Encoding parameters for spatial layers

[Orphis]

It would be interesting to be able to control the encoding parameters for the spatial layers in a similar way we can control simulcast layers. I was thinking of an array of spatial coding parameters nested in the RTCRtpEncodingParameters dictionnary, one entry per layer.

I believe we should be able to have parameters like maxBitrate, maxFramerate, scaleResolutionDownBy and the active flag.

• Bitrate and framerate are quite self explanatory.
• scaleResolutionDownBy should be decreasing for each spatial layer (aka layers have to be bigger). There will be some interaction with the same encoding parameter for simulcast to define.
• active is a bit tricky when there are layer dependencies and should disable that layer and all layers having dependencies on disabled layers. This might not apply in "S" modes.

[aboba]

Related to Issues #3 and #4

[aboba]

@Orphis The ORTC API described dependencies between SVC layers via framerateScale, encodingId and dependencyEncodingIds . However I believe this had some limitations with respect to KEY and KEY_SHIFT modes.

Note that the AV1 Bitstream Specification supports both pre-canned Scalability Modes as well as custom Scalability Structures via the syntax in Section 5.8.6.

[Orphis]

As per discussions we had, let's just try to keep the active boolean and maxBitrate values.
It could look like this:

let encodings = [{
    scalabilityMode = 'L3T3',
    spatialEncodings = [
        { active: true, maxBitrate: 100000 },
        { active: true, maxBitrate: 300000 },
        { active: true, maxBitrate: 600000 },
    ],
}];

[aboba]

The above proposal provides some additional flexibility with respect to spatial ratios as a side effect. For example:

// Provide a full resolution simulcast encoding as well as a quarter-size (thumbnail) encoding
let encodings = [{
    scalabilityMode = 'S3T1',
    spatialEncodings = [
        { active: true },
        { active: false },
        { active: true}
    ],
}];

[aboba]

Additional flexibility can also be provided by defining additional pre-canned modes. For example, popular 16 x 9 resolutions include:

• 1080p (1920 x 1080)
• 720p (1280 x 720)
• 360p (640 x 360)

Since the spatial ratios vary between layers (1.5:1 as well as 2:1), this is not covered by existing modes. However, a new mode could be defined to cover this use case.

[vr000m]

Curious about the status of this issue, what do we need to do to move this forward?

[aboba]

We would need implementer interest to move forward.

@Orphis @HTA -- Is there continued interest in this?

[Orphis]

@vr000m We need the rest of the SVC APIs implemented first to get some implementation experience.
We'll move forward with this then. We have certainly not forgotten about it.

[aboba]

Can we get a PR submitted for discussion at TPAC?