The Resource Group

1700 Pennsylvania Ave. NW Washington DC 20006 US +1-224-465-5954 Qiaobing.Xie@gmail.com

General Audio Video Transport WG Internet-Draft This document defines a simple enhancement to RFC 2198 to support RTP sessions with forward-shifted redundant encodings, i.e., redundant data sent before the corresponding primary data. Forward-shifted redundancy can be used to conceal losses of a large number of consecutive media frames (e.g., consecutive loss of seconds or even tens of seconds of media).

The keywords MUST, MUST NOT, REQUIRED, SHALL, SHALL NOT, SHOULD, SHOULD NOT, RECOMMENDED, NOT RECOMMENDED, MAY, and OPTIONAL, when they appear in this document, are to be interpreted as described in RFC 2119 .

This document defines a simple enhancement to RFC 2198 to support RTP sessions with forward-shifted redundant encodings, i.e., redundant data is sent before the corresponding primary data. Forward-shifted redundancy can be used to conceal losses of a large number of consecutive media frames (e.g., consecutive loss of seconds of media). Such capability is highly desirable, especially in wireless mobile communication environments where the radio signal to a mobile wireless media receiver can be temporarily blocked by tall buildings, mountains, tunnels, etc. In other words, the receiver enters into a shadow of the radio coverage. No new data will be received when the receiver is in a shadow. In some extreme cases, the receiver may have to spend seconds or even tens of seconds in a shadow. The traditional backward-shifted redundant encoding scheme (i.e., redundant data is sent after the primary data), as currently supported by RFC 2198 , is known to be ineffective in dealing with such consecutive frame losses. In contrast, the forward-shifted redundancy, when used in combination with the anti-shadow loss management at the receiver (as described in ), can effectively prevent service interruptions when a mobile receiver runs into such a shadow.

Regardless of the direction of time shift (e.g., forward-shifting or backward-shifting as in RFC 2198) or the encoding scheme (e.g., FEC, or non-FEC), there is always the option of sending the redundant data and the primary data either in the same RTP session (i.e., inband) or in separate RTP sessions (i.e., out-of-band). There are pros and cons for either approach, as outlined below. Inband Approach: Pro: A single RTP session is faster to setup and easier to manage. Pro: A single RTP session presents a simpler problem for NAT/firewall traversal. Pro: Less overall overhead - one RTP/UDP/IP overhead. Con: Lack of flexibility - difficult for middle boxes such as gateways to add/remove the redundant data. Con: Need more specification - special payload formats need to be defined to carry the redundant data inband. Out-of-band Approach: Pro: Flexibility - redundant data can be more easily added, removed, or replaced by a middle box such as a gateway. Pro: Little or none specification - no new payload format is needed. Con: Multiple RTP sessions may take longer to setup and more complexity to manage. Con: Multiple RTP sessions NAT/firewall traverse are harder to address. Con: Bigger overall overhead - more than one RTP/UDP/IP overhead. It is noteworthy that the specification of inband payload formats such as this and RFC 2198 does not preclude a deployment from using the out-of-band approach. Rather, it gives the deployment the choose to use whichever approach deemed most beneficiary under a given circumstance.

In RFC 2198, the timestamp offset in the additional header corresponding to a redundant block is defined as a 14 bits unsigned offset of timestamp relative to timestamp given in the RTP header. As stated in RFC 2198:

"The use of an unsigned offset implies that redundant data must be sent after the primary data, and is hence a time to be subtracted from the current timestamp to determine the timestamp of the data for which this block is the redundancy."

This effectively prevents RFC 2198 from being used to support forward-shifted redundant blocks. In order to support the use of forward-shifted redundant blocks, the media type "fwdred" which allows an optional MIME parameter, "forwardshift", is introduced for indicating the capability and willingness of using forward-shifted redundancy and the base value of timestamp forward-shifting. The base value of "forwardshift" is an integer equal or greater than '0'. In an RTP session which uses forward-shifted redundant encodings, the timestamp of a redundant block in a received RTP packet is determined as follows:

timestamp of redundant block = timestamp in RTP header - timestamp offset in additional header + forward shift base value

(The definition is based on media type "red" defined in RFC 2198 , with the addition of the optional "forwardshift" parameter.) audio fwdred none An unsigned integer can be specified as value. If this parameter is present with a value greater than '0', it indicates that the sender of this parameter will use forward shifting with a base value as specified when sending out redundant data. If this parameter is absent or present with a value of '0', it indicates that the sender of this parameter will not use forward shifting when sending its redundant data, i.e., the sender will have the same behaviors as defined in RFC 2198. This media type is framed binary data (see RFC 4288, Section 4.8) and is only defined for transfer of RTP redundant data frames specified in RFC 2198. See Section 6 "Security Considerations" of RFC 2198. None. RTP redundant data frame format is specified in RFC 2198. It is expected that real-time audio/video applications that want protection against losses of a large number of consecutive frames will be interested in using this type. none Qiaobing Xie <Qiaobing.Xie@motorola.com> COMMON This media type depends on RTP framing, and hence is only defined for transfer via RTP (RFC 3550 ). Transfer within other framing protocols is not defined at this time. Qiaobing Xie IETF Audio/Video Transport working group delegated from the IESG.

The information carried in the MIME media type specification has a specific mapping to fields in the Session Description Protocol (SDP) , which is commonly used to describe RTP sessions. When SDP is used to specify sessions employing the forward-shifted redundant format, the mapping is as follows: The media type ("audio") goes in SDP "m=" as the media name. The media subtype ("fwdred") goes in SDP "a=rtpmap" as the encoding name. The optional parameter "forwardshift" goes in the SDP "a=fmtp" attribute by copying it directly from the MIME media type string as "forwardshift=value". Example of usage of indicating forward-shifted (by 5.1 sec) redundancy:

m=audio 12345 RTP/AVP 121 0 5 a=rtpmap:121 fwdred/8000/1 a=fmtp:121 0/5 forwardshift=40800

Example of usage of indicating sending redundancy without forward-shifting (equivalent to RFC 2198):

m=audio 12345 RTP/AVP 121 0 5 a=rtpmap:121 fwdred/8000/1 a=fmtp:121 0/5 forwardshift=0

The optional "forwardshift" SDP parameter specified in this document is declarative, and all reasonable values are expected to be supported.

The registration of the new MIME subtype "fwdred", as described in , is required.

See Section 6 "Security Considerations" of RFC 2198 .

It is not unusual in a wireless mobile communication environment that the radio signal to a mobile wireless media receiver can be temporarily blocked by tall buildings, mountains, tunnels, etc. for a period of time. In other words, the receiver enters into a shadow of the radio coverage. When the receiver is in such a shadow no new data will be received. In some extreme cases, the receiver may have to spend seconds or even tens of seconds in such a shadow. Without special design considerations to compensate the loss of data due to shadowing, a mobile user may experience an unacceptable level of service interruptions. And traditional redundant encoding schemes (including RFC 2198 and most FEC schemes) are known to be ineffective in dealing with such losses of consecutive frames. However, the employment of forward-shifted redundancy, in combination with the anti-shadow loss concealment at the receiver, as described here, can effectively prevent service interruptions due to the effect of shadowing.

For anti-shadow loss management, the RTP sender simply adds a forward-shifted redundant stream (called anti-shadow or AS stream) while transmitting the primary media stream. The amount of forward-shifting, which should remain constant for the duration of the session, will determine the maximal length of shadows that can be completely concealed at the receiver, as explained below. Except for the fast that it is forward-shifted relative to the primary stream (i.e., the redundant data is sent ahead of the corresponding primary data), the design decision and trade-offs on the quality, encoding, bandwidth overhead, etc. of the redundant stream is not different from the traditional RTP payload redundant scheme. The following diagram illustrates a segment of the transmission sequence of a forward-shifted redundant RTP session, in which the AS stream is forward-shifted by 155 frames. If, for simplicity here, we assume the value of timestamp is incremented by 1 between two consecutive frames, this forward-shifted redundancy can then be indicated with:

forwardshift=155

and the setting of timestamp offset to 0 in all the additional headers. This can mean a 3.1 second of forward shifting if each frame represents 20 ms of original media,

Primary stream AS stream ... | | v v Pkt k+8 [ 111 ] [ 266 ] | | v v Pkt k+7 [ 110 ] [ 265 ] | | v v ^ Pkt k+6 [ 109 ] [ 264 ] | | | | v v Pkt k+5 [ 108 ] [ 263 ] T | | I v v M Pkt k+4 [ 107 ] [ 262 ] E | | v v Pkt k+3 [ 106 ] [ 261 ] | | v v Pkt k+2 [ 105 ] [ 260 ] | | v v Pkt k+1 [ 104 ] [ 259 ] | | v v Pkt k [ 103 ] [ 258 ] | | v v Transmit first Figure 1. An example of forward-shifted redundant RTP packet transmission.

The anti-shadow receiver is illustrated in the following diagram.

+---------+ normal mode sw1 | media | media Primary stream ======================o___o==>| decoder |===> output AS stream ---- +---------+ device | AS mode o | +---------+ | | | anti- | | ------->| shadow |---- | buffer | +---------+ | V expired frames discarded Figure 2. Anti-shadow RTP receiver.

The anti-shadow receiver operates between two modes - "normal mode" and "AS mode". When the receiver is not in a shadow (it can easily tell that if it is still receiving new data), the receiver operates in the normal mode. Otherwise, it operates in the AS mode.

In the normal mode, after receiving a new RTP packet that contains the primary data and forward-shifted AS data, the receiver passes the primary data directly to the appropriate media decoder for play-out (a de-jittering buffer may be used before the play-out, but for simplicity we assume none is used here), while the received AS data is stored in an anti-shadow buffer. Moreover, data stored in the anti-shadow buffer will be continuously checked to determine whether it has expired. If a redundant data in the anti-shadow buffer is found to have a timestamp older (i.e., smaller) than that of the last primary frame passed to the media decoder, it will be considered expired and be purged from the anti-shadowing buffer. The following example illustrates the operation of the anti-shadow buffer in normal mode. We use the same assumption as in Figure 1, and assume that the initial timestamp value is 103 when the session starts.

Timestamp Timestamp Time being of media in (in ms) played out AS buffer Note ------------------------------------------------------------------ t < 0 -- (buffer empty) ... t=0 103 258 (hold 1 AS frame) t=20 104 258-259 (hold 2 AS frames) t=40 105 258-260 (hold 3 AS frames) ... t=3080 257 258-412 (full, hold 154 AS frames) t=3100 258 259-413 (full, frame 258 purged) t=3120 259 260-414 (full, frame 259 purged) ... t=6240 415 416-570 (always holds 3.08 sec worth of redundant data) ... Figure 3. Example of anti-shadow buffer operation in normal mode.

At the beginning of the session (t=0), the anti-shadow buffer will be empty. When the first primary frame is received, the play-out will start immediately, and the first received AS frame is stored in the anti-shadow buffer. And with the arriving of more forward-shifted redundant frames, the anti-shadow buffer will gradually be filled up. For the example shown in Figure 1, after 3.08 seconds (the amount of the forward-shifting minus one frame) from the start of the session, the anti-shadow buffer will be full, holding exactly 3.08 seconds worth of redundant data, with the oldest frame corresponding to t=3.1 sec and youngest frame corresponding to t=6.18 sec. And it is not difficult to see that in normal mode because of the continuous purge of expired frames and the addition of new frames, the anti-shadowing buffer will always be full holding the next forward-shift amount of redundant frames.

When the receiver enters a shadow (or any other conditions that prevent the receiver from getting new media data), the receiver switches to the anti-shadow mode, in which it simply continues the play-out from the forward-shifted redundant data stored in the anti-shadow buffer. For the example in Figure 3, if the receiver enters a shadow at t=3120, it can continue the play-out by using the forward-shifted redundant frames (ts=260-414) from the anti-shadow buffer. As far as the receiver can move out of the shadow by t=6240, there will be no service interruption. When the shadow condition ends (meaning new data starts to arrive again), the receiver immediately switches back to normal mode of operation, playing out from newly arrived primary frames. And at the same time, the arrival of new AS frames will start to re-fill the anti-shadow buffer. However, if the duration of the shadow is longer than the amount of forward-shifting, the receiver will run out of media frames from its anti-shadow buffer. At that point, service interruption will occur. Anti-shadow loss concealment described above can be readily applied to the streaming of pre-recorded media. Because of the need of generating the forward-shifted anti-shadow redundant stream, to apply anti-shadow loss concealment to the streaming of live media will require the insertion of a delay equal to or greater than the amount of forward-shifting at the source of media.