AV1 Low-Delay Mode for Conversational Video

AV1 low-delay mode is an encoder configuration designed for real-time, bidirectional video communications such as WebRTC calls, live streaming, and interactive cloud applications. This mode optimizes the AV1 codec to prioritize minimal end-to-end latency over maximum compression efficiency, ensuring that video frames are captured, compressed, transmitted, and decoded instantaneously. By disabling frame reordering, enforcing strict buffer limits, and utilizing real-time speed presets, AV1 provides viable high-quality compression suitable for interactive communication environments.

Elimination of Display Delay

Standard video compression typically utilizes bidirectional predicted frames (B-frames), which reference both past and future frames to achieve higher compression ratios. In conversational video, referencing future frames requires buffering and introduces display latency.

AV1 low-delay mode enforces a strictly causal prediction structure:

• No Frame Reordering: The encoder operates in an IPPP (Intra and Predicted frames) or low-delay B configuration where frames are decoded in the exact order they are displayed.
• Lag-in-Frames Set to Zero: The encoder lookahead is disabled (--lag-in-frames=0 in reference implementations like libaom). This prevents the encoder from waiting to analyze future frames before deciding how to compress the current one.

Encoder Preset and Speed Profiles

Real-time video requires the encoding process to execute within the duration of a single frame capture (e.g., within 33.3 milliseconds for a 30 fps stream).

• Real-time Usage Profile: The encoder is set to real-time mode (--usage=realtime).
• CPU Utilization: AV1 provides multiple encoding speed settings. For real-time operations, higher speed presets (typically --cpu-used=6 through --cpu-used=10 in libaom-av1) are selected to reduce computational complexity by simplifying motion estimation, partition search trees, and transform selection.

Rate Control and Buffer Management

Network fluctuations in conversational video can cause latency spikes if the bitrate suddenly surges. Low-delay configurations maintain strict bitstream compliance:

• Constant Bitrate (CBR): Real-time modes typically rely on CBR (--end-usage=cbr) to keep the data rate predictable over the network.
• Low Buffer Tolerances: Buffer sizes are configured to match minimal playout delays. Flags configuring the client buffer size, initial buffer fullness, and target optimal buffer level are tuned to low millisecond windows (e.g., 600ms or lower) so the rate controller aggressively penalizes frame size overshoots.

Error Resilience and Intra-Refresh

Standard periodic keyframes (IDR frames) introduce large instantaneous data spikes that often cause packet loss or buffering on consumer networks.

• Cyclic Intra-Refresh: Instead of sending a complete full-frame keyframe to refresh the stream, the encoder updates a horizontal or vertical slice of the frame with intra-blocks over a succession of frames.
• Error Resilient Flag: Enabling error-resilient mode ensures that frame headers and entropy coding contexts do not depend on previously transmitted frames that may have been lost in transit.

Scalable Video Coding (SVC)

Conversational AV1 frequently integrates temporal and spatial Scalable Video Coding (SVC). Under low-delay configurations, the stream can be split into multiple sub-layers (e.g., lower framerates or resolutions) using internal reference frame buffers. If a network path degrades, middle-mile servers or the decoder can selectively drop higher-layer packets without breaking the decoding loop of the base layer, maintaining interactive latency without stream interruption.