Re: [MMUSIC] Scope of RTP payload types in BUNDLE?

[Paul Kyzivat <[e-mail]>]

Colin,

As I suggested on another thread, ISTM that it should be possible to 
reuse the same PT to map to different payload formats in different 
m-lines of a bundle *if* there is some other characteristic declared in 
SDP and present in packets that can be used to associate the packet to 
one m-line. In that case, after picking the m-line, the mapping from PT 
to payload format for *that* m-line can be used. E.g.,

    v=0
    o=alice 2890844526 2890844526 IN IP4 host.anywhere.com
    s=
    c=IN IP4 host.anywhere.com
    t=0 0
    a=group:bundle X Y
    m=audio 49170 RTP/AVP 96
    a=mid:X
    a=ssrc:1111 cname:[e-mail]
    a=rtpmap:96 AMR-WB/16000
    m=audio 49172 RTP/AVP 96
    a=mid:Y
    a=ssrc:2222 cname:[e-mail]
    a=rtpmap:96 G7291/16000

	Thanks,
	Paul


On 5/27/13 8:29 AM, Colin Perkins wrote:
> There were a number of comments in the call last week, and on the list, about unique payload types in BUNDLE. I'd like to explore this further.
>
> Case A: Within a single RTP session, I think we'd all agree that an offer that uses the same RTP payload type for two payload formats on a single m= line is problematic:
>
>     v=0
>     o=alice 2890844526 2890844526 IN IP4 host.anywhere.com
>     s=
>     c=IN IP4 host.anywhere.com
>     t=0 0
>     m=audio 49170 RTP/AVP 96
>     a=rtpmap:96 AMR-WB/16000
>     a=rtpmap:96 G7291/16000
>
> If this were done the receiver would have no way of distinguishing what payload format is meant by payload type 96. Accordingly, unique payload formats need to be used for each payload format.
>
> Case B: If one were to use two separate m= lines on different ports, in the non-BUNDLE case, then the same RTP payload type can be reused without difficulty:
>
>     v=0
>     o=alice 2890844526 2890844526 IN IP4 host.anywhere.com
>     s=
>     c=IN IP4 host.anywhere.com
>     t=0 0
>     m=audio 49170 RTP/AVP 96
>     a=rtpmap:96 AMR-WB/16000
>     m=audio 49172 RTP/AVP 96
>     a=rtpmap:96 G7291/16000
>
> The implication is that there are two separate RTP sessions, which the receiver can distinguish based on the UDP port on which the packets are received. The mapping from RTP payload types to payload formats is done on a per-RTP session basis.
>
> Case C: This is when we BUNDLE several m= lines on a single UDP port. On the call last week, there seemed to be agreement that the m= lines in such a BUNDLE group comprise a single RTP session (it also matches the definition of an RTP session from RFC 3550: the packets go to a single destination and the m= lines share a single SSRC space). To me this would suggest that the RTP payload type values MUST be unique across the m= lines (or, if the same RTP payload type is used in different m= lines, it MUST map to an identical payload format). The reason is that everything runs over one UDP port, and if a single RTP payload type is mapped to different payload formats, there's no way to distinguish which payload format is meant. This is essentially the same as case A above.
>
> Given this, it's not clear to me that non-unique payload types make sense within a BUNDLE group. They work on different m= lines in the legacy case because those lines form different RTP sessions, and the RTP sessions can be used to scope the payload type to payload format mappings. I don't see that they work in BUNDLE, because there's a single RTP session, and so a single mapping from payload type to payload format.
>