[opendtv] Re: Half Truths - Was More 1080p@60

At 3:54 PM -0500 12/4/07, Manfredi, Albert E wrote:

Craig Birkmaier wrote:


 The EBU studies provide the correct technical basis to justify
 this, and their conclusion is that we should use 720P for
 emission.


Not quite that simple.

The EBU studies I think you're referring to said that 1080 at 50p would
be the best emission option as long as bit rates could be above 10 to 13
Mb/s, using H.264 compression. They said that below that channel
capacity, 720 at 50p was preferable. And they said that 1080i was never
the best choice.
Yes Bert, we have been through all of this before. I guess you did not catch the part about program distributors using each new generation of compression technology to increase program capacity rather than maintaining program quality. At least the Europeans have established a floor for quality that is tolerable.
The notion of compressing HD to less than 10 Mbps is absurd. Talk about a low pass filter!
But the reality is that in many areas of the world this will happen. Average bit rates are not the problem. It is the lack of bandwidth to handle the peaks that is the real culprit behind most of the compression artifact that we see. What we DON'T see is the loss of detail that goes along with these very high compression rates. The guys that squeeze too hard - and that's EVERYONE in the U.S. - understand that people can tell when pictures break up into blocks. So they have learned to use low pass filters ahead of the encoder to reduce the encoder stress. Whne this happens, and it happens alot, you might as well be watching NTSC. I remember the early days of DirecTV when they were trying to squeeze some movies into about 2-3 Mbps ; some of those old westerns looked more like cartoons than film or video because so much detail had been removed.
What the EBU is saying is that you need >13 Mbps just to maintain the AVERAGE level of picture quality at 1080P at a level that is comparable to 720P at the same bit rate. Now WHAT happens with content that has lots of peaks, like sports? Do you honestly believe that the 1080P material is going to hold up as well as 720P material that has about half the samples and motion vectors to deliver? What will happen is that the low pass filters will kick in for the 1080P source and you will get LESS detail than in the 720P version. 



So, the way I read this Ambarella paper is that maybe they are tweaking
H.264 efficiency to the point that the future the EBU was talking about
is now. The numbers they claimed seemed way optimistic, but on the other
hand, the numbers in the EBU report did not seem subsanttially different
from what MPEG-2 compression can do (namely, they used 18, 16, 13, 10, 8
and 6 Mb/s). So the truth probably lies somewhere in the middle.
What they are saying is that they can run an H.264 encoder in closed loop with noise reduction and low pass filters to reach these very aggressive bit rates. This is EXACTLY what happened with SD when the customers got the boxes and set the knobs to squeeze out every bit until the viewers started cancelling their service.
You've got to be kidding about MPEG-2 being comparable to H.264. Just because MPEG-2 is being used deliver crappy HD video today is not a justification for this practice to continue, much less a justification for even more aggressive compression. At least the EBU is trying to determine what is really required to justify using HD in the first place. European broadcasters are using about 2x the number of bits for their SD services when compared with countries like the U.S. I expect they will do the same with HD, while we will get over-compressed HD with quality no better than progressive EDTV. 

How can they get away with this?

Display Oversampling.
The CE guys will flood the market with 1080P displays that are smaller than 60" diagonal. At these smaller sizes most of the quantization noise and artifacts will be suppressed by the inability of the viewer to resolve all of the potential detail in the 1080P formats. And when the low pass filters kick in, the source will still have enough resolution to saturate these smaller displays. But those with 60" and larger displays will see that the stuff they are watching does not even come close to the real potential of 1080P. 



If the Arambella numbers represent some sort of real breakthrough in
H.264 tweaking, then I would think H.264 and a 1080 at 60p transmission
format would also make sense for the ATSC to consider.


Why am I not surprised.

Now consider this...
The Ambarella press release talks about the need for broadcasters to catch up with the capabilities of the displays that the CE industry is selling today. They claim that acquisition gear is up to the task. But the reality is that there are only a few first generation 1080@50/60P cameras available today, and these cameras cannot come close to capturing the full potential of the 1080P format. At best, the noise floor is hit a little above 20 MHz, far short of the 30MHz container for 1080P.
Now what happens to the Ambarella arguments as cameras improve and it is possible to capture more of the detail potential for the 1080P formats? Downsampling to 720P produces even better quality samples with less entropy, thus the 720P encoding efficiency improves. But compressing this extra detail at 1080P requires MORE BITS, thus the >10-13 Mbps average may grow to 13-15 Mbps or higher.
Nothing that Ambarella is saying is some kind of breakthrough. It's all just physics, and the physics say that the EBU is correct. Use 1080P for acquisition, production and archiving. Use 720P to deliver higher quality HD. 

Regards
Craig

P.S.



On a side note, I noticed the small (19" and even less) LCD TVs on store
shelves are now no longer SD or even 1366 X 768. They are now 1440 X
900. This says that large TVs beyond 1080p are just around the corner.
That's when a 1080p transmission format may well start making sense, for
displays of, say, 50" and greater anyway.
Physics Bert. The bigger the screen, the easier it is to see the problems with the compression. 



This stuff is not going to remain static. The CE companies have to have
something new to sell you, after you've owned your set for three or four
years.
They are playing the same old game. Bigger numbers are ALWAYS BETTER. Unfortunatley with video compression this is simply not true. 

Do we need 1080P resolution in displays smaller than 50" diagonal? Perhaps.
Display oversampling is a good thing and it also helps when you try to display Non-Nyquist imagery such as Web content.
Do we need 1080P emission for a mass market OTA distribution system. Not even close... 


----------------------------------------------------------------------
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: