Adaptive Bitrate (ABR) Streaming: Concepts, Protocols, Challenges, and Algorithms

Streaming media delivers multimedia over the network in real‑time, allowing users to watch video or listen to audio while the data is being transferred. Unlike file‑download models, streaming sends media as a continuous packet flow, enabling on‑demand video, live broadcast, short‑form video, and online education on mobile devices.

Because streaming relies heavily on network bandwidth, poor network conditions can cause stalling or playback failure. To mitigate this, providers offer multiple representations of the same content at different resolutions and bitrates, ranging from low‑resolution (e.g., 512×288) to high‑resolution (e.g., 1920×1080), matching bandwidths from a few hundred kbps to several Mbps.

The adaptive bitrate (ABR) technique was introduced to address the trade‑off between maintaining smooth playback and preserving visual quality. ABR aims to maximize average bitrate while minimizing re‑buffering events and bitrate switches, thereby improving the overall Quality of Experience (QoE) measured by video quality, buffering duration, and playback smoothness.

Common streaming protocols include proprietary solutions such as Adobe RTMP, Apple HLS, and Microsoft Smooth Streaming, as well as standardized formats like MPEG‑DASH (Dynamic Adaptive Streaming over HTTP) and CMAF (Common Media Application Format). The industry is converging toward CMAF‑based HLS and DASH to reduce protocol fragmentation.

ABR streaming parameters typically involve segment duration, number of segments, bitrate levels, network throughput, and buffer occupancy. These parameters feed into the decision‑making logic that selects the optimal representation for the next segment.

ABR faces three major challenges: (1) unpredictable network bandwidth, especially on mobile or wireless links; (2) conflicting QoE objectives (e.g., high quality vs. low re‑buffering); and (3) the cascading effect of bitrate decisions, where early choices heavily influence later performance.

Several bandwidth‑prediction methods are employed in ABR systems, including:

Segment‑based Last Bandwidth (SLBW): uses the download speed of the most recent segment as the prediction.

Chunk‑based Sliding Window Moving Average (SWMA): computes an average over a fixed‑size window of recent chunks.

Chunk‑based Exponentially Weighted Moving Average (EWMA): applies exponential decay to give more weight to recent measurements.

Harmonic Mean: reduces the impact of outliers by averaging reciprocals of observed bandwidths.

ABR algorithms can be categorized as:

Buffer‑based algorithms (e.g., BBA, BOLA) that adjust bitrate according to buffer occupancy.

Rate‑based algorithms that estimate available throughput and select the highest sustainable bitrate.

Hybrid algorithms that combine buffer and throughput information.

Machine‑learning‑based approaches (e.g., Pensieve) that use reinforcement learning to optimize QoE.

Standardized implementations such as Microsoft Smooth Streaming (MSS), Apple HLS, and MPEG‑DASH provide reference clients (e.g., dash.js) and define protocol components, enabling interoperable adaptive streaming across devices.

In summary, Adaptive Bitrate Streaming dynamically adapts video quality to network conditions, reducing buffering and improving user experience. It relies on a suite of protocols, prediction algorithms, and bitrate‑selection strategies that continue to evolve with advances in networking and machine learning.