[opendtv] Re: MPs back Ofcom stance on spectrum sale
- From: Craig Birkmaier <craig@xxxxxxxxx>
- To: opendtv@xxxxxxxxxxxxx
- Date: Thu, 24 May 2007 12:01:57 -0400
At 11:51 AM -0400 5/23/07, Manfredi, Albert E wrote:
Let me try again. If you calculate the maximum possible amount of
*information* that can be transferred to a display in 720 at 60p and in
1080 at 60p, you will find that the maximum possible is 2.25X greater in
1080 at 60p than it is in 720 at 60p.
Sorry Bert, but you still do not get it.
You are talking about the number of samples that are transferred to
the display over some unit of time. This has NOTHING to do with the
information content of those samples.
Here are two simple examples that may help with your lack of
comprehension of the concept.
Let's say we want to create a solid color field on a display.
Regardless of the number of samples that will be displayed, we need
only a three bytes of data to represent any color field with 8 bit
samples (you can call this RGB or YUV, whatever). We simply say that
all samples are the same.
With MPEG we would need more data to represent the color field, as we
would need to send the same "information" for each 8X8 DCT block, but
this would still be a very small amount of information.
At the other extreme we could try to encode a random noise pattern
where there is a strong probability that we will have a number of
sample values within the DCT blocks. Since there is NO CORRELATION
between these samples it is virtually impossible to compress such a
signal accurately.
A more realistic example would be a very fine texture - perhaps
shooting a coarse tweed jacket. Each DCT block would contain a wide
range of samples and a number of different high frequency components.
Any attempts to quantize such a block would result in a significant
loss of the high frequency detail. A lower resolution version of this
would contain less detail, thus less information. To make matters
worse, with a 1080@60P camera we would capture more information, but
with more entropy - i.e. noise would cause distortions in the samples
which would actually add to the encoding overhead. IF we resampled
this image to 720P we would reduce the entropy (sampling errors) and
the information content (high frequency details). Thus this image
would compress with significantly improved efficiency.
This means that if you use the same compression algorithm in each
example, the relationship of b/s between the two will be 2.25X, in the
*very worst* case.
NO. THis is only the relationship of the number of samples. It has
NOTHING to do with the information content.
So you don't need to wrap yourself around an axle with considerations
about image content. The number 2.25X assumes that the 720p and 1080p
images EACH contain as much detail as they possibly can. Each is used to
its full extent.
In this extreme case, both would be impossible to compress in a 209
Mbps emission channel. Fortunately, we usually have a great deal of
correlation in the samples that we can exploit with a compression
algorithm. As I mentioned in a previous post in this thread, Higher
resolution images MAY have better correlation of the samples
(although noise is an enemy here). And we are providing more samples
for the same image detail than a lower resolution image, thus the
amount of information in each block may be reduced if the information
is at lower frequencies - then again it could also be increased, as
the filtering of a lower resolution image may eliminate fine details.
Once again the number of samples in each image has little to do with
the information content.
Think of this as TOTAL INFORMATION CONTENT, which is then compressed
down to something much smaller by the compression algorithm. If you use
exactly the same compression algorithm in each case, it will compress
down by the same factor. Again, both images contain as much detail as
they possibly can, in their respective formats.
NOT TRUE. IT will depend on the amount of information and the amount
of entropy in each. IF the lower resolution image is downsampled from
the higher there will be less information and less entropy.
I said that when/as/if a compression algorithm comes along that provides
a true 2:1, or slightly better improvement, compared with H.262, THEN
the transmission of 1080p will be feasible in the existing RF channels.
This would ONLY be true if the information content and entropy did
not increase with the higher resolution format. Just the opposite is
likely to be true.
> 720P will compress more efficiently UNLESS the information
content of the 1080@60P source is equal to or less than the
information content of the 720P source. You can compress both
to the same bit rate, but you will throw away most of the
higher frequency detail in the 1080@60P source, which cancels
out the ONLY reason why we would want to deliver 1080@60P in
the first place.
SO WHAT? This assumes the same compression algorithm is used in both
cases. What you said there is completely obvious, Craig. Now change the
compression algorithm used in the 1080p case to something much better
than what you used in 720p, and what you said above will no longer hold.
SO WHAT? Why would you bother to try and send a higher resolution
format through the channel if you got the same quality or less than a
lower resolution format that compressed more efficiently, and/or
provides more overhead to deal with the peak bit rate requirements
that will cause more stress (artifacts) in the higher resolution
encoding.
THIS WAS THE ENTIRE POINT OF THE EBU DEMO. Somehow you completely missed it.
It seems to me that the PROPER place to place our attention in
terms of improving the overall quality of DTV should be at the
lower end of the spectrum - to improve the delivered quality of
the SD content that represents the vast majority of what is
delivered by broadcasters, cable and DBS today.
Legacy thinking. Just like we should only be concerned with improving on
the sound of 78 RPM records.
Sorry Bert but this is not legacy thinking The reality is that the
vast majority of content that is being delivered today and will be
delivered in the next decade will NOT be HDTV.
To be honest, it seems that the distributors could care less about
delivered image quality at ANY resolution. They are squeezing it as
hard as they can until people stop paying for the service.
Furthermore, the level of resolution should match the application
requirement. We do not need HDTV resolution for MANY applications.
What we need is good quality samples regardless of the resolution of
the format. Good quality images can be scaled for any display.
Artifacts just get magnified when we scale up.
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.
Other related posts:
