You moved a depth camera out onto an arm or a gantry, extended the cable to reach it, and now the stream dies every few minutes: incomplete frames, overflow lines in the viewer log, depth and IR streams that refuse to start until you unplug and replug. Nothing is wrong with the camera. A 5 m passive USB 3.0 run sits outside the signaling budget the interface was designed around. The fix comes in two halves: send fewer bytes across the marginal link, then rebuild the run so it meets the standard again. This page gives you the bandwidth arithmetic for a typical 1280 × 720 depth profile, the likely causes ranked with a test for each, and the cabling decision for short, marginal and long runs.
What the overflow errors are telling you
The failure pattern is distinctive. The camera enumerates, streams start, and after minutes of sustained capture the viewer logs incomplete frames and overflow errors on the USB receive path. Stream start failures follow, and only a physical replug brings the camera back.
That pattern means link errors, not a slow algorithm. USB 3.0 moves data in packets at up to 5 Gbit/s. When the electrical margin is gone, packets arrive corrupted, the host discards them, buffers overflow, and the driver gives up on the stream. A retry loop in your code cannot repair a link that is out of spec; it restarts the same marginal link.
Two details narrow the diagnosis. A camera that drops out the moment the IR stream starts points at power as well as signal, because the IR illuminator adds load on the same cable. A camera that fails on one PC and streams cleanly on another, over the same cable, points at the host controller rather than the camera.
The bandwidth budget: the arithmetic that decides it
Start from the payload, because every fix below is judged against it:
Halve the frame rate and every number halves with it: depth 27.6 MB/s, IR 13.8 MB/s, about 41 MB/s (332 Mbit/s) combined. Disable an IR stream the code never reads and the 30 fps load falls from 83 to 55 MB/s.
For scale, USB 3.0 runs at 5 Gbit/s with 8b/10b coding, so a healthy link moves about 500 MB/s of raw payload, and real bulk transfers land near 350-400 MB/s. An 83 MB/s depth profile uses roughly a fifth of that. Raw capacity was never the problem, which is why the failure arrives minutes into the run instead of at once. It is signal margin on a too-long passive run, consumed by errors under sustained load. The same bus check governs depth and color streams together — the depth and color stream timeout walkthrough shows the test.
Why 5 m of passive cable is out of spec
SuperSpeed signaling is tuned tightly enough that a passive cable has a practical limit of about 3 m (roughly 9.8 ft). Push past it and the documented result is what the log shows: signal degradation and reduced transfer speeds. A run assembled as a 3 m lead + a 2 m extension + the camera's own stub cable is worse still. Every added connector is an impedance discontinuity that reflects part of the signal, and four of them in series spend the margin entirely.
Charging-grade cables fail the same way. They are built to carry a few amps at 5 V, not 5 Gbit/s across twisted, individually shielded pairs, so a cable that charges a phone perfectly can still corrupt depth frames all day. That is also why an extension that looks fine in light use fails only at full stream rate — the pattern behind an extension cable that fails only when the depth stream records.
Likely causes, ranked, with a test for each
Work down the table; the first two rows cover most of these cases.
| Rank | Likely cause | Test | Fix |
|---|---|---|---|
| 1 | Passive run longer than about 3 m | Soak the exact stream profile on one short certified cable for 30 min, then repeat on the full run | One active extension spanning the run, or shorten the run |
| 2 | Daisy-chained or charging-grade cable | Count the connectors; swap in one certified data cable of the same total length | A single continuous, data-rated cable |
| 3 | Port power marginal; the drop coincides with the IR stream starting | Move the camera to a powered USB 3.0 hub and repeat the soak | Powered hub, camera alone on its port group |
| 4 | Marginal host controller: same cable fails on one PC, works on another | Run the identical cable on a second PC | A different port or controller, powered hub in front |
| 5 | Streams enabled that the code never consumes | List enabled streams, resolutions and fps against what the algorithm reads | Disable unused streams, drop fps to 15 |
If row 4 matches, the fix lives on the host side. The bandwidth and power budget for several USB cameras on one host walks the controller math, and the dock and hub path behind a black window covers the hub chain.
Fix 1: cut the data rate the link must carry
Do this first; it costs nothing. In the camera SDK's stream configuration, disable every stream, then re-enable only the ones your code reads. Dropping an unused IR stream takes the load from 83 to 55 MB/s. Setting 15 fps instead of 30 brings the payload to about 41 MB/s, and the error rate falls faster still, because each second now carries half the packets and half the chances to corrupt one.
Expect the overflow count to fall sharply, not necessarily to zero, on a marginal passive run. Treat the lower frame rate as mitigation and rebuild the cable for the cure.
Check the motion budget before you commit. At 15 fps the frame period stretches from 33.3 to 66.7 ms, so a part crossing the field at 500 mm/s moves 500 mm/s × 0.0667 s ≈ 33 mm between frames, against about 17 mm at 30 fps. If the algorithm needs tighter inter-frame motion, rebuild the cable and keep 30 fps. Exposure interacts with the ceiling too — the depth camera frame-rate ceiling walkthrough covers that side.
Fix 2: rebuild the run to spec
One certified cable, up to 3 m
If the arm stroke allows it, shorten the run. A single data-rated USB 3.0 cable at or under 3 m is fully in spec and needs nothing else in the chain. Where the cable flexes every cycle, insist on drag-chain (high-flex) construction and strain relief at the camera connector.
For that duty, put one certified high-flex USB 3.0 cable on the arm:
One active extension for 5 m and beyond
An active USB 3.0 extension carries repeater electronics that regenerate the signal mid-run. Commercial units reach 25-32 ft (7.6-9.8 m), so the full run length is comfortably covered. Keep the chain to two cables: the camera's short lead into the active extension, and the active extension straight into the host.
A powered hub when power or the controller is suspect
A USB 3.0 hub with its own supply regenerates the signal and feeds the camera from its own budget, taking the host port's 900 mA at 5 V out of the equation. Feed the hub with a short certified cable and leave its other ports empty while you test. A hub addresses rows 3 and 4 of the table above; it does not rescue a daisy-chained passive run, so clean the cabling first. The power arithmetic behind row 3 is in why a depth camera disconnects when the IR stream starts.
When the run is fixed and long: change the interface
The fight is winnable at this length. Past about 10 m, stop fighting and change interface. If the station is permanent and the camera choice is still open, pick GigE. 1000BASE-T is specified to 100 m on one cable, and an 83 MB/s depth profile fits inside GigE's usable payload of roughly 118 MB/s with about 30% headroom.
The interface choices live in the shop's 3D sensors range; check the camera finder before you re-cable. Camera-side cabling for one continuous long run:
- Type
- Camera data cables
- Length
- 15 m
The LS-CBL-0199 is a 3 m high-flex GigE cable with screw locks for the camera side. The LS-CBL-0117 is a 15 m RJ45 camera data cable with a locking stud, the shape a fixed run wants.
One boundary case worth knowing: if the run silently falls back to USB 2.0, no stream configuration will save it, because a depth camera on USB 2.0 cannot start its streams at all.
The decision, case by case
- Camera on an arm, 5 m fixed, hardware stays: disable unused streams, set 15 fps, run one active USB 3.0 extension spanning the whole run, and add a powered hub at the host if disconnects persist. Return to 30 fps only after a full-shift soak proves the link clean.
- Run can be kept to 3 m or less: one certified high-flex cable, full resolution, 30 fps, no hub.
- Specifying a fixed station beyond that now: a GigE camera and one long RJ45 run; keep USB 3.0 for short and mobile mounts.
- Errors persist after the cabling is clean: suspect the host controller. Test a second PC before condemning the camera, and update the camera firmware with the manufacturer's own updater first — a two-minute check that rules out a known error class.
Bench checklist before it goes on the arm
- Measure the true run length, including slack for the arm's full stroke; the cable must suit the worst pose, not the bench pose.
- Choose one continuous cable: passive for the short case, active for longer. Count the connectors; the target is the two ends only.
- Give the camera a powered hub or a port group of its own, and share it with nothing during the soak test.
- Soak the exact production profile — same streams, same resolution, same fps — for 30-60 minutes and watch the viewer log for overflow lines.
- Strain-relieve at the camera connector, and use drag-chain-rated cable anywhere the run flexes every cycle.
- Log re-enumerations for the first day: a camera dropping off the bus points at power, while overflow lines point at signal.
To pick cable, hub and mount in one pass, start from the camera cables page and the cable finder, or let the configurator walk the camera-to-host chain in order:
The configurator matches a camera, lens, light, light controller and cables to each other, step by step, and lists only parts that fit what you already picked.
The 3D topic page collects the related guides, from USB-C cable length and data rate limits to streaming depth data over a network.
Questions
How long can a USB 3.0 camera cable be?
Plan on about 3 m for a passive cable carrying SuperSpeed data; past that, signal degradation and reduced transfer speeds are the documented result. With one active USB 3.0 extension in the run, distances of 5-10 m are routine, because repeater electronics regenerate the signal mid-cable. For fixed runs beyond about 10 m, change the camera interface: GigE over Cat 5e or better is specified to 100 m on a single cable.
Why does my depth camera work at 3 m but fail at 5 m?
The longer passive run leaves almost no signal-integrity margin, so link errors accumulate once streaming is sustained. Prove it: soak the identical stream profile on a short certified cable at the bench, then repeat on the longer run. If the short one stays clean and the long one overflows within minutes, the run is the cause — replace it with one active extension rather than joined passive cables.
Does lowering the frame rate fix USB overflow errors?
It usually reduces them sharply but rarely cures the root cause. At 1280 × 720, halving 30 fps to 15 fps takes the depth stream from about 55 MB/s to 28 MB/s and the total payload from 83 MB/s to about 41 MB/s, which a marginal link carries far more reliably. Treat the lower fps as mitigation and rebuild the cabling for the permanent fix.
Is a powered USB hub a good fix for a long camera cable?
Yes, when port power or a marginal host controller shares the blame. A USB 3.0 hub with its own supply regenerates the data signal and feeds the camera from its own budget instead of the host port's 900 mA. Put it close to the host, feed it with a short certified cable, and leave its other ports empty during the test. It will not rescue a daisy-chained passive run on its own.
Can I join two USB 3.0 cables to make a 5 m run?
Avoid it. Every extra connector adds an impedance discontinuity and a reflection point, and joined cables behave far worse than one continuous cable of the same length. If the longer run is unavoidable, buy a single active USB 3.0 extension spanning the whole distance, or shorten the run so that one certified passive cable covers it with margin.
When should I switch the camera to GigE instead of fighting USB 3.0?
Switch when the mount is permanent, the run is fixed and long, and the camera choice is still open. GigE carries the same depth profile comfortably on a single RJ45 cable, with standard machine-vision cables available in 15 m lengths and no practical length ceiling anywhere near a few meters. Keep USB 3.0 for short and mobile mounts where its simplicity wins.


