Re: [AVT] Proposed addition of time alignment for AMR in RFC 3267 bis

[Magnus Westerlund <magnus.westerlund at ericsson.com>]

Hi,

WG chair hat off!

Both David's and Chuck's email seem to me to relate to the same problem, 
which has to do with synchronized playback of multiple streams. Where I 
see two versions, multiple received streams need to be played backed at 
the same time, and the case where multiple units need to play back 
things in a synchronized way. Time alignment as existing today in 
circuit switched networks is not the same problem.

Let look at Time alignment in AMR. Its origin is the circuit switched 
GSM link for voice calls. That link enforces that speech is transmitted 
with a periodic time of 20 ms. The data must be available every 20 ms or 
you miss your time slot. As it is each base station that have this clock 
there would buffering of up to 20 ms in that base station to keep that 
schedule. To optimize this and remove this delay when possible a 
procedure "time-alignment" was created that allows the base station to 
request that the data delivery is done with a different phase to 
minimize the buffering delay.

Things worth noting about time-alignment. It is all about of providing 
data just in time. That exercise is difficult if there is varying delay 
between the transmitting party and the consuming party. That was not a 
problem for the circuit switched networks where also the links between 
the radio network controller and the base station had fixed delay. This 
becomes a bit more problematic in IP/UDP/RTP domain. My opinion is that 
time-alignment procedures in RTP is useless unless you apply the 
requirements of low delay variance to the underlying IP network. These 
requirements I think can be satisfied primarily in a well controlled 
environment, like a mobile operator's core network. Thus time alignment 
applicability is limited to certain scenarios.

The use of time-alignment does not remove the need for a jitter-buffer 
and to achieve high performance, and adaptive jitter-buffer. Which 
results in that you have some interacting mechanisms. The jitter-buffer 
will be part of the delay between the transmitting side and the 
consuming side. Thus any adjustment of the jitter-buffer that result in 
that delay changes results in a new optimal phase. This is an important 
consideration to handle.

I would also like to point out that time-alignment is only usable when 
the there is only one link having fixed time slots to adopt to between 
the AMR encoding entity and the decoding entity. So two GSM or UMTS 
links prevents this from being used.

I would also like to point out that any usage of time alignment or 
similar procedures becomes impossible in cases where one encoding is 
distributed to several receivers. It is strictly a procedure for point 
to point sessions. In multi-party environments one will have to rely on 
receiver side procedures.

If the WG thinks this feature with its limited applicability is worth 
pursuing, I would be in favor for a more general mechanism.

From a glance through of 
http://tools.ietf.org/wg/avt/draft-hdesineni-avt-avpf-ccm-pd-extn-00.txt
I think the authors of this draft is after doing something similar. 
However also in this draft I think there is need for a consideration on 
the applicability of the mechanism. What assumptions it makes of the IP 
transport network. And what possibilities to action the sender and the 
receiver can do to handle the problem(s).

Cheers

Magnus Westerlund

Multimedia Technologies, Ericsson Research EAB/TVA/A
----------------------------------------------------------------------
Ericsson AB                | Phone +46 8 4048287
Torshamsgatan 23           | Fax   +46 8 7575550
S-164 80 Stockholm, Sweden | mailto: magnus.westerlund at ericsson.com

_______________________________________________
Audio/Video Transport Working Group
avt at ietf.org
https://www1.ietf.org/mailman/listinfo/avt