Impact of Temporal Sub-Sampling in AV1 Archives

This article examines how temporal sub-sampling affects compression efficiency, file sizes, and visual quality in AV1 video archives. It explores the mechanics of AV1 inter-frame prediction, explains why halving frame rates does not equate to halving file size, and evaluates the trade-offs between storage reduction and preservation quality. Readers will gain actionable insight into optimizing archival encoding pipelines without degrading original footage fidelity.

Understanding Temporal Sub-Sampling in AV1

Temporal sub-sampling refers to reducing the frame rate of a video stream before encoding—for instance, downsampling native 60 frames per second (fps) capture to 30 fps, or 24 fps to 12 fps. In archival preservation, reducing the frame count is often considered as a method to lower bitrates and save long-term storage costs. However, in modern codecs like AV1, temporal frequency directly interacts with motion-compensated prediction algorithms, leading to complex trade-offs in coding efficiency.

The Cost to Motion Compensation

AV1 achieves high compression ratios primarily through advanced inter-frame prediction tools, such as Overlapped Block Motion Compensation (OBMC), warped motion (affine transformation models), and compound inter-intra prediction. These mechanisms rely on small temporal distances between successive frames.

When a video is temporally sub-sampled:

• Motion Displacements Increase: Increasing the temporal gap between frames forces motion vectors to span larger spatial distances.
• Non-Linear Motion Increases: Complex real-world movement, camera pans, and occlusions become harder to track linearly, increasing the prediction residual.
• Higher Residual Energy: Because the reference frames are further apart in time, the encoder can no longer rely on simple translations. It must allocate significantly more bits to encode high-frequency residual errors via DCT or asymmetric discrete sine transforms (ADST).

Consequently, while there are fewer total frames to encode, the cost in bits to encode each remaining P-frame or B-frame rises sharply.

Diminishing Returns on File Size

A common misconception in video archival is that cutting the frame rate by 50% cuts the storage footprint in half. In AV1 encoding under constant quality modes (such as Constant Rate Factor or CRF):

1. Intra-Frames (Keyframes): Keyframes remain essentially unaffected in size, as they do not reference neighboring frames.
2. Inter-Frames: P- and B-frames grow substantially larger due to the degraded efficiency of the motion estimation pipeline.
3. Net Storage Savings: The actual reduction in storage typically ranges between 15% and 30%, far below the 50% frame reduction. In scenes with erratic motion, the efficiency penalty can be even more severe, yielding negligible storage gains.

Impact on Perceptual Quality and Archival Value

Archival preservation aims to retain original signal integrity for future retrieval or restoration. Applying temporal sub-sampling creates irreversible artifacts:

• Stroboscopic Motion and Judder: High-motion sequences lose temporal smoothness, producing visual judder that cannot be corrected later.
• Encoder Artifacts: Because AV1 must distribute its bit budget to handle larger prediction errors across fewer frames, scenes with rapid motion may exhibit blocking, ringing, or texture blurring on high-contrast edges.
• Synthetic Film Grain Disruption: AV1 features built-in film grain synthesis, which models and strips noise before re-applying it at playback. Sub-sampling changes the temporal consistency of grain across frames, impairing the encoder’s ability to separate true motion from grain structure.

Practical Recommendations for AV1 Archiving

To maximize compression efficiency while safeguarding long-term usability, adhere to the following principles:

• Retain Native Frame Rates for Master Archives: Do not downsample frame rates for long-term preservation masters. Instead of temporal sub-sampling, increase the CRF value or leverage AV1's multi-reference frame capabilities to achieve storage targets without discarding temporal data.
• Reserve Sub-Sampling for Static Content: Temporal sub-sampling is viable only for content categories with minimal motion, such as slides, desktop recordings, security feeds, or talking-head interviews where motion vectors remain small even across dropped frames.
• Tune Motion Estimation Settings: If temporal sub-sampling is mandatory due to strict bandwidth or hardware playback constraints, configure the AV1 encoder to use broader motion estimation search ranges (--motion-mode and deeper lookahead buffers) to mitigate prediction loss across larger temporal intervals.