Connector Grounding and Bonding: Where the Shield Current Actually Goes

Every shielded connector is sold on a number that describes one short length of metal: the shielding effectiveness of the shell, or the transfer impedance of the braid. Both are honest numbers and both are close to meaningless on their own, because the current that the shield is supposed to carry does not stay in the shield. It leaves the braid at the backshell, crosses a termination that somebody made by hand, runs through the connector shell, crosses the flange joint into the enclosure, and from there has to reach a plane or a structure that can absorb it. Any one of those crossings can be an order of magnitude worse than the shield, and the worst one sets the performance of the whole assembly.

This article is about that path rather than about the connector: how it is measured, where it breaks, and what the arithmetic says about the choices that get made in the drawing and on the bench. There are no proprietary numbers here. Every figure below is either computed from inputs printed beside it or is a specification limit you should be reading from your own product document rather than from a blog, including this one.

One framing rule first. A shield does not block current; it offers it a preferred route. Shielding works because the interfering current divides between the shield path and everything else in inverse proportion to impedance, so the discipline is keeping that path's impedance low across the whole frequency range that matters. Bonding is not a separate topic from shielding: it is the part of the shield that is not made of braid.

The Connector Is One Segment of a Path

Procurement asks a question that sounds complete: is the connector shielded? It is not complete. A shielded connector terminated with a drain wire is an unshielded connector with a more expensive shell. A bonded connector installed on an anodised panel is an isolated connector. A system that passes its radiated-emissions test on a bench and fails in the vehicle has usually not changed its connector at all, it has changed the return path behind it.

The practical consequence is that a component-level qualification report is not evidence about the installation. It describes a specimen in a fixture, terminated by the people who designed the termination, bonded to a ground plane nobody had to think about. Your assembly is a different circuit with the same part number on it, and the difference lives in three joints that appear on no drawing: shield-to-backshell, shell-to-chassis, and chassis-to-board.

Three Segments, Three Different Impedances

The path divides naturally into four segments — the same four you will find on any harness drawing — and they fail for completely different reasons. Treating them as one item is what produces the situation where every element meets its own requirement and the assembly still radiates.

The segments are in series, so the largest impedance dominates and the rest is decoration. That is why fixing the connector rarely fixes the problem: the connector is usually the best-behaved segment in the chain, because it is the only one that was designed, tested and inspected.

Why DC Resistance Tells You Almost Nothing

Bonding is normally accepted on a DC resistance measurement — four-wire, micro-ohmmeter, probe on bare metal. Military bonding practice commonly quotes a 2.5 mΩ class limit for an RF bond; treat that as orientation only and take the actual limit from your own specification and drawing, because the applicable value depends on the bond class and on what the bond is for. The measurement is necessary. It is nowhere near sufficient.

The reason is that current does not use the cross-section you think it does. It concentrates in a surface layer whose thickness falls as the square root of frequency, and it falls faster in magnetic materials:

Computed as δ = √(ρ / (π f μ₀ μ_r)) with ρ = 1.72×10^-8 Ω·m for copper, 2.82×10^-8 for aluminium and 1.43×10^-7 for low-carbon steel. Two things follow that matter on a real drawing. First, at 1 MHz a copper path is using the outer 66 micrometres, so a heavy brass shell and a thin plated shell are electrically similar — the alloy contributes almost nothing and the finish contributes almost everything. Second, once a plating is thinner than the skin depth, its resistance stops improving with thickness and is set by the plating thickness alone:

Computed as R_s = ρ/t while the plating is thinner than the skin depth, and R_s = ρ/δ once it is thicker. A 2 µm silver plating is about thirty times more resistive per square than bulk silver at 1 MHz, and it stops improving entirely above roughly 100 MHz because the current has already left the metal behind it. This is the honest version of the argument that thin decorative plating is not an RF finish: the number that matters is plating thickness against skin depth at your highest frequency of concern, not the words “silver plated” on the drawing.

Geometry then makes the DC measurement look even less relevant. A flat copper strap 100 mm long, 10 mm wide and 0.5 mm thick measures about 0.34 mΩ at DC — comfortably inside any bond limit. Its inductance, at roughly a nanohenry per millimetre of length, is tens of nanohenries: several ohms at 100 MHz. The joint passed its test and still fails at a frequency the test was never run at. If a bond has to work above a few hundred kilohertz, specify its width and length, not its alloy.

Transfer Impedance: What the Shield Actually Leaks

Transfer impedance is the honest way to describe a shield, because it is a coupling term rather than a claim. It is defined as the longitudinal voltage appearing inside the shield per unit length, divided by the current flowing on the outside, and it is quoted in milliohms per metre. The series noise voltage it injects into your circuit is then simply:

V = Zt × Icm × L

Twenty millivolts does not sound like much until you remember that it sits in series with a few volts of low-level analogue signal, and that the same twenty millivolts is driving the cable as an antenna. A radiated-emissions test measures a field produced by exactly this voltage applied to whatever length of cable is radiating, which is why a shield with a good transfer impedance and a bad termination still fails.

Transfer impedance has two mechanisms and they point in opposite directions. Through a solid wall the current diffuses, so the coupling falls with frequency. Through the apertures of a braid the coupling rises with frequency, because a braid is a mesh and the mesh gaps behave as small slots. A braid therefore has a transfer impedance that flattens and then climbs, and the crossover is roughly where the aperture term overtakes the diffusion term — typically in the low megahertz for a dense braid and lower for a sparse one. This is why coverage percentage and braid construction matter more than conductor cross-section, and why a foil-plus-drain construction, which is excellent at low frequency, is poor at high frequency: the drain wire is a single inductor and the foil's longitudinal seam is a slot.

Read the transfer impedance curve from the cable datasheet at the band you care about; if it is not published, treat that as a finding about the cable rather than an invitation to assume a value. What you cannot do is substitute DC resistance for it, even though both are quoted in milliohms and appear in the same table. Our MIL-DTL-38999 Series III range is specified this way — braid and shell together, not shell alone.

The Pigtail Penalty, Computed

If the braid is terminated through a short lead — a drain wire, a single strap, or braid twisted into a tail and crimped — the shield current has to pass through that lead's inductance to reach the shell. The numbers are unforgiving, and they are the single most useful calculation in this subject:

Computed as Z = 2πfL with L taken as 1 nH per millimetre of lead, which is an order-of-magnitude rule for a wire or a narrow strap and is accurate enough to make the decision. Compare a 10 mm pigtail at 100 MHz — 6.3 Ω — with a properly banded 360° termination at a couple of milliohms. The ratio is about 2,500 to 1. Current divides inversely with impedance, so approximately 99.96 % of the shield current takes the pigtail and the expensive 360° termination carries essentially nothing. The bond has not failed; it has been bypassed.

That ratio answers the common objection that a short pigtail is fine because it is short: at 10 MHz the same lead is 0.63 Ω against a few milliohms, still a ratio of hundreds. There is no length at which a single lead is acceptable above the low megahertz, which is why the braid is specified as captured around its full circumference.

Where a lead is genuinely unavoidable — a repair, a field re-termination, a shell that cannot take a band — use several short wide straps around the circumference instead of one: the inductances run in parallel and the current has somewhere to go. Keep the length-to-width ratio of any single conductor under about three to one as well, because a flat conductor spreads current across its width rather than crowding it into a filament, and the inductance falls with it.

What “360°” Has to Mean

Termination methods are not ranked by quality so much as by what they actually capture. The distinction that matters is whether the braid is captured as a surface or as a wire.

The common error is a drawing note that says “terminate shield” and a build that satisfies it with a boot and a lead. Both parties are correct on their own terms and the assembly is not shielded. If the requirement is RF bonding, the drawing has to say so and the inspection has to look under the boot, because after the boot is shrunk the termination is permanent and invisible.

There is a second, related confusion worth clearing up: the accessory thread is not a bond. Screwed metal-to-metal threads do provide contact, but the contact is at the thread flanks, it is uncontrolled, it depends on engagement length and lubrication, and it degrades with every mating. Where a connector family is designed for shell-to-shell RF continuity, that continuity comes from the coupling ring and its grounding fingers, not from the accessory thread. Treat the thread as a mechanical attachment and provide the bond deliberately — with the backshell's own ring or with a separate strap. Our M85049 backshell range and the wider backshell accessories are specified by termination method for exactly this reason, the MS3057 cable clamps are a mechanical strain-relief family rather than an RF termination, and the J599 III shells carry the plating that the joint actually relies on.

The Shell-to-Chassis Joint Is a Joint, Not a Surface

Between the connector flange and the enclosure there is a stack: finish, gasket, panel finish, fastener, and whatever the painter did on the day. The bond has to cross all of it, and the finishes make the difference between a joint and an insulator.

The star washer is the usual answer to paint and anodise and works for the reason people think: it displaces the film and makes metal-to-metal contact. What is less often considered is that it does so over a very small area, which is both the mechanism and the liability — a small anode against a large cathode corrodes at a rate that surprises people who expected an electrical part rather than a wear part. Use it deliberately, keep it dry, and inspect it.

The subtler problem is the gasket. A sealing gasket between the flange and the panel does its job by being compressed, and it does that job best when it is soft, thick and continuous — three properties that all remove metal-to-metal contact from the joint. If the same joint has to seal and bond, decide which element carries each function: the gasket seals, the hardware bonds, and the hardware therefore has to be designed, specified and inspected as a conductor. That trade-off, and the preload that makes it work, is covered in more detail in this site's panel attachment hardware guide.

The Panel Cutout Is an Antenna

An enclosure is a shield with holes in it, and its shielding effectiveness is bounded by its longest hole. A cutout is longer than the connector it serves — flange, keyway and clearance all add length — and two cutouts placed close together behave electrically as one longer slot. A narrow slot radiates most efficiently near the frequency where its length is a half wavelength:

f ≈ c / (2 L)

Computed as f = c/(2L) with c = 3×10⁸ m/s. The resonance is the worst case, not the threshold: coupling through a slot rises steadily as its length approaches a significant fraction of a wavelength, so a 200 mm cutout is already a real leak at 100 MHz where it is only a fifteenth of a wavelength, and is a serious one by 500 MHz. This is the arithmetic behind the observation that emissions problems appear at the top of the test band rather than in the middle of it.

Three consequences follow, all cheap at the drawing stage. Do not enlarge a cutout for assembly clearance beyond what the connector requires, because every extra millimetre of slot is paid for in shielding. Do not place two cutouts adjacent without a metal bridge between them, because they merge into one longer aperture. And remember that the shell does not close the aperture: it passes through the hole while the flange sits on the surface, so the only thing closing the slot is the flange-to-panel interface — which is why the gasket is an EMC component as well as a sealing one.

Gaskets That Conduct and Gaskets That Seal

A conductive gasket is a spring with two jobs that pull in opposite directions: sealing wants soft, thick and compliant, conducting wants hard, thin and heavily loaded. A single part has to be chosen with both in view, and compression set is where the compromise shows up in service.

The rule that survives contact with production is to keep the functions separable. If the same gasket seals and bonds, then a compression-set gasket that still passes the leak test can have already lost its bond, and nothing in the routine test will notice. Where both functions are required in one part, the joint needs a compression stop or a controlled groove so that the squeeze is set by metal, not by the gasket's own stiffness, and the bond must be verified on a sample after environmental exposure, not only at first build.

Single-Point, Multipoint, and the Loop You Did Not Draw

Grounding advice contains an apparent contradiction that is not a contradiction once frequency is taken into account. At low frequency, two grounds create a loop, the loop is a transformer, and the induced current shows up as hum and as common-mode noise. At high frequency, a single ground point is a long inductive stub, and the return current takes whatever path it can find — usually the one you did not design. The resolution is that the correct practice changes with the relationship between the loop geometry and the wavelength.

Computed as V = A · 2πf · B_pk for a sinusoidal field. The table makes the frequency dependence obvious and the geometry dependence easy to underestimate: a hundred square centimetres at 1 MHz in a modest field produces more than half a volt of series interference in a loop that nobody drew, because the loop is formed by two bond straps and the chassis between them. At 400 Hz the same loop produces millivolts — which is the regime where single-point grounding earns its reputation, because the loop current is small and the alternative, multiple grounds, permits a much larger circulating current.

The working rule is to control loop area at low frequency and bond extensively at high frequency, deciding the regime by whether the loop is small against a wavelength: below about a twentieth, treat it as a circuit and manage the loop; above it, give the current a short, wide, multipoint path. This is why a bond that is excellent for lightning — a heavy single strap to structure — can be poor for RF, and why an RF bond is not automatically a lightning bond.

One case deserves a sentence of its own. An isolated connector shell — mounted on an insulating gasket or a plastic insert — does not ground the shield at all, and a filtered connector's capacitors need a ground to work into. Isolation is a legitimate way to break a ground loop, but it is a decision about where the current goes, not an absence of a path, and it has to be paired with a deliberate bond elsewhere.

Getting the Current Off the Shell and Into the Board

The final segment is the one most often left to layout. Shield current arrives on the shell and must reach a plane that can absorb it without crossing the signal return. On a board-mounted connector the shell is bonded to a ring of shell-ground pads, and those pads must be stitched into the chassis or shield plane with enough vias that the connection behaves as a surface rather than a point.

The spacing rule of thumb is a twentieth of a wavelength at the highest frequency you care about: about 150 mm at 100 MHz, 30 mm at 500 MHz and 15 mm at 1 GHz. A single via per pad at 1 GHz is a point connection with nanohenries of inductance, and a row of them at 15 mm pitch is a wall. The same arithmetic applies to a board-level shield can: bonded at four corners with long gaps between, it is four inductors and a set of slots, which is why perimeter bonding with a dense via fence outperforms a few large tabs.

Two currents share a connector and must not share a return. Signal return is the reference for the data and must stay continuous under the signal path; shield current is common-mode and must be kept off it, which is the whole reason for the shell-ground ring. When a split plane forces shield current through the signal reference, the symptom is errors that track an external event rather than the data rate — a board that passes on the bench and fails in the enclosure.

Nine Field Symptoms and What to Measure First

Symptoms are ambiguous, so the useful discipline is to convert a symptom into the first measurement that discriminates between mechanisms. The table below is ordered by how often each one turns out to be the answer, not by how interesting it is.

The last row is included because it is the most common misdiagnosis in the whole subject: heat means resistance in the power path, and the bond is not in the power path. The two are measured with the same instrument and answered by different parts of the drawing.

Specifying a Bond Path: A Ten-Line Checklist

If the bond path has to be designed rather than inherited, these ten lines on the drawing do most of the work. None of them is unusual; what is unusual is having all ten in one place.

1. State the bond class and the limit, and state the frequency range it applies over — a DC limit alone does not describe an RF bond.
2. Name the finish on both mating faces, and mask the bond area if the standard finish is anodise or paint.
3. Specify the termination method for the shield by name: band, clamp, saddle, ring — not “terminate shield”.
4. Set a maximum lead length and a maximum length-to-width ratio where a lead is unavoidable.
5. State whether the flange joint must be metal-to-metal, and if a gasket is present, say whether it is conductive.
6. Identify the fastener as a conductor or as an insulator, and choose the washer accordingly.
7. Give the maximum slot length for any aperture in the enclosure, including connector cutouts and their clearance.
8. Give the via pitch at the shell-ground ring for the highest frequency of concern.
9. Require bond resistance measurement at bare metal, four-wire, and define the sample and the interval after environmental exposure.
10. Require a visual inspection of the termination before any boot, tape or potting makes it permanent.

Line ten is the one that pays for itself: almost every termination defect described here is visible for a few seconds during build and invisible afterwards, and the inspection costs less than the rework that follows a failed emissions test. Bond and plating measurements are covered in our quality and test capability page.

Frequently Asked Questions

Is a shielded connector enough on its own?

No. The connector is one segment of a path that also includes the shield termination, the flange joint and the chassis-to-board connection. A shielded connector with a drain-wire termination, or one mounted on an anodised panel without a deliberate bond, behaves as an unshielded connector in the frequency range where the path impedance is dominated by that weak segment — which is usually the range the emissions requirement cares about.

Why does my bond pass the milliohm test and still fail at high frequency?

Because the test is a DC measurement and the failure is inductive. A joint that measures a fraction of a milliohm at DC can present several ohms at 100 MHz once its geometry is taken into account, and the current simply takes the lower-impedance route elsewhere. If a bond has to work above a few hundred kilohertz, control its length and width, and measure it with a method that has some frequency content rather than only with a micro-ohmmeter.

Can I use a drain wire if I keep it very short?

Not as an RF termination. A 10 mm lead is about 10 nH, which is roughly 6 Ω at 100 MHz against a couple of milliohms for a proper 360° band, so the lead carries almost all of the shield current regardless of how short it is. Short leads are less bad than long ones, and several short wide straps distributed around the circumference are better than one, but the only real fix is capturing the braid around its full circumference.

Does the gasket between the connector and the panel break the bond?

It can, and it is one of the most common causes of a bond that exists in the prototype and not in production. A non-conductive gasket removes the metal-to-metal contact that was doing the bonding, so the bond has to be carried by the hardware — screws, inserts and washers — and designed and inspected as a conductor. If both sealing and bonding are needed in the same joint, use a conductive gasket, or provide a deliberate masked bond path alongside the seal.

Should I ground at one point or many?

Both, in the right frequency range. Below roughly a twentieth of a wavelength, manage the loop: a single reference point avoids circulating current and the induced voltage is small. Above it, bond extensively, because a single point is a long inductive stub and the return current will find a path you did not design. The question to ask is not “how many grounds” but “what is the loop area, and is it small against a wavelength at the frequencies present”.

Does a conductive gasket fix a bad flange bond?

It helps only if the joint is designed for it. A conductive gasket is a spring, and it conducts through the compression it retains; if the joint has no compression stop and the gasket takes a permanent set, the bond degrades while the seal may still pass. Where the bond matters, verify resistance on a sample after thermal cycling rather than on the first build, and prefer a design in which metal, not elastomer, sets the squeeze.