AV1 Video Codec: Adoption for Real-Time Communications

The AV1 video codec was standardised by the Alliance for Open Media (AOMedia) in 2018 as the organisation's first codec. Since then, development and implementations have progressed, and the codec has been deployed at large scale for video streaming by companies including YouTube, Netflix and Meta. AOMedia's stated goal is to create license-free technology, which open-source implementations have made easier to achieve.

• Standardised in 2018 by AOMedia
• AOMedia's first codec
• Deployed at scale by YouTube, Netflix and Meta
• Open-source implementations available

Meta states that now is the time to start adopting AV1 for real-time communications in order to deliver higher-quality calling. Multiple implementations are available, and commercial, open-source and hardware-offload support has begun shipping in Apple, MediaTek and Qualcomm chipsets. Because most players are upgrading from H.264, a 20-year-old codec specification, the move represents a large technical leap.

• Hardware-offload support in Apple, MediaTek and Qualcomm chipsets
• Upgrade path from H.264 rather than HEVC/H.265
• AV1 enabled in the Messenger app for video calls

The main reason cited for moving to a more advanced codec is that the same quality experience can be delivered at a much lower bitrate, which benefits users on bandwidth-constrained networks. Quality is compared using the BD-BR (Bjontegaard Delta-Bit Rate) metric, derived from rate-distortion curves generated at multiple bitrates. Comparing Meta's existing H.264-based solution with AV1 shows BD-BR savings of around 30% on typical video-calling content such as a talking head. In a side-by-side Messenger demonstration on two iPhones with bitrate restricted to 100 kbps, the H.264 video appears blurry while the AV1 video is clearer.

• Around 30% BD-BR savings versus H.264
• Measured with rate-distortion curves using Libaom and OpenH264
• Screen-content coding tools
• Reference picture resampling (RPR) to eliminate key frames
• Temporal, spatial and quality scalability layers

Codec licensing is described as one of the larger challenges for real-time communication, because both encoder and decoder support must run on large numbers of end-user devices. The previous codec generation, HEVC/H.265, saw little adoption in real-time use cases, mainly because of its complex licensing situation; since calling services are usually free for consumers, there is little incentive to purchase an expensive IP licence. Meta's account of adopting AV1 also addresses engineering constraints including binary size, CPU usage, memory usage and in-product measurement.

• Licensing complexity limited HEVC/H.265 in RTC
• Encoder and decoder support required on end-user devices
• Binary size constraints
• CPU and memory usage
• In-product measurements

Beyond calling, AV1 features in Meta's wider video infrastructure, which handles billions of video uploads per day across Facebook, Instagram, Messenger, WhatsApp and Oculus. Reducing the storage cost of legacy progressive MP4 files frees space that can instead be used to generate higher-quality encodings such as VP9 and AV1. AV1 has also been a recurring topic at Video @Scale, where the 2019 event included an AV1 panel, a YouTube talk on delivering AV1 and codec switching, and a Facebook talk comparing the compute complexity and compression efficiency of AVC, VP9 and AV1.

• Higher-quality encodings generated as VP9 and AV1
• Storage savings via virtual video files
• AV1 panel at Video @Scale 2019
• Talks on AV1 delivery and codec switching at YouTube