Skip to content

PATTERN Cited by 1 source

LTR proactive and reactive recovery

Combine two LTR activation paths — reactive (on freeze) and proactive (on elevated loss) — to minimize video freezes without waiting for the decode chain to fully break before acting.

Structure

Reactive path (RPSI-triggered)

  1. Receiver experiences a video freeze (decode chain broken).
  2. Receiver sends RPSI (Reference Picture Selection Indication) to the sender.
  3. Sender produces an LTRP frame predicted from the most recently ACKed LTR.
  4. Receiver decodes LTRP → chain resynchronized.

Proactive path (loss-detected)

  1. Sender detects elevated packet loss via continuous feedback channel.
  2. Sender requests periodic LTRP frames from the encoder before a freeze occurs.
  3. If a lost frame would have broken the chain, the next scheduled LTRP bridges the gap.

Fallback

If no ACKed LTR is available in the encoder's bounded reference buffer (size 4), the encoder sends a keyframe instead.

Why dual-path?

The reactive path alone adds one full RTT of latency between freeze detection and recovery. The proactive path is somewhat redundant (generates LTRP frames that may not be needed), but significantly reduces freeze duration by pre-positioning recovery frames during periods of elevated loss.

Implementation detail (Meta AV1)

  • Explicit LTR indicator via proprietary RTP header extension (AV1 bitstream doesn't distinguish LTR from non-LTR).
  • ACK feedback via separate RTP header extension carrying frame_id.
  • LTR frames are combined with periodic higher-quality frames the encoder already emits — no additional quality cost for the LTR.

Source: sources/2026-06-22-meta-adopting-av1-for-real-time-communication-rtc-at-scale

Seen in

Last updated · 559 distilled / 1,651 read