Adaptive bitrate streaming (ABR) is a method of delivering video over the internet in which the same content is encoded at several bitrates and resolutions, split into short segments, and the player chooses the most suitable version for each segment based on current network throughput and device capability. In streaming & broadcasting, it is the technique behind almost every on-demand and large-scale live video service.
ABR exists because internet bandwidth is unpredictable and shared. A viewer on mobile data, home Wi-Fi or fiber may see throughput change from second to second. Instead of forcing one fixed quality, ABR trades a little quality for continuous playback without stalls, and raises quality again when conditions improve.
How adaptive bitrate streaming works
ABR combines server-side preparation and client-side decisions. The server, or more often a CDN, holds ready-made segments at each quality level. The player downloads a manifest, measures how quickly segments arrive, watches its buffer and requests the next segment from the appropriate level.
- Encoding the ladder: the source is encoded into several renditions, for example from 240p to 2160p.
- Segmenting and packaging: each rendition is cut into segments of typically 2 to 6 seconds, aligned at keyframes.
- Publishing a manifest: a playlist or MPD file lists the renditions, their bitrates, resolutions and codecs.
- Measuring and switching: the player estimates throughput and buffer level and picks the next rendition.
- Seamless playback: because segments are aligned, the decoder switches renditions at segment boundaries without visible gaps.
Segments are delivered over ordinary HTTP, which means standard web servers and CDN caches can distribute them at massive scale. This is a key advantage over older streaming protocols that required dedicated media servers and persistent connections.
The bitrate ladder
The bitrate ladder is the set of renditions the player can choose from. Its design decides how smoothly quality adapts and how much storage and encoding time the service needs.
| Rendition | Resolution | Video bitrate | Typical target |
|---|---|---|---|
| Lowest | 426 x 240 | 300 to 400 kbps | Weak mobile connections |
| Low | 640 x 360 | 600 to 800 kbps | Mobile, small screens |
| Medium | 854 x 480 | 1 to 1.5 Mbps | Average mobile, tablets |
| High | 1280 x 720 | 2.5 to 4 Mbps | Laptops, good Wi-Fi |
| Full HD | 1920 x 1080 | 4.5 to 8 Mbps | Televisions, desktops |
| Ultra HD | 3840 x 2160 | 12 to 20 Mbps (HEVC or AV1) | 4K televisions |
Adjacent rungs are usually spaced so that each step is about 1.5 to 2 times the bitrate of the step below. Steps that are too wide cause visible quality jumps; steps that are too narrow waste encoding and storage. Many services use per-title or per-scene encoding, in which the ladder is tuned to the complexity of each title: a simple animation may need far less bitrate at 1080p than a grainy action film.
Protocols and formats for adaptive delivery
ABR is implemented by a small number of HTTP-based streaming protocols. They differ mainly in manifest format, ecosystem and latency options, while sharing the same core idea.
| Protocol | Manifest | Segment formats | Typical latency |
|---|---|---|---|
| HLS | M3U8 playlists | MPEG-TS or fragmented MP4 | 6 to 30 seconds |
| Low-Latency HLS | M3U8 with partial segments | Fragmented MP4 parts | 2 to 5 seconds |
| MPEG-DASH | XML MPD | Fragmented MP4 or WebM | 6 to 30 seconds |
| Low-latency DASH | MPD with chunked transfer | CMAF chunks | 2 to 5 seconds |
CMAF, the Common Media Application Format, allows HLS and DASH to share the same fragmented MP4 segments, so a service can package media once and serve both protocols. Ingest protocols such as RTMP and SRT are not ABR protocols: they carry a single high-quality stream from the encoder to the platform, which then creates the ladder.
Player adaptation algorithms
The player’s adaptation algorithm decides which rendition to request next. Its behavior shapes the viewer’s experience as much as the encoding itself.
| Approach | Main input | Strength | Weakness |
|---|---|---|---|
| Throughput-based | Measured download speed | Fast reaction to bandwidth | Unstable on noisy networks |
| Buffer-based | Seconds of video in the buffer | Few stalls, stable quality | Slow start at low quality |
| Hybrid | Throughput and buffer together | Balanced behavior | More tuning parameters |
| Learning-based | Observed sessions and models | Adapts to real networks | Complex to validate |
Good algorithms avoid oscillation, the constant switching between two levels that viewers perceive as flickering quality. They also start conservatively, ramp up after the buffer is secure and drop down early when the buffer falls toward a safety threshold of a few seconds.
Trade-offs and common mistakes
Every ABR configuration balances latency, quality, stability and cost. Shorter segments react faster and lower latency, but they add request overhead and reduce compression efficiency because each segment must start with a keyframe.
- Misaligned keyframes between renditions break seamless switching; every rendition needs the same GOP boundaries.
- Uncapped VBR renditions can exceed their declared bitrate and mislead the player, causing stalls.
- Too few low rungs leave mobile viewers without a fallback when coverage weakens.
- Oversized top rungs waste CDN traffic on small screens that cannot show the extra detail.
- Ignoring audio renditions and captions leads to manifests that work for video but fail accessibility or language requirements.
Quality monitoring should track startup time, rebuffering ratio, average delivered bitrate and switch frequency. These metrics show whether the ladder and the player logic work together for real audiences.
Understanding it simply
Adaptive bitrate streaming is why online video keeps playing when your internet slows down. The service prepares the same video in several quality versions, from small and blurry to large and sharp. Each version is cut into short pieces of a few seconds.
Your player checks how fast the pieces arrive. If the connection is fast, it fetches the sharp version. If the connection slows, it switches to a lighter version for the next pieces. You may notice the picture soften for a moment, but the video does not stop.
Applied in practice
A university streams its graduation ceremony live to families around the world. The production team sends a single 1080p stream at 8 Mbps over SRT from the venue to a cloud video platform.
The platform transcodes the feed into a five-rung ladder from 360p at 700 kbps to 1080p at 6 Mbps, packages it as CMAF and delivers it over HLS and DASH with 4-second segments. Relatives on fiber watch in full HD, while a grandparent on a rural mobile connection receives mostly 480p. Monitoring shows a rebuffering ratio below one percent across about 3,000 viewers, and nobody needs to choose a quality setting manually.