Introduction

The purpose of this article is to provide recommendations and guidelines concerning the basic framing of Serialized ADM (S-ADM) for broadcast use-cases. The ITU and SMPTE standards are flexible in this area leaving many options to implementors. These guidelines are intended to help vendors achieve interoperability and avoid pit-falls in design. The primary consideration for these guidelines is maintaining S-ADM correctly as video and audio are edited or processed. As experience with S-ADM enabled products grow these guidelines may evolve.

These guidelines are primarily aimed at uncompressed workflows in a facility where an uncompressed audio stream compliant with SMPTE ST 2110-30 is accompanied by a metadata stream compliant with SMPTE ST 2110-41 or 2110-31. The guidelines are also applicable to SDI workflows using SMPTE ST 2116 to carry the S-ADM inside an audio channel.

Metadata Transmission Periodicity and Synchronization

For uncompressed workflows, it is recommended that S-ADM metadata be sent repeatedly and synchronized to the accompanying video, if it exists. If sub video frame latencies are required then a multiple of the video frame rate may be used i.e. 2x or 4x. For non-integer video frame rates the conventional cadence of frame sizes are used. At the end of each cadence, the time-stamp should align with the video frame boundary.

Serialized ADM Frame Format

The timing information described in the frameFormat XML element must reflect the metadata update period avoiding overlaps and gaps. The frameFormatID attribute must be updated for every frame that is transmitted. For uncompressed workflows the duration attribute will typically take the form similar to this example for 25fps:

duration="00:00:00.01920S48000"

Compressed Audio Workflows

For compressed workflows, typically where audio is transmitted over a compressed link, a lower than video frame metadata update period may be used to reduce the metadata bandwidth requirement. The minimum recommended is 1fps. If a change occurs to the metadata, for example, a program transition of gain parameter update, the new metadata may be sent immediately or queued for transmission at the metadata repetition period. In this scenario, receivers should use a "last known good" reversion strategy so that audio can be rendered if metadata is lost.

Workflows with multiple frame rates

It is common for linear workflows involving repository files to ingest content with different frame rates. To obtain consistency downstream, frame rate conversion is required.

If lower frame rate content is ingested into an environment using a higher frame rate, for example, 25fps content ingested into a 50fps environment, then new S-ADM metadata frames should be inserted with timing information for the new higher frame rate. This conversion may typically occur in a file context or during play-out. The frameFormat XML element must be updated to reflect the new repetition rate. For static metadata then a simple frame repeat of the S-ADM is sufficient. For dynamic metadata with moving objects, object position interpolation is required to achieve smoother object movement. The details of performing the object position interpolation are outside the scope of this article.

When performing frame conversion from low frame rates to high frame rates a simple implementation may avoid the metadata frame insertion. The pitfalls of a such a simple implementation are shown below. 

If the S-ADM stream is not populated with the additional frames to match the new higher frame rate then gaps will appear in stream. In the absence of switching, these gaps present no problem as the timing information of the low frame metadata spans two frames. However, if a switch occurs when a gap starts then there is at period immediately after the switch point where no valid metadata exists. This can be remedied by last-known good revision resulting is a mismatch between the audio and metadata. Alternatively if the gap appears at the start of the stream then no audio will be available for the gap period due to a lack of metadata and audio configuration information required for audio rendering.

If higher frame rate content is ingested into a system using a lower frame rate then decimation must occur with metadata being dropped. Decimation should always occur with knowledge of where program boundaries occur, such as in the file domain. If blind decimation occurs it is possible for a program boundary to occur mid-frame making a synchronized audio and metadata change impossible without retaining the high frame rate for the metadata.