Cable Shielding and Grounding for Sensor Signals: One End, Both Ends, or Neither

A 4-20mA level transmitter reads a rock-steady 12.00 mA on the bench calibrator. The technician signs the paperwork and pulls the cable into the plant. Two weeks later the same loop jitters half a milliamp every time the mixer's VFD ramps up. Nobody touched the sensor.

This failure traces back to a single wrong call about shielding or grounding. Ground a shield at one end on a fast-switching VFD cable, and it radiates like an antenna. Ground it at both ends on a slow analog loop, and it becomes the ground loop it was supposed to prevent. The right answer turns on frequency, cable length, and what else shares the tray - change any one of those and the answer flips.

TL;DR: Ground a shield at one end for low-frequency loops like 4-20mA, RTD, and thermocouple wiring below roughly 100 kHz - this breaks the DC path a ground loop needs. Ground it at both ends, with a full 360-degree termination, for VFD, servo, and high-speed digital cable, where an ungrounded end turns the shield into a resonant antenna (NotebookLM, Cable Shielding notebook, 2026).

This guide covers what a shield does, the two ways noise gets into a cable, the one-end-versus-both-ends decision and its exceptions, how to terminate a shield for the frequency you're fighting, routing and separation distances, and what to do when you can't fix the grounding.

What Does a Cable Shield Protect Against?

A cable shield is a grounded conductive layer wrapped around the signal conductors. It intercepts electric-field noise before it reaches them and does little against magnetic-field noise on its own (NotebookLM, Cable Shielding notebook, 2026). Treat it as a Faraday cage for one coupling mechanism only.

You add shielded cable, the noise doesn't go away, and you conclude shielding doesn't work. The noise was mostly magnetic, and shielding was never built to stop it. The twisted pair inside the cable was supposed to handle that job.

Leave both ends ungrounded and the shield does nothing - just an insulated conductor sitting next to your signal wires. Ground one end and it blocks electric-field coupling while breaking the DC path a ground loop needs. Ground both ends and it adds a low-impedance path for high-frequency noise, but any low-frequency ground potential difference now drives current straight through it (NotebookLM, Cable Shielding notebook, 2026).

Citation capsule: A cable shield is a grounded conductive layer that intercepts electric-field (capacitive) coupling before it reaches the signal conductors; it does little against magnetic-field (inductive) coupling, which twisted pairs handle (NotebookLM, Cable Shielding notebook, 2026). Whether you ground that shield at one end, both ends, or not at all depends on the frequency of the noise you're fighting. No single scheme is correct for every application.

Capacitive vs Inductive Coupling: Which One Is Corrupting Your Signal?

Capacitive coupling is driven by a fast voltage change (dV/dt) across a parasitic capacitance between conductors. Inductive coupling is driven by a fast current change (di/dt) through a shared magnetic field (NotebookLM, Cable Shielding notebook, 2026). Fix one and the other keeps getting through.

Capacitive coupling injects a displacement current into your signal wire whenever a nearby conductor's voltage swings fast - a VFD's switching edges are a textbook source. A grounded Faraday shield intercepts that displacement current and routes it to ground before it reaches the signal conductor. Separation helps too: coupling capacitance falls with distance.

Inductive coupling follows a different path. A high-current conductor's changing magnetic field penetrates the loop formed by your signal-forward and signal-return conductors, inducing an EMF proportional to the loop's area. A conventional copper or aluminum shield does little at low frequency. Twisting the two conductors together cancels the induced voltage instead: each half-twist, or lay, sees the field with reversed polarity, and the induced voltages cancel at the receiver (NotebookLM, Cable Shielding notebook, 2026). Tighter, more regular twists reject higher frequencies better. An uneven twist creates a flux-trapping area that breaks the cancellation.

Keeping the loop area small matters as much as the twist. Run send and return conductors close together, and route the pair tight against a grounded metal structure like a conduit or tray rail - both moves shrink the loop and cut pickup and radiated emission. Most industrial noise mixes both mechanisms, which is why instrumentation cable standardizes on shielded twisted pair, combining both defenses in one construction.

Citation capsule: A grounded shield intercepts displacement current and stops capacitive coupling, which scales with dV/dt. Twisting the conductors cancels induced voltage every half-lay and stops inductive coupling, which scales with di/dt (NotebookLM, Cable Shielding notebook, 2026). A standard copper shield does little against low-frequency magnetic fields alone - that job belongs to the twist, or to a high-permeability steel shield where a shield alone must do it.

Should the Shield Go to Ground at One End, Both Ends, or Neither?

Ground the shield at one end for signals and interference below roughly 100 kHz to 1 MHz. Ground it at both ends above that threshold, or once the cable's length approaches a real fraction of the interfering noise's wavelength (NotebookLM, Cable Shielding notebook, 2026). That frequency threshold is the whole rule. The rest of this section works out which side of it your cable sits on.

Single-point grounding suits classic process instrumentation: 4-20mA loops, thermocouples, RTDs, and other slow analog signals, where 50/60 Hz power distribution is the main threat. Leaving one end floating breaks the DC path a ground loop needs. Ground the shield at the signal source to match the voltage reference, or at the receiving end if the source floats (NotebookLM, Cable Shielding notebook, 2026).

Both-ends grounding fits VFD output cable, servo motor cable, high-speed digital links like Industrial Ethernet, and anything sharing space with radiated RF. Once a cable's length reaches roughly one-sixth to one-quarter of the interfering wavelength, a shield grounded at only one end stops acting as a shield and starts acting as a resonant antenna, radiating internal noise and absorbing external RF energy (NotebookLM, Cable Shielding notebook, 2026). Lift one end of a ten-meter cable and it becomes an antenna for anything above roughly 3 MHz.

Both-ends grounding fails on a low-frequency loop because of a parasitic ground loop. Two ground points in an industrial plant rarely sit at the same potential - motors, VFDs, and heavy machinery inject stray current into the earth. That potential difference forces current through any shield bonded at both ends. On a 4-20mA loop it shows up as 50 to 500 mV AC riding on the signal, the fault covered in troubleshooting 4-20mA loop noise (NotebookLM, Cable Shielding notebook, 2026). Heavier applications push the numbers further: a well-meshed grounding system spreads that current across so many parallel paths that any single shield carries a negligible share, a servo cable at full torque can see roughly 1 A with microsecond rise times, and shield currents as high as 5 A are tolerated without noise where the cable bonds straight to chassis rather than into a sensitive circuit's 0V reference (NotebookLM, Cable Shielding notebook, 2026).

IEC 61000-5-2 complicates the traditional one-end rule. It treats most industrial environments as already polluted with high-frequency noise from VFDs, switch-mode supplies, and digital communications, and recommends both-ends bonding by default, paired with a meshed grounding structure and a parallel earth conductor to keep loop current out of the shield rather than lifting an end (NotebookLM, Cable Shielding notebook, 2026). Single-point grounding at the receiver still works for a 4-20mA loop in a quiet cabinet, and it's the simpler install. Near VFDs, servo drives, or dense digital wiring, treat both-ends-plus-PEC as the safer default.

A hybrid termination splits the difference when neither pure approach fits. Hard-ground the shield at the source, and bond the far end through a ceramic capacitor, roughly 10 to 100 nF. The capacitor acts as an open circuit to 50/60 Hz, blocking ground-loop current, and as a short circuit to high-frequency RF, giving that noise the low-impedance path both-ends grounding provides (NotebookLM, Cable Shielding notebook, 2026).

Citation capsule: A parasitic ground loop forms because two earth points in a plant rarely sit at the same potential. Bond a shield at both ends on a low-frequency loop, and that potential difference drives current through it - 50 to 500 mV AC on a 4-20mA loop (NotebookLM, Cable Shielding notebook, 2026). Decision rule: single-point ground below roughly 100 kHz on classic instrumentation loops, both ends with 360-degree termination above that threshold or near VFD and high-speed digital cable, and a capacitor-coupled hybrid when one cable must satisfy both conditions.

How Do You Terminate a Shield Correctly?

A shield termination is only as good as its bond to the enclosure. A wire pigtail is the most common way installers wreck an otherwise sound cable. Terminate the shield with a full 360-degree circumferential connection where the cable enters the enclosure (NotebookLM, Cable Shielding notebook, 2026).

A pigtail termination means unpicking the braid, twisting it into a single lead, and landing it on a terminal. It adds roughly 10 nH of parasitic inductance per centimeter. At low frequency that inductance is negligible. At high frequency, the impedance it presents (X_L = 2πfL) blocks the noise current the shield is supposed to carry to ground and lets it leak into the surrounding circuit. A 25mm pigtail can ruin shielding effectiveness at frequencies as low as 30 MHz; testing has measured a pigtailed connection at roughly 6,000 times worse transfer impedance at 100 MHz than a proper 360-degree termination (NotebookLM, Cable Shielding notebook, 2026).

A 360-degree gland, EMC-rated compression fitting, or metal backshell makes continuous metal-to-metal contact around the braid's full circumference, so the shield acts as an unbroken extension of the enclosure instead of a wire dangling off one point. A saddle clamp that clips the braid against a grounded backplate works as an approximation where a full gland isn't practical - it leaves the braid undisturbed and avoids most of the pigtail's parasitic inductance.

Conduit deserves the same discipline as a cable gland. Steel conduit is the better choice at low frequency: its higher magnetic permeability damps magnetic fields effectively up to roughly 100 kHz, better than aluminum or copper (NotebookLM, Cable Shielding notebook, 2026). Above that, into the megahertz range, aluminum and copper become the more effective shielding materials. Bonding conduit sections with a green ground wire defeats the point. The joints need metal-to-metal contact, and end terminations need the same 360-degree bond a cable gland provides - typically a conduit nut with grub screws that bite through to bare metal.

Citation capsule: A pigtail termination adds roughly 10 nH/cm of parasitic inductance - enough for a 25mm pigtail to ruin shielding at frequencies as low as 30 MHz, and to measure roughly 6,000 times worse transfer impedance than a proper 360-degree termination at 100 MHz (NotebookLM, Cable Shielding notebook, 2026). Terminate every shield with a full circumferential metal-to-metal bond at the enclosure entry - a gland, backshell, or saddle clamp - never a twisted pigtail.

What Cable Routing and Separation Rules Matter Most?

Separation is the cheapest EMC fix, and it works on both coupling mechanisms at once: capacitive coupling falls off with distance, and inductive coupling falls off faster still. IEEE 518 sets numbers: roughly 670 mm (26 in) between bare cable trays carrying sensitive signal and heavy power cable, 460 mm (18 in) if one tray is covered, and 310 mm (12 in) if both run in separate steel conduit (NotebookLM, Cable Shielding notebook, 2026). A simpler field rule: 12 inches minimum from VFD cable for shielded instrumentation cable, widened to 36 inches for unshielded.

These distances aren't fixed once a run gets long. For parallel runs exceeding 100 meters, multiply the base separation by 1.5 to account for cumulative inductive coupling over the extra length. Keep parallel runs short - ten feet or less of close-quarters routing next to a VFD cable is a reasonable target - and cross power and signal cable at 90 degrees to minimize the loop area exposed to the field (NotebookLM, Cable Shielding notebook, 2026).

Cable tray layout matters as much as distance. Stack power on top, control cabling in the middle, and instrumentation on the bottom. That order puts the most susceptible cables farthest from the noise source, keeps rising heat from power cable away from delicate insulation, and puts the cabling technicians touch most often within reach. If different cable classes share a single tray, a grounded steel divider between them removes most of the separation requirement, often cheaper than running a second tray (NotebookLM, Cable Shielding notebook, 2026).

Route VFD power cable and instrument signal cable in separate conduit, no matter how short the run or how clean the load looks. High-frequency switching noise on the VFD conductors couples into anything sharing that raceway, which is why a dedicated conduit for VFD cabling is standard practice (NotebookLM, Cable Shielding notebook, 2026). The same principle applies to 2-wire, 3-wire, and 4-wire sensor wiring: more conductors in a shared raceway means more chances for a noise path to form.

Citation capsule: IEEE 518 sets minimum separation between sensitive signal cable and power cable at roughly 670 mm for bare trays, 460 mm with one covered tray, and 310 mm when both run in separate steel conduit, with a 1.5x multiplier for parallel runs over 100 meters (NotebookLM, Cable Shielding notebook, 2026). Decision rule: stack power on top and signal on the bottom of a shared tray, cross at 90 degrees where paths must intersect, and keep VFD power and instrument signal cable in separate conduit.

When Grounding Can't Be Fixed, Isolation Is the Answer

Every technique above assumes you control the grounding at both ends of a cable run. A plant retrofit, a leased building, or a sensor on someone else's skid often means you don't. When the ground potential difference is out of reach, stop chasing the shield and break the electrical path.

A galvanic signal isolator recreates the signal on the far side of an isolation barrier - optical, capacitive, or transformer-coupled - with no conductive connection between input and output. That barrier blocks the DC path a ground-loop current needs, regardless of how the two ends are wired or grounded. If you've worked through the termination and routing steps above and the loop is still noisy, run the 50 mV AC threshold test from the 4-20mA loop noise troubleshooting guide to confirm a ground loop before reaching for an isolator.

Isolation earns its cost where grounding is unfixable: long runs between buildings on separate grounding systems, sensors that float at a high common-mode voltage, or a control cabinet you don't have authority to rewire. A passive, loop-powered isolator is often the more practical answer than a plant-wide grounding audit, and the fix outlasts the next facility renovation.

Field Checklist Before You Pull Cable

Confirm the frequency first: a slow analog loop, or a VFD, servo, or high-speed digital run. That answer decides one-end versus both-ends before anything else. Check whether the sensor floats or is grounded, since that decides which end gets the single-point bond.

Specify shielded twisted pair, and check the run's length against the wavelength fraction that decides one-end versus both-ends grounding (NotebookLM, Cable Shielding notebook, 2026). Choose termination hardware - a 360-degree gland or backshell, not a pigtail - before the cable arrives on site. Check the separation from the nearest VFD or power cable against IEEE 518 or the 12-inch/36-inch field rule, and route the tray so power sits above signal.

Bond conduit and tray sections metal-to-metal, not with a ground wire alone, wherever either is doing shielding duty. If the two ends of the run sit on grounds you don't control, budget for an isolator instead of a grounding fix you can't make.

Frequently Asked Questions

Should I ground a shield at one end or both ends?

Ground at one end, the source or receiver, for slow analog loops like 4-20mA, RTD, and thermocouple wiring below roughly 100 kHz. Ground at both ends, with 360-degree termination, for VFD, servo, and high-speed digital cables, where high-frequency noise and antenna effects dominate (NotebookLM, Cable Shielding notebook, 2026).

What happens if I ground a shield at both ends on a 4-20mA loop?

Two grounds rarely sit at the exact same potential, so the difference drives a circulating current through the shield by Ohm's law. On a 4-20mA loop that shows up as 50 to 500 mV AC across the signal terminals, riding on top of your measurement (NotebookLM, Cable Shielding notebook, 2026).

Does a pigtail shield termination matter that much?

A pigtail termination adds roughly 10 nH of parasitic inductance per centimeter, and a 25mm pigtail can ruin shielding effectiveness at frequencies as low as 30 MHz. A proper 360-degree gland measures roughly 6,000 times lower transfer impedance at 100 MHz (NotebookLM, Cable Shielding notebook, 2026).

How far should signal cable stay from VFD power cable?

IEEE 518 calls for about 670 mm (26 in) between bare cable trays, 460 mm (18 in) with one covered tray, and 310 mm (12 in) when both run in separate steel conduit. A simpler field rule is 12 inches minimum for shielded cable, 36 inches for unshielded (NotebookLM, Cable Shielding notebook, 2026).

Twisted pair or shielded cable - which one do I need?

Twisting cancels magnetic (inductive) coupling by reversing the loop polarity every lay; a shield intercepts electric (capacitive) coupling. Most industrial signal cable needs both, since plants generate both noise types at once (NotebookLM, Cable Shielding notebook, 2026).

Conclusion

Frequency decides whether a shield goes to one end or both. Slow analog loops get single-point grounding to break the DC ground-loop path. VFD, servo, and high-speed digital cable gets both ends bonded through a 360-degree termination, because an ungrounded end on a long or high-frequency cable stops shielding and starts radiating.

Every noisy loop I've traced back to grounding turned out to be one of two mistakes: a pigtail where a 360-degree gland should have been, or a shield bonded at both ends on a cable that should have had one end floating. Neither mistake is visible on a bench test. Both show up the moment a VFD ramps up fifty feet away.

Get the shield termination and separation distance right at pull time, and most of this guide never comes up again in the field. A signal isolator breaks the path when the ground is out of your control, often cheaper than the grounding audit it replaces. For the base signal these loops carry, see 4-20mA vs 0-10V analog signals.