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

Re: [AVT] retransmission payload format



Hi Colin,

I think we are in agreement on which solution is the more proper one in this case. Using one RTX payload type for each original PT. The original TS goes in the retransmitted packets RTP header exactly as is. All RTCP statistics for the retransmission stream will be screwed up and is simply declared as incorrect.

Does any one see any problems with the proposed approach? Also see comments below.

Regards

Magnus

Colin Perkins wrote:

--> Magnus Westerlund writes:
...

Using multiple payload types with different rates in the retransmission stream and having the original timestamp in the RTP header forces the RTP session of switching the rate also for the retransmission SSRC. And the number of switches can potentially be several for each switch in the original stream. If packets before and just after switch is retransmitted the following might occur: Packet A has one rate of the x's and Packet B of the y's.

Original stream x x A B y y y
Packet losses l l
Retransmission stream A1 B1 A2 B2 A3
RTX losses l l l

In the above scenario the retransmission streams switches 5 times very rapidly on the sender side. The receiver although only sees 2 switches. Still, I don't see this as a desired behavior. If one would use a single static RTX pt with its individual RTP TS the retransmission stream would not switch a single time. The reconstructed packets would simply have there correct RTP TS. So having larger or smaller synchronization problems when switching rate depends if the RTP TS is used to transport the original or not.

Okay, but you'd need to restrict the clock for the retransmission stream,
such that the timing of the original streams can be exactly deduced from
the clock in the retransmission.

Yes, the RTX packet RTP TS MUST be the same as the one in the original packet.


Having a RTP TS based on send time is nicer as it doesn't screw up the RTCP measurements and instead allows them to work as intended. If we sometimes in the future are going to do a new version of RTP we really should consider having two timestamps. One for media synchronization and another one for transport statistics.

However I don't know it these nice things is worth the four byte in extra overhead for each retransmitted packet. Especially as they aren't really needed for the retransmitted stream.

The RTCP jitter statistics are probably not useful for the retransmission
stream anyway?

Agree, and therefore there is no need to make the statistics correct.


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

--

Magnus Westerlund
Multimedia Technologies, Ericsson Research ERA/TVA/A
----------------------------------------------------------------------
Ericsson AB | Phone +46 8 4048287
Torshamsgatan 23 | Fax +46 8 7575550
S-164 80 Stockholm, Sweden | mailto: magnus.westerlund@era.ericsson.se



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