[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

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



I'm wondering if this might be better abstracted into a payload- format independent capability and use a framework like that in draft- wenger-avt-avpf-ccm-02.txt?

Dave.

On Mar 1, 2006, at 5:10 PM, Ejzak, Richard P (Richard) wrote:

Hi all,
The editors of RFC 3267 bis are preparing a new draft for the upcoming meeting and we propose to add an optional capability to perform time alignment for AMR and WB-AMR. Unless there are significant objections to this addition, the editors propose to add time alignment to the new draft next week. Please provide any comments or concerns with the proposed new capability.


Time alignment is useful in certain gateway-to-gateway scenarios and certain gateway-to-VoIP-terminal scenarios. It is the ability for a receiving gateway to request that the encoder shift the time reference for encoding and packetization by a specified amount. When a receiving gateway with strict packet timing limitations, e.g., due to a 3G UMTS Iu interface, is connected to an encoder at a peer endpoint capable of responding to time alignment requests, e.g., with an AMR encoder associated with a PCM interface to the PSTN, this allows the endpoints to reduce delay in the speech path by as much as the 20 ms frame-block duration, depending on the degree of misalignment in the time reference.

Note that time alignment is an optimization that reduces delay by adjusting the sender's packet transmission phase so that the packet arrives just in time for consumption. When the delay/jitter statistics change substantially during a call, delay optimization will benefit from re-invocation of time alignment once the jitter buffer has adapted to the new delay/jitter statistics. However, to avoid repeated requests for time alignment it is recommended to issue a new time alignment request only after the delay/jitter statistics have stabilized. In environments where the delay or jitter continuously varies, the usefulness of time alignment is significantly reduced. In this case the receiver will be out of alignment most of the time, and sent requests may become dated even before the media sender has managed to implement them.

We propose to implement this feature with the following additions to RFC 3267 bis:

1) A new declarative payload format parameter "time-align=1" will inform the peer entity if the sender of the parameter is capable of receiving "time alignment requests" within speech frames.

2) An endpoint that wants to request a change in time alignment at its peer will send a group of time alignment requests in a series of speech frames using currently reserved values of the CMR field. Three values are used to indicate time shifts of 10 ms, 5 ms or 1 ms, respectively. These are transmitted in such a way as to give priority to actual changes in codec mode requests and to allow transmission of time alignment requests during DTX.

3) If the receiver of a group of time alignment requests is currently capable of changing the time alignment, it will send subsequent speech frames using a new (later) time offset, effectively introducing a gap in the media stream (unless it occurs during DTX).

4) The receiver will play out the new frames at the same relative time as previous frames, effectively reducing end-to-end delay to the extent of the time alignment shift implemented by the sender. Note that this assumes that there was and continues to be extra delay in the playout buffer that can be absorbed by the shift in time alignment.

This capability has the potential to be very useful in reducing critical end-to-end delay for certain call scenarios.

BR Richard Ejzak
Lucent Technologies

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

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