Abstract:
Video streaming today accounts for more than 80% of global internet traffic, with demands projected to rise further due to UHD, 8K, VR, and immersive media applications. While newer codecs like AV1 and VVC deliver better compression efficiency, their increased computational complexity significantly raises energy consumption in encoding, storage, transmission, and decoding. This creates a fundamental tension between maintaining high Quality of Experience (QoE) for users and ensuring sustainable, energy-efficient streaming.
This talk presents a holistic framework for energy-aware adaptive streaming that addresses inefficiencies across the entire pipeline. First, a Video Complexity Analyzer (VCA) provides lightweight content features to guide efficient encoding. Second, convex-hull–based optimization enables Pareto-efficient bitrate ladder construction. Third, multi-codec bitrate ladder pruning (MCBE) reduces redundancy across AVC, HEVC, AV1, and VVC, achieving major savings in storage and transmission. Fourth, ML-driven encoding parameter selection balances quality and energy by predicting VMAF and power usage. Finally, a Relative Decoding Energy Index (RDEI) provides device-independent modeling of decoding costs, enabling client-side adaptation.
Together, these methods reduce encoding energy by 46%, storage by 95%, transmission by 78%, and decoding by 15%, all while maintaining QoE within just-noticeable-difference thresholds. The framework paves the way toward scalable, sustainable, and net-zero streaming ecosystems.