Latency is a chain, not a number: sensor readout, encoding, transmission, decoding and display each add delay. Analog links land near 20–40 ms glass-to-glass; digital HD links typically run 60–120 ms; IP and network paths can exceed 200 ms — and different missions break at different points.
- 20–40 msAnalog glass-to-glass
- 60–120 msDigital HD links
- 200+ msIP and network paths
- <40 msProximity FPV requirement
Key takeaways
- Latency accumulates across five stages. Optimising one while ignoring the others moves the bottleneck rather than removing it.
- Encoder settings dominate on digital links, network hops dominate on IP, and display processing is the cost everyone forgets to measure.
- Click-to-track is the quiet exception: tracking runs on the payload, so downlink latency barely affects the control loop.
- Measure glass-to-glass on your own chain. The datasheet number describes one stage, not the system you will fly.
On this page
Where the Milliseconds Go
Five stages sit between photons hitting the sensor and photons leaving your screen, and each contributes.
Sensor readout is usually small and fixed. Encoding is where digital links spend most of their budget, and it is adjustable: low-latency encoder modes trade compression efficiency for speed, which means more bandwidth for the same picture. That trade is the main lever an integrator actually controls.
Transmission varies enormously by link type and, on IP paths, by how many network hops sit in the way. Decoding mirrors encoding. And display processing is the sleeper cost everyone forgets to measure — a monitor or headset with its own processing pipeline can add tens of milliseconds that no amount of tuning upstream will recover.
The practical consequence is that a datasheet figure for any single stage tells you very little. Video encoding and streaming architecture covers H.264, H.265 and RTSP behaviour in depth, but the number that matters is the one you measure end to end.
Typical Figures by Link Type
Ranges rather than points, because configuration moves these substantially.
| Link type | Typical glass-to-glass | Dominant contributor | Suits |
|---|---|---|---|
| Analog AV | 20–40 ms | Almost none — minimal processing | Proximity FPV, racing, close-quarters flying |
| Digital HD (O3 / O4 class) | 60–120 ms | Encoder settings | Cinematic, inspection framing, general piloted work |
| IP / network (ONVIF, RTSP) | 200 ms and above | Network hops and buffering | Control-room monitoring, tethered overwatch, VMS integration |
The ordering is not a quality ranking. Analog wins latency by doing almost no processing, which is also why it cannot deliver an HD image. IP paths carry the most delay and in exchange integrate natively with recording, video walls and permissions — see ONVIF and IP video on drone payloads. Each is the correct answer to a different question.
Mission Budgets
What breaks, and at what point, depends entirely on whether a human control loop closes through the image.
Proximity FPV flying: under 40 ms. The pilot is steering by the picture, and delay directly becomes overshoot. Analog or racing-grade digital is the requirement here, and this is the one mission where the latency figure is genuinely a safety specification.
Manual inspection framing: 100 ms is comfortable. The operator is positioning a camera on a static structure, not threading a gap. The image quality gained from a digital link matters far more than the delay costs.
Click-to-track operation: downlink latency barely matters. Tracking runs on the payload, so the control loop closes on the aircraft between camera and gimbal, and the ground link only carries monitoring video. This is one of AI tracking’s quiet advantages and it is rarely stated in product material — see how AI tracking actually works.
Control-room monitoring via ONVIF: 200 ms and above is routine and fine. Nobody in a control room is flying anything. They are watching a scene and directing a response, and a fifth of a second is invisible in that task.
Decide by who is closing the loop. If a human is steering by the image, latency is a hard requirement. If a human is watching the image, it is a comfort setting. If a machine on the aircraft is closing the loop, the downlink figure is nearly irrelevant.
Designing the Chain
Three decisions cover most of the achievable improvement, and they are made in this order.
- Match the payload variant to the link. The AX-20 series ships in O3, O4, O4 Pro, Avatar/Moonlight and AV variants, and a mismatched unit does not connect at all — see the variant map.
- Enable low-latency encoder profiles for piloted work, accepting the bandwidth cost. Leave them off for recording-oriented missions where compression efficiency is worth more.
- Choose the display deliberately. A headset or monitor with heavy internal processing can undo everything gained upstream.
- Measure glass-to-glass on the assembled chain — point the camera at a running timer, photograph camera and screen together, and read the difference.
That last step is the one teams skip, and it is the only one that produces a number describing the system you actually fly. Matching a gimbal camera to your FPV link covers the variant question, and the integration checklist puts video alongside the mechanical and electrical checks it belongs with.
The Zoom Interaction Nobody Plans For
Latency does not change with zoom level in any meaningful way — the same processing chain runs regardless of focal length. What changes is how much that latency costs you.
At wide angle, the scene moves slowly across the frame and 100 ms of delay is imperceptible. At 40x, a small gimbal or aircraft movement sweeps the scene rapidly, and the same 100 ms means the image on screen is showing where the camera was pointed a noticeable moment ago. The operator over-corrects, the image swings, and the payload feels unresponsive.
This compounds with the stabilisation issue at magnification, where jitter rather than optics caps usable zoom — pointing accuracy at 1,000 metres and hybrid zoom explained cover that side. The practical guidance is to test manual zoom control at the top of the range before committing to a link, because that is where an acceptable latency budget stops being acceptable.
Related reading
- Video Encoding and Streaming Architecture: H.264, H.265, RTSP and SEI
- Multi-Sensor Time Synchronization
- ONVIF and IP Video on Drone Payloads
- Matching a Gimbal Camera to Your FPV Link
- AI Target Tracking on UAV Payloads
- What ±0.01° Gimbal Accuracy Means at 1,000 Metres
- UAV Payload Integration Checklist
- Head-Tracking FPV Gimbals
- AX-20T — 3-axis gimbal in five link variants
- OP-80P — 4K pod with ONVIF output
- OP-90P — multi-sensor pod with AI tracking
- Event and Crowd Safety Drones
FAQ
Does zoom level change latency?
Not meaningfully — the same processing chain runs at every focal length. What changes is how much the delay costs you. Latency you comfortably tolerated at wide angle becomes visible at 40x because the scene moves across the frame much faster, so the picture on screen lags perceptibly behind where the camera is now pointed. Test manual zoom control at the top of the range before committing to a link.
Why does onboard tracking reduce latency sensitivity?
Because the control loop closes on the aircraft, between the camera and the gimbal, rather than through the ground link. The payload sees the target move and corrects immediately; the downlink only carries monitoring video for a human to watch. This means a tracking mission can run happily on a link whose latency would be unacceptable for manual camera control — a genuine and under-advertised advantage of payload-side tracking over ground-application tracking.
How do I measure glass-to-glass latency myself?
Point the camera at a running millisecond timer, then photograph the timer and your display in the same frame. The difference between the two readings is your true end-to-end latency, including the display processing that datasheets never mention. Do it on the assembled system you intend to fly, with the actual monitor or headset — measuring a component tells you about a stage, and only the full chain tells you about the mission.
Questions about the technology? Talk to our engineers — we reply within 2 business days.