The short answer for a 5–10 Gb/s link through a moving joint
For a camera link that bends, twists or rolls continuously inside a neck, wrist or torso joint, shielded micro coax is the safer default at 5–10 Gb/s: per-signal shielding, tighter bend radii and longer reachable runs. Choose a controlled-impedance FPC when the run is short, folds once and must stay flat. Neither choice survives on the cable family name alone — termination and shield grounding decide the outcome.
That answer flips under one condition: the moment the joint actually rotates, the question stops being "coax or FPC" and becomes "which termination, which strand or flex construction, and which acceptance test". Two harnesses built from the same cable can pass or fail the same 100k-cycle joint depending on how the shield lands at the connector shell and how the assembly reacts at its fan-out.
| Routing criterion | Micro coax | FPC / FPC-based harness |
|---|---|---|
| Space inside a joint cavity | Round bundle; needs a service loop and fan-out volume at each end | Flat and thin; wins when the gap is a slot rather than a tube |
| Continuous motion duty | Designed case — constant-motion kits are explicitly positioned for gimbals and robotic arms | Acceptable for static or once-folded routing; weakest at fold lines |
| EMI inside a motor-driven joint | Shield per signal by construction; result depends on where the shield terminates | Needs a dedicated ground layer stack to approach comparable crosstalk behaviour |
Everything quantitative further down is either a supplier statement without a published test method, or absent. No cycle count, no shield-effectiveness figure and no bend-radius multiplier is asserted in this article as a measured result.
Turn your joint spec into three acceptance numbers before comparing cables
Most vendor pages in this niche stop at "coax is better for motion" and leave the engineer to guess. That is a sales position, not a selection tool. The workable approach is to convert your joint drawing into three numbers first, put them on the drawing, and only then ask which cable family can be tested against them.
Metric 1 — electrical
Write down lane count, per-lane data rate, differential impedance target per pair, and the maximum insertion loss you can accept at your actual length. The drawing has to name the impedance requirement and the test method, not only the pinout — a sourcing guide on micro-coax and FPC assembly testing makes exactly this point, noting that for AWG#40 micro coax a bad assumption in the test fixture can turn an inspected lot into a failure event at the buyer (flexipcb.com).
Metric 2 — mechanical
Bend radius, torsion angle, travel length and the cycle target over the robot's service life. Ask for the radius as an absolute figure plus the supplier's test basis — mandrel diameter, cycle rate, twist angle, pass/fail criterion. "High flex" is a marketing adjective; a test setup description is a specification.
Metric 3 — EMC
Shield type, where the shield terminates at each end, and the return path. Inside a dexterous hand, "high-dv/dt motor power and microvolt-level sensor returns share a space measured in millimeters", and the mitigation is the same physics used at larger scale: segregate, shield, and reference, with a defined termination for every shield (roboticscableassembly.com). Unresolved shield termination is a more frequent root cause of intermittent link errors than the cable construction itself.
If a supplier cannot describe the test rig behind a claimed flex life, treat the claim as unverified and ask for a sample you can cycle yourself.
Micro coax vs FPC vs FPC-based harness: where each one actually fails
| Attribute | Micro coax | FPC / FPC-based harness |
|---|---|---|
| Routing volume | Round, flexible bundle; needs fan-out space at both ends | Flat profile; best in slot-shaped cavities |
| Dynamic flex capability | Constant-motion applications are its stated design case | Moderate with a dynamic-rated design; fails at fold lines and crimp-to-flex transitions |
| Impedance control | Per-pair geometry set by the coax; sensitive to termination and fixture definition | Requires a controlled-impedance multi-layer stack with dedicated ground layers |
| Shield / EMI | Shield per signal; effectiveness decided at the termination | Needs dedicated ground planes to reduce crosstalk; no per-signal shield |
| Reachable length | Supplier-stated up to 100 cm camera-to-processor | Supplier-stated as technically limited to 20 cm |
| Termination style | Solder or crimp to fine coax; connector shell grounding | Connector-to-flex; ZIF or board-to-board mating |
| Field repairability | Poor — retermination at the joint is a bench job | Poor to moderate — replacement flex is cheap, access is not |
| Cost driver | Termination labour, fixture, test time | Tooling and flex layer count at low volume |
| Typical failure after cycling | Marginal impedance / return-loss shift, intermittent link errors | Visible conductor crack at the fold line |
The asymmetry buyers miss: the FPC failure is mechanical and usually visible during inspection. The coax failure is often an electrical drift that only shows up as link errors or BER excursions in the field — after the joint has been assembled and the harness is no longer accessible.
The FPCs sold for MIPI CSI-2 camera links are genuinely capable: one vendor's slim FPC uses a 4-layer construction with controlled impedance and dedicated ground layers in a cable 4 mm wide, offered at 10, 20, 30 and 40 cm with a 22-pin, 0.5 mm pitch connector (kurokesu.com). That same vendor positions its I-PEX CABLINE-UA II micro-coax kit for constant-motion cases such as gimbals and robotic arms. Read those two statements together and the split is clear: the FPC is a compact static interconnect, the coax is the motion part.
Where a cell is genuinely unknown for a specific part — cycle count, shield effectiveness — the honest entry is "not published; request test data", not an assumed number.
Length limits and signal integrity: what the published evidence does and does not say
The most-quoted ceilings in this category are supplier statements. One accessory supplier states that micro-coax camera-to-processor links can be supplied in custom lengths up to 100 cm, while FPC cable length is "technically limited to 20 cm", and adds that micro coax allows smaller bending radii (mipi-modules.com). No test method, cable AWG, data rate or measurement condition is published alongside those figures. Treat them as a starting envelope for a conversation, not as a validated limit for your link.
A 2025 TDCommons paper argues that micro coax supports higher signal speeds than FPC and enables extension of the MIPI C-PHY channel length (tdcommons.org). We flag it as a research lead: the full text was not available to us, so nothing from it is presented here as our own validation, and its specific results should be read in the original before you cite them in a design review.
The practical translation matters more than the ceiling. Once your routing exceeds the FPC envelope, the design question shifts from "coax or FPC" to "where does the connector or adapter sit". The same supplier notes that micro-coax-to-board adapters are used in prototypes and functional samples, and that in volume the adapter can be integrated into the design or replaced with an alternative connector (mipi-modules.com). That is a mechanical and cost decision, not a cable decision — and it usually belongs to whoever owns the joint envelope.
Also assume that a 10 cm build and a 60 cm build of "the same" assembly do not share a pass/fail result. Per the assembly test guide, even strip length and fixture definition can change pass/fail outcomes on a 100 mm micro-coax assembly (flexipcb.com). Quote and test at your real length.
Shield and ground: the part that decides whether the link survives motion
Coax carries a shield per signal by construction; an FPC does not, and needs a dedicated ground-layer stack to reduce crosstalk to a comparable level — that is what the 4-layer controlled-impedance slim FPC is doing (kurokesu.com). Neither arrangement survives a joint on its own if the shield has nowhere to land.
Two questions decide it:
- Where does the shield bond at the connector shell, and where at the board? A long pigtail drain is the most common rework source in joint harnesses; ask for the intended bond point in writing on the drawing.
- What is the return path during torsion? If the reference conductor's path changes as the joint rotates, the link sees a moving ground — and that shows up as jitter and intermittent errors, not as a continuity failure.
Keep motor power and camera returns on physically separate routing paths through the joint wherever the envelope permits, shield the sensitive runs, and give every shield a defined termination (roboticscableassembly.com). Then specify the shield termination on the drawing, because that is the only part of this that a supplier can be held to.
Don't buy a cable spec — buy a validation plan you can accept
Four gates, in order, before you release the drawing.
- Prototype gate. Build 5–20 samples and confirm connector fit, strip length, soldering or crimping method, fixture approach and basic electrical results. Capture technician handling notes during these builds — micro coax and thin FPC tails are easy to damage in manual trial builds (flexipcb.com).
- Electrical acceptance. Differential impedance per pair with a named test method, insertion loss and return loss at your top frequency, plus continuity and isolation on every unit. The drawing names the impedance requirement and the test method — not just the pinout.
- Mechanical acceptance. Bend and torsion cycles at your joint's real radius and twist angle, with a stated pass/fail criterion and the number of units tested. Test the harness with the connector and strain relief as shipped, not as a bare cable sample.
- Workmanship. Agree the IPC/WHMA-A-620 class per assembly and require evidence with the lot. Class 3 is not automatic for every joint — a humanoid hand expected to run for years is the textbook Class 3 case, while a simpler mechanism may be fine at a lower class, provided the class is specified and paired with electrical test on every production lot (roboticscableassembly.com).
Stated honestly: this article asserts no shield-effectiveness figure and no cycle count for these assemblies, because method-documented published data for these specific products was not available to us. What replaces it is a test plan you own — which is more defensible in a design review than a borrowed number.
What to put in the RFQ (and what to send with it)
Send a data set, not a description. The assembly drawing, BOM, connector series, cable type and AWG, length, pinout, shield rule, impedance target, test method, quantity, environment and target lead time are the minimum that prevents a re-quote cycle (flexipcb.com).
- Motion envelope sketch — bend radius, torsion angle, travel and cycle target, so the supplier can quote a strand or flex construction instead of a catalogue part.
- Shield termination note — the intended bond point at both ends, and whether the connector shell is part of the ground path.
- Open decisions declared — adapter versus direct connector, service-loop length, repair strategy. Declaring them lets the quote price options rather than assume one.
- Test and evidence requirements — which of the four gates above you require, and what evidence ships with the lot.
- Build volume and stage — prototype-only or heading to repeat production, because that changes fixture investment and lead time.
Then ask for hardware: a bend-and-torsion sample built to your joint's radius and twist angle. A sample you cycled yourself settles in an afternoon what a datasheet argument will not settle in a month.
Request a joint-harness quote and a bend-test sample — send your drawing, motion envelope and cycle target, and specify micro coax or FPC as an open option if you want both priced against the same acceptance criteria. Samples can be built to your joint's bend radius and torsion angle so the validation runs on your geometry, not the supplier's default mandrel. If the joint needs multi-axis motion rather than a single bend axis, say so in the RFQ so the sample is designed for the combined motion from the start.
Frequently asked: the four things engineers ask before shortlisting
How many flex cycles can a micro coax robot camera cable survive in a humanoid joint?
No published, method-documented cycle count is available for the assemblies discussed here, so any figure you are quoted should come with the test setup that produced it. Specify your own cycle target and require bend-and-torsion results at your joint radius.
Is FPC ever the better choice than micro coax for a MIPI CSI-2 camera link?
Yes — for short, flat, static or once-folded runs where slot space is tighter than a round bundle allows. A 4-layer controlled-impedance FPC with dedicated ground layers handles short CSI-2 runs inside constrained cavities well; it is the repeated-motion pass-through that favours coax.
What cable length can I run at 10 Gb/s from camera to processor board — 20 cm with FPC, 100 cm with micro coax?
Those are supplier statements, not validated limits: one supplier cites up to 100 cm for micro coax and a technical limit of 20 cm for FPC, without publishing a test method (mipi-modules.com). Your achievable length depends on data rate, AWG, impedance control and termination — validate at your real length, since even strip length and fixture definition can shift pass/fail.
How do I specify shield termination and grounding for a camera harness that passes through a motor-driven joint?
Name the bond point at the connector shell and at the board, define the return path, and keep motor power separated from camera returns inside the joint. Then require the shield rule on the drawing and verify it at prototype build, because an undefined termination is the usual source of intermittent link errors after cycling.