Interference Mitigation & Shielding

Interference is not a substance that can be “grounded away.” It is energy from a source that reaches a susceptible circuit by a physical path and becomes an error only because the victim responds to it. A successful fix must therefore identify all three: source, coupling path, and victim.

That model also separates three goals that are often conflated. Reducing emissions changes the source; shielding, routing, isolation, and boundary filters break a coupling path; selectivity, linearity, coding, diversity, and cancellation harden the victim. The best fix is the one that removes the measured dominant term without creating a new path elsewhere.

One model: source → coupling path → victim #

flowchart LR
  SRC["Source<br/>switch edge, PA, motor"] --> PATH["Coupling path<br/>wire, shared impedance, E or H field"]
  PATH --> VICT["Victim<br/>sensor input, LNA, ADC, detector"]
  EMIT["Reduce emission<br/>slew, loop area, backoff"] -.-> SRC
  BREAK["Break coupling<br/>bond, shield, filter, isolate"] -.-> PATH
  HARD["Harden victim<br/>balance, selectivity, linearity"] -.-> VICT

The diagram is a debugging contract. “Add a ferrite” is not yet a diagnosis: it is a hypothesis that a cable carries common-mode RF and that added series impedance at the right frequency will reduce that current. A valid experiment names the predicted path, changes one parameter, and measures the result at the victim and at the emission point.

Classify the interference before choosing a part #

The same disturbance needs several labels because each axis answers a different question.

Conducted and radiated are not mutually exclusive #

A disturbance can conduct along a cable and then radiate from it. Conversely, a radiated field can induce common-mode cable current which enters the enclosure through the connector. Calling the problem “radiated” because an antenna found it describes only the final observation; the controllable path may still be a conducted current on a cable.

Near-field coupling divides further:

• capacitive coupling grows with source-to-victim capacitance and $dv/dt$; a fast switch node couples an electric field into a high-impedance victim;
• inductive coupling grows with mutual inductance and $di/dt$; a large current loop couples a magnetic field into a victim loop;
• shared-impedance coupling converts one circuit's current into another circuit's voltage, $v_{mathrm{err}}=i_{mathrm{dist}}Z_{mathrm{shared}}$.

Reducing separation alone may help field coupling and do nothing to a shared return. Adding a metal shield may help an electric field but do little against low-frequency magnetic coupling. Classification precedes treatment.

Common mode and differential mode require different filters #

For two line currents $i_1$ and $i_2$, one convenient decomposition is

$$i_{CM}=\frac{i_1+i_2}{2},\qquad i_{DM}=\frac{i_1-i_2}{2}.$$

The intended signal on a balanced pair is usually differential. Imbalance in driver impedance, receiver impedance, connector geometry, or filtering converts common mode to differential mode, so a perfectly symmetric external field can still become a decision error. This is why component matching and return geometry matter even when the receiver advertises high DC common-mode rejection.

A current probe clamped around both conductors of a pair largely cancels the intended differential currents and reports their residual sum: a direct indicator of common-mode current. Measuring one conductor alone does not make that separation.

What the receiver sees #

Co-channel, adjacent-channel, and blockers #

A co-channel interferer survives any linear filter that also passes the wanted waveform. The available levers are therefore separation, antenna pattern or beamforming, scheduling, transmit power control, orthogonalization, coding, or a reference-assisted canceller.

Adjacent-channel interference has two contributors:

1. the interferer's transmitter leaks energy outside its assigned channel;
2. the victim receiver has finite adjacent-channel selectivity, so some out-of-channel energy reaches the detector.

Improving only one side can leave the other dominant. A very clean transmitter still blocks a receiver with inadequate preselection, and an excellent receiver cannot remove energy the transmitter already placed inside the wanted band.

ETSI TS 138 104 V19.5.0 defines base-station adjacent-channel selectivity as receiving a wanted signal at the assigned channel in the presence of an adjacent signal at a specified offset. It defines in-band blocking similarly but with an unwanted interferer farther from the assigned channel, and requires at least 95% of the reference-channel throughput under the specified test conditions. The distinction is operational: ACS asks whether the channel filter separates neighbours; blocking asks whether a strong signal elsewhere destabilizes the receive chain.

The nonlinear mechanisms that create $2f_1-f_2$, compression, and spectral regrowth are developed in signal distortion. Here the design implication is narrow: put selectivity before the stage that overloads, and give every stage enough P1dB, IIP3, and converter headroom for the largest off-channel input, not merely the wanted signal.

Work the power sum: C/I and SINR #

Carrier-to-interference ratio is useful when interference dominates thermal noise:

$$\frac{C}{I}_{\mathrm{dB}}=P_C-P_I.$$

For several interferers and noise, powers add in linear units, not in dB:

$$P_{N+I,\mathrm{dBm}} =10\log_{10}\!\left(10^{P_N/10}+\sum_k10^{P_{I,k}/10}\right),$$

$$\mathrm{SINR}_{\mathrm{dB}}=P_C-P_{N+I}.$$

All powers in that expression are referred to the same point and bandwidth. An adjacent signal must first be reduced by the measured or specified rejection between its input frequency and the detector-equivalent input.

This extends the thermal link-budget calculation: use the same reference plane and bandwidth, then add every noise and interference contribution linearly before comparing with the modem threshold.

Reduce emissions at the source #

Source treatment attacks energy before it can find an antenna.

• Keep the forward and return paths close. Differential radiation scales with loop area; a return forced around a plane split can turn a small logic signal into a large magnetic loop.
• Control edge rate to the minimum the timing and signal-integrity budget requires. Clock frequency does not bound emissions; a 10 MHz waveform with sub-nanosecond edges has energy far above 10 MHz.
• Dampen ringing at its impedance discontinuity. A source termination or a correctly designed RC/RCD snubber can remove a high-frequency resonance instead of masking it downstream.
• Minimize switch-node area and high-$di/dt$ loops in power converters. Place local energy storage so the pulse current closes locally rather than through a cable or chassis.
• Back off an RF power amplifier when compression or intermodulation is creating adjacent-channel energy. Backoff spends average power and efficiency, so verify EVM and out-of-band emissions with the deployed waveform rather than a single CW tone.
• Avoid unnecessary current imbalance. A differential interface radiates efficiently when parasitic capacitance or unequal termination converts part of its current into common mode.

Break the coupling path #

Twisted pair, coax, fibre, and boundary parts #

Boundary placement is causal. A filter installed inside an enclosure with an unfiltered wire between the connector and the filter has not protected the boundary; it has imported the disturbance and supplied it an internal radiator. Put the current-diverting element at the penetration with a short, low-inductance chassis connection appropriate to the interface and safety requirements.

Shielding is a current-routing problem #

A continuous conductor responds to a field by carrying induced surface current. At high frequency the current is concentrated within the skin depth

$$\delta=\sqrt{\frac{2}{\omega\mu\sigma}},$$

where $\omega$ is angular frequency, $\mu$ permeability, and $\sigma$ conductivity. Increasing conductivity, permeability, frequency, or thickness relative to $\delta$ increases absorption in an ideal sheet. Real enclosures usually fail somewhere else first.

Apertures

Current must detour around a hole. Leakage rises as the aperture's longest dimension becomes significant relative to wavelength; a long narrow slot can therefore be worse than many small holes with the same total area.

Seams and removable covers

An unbonded seam is a distributed slot. Closely spaced fasteners, clean conductive mating surfaces, and a suitable conductive gasket maintain many short current paths instead of one long inductive detour.

Cable and connector entries

A cable crossing the enclosure can bypass the wall. Terminate a broadband cable shield around the connector circumference to the chassis at the boundary; a long pigtail adds inductance and forfeits high-frequency shielding.

Low-frequency magnetic field

A thin good conductor may provide little attenuation when it is thin relative to skin depth. Reduce source and victim loop area, increase separation, change orientation, or use an engineered high-permeability flux path rather than assuming an aluminium box is universal.

NASA MSFC-HDBK-3697 states the practical enclosure rule directly: unbonded joints and seams are RF apertures, while effective high-frequency bonds need low inductance and large contact area. IEC 61000-5-7 exists because enclosure shielding must be measured as a frequency-dependent property of the assembled mechanical boundary, not inferred from the sheet-metal alloy.

Grounding, bonding, protective earth, and signal reference #

These words describe different jobs:

• protective earth (PE) is a safety path intended to keep accessible conductive parts from remaining hazardous under a fault;
• bonding joins conductive parts so voltage difference and impedance between them meet a defined current and frequency requirement;
• signal reference is the circuit node against which a signal is measured;
• functional earthing is an installation choice made for operation or EMC, subject to the applicable product and installation rules.

At RF, “ground” is not an infinite sink. A conductor with milliohms of DC resistance can have substantial inductive impedance, $Z\approx j\omega L$. That is why a short, wide bond or direct metal-to-metal contact can outperform a long round wire even though both read zero ohms on a handheld meter. The desired current path must be drawn explicitly.

There is no universal “connect every shield at one end” rule. Low-frequency measurement systems may need isolation or a controlled single-point reference to prevent signal-band loop current, while broadband RF shielding needs a low-impedance termination at the boundary. Choose from the interface standard, frequency range, common-mode voltage, fault path, and measured current—not folklore.

Harden the victim #

Selectivity and linearity before gain #

A receiver preselector, duplexer, or notch filter is valuable only if it precedes the stage being overloaded. A beautiful digital notch after the ADC cannot restore samples clipped by a blocker. Likewise, an LNA with the lowest available noise figure may be the wrong first stage if its IIP3 or compression point cannot tolerate the local transmitter environment.

Practical levers include:

• a narrower RF or IF passband consistent with acquisition error, drift, and the wanted waveform;
• a notch for a stable strong interferer, checked for group-delay and wanted-band loss;
• higher-linearity LNA, mixer, VGA, and ADC choices;
• an attenuator or gain-state change that trades noise figure for blocker headroom;
• balanced sensor and data inputs with impedance symmetry preserved through the filter;
• hysteresis, debounce, blanking, or temporal validation for impulsive control inputs when the application can tolerate the latency;
• FEC and interleaving when residual errors are random or burst-like rather than persistent overload.

The trade must be measured at the decision output. A 6 dB input attenuator may remove 20 dB of intermodulation in a cubic region but also raises the thermal sensitivity floor by 6 dB. The correct gain state depends on whether blocker headroom or receiver noise is binding.

Spatial, frequency, time, and code separation #

System-level separation changes how often or how strongly source and victim overlap:

• spatial: antenna directivity, polarization, physical separation, beamforming, or a spatial null reduces coupling without spending bandwidth;
• frequency: guard bands, retuning, hopping, and dynamic channel selection avoid occupied or distorted regions;
• time: scheduling, blanking during a known local pulse, or interleaving prevents persistent collision;
• code: correlation separates users whose codes retain low cross-correlation under the actual timing and multipath conditions.

These are not interchangeable. Frequency diversity does not help a broadband jammer; a spatial null fails when desired and interfering signals share the same direction; time blanking loses duty cycle; orthogonal codes lose orthogonality under delay, frequency offset, or near–far imbalance. The correlation and acquisition limits are developed in spread spectrum and orthogonal codes, while independent-path combining belongs to fading and diversity.

Active cancellation needs a usable reference #

An adaptive canceller observes a reference $r[n]$ correlated with the interference, models its path with weights $\mathbf w$, and subtracts the estimate from the primary input $d[n]$:

$$e[n]=d[n]-\mathbf w^T\mathbf r[n].$$

The Widrow adaptive-noise-cancellation result is powerful because the reference need not equal the interference at the victim; it needs enough correlation for the adaptive filter to learn the intervening path. Three constraints decide whether it works:

1. the reference must observe the interferer but not contain a significant copy of the desired signal, or the canceller will erase wanted energy;
2. the model and adaptation rate must track the real coupling path without chasing noise;
3. cancellation must occur before irreversible overload. Digital cancellation after an ADC cannot undo LNA compression or clipped samples; strong self-interference may require RF/analogue cancellation first and digital cleanup second.

Always compare cancellation depth over the occupied bandwidth, not at one tone. Delay error, phase noise, nonlinear products, and path motion make wideband cancellation much harder than nulling a stationary carrier.

A regulatory example: why “36 dBm EIRP” is not a universal rule #

The current 47 CFR 15.247 text illustrates why a slogan is not a spectrum plan. For digitally modulated systems in 902–928, 2400–2483.5, and 5725–5850 MHz, paragraph (b)(3) gives a 1 W (30 dBm) maximum conducted-output-power basis. Paragraph (b)(4) bases that limit on antenna gain not exceeding 6 dBi and, except for paragraph (c), reduces conducted power one-for-one for gain above 6 dBi.

For a 15 dBi antenna in the ordinary non-exempt case, including point-to-multipoint operation:

$$P_{\mathrm{cond}}=30-(15-6)=21\ \mathrm{dBm},\qquad \mathrm{EIRP}=21+15=36\ \mathrm{dBm}.$$

But paragraph (c)(1)(i) gives systems used exclusively for fixed point-to-point operation in 2400–2483.5 MHz a reduction of only 1 dB per 3 dB of antenna gain above 6 dBi. The same 15 dBi antenna therefore gives

$$P_{\mathrm{cond}}=30-\frac{15-6}{3}=27\ \mathrm{dBm},\qquad \mathrm{EIRP}=27+15=42\ \mathrm{dBm}.$$

Paragraph (c)(1)(ii) permits fixed point-to-point systems in 5725–5850 MHz to use gain above 6 dBi without the corresponding conducted-power reduction; under this antenna-power arithmetic, 30 dBm conducted plus 15 dBi is 45 dBm EIRP. Paragraph (c)(1)(iii) explicitly excludes point-to-multipoint, omnidirectional, and specified co-located same-information arrangements from “fixed point-to-point.” PSD, unwanted-emission, equipment-authorization, RF-exposure, and other applicable requirements remain separate constraints.

EMC pre-compliance and diagnostic workflow #

Compliance and debugging answer different questions. CISPR basic standards define reproducible disturbance measurement methods; product-family or generic standards select limits, detector settings, configurations, and operating modes. IEC immunity standards inject defined disturbances and apply performance criteria. A near-field probe is excellent for localization and poor evidence of formal compliance.

1. Define the failure and reference plane. Record the victim metric—BER, packet loss, EVM, ADC error, false reset, audible artifact—and the exact equipment mode, cable set, load, bandwidth, detector, and software state.
2. Make a background measurement. Measure with the DUT off before attributing a spectral line to it. Then capture the DUT baseline with repeatable cable routing and all relevant ports exercised.
3. Separate conducted from radiated evidence. Use the applicable LISN/AMN or coupling network and CISPR receiver method for conducted ports; use the defined antenna distance/site for radiated measurements. CISPR 16-2-1 emphasizes conducted disturbances, especially 9 kHz–30 MHz; CISPR 16-2-3 covers radiated disturbance methods from 9 kHz to 18 GHz. The target product standard decides what applies.
4. Measure cable common-mode current. Clamp around the whole cable, scan position for a local maximum, and record probe transfer impedance and bandwidth. Rank cables before opening the enclosure.
5. Localize with near-field probes. Start with a larger magnetic loop for sensitivity, then a smaller loop for spatial resolution. Rotate it and hold distance and orientation constant for before/after comparisons. Use an electric-field probe for high-impedance switch nodes, capacitive paths, and enclosure leaks.
6. Correlate in time and operating mode. Trigger on the converter switch, motor event, radio burst, or error flag. If a line moves with clock frequency or disappears in a known firmware mode, the source hypothesis becomes testable.
7. Perturb one path at a time. A temporary clamp ferrite, bonded foil over one seam, shorter cable, alternate isolated interface, lower edge-rate setting, or external preselector is an A/B experiment. Record the source level, cable current, far-field trend, and victim metric; do not treat one local reduction as success.
8. Probe immunity, not only emissions. IEC 61000-4-6:2023 evaluates conducted RF disturbances induced on connected wires from 150 kHz to 80 MHz; IEC 61000-4-3:2020 provides the radiated RF field immunity method. Exercise all required ports and monitor functional performance while sweeping. Formal level setting, field uniformity, coupling devices, safety, and performance criteria belong to the applicable standard and qualified test setup.
9. Re-test the complete configuration. Replace temporary fixes with production geometry and components, repeat worst-case modes and cable layouts, then run the applicable formal emissions and immunity tests. Pre-compliance narrows risk; it is not a substitute for the required conformity procedure.

Failure signatures #

The reliable order is therefore classify → quantify → localize → perturb → harden → repeat the standards-based test. A good EMC fix is not the one with the largest ferrite or thickest enclosure; it is the one that removes a demonstrated source–path–victim chain while preserving signal integrity, receiver sensitivity, safety, and the required operating mode.