Guidelines for Implementation: DASH-IF Interoperability Points

Living Document,

This version:
https://dashif.org/guidelines/
Issue Tracking:
GitHub
Inline In Spec
Editors:
DASH Industry Forum

1. Purpose

The scope of the DASH-IF InterOperability Points (IOPs) defined in this document is to provide support interoperable services for high-quality video distribution based on MPEG-DASH and related standards. The specified features enable relevant use cases including on-demand and live services, ad insertion, content protection and subtitling. The integration of different media codecs into DASH-based distribution is also defined.

The guidelines are provided in order to address DASH-IF members' needs and industry best practices. The guidelines provide support the implementation of conforming service offerings as well as the DASH client implementation. While alternative interpretations may be equally valid in terms of standards conformance, services and clients created following the guidelines defined in this document can be expected to exhibit highly interoperable behavior between different implementations.

Any identified bugs or missing features may be submitted through the DASH-IF issue tracker at https://gitreports.com/issue/Dash-Industry-Forum/DASH-IF-IOP.

2. Interpretation

Requirements in this document describe required service and client behaviors that DASH-IF considers interoperable:

  1. If a service provider follows these requirements in a published DASH service, that service is likely to experience successful playback on a wide variety of clients and exhibit graceful degradation when a client does not support all features used by the service.

  2. If a client implementer follows the client-oriented requirements described in this document, the client plays the content conforming to this document.

This document uses statements of fact when describing normative requirements defined in referenced specifications such as [MPEGDASH] and [MPEGCMAF]. [[RFC2119!]] statements (e.g. "SHALL", "SHOULD" and "MAY") are used when this document defines a new requirement or further constrains a requirement from a referenced document. In order to clearly separate the requirements of referenced specifications vs. the additional requirements set by this document, the normative statements in each section of this document are separated into two different groups, ones starting with "(referenced specification) requires/recommends:" and the ones starting with "This document requires/recommends:". See also Conformance.

All DASH presentations are assumed to be conforming to an IOP. A service may explicitly signal itself as conforming by including the string https://dashif.org/guidelines/ in MPD@profiles.

There is no strict backward compatibility with previous versions - best practices change over time and what was once considered sensible may be replaced by a superior approach later on. Therefore, clients and services that were conforming to version N of this document are not guaranteed to conform to version N+1.

This is a document made available by DASH-IF. The technology embodied in this document may involve the use of intellectual property rights, including patents and patent applications owned or controlled by any of the authors or developers of this document. No patent license, either implied or express, is granted to you by this document. DASH-IF has made no search or investigation for such rights and DASH-IF disclaims any duty to do so. The rights and obligations which apply to DASH-IF documents, as such rights and obligations are set forth and defined in the DASH-IF Bylaws and IPR Policy including, but not limited to, patent and other intellectual property license rights and obligations. A copy of the DASH-IF Bylaws and IPR Policy can be obtained at http://dashif.org/.

The material contained herein is provided on an "AS IS" basis and to the maximum extent permitted by applicable law, this material is provided AS IS, and the authors and developers of this material and DASH-IF hereby disclaim all other warranties and conditions, either express, implied or statutory, including, but not limited to, any (if any) implied warranties, duties or conditions of merchantability, of fitness for a particular purpose, of accuracy or completeness of responses, of workmanlike effort, and of lack of negligence.

In addition, this document may include references to documents and/or technologies controlled by third parties. Those third party documents and technologies may be subject to third party rules and licensing terms. No intellectual property license, either implied or express, to any third party material is granted to you by this document or DASH-IF. DASH-IF makes no any warranty whatsoever for such third party material.

Note that technologies included in this document and for which no test and conformance material is provided, are only published as a candidate technologies, and may be removed if no test material is provided before releasing a new version of this guidelines document. For the availability of test material, please check http://www.dashif.org.

4. DASH and related standards

DASH is a set of manifest and media formats for adaptive media delivery defined by [MPEGDASH]. Dynamic Adaptive Streaming over HTTP (DASH) is initially defined in the first edition of ISO/IEC 23009-1 which was published in April 2012 and some corrections were done in 2013. In May 2014, ISO/IEC published the second version of ISO/IEC 23009-1 that includes additional features and provide additional clarifications. ISO/IEC published the third and fourth editions of ISO/IEC 23009-1 in 2019 and 2020.

ISO/IEC also published the 1st and 2nd edition of ISO/IEC 23000-19 'Common media application format (CMAF) for segmented media' [MPEGCMAF] in 2018 and 2019. CMAF defines segment and chunk format based on ISO Base Media File Format, optimized for streaming delivery. CMAF defines a set of well defined constraints that allows interoperability for media deliverable objects, which are compatible with [MPEGDASH].

This document is based on the 4th edition DASH [MPEGDASH] and 2nd edition CMAF [MPEGCMAF] specifications.

DASH together with related standards and specifications is the foundation for an ecosystem of services and clients that work together to enable audio/video/text and related content to be presented to end-users.

This document connects DASH with international standards, industry specifications and DASH-IF guidelines.

[MPEGDASH] defines a highly flexible set of building blocks that needs to be constrained to a meaningful subset to ensure interoperable behavior in common scenarios. This document defines constraints that limit DASH features to those that are considered appropriate for use in interoperable clients and services.

This document was generated in close coordination with [DVB-DASH]. The features are aligned to the extent considered reasonable. The tools and features are aligned to the extent considered reasonable. In addition, DASH-IF worked closely with ATSC to develop a DASH profile for ATSC3.0 for broadcast distribution [ATSC3].

Clients consuming DASH content will need to interact with the host device’s media platform. While few constraints are defined on these interactions, this document does assume that the media platform implements APIs that are equivalent to the popular Media Source Extensions (MSE) and Encrypted Media Extensions (EME).

4.1. Relationship to the previous versions of this document

There is no strict backward compatibility with previous versions of this document - best practices change over time and what was once considered sensible may be replaced by a superior approach later on. Therefore, clients and services that were conforming to version N of this document are not guaranteed to conform to version N+1.

The initial two versions of this document where based on the first edition of ISO/IEC 23009-1. Version 4.3 was mostly relying on the third edition of ISO/IEC 23009-1.

This version of the document relies on the 4th edition of ISO/IEC 23009-1 that was technically frozen in July 2019 and is expected to be published by the end of 2019 as ISO/IEC 23009-1:2020.

4.2. Structure of a DASH presentation

[MPEGDASH] specifies the structure of a DASH presentation, which consists primarily of:

  1. The manifest or MPD, which describes the content and how it can be accessed.

  2. Data containers that clients will download over the course of a presentation in order to obtain media samples.

Relationships of primary DASH data structure and the standards they are defined in.

The MPD is an XML file that follows a schema defined by [MPEGDASH]. This schema defines various extension mechanisms for 3rd parties. This document defines some extensions, as do other industry specifications.

[MPEGDASH] defines two data container formats, one based on [ISOBMFF] and the other [MPEG2TS]. However, only the former is used in modern solutions. This document only supports services using the [ISOBMFF] container format.

[!MPEGCMAF] is the constrained media format based on [ISOBMFF], specifically designed for adaptive streaming. This document uses [MPEGCMAF] compatible data containers.

Note: The relationship to [MPEGCMAF] is constrained to the container format. In particular, there is no requirement to conform to [MPEGCMAF] media profiles.

The data container format defines the physical structure of the following elements described by the MPD:

  1. Each representation in the MPD references an initialization segment.

  2. Each representation in the MPD references any number of media segments.

  3. Some representations in the MPD may reference an index segment, depending on the addressing mode used.

Note: HLS (RFC8216) also support ([MPEGCMAF]). Therefore, under certain constraints, the content encoded in ([MPEGCMAF]) can be delivered using MPD or HLS m3u8 manifest format.

[MPEGDASH] [MPEGCMAF] [ISOBMFF]
(media) segment, subsegment CMAF segment
initialization segment CMAF header
index segment, segment index segment index box (sidx)
Quick reference of closely related terms in different standards.

Note: [MPEGDASH] has the concept of "segment" (URL-addressable media object) and "subsegment" (byte range of URL-addressable media object), whereas [MPEGCMAF] does not make such a distinction. This document uses [MPEGCMAF] segment terminology, with the term segment in this document being equivalent to "CMAF segment" which in turns means "DASH media segment or media subsegment", depending the employed DASH profile.

5. Interoperability requirements

The DASH-related standards enable various options for each feature supported by these standards. Limiting options and in some cases additional constraints are needed to establish interoperable behavior between service offerings and client implementations.

This chapter defines the requirements that enable DASH services and clients to provide interoperable behavior. To be compliant to a feature in this document, each service offering or client must conform to specific requirements of that feature, outline in this document.

Need to add a paragraph on interoperability on baseline, if we have any

5.1. CMAF and ISO BMFF Requirements

Media segments SHALL be compliant to [MPEGDASHCMAFPROFILE].

Note: [MPEGDASHCMAFPROFILE] defines the media segment format using [MPEGCMAF], which is largely based on [ISOBMFF].

5.2. Timing model

The purpose of this chapter is to give a holistic overview of DASH presentation timing and related segment addressing. It is not intended to provide details of the timing model and all possible uses of the attributes in [MPEGDASH].

In order to achieve higher interoperability, DASH-IF’s Implementation Guidelines allow considerably limited options than the ones provided by [MPEGDASH], constraining services to a specific set of reasonably flexible behaviors that are highly interoperable with modern client platforms. This chapter covers the timing model and related segment addressing schemes for these common use-cases.

5.2.1. Conformance requirements

This document adds additional constraints to [MPEGDASH] timing requirements.

To be conformant to this document:

5.2.2. MPD Timeline

[MPEGDASH] defines DASH general timing model in its clause 4.3.

The MPD defines the MPD timeline of a Media Presentation, which serves as the baseline for all scheduling decisions made during DASH presentation playback.

There exist two types of Media Presentations, indicated by the MPD@type.

The playback of a static MPD (defined in [MPEGDASH] as a MPD with MPD@type="static") does not depend on the mapping of the MPD timeline to real time. This means that entire presentation is available at any time and a client can play any part of the presentation at any time (e.g. it can start playback at any time and seek freely within the entire presentation).

The MPD timeline of a dynamic MPD (defined in [MPEGDASH] as a MPD with MPD@type="dynamic") has a fixed mapping to wall clock time, with each point on the MPD timeline corresponding to a point in real time. This means that segments of the presentation get available over time. Clients can introduce an additional offset with respect to wall clock time for the purpose of maintaining an input buffer to cop with network bandwidth fluctuations.

Note: In addition to mapping the MPD timeline to wall clock time, a dynamic MPD can be updated during the presentation. Updates may add new periods and remove or modify existing ones including adding new segments with progress in time, though some restrictions apply. See § 5.2.9.5 MPD updates.

The time zero on the MPD timeline of a dynamic MPD is mapped to the point in wall clock time indicated by MPD@availabilityStartTime.

The ultimate purpose of the MPD is to enable the client to obtain media samples for playback. Additionally it may possibly dynamically switch between different bitrate of the same content to adopt to the network bandwidth fluctuation. The following data structures are most relevant to locating and scheduling the samples:

  1. The MPD consists of consecutive periods which map data onto the MPD timeline.

  2. Each period contains of one or more representations, each of which provides media samples inside a sequence of media segments.

  3. Representations within a period are grouped in adaptation sets, which associate related representations and decorate them with metadata.

The primary elements described by an MPD.

5.2.3. Periods

An MPD defines an ordered list of one or more consecutive periods. A period is both a time span on the MPD timeline and a definition of the data to be presented during this time span. Period timing is relative to the zero point of the MPD timeline.

An MPD is a collection of consecutive periods.

Common reasons for defining multiple periods are:

Periods are self-contained - a client is not required to know the contents of another period in order to correctly present a period. Knowledge of the contents of different periods may be used by a client to achieve seamless period transitions, especially when working with period-connected representations.

The below static MPD consists of two 20-second periods. The duration of the first period is calculated using the start point of the second period. The total duration of the presentation is 40 seconds.
<MPD xmlns="urn:mpeg:dash:schema:mpd:2011" type="static">
  <Period>
    ...
  </Period>
  <Period start="PT20S" duration="PT20S">
    ...
  </Period>
</MPD>

Parts of the MPD structure that are not relevant for this chapter have been omitted - this is not a fully functional MPD file.

[MPEGDASH] clause 5.3.2 defines the period’s requirements in MPD authoring. Among others it requires the followings:

  1. All periods are consecutive and non-overlapping. A period may have a duration of zero.

Note: A period with a duration of zero might, for example, be the result of ad-insertion logic deciding not to insert any ad.

  1. In a static MPD, the first period starts at the time zero of the MPD timeline. In a dynamic MPD, the first period starts at or after the zero point of the MPD timeline.

  2. In a static MPD, either the last period has a Period@duration or MPD@mediaPresentationDuration exists.

  3. In a dynamic MPD, the last period may have a Period@duration, in which case it has a fixed duration. If without Period@duration, the last period in a dynamic MPD has an unknown duration, which allows to extend the timeline indefinitely.

Note: In a dynamic MPD, a period with an unknown duration may be converted to fixed-duration by an MPD update. Periods in a dynamic MPD can also be shortened or removed entirely under certain conditions. However, Media Presentation is defined until (current wall clock time + MPD@minimumUpdatePeriod), by which the current MPD is still valid. See § 5.2.9.5 MPD updates.

  1. MPD@mediaPresentationDuration may be present. If present, it accurately matches the duration between the time zero on the MPD timeline and the end of the last period. Clients must calculate the total duration of a static MPD by adding up the durations of each period and must rely on the presence of MPD@mediaPresentationDuration.

Note: This calculation is necessary because the durations of XLink periods can only be known after the XLink is resolved. Therefore it is impossible to always determine the total MPD duration on the service side as only the client is guaranteed to have access to all the required knowledge.

5.2.4. Representations

A representation is a sequence of segment as defined by [MPEGDASH] 5.3.1 and 5.3.5. A Representation element is a collection of these segment references and a description of the samples within the referenced media segments.

In practice, each representation usually belongs to exactly one adaptation set and often belongs to exactly one period, although a representation may be connected with a representation in another period.

Each segment reference addresses a media segment that corresponds to a specific time span on the sample timeline. Each media segment contains samples for a specific time span on the sample timeline.

Note: Simple addressing allows the actual time span of samples within a media segment to deviate from the corresponding time span described in the MPD ([MPEGDASH] 7.2.1). All timing-related clauses in this document refer to the timing described in the MPD (i.e. according to MPD timeline)unless otherwise noted.

The exact mechanism used to define segment references depends on the addressing mode used by the representation.

This document requires the following additional requirement:

As recommended by [MPEGDASH] 7.2.1:

This document additionally requires:

In a static MPD, the entire period must be covered with media segments.
In a dynamic MPD, the time shift buffer determines the set of required segment references in each representation. Media segments filled with gray need not be referenced due to falling outside the time shift buffer, despite falling within the bounds of a period.

Note: In a dynamic MPD, each Media segments only become available when its end point is within their availability window (This time may need to be adjusted by availabilityTimeOffset (need to be defined) and @availabilityTimeComplete values) ([MPEGDASH] 5.3.9.5.1 and 5.3.5.3). It is a valid situation that a media segment is required to be referenced but is not yet available.

As required by [MPEGDASH] 5.3.9.5.3:

As allowed by [MPEGDASH] 7.2.1:

An unnecessary segment reference is one that is not defined as required by this chapter.

This document requires the following additional requirements to [MPEGDASH]:

  1. The segment reference is for future content and will eventually become necessary.

  2. The segment reference is defined via indexed addressing.

  3. The segment reference is defined by an <S> element that defines multiple references using S@r, some of which are necessary.

  4. Removal of the segment reference is not allowed by content removal constraints.

This document also requires the following requirements for clients:

Media segments and samples need not align with period boundaries. Some samples may be entirely outside a period (marked gray) and some may overlap the period boundary (yellow).

Note: In the end, which samples are presented is entirely up to the client. It may sometimes be impractical to present media segments only partially, depending on the capabilities of the client platform, the type of media samples involved and any dependencies between samples.

5.2.5. Sample timeline

The samples within a representation exist on a linear sample timeline defined by the encoder that created the samples. One or more sample timelines are mapped onto the MPD timeline by metadata stored in or referenced by the MPD ([MPEGDASH] 7.3.2).
Sample timelines are mapped onto the MPD timeline based on parameters defined in the MPD.

Note: A sample timeline is linear - encoders are expected to use an appropriate timescale and sufficiently large timestamp fields to avoid any wrap-around. If wrap-around does occur, a new period must be started in order to establish a new sample timeline.

The sample timeline is formed after applying any [ISOBMFF] edit lists ([MPEGDASH] 7.3.2).

This document additionally requires:

Note: While optional in [MPEGDASH], the presence of the @timescale attribute is required by the interoperable timing model because the default value of 1 is unlikely to match any real-world content and is far more likely to indicate an unintentional content authoring error.

@presentationTimeOffset is the key component in establishing the relationship between the MPD timeline and a sample timeline.

The point on the sample timeline indicated by @presentationTimeOffset is equivalent to the period start point on the MPD timeline ([MPEGDASH] Table 15). The value is provided by SegmentTemplate@presentationTimeOffset or SegmentBase@presentationTimeOffset, depending on the addressing mode, and has a default value of 0 timescale units.

Note: To transform a sample timeline position SampleTime to an MPD timeline position, use the formula MpdTime = Period@start + (SampleTime - @presentationTimeOffset) / @timescale.

5.2.6. Clock drift is forbidden

Some encoders experience clock drift - they do not produce exactly 1 second worth of output per 1 second of input, either stretching or compressing the sample timeline with respect to the MPD timeline.

This document adds the following requirement:

If a packager receives input from an encoder at the wrong rate, it must take corrective action. For example, it might:

  1. Drop a span of content if input is produced faster than real-time.

  2. Insert regular padding content if input is produced slower than real-time. This padding can take different forms:

    • Silence or a blank picture.

    • Repeating frames.

    • Insertion of short-duration periods where the affected representations are not present.

Of course, such after-the-fact corrective actions can disrupt the end-user experience. The optimal solution is to fix the defective encoder.

5.2.7. Media segments

A media segment is an HTTP-addressable data structure that contains one or more media samples.

Note: Different media segments may be different byte ranges accessed on the same URL.

[MPEGCMAF] requires that Media segments contain one or more consecutive media samples, and consecutive media segments in the same representation contain consecutive media samples.

[MPEGDASH] 7.2.1 requires the followings:

[MPEGCMAF] 7.3.4 and [MPEGDASHCMAFPROFILE] requires the following:

5.2.7.1. Media segment duration deviation

When using simple addressing, the samples contained in a media segment may cover a different time span on the sample timeline than what is indicated by the nominal timing in the MPD timeline. This deviation is defined as the offset between the edges of the nominal time span (as defined by MPD timeline) and the edges of the true time span (as defined by [=sample timeline], and is calculated separately for each edge.

In simple addressing, a media segment may cover a different time span on the sample timeline than what is indicated by the nominal timing in the MPD timeline. Red boxes indicate samples.

[MPEGDASH] 7.2.1 requires: The duration deviation is no more than 50% of the nominal media segment duration and may be in either direction.

This document also recommends:

Note: [DVB-DASH] defines some relevant constraints in section 4.5. Consider obeying these constraints to be compatible with [DVB-DASH].

5.2.7.2. Segments must be aligned

Media segments are said to be aligned if the earliest presentation time of all media segments on the sample timeline is equal in all representations that belong to the same adaptation set.

[MPEGDASHCMAFPROFILE] requires:

5.2.8. Period connectivity

The precise definition of Period connectivity can found in [MPEGDASH] 5.3.2.4. However, generally speaking, in certain circumstances content may be offered such that a representation is technically compatible with the content of a representation in a previous period. Such representations are period-connected.

Any subset of the representations in a period may be period-connected with their counterparts in a future or past period. Period connectivity may be chained across any number of periods.

Note: Connectivity is generally achieved by using the same encoder to encode the content of multiple periods using the same settings. Keep in mind, however, that decryption is also a part of the client media pipeline - it is not only the codec parameters that are configured by the initialization segment; different decryption parameters are likely to break connectivity that would otherwise exist.

For signaling the period connectivity between representation of two periods in a MPD, [MPEGDASH] 5.3.2.4 requires:

Representations can be signaled as period-connected, enabling client optimizations. Arrows on diagram indicate direction of connectivity reference (from future to past), with the implied message being "the client can use the same decoder it used where the arrow points to".

Note: Not all representations in an adaptation set need to be period-connected. For example, if a new period is introduced to add a representation that contains a new video quality level, all other representations will likely be connected but not the one that was added.

Note that [MPEGDASH] allows:

The same media segment will often exist in two periods at a period-connected transition. On the diagram, this is segment 4.

This document recommends:

This document also recommends:

Note: The exact mechanism that ensures seamless playback depends on client capabilities and will be implementation-specific. Any shared media segment overlapping the period boundary may need to be detected and deduplicated to avoid presenting it twice.

5.2.8.1. Period continuity

In addition to period connectivity, [MPEGDASH] 5.3.2.4 defines period continuity, which is a special case of period connectivity where the two samples on the boundary between the connected representations are consecutive on the same sample timeline. Continuity implies connectivity.

Note: The above can only be true if the sample boundary exactly matches the period boundary.

For signaling the period continuity, [MPEGDASH] 5.3.2.4 requires:

This document requires:

This document requires:

5.2.9. Dynamic MPDs

This section only applies to dynamic MPDs.

Three main factors differentiate them from static MPDs:

  1. The segments described in a dynamic MPD may become available over time, i.e. not all segments are available.

  2. Playback of a dynamic MPD is synchronized to a real time clock (with some amount of client-chosen time shift allowed).

  3. A dynamic MPD may change over time, with clients retrieving new snapshots of the MPD when the validity duration of the previous snapshot expires.

[MPEGDASH] 5.4.1 requires:

The MPD validity duration starts when the MPD download is initiated by a client, which may be some time after it is generated/published!

This document requires: DASH clients SHALL support the presentation of dynamic MPDs.

5.2.9.1. Real time clock synchronization

It is critical to synchronize the clocks of the client with the clock of service when using a dynamic MPD. The time indicated by the clock does not necessarily need to match some universal standard as long as the two are mutually synchronized.

The use of UTCTiming is optional in [MPEGDASH].

This document requires:

The use of a "default time source" is not allowed. The mechanism of time synchronization must always be explicitly defined in the MPD by every service.

This document requires:

We could benefit from some detailed examples here, especially as clock sync is such a critical element of live services.

5.2.9.2. Availability

A media segment is available when an HTTP request to acquire the media segment can be started and successfully performed to completion by a client. During playback of a dynamic MPD, new media segments continuously become available and stop being available with the passage of time. [MPEGDASH] defines the segment availability times of a segment as the duration in wall-clock time in which that segment is available.

An availability window is a time span on the MPD timeline that determines which media segments can be expected to be available. Each representation has its own availability window. Consequently, availability window at each moment is defined by the union of segment availability times of all available segments at that moment.

A segment start point (referred to as MPD start time of a segment in [MPEGDASH]) is the presentation start time of the segment in MPD timeline.

A segment end point is defined is the presentation end time of the segment in MPD timeline.

[!MPEGDASH]] requires:

It is the responsibility of the service to ensure that media segments are available to clients when they are described as available by the MPD. Consider that the criterium for availability is a successful download by clients, not successful publishing from a packager.

The availability window is calculated as follows:

  1. Let now be the current wall clock time according to the synchronized clock.

  2. Let AvailabilityWindowStart be now - MPD@timeShiftBufferDepth.

    • If MPD@timeShiftBufferDepth is not defined, let AvailabilityWindowStart be MPD@availabilityStartTime.

  3. Let TotalAvailabilityTimeOffset be the sum of all @availabilityTimeOffset values that apply to the representation (those directly on the Representation element and any of its ancestors).

  4. The availability window is the time span from AvailabilityWindowStart to now + TotalAvailabilityTimeOffset.

The availability window determines which media segments can be expected to be available, based on where their segment end point lies.

This document requires:

5.2.9.3. Time shift buffer

The time shift buffer is a time span on the MPD timeline that defines the set of media segments that a client is allowed to present at the current moment in time according to the synchronized clock (now).

This is the mechanism by which clients can introduce a time shift (an offset) between real time and the MPD timeline when presenting dynamic MPDs. The time shift is zero when a client always chooses to play back the media segment at the end point of the time shift buffer. By playing back media segments from further in the past, a time shift is introduced.

Note: A time shift of 30 seconds means that the client starts presenting a media segment at the moment when its position on the MPD timeline reaches a distance of 30 seconds from the end of the time shift buffer.

The following additional factors further constrain the set of media segments that can be presented at the current time and can force a client to introduce a time shift:

  1. § 5.2.9.2 Availability - not every media segment in the time shift buffer is guaranteed to be available.

  2. § 5.2.9.4 Presentation delay - the service may define a delay that forbids the use of a section of the time shift buffer.

The time shift buffer extends from now - MPD@timeShiftBufferDepth to now. In the absence of MPD@timeShiftBufferDepth the start of the time shift buffer is MPD@availabilityStartTime.

Media segments overlapping the time shift buffer may potentially be presented by a client, if other constraints do not forbid it.

This document requires:

A dynamic MPD SHALL contain a period that ends at or overlaps the end point of the time shift buffer, except when reaching the end of live content in which case the last period MAY end before the end of the time shift buffer.

5.2.9.4. Presentation delay

There is a natural conflict between the availability window and the time shift buffer. It is legal for a client to present media segments as soon as they overlap the time shift buffer, yet such media segments might not yet be available.

Furthermore, the delay between media segments entering the time shift buffer and becoming available might be different for different representations that use different media segment durations. This difference may also change over time if a representation does not use a constant media segment duration.

This document requires:

[MPEGDASH] allows:

This document requires:

Note: As different clients might use different algorithms for calculating the presentation delay, providing MPD@suggestedPresentationDelay enables services to roughly synchronize the playback start position of clients.

The effective time shift buffer is the time span from the start of the time shift buffer to now - PresentationDelay.

Media segments that overlap the effective time shift buffer are the ones that may be presented at time now. Two representations with different segment lengths are shown. Diagram assumes @availabiltiyTimeOffset=0.

This document requires:

5.2.9.5. MPD updates

Dynamic MPDs may change over time. The nature of the change is not restricted unless such a restriction is explicitly defined.

Some common reasons to make changes in dynamic MPDs:

[MPEGDASH] 5.4.1 requires the following restrictions for MPD updates:

Additional restrictions on MPD updates are defined by other parts of this document.

This document requires:

This document also requires:

5.2.9.5.1. Adding content to the MPD

[!MPEGDASH]] allows two mechanisms for adding content:

Multiple content adding mechanisms may be combined in a single MPD update. An MPD update that adds content may be combined with an MPD update that removes content.

MPD updates can add both segment references and periods (additions highlighted in blue).

This document requires:

Note: The duration of the last period cannot change as a result of adding segment references. A live service will generally use a period with an unlimited duration to continuously add new segment references.

When using simple addressing or explicit addressing, it is possible for a period to define an infinite sequence of segment references that extends to the end of the period (e.g. using SegmentTemplate@duration or r="-1"). Such self-extending reference sequences are equivalent to explicitly defined segment reference sequences that extend to the end of the period and clients MAY obtain new segment references from such sequences even between MPD updates.

5.2.9.5.2. Removing content from the MPD

Removal of content is only allowed if the content to be removed is not yet available to clients and guaranteed not to become available until clients receive the MPD update. See § 5.2.9.2 Availability.

To determine the content that may be removed, let EarliestRemovalPoint be availability window end + MPD@minimumUpdatePeriod.

Note: As each representation has its own availability window, so does each representation have its own EarliestRemovalPoint.

MPD updates can remove both segment references and periods (removals highlighted in red).

An MPD update removing content MAY remove any segment references to media segments that start after EarliestRemovalPoint at the time the update is published.

Media segments that overlap or end before EarliestRemovalPoint might be considered by clients to be available at the time the MPD update is processed and therefore SHALL NOT be removed by an MPD update.

The following mechanisms exist removing content:

Multiple content removal mechanisms MAY be combined in a single MPD update.

Note: When using indexed addressing or simple addressing, removal of segment references from the end of the period only requires changing Period@duration. When using explicit addressing, pruning some S elements may be appropriate to avoid leaving unnecessary segment references.

Clients SHALL NOT fail catastrophically if an MPD update removes already buffered data but MAY incur unexpected time shift or a visible transition at the point of removal. It is the responsibility of the service to avoid removing data that may already be in use.

In addition to editorial removal from the end of the MPD, content naturally expires due to the passage of time. Expired content also needs to be removed:

An MPD update that removes content MAY be combined with an MPD update that adds content.

5.2.9.5.3. End of live content

Live services can reach a point where no more content will be produced - existing content will be played back by clients and once they reach the end, playback will cease.

This document requires:

5.2.9.6. MPD refreshes

To stay informed of the MPD updates, clients need to perform MPD refreshes at appropriate moments to download the updated MPD snapshots.

This document requires:

  1. When an MPD snapshot is downloaded, it is valid for the present moment and at least MPD@minimumUpdatePeriod after that.

  2. A client can expect to be able to successfully download any media segments that the MPD defines as available at any point during the MPD validity duration.

  3. The clients MAY refresh the MPD at any point. Typically this will occur because the client wants to obtain more segment references or make more media segments (for which it might already have references) available by extending the MPD validity duration.

    • This may result in a different MPD snapshot being downloaded, with updated information.

    • Or it may be that the MPD has not changed, in which case its validity period is extended to now + MPD@minimumUpdatePeriod.

Note: There is no requirement that clients poll for updates at MPD@minimumUpdatePeriod interval. They can do so as often or as rarely as they wish - this attribute simply defines the MPD validity duration.

Services may publish in-band events to explicitly signal MPD validity instead of expecting clients to regularly refresh on their own initiative. This enables finer control by the service but might not be supported by all clients.

This document requires:

5.2.10. Timing of stand-alone IMSC1 and WebVTT text files

Some services store text adaptation sets in stand-alone IMSC1 or WebVTT files, without segmentation or [ISOBMFF] encapsulation.

This document requires:

IMSC1 subtitles in stored in a stand-alone XML file.
<AdaptationSet mimeType="application/ttml+xml" lang="en-US">
  <Role schemeIdUri="urn:mpeg:dash:role:2011" value="subtitle" />
  <Representation>
    <BaseURL>subtitles_en_us.xml</BaseURL>
  </Representation>
</AdaptationSet>

Parts of the MPD structure that are not relevant for this chapter have been omitted - this is not a fully functional AdaptationSet element.

5.2.11. Forbidden techniques

Some aspects of [MPEGDASH] are not compatible with the interoperable timing model defined in this document. In the interest of clarity, they are explicitly listed here:

5.2.12. Examples

This section is informative.

5.2.12.1. Offer content with imperfectly aligned tracks

It may be that for various content processing workflow reasons, some tracks have a different duration from others. For example, the audio track might start a fraction of a second before the video track and end some time before the video track ends.

Content with different track lengths, before packaging as DASH.

You now have some choices to make in how you package these tracks into a DASH presentation that conforms to this document. Specifically, there exists the requirement that every representation must cover the entire period with media samples.

Content may be cut (indicated in black) to equalize track lengths.

The simplest option is to define a single period that contains representations resulting from cutting the content to match the shortest common time span, thereby covering the entire period with samples. Depending on the nature of the data that is removed, this may or may not be acceptable.

Content may be padded (indicated in green) to equalize track lengths.

If you wish to preserve track contents in their entirety, the most interoperable option is to add padding samples (e.g. silence or black frames) to all tracks to ensure that all representations have enough data to cover the entire period with samples. This may require customization of the encoding process, as the padding must match the codec configuration of the real content and might be impractical to add after the real content has already been encoded.

New periods may be started at any change in the set of available tracks.

Another option that preserves track contents is to split the content into multiple periods that each contain a different set of representations, starting a new period whenever a track starts or ends. This enables you to ensure every representations covers its period with samples. The upside of this approach is that it can be done easily, requiring only manipulation of the MPD. The downside is that some clients may be unable to seamlessly play across every period transition.

You may combine the different approaches, cutting in some places (black), padding in others (green) and defining multiple periods as needed.

You may wish to combine the different approaches, depending on the track, to achieve the optimal result.

Some clients are known to fail when transitioning from a period with audio and video to a period with only one of these components. You should avoid such transitions unless you have exact knowledge of the capabilities of your clients.

5.2.12.2. Split a period

There exist scenarios where you would wish to split a period in two. Common reasons would be:

This example shows how an existing period can be split in a way that clients capable of seamless period-connected playback do not experience interruptions in playback among representations that are present both before and after the split.

Our starting point is a presentation with a single period that contains an audio representation with short samples and a video representation with slightly longer samples, so that media segment start points do not always overlap.

Presentation with one period, before splitting. Blue is a segment, yellow is a sample. Duration in arbitrary units is listed on samples. Segment durations are taken to be the sum of sample durations. presentationTimeOffset may have any value - it is listed because will be referenced later.

Note: Periods may be split at any point in time as long as both sides of the split remain in conformance to this document (e.g. each contains at least 1 media segment). Furthermore, period splitting does not require manipulation of the segments themselves, only manipulation of the MPD.

Let’s split this period at position 220. This split occurs during segment 3 for both representations and during sample 8 and sample 5 of the audio and video representation, respectively.

The mechanism that enables period splitting in the middle of a segment is the following:

After splitting the example presentation, we arrive at the following structure.

Presentation with two periods, after splitting. Audio segment 3 and video segment 3 are shared by both periods, with the connectivity signaling indicating that seamless playback with de-duplicating behavior is expected from clients.

If indexed addressing is used, both periods will reference all segments as both periods will use the same unmodified index segment. Clients are expected to ignore media segments that fall outside the period bounds.

Simple addressing has significant limitations on alignment at period start, making it unsuitable for some multi-period scenarios. See § 5.3.4.2 Moving the period start point (simple addressing).

Other periods (e.g. ads) may be inserted between the two periods resulting from the split. This does not affect the addressing and timing of the two periods.

5.2.12.3. Change the default_KID

In encrypted content, the default_KID of a representation might need to be changed at certain points in time. Often, the changes are closely synchronized in different representations.

To perform the default_KID change, start a new period on every change, treating each representation as an independently changing element. With proper signaling, clients can perform this change seamlessly.

What about period connectivity? #238

A change in default_KID starts a new period. Orange indicates audio and yellow video representation.

The same pattern can also be applied to other changes in representation configuration.

5.3. Segment addressing modes

This section defines the addressing modes that can be used for referencing media segments, initialization segments and index segments in interopreable DASH presentations.

Addressing modes not defined in this chapter SHALL NOT be used by DASH services. Clients SHOULD support all addressing modes defined in this chapter.

All representations in the same adaptation set SHALL use the same addressing mode. Representations in different adaptation sets MAY use different addressing modes. Period-connected representations SHALL use the same addressing mode in every period.

You SHOULD choose the addressing mode based on the nature of the content:

Content generated on the fly

Use explicit addressing.

Content generated in advance of publishing

Use indexed addressing or explicit addressing.

A service MAY use simple addressing which enables the packager logic to be very simple. This simplicity comes at a cost of reduced applicability to multi-period scenarios and reduced client compatibility.

Note: Future updates to [MPEGDASH] are expected to eliminate the critical limitations of simple addressing, enabling a wider range of applicable use cases.

Update to match [MPEGDASH] 4th edition.

Indexed addressing enables all data associated with a single representation to be stored in a single CMAF track file from which byte ranges are served to clients to supply media segments, the initialization segment and the index segment. This gives it some unique advantages:

5.3.1. Indexed addressing

A representation that uses indexed addressing consists of a CMAF track file containing an index segment, an initialization segment and a sequence of media segments.

Note: This addressing mode is sometimes called "SegmentBase" in other documents.

Clauses in section only apply to representations that use indexed addressing.

Note: [MPEGDASH] makes a distinction between "segment" (HTTP-addressable entity) and "subsegment" (byte range of an HTTP-addressable entity). This document does not make such a distinction and has no concept of subsegments. Usage of "segment" here matches the definition of CMAF segment [MPEGCMAF].

Indexed addressing is based on an index segment that references all media segments.

The MPD defines the byte range in the CMAF track file that contains the index segment. The index segment informs the client of all the media segments that exist, the time spans they cover on the sample timeline and their byte ranges.

Multiple representations SHALL NOT be stored in the same CMAF track file (i.e. no multiplexed representations are to be used).

At least one Representation/BaseURL element SHALL be present in the MPD, containing a URL pointing to the CMAF track file.

The SegmentBase@indexRange attribute SHALL be present in the MPD. The value of this attribute identifies the byte range of the index segment in the CMAF track file. The value is a byte-range-spec as defined in [RFC7233], referencing a single range of bytes.

The SegmentBase@timescale attribute SHALL be present and its value SHALL match the value of the timescale field in the index segment (in the [ISOBMFF] sidx box) and the value of the timescale field in the initialization segment (in the [[!ISOBMFF tkhd box)]]).

The SegmentBase/Initialization@range attribute SHALL identify the byte range of the initialization segment in the CMAF track file. The value is a byte-range-spec as defined in [RFC7233], referencing a single range of bytes. The Initialization@sourceURL attribute SHALL NOT be used.

Below is an example of common usage of indexed addressing.

The example defines a timescale of 48000 units per second, with the period starting at position 8100 (or 0.16875 seconds) on the sample timeline. The client can use the index segment referenced by indexRange to determine where the media segment containing position 8100 (and all other media segments) can be found. The byte range of the initialization segment is also provided.

<MPD xmlns="urn:mpeg:dash:schema:mpd:2011">
  <Period>
    <AdaptationSet>
      <Representation>
        <BaseURL>showreel_audio_dashinit.mp4</BaseURL>
        <SegmentBase timescale="48000" presentationTimeOffset="8100" indexRange="848-999">
          <Initialization range="0-847"/>
        </SegmentBase>
      </Representation>
    </AdaptationSet>
  </Period>
</MPD>

Parts of the MPD structure that are not relevant for this chapter have been omitted - this is not a fully functional MPD file.

5.3.2. Structure of the index segment

The index segment SHALL consist of a single Segment Index Box (sidx) as defined by [ISOBMFF]. The field layout is as follows:

aligned(8) class SegmentIndexBox extends FullBox('sidx', version, 0) {
  unsigned int(32) reference_ID;
  unsigned int(32) timescale;

  if (version==0) {
    unsigned int(32) earliest_presentation_time;
    unsigned int(32) first_offset;
  }
  else {
    unsigned int(64) earliest_presentation_time;
    unsigned int(64) first_offset;
  }

  unsigned int(16) reserved = 0;
  unsigned int(16) reference_count;

  for (i = 1; i <= reference_count; i++)
  {
    bit (1) reference_type;
    unsigned int(31) referenced_size;
    unsigned int(32) subsegment_duration;
    bit(1) starts_with_SAP;
    unsigned int(3) SAP_type;
    unsigned int(28) SAP_delta_time;
  }
}

The values of the fields are determined as follows:

reference_ID

The track_ID of the [ISOBMFF] track that contains the data of this representation.

timescale

Same as the timescale field of the Media Header Box and same as the SegmentBase@timescale attribute in the MPD.

earliest_presentation_time

The start timestamp of the first media segment on the sample timeline, in timescale units.

first_offset

Distance from the end of the index segment to the first media segment, in bytes. For example, 0 indicates that the first media segment immediately follows the index segment.

reference_count

Total number of media segments referenced by the index segment.

reference_type

0

referenced_size

Size of the media segment in bytes. Media segments are assumed to be consecutive, so this is also the distance to the start of the next media segment.

subsegment_duration

Duration of the media segment in timescale units.

starts_with_SAP

1

SAP_type

Either 1 or 2, depending on the sample structure in the media segment.

SAP_delta_time

0

We need to clarify how to determine the right value for SAP_type. #235

5.3.2.1. Moving the period start point (indexed addressing)

When splitting periods in two or performing other types of editorial timing adjustments, a service might want to start a period at a point after the "natural" start point of the representations within.

For representations that use indexed addressing, perform the following adjustments to set a new period start point:

  1. Update SegmentBase@presentationTimeOffset to indicate the desired start point on the sample timeline.

  2. Update Period@duration to match the new duration.

5.3.3. Explicit addressing

A representation that uses explicit addressing consists of a set of media segments accessed via URLs constructed using a template defined in the MPD, with the exact time span covered by each media segment described in the MPD.

Note: This addressing mode is sometimes called "SegmentTemplate with SegmentTimeline" in other documents.

Clauses in section only apply to representations that use explicit addressing.

Explicit addressing uses a segment template that is combined with explicitly defined time spans for each media segment in order to reference media segments, either by start time or by sequence number.

The MPD SHALL contain a SegmentTemplate/SegmentTimeline element, containing a set of segment references, which satisfies the requirements defined in this document. The segment references exist as a sequence of S elements, each of which references one or more media segments with start time S@t and duration S@d timescale units on the sample timeline. The SegmentTemplate@duration attribute SHALL NOT be present.

To enable concise segment reference definitions, an S element may represent a repeating segment reference that indicates a number of repeated consecutive media segments with the same duration. The value of S@r SHALL indicate the number of additional consecutive media segments that exist.

Note: Only additional segment references are counted, so S@r=5 indicates a total of 6 consecutive media segments with the same duration.

The start time of a media segment is calculated from the start time and duration of the previous media segment if not specified by S@t. There SHALL NOT be any gaps or overlap between media segments.

The value of S@r is nonnegative, except for the last S element which MAY have a negative value in S@r, indicating that the repeated segment references continue indefinitely up to a media segment that either ends at or overlaps the period end point.

Updates to a dynamic MPD MAY add more S elements, remove expired S elements, increment SegmentTemplate@startNumber, add the S@t attribute to the first S element or increase the value of S@r on the last S element but SHALL NOT otherwise modify existing S elements.

The SegmentTemplate@media attribute SHALL contain the URL template for referencing media segments, using either the $Time$ or $Number$ template variable to unique identify media segments. The SegmentTemplate@initialization attribute SHALL contain the URL template for referencing initialization segments.

If using $Number$ addressing, the number of the first segment reference is defined by SegmentTemplate@startNumber (default value 1). The S@n attribute SHALL NOT be used - segment numbers form a continuous sequence starting with SegmentTemplate@startNumber.

Below is an example of common usage of explicit addressing.

The example defines 225 media segments starting at position 900 on the sample timeline and lasting for a total of 900.225 seconds. The period ends at 900 seconds, so the last 0.225 seconds of content is clipped (out of bounds samples may also simply be omitted from the last media segment). The period starts at position 900 which matches the start position of the first media segment found at the relative URL video/900.m4s.

<MPD xmlns="urn:mpeg:dash:schema:mpd:2011">
  <Period duration="PT900S">
    <AdaptationSet>
      <Representation>
        <SegmentTemplate timescale="1000" presentationTimeOffset="900"
            media="video/$Time$.m4s" initialization="video/init.mp4">
          <SegmentTimeline>
            <S t="900" d="4001" r="224" />
          </SegmentTimeline>
        </SegmentTemplate>
      </Representation>
    </AdaptationSet>
  </Period>
</MPD>

Parts of the MPD structure that are not relevant for this chapter have been omitted - this is not a fully functional MPD file.

Below is an example of explicit addressing used in a scenario where different media segments have different durations (e.g. due to encoder limitations).

The example defines a sequence of 11 media segments starting at position 120 on the sample timeline and lasting for a total of 95520 units at a timescale of 1000 units per second (which results in 95.52 seconds of data). The period starts at position 810, which is within the first media segment, found at the relative URL video/120.m4s. The fifth media segment repeats once, resulting in a sixth media segment with the same duration.

<MPD xmlns="urn:mpeg:dash:schema:mpd:2011">
  <Period>
    <AdaptationSet>
      <Representation>
        <SegmentTemplate timescale="1000" presentationTimeOffset="810"
            media="video/$Time$.m4s" initialization="video/init.mp4">
          <SegmentTimeline>
            <S t="120" d="8520"/>
            <S d="8640"/>
            <S d="8600"/>
            <S d="8680"/>
            <S d="9360" r="1"/>
            <S d="8480"/>
            <S d="9080"/>
            <S d="6440"/>
            <S d="10000"/>
            <S d="8360"/>
          </SegmentTimeline>
        </SegmentTemplate>
      </Representation>
    </AdaptationSet>
  </Period>
</MPD>

Parts of the MPD structure that are not relevant for this chapter have been omitted - this is not a fully functional MPD file.

5.3.3.1. Moving the period start point (explicit addressing)

When splitting periods in two or performing other types of editorial timing adjustments, a service might want to start a period at a point after the "natural" start point of the representations within.

For representations that use explicit addressing, perform the following adjustments to set a new period start point:

  1. Update SegmentTemplate@presentationTimeOffset to indicate the desired start point on the sample timeline.

  2. Update Period@duration to match the new duration.

  3. Remove any unnecessary segment references.

  4. If using the $Number$ template variable, increment SegmentTemplate@startNumber by the number of media segments removed from the beginning of the representation.

Note: See § 5.2.4 Representations and § 5.2.9.5.2 Removing content from the MPD to understand the constraints that apply to segment reference removal.

5.3.4. Simple addressing

Once we have a specific @earliestPresentationTime proposal submitted to MPEG we need to update this section to match. See #245. This is now done in [MPEGDASH] 4th edition - need to synchronize this text.

A representation that uses simple addressing consists of a set of media segments accessed via URLs constructed using a template defined in the MPD, with the nominal time span covered by each media segment described in the MPD.

Simple addressing defines the nominal time span of each media segment in the MPD. The true time span covered by samples within the media segment can be slightly different than the nominal time span. See § 5.3.4.1 Inaccuracy in media segment timing when using simple addressing.

Note: This addressing mode is sometimes called "SegmentTemplate without SegmentTimeline" in other documents.

Clauses in section only apply to representations that use simple addressing.

Simple addressing uses a segment template that is combined with approximate first media segment timing information and an average media segment duration in order to reference media segments, either by start time or by sequence number.

The SegmentTemplate@duration attribute SHALL define the nominal duration of a media segment in timescale units.

The set of segment references SHALL consist of the first media segment starting exactly at the period start point and all other media segments following in a consecutive series of equal time spans of SegmentTemplate@duration timescale units, ending with a media segment that ends at or overlaps the period end time.

The SegmentTemplate@media attribute SHALL contain the URL template for referencing media segments, using either the $Time$ or $Number$ template variable to uniquely identify media segments. The SegmentTemplate@initialization attribute SHALL contain the URL template for referencing initialization segments.

If using $Number$ addressing, the number of the first segment reference is defined by SegmentTemplate@startNumber (default value 1).

Below is an example of common usage of simple addressing.

The example defines a sample timeline with a timescale of 1000 units per second, with the period starting at position 900. The average duration of a media segment is 4001. Media segment numbering starts at 800, so the first media segment is found at the relative URL video/800.m4s. The sequence of media segments continues to the end of the period, which is 900 seconds long, making for a total of 225 defined segment references.

<MPD xmlns="urn:mpeg:dash:schema:mpd:2011">
  <Period duration="PT900S">
    <AdaptationSet>
      <Representation>
        <SegmentTemplate timescale="1000" presentationTimeOffset="900"
            media="video/$Number$.m4s" initialization="video/init.mp4"
            duration="4001" startNumber="800" />
      </Representation>
    </AdaptationSet>
  </Period>
</MPD>

Parts of the MPD structure that are not relevant for this chapter have been omitted - this is not a fully functional MPD file.

5.3.4.1. Inaccuracy in media segment timing when using simple addressing

When using simple addressing, the samples contained in a media segment MAY cover a different time span on the sample timeline than what is indicated by the nominal timing in the MPD, as long as no constraints defined in this document are violated by this deviation.

Simple addressing relaxes the requirement on media segment contents matching the sample timeline. Red boxes indicate samples.

The allowed deviation is defined as the maximum offset between the edges of the nominal time span (as defined by the MPD) and the edges of the true time span (as defined by the contents of the media segment). The deviation is evaluated separately for each edge.

This allowed deviation does not relax any requirements that do not explicitly define an exception. For example, periods must still be covered with samples for their entire duration, which constrains the flexibility allowed for the first and last media segment in a period.

The deviation SHALL be no more than 50% of the nominal media segment duration and MAY be in either direction.

Note: This results in a maximum true duration of 200% (+50% outward extension on both edges) and a minimum true duration of 1 sample (-50% inward from both edges would result in 0 duration but empty media segments are not allowed).

Allowing inaccurate timing is intended to enable reasoning on the sample timeline using average values for media segment timing. If the addressing data says that a media segment contains 4 seconds of data on average, a client can predict with reasonable accuracy which samples are found in which media segments, while at the same time the service is not required to publish per-segment timing data in the MPD. It is expected that the content is packaged with this contraint in mind (i.e. every segment cannot be inaccurate in the same direction - a shorter segment now implies a longer segment in the future to make up for it).

Consider a media segment with a nominal start time of 8 seconds from period start and a nominal duration of 4 seconds, within a period of unlimited duration.

The following are all valid contents for such a media segment:

Near period boundaries, all the constraints of timing and addressing must still be respected! Consider a media segment with a nominal start time of 0 seconds from period start and a nominal duration of 4 seconds. If such a media segment contained samples from 1 to 5 seconds (offset of 1 second away from zero point at both ends, which is within acceptable limits) it would be non-conforming because of the requirement in § 5.2.7 Media segments that the first media segment contain a media sample that starts at or overlaps the period start point. This severely limits the usefulness of simple addressing.

5.3.4.2. Moving the period start point (simple addressing)

When splitting periods in two or performing other types of editorial timing adjustments, a service might want to start a period at a point after the "natural" start point of the representations within.

Simple addressing is challenging to use in such scenarios. You SHOULD convert simple addressing representations to use explicit addressing before adjusting the period start point or splitting a period. See § 5.3.4.3 Converting simple addressing to explicit addressing.

The rest of this chapter provides instructions for situations where you choose not to convert to explicit addressing.

To move the period start point, for representations that use simple addressing:

Note: If you are splitting a period, also keep in mind the requirements on period end point sample alignment for the period that remains before the split point.

Finding a suitable new start point that conforms to the above requirements can be very difficult. If inaccurate timing is used, it may be altogether impossible. This is a limitation of simple addressing.

Having ensured conformance to the above requirements for the new period start point, perform the following adjustments:

  1. Update SegmentTemplate@presentationTimeOffset to indicate the desired start point on the sample timeline.

  2. If using the $Number$ template variable, increment SegmentTemplate@startNumber by the number of media segments removed from the beginning of the representation.

  3. Update Period@duration to match the new duration.

5.3.4.3. Converting simple addressing to explicit addressing

It may sometimes be desirable to convert a presentation from simple addressing to explicit addressing. This chapter provides an algorithm to do this.

Simple addressing allows for inaccuracy in media segment timing. No inaccuracy is allowed by explicit addressing. The mechanism of conversion described here is only valid when there is no inaccuracy. If the nominal time spans in original the MPD differ from the true time spans of the media segments, re-package the content from scratch using explicit addressing instead of converting.

To perform the conversion, execute the following steps:

  1. Calculate the number of media segments in the representation as SegmentCount = Ceil(AsSeconds(Period@duration) / ( SegmentTemplate@duration / SegmentTemplate@timescale)).

  2. Update the MPD.

    1. Add a single SegmentTemplate/SegmentTimeline element.

    2. Add a single SegmentTimeline/S element.

    3. Set S@t to equal SegmentTemplate@presentationTimeOffset.

    4. Set S@d to equal SegmentTemplate@duration.

    5. Remove SegmentTemplate@duration.

    6. Set S@r to SegmentCount - 1.

Below is an example of a simple addressing representation before conversion.
<MPD xmlns="urn:mpeg:dash:schema:mpd:2011">
  <Period duration="PT900S">
    <AdaptationSet>
      <Representation>
        <SegmentTemplate timescale="1000" presentationTimeOffset="900"
            media="video/$Number$.m4s" initialization="video/init.mp4"
            duration="4001" startNumber="800" />
      </Representation>
    </AdaptationSet>
  </Period>
</MPD>

As part of the conversion, we calculate SegmentCount = Ceil(900 / (4001 / 1000)) = 225.

After conversion, we arrive at the following result.

<MPD xmlns="urn:mpeg:dash:schema:mpd:2011">
  <Period duration="PT900S">
    <AdaptationSet>
      <Representation>
        <SegmentTemplate timescale="1000" presentationTimeOffset="900"
            media="video/$Number$.m4s" initialization="video/init.mp4"
            startNumber="800">
          <SegmentTimeline>
            <S t="900" d="4001" r="224" />
          </SegmentTimeline>
        </SegmentTemplate>
      </Representation>
    </AdaptationSet>
  </Period>
</MPD>

Parts of the MPD structure that are not relevant for this chapter have been omitted - the above are not fully functional MPD files.

5.4. Adaptation set contents

Adaptation sets SHALL contain media segments compatible with a single decoder, although services MAY require the decoder to be re-initialized when switching to a new representation. See also § 6.4 Bitstream switching.

All representations in the same adaptation set SHALL have the same timescale, both in the MPD and in the initialization segment tkhd boxes.

[ISOBMFF] edit lists SHALL be identical for all representations in an adaptation set.

Note: [DVB-DASH] defines some relevant constraints in section 4.5. Consider obeying these constraints to be compatible with [DVB-DASH].

5.5. Adaptation set types

Each adaptation set SHALL match exactly one category from among the following:

What exactly is metadata @codecs supposed to be? https://github.com/Dash-Industry-Forum/DASH-IF-IOP/issues/290

The adaptation set type SHALL be used by a DASH client to identify the appropriate handler for rendering. Typically, a DASH client selects at most one adaptation set of each type.

In addition, a DASH client SHOULD use the value of the @codecs parameter to determine whether the underlying media playback platform can play the media contained within the adaptation set.

See § 11 Media coding technologies for detailed codec-specific constraints.

5.6. Video adaptation set constraints

All representations in the same video adaptation set SHALL be alternative encodings of the same source content, encoded such that switching between them does not produce visual glitches due to picture size or aspect ratio differences.

An illustration here would be very useful.

https://github.com/Dash-Industry-Forum/DASH-IF-IOP/issues/284

To avoid visual glitches you must ensure that the sample aspect ratio is set correctly. For reasons of coding efficiency and due to technical constraints, different representations might use a different picture aspect ratio. Each representation signals a sample aspect ratio (e.g. in an [MPEGAVC] aspect_ratio_idc) that is used to scale the picture so that every representation ends up at the same display aspect ratio. The formula is display aspect ratio = picture aspect ratio / sample aspect ratio.

In the MPD, the display aspect ratio is AdaptationSet@par and the sample aspect ratio is Respresentation@sar. The picture aspect ratio is not directly present but is derived from Representation@width and Representation@height.

The encoded picture SHALL only contain the active video area, so that clients can frame the height and width of the encoded video to the size and shape of their currently selected display area without extraneous padding in the decoded video, such as "letterbox bars" or "pillarbox bars".

Representations in the same video adaptation set SHALL NOT differ in any of the following parameters:

If different video adaptation sets differ in any of the above parameters, these parameters SHOULD be signaled in the MPD on the adaptation set level by a supplemental property descriptor or an essential property descriptor with @schemeIdUri="urn:mpeg:mpegB:cicp:<Parameter>" as defined in [iso23001-8] and <Parameter> being one of the following: ColourPrimaries, TransferCharacteristics, or MatrixCoefficients. The @value attribute SHALL be set as defined in [iso23001-8].

Why is the above a SHOULD? If it matters enough to signal, we should make it SHALL? https://github.com/Dash-Industry-Forum/DASH-IF-IOP/issues/286

In any video adaptation set, the following SHALL be present:

Note: @width and @height indicate the number of encoded pixels. @par indicates the final intended display aspect ratio and @sar is effectively the ratio of aspect ratios (ratio of @width x @height to @par).

Given a coded picture of 720x576 pixels with an intended display aspect ratio of 16:9, we would have the following values:

This chapter already includes changes from #274

In any video adaptation set, the following SHOULD NOT be present and SHALL be ignored by clients if present:

The above min/max values are trivial to determine at runtime, so can be calculated by the client when needed.

@scanType SHOULD NOT be present and if present SHALL have the value progressive. Non-progressive video is not interoperable.

5.7. Audio adaptation set constraints

AdaptationSet@lang SHALL be present on every audio adaptation set.

@audioSamplingRate SHALL be present either on the adaptation set or representation level (but not both).

The AudioChannelConfiguration element SHALL be present either on the adaptation set or representation level (but not both). The scheme and value SHALL conform to ChannelConfiguration as defined in [iso23001-8].

5.8. Text adaptation set constraints

Text adaptation sets SHOULD be annotated using descriptors defined by [MPEGDASH], specifically Role, Accessibility, EssentialProperty and SupplementalProperty descriptors.

Guidelines for annotation are provided in § 7 Content annotation and selection and section 7.1.2 of [DVB-DASH].

5.9. Accessing resources over HTTP

[MPEGDASH] defines the structure of DASH presentations. Combined with an understanding of the [addressing modes], this enables DASH clients to determine a set of HTTP requests that must be made to acquire the resources needed for playback of a DASH presentation. This section defines rules for performing the HTTP requests and signaling the relevant parameters in an interoperable manner.

https://github.com/Dash-Industry-Forum/DASH-IF-IOP/issues/333

5.9.1. MPD URL resolution

A service MAY use the MPD/Location element to redirect clients to a different URL to perform MPD refreshes. HTTP redirection MAY be used when responding to client requests.

A DASH client performing an MPD refresh SHALL determine the MPD URL according to the following algorithm:

  1. If at least one MPD/Location element is present, the value of any MPD/Location element is used as the MPD URL. Otherwise the original MPD URL is used as the MPD URL.

  2. If the HTTP request results in an HTTP redirect using a 3xx response code, the redirected URL replaces the MPD URL.

The MPD URL as defined by the above algorithm SHALL be used as an implicit base URL for media segment requests.

Any present BaseURL element SHALL NOT affect MPD location resolution.

5.9.2. Segment URL resolution

A service MAY publish media segments on URLs unrelated to the MPD URL. A service MAY use multiple BaseURL elements on any level of the MPD to offer content on multiple URLs (e.g. via multiple CDNs). HTTP redirection MAY be used when responding to client requests.

For media segment requests, the DASH client SHALL determine the URL according to the following algorithm:

  1. If an absolute media segment URL is present in the MPD, it is used as-is (after template variable substitution, if appropriate).

  2. If an absolute BaseURL element is present in the MPD, it is used as the base URL.

  3. Otherwise the MPD URL is used as the base URL, taking into account any MPD URL updates that occurred due to MPD refreshes.

  4. The base URL is combined with the relative media segment URL.

Note: The client may use any logic to determine which BaseURL to use if multiple are provided.

The same logic SHALL be used for initialization segments and index segments.

What do relative BaseURLs do? Do they just incrementally build up the URL? Or are they ignored? This algorithm leaves it unclear, only referencing absolute BaseURLs. We should make it explicit.

5.9.3. Conditional MPD downloads

It can often be the case that a live service signals a short MPD validity period to allow for the possibility of terminating the last period with minimal end-to-end latency. At the same time, generating future segment references might not require any additional information to be obtained by c7lients. That is, a situation might occur where constant MPD refreshes are required but the MPD content rarely changes.

Clients using HTTP to perform MPD refreshes SHOULD use conditional GET requests as specified in [RFC7232] to avoid unnecessary data transfers when the contents of the MPD do not change between refreshes.

5.9.4. Expanding URL template variables

This section clarifies expansion rules for URL template variables such as $Time$ and $Number, defined by [MPEGDASH].

The set of string formatting suffixes used SHALL be restricted to %0[width]d.

Note: The string format suffixes are not intended for general-purpose string formatting. Restricting it to only this single suffix enables the functionality to be implemented without a string formatting library.

5.10. Minimum buffer time signaling

The text here is technically correct but could benefit from being reworded in a simpler and more understandable way. If anyone finds themselves with the time, an extra pass over this would be helpful.

The MPD contains a pair of values for a bandwidth and buffering description, namely the Minimum Buffer Time (MBT) expressed by the value of MPD@minBufferTime and bandwidth (BW) expressed by the value of Representation@bandwidth. The following holds:

In a simple and straightforward implementation, a DASH client decides downloading the next segment based on the following status information:

The task of the client is to select a suitable Representation i.

The relevant issue is that starting from a SAP on, the DASH client can continue to playout the data. This means that at the current time it does have buffer data in the buffer. Based on this model the client can download a Representation i for which BW[i] ≤ rate*buffer/MBT without emptying the buffer.

Note that in this model, some idealizations typically do not hold in practice, such as constant bitrate channel, progressive download and playout of Segments, no blocking and congestion of other HTTP requests, etc. Therefore, a DASH client should use these values with care to compensate such practical circumstances; especially variations in download speed, latency, jitter, scheduling of requests of media components, as well as to address other practical circumstances.

One example is if the DASH client operates on media segment granularity. As in this case, not only parts of the media segment (i.e., MBT worth of data) needs to be downloaded, but the entire Segment, and if the MBT is smaller than the media segment duration, then rather the media segment duration needs to be used instead of the MBT for the required buffer size and the download scheduling, i.e. download a Representation i for which BW[i] ≤ rate*buffer/max_segment_duration.

5.11. Large timescales and time values

[ECMASCRIPT] is unable to accurately represent numeric values greater than 253 using built-in types. Therefore, interoperable services cannot use such values.

All timescales are start times used in a DASH presentations SHALL be sufficiently small that no timecode value exceeding 253 will be encountered, even during the publishing of long-lasting live services.

Note: This may require the use of 64-bit fields, although the values must still be limited to under 253.

5.12. MPD size

No constraints are defined on MPD size, or on the number of elements. However, services SHOULD NOT create unnecessarily large MPDs.

Note: [DVB-DASH] defines some relevant constraints in section 4.5. Consider obeying these constraints to be compatible with [[DVB DASH]].

5.13. Representing durations in XML

All units expressed in MPD fields of datatype xs:duration SHALL be treated as fixed size:

MPD fields having datatype xs:duration SHALL NOT use the year and month units and SHOULD be expressed as a count of seconds, without using any of the larger units.

6. Commonly used features

This chapter describes some features of DASH presentations in their common implementations.

Not every DASH client will support each of these features. Compatibility of different clients and services can verified by comparing the feature sets supported by clients and used by services (and may require experimentation and testing).

6.1. Seamless switching

A key feature of DASH is the ability for clients to seamlessly switch between compatible representations at predetermined points on the MPD timeline, enabling content from different representations to be interleaved according to the wishes of the client. This enables adaptive streaming - changing the active quality level in accordance with dynamically changing network conditions. Most DASH presentations define switching points at 1-10 second intervals.

Note: Decoder reinitialization during representation switches may result in visible or audible artifacts on some clients.

There are IDR-like SAPs (i.e. SAPs of type 1 or 2) at the start of each media segment. This enables seamless switching. The presence of such SAPs is be signaled in the MPD by providing a value of 1 or 2, depending on the sample structure of the media segments, for either AdaptationSet@subsegmentStartsWithSAP (if indexed addressing is used) or AdaptationSet@segmentStartsWithSAP (if any other addressing mode is used).

We need to clarify how to determine the right value for startsWithSAP. #235

Add a reference here to help readers understand what are "IDS-like SAPs (i.e. SAPs of type 1 or 2)".

See also § 6.4 Bitstream switching.

6.2. Preview thumbnails for seeking and navigation

Clients may wish to show timecode-associated preview thumbnails as part of the seeking experience. A typical use case is for enhancing a scrub bar with visual cues. Services that wish to support this SHOULD provide an adaptation set with thumbnails.

The thumbnails are published as a sequence of jpeg/png images containing grids of thumbnails. One grid of thumbnails is one media segment. To ensure efficent transfer, a thumbnail media segment SHOULD be at least 1 minute in duration.

A thumbnail adaptation set MAY offer multiple representations with different spatial resolutions.

The addressing mode SHALL be restricted to simple addressing with only the $Number$ templating variable.

Note: The constraint on allowed addressing modes exists to limit the effort required to implement this feature in clients.

Detailed requirements on the thumbnail representations are defined in § 11.5 Thumbnail images.

6.3. Trick mode

Trick modes are used by DASH clients in order to support fast forward, seek, rewind and other operations in which typically the media, especially video, is displayed in a speed other than the normal playout speed. In order to support such operations, it is recommended that the content author adds representations at lower frame rates in order to support faster playout with the same decoding and rendering capabilities.

However, representations targeted for trick modes are typically not be suitable for regular playout. If the content author wants to explicitly signal that a representation is only suitable for trick mode cases, but not for regular playout, the service SHOULD be structured as follows:

If an adaptation set in annotated with the essential property descriptor with URI http://dashif.org/guidelines/trickmode then the DASH client SHALL NOT select any of the contained representations for regular playout.

6.4. Bitstream switching

Bitstream switching if a feature that allows a switched sequence of media segments from different representations in the same adaptation set to be decoded without resetting the decoder at switch points by ensuring that the resulting stream of media segments can be successfully decoded without the decoder even being aware of a switch.

An adaptation set that supports bitstream switching is a bitstream switching adaptation set.

The AdaptationSet@bitstreamSwitching attribute SHOULD be set to true on a bitstream switching adaptation set. Services SHALL NOT require clients to support bitstream switching in order to correctly present a bitstream switching adaptation set.

The [ISOBMFF] track_id SHALL be equal for all representations in the same bitstream switching adaptation set.

The AdaptationSet@codecs attribute SHALL be present on a bitstream switching adaptation set and indicate the maximum profile and level of any representation.

The Representation@codecs attribute MAY be present on representations that belong to a bitstream switching adaptation set. If present, it SHALL indicate the maximum profile and level of any media segment in the representation.

Allowing Representation@codecs to be absent might make it more difficult to make bitstream-switching-oblivious clients. If we require Representation@codecs to always be present, client developer life could be made simpler.

Clients that support bitstream switching SHALL initialize the decoder using the initialization segment of the representation with the highest Representation@bandwidth in a bitstream switching adaptation set.

Note: A bitstream switching adaptation set fulfills the requirements of [DVB-DASH].

6.5. Switching across adaptation sets

Note: This technology is expected to be available in [MPEGDASH] Amd 4. Once published by MPEG, this section is expected to be replaced by a reference to the MPEG-DASH standard.

Representations in two or more adaptation sets may provide the same content. In addition, the content may be time-aligned and may be offered such that seamless switching across representations in different adaptation sets is possible. Typical examples are the offering of the same content with different codecs, for example H.264/AVC and H.265/HEVC and the content author wants to provide such information to the receiver in order to seamlessly switch representations across different adaptation sets. Such switching permission may be used by advanced clients.

A content author may signal such seamless switching property across adaptation sets by providing a supplemental property descriptor along with an adaptation set with @schemeIdUri set to urn:mpeg:dash:adaptation-set-switching:2016 and the @value is a comma-separated list of adaptation set IDs that may be seamlessly switched to from this adaptation set.

If the content author signals the ability of adaptation set switching and as @segmentAlignment or @subsegmentAlignment are set to true for one adaptation set, the (sub)segment alignment shall hold for all representations in all adaptation sets for which the @id value is included in the @value attribute of the supplemental property descriptor.

As an example, a content author may signal that seamless switching across an H.264/AVC adaptation set with AdaptationSet@id="264" and an HEVC adaptation set with AdaptationSet@id="265" is possible by adding a supplemental property descriptor to the H.264/AVC adaptation set with @schemeIdUri set to urn:mpeg:dash:adaptationset-switching:2016 and the @value="265" and by adding a supplemental property descriptor to the HEVC adaptation set with @schemeIdUri set to urn:mpeg:dash:adaptationset-switching:2016 and the @value="264".

In addition, if the content author signals the ability of adaptation set switching for:

What is the above talking about?

Note: This constraint may result that the switching may only be signaled with one adaptation set, but not with both as for example one adaptation set signaling may include all spatial resolutions of another one, whereas it is not the case the other way round.

Some XML elements in an MPD may be external to the MPD itself, delay-loaded by clients based on different triggers. This mechanism is called XLink and it enables client-side MPD composition from different sources. For the purposes of timing and addressing, it is important to ensure that the duration of each period can be accurately determined both before and after XLink resolution.

Note: XLink functionality in DASH is defined by [MPEGDASH] and [XLINK]. This document provides a high level summary of the behavior and defines interoperability requirements.

XLink elements are those in the MPD that carry the xlink:href attribute. When XLink resolution is triggered, the client will query the URL referenced by this attribute. What happens next depends on the result of this query:

Non-empty result containing a valid XML fragment

The entire XLink element is replaced with the query result. A single XLink element MAY be replaced with multiple elements of the same type.

Empty result or query failure

The XLink element remains as-is with the XLink attributes removed.

When XLink resolution is triggered depends on the value of the xlink:actuate attribute. A value of onLoad indicates resolution at MPD load-time, whereas a value of onRequest indicates resolution on-demand at the time the client wishes to use the element. The default value is onRequest.

Services SHALL publish MPDs that conform to the requirements in this document even before XLink resolution. This is necessary because the behavior in case of XLink resolution failure is to retain the element as-is.

The below MPD example contains an XLink period. The real duration of the XLink period will only become known once the XLink is resolved by the client and the XLink element replaced with real content.

The first period has an explicit duration defined because the XLink resolver has no knowledge of the MPD and is unlikely to know the appropriate value to define for the second period’s Period@start (unless this data is provided in the XLink URL as a parameter).

The explicitly defined duration of the second period will only be used as a fallback if the XLink resolver decides not to define a period. In this case the existing element in the MPD is preserved.

<MPD xmlns="urn:mpeg:dash:schema:mpd:2011" xmlns:xlink="http://www.w3.org/1999/xlink" type="static">
  <Period duration="PT30S">
    ...
  </Period>
  <Period duration="PT0S" xlink:href="https://example.com/256479/clips/53473/as_period">
  </Period>
</MPD>

After XLink resolving, the entire <Period> element will be replaced, except when the XLink result is empty, in which case the client preserves the existing element (which in this case is a period with zero duration, ignored by clients).

Parts of the MPD structure that are not relevant for this chapter have been omitted - this is not a fully functional MPD file.

6.7. Update signaling via in-band events

Services MAY signal the MPD validity duration by embedding in-band messages into representations instead of specifying a fixed validity duration in the MPD. This allows services to trigger MPD refreshes at exactly the desired time and to avoid needless MPD refreshes.

The rest of this chapter only applies to services and clients that use in-band MPD validity signaling.

Services SHALL define MPD@minimumUpdatePeriod=0 and add an in-band event stream to every audio representation or, if no audio representations are present, to every video representation. The in-band event stream MAY also be added to other representations. The in-band event stream SHALL be identical in every representation where it is present.

The in-band event stream SHALL be signaled on the adaptation set level by an InbandEventStream element with @scheme_id_uri="urn:mpeg:dash:event:2012" and a @value of 1 or 3, where:

Services SHALL update MPD@publishTime to an unique value after every MPD update.

Note: MPD@publishTime is merely a version label. The value is not used in timing calculations.

Using in-band signaling and MPD@minimumUpdatePeriod=0, each media segment increases the validity period of the MPD by the duration of the media segment by default. When a validity event arrives, it carries the validity end timestamp of the MPD, enabling the client to determine when a new MPD refresh is needed.

For a detailed definition of the mechanism and the event message data structures, see [MPEGDASH]. This chapter is merely a high level summary of the most important aspects relevant to interoperability.

Illustration of MPD expiration signaling using in-band events.

Services SHALL emit in-band events as [MPEGDASH] emsg boxes to signal the MPD validity duration using the following logic:

The in-band events used for signaling MPD validity duration SHALL have scheme_id_uri and value matching the InbandEventStream element. Clients SHALL NOT use in-band events for MPD validity update signaling if these fields on the events do not match the InbandEventStream element or if the InbandEventStream element is not present in the MPD.

In-band events with value=3 SHALL provide an updated MPD in the event’s mpd field as UTF-8 encoded text without a byte order mark.

Clients MAY perofrm MPD refreshes or process an event-embedded MPD immediately upon reading the event, without waiting for the moment signaled by the event timestamp. Services SHALL ensure that an updated MPD is available and valid starting from the moment a validity event is signaled.

Multiple media segments MAY signal the same validity update event (identified by matching id field on event), enabling the signal to be delivered several segments in advance of the MPD expiration.

In-band MPD validity events SHALL NOT be signaled in a static MPD but MAY be present in the media segments referenced by a static MPD, in which case they SHALL be ignored by clients.

Note: The above may happen when a live service is converted to an on-demand service for catchup/recording purposes.

6.8. Specifying initial position in presentation URL

This section could use another pass to make it easier to read.

By default, a client would want to start playback from the start of the presentation (if MPD@type="static") or from near the live edge (if MPD@type="dynamic"). However, in some situations it may be desirable to instruct clients to start playback from a specific position. In live services, where content has a fixed mapping to real time, this means an initial time-shift is applied.

The interoperable mechanism for this is to add an MPD anchor to the presentation URL. Details of this feature are defined in [MPEGDASH], with this chapter offering a summary of the feature and constraining its use to interoperable cases.

An initial position MAY be signalled to the DASH client by including an MPD anchor in the presentation URL. If an anchor is used, it SHALL be specified with one of the following sets of parameters:

The t parameter indicates offset from period start or a moment in real-time, with period referencing a Period@id (defaulting to the first period).

The value of Period@id must be URL-encoded.

The time indicated using the t parameter SHALL be a single npttime value as specified in [media-frags]. This is a narrower definition than accepted by [MPEGDASH].

To start from the beginning of the first period the following would be added to the end of the MPD URL provided to the DASH client: #t=0

To start with a fixed offset from the start of a specific period, in this case 50 minutes from the beginning of the period with ID program_part_2, use the following syntax: #period=program_part_2&t=50:00

When accessing a live service, you can instruct the client to use an initial time-shift so that content from a specific moment is played back by providing a POSIX timestamp with the t parameter. For example, starting playback from Wed, 08 Jun 2016 17:29:06 GMT would be expressed as #t=posix:1465406946. Starting playback from the live edge can be signaled as #t=posix:now.

When referencing a moment in real time using t=posix, the period parameter SHALL NOT be used.

How do leap seconds tie into this? See #161

7. Content annotation and selection

[MPEGDASH] enables a service to annotate adaptation sets to enable clients to make an informed decision on which adaptation set to select for presentation from among the alternatives offered for each adaptation set type. The selection is based on client capabilities, client preferences, user preferences and possibly also interactive choices presented to the user. Typically, the signalling and selection is independent of the codec in use.

This chapter defines requirements and recommendations for annotating adaptation sets with interoperable descriptive information.

A service may offer multiple adaptation sets of the same type to provide the same content in different encodings or different source formats (e.g. one adaptation set encoded from a standard dynamic range master and another encoded from a high dynamic range video master). Alternatively, adaptation sets may describe different content (e.g. different languages or different camera views).

Note: While the typical situation is that a client selects one adaptation set per adaptation set type, there may be cases where multiple adaptation sets of the same type are chosen for playback (e.g. § 6.5 Switching across adaptation sets).

Proper annotation of adaptation sets in MPDs is essential in order to enable interoperable client implementations.

7.1. Annotations for content selection

[MPEGDASH] provides many options for annotating adaptation sets. This document defines a restricted subset considered interoperable by DASH-IF members.

The table below lists the permitted annotations for each adaptation set type. It is expected that interoperable DASH clients recognize the descriptors, elements, and attributes as documented in this chapter.

Content selection annotations SHALL be defined by a service in sufficient detail to differentiate every adaptation set from others of the same type. A service SHALL limit content selection annotations to those defined in this chapter.

Many of these annotations are defined by [MPEGDASH]. Other organizations may define additional descriptors or elements. Some are defined by IOP.

Note: Supplemental property descriptors are intended for presentation optimization and are intentionally not listed as annotations to be used for content selection.

Attribute or element Use Usage requirements
@profiles O If not present, it is inherited from the MPD or period. This may be used for example to signal extensions for new media profiles in the MPD.
@group OD
default=unique (see [MPEGDASH])
The attribute MAY be used. If present, it SHALL be greater than 0.

The value SHALL be different for different adaptation set types and MAY be different for adaptation sets of the same type.

This attribute enables a service to define logical groupings of adaptation sets. A client SHALL select either zero or one adaptation sets from each group.

@selectionPriority OD
default=1
This attribute SHOULD be used to expresses the preferences of the service on selecting adaptation sets for which the DASH client does make a decision otherwise.

Examples include two video codecs providing the same content, but one of the two provides higher compression efficiency and is therefore preferred.

ContentProtection 0...N If this element is present, then the content is protected. If not present, no content protection is applied.

See § 12 Content protection and security

EssentialProperty 0...N Defines an annotation that is considered essential for processing the adaptation set. See also essential property descriptor.

Clients SHALL NOT select adaptation sets that are annotated with any instances of this element that are not understood by the client.

The following schemes are expected to be recognized by a client independent of the adaptation set type:

Viewpoint 0...N Indicates that adaptation set differentiates by a different viewpoint or combination of viewpoints.

If present then all adaptation sets of the same type SHALL carry this descriptor with the same @schemeIdUri and different @value.

Label 0...N Provides a textual description of the content. This element SHOULD be used if content author expects a client to support a UI for selection.

If present then all adaptation sets of the same type SHALL carry this element with different values.

This element SHALL NOT be used as the sole differentiating element, as scenarios with no user interaction must still lead to umanbiguous selection.

Content selection annotations for any adaptation set type.

The following annotations are specific to an adaptation set type.

https://github.com/Dash-Industry-Forum/DASH-IF-IOP/issues/274

Attribute or element Use Usage requirements specific to video adaptation sets
@codecs 1...N Defines the codec that is necessary to present one or more representations in an adaptation set.

This attribute can be present on either the adaptation set level (as a single value) or the representation level (in which case multiple values might be present).

See § 11 Media coding technologies for a description of interoperable codecs.

@par M The display aspect ratio at which content is intended to be displayed.
@maxWidth O This attribute should be present to express the maximum width in samples after decoder sample cropping of any representation contained in the adaptation set.

The value should be the maximum horizontal sample count of any SPS in the contained bitstream.

@maxHeight O This attribute should be present to express the maximum height in pixel of any representation contained in the adaptation set.

The value should be the maximum horizontal sample count of any SPS in the contained bitstream.

@maxFrameRate O This attribute should be present to express the maximum frame rate, i.e. the maximum value of any entry in the decoder configuration record of the signaled frame rate, if constant frame rate is provided. contained in the adaptation set.
EssentialProperty 0...N Defines an annotation that is considered essential for processing the adaptation set. See also essential property descriptor.

Clients SHALL NOT select adaptation sets that are annotated with any instances of this element that are not understood by the client.

The following schemes are expected to be recognized by a client for video adaptation sets:

  • urn:mpeg:mpegB:cicp:<Parameter> as defined in [iso23001-8] and <Parameter> being one of the following: ColourPrimaries, TransferCharacteristics, or MatrixCoefficients

Accessibility 0...N Defines the type of accessibility-relevant content present in the adaptation set.

The set of descriptors SHALL be restricted to the following:

Role 0...N Defines the role of the content in the adaptation set.

The set of descriptors SHALL be restricted to the "Role" scheme as defined by [MPEGDASH] with @schemeIdUri="urn:mpeg:dash:role:2011" MAY be used for differentiation. A client is expected to recognize the following values when this scheme is used in the Role descriptor:

  • caption

  • subtitle

  • main

  • alternate

  • supplementary

  • sign

  • emergency

Clients SHALL consider there to be an implicit Role descriptor with the "Role" scheme and the value main if no explicitly defined Role descriptor with the "Role" scheme is present.

Annotations for video adaptation sets.
Attribute or element Use Usage requirements specific to audio adaptation sets
@codecs 1...N Defines the codec that is necessary to present one or more representations in an adaptation set.

This attribute can be present on either the adaptation set level (as a single value) or the representation level (in which case multiple values might be present).

See § 11 Media coding technologies for a description of interoperable codecs.

@lang M The language of the audio stream.
@audioSamplingRate M The audio sampling rate.
AudioChannelConfiguration 1...N specifies information about the audio channel configuration. The following schemes are expected to be recognized by a client:
EssentialProperty 0...N Defines an annotation that is considered essential for processing the adaptation set. See also essential property descriptor.

Clients SHALL NOT select adaptation sets that are annotated with any instances of this element that are not understood by the client.

The following schemes are expected to be recognized by a client for audio adaptation sets:

  • urn:mpeg:dash:audio-receiver-mix:2014 as defined in [MPEGDASH]

Accessibility 0...N Defines the type of accessibility-relevant content present in the adaptation set.

The set of descriptors SHALL be restricted to the "Role" scheme as defined by [MPEGDASH], with @schemeIdUri="urn:mpeg:dash:role:2011". A client is expected to recognize the following values when this scheme is used in the Accessibility descriptor:

  • description

  • enhanced-audio-intelligibility

Role 0...N The set of descriptors SHALL be restricted to the "Role" scheme as defined by [MPEGDASH] with @schemeIdUri="urn:mpeg:dash:role:2011" MAY be used for differentiation. A client is expected to recognize the following values when this scheme is used in the Role descriptor:
  • main

  • alternate

  • supplementary

  • commentary

  • dub

  • emergency

Clients SHALL consider there to be an implicit Role descriptor with the "Role" scheme and the value main if no explicitly defined Role descriptor with the "Role" scheme is present.

Annotations for audio adaptation sets.
Attribute or element Use Usage requirements specific to text adaptation sets
@codecs 0...N Defines the codec that is necessary to present one or more representations in an adaptation set.

This attribute can be present on either the adaptation set level (as a single value) or the representation level (in which case multiple values might be present).

The attribute SHALL be present, except when IOP does not define a @codecs value for the used text codec and encapsulation mode combination, in which case it SHALL be omitted.

See § 11 Media coding technologies for a description of interoperable codecs.

@lang M The text language.
Accessibility 0...N Defines the type of accessibility-relevant content present in the adaptation set.

The set of descriptors SHALL be restricted to the "Role" scheme as defined by [MPEGDASH], with @schemeIdUri="urn:mpeg:dash:role:2011". A client is expected to recognize the following values when this scheme is used in the Accessibility descriptor:

  • sign

  • caption

Role 0...N Defines the role of the content in the adaptation set.

The set of descriptors SHALL be restricted to the "Role" scheme as defined by [MPEGDASH] with @schemeIdUri="urn:mpeg:dash:role:2011" MAY be used for differentiation. A client is expected to recognize the following values when this scheme is used in the Role descriptor:

  • main

  • alternate

  • subtitle

  • supplementary

  • commentary

  • dub

  • description

  • emergency

Clients SHALL consider there to be an implicit Role descriptor with the "Role" scheme and the value main if no explicitly defined Role descriptor with the "Role" scheme is present.

Annotations for text adaptation sets.

7.2. Content model

In order to support the content author in providing content in a consistent manner, this chapter provides a conceptual content model for DASH content in one period of an MPD. The content may be described by an asset identifier as a whole and may contain different adaptation set types.

Model for content selection.

Within each adaptation set type, the content author may want to offer alternative content that is time-aligned but where each alternative represents different content (e.g. multiple camera angles). Automatic selection of the alternative content is not expected to be done by the DASH client as the client would not have sufficient information to make such decisions. However, the selection is expected to be done by communication with an application or the user, typically using a user interface appropriate for selection.

In the absence of user indication to select from among the alternatives, the DASH client still needs to select content to be presented. A DASH service must therefore signal the preferred default content. The preferred content is referred to as main content, whereas any content that is not main content is referred to as alternative content. There may be multiple alternatives which may need to be distinguished. See § 7.2.1 Signaling alternative content.

Furthermore, it may be that content of different [[#adaptation-set-types|adaptation set types] is linked by the content author, to express that two content of different adaptation set type are preferably played together. We define associated content for this purpose. As an example, there may be a main commentator associated with the main camera view, but for a different camera view, a different associated commentary is provided. See § 7.2.2 Signaling associated content.

In addition to semantical content level differentiation, each alternative content may be provided in different variants, based on content preparation properties (downmix, subsampling, translation, suitable for trick mode, etc.), client preferences (decoding or rendering preferences, e.g. codec), client capabilities (DASH profile support, decoding capabilities, rendering capabilities) or user preferences (accessibility, language, etc.). Both main content and alternative content in all their variants are differentiated in the MPD as defined in § 7.1 Annotations for content selection.

7.2.1. Signaling alternative content

If a period contains alternative content for one adaptation set type , then the alternatives SHALL be differentiated according to § 7.1 Annotations for content selection and one of the alternatives SHALL be provided as main content.

Main content is signaled by using the Role descriptor with scheme urn:mpeg:dash:role:2011 and value set to main. Alternative content is signaled by using the Role descriptor with scheme urn:mpeg:dash:role:2011 and value set to alternative.

7.2.2. Signaling associated content

A Viewpoint descriptor with the same @schemeIdUri and @value SHALL be used by services to signal associated content.

Clients SHALL use identical Viewpoint descriptors for determining associated content even if they do not understand the @schemeIdUri.

7.3. Client processing reference model

The following client model serves two purposes:

In the model it is assumed that the client can get sufficient information on at least the following properties:

Note: If any of these functionalities are not fulfilled, then the client may still be functional, but it may not result in the full experience as provided by the content author. As an example, if the DASH client cannot determine the preferred language, it may just use the selection priority for language selection.

The DASH client uses the MPD and finds the period that it likes to join, typically the first one for On-Demand content and the one at the live edge for live content. In order to select the media to be played, the DASH client assumes that the content is offered according to the content model above.

  1. The DASH client looks for main content, i.e. any adaptation set with annotation Role@schemeIdUri="urn:mpeg:dash:role:2011" and Role@value="alternative" is excluded initially for selection. Note that in this model it is assumed that immediate startup is desired. If the DASH client wants to go over the alternatives upfront before starting the service, then the sequence is slightly different, but still follows the remaining principles.

  2. DASH Client checks each adaptation set for the supported capabilities of the platform. If any of the capabilities are not supported, then the adaptation set is excluded from the selection process.

    • Codec support

    • DRM support

    • Rendering capabilities

  3. The DASH client checks if it supports for CEA-608 rendering as defined in clause § 11.8 CEA-608/708 Digital Television (DTV) Closed Captioning. If not supported, any accessibility descriptor with @schemeIdUri="urn:scte:dash:cc:cea-608:2015" is removed. Note that the adaptation set is maintained as it may used for regular video decoding.

  4. DASH Client checks is there are any specific settings for accessibility in the user preferences

    • If captions are requested by the system, the DASH client extracts

      • all video adaptation sets that have an Accessibility descriptor assigned with either the @schemeIdUri="urn:mpeg:dash:role:2011" and @value="caption" or @schemeIdUri="urn:scte:dash:cc:cea-608:2015" (burned-in captions and SEI-based), as well as

        • all text adaptation sets that have an Accessibility descriptor assigned with either the @schemeIdUri="urn:mpeg:dash:role:2011" and @value="caption"

        • and makes those available for adaptation sets that can be selected by the DASH client for caption support.

      • If multiple text adaptation sets remain, the DASH client removes all adaptation sets from the selection that are not in the preferred language, if language settings are provided in the system. If no language settings in the system are provided, or none of the adaptation sets meets the preferred languages, none of the adaptation sets are removed from the selection. Any adaptation sets that do not contain language annotation are removed, if any of the remaining adaptation sets provides proper language settings.

      • If still multiple text adaptation sets remain, then the ones with the highest value of @selectionPriority are chosen.

      • If still multiple text adaptation sets remain, then the DASH client makes a random choice on which caption to enable.

    • else if no captions are requested

      • the Accessibility element signaling captions may be removed from the adaptation set before continuing the selection.

    • If sign language is requested

      • all video adaptation sets that have an Accessibility descriptor assigned with @schemeIdUri="urn:mpeg:dash:role:2011" and @value="sign" are made available for sign language support.

    • else if no sign language is requested

      • the adaptation set signaling sign language with the Accessibility element may be removed from the adaptation set before continuing the selection

    • If audio descriptions are requested

      • all video adaptation sets that have an Accessibility descriptor assigned with @schemeIdUri="urn:mpeg:dash:role:2011" and @value="description" are made available for audio description support.

    • else if no audio descriptions are requested

      • the adaptation set signaling audio descriptions with the Accessibility element may be removed from the adaptation set before continuing the selection.

    • If enhanced audio intelligibility is requested

      • all audio adaptation sets that have an Accessibility descriptor assigned with @schemeIdUri="urn:mpeg:dash:role:2011" and @value="enhanced-audio-intelligibility" are made available for enhanced audio intelligibility support.

    • else if no enhanced audio intelligibility is requested

      • the Accessibility element may be removed from the adaptation set before continuing the selection.

  5. If video rendering is enabled, based on the remaining video adaptation sets the client selects one as follows:

  6. If audio rendering is enabled, based on the remaining audio adaptation sets the client selects one as follows:

  7. If text rendering is enabled, based on the text adaptation sets the client selects one as follows:

    • Any adaptation set for which an essential property descriptor is present for which the scheme or value is not understood by the DASH client, is excluded from the selection

    • If multiple text adaptation sets remain, the DASH client removes all adaptation sets from the selection that are not in the preferred language, if language settings are provided in the system. If no language settings in the system are provided, or none of the adaptation sets meets the preferred languages, none of the adaptation sets are removed from the selection. Any adaptation set that does not contain language annotation are removed, if any of the remaining adaptation sets provides proper language settings.

    • If still multiple text adaptation sets remain, then the ones with the highest value of @selectionPriority are chosen.

    • If still multiple text adaptation sets remain, then the DASH client makes a choice for itself, possibly on a random basis.

  8. If the DASH client has the ability to possibly switch to alternative content, then alternative content may be selected either through the Label function or the Viewpoint functionality. This selection may be done dynamically during playout and the DASH client is expected to switch to the alternative content. Once all alternative content is selected, the procedures following from step 2 onwards apply.

  9. At period boundary a DASH client initially looks for period continuity or connectivity, i.e. does the period include an adaptation set that is a continuation of the existing one. If not present it will go back to step 1 and execute the decision logic.

8. On-demand services

An on-demand service is one that is published with a static MPD (MPD@type="static").

On-demand services have an infinite availability window and have no fixed mapping to real time - clients may present any part at any time and may use trick mode support to alter the playback rate.

Note: An on-demand service may be created by transforming what was previously a live service into an on-demand service for viewing as a catch-up presentation or a recording. See § 9.10 Converting a live service to an on-demand service.

On-demand services MAY use any addressing mode or even a combination of multiple addressing modes.

MPD elements or attributes only relevant for dynamic MPDs SHALL NOT be present in MPDs of on-demand services. Clients SHALL ignore any such elements or attributes if present.

8.1. Surviving transforming boxes and other adaptation middleboxes

A number of video transcoding proxies (aka "middleboxes") are deployed on the Internet that may silently transcode DASH presentations. Specifically, a middlebox may see a video/mp4 HTTP response, transcode that video into a different format (perhaps using a lower bitrate or a different codec), then forward the transcoded video to the DASH client. This will break byte range based operations, as byte ranges from the original video are not valid in the transcoded video.

If such a threat is encountered, the following options may prevent proxies from transcoding DASH presentations:

insert reference to encryption.

In all cases the operational impacts on caching and implementations should be considered when using any of the above technologies. The same methods may also need to be applied to prevent middleboxes manipulating the MPD.

9. Live services

A live service is one that is published with a dynamic MPD (MPD@type="dynamic").

Live services have a strict mapping between the MPD timeline and real time and are often available only for a limited time. The MPD of a live service may change over time, for example as more content is appended and expired content is removed. Clients are forced into a timed schedule for the playout, such that they follow the schedule as desired by the content author (with some amount of client-controlled time-shift allowed).

A live service has a live edge, which is the most recent moment on the MPD timeline for which the MPD guarantees that media segments are available for all representations. See § 9.7 Determining the live edge.

Live services MAY use either explicit addressing or simple addressing or a combination of the two. Indexed addressing is not meaningful in a live service context.

Note: In previous versions of IOP a distinction was made between "simple live" and "main live" services. The latter simply refers to live services that signal MPD updates using in-band events.

There are multiple different types of live services:

Scheduled playback of prepared content

The content is prepared in advance but playback is scheduled for a specific time span in real time.

MPD-controlled live service

The content is generated on the fly and the MPD receives constant updates to reflect the latest state of the service offering. The DASH client behavior is driven solely by the MPD contents, which it regularly refreshes.

MPD- and segment-controlled live service

The content is generated on the fly and clients are kept informed of MPD validity by in-band events in the media segments. MPD downloads are only initiated when the need for updates is detected. Services can signal the need for updates on short notice.

For initial access to the service and joining the service, an MPD is required. MPDs may be accessed at join time or may have been provided earlier, for example along with an Electronic Service Guide. An MPD anchor MAY be used when referencing the MPD to specify an initial time-shift that clients are expected to apply.

Note: Support for MPD anchors is an optional client feature - a service should consider clients that lack an implementation.

The initial MPD or join MPD is accessed and processed by the client and, having an accurate clock that is synchronized with the server, the client can analyze the MPD and extract suitable information in order to initiate playback of the service. This includes, but is not limited to:

The MPD may be updated on the server based on certain rules and clients consuming the service are expected to update MPDs based on certain triggers. The triggers may be provided by the MPD itself or by information included in media segments. See § 5.2.9.5 MPD updates and § 6.7 Update signaling via in-band events.

9.1. Selecting the time shift buffer size

Recommended configuration for time shift buffer size:

If playback should only occur near the live edge, without significant time shift possibility.

MPD@timeShiftBufferDepth SHOULD be short but with a lower limit of 4 times media segment duration or 6 seconds (whichever is largest). This gives the client some opportunity to time-shift for buffering purposes, to overcome difficult network conditions.

If playback is not limited to near-live-edge.

MPD@timeShiftBufferDepth MAY have an arbitrarily large value, including a value greater than the total duration of periods in the presentation.

9.2. Selecting the suggested presentation delay

Recommended configuration for presentation delay:

If the service wishes to explicitly synchronize playback of different clients.

MPD@suggestedPresentationDelay SHOULD be set to the desired presentation delay but with a lower limit of 4 seconds or 2-4 times the media segment duration (whichever is largest).

If the service does not wish to explicitly synchronize playback of different clients.

Omit MPD@suggestedPresentationDelay and let each client determine the optimal presentation delay based on its own heuristics (which may lead different clients to choosing a different presentation delay).

The limitations imposed by the following factors SHOULD be considered when selecting the value for the presentation delay:

9.3. Selecting the media segment duration

The media segment duration SHOULD be between 1 and 10 seconds. The duration influences the end-to-end latency but also the switching and random access granularity as in DASH-264/AVC each media segment starts with a stream access point which can also be used as a switching point. The service provider should set the value taking into account at least the following:

9.4. Safety margins in availability timing

There exists unavoidable jitter and occur occasional delays in most content delivery architectures. A DASH client SHOULD avoid being too aggressive in requesting media segments as soon as they become available. If a DASH client observes issues, such as 404 responses, it SHOULD back up slightly in the requests.

Services SHALL be published so that all timing promises made by the MPD hold under normal operating conditions. Services MAY indicate an availability window that includes a safety margin. However, such a safety margin will lead to increased end-to-end latency, so it is a balance to be taken into account.

If a service wishes to impose a safety margin of N seconds, it SHOULD offset MPD@availabilityStartTime into the future by N seconds when starting the presentation.

9.5. Selecting the minimum update period

The minimum update period signals that MPD@minimumUpdatePeriod worth of future media segments are guaranteed to become available over that time span after retrieving the MPD.

Setting the value of the minimum update period primarily affects two main aspects of a service:

The downside of a small minimum update period is that a large number of MPD download requests will be made by clients. This overhead can be minimized by conditional GET requests and/or in-band MPD update signaling.

If in-band MPD validity signaling is used, MPD@minimumUpdatePeriod SHALL be 0.

9.6. Robust and seamless period transitions

Multilanguage live services are likely to encounter experience transitions from one period to another. For example, due to changes in the set of available audio/text languages or due to ad insertion.

To ensure robust client operation at period transitions, ensure that all the requirements of the timing model are satisfied. In particular, periods must be fully covered by media segment references and media samples, including immediately before/after a period transition. No gaps can occur in any representation!

In many of these cases, some adaptation sets are likely to continue seamlessly across period boundaries, in which case they SHOULD be marked as period-connected or period-continuous.

9.7. Determining the live edge

If a service does not declare a suggested presentation delay or if the client chooses to ignore it, the client will likely want to know the position of the live edge in order to perform its own presentation delay calculations.

The live edge is affected by the following factors:

Accordingly, the live edge can be calculated as follows:

  1. Determine the maximum media segment length segment_length_max for each representation.

  2. Determine the availability window end position availability_end for each representation.

  3. Determine the minimum guaranteed start of the most recent available media segment available_segment_start for each representation as available_segment_start = availability_end - segment_length_max.

  4. The live edge is min(available_segment_start).

A client MAY exclude some representations from live edge calculation if it considers them optional for successful playback. For example, trick mode representations may become available in a delayed fashion and would needlessly delay the live edge. See also § 9.8 Trick mode for live services.

Note: When determining the presentation delay a client should also consider other aspects besides the live edge such as clock synchronization accuracy, expected network performance jitter and desired buffer size.

See also § 9.4 Safety margins in availability timing.

9.8. Trick mode for live services

If trick mode is to be supported for live services, the trick mode representations SHOULD be offered using the same media segment duration as in the main adaptation set or each media segment duration should aggregate an integer multiple of the media segments in the main adaptation set.

The content author needs to find a balance between the media segment duration affecting the amount of requests in fast forward or fast rewind and the availability timing of trick mode media segments. Longer media segment durations for the trick mode representation delay the availability time of such media segments by the duration of the media segment - i.e. at the live edge the trick mode may not be fully supported.

Based on this it is a content author’s decision to provide one or more of the following alternatives for trick mode for live services:

Combinations of different alternatives are possible.

If a client wants to access a trick mode adaptation set in a live service, it SHOULD attempt to minimize the amount of requests to the network by preferring media segments with longer duration (if multiple choices are provided).

If a service is converted from live to on-demand, trick mode adaptation sets SHOULD be converted to use indexed addressing.

9.9. DVB-DASH alignment

For alignment with [DVB-DASH], the following should be considered:

[DVB-DASH] also provides recommendations in order to apply weights and priorities to different networks in a multi-BaseURL offering in section 10.8.2.1.

9.10. Converting a live service to an on-demand service

The major difference between live and on-demand services is that live services have their timeline mapped to a real time clock and have an MPD that may change. This behavior is signaled by MPD@type="dynamic". To transform a live service to an on-demand service, it may often be sufficient to set MPD@type="static" and to remove any signaling in the MPD that is restircted to dynamic MPDs.

There is no need to alter media segments when transforming a live service to an on-demand service.

Consider the time span of available content. A live service has a time shift buffer that may only allow a recent time span of content to be presented as a live service. If you wish to publish a larger time span as a recording, creating a separate on-demand variant of the MPD in parallel with the on-demand variant may be sensible.

A live service MAY be converted to an on-demand service without changing the URL, by simply replacing the dynamic MPD with a static MPD. Maintaining the same URLs for media segments might be beneficial in terms of CDN efficiency.

See also § 5.2.9.5.3 End of live content.

9.11. Reliable and consistent-delay live service

This and everything below needs to be updated to conform to timing model

Needs proper Bikeshed formatting and references

A service provider wants to run a live DASH service according to the below Figure 8. As an example, a generic encoder for a 24/7 linear program or a scheduled live event provides aproduction encoded stream. Such streams typically includ inband events to signal program changes, ad insertion opportunities and other program changes. An example for such signalling are SCTE-35 [54] messages. The stream is then provided to one or more Adaptive Bitrate (ABR) encoders, which transcodes the incoming stream into multiple bitrates and also conditions the stream for segmentation and program changes. These multiple encoders may be used for increased ABR stream density and/are then distributed downstream for redundancy purposes. The resultant streams are received by theDASH generation engines that include: MPD generator, packager and segmenter.Typically the following functions are applied by the MPD packager:

Downstream, the segments may be hosted on a single origin server, or in one or multiple CDNs. The MPD may even be further customized downstream, for example to address specific receivers. Customization may include the removal of certain Adaptation Sets that are not suitable for the capabilities of downstream clients. Specific content may be spliced based on regional services, targeted ad insertion, media blackouts or other information. Events carried from the main encoder may be interpreted and removed by the MPD packager, or they may be carried through for downstream usage. Events may also added as MPD events to the MPD.

In different stages of the encoding and distribution, errors may occur (as indicated by lightning symbols in the diagram), that for itself need to be handled by the MPD Generator and packager, the DASH client, or both of them. The key issue for this section is the ability for the DASH Media Presentation Generator as shown in to generate services that can handle the incoming streams and provide offerings such that DASH clients following DASH-IF IOPs can support.

Hence this section primarily serves to provide guidelines for implementation on MPD Generators and Packagers.

Example live service deployment architecture.

The following scenarios are considered in the service setup:

Check and align references in original text.

The subchapters here outline some possibilities for solving the above challenges.

9.11.1. Consistent latency

The scenario does not ask for very low latency, but for consistent latency. Latency can primarily be controlled by the following means:

9.11.2. Unanticipated new periods

An MPD has a certain duration after download during which the service guarantees that the information within remains valid, signaled by MPD@minimumUpdatePeriod. To avoid that the clients take future segment existence for granted even if a sudden change on the service offering is necessary, the MPD service provider must set to the MPD@minimumUpdatePeriod to a low value.

In the most conservative case, [[#live-mup-zero|the MPD author sets the MPD@minimumUpdatePeriod to 0]]. Then no promise for future segments is provided. The DASH client is forced to revalidate the MPD prior to any new Segment request.

For controlling future MPD validity, basically two options exist:

  1. Client downloads a fresh MPD before every Segment request (or batch of requests), preferably using a conditional GET in order to avoid unnecessary downlink traffic and processing in the client.

  2. Client relies on MPD validity expiration events in event messages, if content provider announces those in the MPD and by this, it can revalidate.

The two methods are not mutually exclusive.

9.11.3. Media segment duration variations

Variable media segment durations need to be correctly signed in the MPD. The mechanism depends on the addressing mode:

  1. Simple addressing allows for a deviation of up to 50% segment duration in segment start points, allowing for some drift to be compensated.

  2. Explicit addressing allows the duration of each media segment to be defined explicitly.

Media segments SHALL NOT have a duration greater than MPD@maxSegmentDuration in any situation.

9.11.4. Losses and operational failures

One of the most complex aspects are occasional operational issues, such as losses, outages, failovers of input streams, encoders, packagers and distribution links. Section 4.8 provides detailed overview on available tools that should be used by network service offering and clients in order to deal with operational issues. Several types of losses may occur:

Examples of different types of data loss.

Losses may occur in the middle of a Segment, at the end of a Segment, at the start of a new Segment. At the elementary stream level, losses may be within a compressed access unit (AU), producing a syntactically corrupt bitstream, or may be the result of the ABR encoder simply not encoding a source frame in which case the duration of the prior AU is extended producing a conforming bitstreams. Losses may impact an entire Segment or may just impact a part of the Segment. Typically, service oriented losses will occur until the next Random access point, i.e. a loss is to be signaled from the start of the lost sample up to the next random access point, typically coinciding with the start of a new Segment.

IOP defines some basic constraints in the timing model:

Deviation from these constraints is not allowed, even in case of data loss. This means that theer are basically two options:

  1. A service MAY replace lost data with padding data.

  2. A service MAY start a new period when the data loss starts and ends, removing the affected representations for the duration of the loss.

Of course, it is not possible for a service to compensate for data loss in the CDN layer. Clients are expected to survive arbitrary 404 errors that occur due to CDN faults, either by retrying, switching to another CDN (base URL), switching to another representation or automatically seeking forward.

Is there something that goes into more depth about 404s? These statements need a better home.

9.11.5. Minimizing MPD updates

MPD updates, the frequency of MPD updates and the actions included in MPD updates are different ones, and their effects may have different impacts on deployments. To avoid confusion on the generally overloaded term, some more details are discussed in the following section. In non-DASH adaptive streaming solutions, MPD updates result in the following additional processing and delivery overhead:

  1. The client sends an uplink requests for the MPD. At least from a CDN perspective, this is issue is considered less critical, typically the bounds of operation are reached by throughout, not by the number of requests.

  2. The server needs to send a full MPD with every request, which for itself causes overhead from all the way of the origin server to the client. This is in particular relevant if the manifest contains a list of URLs, and some timeshift buffer is maintained.

  3. Yet another aspect is the regular parsing and processing of the manifest in the client. Whereas the processing is likely less of a burden, the consistency across two parsing instances is relevant and requires to keep state.

  4. MPD updates may also result in writing a new MPD on the server. This may be less problematic for certain cases, especially for unicast, but it results in significant overhead if DASH formats are used for broadcast.

DASH-IF IOP provides different means to avoid one or the more of the above issues. Assuming that the MPD@minimumUpdatePeriod is set to a low value for reasons documented above, then issues mentioned above can be addressed by the following means in DASH-IF IOP:

  1. Client Requests: can be avoided by signalling inband that an MPD is has expired. The most obvious tool is the use of Inband Events with MPD expiry. However, this requires inband events being added during packaging.

  2. Sending Full MPD: Instead of requesting the full MPD, the client can support this operation by issuing a conditional GET. If the MPD has not changed, no MPD needs to be sent and the downlink rate is small. However, this requires the usage of @duration or SegmentTimeline with @r=-1.

  3. MPD Parsing and Processing: This can be avoided by using either of the solutions documented above.

  4. MPD writing on server: This goes hand-in-hand with 2, i.e. the usage of @duration or SegmentTimeline with @r=-1.

Generally, DASH-IF IOP provide several tools to address different aspects of minimizing MPD updates. Based on the deployment scenario, the appropriate tools should be used. However, it is preferable that DASH clients support different tools in order to provide choices for the service offering.

9.11.6. Proposed service configuration and MPD generation logic

The core concept is the availability of a segment stream at the input to a packager. The segment stream may be made available as individual segments or as boundary markers in a continuous stream. In addition, the stream may contain information that is relevant for the packager, such as program changes. The segment stream determines for each segment the earliest presentation time, the presentation duration, as well as boundaries in program offerings.

Furthermore, it is assumed that multiple bitrates may exist that are switchable. In the following we focus on one segment stream, but assume that in the general case multiple bitrates are available and the encoding and segment streams are generated such that they can be switched.

The high-level assumptions for the service are summarized in 4.11.2. Based on these assumptions, a more detailed model is provided.

The different scenarios are summarized in Figure 10. For the third part, it shows the notion of the change lead time. Segment of Period with index j are provided. In this case, at the start of segment (j, i+1) (i.e. its earliest presentation time) an indication is provided that the media will change after segment (j, i+2), i.e. the change lead time is d[j, i+1] + d[j, i+2]. A new Period j+1 is generated that starts with a new segment numbering.

Different properties of a segment stream.

Based on the discussions in 4.11.2, proposed service configuration for such a service are proposed. The service configuration differentiates two deployment scenarios:

  1. Clients implementing the simple live client, i.e. no emsg support and no segment parsing is implemented.

  2. Clients implementing the main client, i.e. emsg is supported and segment parsing is implemented.

9.11.6.1. Service configuration for simple live

Assuming that the input stream is a segment stream with the properties documented above is received by the DASH packager.

The DASH packager may operate as follows:

9.11.6.2. Service configuration for main live

Assuming that the input stream is a segment stream with the properties documented above is received by the DASH packager.

The DASH packager may operate as follows:

The DASH client having received an MPD that signals gaps is expected to either look for alternative Representations that are not affected by the loss, or if not possible, do some appropriate error concealment. The DASH client also should go back regularly to check for MPD updates whether the Representation gets available again.

10. Ad insertion

Needs to be checked for conformance with timing model.

Needs proper Bikeshed formatting and referencing

Needs deduplication of DASH concepts that are re-defined here.

This section provides recommendations for implementing ad insertion in DASH. Specifically, it defines the reference architecture and interoperability points for a DASH-based ad insertion solution.

The baseline reference architecture addresses both server-based and app-based scenarios. The former approach is what is typically used for Apple HLS, while the latter is typically used with Microsoft SmoothStreaming and Adobe HDS.

The following definitions are used in this section:

Ad Break

A location or point in time where one or more ads may be scheduled for delivery; same as avail and placement opportunity.

Ad Decision Service

functional entity that decides which ad(s) will be shown to the user. It interfaces deployment-specific and are out of scope for this document.

Ad Management Module

logical service that, given cue data, communicates with the ad decision service and determines which advertisement content (if at all) should be presented during the ad break described in the cue data.

Cue

indication of time and parameters of the upcoming ad break. Note that cues can indicate a pending switch to an ad break, pending switch to the next ad within an ad break, and pending switch from an ad break to the main content.

CDN node

functional entity returning a segment on request from DASH client. There are no assumptions on location of the node.

Packager

functional entity that processes conditioned content and produces media segments suitable for consumption by a DASH client. This entity is also known as fragmenter, encapsulater, or segmenter. Packager does not communicate directly with the origin server – its output is written to the origin server’s storage.

Origin

functional entity that contains all media segments indicated in the MPD, and is the fallback if CDN nodes are unable to provide a cached version of the segment on client request. Splice Point: point in media content where its stream may be switched to the stream of another content, e.g. to an ad.

MPD Generator

functional entity returning an MPD on request from DASH client. It may be generating an MPD on the fly or returning a cached one.

XLink resolver

functional entity which returns one or more remote elements on request from DASH client.

DASH ad insertion relies on several DASH tools defined in [MPEGDASH], which are introduced in this section. The correspondence between these tools and ad insertion concepts are explained below.

10.1. Remote elements

Remote elements are elements that are not fully contained in the MPD document but are referenced in the MPD with an HTTP-URL using a simplified profile of XLink.

A remote element has two attributes, @xlink:href and @xlink:actuate. @xlink:href contains the URL for the complete element, while @xlink:actuate specifies the resolution model. The value onLoad requires immediate resolution at MPD parse time, while onRequest allows deferred resolution at a time when an XML parser accesses the remote element. In this text we assume deferred resolution of remote elements, unless explicitly stated otherwise. While there is no explicit timing model for earliest time when deferred resolution can occur, the specification strongly suggests it should be close to the expected playout time of the corresponding Period. A reasonable approach is to choose the resolution at the nominal download time of the Segment.

XLink resolution

Resolution (a.k.a. dereferencing) consists of two steps. Firstly, a DASH client issues an HTTP GET request to the URL contained in the @xlink:href, attribute of the in-MPD element, and the XLink resolver responds with a remote element entity in the response content. In case of error response or syntactically invalid remote element entity, the @xlink:href and @xlink:actuate attributes the client shall remove the in-MPD element.

If the value of the @xlink:href attribute is urn:mpeg:dash:resolve-to-zero:2013, HTTP GET request is not issued, and the in-MPD element shall be removed from the MPD. This special case is used when a remote element can be accessed (and resolved) only once during the time at which a given version of MPD is valid.

If a syntactically valid remote element entity was received, the DASH client will replace in-MPD element with remote period entity. Once a remote element entity is resolved into a fully specified element, it may contain an @xlink:href attribute with @xlink:actuate set to onRequest, which contains a new XLink URL allowing repeated resolution. Note that the only information passed from the DASH client to the XLink resolver is encoded within the URL. Hence there may be a need to incorporate parameters into it, such as splice time (i.e., PeriodStart for the remote period) or cue message.

Note: In ISO/IEC 23009-1:2014/Cor.3 it is clarified that if multiple top-level remote elements are included, the remote element entity is not a valid XML document.

10.2. Periods

Periods are time-delimited parts of a DASH Media Presentation. The value of PeriodStart can be explicitly stated using the Period@start attribute or indirectly computed using Period@duration of the previous Periods.

Precise period duration of period i is given by PeriodStart(i+1) – PeriodStart(i). This can accommodate the case where media duration of period i is slightly longer than the period itself, in which case a client will schedule the start of media presentation for period i+1 at time PeriodStart(i+1).

Representation@presentationTimeOffset specifies the value of the presentation time at PeriodStart(i).

10.3. Segment availability

In case of dynamic MPDs, Period-level BaseURL@availabilityTimeOffset allow earlier availability start times. A shorthand notation @availabilityTimeOffset="INF" at a Period-level BaseURL indicates that the segments within this period are available at least as long as the current MPD is valid. This is the case with stored ad content. Note that DASH also allows specification of @availabilityTimeOffset at Adaptation Set and Representation level.

10.4. Seamless transition

The DASH specification says nothing about Period transitions – i.e., there are no guarantees for seamless continuation of playout across the period boundaries. Content conditioning and receiver capability requirements should be defined for applications relying on this functionality. However, Period continuity or connectivity should be used and signaled as defined in section 3.2.12 and ISO/IEC 23009-1:2014/Amd.3 [4].

10.5. Period labeling

Period-level AssetIdentifier descriptors identify the asset to which a given Period belongs. Beyond identification, this can be used for implementation of client functionality that depends on distinguishing between ads and main content (e.g. progress bar and random access).

10.6. DASH events

DASH events are messages having type, timing and optional payload. They can appear either in MPD (as period-level event stream) or inband, as ISO-BMFF boxes of type emsg. The emsg boxes shall be placed at the very beginning of the Segment, i.e. prior to any media data, so that DASH client needs a minimal amount of parsing to detect them.

DASH defines three events that are processed directly by a DASH client: MPD Validity Expiration, MPD Patch and MPD Update. All signal to the client that the MPD needs to be updated – by providing the publish time of the MPD that should be used, by providing an XML patch that can be applied to the client’s in-memory representation of MPD, or by providing a complete new MPD. For details please see section 4.5.

User-defined events are also possible. The DASH client does not deal with them directly – they are passed to an application, or discarded if there is no application willing or registered to process these events. A possible client API would allow an application to register callbacks for specific event types. Such callback will be triggered when the DASH client parses the emsg box in a Segment, or when it parses the Event element in the MPD.

In the ad insertion context, user-defined events can be used to signal information, such as cue messages (e.g. SCTE 35 [54])

10.7. MPD updates

If MPD@minimumUpdatePeriod is present, the MPD can be periodically updated. These updates can be synchronous, in which case their frequency is limited by MPD@minimumUpdatePeriod. In case of the main live profiles MPD updates may be triggered by DASH events. Fir details refer to section 4.5.

When new period containing stored ads is inserted into a linear program, and there is a need to unexpectedly alter this period the inserted media will not carry the emsg boxes – these will need to be inserted on-the-fly by proxies. In this case use of synchronous MPD updates may prove simpler.

MPD@publishTime provides versioning functionality: MPD with later publication times include all information that was included all MPDs with earlier publication times.

10.8. Session information

In order to allow fine-grain targeting and personalization, the identity of the client/viewer, should be known i.e. maintain a notion of a session.

HTTP is a stateless protocol, however state can be preserved by the client and communicated to the server.

The simplest way of achieving this is use of cookies. According to RFC 6265 [41], cookies set via 2xx, 4xx, and 5xx responses must be processed and have explicit timing and security model.

10.9. Tracking and reporting

The simplest tracking mechanism is server-side logging of HTTP GET requests. Knowing request times and correspondence of segment names to content constitutes an indication that a certain part of the content was requested. If MPDs (or remote element entities) are generated on the fly and identity of the requester is known, it is possible to provide more precise logging. Unfortunately this is a non-trivial operation, as same user may be requesting parts of content from different CDN nodes (or even different CDNs), hence log aggregation and processing will be needed.

Another approach is communicating with existing tracking server infrastructure using existing external standards. An IAB VAST-based implementation is shown in section 5.3.3.7.

DASH Callback events are defined in ISO/IEC 23009-1:2014 AMD3 [4], are a simple native implementation of time-based impression reporting (e.g., quartiles). A callback event is a promise by the DASH client to issue an HTTP GET request to a provided URL at a given offset from PeriodStart. The body of HTTP response is ignored. Callback events can be both, MPD and inband events.

10.10. Ad insertion architectures

The possible architectures can be classified based on the location of component that communicates with the ad decision service: a server-based approach assumes a generic DASH client and all communication with ad decision services done at the server side (even if this communication is triggered by a client request for a segment, remote element, or an MPD. The app-based approach assumes an application running on the end device and controlling one or more generic DASH clients.

Yet another classification dimension is amount of media engines needed for a presentation – i.e., whether parallel decoding needs to be done to allow seamless transition between the main and the inserted content, or content is conditioned well enough to make such transition possible with a single decoder.

Workflows can be roughly classified into linear and elastic. Linear workflows (e.g., live feed from an event) has ad breaks of known durations which have to be taken: main content will only resume after the end of the break and the programmer / operator needs to fill them with some inserted content. Elastic workflows assume that the duration of an ad break at a given cue location not fixed, thus the effective break length can vary (and can be zero if a break is not taken).

10.11. Server-based architecture

Server-based architecture

In the server-based model, all ad-related information is expressed via MPD and segments, and ad decisions are triggered by client requests for MPDs and for resources described in them (Segments, remote periods).

The server-based model is inherently MPD-centric – all data needed to trigger ad decision is concentrated in the MPD. In case where ad break location (i.e., its start time) is unknown at the MPD generation time, it is necessary to rely on MPD update functionality. The two possible ways of achieving these are described in 5.1.3.5.

In the live case, packager receives feed containing inband cues, such as MPEG-2 TS with SCTE 35 cue messages [54]. The packager ingests content segments into the CDN. In the on demand case, cues can be provided out of band.

Ad management is located at the server side (i.e., in the cloud), thus all manifest and content conditioning is done at the server side.

10.11.1. Implementation basics

A single ad is expressed as a single Period element.

Periods with content that is expected to be interrupted as a result of ad insertion should contain explicit start times (Period@start), rather than durations. This allows insertion of new periods without modifying the existing periods. If a period has media duration longer then the distance between the start of this period and the start of next period, use of start times implies that a client will start the playout of the next period at the time stated in the MPD, rather than after finishing the playout of the last segment.

An upcoming ad break is expressed as Period element(s), possibly remote.

Remote Periods are resolved on demand into one or more than one Period elements. It is possible to embed parameters from the cue message into the XLink URL of the corresponding remote period, in order to have them passed to the ad decision system via XLink resolver at resolution time.

In an elastic workflow, when an ad break is not taken, the remote period will be resolved into a period with zero duration. This period element will contain no adaptation sets.

If a just-in-time remote Period dereferencing is required by use of @xlink:actuate="onRequest", MPD update containing a remote period should be triggered close enough to the intended splice time. This can be achieved using MPD Validity events and full-fledged MPD update, or using MPD Patch and MPD Update events (see sec. 5.1.3.5 and 5.1.3.4). However, due to security reasons MPD Patch and MPD Update events should only be used with great care.

In case of Period@xlink:actuate="onRequest", MPD update and XLink resolution should be done sufficiently early to ensure that there are no artefacts due to insufficient time given to download the inserted content. Care needs to be taken so that the client is given a sufficient amount of time to (a) request and receive MPD update, and (b) dereference the upcoming remote period.

Note: It may be operationally simpler to avoid use of Period@xlink:actuate="onRequest", dereferencing in case of live content.

10.11.3. Timing and dereferencing

The only interface between DASH client and the XLink resolver is the XLink URL (i.e., the Period@xlink:href attribute).After resolution, the complete remote Period element is replaced with Period element(s) from the remote entity (body of HTTP response coming from XLink resolver). This means that the XLink resolver is (in the general case) unaware of the exact start time of the ad period.

In case of linear content, start of the ad period is only known a short time before the playback. The recommended implementation is to update the MPD at the moment the start of the ad period is known to the MPD generator.

The simplest approach for maintaining time consistency across dereferencing is to have the MPD update adding a Period@duration attribute to the latest (i.e., the currently playing) main content period. This means that the MPD resolver needs to include the Period@duration attribute into each of the Period elements returned in the remote entity. The downside of this approach is that the DASH client needs to be able to update the currently playing period.

An alternative approach is to embed the desired value of Period@start of the first period of the remote entity in the XLink URL (e.g., using URL query parameters). This approach is described in clause 5.3.5. The downside of this alternative approach is that the DASH specification does not constrain XLink URLs in any way, hence the XLink resolver needs to be aware of this URL query parameter interface defined in clause 5.3.5.

10.11.4. Asset identifiers

AssetIdentifier descriptors identify the asset to which a Period belongs. This can be used for implementation of client functionality that depends on distinguishing between ads and main content (e.g. progress bar).

Periods with same AssetIdentifier should have identical Adaptation Sets, Initialization Segments and same DRM information (i.e., DRM systems, licenses). This allows reuse of at least some initialization data across periods of the same asset, and ensures seamless continuation of playback if inserted periods have zero duration. Period continuity or connectivity should be signaled, if the content obeys the rules.

Using an asset identifier

10.11.5. MPD updates

MPD updates are used to implement dynamic behavior. An updated MPD may have additional (possibly – remote) periods. Hence, MPD update should be triggered by the arrival of the first cue message for an upcoming ad break. Ad breaks can also be canceled prior to their start, and such cancellation will also trigger an MPD update.

Frequent regular MPD updates are sufficient for implementing dynamic ad insertion. Unfortunately they create an overhead of unnecessary MPD traffic – ad breaks are rare events, while MPD updates need to be frequent enough if a cue message is expected to arrive only several seconds before the splice point. Use of HTTP conditional GET requests (i.e., allowing the server to respond with "304 Not Modified" if MPD is unchanged) is helpful in reducing this overhead, but asynchronous MPD updates avoid this overhead entirely.

DASH events with scheme "urn:mpeg:dash:event:2013" are used to trigger asynchronous MPD updates.

The simple mapping of live inband cues in live content into DASH events is translating a single cue into an MPD Validity expiration event (which will cause an MPD update prior to the splice time). MPD Validity expiration events need to be sent early enough to allow the client request a new MPD, resolve XLink (which may entail communication between the resolver and ADS), and, finally, download the first segment of the upcoming ad in time to prevent disruption of service at the splice point.

If several emsg boxes are present in a segment and one of them is the MPD Validity Expiration event, emsg carrying it shall always appear first.

10.11.6. MPD events

In addition to tracking events (ad starts, quartile tracking, etc.) the server may also need to signal additional metadata to the video application. For example, an ad unit may contain not only inline linear ad content (that is to be played before, during, or after the main presentation), it may also contain a companion display ad that is to be shown at the same time as the video ad. It is important that the server be able to signal both the presence of the companion ad and the additional tracking and click-through metadata associated with the companion.

With that said, there is no need to have a generic DASH client implement this functionality – it is enough to provide opaque information that the client would pass to an external module. Event @schemeIdUri provides us with such addressing functionality, while MPD events allow us to put opaque payloads into the MPD.

10.11.7. Workflows

In the workflows below we assume that our inputs are MPEG-2 transport streams with embedded SCTE 35 cue messages [54]. In our opinion this will be a frequently encountered deployment, however any other in-band or out-of-band method of getting cue messages and any other input format lend themselves into the same model.

10.11.8. Linear workflow

A real-time MPEG-2 TS feed arrives at both packager and MPD generator. While real-time multicast feeds are a very frequently encountered case, the same workflow can apply to cases such as ad replacement in a pre-recorded content (e.g., in time-shifting or PVR scenarios).

MPD generator generates dynamic MPDs. Packager creates DASH segments out of the arriving feed and writes them into the origin server. Client periodically requests the MPDs so that it has enough time to transition seamlessly into the ad period.

Packager and MPD generator may be tightly coupled (e.g. co-located on the same physical machine), or loosely coupled as they both are synchronized only to the clock of the feed.

Live workflow
10.11.8.1. Cue interpretation by the MPD generator

When an SCTE 35 cue message indicating an upcoming splice point is encountered by the MPD generator, the latter creates a new MPD for the same program, adding a remote period to it.

The Period@start attribute of the inserted period has splice_time() translated into the presentation timeline. Parameters derived from the cue message are inserted into the Period@xlink:href attribute of the inserted period. Examples below show architectures that allow finer targeting.

Immediate ad decision.

MPD generator keeps an up-to-date template of an MPD. At each cue message arrival, the generator updates its template. At each MPD request, the generator customizes the request based on the information known to it about the requesting client. The generator contacts ad decision server and produces one or more non-remote ad periods. In this case XLink is not needed.

Stateful cue translation.

MPD generator keeps an up-to-date template of an MPD. At each cue message arrival, the generator updates its template. At each MPD request, the generator customizes the request based on the information known to it about the requesting client.

The operator targets separately male and female audiences. Hence, the generator derives this from the information it has regarding the requesting client (see 5.1.3.6), and inserts an XLink URL with the query parameter ?gender=male for male viewers, and ?gender=female for the female viewers.

Note that this example also showcases poor privacy practices – would such approach be implemented, both parameter name and value should be encrypted or TLS-based communication should be used

Stateless cue translation.

At cue message arrival, the MPD generator extracts the entire SCTE 35 splice_info_section (starting at the table_id and ending with the CRC_32) into a buffer. The buffer is then encoded into URL-safe base64url format according to RFC 4648 [60], and inserted into the XLink URL of a new remote Period element. splice_time is translated into Period@start attribute. The new MPD is pushed to the origin.

Note: this example is a straightforward port of the technique defined for SCTE 67 [55], but uses base64url and not base64 encoding as the section is included in a URI.

10.11.8.2. Cue interpretation by the packager

Cue interpretation by the packager is optional and is an optimization, rather than core functionality. On reception of an SCTE 35 cue message signaling an upcoming splice, an emsg with MPD Validity Expiration event is inserted into the first available segment. This event triggers an MPD update, and not an ad decision, hence the sum of the earliest presentation time of the emsgbearing segment and the emsg.presentation_time_delta should be sufficiently earlier than the splice time. This provides the client with sufficient time to both fetch the MPD and resolve XLink.

splice_time() of the cue message is translated into the media timeline, and last segment before the splice point is identified. If needed, the packager can also finish the segment at the splice point and thus having a segment shorter than its target duration.

10.11.8.3. Multiple cue messages

There is a practice of sending several SCTE 35 cue messages for the same splice point (e.g., the first message announces a splice in 6 seconds, the second arrives 2 seconds later and warns about the same splice in 4 seconds, etc.). Both the packager and the MPD generator react on the same first message (the 6-sec warning in the example above), and do nothing about the following messages.

10.11.8.4. Cancelation

It is possible that the upcoming (and announced) insertion will be canceled (e.g., ad break needed to be postponed due to overtime). Cancelation is announced in a SCTE 35 cue message.

When cancelation is announced, the packager will insert the corresponding emsg event and the MPD generator will create a newer version of the MPD that does not contain the inserted period or sets its duration to zero. This implementation maintains a simpler less-coupled server side system at the price of an increase in traffic.

10.11.8.5. Early termination

It is also possible that a planned ad break will need to be cut short – e.g., an ad will be cut short and there will be a switch to breaking news. The DASH translation of this would be creating an emsg at the packager and updating the MPD appropriately. Treatment of early termination here would be same as treatment of a switch from main content to an ad break.

It is easier to manipulate durations when Period@duration is absent and only Period@start is used – this way attributes already known to the DASH client don’t change.

10.11.8.6. Informational cue messages

SCTE 35 can be used for purposes unrelated to signaling of placement opportunities. Examples of such use are content identification and time-of-day signaling. Triggering MPD validity expiration and possibly XLink resolution in this case may be an overreaction.

10.11.8.7. Ad decision
Ad decision

A client will attempt to dereference a remote period element by issuing an HTTP GET for the URL that appears in Period@xlink:href. The HTTP server responding to this request (XLink resolver) will contact the ad decision service, possibly passing it parameters known from the request URL and from client information available to it from the connection context. In case described in 5.3.3.2.1.3, the XLink resolver has access to a complete SCTE 35 message that triggered the splice.

The ad decision service response identifies the content that needs to be presented, and given this information the XLink resolver can generate one or more Period elements that would be then returned to the requesting DASH client.

A possible optimization is that resolved periods are cached – e.g. in case of 5.3.3.2.1.1 "male" and "female" versions of the content are only generated once in T seconds, with HTTP caching used to expire the cached periods after T seconds.

10.11.9. On demand workflow

In a VoD scenario, cue locations are known ahead of time. They may be available multiplexed into the mezzanine file as SCTE 35 or SCTE 104, or may be provided via an out-of-band EDL.

In VoD workflows both cue locations and break durations are known, hence there is no need for a dynamic MPD. Thus cue interpretation (which is same as in 5.3.3.2) can occur only once and result in a static MPD that contains all remote elements with all Period elements having Period@start attribute present in the MPD.

In elastic workflows ad durations are unknown, thus despite our knowledge of cue locations within the main content it is impossible to build a complete presentation timeline. Period@duration needs to be used. Remote periods should be dereferenced only when needed for playout. In case of a “jump” – random access into an arbitrary point in the asset – it is a better practice not to dereference Period elements when it is possible to determine the period from which the playout starts using Period@duration and asset identifiers. The functionality described in 5.3.3.2 is sufficient to address on-demand cases, with the only difference that a client should be able to handle zero-duration periods that are a result of avails that are not taken.

10.11.9.1. Capture to VoD

Capture to VoD use case is a hybrid between pure linear and on demand scenarios: linear content is recorded as it is broadcast, and is then accessible on demand. A typical requirement is to have the content available with the original ad for some time, after which ads can be replaced.

There are two possible ways of implementing the capture-to-VoD workflow.

The simplest is treating capture-to-VoD content as plain VoD, and having the replacement policy implemented on the XLink resolver side. This way the same Period element(s) will be always returned to the same requester within the window where ad replacement is disallowed; while after this window the behavior will be same as for any on-demand content. An alternative implementation is described in 5.3.3.5 below.

10.11.9.2. Slates and ad replacement

A content provider (e.g., OTT) provides content with ad breaks filled with its own ads. An ISP is allowed to replace some of these with their own ads. Conceptually there is content with slates in place of ads, but all slates can be shown and only some can be replaced.

An ad break with a slate can be implemented as a valid in-MPD Period element that also has XLink attributes. If a slate is replaceable, XLink resolution will result in new Period element(s), if not – the slate is played out.

10.11.9.3. Blackouts and alternative content

In many cases broadcast content cannot be shown to a part of the audience due to contractual limitations (e.g., viewers located close to an MLB game will not be allowed to watch it, and will be shown some alternative content). While unrelated to ad insertion per se, this use case can be solved using the same “default content” approach, where the in-MPD content is the game and the alternative content will be returned by the XLink resolver if the latter determines (in some unspecified way) that the requester is in the blackout zone.

10.11.9.4. Tracking and reporting

A Period, either local or a remote entity, may contain an EventStream element with an event containing IAB VAST 3.0 Ad element [53]. DASH client does not need to parse the information and act accordingly – if there is a listener to events of this type, this listener can use the VAST 3.0 Ad element to implement reporting, tracking and companion ads. The processing done by this listener does not have any influence on the DASH client, and same content would be presented to both “vanilla” DASH client and the player in which a VAST module registers with a DASH client a listener to the VAST 3.0 events. VAST 3.0 response can be carried in an Event element where EventStream@schemeIdUri value is http://dashif.org/identifiers/vast30.

An alternative implementation uses DASH Callback events to point to the same tracking URLs. While DASH specification permits both inband and MPD Callback events, inband callback events shall not be used.

10.11.10. Examples

MPD with mid-roll ad breaks and default content.

In this example, a movie (“Top Gun”) is shown on a linear channel and has two mid-roll ad breaks. Both breaks have default content that will be played if the XLink resolver chooses not to return new Period element(s) or fails.

In case of the first ad break, SCTE 35 cue message is passed completely to the XLink resolver, together with the corresponding presentation time.

In case of the second ad break, proprietary parameters u and z describe the main content and the publishing site.

<MPD xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
    xmlns="urn:mpeg:dash:schema:mpd:2011"
    xsi:schemaLocation="urn:mpeg:dash:schema:mpd:2011 DASH-MPD.xsd"
    type="dynamic"
    minimumUpdatePeriod="PT2S"
    timeShiftBufferDepth="PT600S"
    minBufferTime="PT2S"
    profiles="urn:mpeg:dash:profile:isoff-live:2011"
    availabilityStartTime="2012-12-25T15:17:50">
    <BaseURL>http://cdn1.example.com/</BaseURL>
    <BaseURL>http://cdn2.example.com/</BaseURL>

    <Period start="PT0.00S" duration="PT600.6S" id="movie period #1">
        <AssetIdentifier schemeIdUri="urn:org:dashif:asset-id:2013"
            value="md:cid:EIDR:10.5240%2f0EFB-02CD-126E-8092-1E49-W"/>
        <AdaptationSet mimeType="video/mp4" codecs="avc1.640828"
            frameRate="24000/1001" segmentAlignment="true" startWithSAP="1">
            <BaseURL>video_1/</BaseURL>
            <SegmentTemplate timescale="90000" initialization="$Bandwidth%/init.mp4v"
                media="$Bandwidth$/$Number%05d$.mp4v"/>
            <Representation id="v0" width="320" height="240" bandwidth="250000"/>
            <Representation id="v1" width="640" height="480" bandwidth="500000"/>
            <Representation id="v2" width="960" height="720" bandwidth="1000000"/>
        </AdaptationSet>
    </Period>

    <Period duration="PT60.6S" id="ad break #1"
            xlink:href="https://adserv.com/avail.mpd?scte35-time=PT600.6S&                  scte35-cue=DAIAAAAAAAAAAAQAAZ_I0VniQAQAgBDVUVJQAAAAH+cAAAAAA%3D%3D"
            xlink:actuate="onRequest" >

        <AdaptationSet mimeType="video/mp4" codecs="avc1.640828"
                       frameRate="30000/1001"
            segmentAlignment="true" startWithSAP="1">
            <BaseURL availabilityTimeOffset="INF">default_ad/</BaseURL>
            <SegmentTemplate timescale="90000" initialization="$Bandwidth%/init.mp4v"
                media="$Bandwidth%/$Time$.mp4v"/>
            <Representation id="v0" width="320" height="240" bandwidth="250000"/>
            <Representation id="v1" width="640" height="480" bandwidth="500000"/>
            <Representation id="v2" width="960" height="720" bandwidth="1000000"/>
        </AdaptationSet>
    </Period>

    <!—Movie, cont’d -->
    <Period duration="PT600.6S" id="movie period #2">
        <AssetIdentifier schemeIdUri="urn:org:dashif:asset-id:2013"
            value="md:cid:EIDR:10.5240%2f0EFB-02CD-126E-8092-1E49-W"/>
        <AdaptationSet mimeType="video/mp4" codecs="avc1.640828"
                       frameRate="24000/1001"
            segmentAlignment="true" startWithSAP="1">
            <BaseURL>video_2/</BaseURL>
            <SegmentTemplate timescale="90000" initialization="$Bandwidth%/init.mp4v"
                media="$Bandwidth%/$Time$.mp4v"/>
            <Representation id="v0" width="320" height="240" bandwidth="250000"/>
            <Representation id="v1" width="640" height="480" bandwidth="500000"/>
            <Representation id="v2" width="960" height="720" bandwidth="1000000"/>
        </AdaptationSet>
    </Period>

    <Period duration="PT60.6S" id="ad break #2"
        xlink:href=”https://adserv.com/avail.mpd?u=0EFB-02CD-126E-8092-1E49-W&z=spam”
        xlink:actuate="onRequest" >

        <AdaptationSet mimeType="video/mp4" codecs="avc1.640828"
                       frameRate="30000/1001"
            segmentAlignment="true" startWithSAP="1">
            <BaseURL availabilityTimeOffset="INF">default_ad2/</BaseURL>
            <SegmentTemplate timescale="90000" initialization="$Bandwidth%/init.mp4v"
                media="$Bandwidth%/$Time$.mp4v"/>
            <Representation id="v0" width="320" height="240" bandwidth="250000"/>
            <Representation id="v1" width="640" height="480" bandwidth="500000"/>
            <Representation id="v2" width="960" height="720" bandwidth="1000000"/>
        </AdaptationSet>
    </Period>
</MPD>

10.11.11. Use of query parameters

Parameters can be passed into the XLink resolver as a part of the XLink URL. Clause 5.3.3.2.1.3 shows an example of this approach when an SCTE 35 cue message is embedded into the XLink URL.

This approach can be generalized and several parameters (i.e., name-value pairs) can be defined. SCTE 214-1 2016 [56] takes this approach and defines parameters expressing splice time (i.e., Period@start of the earliest ad period), SCTE 35 cue message, and syscode (a geolocation identifier used in US cable industry). The first two parameters are also shown in example in clause 5.3.4.1 of this document.

Note: Effectively this creates a RESTful API for XLink dereferencing. While discussion above implies that these parameters are embedded by the MPD generator into the XLink URL, the parameter values may as well be calculated by the client or the embedded values may be modified by the client.

Note: The same RESTful API approach can be used with MPD URLs as well.

Note: More parameters may be defined in the future version of these guidelines.

10.12. App-based architecture

App-based architecture

Inputs in this use case are same as the ones described in sec. 5.3. At the packaging stage, cues are translated into a format readable by the app or/and DASH client and are embedded into media segments or/and into the manifest.

Ad management module is located at the client side. The DASH client receives manifest and segments, with cues embedded in either one of them or in both.

Cue data is passed to the ad management module, which contacts the ad decision service and receives information on content to be played. This results in an MPD for an inserted content and a splice time at which presentation of main content is paused and presentation of the inserted content starts.

Note that this architecture does not assume multiple decoders – with careful conditioning it is possible to do traditional splicing where inserted content is passed to the same decoder. In this case it is necessary to keep a player state and be able to initialize a player into this state.

10.12.1. Implementation basics

Each ad decision results in a separate MPD. A single MPD contains either main content or inserted content; existence of multiple periods or/and remote periods is possible but not essential.

10.12.2. SCTE 35 events

Cue messages are mapped into DASH events, using inband emsg boxes and/or in-MPD events. Note that SCTE 35 cue message may not be sufficient by itself.

The examples below show use of SCTE 35 in user-defined events, and presentation time indicates the timing in within the Period.

Figure 18 below shows the content of an emsg box at the beginning of a segment with earliest presentation time T. There is a 6-sec warning of an upcoming splice – delta to splice time is indicated as 6 seconds – and duration is given as 1 minute. This means that an ad will start playing at time T + 6 till T + 66. This example follows a practice defined in SCTE 214-3 [57].

Inband carriage of SCTE 35 cue messages

Figure 19 below shows the same example with an in-MPD SCTE35 cue message. The difference is in the in-MPD event the splice time is relative to the Period start, rather than to the start of the event-carrying segment. This figure shows a one-minute ad break 10 minutes into the period.

<EventStream schemeIdUri="urn:scte:scte35:2014:xml+bin">
    <Event timescale="90000" presentationTime="54054000" duration="5400000" id="1">
        <scte35:Signal>
             <scte35:Binary>
                 /DAIAAAAAAAAAAAQAAZ/I0VniQAQAgBDVUVJQAAAAH+cAAAAAA==
             </scte35:Binary>
         </scte35:Signal>
    </Event>
</EventStream>
In-MPD carriage of SCTE 35 cue message

Note: for brevity purposes SCTE 35 2014 allows use of base64-encoded section in Signal.Binary element as an alternative to carriage of a completely parsed cue message.

Normative definitions of carriage of SCTE 35 cue messages are in ANSI/SCTE 214-1 [56] sec 6.8.4 (MPD) and SCTE 214-3 [57] sec 8.3.3.

10.12.3. Asset identifiers

See sec. 5.3.2.2 for details.

10.12.4. Linear workflow

Linear workflow for app-driven architecture

A real-time MPEG-2 TS feed arrives at a packager. While real-time multicast feeds are a very frequently encountered case, the same workflow can apply to cases such as ad replacement in a pre-recorded content (e.g., in time-shifting or PVR scenarios).

Packager creates DASH segments out of the arriving feed and writes them into the origin server. The packager translates SCTE 35 cue messages into inband DASH events, which are inserted into media segments.

MPD generator is unaware of ad insertion functionality and the packager does the translation of SCTE 35 cue messages into inband user-defined DASH events. On reception of an SCTE 35 cue message signaling an upcoming splice, a emsg with a translation of the cue message in its emsg.message_data[] field is inserted into the most recent Segment. This event triggers client interaction with an ad decision server, hence the sum of the earliest presentation time of the emsg-bearing segment and the emsg.presentation_time_delta should be a translation of splice_time() into the media timeline.

An alternative implementation which is more compatible with server-based architecture in section 5.3, an MPD generator can generate separate MPDs for both server-based and app-based architectures creating remote periods for server-based and in-MPD SCTE 35 events for app-based architectures, while a packager can insert inband MPD validity expiration events.

A DASH client will pass the event to the app controlling it (e.g., via a callback registered by the app). The app will interpret the event and communicate with the ad decision server using some interface (e.g., VAST). This interface is out of the scope of this document.

The communication with ad decision service will result in an MPD URL. An app will pause the presentation of the main content and start presentation of the inserted content. After presenting the inserted content the client will resume presentation of the main content. This assumes either proper conditioning of the main and inserted content or existence of separate client and decoder for inserted content. The way pause/resume is implemented is internal to the API of the DASH client. Interoperability may be achieved by using the DASH MPD fragment interface, see ISO/IEC 23009-1 [4], Annex C.4

10.12.5. On demand workflow

As in the server-based case, functionality defined for the live case is sufficient. Moreover, the fact that that app-based implementation relies heavily on app’s ability to pause and resume the DASH client, support for elastic workflows is provided out of the box.

In the on demand case, as cue locations are well-known, it is advantageous to provide a static MPD with SCTE 35 events than run a dynamic service that relies on inband events.

10.13. AssetIdentifier extensions

What are "extensions"? Move this to features/constraints chapters?

AssetIdentifier descriptor shall be used for distinguishing parts of the same asset within a multi-period MPD, hence it shall be used for main content and may be used for inserted content. In order to enable better tracking and reporting, unique IDs should be used for different assets.

Use of EIDR and Ad-ID identification schemes is recommended. The value of @schemeIdUri set to "urn:eidr" signals use of EIDR. The value of @value attribute shall be a valid canonical EIDR entry as defined in [67].

Use of Ad-ID for asset identification is signaled by setting the value of @schemeIdUri to "urn:smpte:ul:060E2B34.01040101.01200900.00000000" ("designator" URN defined in SMPTE 2092-1 [68]). The value of @value attribute shall be a canonical full Ad-ID identifier as defined in SMPTE 2092-1 [68].

Other schemes may be used, including user private schemes, by using appropriately unique values of @schemeIdUri.

In the absence of other asset identifier schemes, a DASH-IF defined scheme may be used with the value of @schemeIdUri set to "urn:org:dashif:asset-id:2014". If used, the value of @value attribute descriptor shall be a MovieLabs ContentID URN ([58], 2.2.1) for the content. It shall be the same for all parts of an asset. Preferred schemes are EIDR (main content) and AdID (advertising).

If a Period has one-off semantics (i.e., an asset is completely contained in a single period, and its continuation is not expected in the future), the author shall not use asset identifier on these assets.

Periods that do not contain non-remote AdaptationSet elements, as well as zero-length periods shall not contain the AssetIdentifier descriptor.

An MPD may contain remote periods, some of which may have default content. Some of which are resolved into multiple Period elements.

After dereferencing MPD may contain zero-length periods or/and remote Periods.

In case of Period@xlink:actuate="onRequest", MPD update and XLink resolution should be done sufficiently early to ensure that there are no artefacts due to insufficient time given to download the inserted content.

Period@xlink:actuate="onRequest" shall not be used if MPD@type ="dynamic" 5

10.15. User-defined event extensions

10.15.1. Cue message

Cue messages used in app-driven architecture shall be SCTE 35 events [54]. SCTE 35 event carriage is defined in ANSI/SCTE 214-1 (MPD) and ANSI/SCTE 214-3 (inband). For MPD events, the XML schema is defined in SCTE 35 2014 [54] and allows either XML representation or concise base64-coded representation.

NOTE: PTS offset appearing in SCTE 35 shall be ignored, and only DASH event timing mechanism may be used to determine splice points.

10.15.2. Reporting

MPD events with embedded IAB VAST 3.0 [53] response may be used for reporting purposes.

If only time-based reporting is required (e.g., reporting at start, completion, and quartiles), use of DASH callback event may be a simpler native way of implementing tracking. Callback events are defined in ISO/IEC 23009-1:2014 AMD3 [4].

10.15.3. Ad insertion event streams

Recommended Event Stream schemes along with their scheme identifier for app-driven ad insertion are:

  1. "urn:scte:scte35:2013:bin" for inband SCTE 35 events containing a complete SCTE 35 section in binary form, as defined in ANSI/SCTE 214-3.

  2. “urn:scte:scte35:2014:xml+bin” for SCTE 35 MPD events containing only base64 cue message representation, as defined in ANSI/SCTE 214-1. NOTE: the content of Event element is an XML representation of the complete SCTE 35 cue message, that contains Signal.Binary element rather than the Signal.SpliceInfoSection element, both defined in SCTE 35 2014.

  3. "http://dashif.org/identifiers/vast30" for MPD events containing VAST3.0 responses [53].

  4. urn:mpeg:dash:event:callback:2015 for DASH callback events.

11. Media coding technologies

This chapter describes the constraints that apply to media codecs when used in interoperable services.

Services SHALL use only the media codecs described in this chapter, in conformance with the requirements defined here.

Clients MAY support any set of codecs described in this chapter and SHALL NOT attempt to play back representations for which they do not have codec support.

11.1. H.264 (AVC)

The H.264 (AVC) codec [MPEGAVC] MAY be used by services for video adaptation sets. Clients SHOULD support this codec.

For representations up to 1280x720p resolution and up to 30 fps, the H.264 (AVC) Progressive High Profile Level 3.1 decoder SHALL be used.

For representations up to 1920x1080p resolution and up to 30 fps, the H.264 (AVC) Progressive High Profile Level 4.0 decoder SHALL be used.

The encapsulation of H.264 data in DASH containers SHALL conform to [iso14496-15].

Clients SHALL support SPS/PPS storage both in the initialization segment (sample entry avc1) and inband storage (sample entry avc3). Services MAY use either form.

Note: Use of avc3 is one of the factors that enables bitstream switching.

The below table lists examples of @codecs strings for H.264 (AVC) that match the decoders defined in this chapter.

Profile Level @codecs
H.264 (AVC) Progressive High Profile 3.1 avc1.64Y01F
avc3.64Y01F
4.0 avc1.64Y028
avc3.64Y028
Example @codecs strings for H.264 (AVC)

Note: Other @codecs strings may also be compatible (a higher level decoder can typically decode content intended for a lower level decoder).

For a detailed description on how to derive the signaling for the codec profile for H.264/AVC, see [DVB-DASH] section 5.1.3.

11.2. H.265 (HEVC)

The H.265 (HEVC) codec [MPEGHEVC] MAY be used by services for video adaptation sets.

For representations up to 1280x720p at up to 30 fps, the HEVC Main Profile Main Tier Level 3.1 decoder SHALL be used.

For representations up to 2048x1080 at up to 60 fps at 8-bit frame depth, the HEVC Main Profile Main Tier Level 4.1 decoder SHALL be used.

For representations up to 2048x1080 at up to 60 fps at 10-bit frame depth, the HEVC Main10 Profile Main Tier Level 4.1 decoder SHALL be used.

The encapsulation of H.265 data in DASH containers SHALL conform to [iso14496-15].

Clients SHALL support VPS/SPS/PPS storage both in the initialization segment (sample entry hvc1) and inband storage (sample entry hev1). Services MAY use either form.

Note: Use of hev1 is one of the factors that enables bitstream switching.

Where does UHD fit? Why is it in a separate chapter? We should unify.

The [ISOBMFF] sync sample signaling and [MPEGDASH] SAP type signaling SHALL be derived from the following table.

NAL unit type [ISOBMFF] sync sample flag [MPEGDASH] SAP type
IDR_N_LP true 1
IDR_W_RADL true 2 (if the IRAP has associated RADL pictures)
1 (if the IRAP has no associated RADL pictures)
BLA_N_LP true 1
BLA_W_RADL true 2 (if the IRAP has associated RADL pictures)
1 (if the IRAP has no associated RADL pictures)
BLA_W_LP false 3 (if the IRAP has associated RASL pictures)
true 2 (if the IRAP has no associated RASL pictures but has associated RADL pictures
true 1 (if the IRAP has no associated leading pictures)
CRA false 3 (if the IRAP has associated RASL pictures)
true 2 (if the IRAP has no associated RASL pictures but has associated RADL pictures)
true 1 (if the IRAP has no associated leading pictures)
Signaling dependent on HEVC IRAP pictures in [ISOBMFF] and [MPEGDASH].

IOP requires that each media segment start with SAP type 1 or 2. If the above table indicates SAP type 3, the content is not conforming to IOP.

When the table above lists multiple possible values for a given NAL unit type and the entity creating the signaling is not able to determine correctly which values to use, it SHALL use the first value listed in the table for that NAL unit type.

The below table lists examples of @codecs strings for H.265 (HEVC) that match the decoders defined in this chapter.

Profile Level @codecs
HEVC Main 3.1 hev1.1.2.L93.B0
hvc1.1.2.L93.B0
4.1 hev1.1.2.L123.B0
hvc1.12.L123.B0
HEVC Main-10 4.1 hev1.2.4.L123.B0
hvc1.2.4.L123.B0
Example @codecs strings for H.265 (HEVC)

Note: Other @codecs strings may also be compatible (a higher level decoder can typically decode content intended for a lower level decoder).

For a detailed description on how to derive the signaling for the codec profile for H.265/HEVC, see [DVB-DASH] section 5.2.2.

11.3. Decoder configuration with H.264 and H.265

This chapter applies only to video adaptation sets that use H.264 or H.265.

All initialization segments in the same video adaptation set SHALL use the same sample description (i.e. no mixing of avc1 and avc3 is allowed).

In representations using avc1 or hvc1 sample description:

In representations using avc3 or hev1 sample description:

11.4. Bitstream switching with H.264 and H.265

This chapter applies only to bitstream switching adaptation sets that use H.264 or H.265.

All representations SHALL be encoded using the avc3 or hev1 sample description.

The first presented sample’s composition time SHALL equal the first decoded sample’s decode time, which equals the baseMediaDecodeTime in the Track Fragment Decode Time Box (tfdt).

Note: This requires the use of negative composition offsets in a v1 Track Run Box (trun) for video samples, otherwise video sample reordering will result in a delay of video relative to audio.