Network Working Group S. Wenger Internet Draft Y.-K. Wang Document: draft-wenger-avt-rtp-svc-03.txt T. Schierl Expires: April 2007 October 2006 RTP Payload Format for SVC Video Status of this Memo By submitting this Internet-Draft, each author represents that any applicable patent or other IPR claims of which he or she is aware have been or will be disclosed, and any of which he or she becomes aware will be disclosed, in accordance with Section 6 of BCP 79. Internet-Drafts are working documents of the Internet Engineering Task Force (IETF), its areas, and its working groups. Note that other groups may also distribute working documents as Internet- Drafts. Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress." The list of current Internet-Drafts can be accessed at http://www.ietf.org/ietf/1id-abstracts.txt. The list of Internet-Draft Shadow Directories can be accessed at http://www.ietf.org/shadow.html. This Internet-Draft will expire on April 20, 2007. Copyright Notice Copyright (C) The Internet Society (2006). INTERNET-DRAFT RTP Payload Format for SVC Video October 2006 Abstract This memo describes an RTP Payload format for the scalable extension of the ITU-T Recommendation H.264 video codec which is the technically identical to ISO/IEC International Standard 14496-10 video codec. The RTP payload format allows for packetization of one or more Network Abstraction Layer Units (NALUs), produced by the video encoder, in each RTP payload. The payload format has wide applicability, as it supports applications from simple low bit-rate conversational usage, to Internet video streaming with interleaved transmission, to high bit-rate video-on-demand. Wenger, Wang, Schierl Standards Track [page 2] INTERNET-DRAFT RTP Payload Format for SVC Video October 2006 Table of Content RTP Payload Format for SVC Video...............................1 1. Introduction..............................................5 1.1. SVC -- the scalable extensions of H.264/AVC................5 2. Conventions...............................................5 3. The SVC Codec.............................................6 3.1. Overview................................................6 3.2. Parameter Set Concept....................................7 3.3. Network Abstraction Layer Unit Header......................7 4. Scope...................................................11 5. Definitions and Abbreviations .............................11 5.1. Definitions............................................11 5.2. Abbreviations..........................................14 6. RTP Payload Format.......................................14 6.1. Design Principles.......................................14 6.2. RTP Header Usage........................................15 6.3. Common Structure of the RTP Payload Format................16 6.4. NAL Unit Header Usage...................................17 6.5. Packetization Modes.....................................18 6.6. Decoding Order Number (DON)..............................18 6.7. Single NAL Unit Packet..................................19 6.8. Aggregation Packets.....................................19 6.9. Fragmentation Units (FUs)................................19 6.10. Payload Content Scalability Information (PACSI) NAL Unit..19 7. Packetization Rules ......................................22 8. De-Packetization Process (Informative).....................22 9. Payload Format Parameters.................................22 9.1. MIME Registration.......................................23 9.2. SDP Parameters .........................................25 9.2.1. Mapping of MIME Parameters to SDP.......................25 9.2.2. Usage with the SDP Offer/Answer Model...................25 9.2.3. Usage with Session and SSRC multiplexing.................26 9.2.4. Usage in Declarative Session Descriptions................26 9.3. Examples...............................................26 9.4. Parameter Set Considerations.............................26 10. Security Considerations.................................26 11. Congestion Control......................................26 12. IANA Consideration......................................27 13. Informative Appendix: Application Examples................27 13.1. Introduction..........................................28 Wenger, Wang, Schierl Standards Track [page 3] INTERNET-DRAFT RTP Payload Format for SVC Video October 2006 13.2. Layered Multicast.....................................28 13.3. Streaming of an SVC scalable stream.....................29 13.4. Multicast to MANE, SVC scalable stream to endpoint........30 13.5. SSRC Multiplexing in case of using SRTP .................32 13.6. Scenarios currently not considered for complexity reasons.34 13.7. Scenarios currently not considered for being unaligned with IP philosophy...............................................34 14. Acknowledgements........................................36 15. References.............................................36 15.1. Normative References...................................36 15.2. Informative References.................................37 16. Author's Addresses......................................37 17. Intellectual Property Statement..........................38 18. Disclaimer of Validity..................................38 19. Copyright Statement.....................................38 20. RFC Editor Considerations................................39 21. Open Issues............................................39 22. Changes Log............................................39 Wenger, Wang, Schierl Standards Track [page 4] INTERNET-DRAFT RTP Payload Format for SVC Video October 2006 1. Introduction 1.1. SVC -- the scalable extensions of H.264/AVC This memo specifies an RTP [RFC3550] payload format for a forthcoming new mode of the H.264/AVC video codec, known as Scalable Video Coding (SVC). Formally, SVC will take the form of an Amendment to ISO/IEC 14496 Part 10 [MPEG4-10], and likely as one or more new Annexes of ITU-T Rec. H.264 [H.264]. It is planned to keep the technical alignment between the two mentioned specifications, as well as backward compatibility with previous versions of H.264/AVC. The current working draft of SVC is available for public review [SVC]. In this memo, SVC is used as an acronym for the mentioned scalable extensions of H.264/AVC. SVC covers all of H.264/AVC's applications, ranging from all forms of digital compressed video from, low bit-rate Internet streaming applications to HDTV broadcast and Digital Cinema applications with nearly lossless coding. This memo tries to follow a backward compatible enhancement philosophy similar to what the video coding standardization committees implement, by keeping as close an alignment to the H.264/AVC payload RFC [RFC3984] as possible. It basically documents the enhancements relevant from an RTP transport viewpoint, defines signaling support for SVC, and deprecates the single NAL unit packetization mode of RFC 3984. 2. Conventions The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in BCP 14, RFC 2119 [RFC2119]. This specification uses the notion of setting and clearing a bit when bit fields are handled. Setting a bit is the same as assigning that bit the value of 1 (On). Clearing a bit is the same as assigning that bit the value of 0 (Off). Wenger, Wang, Schierl Standards Track [page 5] INTERNET-DRAFT RTP Payload Format for SVC Video October 2006 3. The SVC Codec 3.1. Overview SVC provides scalable video bitstreams. In SVC, a scalable video bitstream contains a base layer conforming to the existing profiles of H.264 as defined in [H.264] and one or more enhancement layers. An enhancement layer may enhance the temporal resolution (i.e. the frame rate), the spatial resolution, or the quality of the video content represented by the lower layer or part thereof. The scalable layers can be aggregated to a single RTP packet stream, or transported independently. The concept of video coding layer (VCL) and network abstraction layer (NAL) is inherited from H.264. The VCL contains the signal processing functionality of the codec; mechanisms such as transform, quantization, motion-compensated prediction, loop filtering and inter-layer prediction. A coded picture of a base or enhancement layer consists of one or more slices. The Network Abstraction Layer (NAL) encapsulates each slice generated by the VCL into one or more Network Abstraction Layer Units (NAL units). Please consult RFC 3984 for a more in-depth discussion of the NAL unit concept. SVC specifies the decoding order of these NAL units. [Edt. Note: The definition of a ''coded picture'' is currently under discussion in JVT. For now, we apply the same definition as in the AVC specification within a give scalable layer. That is, a ''coded picture'' consists of all the coded slices having identical values of dependency_id, quality_level and redundant_pic_cnt, respectively, in one access unit.] The term ''Layer'' in Video Coding Layer and Network Abstraction Layer refers to a conceptual distinction, and is closely related to syntax layers (block, macroblock, slice, ... layers). ''Layer'' here describes a syntax level of the bitstream in contrast to the meaning of layer as a nested part of the bitstream which may be discarded. It should not be confused with base and enhancement layers. Wenger, Wang, Schierl Standards Track [page 6] INTERNET-DRAFT RTP Payload Format for SVC Video October 2006 The concept of temporal scalability is not newly introduced by SVC, as H.264 already supports it. In [H.264], sub-sequences have been introduced in order to allow optional use of temporal layers. [SVC] extends this approach by advertising the temporal layer information within the NAL unit header, or suffix NAL units, as discussed in section 3.3 and [SVC]. By our definition, the base layer may be scalable in the temporal dimension (only). The concept of scaling the visual content quality in the granularity of complete enhancement layers, i.e. through omitting the transport and decoding of entire enhancement layers, is denoted as coarse- grained scalability (CGS). This is what is commonly understood as scalability in the IETF community. According to SVC, a CGS layer may be a spatial or quality (SNR) enhancement layer. In some cases, the bit rate of a given enhancement layer may be reduced by truncating bits from individual NAL units. Truncation leads to a graceful degradation of the video quality of the reproduced enhancement layer. This concept is known as Fine Granularity Scalability (FGS). In SVC, FGS is provided by a concept known as progressive refinement slices. 3.2. Parameter Set Concept The parameter set concept is inherited from [H.264]. Please see section 1.2 of RFC 3984 for more details. In SVC, pictures from different layers may use the same sequence or picture parameter set, but may also use different sequence or picture parameter sets. If different sequence or picture parameter sets are used, then, at any time instant during the decoding process, there may be more than one active sequence or picture parameter set. Any specific active sequence parameter set remains unchanged throughout a coded video sequence in the layer in which the active sequence parameter set is referred to. The active picture parameter set remains unchanged within a coded picture. 3.3. Network Abstraction Layer Unit Header Wenger, Wang, Schierl Standards Track [page 7] INTERNET-DRAFT RTP Payload Format for SVC Video October 2006 An SVC NAL unit consists of a header of four bytes and the payload byte string. SVC extends by that the NAL unit header defined in [H.264] by three additional bytes. The header indicates the type of the NAL unit, the (potential) presence of bit errors or syntax violations in the NAL unit payload, information regarding the relative importance of the NAL unit for the decoding process, the layer decoding dependency information, and FGS fragmentation information. This RTP payload specification is designed to be unaware of the bit string in the NAL unit payload. The NAL unit header co-serves as the payload header of this RTP payload format. The payload of a NAL unit follows immediately. The syntax and semantics of the NAL unit header are formally specified in [SVC], but the essential properties of the NAL unit header are summarized below. The first byte of the NAL unit header has the following format (the bit fields are the same as in [H.264] and [RFC3984], while the semantics have changed slightly, in a backward compatible way): +---------------+ |0|1|2|3|4|5|6|7| +-+-+-+-+-+-+-+-+ |F|NRI| Type | +---------------+ F: 1 bit forbidden_zero_bit. H.264 declares a value of 1 as a syntax violation. NRI: 2 bits nal_ref_idc. A value of 00 indicates that the content of the NAL unit is not used to reconstruct reference pictures for inter picture prediction. Such NAL units can be discarded without risking the integrity of the reference pictures in the same layer. Values greater than 00 indicate that the decoding of the NAL unit is required to maintain the integrity of the reference pictures. Type: 5 bits Wenger, Wang, Schierl Standards Track [page 8] INTERNET-DRAFT RTP Payload Format for SVC Video October 2006 nal_unit_type. This component specifies the NAL unit payload type as defined in table 7-1 of [SVC], and later within this memo. For a reference of all currently defined NAL unit types and their semantics, please refer to section 7.4.1 in [SVC]. Previously, NAL unit types 20 and 21 (among others) have been reserved for future extensions. SVC is using these two NAL unit types. They indicate the presence of three more bytes as shown below. +---------------+---------------+---------------+ |0|1|2|3|4|5|6|7|0|1|2|3|4|5|6|7|0|1|2|3|4|5|6|7| +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |RR | PRID | TL | DID | QL|R|B|U|D|G|L| O | +---------------+---------------+---------------+ RR: 2 bits reserved_zero_two_bits. Reserved bits for future extension. RR MUST be zero. PRID: 6 bits simple_priority_id. This component specifies a priority identifier for the NAL unit. A lower value of PRID indicates a higher priority. TL: 3 bits temporal_level indicates the temporal layer (or frame rate) hierarchy. Informally put, a layer consisted of pictures of a smaller temporal_level value has a smaller frame rate. A given temporal layer typically depends on the lower temporal layers (i.e. the temporal layers with smaller temporal_level values) but never depends on any higher temporal layer. DID: 3 bits dependency_id denotes the inter-layer coding dependency hierarchy. At any temporal location, a picture of a smaller dependency_id value may be used for inter-layer prediction for coding of a picture of a larger dependency_id value, while a picture of a larger dependency_id value is disallowed to be used for inter-layer prediction for coding of a picture of a smaller dependency_id value. Wenger, Wang, Schierl Standards Track [page 9] INTERNET-DRAFT RTP Payload Format for SVC Video October 2006 QL: 2 bits quality_level designates the quality level hierarchy of a progressive refinement (PR) or quality (SNR) enhancement layer slice. At any temporal location and with identical dependency_id value, a picture with quality_level equal to ql uses a picture with quality_level equal to ql-1 for inter-layer prediction. R: 1 bit reserved_zero_bit. Reserved bit for future extension. R MUST be zero. B: 1 bit layer_base_flag indicates that no inter-layer prediction (of coding mode, motion, sample value, and/or residual prediction) is used for the current slice otherwise inter-layer prediction may be used. U: 1 bit use_base_prediction_flag indicates that the base representation of the reference pictures (i.e. only NAL units of the reference pictures with QL equal to zero are used for inter prediction) is used during the inter prediction process. D: 1 bit discardable_flag. A value of 1 indicates that the content of the NAL unit with dependency_id equal to currDependencyId is not used in the decoding process of NAL units with dependency_id larger than currDependencyId. Such NAL units can be discarded without risking the integrity of higher scalable layers with larger values of dependency_id. discardable_flag equal to 0 indicates that the decoding of the NAL unit is required to maintain the integrity of higher scalable layers with larger values of dependency_id. G: 1 bit fragmented_flag indicates that the current NAL unit is fragmented, which may be the case for partitions of an FGS (progressive refinement) slice. L: 1 bit last_fragemented_flag indicates, that the NAL unit is the last fragment of a fragmented NAL unit. Wenger, Wang, Schierl Standards Track [page 10] INTERNET-DRAFT RTP Payload Format for SVC Video October 2006 O: 2 bits fragemnet_order indicates the order in which the NAL units with fragmented_flag equal to 1 shall be ordered before the parsing process is started, starting from lower values. This memo introduces the same additional NAL unit types as RFC 3984, which are presented in section 6.3. The NAL unit types defined in this memo are marked as unspecified in [SVC]. Moreover, this specification extends the semantics of F, NRI, PRID, D, TL, DID and QL as described in section 6.4. 4. Scope This payload specification can only be used to carry the "naked" SVC NAL unit stream over RTP, and not the byte stream format according to Annex B of [SVC]. Likely, the applications of this specification will be in the IP based multimedia communications fields including conversational multimedia, video telephony or video conferencing, Internet streaming and TV over IP. This specification allows, in a given RTP session, to encapsulate NAL units belonging to o the base layer only, detailed specification in [RFC3984], or o one or more enhancement layers, or o the base layer and one or more enhancement layers 5. Definitions and Abbreviations 5.1. Definitions This document uses the definitions of [SVC] and [H.264]. The following terms, defined in [SVC], are summed up for convenience: scalable bitstream: An SVC compliant bit stream containing a base layer and at least one enhancement layer. suffix NAL unit: A NAL unit that immediately follows another NAL unit in decoding order and contains descriptive information of the preceding NAL unit, which is referred to as the associated NAL unit. Wenger, Wang, Schierl Standards Track [page 11] INTERNET-DRAFT RTP Payload Format for SVC Video October 2006 A suffix NAL unit shall have nal_ref_idc equal to 20 or 21, shall have dependency_id and quality_level both equal to 0, and shall not contain a coded slice. A suffix NAL unit belongs to the same coded picture as the associated NAL unit. A suffix NAL unit may be used for indicating temporal levels within the base layer. base layer: The base layer is typically representing the minimal spatial resolution and, or minimal quality of an SVC bitstream. The base layer must be fully complying with [H.264]. The base layer is independently decodable without the requirement of using any other layer of the SVC bitstream. In SVC context each slice NAL unit in the base layer is associated with a suffix NAL unit, which has a four-byte NAL unit header containing all the syntax elements described in section 3.3. [Edt. Note: The definition of ''base layer'' is not deadly clear, mainly because of temporal scalability. One definition is to call all the coded pictures in the lowest inter-layer coding hierarchy (i.e. having both dependency_id and quality_level equal to 0) as the base layer. This concept works perfectly if there is no temporal scalability. Another definition is to call all the coded pictures having temporal_level, dependency_id and quality_level all equal to 0 as the base layer. Yet another definition is to define the layer for which the bitstream of the scalable layer representation is non-scalable as the base layer. However, the absolutely non-scalable stream is the bitstream consisting of only one IDR picture having both dependency_id and quality_level equal to 0.] operation point: An operation point of a SVC bitstream represents a certain level of temporal, spatial and quality scalability. An operation point contains all NAL units required for restoring a valid bitstream (conforming to [SVC]) up to a certain SVC layer. The operation point is further described by simple_priority_id, temporal_level, dependency_id, and quality_level values of that layer. scalable enhancement layer: An SVC enhancement layer is identified by simple_priority_id, temporal_level, dependency_id, and quality_level as defined in [SVC] and summarized in section 3.3. Wenger, Wang, Schierl Standards Track [page 12] INTERNET-DRAFT RTP Payload Format for SVC Video October 2006 access unit: A set of NAL units pertaining to a certain temporal location. An access unit includes the slice data of the pictures of all scalable layers at that temporal location and possibly other associated data, e.g. SEI messages and parameter sets. coded video sequence: A sequence of access units that consists, in decoding order, of an instantaneous decoding refresh (IDR) access unit followed by zero or more non-IDR access units including all subsequent access units up to but not including any subsequent IDR access unit. IDR access unit: An access unit in which all the primary coded pictures are IDR pictures. Such an access unit allows for random access to any layer combination. IDR picture: A coded picture with the property that the decoding of this coded picture and all the following coded pictures in decoding order, with the same value of dependency_id, can be performed without inter prediction from any picture prior to the coded picture in decoding order with the same value of dependency_id. Thus an IDR picture allows for random access to the scalable layer, which it belongs to. An IDR picture causes a "reset" in the decoding process of the scalable layer containing the IDR picture. progressive refinement (PR) slice: A progressive refinement slice is contained in an SVC NAL unit that may be truncated since the end of the slice header for bit-rate and quality reduction. PR slices provide Fine Granularity Scalability (FGS). The following terms are itemized for clarification on RTP multiplexing strategies. For further information and discussion on RTP multiplexing, we refer to section 5.2 of [RFC3550]: RTP packet stream: A sequence of RTP packets with increasing sequence numbers, identical PT and SSRC, carried in one RTP session, and utilized to transport an integer number of SVC layers (which may be FGS scalable). Wenger, Wang, Schierl Standards Track [page 13] INTERNET-DRAFT RTP Payload Format for SVC Video October 2006 Single-Sender RTP Session: an (perhaps multicasted) RTP session in which all RTP packet streams in the session stem from entities that are in close cooperation, and can coordinate SSRC values. By definition, in Single-Sender RTP Sessions, SSRC collisions on the forward media path cannot occur. Note that, in practice, the ''entities in close cooperation'' likely run on the same machine and communicate through non-protocol means, or they communicate by protocols outside the RTP/SIP/SDP environment. Session multiplexing: The scalable SVC bitstream is distributed onto different RTP sessions, whereby each RTP session carries one RTP packet stream. Each RTP session requires a separate signaling and has a separate Timestamp, Sequence Number, and SSRC space. Dependency between sessions MUST be signaled according to [SDPsiglay]. SSRC multiplexing: The scalable SVC bitstream is distributed in a single RTP session, but that session comprises more than one RTP packet stream, identified by its SSRC. The use of SSRC multiplexing MUST be signaled according to [SDPsiglay]. 5.2. Abbreviations In addition to the abbreviations defined in [RFC3984], the following ones are defined. CGS: Coarse Granularity Scalability FGS: Fine Granularity Scalability 6. RTP Payload Format 6.1. Design Principles The authors observed the following design principles: o Backward compatibility with RFC 3984 wherever possible. o As the SVC base layer is H.264/AVC compatible, we assume the base layer (when transmitted in its own session) to be Wenger, Wang, Schierl Standards Track [page 14] INTERNET-DRAFT RTP Payload Format for SVC Video October 2006 encapsulated using RFC 3984. Requiring this has the desirable side effect that it can be used by RFC 3984 legacy devices. o MANEs are signaling aware and rely on signaling information. MANEs have state. o MANEs can terminate RTP sessions, and create different RTP sessions with perhaps modified content. This form of a MANE acts as an RTP mixer. Mixer-MANEs necessarily need to be in the SRTP security context. o MANEs can also perform very limited functionality, namely aggregate multiple RTP packet streams into a single RTP stream within the same session, by utilizing SSRC multiplexing. In this case, a MANE acts as a translator, and does not necessarily need to be in the security context. o Packet integrity needs to be preserved end-to-end (whereby end-to-end can mean endpoint to endpoint but also endpoint to MANE, if (and only if) the MANE acts as a Mixer). o In case of layered multicast transmission as motivated in section 13.2, SVC layers are transported in different RTP sessions (Session multiplexing). If the application should require a layered transmission on session level, the SVC layers are transported in different RTP packet streams within a single RTP session, each stream identified by a unique SSRC (SSRC multiplexing). SSRC multiplexing may further allow for adaptation of an RTP session in the security context, further discussion can be found in section 13.5. 6.2. RTP Header Usage Please see section 5.1 of RFC 3984 [RFC3984]. The following applies in addition. Wenger, Wang, Schierl Standards Track [page 15] INTERNET-DRAFT RTP Payload Format for SVC Video October 2006 When different layers of a SVC bitstream are transported over more than one RTP session, e.g. in layered multicast, for which the use case is given in 13.2, SSRC multiplexing, as described below, MAY be applied. When SSRC multiplexing is in use the same IP address and port number are shared between all RTP streams and all layers, while the relative importance for the decoding process of each RTP stream and/or layer is differentiated by the SSRC values. The SSRC value space is evenly allocated to a number of sub value spaces, with the number of sub value spaces being equal to the number of RTP packet streams forming the RTP session for which SSRC multiplexing is used. The first RTP packet stream conveying the lowest layers is mapped to the first sub SSRC value space with the lowest SSRC values, the second RTP packet stream conveying the second lowest layers is mapped to the second sub SSRC value space with the second lowest SSRC values, and so on. For the RTP packets of a certain RTP packet stream, the SSRC value is randomly selected from the corresponding sub SSRC value space. This way, a packet with a higher SSRC value contains data belonging to higher layers or layers of lower transport priority. SSRC multiplexing as discussed above, in conjunction with multicast from multiple senders requires that a) all streams SSRC multiplexed in the same session carry data of the same layered bitstream, and b) that the different senders are aware (by unspecified means of signaling) of the relative importance of the RTP packet streams they emit. Otherwise, it would be impossible to enforce the allocation of SSRC numbering spaces according to the importance for the decoding process. In other words, SSRC multiplexing as discussed above works only for Single-Sender RTP sessions. Note: in practice, it appears that SSRC multiplexing, due to the above limitation, results in requiring a single entity to send all RTP packet streams. No signaling means are currently available that would allow different senders to coordinate the SSRC value spaces to use. 6.3. Common Structure of the RTP Payload Format Please see section 5.2 of RFC 3984 [RFC3984]. Wenger, Wang, Schierl Standards Track [page 16] INTERNET-DRAFT RTP Payload Format for SVC Video October 2006 6.4. NAL Unit Header Usage The structure and semantics of the NAL unit header were introduced in section 3.3. This section specifies the semantics of F, NRI, PRID, D, TL, DID, QL, B, U, G, L, and O according to this specification. The semantics of F specified in section 5.3 of [RFC3984] also applies herein. For NRI, for the bitstream that is compliant with [H.264], the semantics specified in section 5.3 of [RFC3984] are applicable, otherwise only the semantics specified in SVC [SVC] is applicable. For PRID, the semantics specified in [SVC] applies. MANEs implementing unequal error protection may use this information to protect NAL units with smaller PRID values better than those with larger PRID values, for example by including only the more important NAL units in a FEC protection mechanism. The desirable transport priority increases as the PRID value increases. For D, MANEs may use this information to protect NAL units with D equal to 0 better than NAL units with D equal to 1. Furthermore a MANE or a receiver may determine whether a given NAL unit is required for successfully decoding a certain operation point of the SVC bitstream. For TL, DID and QL, in addition to the semantics specified in [SVC], according to this memo, values of TL, DID or QL indicate the relative priority in their respective dimension. A higher value of TL, DID or QL indicates a higher priority if the other two components are identical correspondingly. MANEs may use this information to protect more important NAL units better than less important NAL units. Informative note: PRID, D, TL, DID, and QL, in combination, provide complete information of the relative priority of a NAL unit compared to any other NAL unit. [Edt. note: examples may be provided in Informative Appendix 13 in future versions.] Wenger, Wang, Schierl Standards Track [page 17] INTERNET-DRAFT RTP Payload Format for SVC Video October 2006 For B, in addition to the semantics specified in [SVC], according to this memo, a MANE or receiver may use this information in order to identify the [H.264] conforming base layer NAL units (if marked by a suffix NAL unit) and may determine the temporal layer (by the TL value of the suffix NAL unit) of it. Thus it allows for generating an outgoing RTP stream, with a certain temporal scalability layer that conforms to [RFC3984] and [H.264]. For U, the semantics specified in [SVC] apply. For G, L and O, in addition to the semantics specified in [SVC], according to this memo, a MANE or receiver may detect a fragmented PR slice by G, L and O. Using this knowledge may let the MANE do FGS adaptation on the PR slice, by forwarding not all of the fragments in fragement_order (O). 6.5. Packetization Modes Please see section 5.4 of RFC 3984 [RFC3984]. The single NAL unit packetization mode SHALL NOT be used. Informative note: The non-interleaved mode allows an application to encapsulate a single NAL unit in a single RTP packet. Historically, the single NAL unit mode has been included into [RFC3984] only for compatibility with ITU-T Rec. H.241 Annex A. There is no point in carrying this historic ballast towards a new application space such as the one provided with SVC. More technically speaking, the implementation complexity increase for providing the additional mechanisms of the non-interleaved mode (namely STAPs) is so minor, and the benefits are so great, that we require STAP implementation. 6.6. Decoding Order Number (DON) Please see section 5.5 of RFC 3984 [RFC3984]. The following applies in addition. When different layers of a SVC bitstream are transported in more than one RTP packet stream (regardless of the use of session or SSRC multiplexing, or a combination thereof), the interleaved packetization mode MUST be used, and the DON values of all the NAL Wenger, Wang, Schierl Standards Track [page 18] INTERNET-DRAFT RTP Payload Format for SVC Video October 2006 units MUST indicate the correct NAL unit decoding order over all the RTP packet streams. If Session multiplexing is used, each session MUST signal the same value for the (marked as optional, but for this use case mandatory) MIME parameters sprop-interleaving-depth, sprop- max-don-diff, sprop-deint-buf-req, and sprop-init-buf-time. Further these values must be valid for the reception capabilities over all sessions. A receiver MUST signal the same (marked as optional, but for this use case mandatory) MIME parameter deint-buf-cap for all sessions used for Session multiplexing. 6.7. Single NAL Unit Packet Please see section 5.6 of RFC 3984 [RFC3984]. 6.8. Aggregation Packets Please see section 5.7 of RFC 3984 [RFC3984]. 6.9. Fragmentation Units (FUs) Please see section 5.8 of RFC 3984 [RFC3984]. 6.10. Payload Content Scalability Information (PACSI) NAL Unit A new NAL unit type is specified, and referred to as payload content scalability information (PACSI) NAL unit. The PACSI NAL unit, if present, MUST be the first NAL unit in an aggregation packet, and it MUST NOT be present in other types of packets. The PACSI NAL unit indicates scalability characteristics that are common for all the remaining NAL units in the payload, thus making it easier for MANEs to decide whether to forward or discard the packet. Senders MAY create PACSI NAL units and receivers can ignore them. Informative note: The NAL unit type for the PACSI NAL unit is selected among those values that are unspecified in the H.264/AVC specification and in RFC 3984 -- and therefore are ignored by receiver. Hence an SVC stream, even when including PACSI NAL units, can be processed with RFC 3984 receivers and H.264/AVC decoders. Wenger, Wang, Schierl Standards Track [page 19] INTERNET-DRAFT RTP Payload Format for SVC Video October 2006 When the first aggregation unit of an aggregation packet contains a PACSI NAL unit, there MUST be at least one additional aggregation unit present in the same packet. The RTP header fields are set according to the remaining NAL units in the aggregation packet. When a PACSI NAL unit is included in a multi-time aggregation packet, the decoding order number for the PACSI NAL unit MUST be set to indicate that the PACSI NAL unit is the first NAL unit in decoding order among the NAL units in the aggregation packet or the PACSI NAL unit has an identical decoding order number to the first NAL unit in decoding order among the remaining NAL units in the aggregation packet. The structure of PACSI NAL unit is exactly the same as the four-byte SVC NAL unit header specified in 3.3, and reproduced here once more for convenience:. +---------------+---------------+---------------+---------------+ |0|1|2|3|4|5|6|7|0|1|2|3|4|5|6|7|0|1|2|3|4|5|6|7|0|1|2|3|4|5|6|7| +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |F|NRI| Type |RR | PRID | TL | DID | QL|R|B|U|D|G|L| O | +---------------+---------------+---------------+---------------+ The values of the fields in PACSI NAL unit MUST be set as follows. o The F bit MUST be set to 1 if the F bit in at least one remaining NAL unit in the payload is equal to 1. Otherwise, the F bit MUST be set to 0. o The NRI field MUST be set to the highest value of NRI field among all the remaining NAL units in the payload. o The Type field MUST be set to 30. o The RR field or reserved_zero_two_bits field (2 bits) MUST be set to 0. o The PRID field MUST be set to the lowest value of the PRID values associated with all the remaining NAL units in the payload. Wenger, Wang, Schierl Standards Track [page 20] INTERNET-DRAFT RTP Payload Format for SVC Video October 2006 o The TL field MUST be set to the lowest value of the TL values associated with all the remaining NAL units in the payload. o The DID field MUST be set to the lowest value of the DID values associated with all the remaining NAL units in the payload. o The QL field MUST be set to the lowest value of the QL values associated with all the remaining NAL units in the payload. o The R field or reserved_zero_bit field (1 bit) MUST be set to 0. o The B field or layer_base_flag field (1 bit) MUST be set to 1 if the layer_base_flag associated with all the remaining NAL units in the payload is equal to 1. Otherwise, layer_base_flag MUST be set to 0. o The U field or use_base_prediction_flag field (1 bit)MUST be set to 1 if the use_base_prediction_flag associated with all the remaining NAL units in the payload is equal to 1. Otherwise, use_base_prediction_flag MUST be set to 0. o The D bit MUST be set to 0 if the D value associated with at least one remaining NAL unit in the payload is equal to 0. Otherwise, the D bit MUST be set to 1. o The G field or fragmented_flag field (1 bit) MUST be set to 1 if the fragmented_flag associated with all the remaining NAL units in the payload is equal to 1. Otherwise, fragmented_flag MUST be set to 0. o The L field or last_fragment_flag field (1 bit) MUST be set to 1 if the last_fragment_flag associated with all the remaining NAL units in the payload is equal to 1. Otherwise, last_fragment_flag MUST be set to 0. o The O field or fragment_order field (2 bits) MUST be set to the lowest value of frame_order associated with all the remaining NAL units in the payload. Wenger, Wang, Schierl Standards Track [page 21] INTERNET-DRAFT RTP Payload Format for SVC Video October 2006 7. Packetization Rules Please see section 6 of RFC 3984 [RFC3984]. The following rules apply in addition. The single NAL unit mode SHALL NOT be used. (See also section 6.5 for the motivation). When a suffix NAL unit is encapsulated for transmission, it SHOULD be aggregated to the same transmission packet as the NAL unit preceding the suffix NAL unit in decoding order. When different layers of a SVC bitstream are transported in more than one RTP packet stream, the interleaved packetization mode MUST be used. 8. De-Packetization Process (Informative) Please see section 7 of RFC 3984 [RFC3984]. The following rules apply in addition. [Edt. Do we need here more information about cross layer DON? Maybe in the next version.] 9. Payload Format Parameters [Edt. note: this section 9 and its subsections will be updated according to the changes listed below, a little later in the process. For now, we just list the adjustments necessary, so not to bury any new information in the RFC 3984 text.] Section 8 of [RFC3984] applies with the following modification. The sentence ''The parameters are specified here as part of the MIME subtype registration for the ITU-T H.264 | ISO/IEC 14496-10 codec.'' is replaced with Wenger, Wang, Schierl Standards Track [page 22] INTERNET-DRAFT RTP Payload Format for SVC Video October 2006 ''The parameters are specified here as part of the MIME subtype registration for the SVC codec.'' 9.1. MIME Registration Editor's note: this needs to be updated by copy-pasting the RFC 3984 MIME registration into this document, so to make it self-contained. Will be done later in the process. The MIME subtype for the SVC codec is allocated from the IETF tree. The receiver MUST ignore any unspecified parameter. Media Type name: video Media subtype name: H.264-SVC Required parameters: none OPTIONAL parameters: The optional MIME parameters specified in [RFC3984] apply, with the following constraints (to be edited in at the appropriate time): sprop-interleaving-depth: In case of using Session multiplexing, the same sprop-interleaving- depth value MUST be signaled for all sessions and MUST be valid over all sessions of the multiplex. sprop-max-don-diff: In case of using Session multiplexing, the same sprop-max-don-diff value MUST be signaled for all sessions and MUST be valid over all sessions of the multiplex. sprop-deint-buf-req: In case of using Session multiplexing, the same sprop-deint-buf-req value MUST be signaled for all sessions and MUST be valid over all sessions of the multiplex. sprop-init-buf-time: Wenger, Wang, Schierl Standards Track [page 23] INTERNET-DRAFT RTP Payload Format for SVC Video October 2006 In case of using Session multiplexing, the same sprop-init-buf-time value MUST be signaled for all sessions and MUST be valid over all sessions of the multiplex. deint-buf-cap: In case of using Session multiplexing, the same deint-buf-cap value MUST be signaled by the receiver for all sessions and MUST be valid over all sessions of the multiplex. In addition the following optional MIME parameters apply: sprop-scalability-info: This parameter MAY be used to convey the NAL unit containing the scalability information SEI message that MUST precede any other NAL units in decoding order. The parameter MUST NOT be used to indicate codec capability in any capability exchange procedure. The value of the parameter is the base64 representation of the NAL unit containing the scalability information SEI message as specified in [SVC]. sprop-transport-priority: This parameter MAY be used to signal the transport priority indicator value(s) in terms of second and third bytes of the SVC NAL unit header for one or more SVC layer(s) conveyed in one RTP session. A transport priority indicator is base64 coded. If more than one layer is transmitted within one RTP session, the transport priority indicator value of each layer MUST be itemized with decreasing importance for decoding and MUST be comma-separated. Encoding considerations: This type is only defined for transfer via RTP (RFC 3550). Security considerations: See section 9 of this specification. Public specification: Please refer to section 15 of this specification. Wenger, Wang, Schierl Standards Track [page 24] INTERNET-DRAFT RTP Payload Format for SVC Video October 2006 Additional information: None File extensions: none Macintosh file type code: none Object identifier or OID: none Person & email address to contact for further information: Intended usage: COMMON Author: Change controller: IETF Audio/Video Transport working group delegated from the IESG. 9.2. SDP Parameters 9.2.1. Mapping of MIME Parameters to SDP The MIME media type video/SVC string is mapped to fields in the Session Description Protocol (SDP) as follows: * The media name in the "m=" line of SDP MUST be video. * The encoding name in the "a=rtpmap" line of SDP MUST be SVC (the MIME subtype). * The clock rate in the "a=rtpmap" line MUST be 90000. * The OPTIONAL parameters "profile-level-id", "max-mbps", "max-fs", "max-cpb", "max-dpb", "max-br", "redundant-pic-cap", "sprop- parameter-sets", "parameter-add", "packetization-mode", "sprop- interleaving-depth", "deint-buf-cap", "sprop-deint-buf-req", "sprop-init-buf-time", "sprop-max-don-diff", "max-rcmd-nalu- size'', ''sprop-transport-priority'', and ''sprop-scalability- info'', when present, MUST be included in the "a=fmtp" line of SDP. These parameters are expressed as a MIME media type string, in the form of a semicolon separated list of parameter=value pairs. 9.2.2. Usage with the SDP Offer/Answer Model TBD. Wenger, Wang, Schierl Standards Track [page 25] INTERNET-DRAFT RTP Payload Format for SVC Video October 2006 9.2.3. Usage with Session and SSRC multiplexing If Session or SSRC multiplexing is used, the rules on signaling media decoding dependency in SDP as defined in [SDPsiglay] apply. Further the use of SSRC multiplexing must be signaled according to [SDPsiglay]. 9.2.4. Usage in Declarative Session Descriptions TBD. 9.3. Examples TBD. 9.4. Parameter Set Considerations Please see section 10 of RFC 3984 [RFC3984]. 10. Security Considerations Please see section 11 of RFC 3984 [RFC3984]. 11. Congestion Control Within any given RTP session carrying payload according to this specification, the provisions of section 12 of RFC 3984 [RFC3984] apply. One key motivation for the recent attention to scalable codecs has been the increasing awareness of media codec designers to network congestion. While CGS scalability cannot reduce congestion for the transport path of a given RTP session, MANEs and layered multicast technologies can be used to alleviate congestion on a larger scale. FGS scalability can be helpful to reduce session bandwidth both end- to-end (with pre-coded content) and in network segments, again assuming the use of MANEs. MANEs MAY alleviate congestion on their outgoing network path by Wenger, Wang, Schierl Standards Track [page 26] INTERNET-DRAFT RTP Payload Format for SVC Video October 2006 a) removing the NAL units belonging to hierarchically ''highest'' enhancement layer (or set of enhancement layers) from an RTP stream carrying base and enhancement layers. b) removing some or all bits of a given FGS NAL unit as long as the remaining bits still form a conforming SVC NAL unit. [Edt. Note: In the following paragraph, ''translator'' and ''mixer'' are not used consistently with RFC 3550. What we think we would need is a ''mixer'' that mixes only a single input in a single output (as a mixer terminates sessions). A ''Translator'' (that does not terminate the RTP session) carries certain unnecessary baggage which appears to make it undesirable for MANEs. The following paragraph can either be fixed into RFC 3550 style and logic (thereby removing an operation point we consider desirable), or we would need to explain in detail what we want to do (not really congestion control related and long). Perhaps we refer to the detailed discussions in the CCM draft... Added to open issues. In both cases, the incoming RTP session is terminated in the MANE, and a second RTP session originates at the MANE. The MANE acts as an RTP translator. The concept of scalability keeps the implementation and computational effort within the MANE low, and avoids expensive and delay-intensive full transcoding (in the sense of reconstruction and re-encoding).] When scalable layers are transported in their own RTP sessions, an RTP receiver SHOULD unsubscribe to one or more enhancement layers when it senses congestion, similar to what has been described in [McCanne/Vetterli]. This behavior could perhaps be sufficient to ease the network load to an acceptable level of congestion. Nevertheless, it MUST follow the mechanisms described in section 12 of [RFC3984]. 12. IANA Consideration [Edt. Note: A new MIME type should be registered from IANA.] 13. Informative Appendix: Application Examples Wenger, Wang, Schierl Standards Track [page 27] INTERNET-DRAFT RTP Payload Format for SVC Video October 2006 13.1. Introduction Scalable video coding is a concept that has been around at least since MPEG-2 [MPEG2], which goes back as early as 1993. Nevertheless, it has never gained wide acceptance; perhaps partly because applications didn't materialize in the form envisioned during standardization. MPEG and JVT, respectively, performed a requirement analysis before the SVC project was launched. Dozens of scenarios have been studied. While some of the scenarios appear not to follow the most basic design principles of the Internet -- and are therefore not appropriate for IETF standardization -- others are clearly in the scope of IETF work. Of these, this draft chooses the following subset for immediate consideration. Note that we do not reference the MPEG and JVT documents directly; partly, because at least the MPEG documents have a limited lifespan and are not publicly available, and partly because the language used in these documents is inappropriately video centric and imprecise, when it comes to protocol matters. With these remarks, we now introduce three main application scenarios that we consider as relevant, and that are implementable with this specification. 13.2. Layered Multicast This well-understood form of the use of layered coding [McCanne/Vetterli] implies that all layers are individually conveyed in their own RTP packet streams, each carried in its own RTP session using the IP (multicast) address and port number as the single demultiplexing point. Receivers ''tune'' into the layers by subscribing to the IP multicast, normally by using IGMP [IGMP]. Layered Multicast has the great advantage of simplicity and easy implementation. However, it has also the great disadvantage of utilizing many different transport addresses. While we consider this not to be a major problem for a professionally maintained content server, receiving client endpoints need to open many ports to IP multicast addresses in their firewalls. This is a practical Wenger, Wang, Schierl Standards Track [page 28] INTERNET-DRAFT RTP Payload Format for SVC Video October 2006 problem from a firewall/NAT viewpoint. Furthermore, even today IP multicast is not as widely deployed as many wish. We consider layered multicast an important application scenario for three reasons. First, it is well understood and the implementation constraints are well known. There may well by large scale IP networks outside the immediate Internet context that may wish to employ layered multicast in the future. One possible example could be a combination of content creation and core-network distribution for the various mobile TV services, e.g. those being developed by 3GPP (MBMS) [MBMS] and DVB (DVB-H) [DVB-H]. Finally, when one base and one enhancement layer is in use and are being conveyed separately, that represents one operation point of layered multicast. 13.3. Streaming of an SVC scalable stream In this scenario, a streaming server has a repository of stored SVC coded layers for a given content. At the time of streaming, and according to the capabilities and connectivity of the client(s), the streaming server generates a scalable stream. This scalable stream is served to the client(s). Both unicast and multicast serving is possible. At the same time, the streaming server may use the same repository of stored layers to compose different streams (with a different set of layers) intended for different audiences. As every endpoint receives only a single SVC RTP session, the number of firewall pinholes can be optimized. In fact, only a single firewall pinhole is required. The main difference between this scenario and straightforward simulcasting lies in the architecture and the requirements of the streaming server, and is therefore out of the scope of IETF standardization. However, compelling arguments can be made why such a streaming server design makes sense. One possible argument is related to storage space and channel bandwidth. Another is bandwidth adaptivity without transcoding -- a considerable advantage in a congestion controlled network. When the streaming server learns about congestion, it can reduce sending bitrate by choosing fewer layers when composing the layered stream. SVC is designed to gracefully support both bandwidth rampdown and bandwidth rampup with Wenger, Wang, Schierl Standards Track [page 29] INTERNET-DRAFT RTP Payload Format for SVC Video October 2006 a considerable dynamic range. This payload format is designed to allow for bandwidth flexibility in the mentioned sense, both for CGS and FGS layers. While, in theory, a transcoding step could achieve a similar dynamic range, the computational demands are impractically high and video quality is typically lowered -- therefore, few (if any) streaming servers implement full transcoding. 13.4. Multicast to MANE, SVC scalable stream to endpoint This final scenario is a bit more complex, and designed to optimize the network traffic in a core network, while still requiring only a single pinhole in the endpoint's firewall. One of its key applications is the mobile TV market. Consider a large IP network, e.g. the core network of 3GPP. Streaming servers within this core network can be assumed to be professionally maintained. We assume that these servers can have many ports open to the network and that layered multicast is a real option. Therefore, we assume that the streaming server multicasts SVC scalable layers, instead of simulcasting different representations of the same content at different bit rates. Also consider many endpoints of different classes. Some of these endpoints may not have the processing power or the display size to meaningfully decode all layers; other may have these capabilities. Users of some endpoints may not wish to pay for high quality and are happy with a base service, which may be cheaper or even free. Other users are willing to pay for high quality. Finally, some connected users may have a bandwidth problem in that they can't receive the bandwidth they would want to receive -- be it through congestion, connectivity, change of service quality, or for whatever other reasons. However, all these users have in common that they don't want to be exposed too much, and therefore the number of firewall pinholes need to be small. This situation can be handled best by introducing middleboxes close to the edge of the core network, which receive the layered multicast streams and compose the single SVC scalable bit stream according to the needs of the endpoint connected. These middleboxes are called MANEs throughout this specification. In practice, we envision the MANE to be part of (or at least physically and topologically close Wenger, Wang, Schierl Standards Track [page 30] INTERNET-DRAFT RTP Payload Format for SVC Video October 2006 to) the base station of a mobile network, where all the signaling and media traffic necessarily are multiplexed on the same physical link. This is why we do not worry too much about decomposition aspects of the MANE as such. MANEs necessarily need to be fairly complex devices. They certainly need to understand the signaling, so, for example, to associate the PT octet in the RTP header with the SVC payload type. A MANE may terminate the multicasted layered RTP sessions incoming from the core network side, and create new RTP sessions (perhaps even multicast sessions) to the endpoints connected to them. In RTP terminology, these types of MANEs are RTP mixers. This implies, per RFC 3550, a very loose relationship between the incoming and outgoing RTP sessions. In particular, there is no direct relationship between the incoming and outgoing RTP sequence numbers, RTP timestamps, payload types used, etc. Mixer-based MANEs are conceptually easy to implement and can offer powerful features, primarily because they necessarily can ''see'' the payload (including the RTP payload headers), utilize the wealth of layering information available therein, and manipulate it. While a mixer-based MANE operation in its most trivial form (combining multiple RTP packet streams into a single one) can be implemented comparatively simply -- reordering the incoming packets according to the DON and sending them in the appropriate order -- more complex forms can also be envisioned. For example, a mixer- type MANE can be optimizing the outgoing RTP stream to the MTU size of the outgoing path by utilizing the aggregation and fragmentation mechanisms of this memo. A MANE can also act as a translator. In this case, we envision its functionality to be limited to the manipulation of the transport addresses, so to enable SSRC multiplexing. The most compelling use case appears to be to forward multiple incoming RTP packets streams (conveyed to their own transport addresses) to a single firewall pinhole. The translator variant of the MANE does not terminate RTP sessions, but rather ''translate'' them in a very simple way -- by changing the transport address -- so to SSRC-multiplex multiple Wenger, Wang, Schierl Standards Track [page 31] INTERNET-DRAFT RTP Payload Format for SVC Video October 2006 sessions onto a single transport address. What sounds trivial at the first glance is in reality a highly complex process primarily due to the need of appropriate RTCP processing. This is particularly true when individual packets are intentionally being pruned or removed from the incoming session, which may be necessary to support FGS. Translator-based MANEs appear to be able to offer a limited amount of functionality without being in the security context, which opens up additional application range. Whether this form of a Translator based MANE is actually feasible, and whether it offers sufficient benefits to warrant the additional specification burden is open for discussion, and input is solicited. While the implementation complexity of either case of a MANE, as discussed above, is fairly high, the computational demands are comparatively low. In particular, SVC and/or this specification contain means to easily generate the correct inter-layer decoding order of NAL units. It is also simple to identify the fine granularity scalable bits in a given NAL unit. No serious bit- oriented processing is required and no significant state information (beyond that of the signaling and perhaps the SVC sequence parameter sets) need to be kept. 13.5. SSRC Multiplexing in case of using SRTP When SRTP is in use, it is not possible to take advantage of the in- band information (SEI messages, NAL unit headers, PACSI NAL units) when processing layered streams. Therefore, a MANE outside the security context cannot make informed decisions when aggregating information. Some relevant information must be available in the RTP header to make meaningful decisions. The first, and most obvious, choice is to map SSRC values directly to certain layers by the means of signaling. As MANEs need to be in the signaling context, this appears to be sensible. However, it requires a per-SSRC signaling mechanism -- a demultiplexing point that is currently not envisioned in SDP. Wenger, Wang, Schierl Standards Track [page 32] INTERNET-DRAFT RTP Payload Format for SVC Video October 2006 A second design choice is to somehow make available the information about the properties of a specific layer -- to the extent a MANE can make a meaningful decision -- in the SSRC value. In other words, SSRC is no more fully randomly chosen, but selected based on context. This is possible only when limiting the scope to a single sender to a multicast group, because the various senders have no means to coordinate their choice of SSRC values. In practice, that's not a major limitation. Any form of such a selection of SSRC values has two major drawbacks: First, without a sufficiently large random component the probability for SSRC collisions increases to a point that becomes unacceptable. We address this point by discouraging the use of multi-sender multicast. When only a single sender emits packets in a given RTP session, it can be expected that this sender is able to avoid SSRC collisions. In addition, we require a sufficiently large random component in the SSRC generation, which is constant for each layer stemming from the same sender. While the probability for SSRC collisions is still lowered, the random component can be kept as large as 26 bits assumes that the SVC bitstream in question contains 64 layers. Second, and more critical, a straightforward copy of values known to be present at fixed locations in the RTP payload would make it easy for codebreakers to attack an SRTP encrypted stream, because an unencrypted representation of a encrypted known value would both be present in the same packet. This is outright unacceptable from a security viewpoint. Therefore, we do not allow to simply copy information from the bitstream into the SSRC field. Instead, we rely on a non-reversible function, that also necessarily contains the aforementioned random component, that, when executed, indicates the relative priority difference between two layers (signaled by two SSRC values). The SSRC value space is evenly allocated to a number of sub value spaces, with the number of sub value spaces being equal to the number of RTP sessions for which SSRC multiplexing is used. Then the first RTP session conveying the lowest layers is mapped to the first sub SSRC value space with the lowest SSRC values, and the second RTP session conveying the second lowest layers is mapped to the second sub SSRC value space with the second lowest SSRC values, Wenger, Wang, Schierl Standards Track [page 33] INTERNET-DRAFT RTP Payload Format for SVC Video October 2006 and so on. For the RTP packets of a certain RTP session, the SSRC value is randomly selected from the corresponding sub SSRC value space. This way, a packet with a higher SSRC value contains data belonging to higher layers or layers of lower transport priority. A translator-based MANE can make use of the aforementioned SSRC values as follows. Suppose that the MANE has identified, through sensed congestion or other unspecified means, that it needs to discard packets belonging to higher layers, say K of the N buffered packets, to maintain a packet sending rate, it identifies the K packets with the highest SSRC values, and discards them. 13.6. Scenarios currently not considered for complexity reasons -- vacat -- 13.7. Scenarios currently not considered for being unaligned with IP philosophy Remarks have been made that the current draft does not take into consideration at least one application scenario which some JVT folks consider important. In particular, their idea is to make the RTP payload format (or the media stream itself) self-contained enough that a stateless, non signaling aware device can ''thin'' an RTP session to meet the bandwidth demands of the endpoint. They call this device a ''Router'' or ''Gateway'', and sometimes a MANE. Obviously, it's not a Router or Gateway in the IETF sense. To distinguish it from a MANE as defined in RFC 3984 and in this specification, let's call it a MDfH (Magic Device from Heaven). To simplify discussions, let's assume point-to-point traffic only. The endpoint has a signaling relationship with the streaming server, but it is known that the MDfH is somewhere in the media path (e.g. because the physical network topology ensures this). It has been requested, at least implicitly through MPEG's and JVT's requirements document, that the MDfH should be capable to intercept the SVC scalable bit stream, modify it by dropping packets or parts thereof, and forwarding the resulting packet stream to the receiving endpoint. It has been requested that this payload specification contains protocol elements facilitating such an operation, and the Wenger, Wang, Schierl Standards Track [page 34] INTERNET-DRAFT RTP Payload Format for SVC Video October 2006 argument has been made that the NRI field of RFC 3984 serves exactly the same purpose. The authors of this I-D do not consider the scenario above to be aligned with the most basic design philosophies the IETF follows, and therefore have not addressed the comments made (except through this section). In particular, we see the following problems with the MDfH approach): - As the very minimum, the MDfH would need to know which RTP streams are carrying SVC. We don't see how this could be accomplished but by using a static payload type. None of the IETF defined RTP profiles envision static payload types for SVC, and even the de- facto profiles developed by some application standard organizations (3GPP for example) do not use this outdated concept. Therefore, the MDfH necessarily needs to be at least ''listening'' to the signaling. - If the RTP packet payload were encrypted, it would be impossible to interpret the payload header and/or the first bytes of the media stream. We understand that there are crypto schemes under discussion that encrypt only the last n bytes of an RTP payload, but we are more than unsure that this is fully in line with the IETF's security vision. Even if the above two problems would have been overcome through standardization outside of the IETF, we still foresee serious design flaws: - An MDfH can't simply dump RTP packets it doesn't want to forward. It either needs to act as a full RTP Translator (implying that it patches RTCP RRs and such), or it needs to patch the RTP sequence numbers to fulfill the RTP specification. Not doing either would, for the receiver, look like the gaps in the sequence numbers occurred due to unintentional erasures, which has interesting effects on congestion control (if implemented), will break pretty much every meta-payload ever developed, and so on. (Many more points could be made here). - An MDfH also can't ''prune'' FGS packets. Again, doing so would not be compatible with meta payloads, and would mess up RTCP RRs and congestion control (if the congestion control is based on Wenger, Wang, Schierl Standards Track [page 35] INTERNET-DRAFT RTP Payload Format for SVC Video October 2006 octet count and not on packet count; there are discussions related to the former at least in the context of TFRC). In summary, based on our current knowledge we are not willing to specify protocol mechanisms that support an operation point that has so little in common with classic RTP use. 14. Acknowledgements Funding for the RFC Editor function is currently provided by the Internet Society. Further, the author Thomas Schierl of Fraunhofer HHI is sponsored by the European Commission under the contract number FP6-IST-0028097, project ASTRALS. 15. References 15.1. Normative References [RFC3550] Schulzrinne, H., Casner, S., Frederick, R., and V. Jacobson, "RTP: A Transport Protocol for Real-Time Applications", STD 64, RFC 3550, July 2003. [MPEG4-10] ISO/IEC International Standard 14496-10:2003. [H.264] ITU-T Recommendation H.264, "Advanced video coding for generic audiovisual services", May 2003. [SDPsiglay] Schierl, T., ''Signaling media decoding dependency in Session Description Protocol (SDP)'', IETF internet draft draft-schierl-mmusic-layered-codec-01, October 2006. [SVC] Joint Video Team, ''Annex G of Joint Draft 7 of SVC Amendment (with proposed changes)'', available from http://ftp3.itu.ch /av-arch/jvt-site/2006_07_Klagenfurt/JVT-T202.zip , July 2006 [RFC3984] Wenger, S., Hannuksela, M, Stockhammer, T, Westerlund, M, Singer, D, ''RTP Payload Format for H.264 Video'', RFC 3984, February 2005 Wenger, Wang, Schierl Standards Track [page 36] INTERNET-DRAFT RTP Payload Format for SVC Video October 2006 [RFC2119] Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, March 1997. 15.2. Informative References [DVB-H] DVB - Digital Video Broadcasting (DVB); DVB-H Implementation Guidelines, ETSI TR 102 377, 2005 [IGMP] Cain, B., Deering S., Kovenlas, I., Fenner, B. and Thyagarajan, A., ''Internet Group Management Protocol, Version 3'', RFC 3376, October 2002. [McCanne/Vetterli] V. Jacobson, S. McCanne and M. Vetterli. Receiver- driven layered multicast. In Proc. of ACM SIGCOMM'96, pages 117--130, Stanford, CA, August 1996. [MBMS] 3GPP - Technical Specification Group Services and System Aspects; Multimedia Broadcast/Multicast Service (MBMS); Protocols and codecs (Release 6), December 2005. [MPEG2] ISO/IEC International Standard 13818-2:1993. [SRTP] Baugher, M., McGrew, D, Naslund, M, Carrara, E, Norrman, K, ''The secure real-time transport protocol (SRTP)'', RFC 3711, March 2004. 16. Author's Addresses Stephan Wenger Phone: +358-50-486-0637 Nokia Research Center Email: stewe@stewe.org P.O. Box 100 FIN-33721 Tampere Finland Ye-Kui Wang Phone: +358-50-486-7004 Nokia Research Center Email: ye-kui.wang@nokia.com P.O. Box 100 FIN-33721 Tampere Finland Thomas Schierl Phone: +49-30-31002-227 Fraunhofer HHI Email: schierl@hhi.fhg.de Einsteinufer 37 Wenger, Wang, Schierl Standards Track [page 37] INTERNET-DRAFT RTP Payload Format for SVC Video October 2006 D-10587 Berlin Germany 17. Intellectual Property Statement The IETF takes no position regarding the validity or scope of any Intellectual Property Rights or other rights that might be claimed to pertain to the implementation or use of the technology described in this document or the extent to which any license under such rights might or might not be available; nor does it represent that it has made any independent effort to identify any such rights. Information on the procedures with respect to rights in RFC documents can be found in BCP 78 and BCP 79. Copies of IPR disclosures made to the IETF Secretariat and any assurances of licenses to be made available, or the result of an attempt made to obtain a general license or permission for the use of such proprietary rights by implementers or users of this specification can be obtained from the IETF on-line IPR repository at http://www.ietf.org/ipr. The IETF invites any interested party to bring to its attention any copyrights, patents or patent applications, or other proprietary rights that may cover technology that may be required to implement this standard. Please address the information to the IETF at ietf-ipr@ietf.org. 18. Disclaimer of Validity This document and the information contained herein are provided on an "AS IS" basis and THE CONTRIBUTOR, THE ORGANIZATION HE/SHE REPRESENTS OR IS SPONSORED BY (IF ANY), THE INTERNET SOCIETY AND THE INTERNET ENGINEERING TASK FORCE DISCLAIM ALL WARRANTIES, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO ANY WARRANTY THAT THE USE OF THE INFORMATION HEREIN WILL NOT INFRINGE ANY RIGHTS OR ANY IMPLIED WARRANTIES OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. 19. Copyright Statement Wenger, Wang, Schierl Standards Track [page 38] INTERNET-DRAFT RTP Payload Format for SVC Video October 2006 Copyright (C) The Internet Society (2006). This document is subject to the rights, licenses and restrictions contained in BCP 78, and except as set forth therein, the authors retain all their rights. 20. RFC Editor Considerations none 21. Open Issues 1. Need to double check MANE, Mixers, and Translators throughout the document (consistently with RFC 3550). 2. Packetization rules need work. 3. Alignment with the SVC specification (ongoing) 4. In context of SSRC multiplexing: make consistent higher/lower layers vs. RTP packet streams of higher/lower importance. 22. Changes Log From -00 to -01 - 04.02.2006, StW: Added details to scope - 04.02.2006, StW: Added short subsection 6.1 ''Design Principles'' - 04.02.2006, StW: Added section 15, ''Application Examples'' - 06.02 - 03.03.2006, YkW: Various modifications throughout the document - 13.02.2006 - 03.03.2006 , ThS: Added definitions and additional information to section 3.3, 5.1, 7 and 8, parameters in section 9.1 and added section 14 for NAL unit re-ordering for layered multicast. Further modifications throughout the document From -01 to -02 - 06.03.2006, StW: Editorial improvements - 26.05.2006, YkW: Updated NAL unit header syntax and semantics according to the latest draft SVC spec - 20.06.2006, Miska/YkW: Added section 6.10 ''Payload Content Scalability Information (PACSI) NAL Unit'' - 20.06.2006, YkW: Updated the NAL unit reordering process for layered multicast (removed the old section 14 ''Informative Appendix: NAL Unit Wenger, Wang, Schierl Standards Track [page 39] INTERNET-DRAFT RTP Payload Format for SVC Video October 2006 Re-ordering for Layered Multicast'' and added the new section 13 ''NAL Unit Reordering for Layered Multicast'') From -02 to -03 - 05.09.2006, YkW: Updated the NAL unit header syntax, definitions, etc., according to the foreseen July JVT output. Updated possible MANE adaptation operations according to SPID, TL, DID and QL. Clarified the removal of single NAL unit packetiztaion mode. Added the support of SSRC multiplexing in layered multicast. - 08.09.2006, StW: Editorial changes throughout the document - 08.09.2006, YkW: Added the packetization rule for suffix NAL unit. - 19.09.2006, YkW: Moved/updated SSRC multiplexing support to section 6.2 ''RTP header usage''. Moved/updated the cross layer DON constraint to Section 6.6 ''Decoding order number''. Moved/updated the packetization rule when a SVC bistream is transported over more than one RTP session to Section 7 ''Packetization rules''. Removed Section 13 ''Support of layered multicast''. - 16.10, TS: Added detailed four-byte NAL unit header description. Change ''AVC'' to ''H.264'' conforming to 3984. Modifications throughout the document. Extended description of 3rd byte of PACSI NAL unit. Corrected terms RTP session and RTP packet stream in case of SSRC multiplexing. Added terms in definition section on RTP multiplexing. Constraints on optional MIME parameters of 3984 for cross-layer DON (DON section and MIME parameters). Copied parts of SI paper regarding mixer, translator and SSRC mux with SRTP to section application examples. Added section on SDP usage with Session and SSRC multiplexing. Added points in Design principles on translator/mixer and RTP multiplexing. Added additional founding information in Ack- section. Corrected reference for SVC and added reference for generic signaling. 17.10, StW: Fixed many editorials, clarified MANE, mixer, translator and RTP packet stream throughout doc (hopefully consistently) 18.10., removed comments, clarified B-Bit, changed definition of base- layer (do not need to be of the lowest temporal resolution), Wenger, Wang, Schierl Standards Track [page 40]