Framework Adapts Video Quality and Chunk Duration to Reduce Streaming Requests

Standard HTTP Adaptive Streaming platforms typically adjust video transmission bitrates while relying on a fixed chunk duration. Under this conventional approach, shorter video chunks facilitate faster adaptation during network fluctuations but generate a higher volume of HTTP requests, whereas longer video chunks decrease overall request frequency while slowing the system’s ability to respond to changing network conditions.

To address this operational trade-off, researchers Yeganeh Chatri and Behzad Akbari developed Safety-Projected Factorized ACTBR. The proposed application-layer framework introduces dynamic control over both video quality selection and logical chunk duration, allowing streaming clients to adjust how much content is requested in individual transmissions based on prevailing delivery constraints.

Under the ACTBR architecture, video content is segmented into aligned two-second Dynamic Adaptive Streaming over HTTP media units.

The underlying control mechanism relies on a factorized reinforcement learning design. A source bitrate-only Deep Deterministic Policy Gradient controller is initially trained using fixed two-second requests, after which its quality actor network is transferred and frozen while a dedicated duration actor is trained to manage logical segment sizing independently.

To prevent performance degradation during streaming sessions, the framework integrates an active safety mechanism. A conservative safety projection actively limits unsafe chunk-duration selections before each network request is executed, bounding transmission risks before requests are dispatched across the connection.

Evaluation of the framework was conducted in a dedicated simulation environment utilizing exact variable-bit-rate media-unit sizes alongside measured mobile bandwidth traces. The experimental setup incorporated forty mobile bandwidth traces, three DASH video sequences, fixed training–validation–test data splits, and five independent training seeds to measure stability and performance across runs.

Compared to the baseline bitrate-only controller operating with fixed two-second requests, ACTBR reduced the total number of HTTP requests by 6.55 percent. This efficiency gain shifted the mean time-weighted nominal representation bitrate by only -0.0099 megabits per second, while incurring a 0.8528-point reduction in mean quality of experience and a 0.1341-second increase in mean rebuffering duration.

Comparative testing against fixed chunk durations revealed that four-, six-, and ten-second configurations reduced HTTP request counts more substantially but resulted in progressively larger quality of experience penalties. Further evaluations spanning variable-bit-rate and constant-bit-rate environments, shared-bottleneck scenarios, and runtime profiles demonstrated consistent trade-offs and confirmed sub-millisecond central processing unit inference speeds.

Subject of Research: Adaptive bitrate and logical chunk-duration control for HTTP adaptive streaming

Article Title: ACTBR: safety-projected adaptive bitrate and logical chunk-duration control for HTTP adaptive streaming

Article References: Chatri, Y., & Akbari, B. (2026). ACTBR: safety-projected adaptive bitrate and logical chunk-duration control for HTTP adaptive streaming. Scientific Reports. https://doi.org/10.1038/s41598-026-77919-2

Image Credits: AI Generated

DOI: 10.1038/s41598-026-77919-2

Keywords: HTTP adaptive streaming, adaptive bitrate, logical chunk duration, video streaming, ACTBR, Dynamic Adaptive Streaming over HTTP, DASH, Deep Deterministic Policy Gradient, DDPG, quality of experience, rebuffering, mobile bandwidth traces, safety projection, variable bit rate, constant bit rate, sub-millisecond inference, network adaptation