A servo axis runs clean on the bench and throws encoder faults on the machine. Nothing changed except that the feedback cable now runs 4 metres alongside a motor power cable in a shared duct, and the drive is switching hard at the carrier frequency. The cable was ordered as "shielded". It is shielded. The shield is terminated as a 50 mm pigtail crimped into a spare connector pin.
That is the single most common failure in servo encoder feedback cable shielding construction, and it is worth stating plainly before anything else: how the shield is terminated matters more than what the shield is made of. You can buy a 95%-coverage tinned copper braid over individually foiled pairs and destroy its usefulness with 50 mm of pigtail. Everything below is written to help you put the construction — and the termination — onto a drawing that a harness shop can quote against, instead of a one-word requirement they will interpret in whichever way is cheapest.
Why the pigtail undoes the shield
A shield works by giving induced current somewhere to go other than your signal conductors, and by giving common-mode current a low-impedance path back to its source. "Low impedance" is the operative phrase, and impedance rises with frequency.
A pigtail — the shield gathered into a twisted tail and landed on a single pin — is a short thin conductor. At DC it measures near zero. At the frequencies that matter on a drive, its inductance dominates, and a few tens of nanohenries in series with your shield connection is enough to lift the shield's potential at the connector and re-radiate the very current you were trying to intercept. The rule of thumb machine builders use is that pigtail length should be measured in millimetres and treated as a defect above roughly 25 mm; the honest version is that any pigtail is a compromise and the length is how much you are compromising.
The alternative is a 360-degree termination: the braid is combed back over a ferrule or a conductive backshell and clamped so that the shield connects to the connector shell around its full circumference. The path is short, wide and low-inductance. Metal-shell circular connectors and metal-shell RJ45 housings are built for exactly this, and it is the reason motion-control drawings specify metal shells for feedback connectors even where a plastic shell would mate perfectly well.
Two further points that belong on the drawing rather than in a phone call:
- Which end the shield lands on. Grounding the shield at one end avoids a low-frequency ground loop; grounding at both ends is far better against high-frequency noise. The common motion-control convention is a solid 360-degree connection at the drive end, with the motor or encoder end handled per the drive maker's instruction — sometimes bonded, sometimes capacitively coupled, sometimes left floating. Do not leave this to the harness shop. It is a system decision, and it must appear on your drawing as an explicit statement per end.
- Whether the shield is bonded to a connector pin at all. If your drive expects the shield on the shell and your harness lands it on pin 9, the cable will pass every continuity test and still fail on the machine.
Pair twist, drain wires and the details that get substituted
Below the shield sit three construction details that buyers routinely leave unspecified and suppliers routinely decide for you.
Pairing has to match the drive's differential pairs
Incremental encoder feedback runs as differential pairs — A and A-inverted, B and B-inverted, Z and Z-inverted — and absolute encoders run clock and data pairs plus power. Differential signalling only rejects common-mode noise if the two conductors of a pair see the same interference, which requires them to be twisted as a pair for the full length.
Here is the trap. A cable can have exactly the right conductor count, exactly the right colours, and be twisted into pairs that do not correspond to the drive's signal pairs. Continuity passes. Colour-to-pin passes. Noise rejection is gone, because A and A-inverted are now in different twists. Specify the pairing explicitly on the wire list: pair number, the two colours in it, and which pin each lands on. Then require the first article to be checked against that pairing, not just against pin-out.
Twist lay length matters too — a shorter, tighter lay gives better rejection, and a good spec states either the cable's part number or a maximum lay length rather than the word "twisted".
Foil, braid, or both, and what the drain wire is for
Foil shields give near-100% coverage and work well at high frequency but are mechanically fragile and cannot carry much current. Braid gives lower coverage — typically quoted as a percentage, and coverage below the mid-80s starts to leak at high frequency — but is robust and terminates properly.
The common construction for encoder feedback is individually foiled pairs (so the pairs do not couple into each other) inside an overall braid (which handles the external field and gives you something mechanically sound to terminate). Each foiled pair gets a drain wire: an uninsulated conductor laid in continuous contact with the conductive side of the foil, because you cannot crimp or clamp aluminium-polyester foil directly. Two things follow that are easy to get wrong in assembly: the foil has a conductive side and a non-conductive side, so foil orientation is real and reversible; and the drain wires from separate pairs should not be casually twisted together and dumped on one pin unless your drive documentation says so.
Power and feedback do not belong in the same jacket
A hybrid cable that carries motor power and encoder feedback in one jacket is a real product, but it is a designed product with double shielding and specified separation, not something to improvise. If you are specifying a discrete feedback cable, say on the drawing that it is not to be combined with power conductors, and specify physical separation from motor cables in the installation note. It costs nothing to write and it removes the cheapest substitution available to a supplier.
Flex life: the spec that separates a cable from a drag-chain cable
If the cable moves — a cable carrier, a rotary axis, a gantry — the construction requirements change completely, and this is where a cheap substitution is most expensive.
A continuous-flex cable is built differently from a static one: finer strand counts with a short lay, bundle stranding rather than layer stranding so the conductors do not saw against each other, a filler core to hold geometry, a shield braid laid at an angle that tolerates bending, and a jacket that does not cold-set. Polyurethane jackets are the usual choice for cable carriers; standard PVC is not. Minimum bend radius for continuous flex is commonly specified in the range of roughly 7.5 to 10 times the outside diameter, and torsional applications have their own separate rating — a cable rated for flexing is not automatically rated for twisting.
Two assembly-side details matter as much as the cable itself, and are exactly the sort of thing to nail down before a sample run:
- The shield must be terminated without cutting the braid short or heating it. A soldered shield joint on a flex cable creates a stiff point that becomes the fatigue origin.
- Strain relief must clamp the jacket, not the conductors, and the transition from clamped to free has to be gradual. Most drag-chain feedback cable failures happen within 100 mm of a connector.
| Construction choice | What it buys you | When to specify it | Put this on the drawing |
|---|---|---|---|
| Overall braid only | Basic external field rejection, robust termination | Short runs, low-noise environments, signal cables away from drives | Braid material and coverage %, e.g. tinned copper, ≥85% |
| Foiled pairs + overall braid | Pair-to-pair isolation plus external rejection; the motion-control default | Encoder feedback, absolute-encoder clock/data, any run parallel to motor cables | Per-pair foil with drain wire, plus overall braid coverage % |
| 360° shell termination | Low-inductance shield path that actually works above a few MHz | Any feedback cable on a switching drive | "Shield terminated 360° to connector shell via ferrule/backshell; pigtails not accepted" |
| Pigtail termination | Cheap and fast | Low-frequency signals only, and even then reluctantly | If unavoidable, state a maximum pigtail length in mm |
| Shield bonded one end / both ends | Ground-loop control vs high-frequency performance | Decided by your drive documentation, not by the cable shop | Explicit statement per end: drive end, motor/encoder end |
| Continuous-flex construction | Survives a cable carrier for millions of cycles | Any moving axis | Flex or torsion rating, minimum bend radius as a multiple of OD, jacket material |
| Defined pair map | Real common-mode rejection | Every differential feedback cable | Pair number, colour pair, pin at each end — as a table, not prose |
What a supplier listing tells you, and where the questions start
Servo and feedback cabling is a normal line for a Shenzhen control-cable shop, and the public listings are usually a starting point rather than an answer. Huisheng Electronics is a fair worked example: a Longhua District, Shenzhen company incorporated in 2016, with company name, credit code, registration status and registered address checked against the Chinese corporate registry in August 2026, and a registered scope covering production plus import and export. Its capability listing names servo power and feedback cables, encoder lines and network-style signal cables for motion-control builds, alongside control-cabinet wiring for machinery, fans, power supplies and motor drives built to the buyer's drawing. Its product listing includes an RJ45 servo signal cable and a servo feedback cable with a breakout. The published equipment roster runs to more than twenty machines and includes withstand-voltage, insulation and cable-continuity testers on the outgoing-QC side. Stated terms are a 1,000-piece MOQ, samples dispatched in 7–10 days, and volume in 15–25 days to the warehouse.
Every item in that paragraph is a capability statement. None of it answers the construction questions this article is about, and no supplier listing of this type does. What braid coverage is available, whether foil-plus-braid stock is held, whether shields are terminated 360-degree in-house or pigtailed, whether the metal-shell RJ45 assemblies land the shield on the shell, and whether continuous-flex or torsion-rated cable is offered at all — those are open, and they are the five questions worth asking before you send a drawing. Continuity and insulation testers on a roster confirm the shop can prove a cable is correctly wired; they do not confirm noise performance, which no standard outgoing test measures. Treat any capacity or capability figure from a small supplier as supplier-stated until an audit or a live video walkthrough backs it up. Shenzhen has depth in this category — the wider picture is in the Shenzhen electronics manufacturing sourcing guide — but depth in a city is not evidence about one shop.
Common questions
Can I just send a competitor's cable and ask for a copy?
You can, and it is a reasonable starting point, but a physical sample under-specifies exactly the things that fail. Shield coverage, foil orientation, drain wire placement, lay length and flex rating are not visible without destroying the cable, and the termination method inside the backshell is not visible at all. Send the sample and a drawing that states the construction, the pair map and the termination requirement. The general discipline for turning an intent into a document a Chinese factory can quote is set out in the guide to writing a product spec sheet for a Chinese factory.
What should I test on the first samples?
Beyond continuity and pin-out, check the three things a supplier can substitute invisibly. Cut one sample open and confirm the shield construction and drain wires match the drawing. Ring out the pairing to confirm each differential pair is genuinely twisted together — a simple wire-map tester will not tell you this, so ask for the cable part number and check its datasheet. Then run one on the actual machine, next to the actual motor cable, at the actual carrier frequency. Bench-testing a feedback cable proves it is wired; only the machine proves it is quiet. Ordering that first batch properly is its own discipline, covered in the notes on the sample order process with a China factory.
Is a higher braid coverage percentage the better buy?
Higher coverage generally improves high-frequency performance, but it adds cost, stiffness and outside diameter, and on a moving axis a heavy braid can shorten flex life. The bigger gains for most machine builders come from fixing the termination and the routing first. A 85% braid terminated 360-degree at the drive end and routed away from the motor cable will out-perform a 95% braid on a pigtail in a shared duct, at lower cost.
What to ask the supplier next
Put these in the RFQ alongside the drawing, and treat the answers as part of the technical evaluation rather than the commercial one:
- What cable do you propose, by manufacturer part number? A construction described in words can be substituted; a part number cannot, quietly.
- Braid coverage and shield architecture — overall braid only, or individually foiled pairs with drain wires plus an overall braid? Confirm what is held in stock versus what needs to be bought in.
- How will the shield be terminated at each end? Ask for photographs of a completed 360-degree termination from a previous job, and state in the purchase order whether pigtails are acceptable and at what maximum length.
- Which end is the shield bonded at, and how is the other end handled? Confirm this matches your drive documentation.
- The pair map, confirmed back to you as a table: pair number, colours, pin at each end.
- Flex rating — is the proposed cable rated for continuous flex or torsion, what is the minimum bend radius as a multiple of outside diameter, and what jacket material?
- Connector shell material on RJ45-style and circular feedback connectors, and whether the shield lands on the shell or on a pin.
- What is tested at outgoing inspection, in writing — continuity and insulation resistance are the usual answers, and neither is a noise test, so agree in advance who validates electrical noise performance and on which machine.