Skip to content

PATTERN Cited by 1 source

Adaptive temporal-layer activation

Enable temporal scalability layers only when the network exhibits packet loss, and disable them when conditions are stable — trading compression efficiency for resilience dynamically rather than paying the efficiency cost continuously.

Structure

  1. Normally encode with a tight prediction chain (each frame references the immediately preceding frame) for maximum compression efficiency.
  2. Monitor network loss feedback from the peer.
  3. When loss rate rises above threshold → enable temporal layers (base layer + enhancement layer). FEC protects base-layer data.
  4. When network recovers → disable temporal layers, return to tight prediction chain.
  5. The switch between modes must handle the reference-chain transition cleanly.

Why not always-on TL?

Temporal layers reduce compression efficiency because the base layer references only every Nth frame (looser temporal correlation). On stable networks (the majority of call duration for many users), this is wasted efficiency — quality degrades at a given bitrate for no benefit.

Why not always-off TL?

Under packet loss, a tight prediction chain means any lost frame breaks the decode chain entirely. Recovery requires retransmission (adds RTT of latency) or a keyframe (causes bitrate spike + congestion). Temporal layers let the base layer survive loss gracefully.

Canonical instance

Meta's AV1 RTC deployment (Source: sources/2026-06-22-meta-adopting-av1-for-real-time-communication-rtc-at-scale): - Two-layer structure (TL0 base + TL1 enhancement). - FEC applied to base layer only. - Enhancement-layer retransmissions handled conservatively (skip when RTT is high). - Activation/deactivation driven by network feedback.

Seen in

Last updated · 559 distilled / 1,651 read