Period connectivity - all representations required or only some allowed?

[sandersaares]

Consider an adaptation set AS and representations R1-R4 as follows.

AS in period A:

• R1
• R2
• R3

AS in period B:

• R1 (connected to period A)
• R2 (connected to period A)
• R4 (not connected)

Is this AS period-connected?

DASH does not explicitly state this, as far as I can tell. I would guess clients would not expect connectivity to apply if only a subset of representations are connected, so I suggest DASH-IF notify MPEG of this ambiguity and propose to constrain it to only cases where all the contents of an adaptation set are connected.

[haudiobe]

How can R4 be not connected if R1 and R2 are?

[sandersaares]

Question answered in IOP call: connectivity is an adaptation set level concept, there should not be nonconnected representations appearing/disappearing. This will be integrated into timing model text as part of Dash-Industry-Forum/Guidelines-TimingModel#2

Answer was found but I leave this issue open as reminder that we should ask MPEG to clarify this.

[sandersaares]

Here is a scenario where representation-level is desirable:

1. I encounter segment loss between encoder and packager. The 720p representation has a gap!
2. I do not want gaps in my MPD, so I remove the 720p representation. All others remain connected and I want to signal this.
3. But oh no, I cannot signal this if adaptation set level continuity is expected (all representations must exist).

From this viewpoint, I can see a valuable use case for allowing the set of representations to be different, as long as the difference is not additive (no new decoder requirements in the period that follows).