Sony can work with an image that is 960
pixels wide by 1080 pixels high. To obtain an image aspect
ratio of 16:9, the pixels from these CCDs could not be square. One
ninth of 1080 is 120 which when multiplied by 16 indicates that
there should be 1920 pixels in each row. By making each pixel twice
as wide, only 960 elements are required.
While
double-width CCD elements might partially account for the FX1's
light sensitivity value of 3 Lux (obtained with gain at +18dB) the
real player may be SonyÕs use of micro-lens CCDs. Each CCD element
has a tiny lens that gathers light and focuses it on the center
of the CCD. (Which implies the extra width of each element, on either
side of the centered lens, does not play a big role in increasing
light sensitivity.) See Diagram 1, from the Sony Japan website.

Diagram
1: Micro Lens on Each CCD Element
Both
the JVC "HD-1" and Sony's "HD-2" HDV definitions utilize MainProfile@High1440 (MP@H-14) MPEG-2. This format
accepts a vertical pixel count of up to 1080 and a horizontal pixel
count of up to 1440. JVC works with 720 by 1280 while Sony works
with 1080 byÑ960? That's certainly not what I expected. In my Future
of HDVÑPart 2 for Video systems magazine, I stated "it
is customary to use CCD's that have at least the number of column
as the maximum number of columns supported by the video format."
I certainly expected a CCD element count of at least 1440 pixels.
What happened? It turns out Sony is using a technology that can
provide 1440 pixels from CCDs with only 960 elements.
By
offsetting the green CCD one-half element spacing from the red and
blue CCDs, a source of additional luminance information is created.
By combining output from all three CCDs, horizontal resolution is
increased by up to 150-percent. And, indeed when 960 is multiplied
by 1.5, the result is 1440. This how Sony obtains a 1440x1080 pixel
luminance image that is encoded as ML@MP-14 MPEG-2. Naturally, that
leads to the question of how "real" is the extra resolution
obtained by using Òpixel offsetÓ technology.
The
simple answer is that when resolution tests are performed, the horizontal
resolution will be that expected from a 1440 element wide CCD. (I
would guesstimate about 1,000 TV-lines.) The complex answer is that
with pixel offset technology the effective horizontal resolution
is a function of the colors, the color patterns, and the motion,
of objects in a scene. To demonstrate this variability I created
multiple diagrams that show effective horizontal resolution of a
very imaginary sceneÑa picket fence that is imaged by the a tiny
8 element portion of the CCDs. While not at all mathematically accurate,
they do represent how resolution changes by the color the fence
is painted. If you look closely at the top row of each diagram,
youÕll see a grayscale pattern that is a representation of the fence.
Your eyes should be able to see differences in the clarity of the
fence pattern. You will note that, as expected, the highest resolution
is from a black and white fence. This is why test patterns will
indicate full horizontal resolution.







The
4:2:0 chroma data are obtained from sub-sampled chroma information
from the three CCDs. The two color components (Y-R) and (Y-B) are
obtained from (G-R) and (G-B), respectively. ItÕs not yet clear
if the offset green component has any effect on chroma definition.
Likely, it does not because MPEG-2 chroma resolution at 4:2:0 is
not that highÑa potential limitation on the use of chroma keys.
Conversion
of the three signals from the CCDs is accomplished by a 14-bit A/D.
The digital data are then processed by a 14-bit DXP. The wide word-length
converters and DXP will hopefully prevent highlights from blowing-out.
The DXP also supports gain settings of 0, +3, +6, +9, +12, +15,
and +18dB. Shutter-speeds range from 1/4 second to 1/10,000 second.
The ITU Rec. 709 HD colorspace is employed by the HDR-FX1.
Sony
has developed a set of four LSI chips that handle the processing
chores. These include: a Base-band signal processor (2,000,000 transistors
with an 18Mbit DRAM), the HD-MPEG Video Encoder (1,500,000 transistors),
an HD-MPEG Video Decoder (700,000 transistors), and an HDV Streaming
Processor (1,200,000 transistors). The MPEG Video Decoder also decodes
720p30 so you should be able to play tapes from current JVC camcorders
on the HDR-FX1. The 720p30 will be output as 720p60.
4. Two-channel
audio can be obtained from the built in stereo mic or via a 1/8-inch
mini-jack. An audio level control adjusts the level for both channels
simultaneously. While some will object to this limitation, the HVR-Z1
version does have dual audio level controls. The Z1 also has dual XLR connections. (IBC photograph courtesy of Filip
Vandoorne.)
5.
DV audio is carried as PCM data while HDV audio is carried
as MPEG-1 Layer 2 data (MP2). This precursor to MP3 is, like MP3,
a ÒperceptualÓ encoding system that discards audio information that
DSP computations indicate will be masked by other audio information.
In short, it is a lossy encoding system that may not meet field-recording
requirements.
6.
The FX1 offers a CinemaTone mode. A linear gamma is often
available on camcorders designed to shoot Òvideo for film.Ó The
linear CCD signal is matched to the linear (B-to-C) portion of film
gamma (see Diagram 9).

Diagram
9: Film Gamma
The
FX1 offers a single CinemaTone mode where the gamma curve (pink
in the diagram below) gracefully prevents highlights from exceeding
about 100IREÑthereby preventing the total loss of highlight detail
when portions of the image are too bright. This function simulates
increased light latitude expected from film. The portion of the
gamma curve below 80IRE is relatively linear compared to the ÒnormalÓ
(gray) concave "video" gamma curve. This slightly crushes
blacks thereby decreasing shadow detail. Doing so can increase the
richness of colors.

The
HVR-Z1 offers two gamma settings: Gamma 1Ñthe
gamma curve (pink) gracefully prevents highlights from exceeding
about 100IRE. Gamma 2Ñin addition to preventing blown highlights,
black stretch (blue) is applied to prevent dark grays from being
crushed into black.
7.
The FX1 employs "interlace scan dual-line" CCDs
that add pairs of rows within the chipsÑthereby automatically outputting
a field every 1/50th or 1/60th second. (The
driving logic causes alternate fields to automatically contain only
odd or even lines.) This type of CCD perfectly supports interlace
scanning as it can read-out 540-lines every 1/50th or
1/60th second. Effective vertical resolution is approximately
850-lines. In CineFrame 30 (FX1/Z1) and CineFrame 25 (FX1e/Z1),
smart deinterlacing is used to create video that has a temporal
resolution of either 30fps or 25fps.
Diagram 10 shows
how the deinterlacing of 1080i60 video. Red text indicates the five
deinterlaced samples per five interlaced frames. Deinterlacing can
be accomplished in several ways.

A
simple technique is to measure the difference between the two fields
in a frame. If there is little differenceÑhence, no motionÑthe frame
is passed without change. Effective vertical resolution remains
about 800-lines. When there is a differenceÑthe upper field is copied
and replaces the lower field to create a new frame. In this case,
effective vertical resolution is reduced to about 400-lines. When
this technique is used, effective vertical resolution varies with
the amount of motion. If there is motion in the image, resolution
can be halved for the entire frame.
A
better technique uses an intelligent deinterlace process. Motion
is detected on a line-by-line basis. Areas with no motion will carry-forward
all lines. Areas with movement, however, will carry-forward only
information from one field. The result is a frame that has, in static
parts, about 800-lines of vertical resolution without interlace
artifacts. Objects in motion, however, will have only approximately
400-lines of effective vertical resolution.
Using
more advanced technique, motion is detected on a pixel-by-pixel
basis. Pixels, making up an object that has no motion, will be obtained
from both fields. Pixels, making up an object that has motion, will
be obtained from only one field. The result is a frame that has,
in static parts, about 800-lines of vertical resolution without
interlace artifacts. Objects in motion, however, will have only
approximately 400-lines of effective vertical resolution.
A
very intriguing possibility is that deinterlacing is done during
MPEG-2 encoding. By measuring motion between fieldsÑusing
the encoderÕs motion tracking logicÑthe following may occur:
I-frame:
when "objects" do not move between fieldsÑboth even and
odd lines within such objects are encoded thereby supplying them
with full effective vertical resolution. When objects moveÑonly
the even field is encoded so such objects have half the possible
resolution.
P
and B-frame: when "objects" do not move between fieldsÑthey
simply are not encoded. When objects moveÑonly the even field is
encoded so such objects have half the possible resolution.
The Table below
shows how this could be done. Blue text indicates fields that are
copies of even fields.

The
result is that for moving objects, the samples are either 1/25th
or 1/30th second apart exactly as they should be.
When
CineFrame 30 is selected, you must a shutter-speed of either
1/30th or 1/60th. You should not
shoot in AUTO or use AE mode as the shutter-speed may rise above
1/60th and cause excessive motion strobing. The FX1e
and Z1 offer a CineFrame 25 mode that deinterlaces 1080i50 to provide
25fps video. CineFrame 25, using either a 1/25th or a 1/50th second
shutter-speed, is ideal for shooting video that will be transferred
to film. For both CineFrame 25 and CineFrame 30, when there is little
motion within the frame, you can choose the faster of the two speeds.
When there is motion, you should choose the slower speed to add
motion blur.
While
the downside of CineFrame 30 is reduced effective vertical resolution
on moving objectsÑthe upside is an image with no interlace combing because
the video, while not progressive, has no interlace artifacts. Although it is ideal for video that will be shown on computer
screensÑit may not be ideal for shooting interlace material because
of motion strobing. Some, of course, may feel the lower temporal
resolution and moderate strobing creates an in-camera film look.
In
theory, the eye will not notice the loss of resolution because moving
objects are blurred anyway. However, it remains to seen how the
deinterlacer handles diagonal-lines that are in motion. Typically,
these lines take on a staircase look.
The
FX1e and Z1 provide CineFrame 25 mode that deinterlaces 1080i50
to provide 25fps video. CineFrame 25, using a 1/50th
second shutter-speed, is ideal for shooting video that will be transferred
to film.
8. According
to Sony USA, in CineFrame 24 mode, the camcorder uses DXP (i.e.,
DSP) to Òsynthesize a 24Hz temporal rate video from 1080i60.Ó According
to Sony, the same intelligent deinterlacing employed in CineFrame
30, is used to create 1080 video with 30 frames-per-second. Then,
according to Sony, 24fps is synthesized
from the 30fps video. (Region 50 camcorders do not offer CineFrame
24.)
To
record 24fps video to tape as 1080i60, pulldown must be applied.
When 2:3:2:3 pulldown is applied to the 24 samples, 1080i60 video
is generated. As shown in Diagram 11,
2:3:2:3 pulldown is applied to the 24 samples to generate 1080i60
video. (Six times each second, four samples are converted to five
frames, yielding 30 frames.) By applying pulldown, 24fps video is
carried as 1080i60 HDV video. With 2:3:2:3 pulldown, two judder
frames (number 3 and number 4) are included within every five video
frames.

Given
the difficulty of converting 30fps to 24fps, it is possible that
Sony has not yet fully disclosed how 24fps video is generated. (That
may await a presentation at Sundance.) There are several alternate
schemes that could be used:
a)
The imaging system could be clocked at 48Hz, rather than 50Hz or
60Hz, yielding 1080i48 interlaced video. Then intelligent deinterlacing
would be used to create 24 frames-per-second video that has no interlace
artifacts. Once 24fps is obtained, 2:3:2:3 pull-down is applied.
This would yield frames that are exactly 1/24th
second apart. However, were the Sony to use this method, the available
shutter-speed should be either 1/48th or 1/50th
second. It is not.
b)
A very intriguing possibility is that
deinterlacing, rate conversion, and 2:3:2:3 pulldown are done during
MPEG-2 encoding. By measuring motion between fieldsÑusing
the encoderÕs motion tracking logicÑthe following can occur:
I-frame:
if "objects" do not move between fieldsÑboth even and
odd lines within such objects are encoded thereby supplying these
objects with full effective vertical resolution. When objects moveÑonly
their odd field is encoded so such objects have half their possible
resolution.
P and B-frame:
if "objects" do not move between fieldsÑthey simply are
not encoded. When objects moveÑonly their odd or even field
are encoded so the objects have half their possible resolution.
Fields are distributed via 2:3:2:3 pulldown in a way that results
in moving objects having a frame-rate of 24fps. Naturally, the encoded
video can be played on 1080i60 monitorsÑor down-converted to 480i
or 480p. ws how this could be done.

Green
text indicates fields that are copies of odd fields. Blue text indicates
fields that are copies of even fields. Red text indicates judder
(split) frames. The pattern of BPB above is repeated three additional times
(plus a final B) in every 15-frame GOP (IBBPB BPB BPB BPB
B-I).
By
choosing fields from 60i interlace video that temporally match where
24fps would occurÑ60i video is converted to 24fps video. The period
between samples that should be exactly 1/24th second
apartÑare not perfectly equally spaced in this implementation.
When
CineFrame 24 is selected, you can must a shutter-speed 1/60th.
You should not shoot in AUTO or use AE mode as the shutter-speed
may rise above 1/60th and cause excessive strobing. (Unfortunately,
the FX1 and Z1 do not offer the option of a 1/30th shutter-speed
option.)
The
advantages of Sony's approach to 24fps video are twofold: light
sensitivity is increased by 6dB (1 stop) while image noise is reduced
compared to utilizing progressive scanning. It remains to be seen
how the relatively high shutter-speed of 1/60th second
will affect shooting for film. Shooting CineFrame 25 has the advantage
of a 1/50th shutter-speed, which is very close to the
typical film shutter-speed of 1/48th second. While some
may feel CineFrame 24 has a film look, the mode may best used with
reverse 2:3:2:3 pulldown that can be used to recover 24 frames in
each second of video.
In
theory, the eye will not notice the loss of resolution because moving
objects are blurred anyway. However, it remains to seen how the
deinterlacer handles diagonal-lines that are in motion. Typically,
these lines take on a staircase look.
Diagram
12 shows how Reverse 2:3:2:3 pulldown can be applied to CineFrame
24 video. Clearly, the four samples can be recovered.

9. One of the
most important features of the HDR-FX1 is its 3.5-inch, widescreen
color LCD that has 250,800 pixels.
The color LCD viewfinder has 252,000 pixels. To aid focusing, you
can push a button and the LCD resolution is zoomed by a factor of
four. Very clever.
Although
the HDR-FX1 looks quite large, it weighs only 2kg (4.4 pounds) Ñwithout
battery and tape. Analog component output, i.LINK, and a LANC port
are some of the connections supported by the FX1. Although a 720p30
tape can be played and output via component analogÑthe 720p30 Transport
Stream cannot be output by the FX1Õs i.LINK port.
Three
batteries are available: an NP-F570 provides 65 minutes of recording
time in both HDV and DV mode; an NP-F770 provides 130 minutes of
recording time in HDV mode and 240 minutes of recording time in
DV mode. The NP-F970 provides a maximum of 205 minutes of recording
in HDV mode and 215 minutes of recording time in DV mode.
The
FX1 not only looks to be an accomplished DV camcorder, it offers
"over-sampled" NTSC video. Moreover, the Z1 offers the
unique ability to shoot NTSC and PAL DV plus 1080i60 and 1080i50.
In fact, Sony notes there are over forty enhancements provided by
the HVR-Z1.
Without
doubt, the HDR-FX1 and HVR-Z1 will be the camcorder of 2005. It
will initiate the rapid move to HD the way the Sony VX1000 drove
the move from analog to digital.