Adaptive bitrate allows you to use a video call even when the network quality is poor by reducing the amount of data transmitted over the network. Smoother video and less frames, but the conversation still goes on.
This guide will cover how the system detects trouble, what it changes first and where it fails. It also includes the steps to take if the connection is just too weak.
What Is Adaptive Bitrate in Video Calling?
Adaptive bitrate is a live feedback loop that allows your video data to be adjusted according to your current internet speed. The sender constantly adapts the quality of the video connection, rather than locking it at a single setting.
Quick terms to keep in mind:
- Bitrate: How much video information is transmitted per second.
- Adjustment levers: Resolution (sharpness), frame rate (smoothness), and compression (how tightly data is packed).
How Does Adaptive Bitrate Work?
The sender dynamically adjusts the target bitrate and then repeatedly verifies it according to the feedback from the connection.
When the internet speed drops: The encoder reduces the target bitrate and thus the call will continue to move, but at a lower resolution, a lower frame rate or less compression detail.
When the internet speed improves: The encoder gradually increases the target, as a rapid increase may cause internet speeds to drop again.
Adaptive Bitrate in Video Calls vs. Video Streaming
Both use adaptive bitrate ideas, but live calls and video streaming work differently:
Video streaming (HLS and DASH): Video is split into chunks and buffered ahead of time. A delay of 2 seconds or more is fine when you’re just watching. But in a live call, it makes talking back and forth harder.
Video calling (WebRTC): Live video is manipulated frame by frame with minimum delay to preserve natural and real-time video conversations.
Why low latency matters?
Audio and video delay of 150 to 200 milliseconds or more are too much for two-way conversations. People interrupt, not nod, and react immediately, so a video calling app can’t take a 3-second breather for safety buffer. The goal of adaptive bitrate algorithms in WebRTC is to have a sub-second latency period, compressing at every 50 milliseconds so the conversation remains live.
How Video Calling Apps Detect Network Problems
Video applications monitor network quality by measuring the actual time packets are received over the network and the packet loss ratio (PLR) over the network. Routers begin to queue data before a network drops packets altogether, so there are measurable delays in packet arrival.
Measuring network conditions
There are three key indicators:
- Packet arrival times: If the intervals are even, then the arrival time should be even. If the gaps are getting bigger, there is a queue forming somewhere on the path.
- Packet loss: Packets missing indicate that a router has dropped data or a wireless connection is dropping.
- Network delay: Round-trip time shows how long a signal takes to go out and return.
How bandwidth estimation works
Bandwidth estimation is how a call figures out the most data it can safely send without overloading the network. The sender keeps adjusting that maximum based on feedback from the receiver. Today’s WebRTC browsers do this using Google Congestion Control (GCC), which is a combination of a delay-based controller and a loss-based controller.
They both calculate a safe maximum bitrate every few milliseconds and the sender chooses the minimum of the two values, so as not to overload the router.
What is Transport-Wide Congestion Control (TWCC)?
Transport-Wide Congestion Control (TWCC) is a way for the receiver to tell the sender how well its data is arriving. The sender numbers every packet it sends, and the receiver reports back exactly when each one arrived. From this, the sender can tell whether the network is slowing down or getting overloaded.
TWCC differs from more traditional methods like Receiver Estimated Maximum Bitrate (REMB). It’s the receiver that computes bandwidth per media track by shifting these calculations from the receiver to the sender’s side.
Combining audio and video packets in the same sequence system provides the sender with more rapid and precise congestion detection.
How Video Calls Adjust When the Connection Gets Weak
If the bandwidth is reduced, then the calling apps preserve the spoken audio first and the video data second. Engineers believe in audio as a requirement for communication and video as an add-on feature.
Why Video Quality Drops Before Audio Quality?
Video is the first to suffer as it is the most data-hungry of all media types and people can still follow a conversation with bad video. It is impossible to follow with faulty sound.
The capacity is released immediately when decreasing video bitrate. Normally, a weak connection is used to send a call containing only voice. VoIP audio's bandwidth requirement is even smaller than video—Zoom's own documentation states it requires 60-80 kbps.
Audio codecs can also adjust. In WebRTC, the default option is Opus, which can be used from approximately 6 kbps to 510 kbps. Optimizes bitrate and bandwidth based on conditions, ensuring speech is still easily understood even at low levels.
In fact, it's a common practice to prioritise audio. Audio issues are easily noticeable and wearisome. A garbled speech takes effort for listeners to understand, but it doesn't take much effort for a pixelated face.
How Video Calls Balance Resolution and Frame Rate?
Video encoders take the video data usage down by lowering the video resolution, lowering the frame rate, or both, depending on the video content. Even video track parameters like motion versus detail are used by WebRTC tracks to determine what to "drop" first.
Lower resolution: While going from 720p to 360p, the resolution drops by 75 percent and motion is maintained. This is the best choice for a talking-head call, when the movement is more important than fine text.
Lower frame rate: It reduces the frame rate from 30 fps to 15 fps, which decreases data rate without compromising image sharpness. When sharing screens with apps, they take this route, and spreadsheets and documents remain readable.
Balance both: When the network drops, the encoder reduces the resolution and frame rate at the same time so as to keep video stream alive at low bitrates.
The W3C content hint specification provides an app with a way to mark a video track as “motion” or “detail”. The browser will then know what to protect. A slide deck marked as detail keeps its sharpness, while a camera feed marked as motion keeps its fluidity.
How Much Bandwidth Does a Video Call Need?
The basic one-to-one video call requires approximately 600kbps and a group video call (gallery view) requires up to 4Mbps. Bandwidth is determined by whether you are sending or receiving video, the number of participants tiles visible on screen and the hardware encoding efficiency.
| Call Type and Quality Level | Typical Upload Bandwidth | Typical Download Bandwidth |
|---|---|---|
| One-to-one video call (Standard HD) | 600 kbps | 600 kbps |
| Group video call (Standard) | 800 kbps – 1.0 Mbps | 1.0 Mbps |
| 720p HD video call | 1.2 – 2.6 Mbps | 1.2 – 1.8 Mbps |
| 1080p HD video call | 1.8 – 3.8 Mbps | 1.8 – 3.0 Mbps |
| Gallery view (25 to 49 video tiles) | 1.5 – 3.8 Mbps | 2.0 – 4.0 Mbps |
These statistics are subject to change depending on the device processor, background traffic and active screen sharing on the Zoom platform. VOIP calls that are only audio-based tend to use only 60 to 80 kbps.
How Simulcast Helps Group Video Calls Adapt to Different Networks?
Simulcast allows a single caller to broadcast several video quality streams simultaneously, enabling a central server to adjust video quality based on individual viewers' internet connection speeds. If not, an individual with a slow mobile connection would slow the video for all of the other people in the room.
The problem with sending one video quality level
When the sender only sends one 1080p stream, weak connections may not be able to download the data quickly enough and the data freezes frequently. On the other hand, if the sender reduces the stream to 240p for that one low-speed customer, then everyone on a high-speed internet connection will have blurry video.
How simulcast works
The sender encodes the same video in multiple quality levels (usually 3), and uploads each of these versions. Each layer is sent to a server, known as a selective forwarding unit (SFU). The SFU does not re-encode anything, hence delay is minimized, and server cost is minimized.
How the server selects video quality?
The SFU observes the viewer's relationship and transmits the layer that matches. The high layer is given to a participant who is on the fibre. The low one is received by a participant on a weak signal. The SFU switches a viewer to a lower layer if the viewers connection becomes worse in the middle of a call without notifying the sender.
Simulcast vs. scalable video coding (SVC)
In Scalable Video Coding (SVC), multiple spatial and temporal layers are packed into a single encoded stream, while simulcast sends multiple completely independent video streams.
SVC is based on a base layer of basic video content with enhancement layers that add additional frames and sharpness to the video. Simulcast requires more upload bandwidth than SVC, but it is more widely supported by older browsers and hardware encoders.
What Happens When Video Packets Are Lost?
Video apps also use targeted retransmissions, error-correction codes and a short jitter buffer to compensate for losses. Internet traffic is carried over the same paths and there will be some packet loss even where the internet connection is good.
Retransmission and error correction
There are 2 tools to take care of lost video data. A negative acknowledgement (NACK) is a request by the receiver asking the sender to re-send a particular packet that was missing.
Forward error correction (FEC) is different. The sender redundantly encodes the data at the start, which means the receiver can regenerate a missing packet without having to request it. FEC costs bandwidth all the time but avoids waiting.
How audio continues during packet loss
Audio implements a technique called packet-loss concealment, in which the decoder infers the missing sound from previous sounds. Opus also supports in-band FEC, which is a lower-quality copy of the previous frame in the next packet. If one goes missing, it may be replaced with the next one.
Why network delay makes recovery harder
A NACK requires both a request out and a packet back. On a long or slow path, the resent packet may arrive after its moment has passed. The decoder has already moved on and hence the data is not useful. That is the reason that high-delay connections rely on FEC, and not retransmission.
How jitter buffers keep audio and video smooth
Packets do not come in at regular intervals. They are temporarily stored in a jitter buffer and then output at a constant rate. The price of this is delayed. A larger buffer will reduce the variation more, but will make the call feel slower.
How to recognise poor call quality?
Look for these indicators:
- Packet loss above a few percent
- High jitter, resulting in choppy or robotic voice quality.
- Video frame rate that is significantly below the usual 24-30 frames per second.
- Rising round-trip time
Developers can read these through statistics APIs such as the one Stream documents. These are typically detected by users as freezing, blurring or loss of sound.
Why Adaptive Bitrate Cannot Fix Every Connection Problem?
Adaptive bitrate requires a minimum baseline of stable bandwidth and local processing power to function. If network or hardware problems exceed a certain level, the bitrate will not save the call.
- Extremely slow connections: Internet speeds of less than 100 kbps will render even highly compressed video useless, and only basic voice transport will be available.
- Severe packet loss: When a network loses 15-20% of packets, at random times, congestion control loops break and video freezes permanently.
- Limited device processing power: Encoding multiple simulcast layers or decoding high-resolution video heats up older laptops and phones, causing CPU throttling that drops frames regardless of internet speed.
- Poor Wi-Fi coverage: Sudden bandwidth spikes due to signal disturbance, thick walls and household members streaming 4K video outpaces ABR reaction time.
- Quality versus latency trade-off: Pushing bitrate down aggressively keeps latency low, but over-compression eventually turns faces and shared text into unreadable blurs.
Best Practices for Developers Building Video Calling Apps
First, create the network, then, build the features. The key question is: a call that remains connected on a weak signal is more important than any additional feature. You can write this layer yourself on WebRTC, or start from an SDK that already includes it, such as MirrorFly.
Using an SDK like MirrorFly: MirrorFly's video call SDK dynamically varies video stream quality based on the network condition using Adaptive Bitrate. If your users are on multiple devices, its SDKs include Android, iOS, Web, Flutter, and React Native. It also can be deployed in the cloud as well as on-premise, which is important if data needs to remain on your own infrastructure.
Before committing to any SDK, see how they deal with each practice below.
- Implement bandwidth estimation and congestion control. Use feedback from the receiver (e.g. TWCC) to estimate capacity, and reduce your send rate as delay begins to increase. This prevents queues building up before losing packets.
- Use adaptive video encoding and suitable bitrate limits. Give a lower and upper bound on each resolution. A ceiling prevents the encoder from overwhelming a shared link and a floor prevents the encoder from showing unviewable video.
- Support simulcast or SVC where appropriate. When using group calls with mixed networks, use simulcast, and use SVC if available on your platform. These are usually not required on a one-to-one call.
- Prioritise audio and manage video streams efficiently. Protect audio bitrate first, and pause or downscale video tiles that nobody is looking at. The bandwidth should not be consumed by offscreen participants.
- Monitor packet loss, jitter, round-trip time, and frame rate. Record these values on each call, rather than an average. The average conceals the bad short periods that users complain about.
- Test across different network conditions and device capabilities. Reduce the bandwidth and introduce packet loss on mobile phones and on older laptops. A quick office network conceals issues that users of mobile devices will discover in just minutes.
Conclusion
Adaptive bitrate maintains a video call by measuring delay and loss, estimating capacity, and reducing the video quality before audio quality is impacted. This is supported with tools like TWCC, simulcast, NACK and forward error correction but nothing can beat a stable connection.
The fixes are easy for users. Get closer to the router, close background applications and turn off the camera when not in use.
On the developer side, they should estimate the bandwidth, prioritize audio, and test at low bandwidth. You can build this layer on WebRTC yourself, or start from an SDK such as MirrorFly that includes adaptive bitrate. Either way, test on real devices with throttled networks before launch.

Top comments (0)