[opendtv] Re: Receiver costs too expensive in the Brazilian DTV system:
- From: Tom Barry <trbarry@xxxxxxxxxxx>
- To: opendtv@xxxxxxxxxxxxx
- Date: Tue, 11 Sep 2007 12:50:14 -0400
Craig Birkmaier wrote:
> Cross fades and fade-to-block (black) are the most difficult
> pathological cases for any digital compression algorithm. Cross
> dissolves would be better accomplished by sending two overlapping
> streams and doing the transition in the receiver. Fades should be done
> in the receiver as this only requires attenuation of a single stream.
> The fact that this is not being done is a strong indication that content
> providers are fighting the logical approach to do more with the
> processing power in the receiver/STB. But I digress...
I've suggested before that even for MPEG-2 a good encoder could probably
recognize and approximate fades by linearly decreasing the luma in all
pixels of any block. The differences due to fade after motion
compensation and dct would then be restricted to only the average luma,
1 component out of 96 in 4:2:0 mpeg-2 encoding. A newer codec could
even implement global luma compensation.
- Tom
At 9:04 PM -0400 9/10/07, Albert Manfredi wrote:
So just for grins, I just now conducted a close inspection of the
WETA-DT multiplex, which is the most densely populated that I receive
so far. One HD and three SD channels. I was standing no more than 2'
from the 26" screen.
Bert has correctly identified the problem "he" is having relative to the
visibility of compression artifacts. He is watching all this digital
stuff on a screen that is incapable of resolving all of the detail (and
artifacts) at ANY viewing distance.At least he acknowledges that at 2
feet he can see a difference - but at a normal viewing distance he might
as well be watching NTSC.
The cuts seemed artifact-free to me, completely.
As they should be. The only time that one might expect to see artifacts
at a cut transition is if both the incoming and outgoing streams are bit
starved. Usually a new I-frame takes care of the problem.
They showed a closeup of the legs of many individuals on treadmills,
filling the screen. From that close viewing distance, yes, I could see
macroblocking on the legs as they moved quickly back and forth. But
that was the only time I noticed anything that obvious, and I'm sure
it would have been less obvious from a normal viewing distance.
Not on a 40" screen or larger, as is required to BEGIN to see the detail
in an HD source.
In one fade, a man's head came on while it was moving to one side. I
got an instantaneous macroblocking event, and then it was fine.
Cross fades and fade-to-block (black) are the most difficult
pathological cases for any digital compression algorithm. Cross
dissolves would be better accomplished by sending two overlapping
streams and doing the transition in the receiver. Fades should be done
in the receiver as this only requires attenuation of a single stream.
The fact that this is not being done is a strong indication that content
providers are fighting the logical approach to do more with the
processing power in the receiver/STB. But I digress...
Okay, I wasn't watching football, but what I saw, to me, was very
acceptable. And no comparison at all with NTSC. And remember, I was
right up close to the screen.
Yes. "most of the time" compressed digital is better than NTSC, unless
of course it is pre-filtered to make the stream fit into a multiplex, as
is the case with the sub-channels here in Gainesville.
The SD channels were probably well pre-filtered, because they seemed
to lack any obvious artifacts. And from up close, they were certainly
not as sharp as the HD channel, but again, no comparison at all with
NTSC, which is essentially unwatchable that close up.
Apples and oranges. NTSC has it's own set of artifacts that are visible
even with the very best reception. But it can deliver very good images.
Digital SD streams may show many of these artifact as well, as
interlacing is the major culprit here. But a prefiltered source may
eliminate many of the higher frequency vertical details that make the
interlacing artifacts visible. I have seen MANY MANY digital streams
that are far inferior to NTSC. Just tune into the CW sub-channel of the
ABC affiliate here in Gainesville and you can see this 24/7.
It's not about better quality images, its ALL about more channels
in less bandwidth to make more money. That's all it is about.
I think it is about both. People are demanding and will continue to
demand better quality simply because they are buying plasmas and LCDs
in droves, in sizes far bigger than their old CRTs. And I, for one,
welcome the greater choice as well.
This may well drive the demand for more HD programming, but it is doing
NOTHING to improve delivered image quality. The only digital sources
that are properly encoded today are DVDs. When broadcasters learn that
compression should be part of the production process - as the movie guys
have learned - rather than an emission process, we may see better
delivered image quality. Managing the emission multiplex is a data
processing problem. not an image processing or compression problem.
Programs should be delivered pre-compressed with good quality, and the
broadcasters should just deliver the bits.
Imagine buying a software application by downloading the bits via the
Internet. Then having the seller tell you that some of the features of
the application may not work because of IP packet errors. It is EASY to
use digital compression to deliver high quality - the current approach
being used by ALL program distributors is the problem - they do not
understand how to properly implement a digital TV delivery system.
My conclusion is that with MPEG-2, while more perfect 1080i might
require an average 16 Mb/s, this is by no means a hard number. I'm not
saying that a comparison with H.264 would show no difference. I'm
simply saying that these are soft numbers. Yeah, if we try squeezing
that down to less than 1.5 or so Mb/s, MPEG-2 will fall apart
completely, and H.264 will not. But those images would not be so great
on most TV-size screens anyway.
subject line.
OF COURSE the numbers are all SOFT. This is an entropy coding system.
Bit rate requirements vary with scene complexity and motion content. The
proper way to manage a multiplex is to assure that all needed bits are
delivered by the time they are needed. If you know what the requirements
are for each sub-stream in advance this is relatively easy to accomplish
- it becomes a file delivery problem.
This is a bit more complex with live programming, in which case the
only way to assure good delivered image quality is to allow enough
overhead to deal with peak bit rate requirements. But most of what is
broadcast is NOT live.
Regards
Craig
----------------------------------------------------------------------
You can UNSUBSCRIBE from the OpenDTV list in two ways:
- Using the UNSUBSCRIBE command in your user configuration settings at
FreeLists.org
- By sending a message to: opendtv-request@xxxxxxxxxxxxx with the word
unsubscribe in the subject line.
--
Tom Barry trbarry@xxxxxxxxxxx
----------------------------------------------------------------------
You can UNSUBSCRIBE from the OpenDTV list in two ways:
- Using the UNSUBSCRIBE command in your user configuration settings at FreeLists.org
- By sending a message to: opendtv-request@xxxxxxxxxxxxx with the word
unsubscribe in the subject line.
Other related posts:
