Adaptive Quality and Energy Enhancement in Video Streaming with RecABR
Daniele Lorenzi, Farzad Tashtarian, Christian Timmerer · 2025
HTTP Adaptive Streaming (HAS) is the dominant approach for delivering video content on the Internet. However, the increasing environmental concerns call for more sustainable methods to reduce HAS's carbon footprint [1, 2]. Currently, adaptive bitrate (ABR) algorithms in HAS primarily focus on maximizing video quality, often overlooking the energy costs associated with higher bitrate selections [3, 4]. The adoption of deep learning techniques, like super-resolution (SR), for video quality enhancement can significantly improve quality but also increase energy consumption for devices already expending substantial energy on streaming tasks [5]. Figure 1 illustrates this trend for both laptop and smartphone, using representations from Ancient Thought [6] encoded according to [7]. The bitrates range from the lowest (r1) to the highest (r5-r9), capped by device display resolutions (1080p-2160p), as upscaling beyond the native display resolution is both energy-intensive and impractical. SR techniques are particularly beneficial in competitive network scenarios where multiple devices share a common link and can end up over- or underestimating the throughput. When this happens, devices equipped with a graphics processing unit (GPU) can apply SR to enhance visual quality of low-bitrate content in real-time, unlike those with only central processing units (CPUs). These enhancements are most effective when clients are aware of quality gains, allowing them to apply SR selectively when the quality boost justifies the energy cost. It is worth noting that for video on demand (VoD) content, quality and energy can be precomputed prior to streaming.