[opendtv] Re: 1080P Question
- From: "Stessen, Jeroen" <jeroen.stessen@xxxxxxxxxxx>
- To: "opendtv@xxxxxxxxxxxxx" <opendtv@xxxxxxxxxxxxx>
- Date: Thu, 10 Sep 2009 09:09:06 +0200
Hello,
Mike Tsinberg wrote:
Ø No, common mistake created by marketing.
And you thought that you should add to the confusion ? ;-)
Ø Both current HDTV (rec 709) and xvYCC are 8 bit systems.
Ø The color gamut for HDTV, of each Red, Green and Blue colors is designed
within BT.709 standard for 16 to 236 level systems or 220 quantization levels.
16 to 236 is a legacy issue for component YCrCb.
It's 16..235 for Y' and 16..240 for Cb,Cr. And it's not a legacy issue, it is
an inherent necessity
for proper sampling and reconstruction. These are the reference levels for
nominal black and
white, i.e. for large areas. The other levels are used for the overshoots and
undershoots that
occur during proper anti-aliasing filtering. A filter without ringing would
have too little bandwidth.
You can't have proper (smooth) motion portrayal without anti-aliasing. If you
cut off the over-
and undershoots in order to limit the range, then that clipping is a non-linear
operation and it will
cause aliasing. There will be jittering and flickering on all sharp moving
edges.
Of course somewhere in the video path we still have to get rid of these extra
levels, particularly
the below-black levels. Traditionally this was done --after reconstruction-- in
the analog time-
continuous domain, where clipping will not cause aliasing anymore. Clipping off
the overshoot
was not necessary, as CRTs could handle the peak white. With digital pixelated
displays this
becomes a different story. Then the only solution seems to be that the spectrum
at the display
be not too full, i.e. an oversampled display or an underutilized transmission
channel.
Neither seem to be attractive. It helps that full-HD is already a bit better
than we really need,
so indeed we see much smoother motion with HDTV. I find this very pleasing.
Ø A fully digital interface, such as HDMI, is capable of a full 0 to 255
performance without increasing the 8 bits per color specification.
This is a common mistake created by the computer community. As in the beginning
they were
dealing only with static images, or perhaps images that move only in discrete
speeds of multiple
pixels per frame, aliasing was not yet an issue. Also, since most of the chain
was still analog
(CRT !) it was common to calibrate the black and white levels in the analog
domain (potmeters
in the CRT monitor). With an all-digital chain you need some headroom and
footroom for
calibration. It's not wise to use the full (8-bits) range for your signal over
the entire chain.
Ø The xvYCC is a new color gamut standard that uses the full range of values
from
Ø 0 to 255 or 256 quantization levels in an 8-bit space to represent colors. I
included a
Ø picture of color space comparison.
Technically the levels Y' = 0 and Y' = 255 are probably still illegal, as they
are reserved for
synchronisation. Besides that, the illustration is plain wrong. Even the
co-ordinates for the
Rec.709 primaries are very very wrong. And the color space of xvYCC can not
even be
described as a simple triangle in (x,y) space. It depends on the luminance. If
you combine a
low value for Y' with large absolute values for Cb,Cr then the color space is
even infinite.
Or at least it goes way beyond the visible colors. Katoh and I have illustrated
it nicely during
our presentations at the HPA Tech Retreat 2007. I suggest that you look it up.
Essentially the extension of the xvYCC color space relative to Rec.709 has two
parts:
- formally illegal combinations of legal Y'CbCr values that will lead
to R'G'B' values
that are < 0 or > 1, which I like to call wide-gamut c.q. bright-gamut colors,
and
- all combinations of formerly illegal Y'CbCr values, that will lead
to even more
extreme R'G'B' values.
It's only the latter colors, typically very bright colors, that eat into the
space that should have
been reserved for ringing (anti-aliasing). Typically they represent fluorescent
objects or
light sources. Yes, this does allow a bit more "sparkle" in the picture, but
they are not so
relevant to the wide-gamut issue. They can be avoided by proper scaling of the
signal range.
Ø Deep color increases bit depth from 24 bits per pixel (8 bit per color) to
36 bits
(12 bit per color). Either 709 or xvYCC can be used with deep color. Perhaps
"deep color" is misleading. It should be called probably "accurate or deep gray"
or something.
Correct about the meaning of deep color. It is my opinion that a video chain
allows only
one 8-bit bottleneck, and then only if the original signal comes with enough
(analog ?!)
noise to dither away the quantisation error of the 8-bit quantizer. If there
are multiple
processing steps with multiple 8-bit bottleneck between them, then the
quantisation noise
will soon be unacceptable. Therefore, every interface that can be upgraded to
more than
8-bits is very welcome. Internal interfaces (i.e. on a chip or on a PCB) are
easy. The
external HDMI interface was a logical candidate for an upgrade of the standerd.
It used to be that contrast control was in the analog domain (again: CRT), so
the video
path before that had a constant amplitude and 8-bits would do. Now that
everything has
been moved to the digital domain, we need more precision at the display input.
That
picture is changing again with the introduction of backlight dimming, where
extra bits
can be carried over the backlight intensity signal. And it will be changing
again with the
introduction of High Dynamic Range signals and displays, where an entire
intensity map
(of low resolution) will be carried in parallel with the video path. But most
likely these
signals will not be separated over the HDMI interface, so it is there that a
high precision
will remain essential for a combined HDR signal.
By the way... HDMI has always been capable of 12-bits precision. That's because
it
re-uses the hardware of the DVI interface, which can do 8-bits R'G'B'. HDMI can
do
(also) Y'CbCr 4:2:2, and with 12-bits components this also adds up to an
average of 24
bits per pixel. It costs nothing ! The "deep color" extension means that this
can also be
enabled for 4:4:4 components, but at the cost of more bits per second over the
link.
Groeten,
-- Jeroen
Jeroen H. Stessen
Specialist Picture Quality
Philips Consumer Lifestyle
Advanced Technology (Eindhoven)
High Tech Campus 37 - room 8.042
5656 AE Eindhoven - Nederland
Office mobex: (27) 99650
VoipBuster: +31.85.785.3611 (VoIP-in to PC)
Mobile 1: +31.6.1389.2492 (office gsm)
Mobile 2: +31.6.4468.0021 (personal gsm)
E-mail 1:
Jeroen.Stessen@xxxxxxxxxxx<mailto:Jeroen.Stessen@xxxxxxxxxxx>
E-mail 2:
Jeroen.Stessen@xxxxxxxxxx<mailto:Jeroen.Stessen@xxxxxxxxxx>
Website: www.consumer.philips.com<http://www.consumer.philips.com>
________________________________
The information contained in this message may be confidential and legally
protected under applicable law. The message is intended solely for the
addressee(s). If you are not the intended recipient, you are hereby notified
that any use, forwarding, dissemination, or reproduction of this message is
strictly prohibited and may be unlawful. If you are not the intended recipient,
please contact the sender by return e-mail and destroy all copies of the
original message.
Other related posts:
