[opendtv] Re: Food for thought

  • From: Ron Economos <k6mpg@xxxxxxxxxxx>
  • To: opendtv@xxxxxxxxxxxxx
  • Date: Sat, 24 Feb 2007 04:29:07 -0800

Channel acquisition is a function of how often a
sequence header (or a sequence parameter set
in H.264) appears in the bitstream. A decoder
can't start decoding until it knows the format
(resolution, frame rate, etc.) of the bitstream.

Usually, a sequence header is sent at the start of
every GOP. So although it's true that channel
acquisition is not a necessarily a function of
GOP structure, in practice, it is.

The BBC is using the usual 1/2 second GOP
(actually 0.48 seconds, because it's 12 frames
at 25 fps).

I've attached an H.264 analyzer screen shot
of a BBC trial bitstream. On the right, it
shows the 12 frame GOP structure (I = blue,
P = red and B = yellow).

I've turned on the slice stucture display. This
shows that the frame contains six slices, which
is the mark of a 1st gen encoder (most likely
a Tandberg EN5990). Second generation
encoders are single slice per frame, which
is more efficient (since intra prediction and
the deblocking filter can't span slices).

BTW, the BBC HD multiplex is 24.128 Mbps
(64-QAM, 2/3 FEC and 1/32 GI).

Ron

John Shutt wrote:


Bert,
Although I think I'm overstuffed with all this food for thought, I would like to point out that even in the worst case scenario of a 30 frame GOP MPEG structure, an I frame comes along once every second for 30 FPS and once every 1/2 second for 60 FPS video.
We don't use anything longer than a 15 frame GOP for our stuff, and I don't know of anybody else who goes longer than 15 either. 

Tuning delay is not related to MPEG GOP structure.
As for your HD Radio tuning delay from the main channel to the secondary channel, look to your particular radio for the explanation (or the chipset it uses) of the 5 second delay in tuning a subchannel, not for anything inherent in the HD format. 

John
----- Original Message ----- From: "Manfredi, Albert E" <albert.e.manfredi@xxxxxxxxxx> 


Indeed, and this is a sensible point, IMO. For sure, encoders will
improve, as MPEG-2 encoders have. But I still think that practical
deployed systems will have to operate at less than their theoretical
ideal, simply because of the sync-up time problem.

A somewhat unrelated example is HD Radio. If you tune to a primary
channel, which comes also in analog, the tuning experience is fine. The
analog comes in right away, the digital seamlessly replaces the analog
in a few seconds.

Try tuning immediately to a subchannel, which has no analog equivalent.
I just don't think that kind of delay can be acceptable, in an
all-digital radio. The wait must be more than 5 seconds. I don't think
DTV tuning would be considered acceptable with that sort of delay
either, yet that is the primary mechanism which provides the codec
efficiencies (when quality is kept constant).

JPEG image

Other related posts:

  1. Home
  2. opendtv
  3. Archive
  4. 02-2007