Electronics Guide

Maxwell's Equations and the Quasi-Static Regime

Maxwell's equations govern every electromagnetic effect in a converter, but solving them in full generality is rarely necessary. Most power electronic analysis uses a quasi-static approximation, in which the displacement current term is neglected for magnetic problems, or conduction and induction are neglected for electric problems, so that fields respond instantaneously to their sources. The approximation holds when the largest physical dimension of the structure is small compared with the wavelength of interest, a condition usually stated as one tenth of a wavelength or less.

The frequency that matters is not the switching frequency but the highest frequency with significant spectral energy, which the switching edge rate sets. A trapezoidal edge has a spectral knee near 0.35 divided by the rise time, so a 10-nanosecond transition carries meaningful content to roughly 35 MHz and a 2-nanosecond wide-bandgap transition to well beyond 100 MHz. At 100 MHz the free-space wavelength is 3 meters, which places a 30-centimeter cable harness or busbar squarely outside the quasi-static regime. This is why radiated emissions, transmission-line behavior on gate loops, and full-wave analysis become unavoidable in fast-switching designs even though the switching frequency itself may be only a few hundred kilohertz.

Numerical Methods for Field Analysis

Computational electromagnetics offers several families of solvers, and choosing the right one matters more than refining the wrong one. The finite element method discretizes the entire problem volume into elements and solves for the field at nodes or edges. It handles arbitrary geometry, inhomogeneous regions, and nonlinear material curves, which makes it the standard tool for magnetic components with saturable cores and distributed air gaps. Its cost is the volume mesh, which must extend into the surrounding air and be refined wherever fields vary rapidly.

The boundary element method discretizes only the surfaces bounding each region, reducing a three-dimensional problem to a two-dimensional mesh. It suits open-boundary problems such as capacitance extraction and external field prediction, but it assumes each region is linear and homogeneous, so it cannot represent a saturating core directly. The method of moments, a related surface-integral technique, dominates radiation and antenna analysis and is well suited to wire harnesses and enclosure currents. The finite-difference time-domain method marches a structured grid forward in time and delivers broadband results from a single run, which is useful for transient and radiated-emission studies. Hybrid formulations couple a finite element region around the nonlinear magnetics to an integral-equation treatment of the open space beyond, capturing both without meshing the entire room.

Interference Sources and Coupling Mechanisms

Every switching converter contains at least one node whose voltage swings the full bus in nanoseconds and at least one loop whose current changes just as fast. The switch node is the dominant electric-field source: its dv/dt, which commonly reaches 10 to 50 volts per nanosecond in silicon carbide and gallium nitride designs, drives displacement current through every parasitic capacitance connected to it, including the capacitance from a transistor tab to its heatsink and from a transformer primary to its secondary. The commutation loop is the dominant magnetic-field source: its di/dt drives current through stray inductance, producing overshoot and ringing that appear directly in the emissions spectrum.

These sources reach a victim by four coupling mechanisms. Conductive coupling shares an impedance, typically a common ground return. Capacitive coupling transfers energy through electric fields between conductors at different potentials. Inductive coupling transfers energy through magnetic fields linking adjacent loops. Radiative coupling dominates once the separation exceeds roughly a sixth of a wavelength, at which point loops and cables act as antennas. Diagnosing an EMI problem means identifying the source, the coupling path, and the victim, because breaking any one of the three solves the problem, and the cheapest fix is not always at the source.

Material Properties and Modeling

Simulation accuracy is limited far more often by material data than by mesh density. Soft magnetic materials exhibit nonlinear B-H curves, hysteresis, and losses that depend on frequency, flux density, temperature, and waveform shape. Core loss is usually fitted with the Steinmetz relation, in which volumetric loss varies as frequency raised to an exponent between roughly 1.1 and 1.7 and peak flux density raised to an exponent between roughly 2 and 3 for typical ferrites. Because the Steinmetz form is derived from sinusoidal excitation, applying it to the trapezoidal flux waveforms of a converter requires an extension such as the improved generalized Steinmetz equation, which integrates loss over the instantaneous rate of change of flux and accounts for duty-cycle asymmetry.

Conductors require equal care. The skin depth in copper is approximately 66 millimeters divided by the square root of the frequency in hertz, giving about 0.21 millimeters at 100 kHz and 66 micrometers at 1 MHz. Any winding whose conductor thickness approaches or exceeds this value carries a strongly non-uniform current, and the proximity effect from neighboring turns and layers makes matters worse: Dowell's one-dimensional analysis shows that AC resistance in a multilayer winding grows rapidly with layer count, which is why interleaving, foil, and Litz wire appear in high-frequency magnetics. Dielectric properties matter for parasitic capacitance and for insulation stress, and both permittivity and loss tangent vary with frequency and temperature. Vendor datasheets rarely span the full operating envelope, so measured characterization at the intended frequency and temperature is often the decisive step in obtaining a trustworthy model.

Static and Low-Frequency Analysis

Magnetostatic analysis solves for the field under DC excitation and answers the questions asked earliest in a magnetic design: what is the inductance, where does the core saturate, how much energy is stored in the gap, and how far does fringing flux extend into the windings. Because gapped cores store most of their energy in and around the gap, a magnetostatic solution also reveals where fringing flux will intercept conductors and create localized heating. Electrostatic analysis returns the capacitance matrix of a winding, busbar, or module and highlights field concentrations at edges and triple points where insulation coordination is at risk.

Time-harmonic eddy-current analysis solves the sinusoidal steady state at a single frequency, yielding frequency-dependent resistance and inductance, eddy-current loss in windings and structural metal, and shielding effectiveness of conductive plates. It is orders of magnitude faster than a transient run, and sweeping it across frequency produces the impedance data needed for filter design and for equivalent-circuit fitting. Its limitation is linearity: a saturating core cannot be represented exactly, so a nonlinear time-harmonic approximation or a transient run is required when flux density approaches saturation.

Transient and Field-Circuit Co-Simulation

Time-domain analysis captures what harmonic methods cannot: saturation, hysteresis, inrush, short-circuit forces, and the true non-sinusoidal waveforms of a switching converter. The cost is severe, because the time step must resolve the fastest transition while the run must span at least one switching period, and the mesh must resolve a skin depth that shrinks as the square root of frequency. Practical models therefore use impedance boundary conditions on thick conductors, symmetry planes, and reduced-order regions wherever the detailed field is not the quantity of interest.

Field-circuit co-simulation couples an electromagnetic solver to a circuit simulator so that the converter's control and semiconductor behavior drive the field model and the field model returns the resulting terminal behavior. This matters for problems in which the two are genuinely coupled, such as saturation-dependent current ripple in a boost inductor, cross-regulation in a multiple-output transformer, or the interaction between a gate loop's stray inductance and the transistor's transfer characteristic. Where the coupling is weak, the far cheaper approach is to extract an equivalent circuit once and simulate it in the circuit domain alone.

Thermal-Electromagnetic Coupling

Losses raise temperature and temperature changes losses, so the two analyses are inseparable in any design near its thermal limit. Copper resistivity rises about 0.39 percent per kelvin, which increases conduction loss and can push a marginal winding into thermal runaway. Ferrite core loss typically falls to a minimum somewhere between 60 and 100 degrees Celsius depending on the material grade and then rises again, and saturation flux density falls monotonically with temperature toward the Curie point, so a core sized at room temperature may saturate when hot.

A coupled solution computes losses from the electromagnetic model, maps them as volumetric heat sources into a thermal model, solves for the temperature field, updates the temperature-dependent material properties, and repeats until the result stops moving. Two or three iterations usually suffice. The payoff is a prediction of hot-spot temperature rather than average temperature, which is what determines insulation life and core margin.

Parasitic Extraction

Parasitic extraction converts geometry into the lumped elements that explain a circuit's high-frequency behavior. The partial element equivalent circuit method is the workhorse for this task: it partitions conductors into segments, computes partial self- and mutual inductances along with coefficients of potential, and assembles them into a network that a circuit simulator can accept directly. The result is a model that reproduces commutation-loop inductance, busbar current sharing, and the resonances that appear in switching waveforms.

The extracted quantities that matter most in power electronics are the commutation-loop inductance, which sets voltage overshoot at turn-off and which modern module layouts drive below 10 nanohenries; the common-mode capacitance from switching nodes to the chassis or heatsink, which sets the ground-current component of conducted emissions; the interwinding capacitance of isolation transformers, which carries common-mode current across the isolation barrier; and the equivalent series inductance of filter capacitors, which determines the frequency above which a capacitor stops behaving like one. Reduced-order or broadband equivalent circuits fitted to extracted impedance data allow these effects to be simulated at circuit speed while retaining field-level accuracy.

Source Reduction

Suppressing interference where it originates is almost always cheaper than filtering it later. Slowing the gate drive reduces dv/dt and the high-frequency content of the spectrum, but it increases switching loss, so the gate resistor becomes an explicit trade between efficiency and emissions margin. Soft-switching topologies, including resonant and quasi-resonant converters, force transitions to occur at zero voltage or zero current and so remove the fast edge altogether, at the cost of higher circulating current and a narrower operating range.

Spread-spectrum modulation dithers the switching frequency so that harmonic energy spreads across a band rather than concentrating in a line. It is important to understand what this achieves: the technique reduces the amplitude a receiver measures within its resolution bandwidth, which for the 150 kHz to 30 MHz conducted range is 9 kHz, but it does not reduce the total energy emitted. It therefore helps a product pass a limit expressed in a defined bandwidth while doing nothing for a victim whose bandwidth is wider than the dither. Careful dead-time control suppresses the parasitic oscillations that follow reverse recovery, and snubbers damp the ringing that stray inductance and device capacitance would otherwise sustain.

Path Interruption

Filters block the propagation path. A converter's input filter typically combines a common-mode choke, X capacitors across the line for differential-mode noise, and Y capacitors from each line to the protective earth for common-mode noise. The Y capacitance cannot be increased without limit, because it forms a path for mains-frequency leakage current to the chassis, and product safety standards cap the touch current that a device may produce; medical and portable equipment operate under considerably stricter caps than general-purpose equipment. This safety constraint, rather than attenuation theory, usually sets the achievable common-mode corner frequency.

Effective filter design requires knowing the noise's source impedance and the mode it travels in, because an L, pi, or T arrangement attenuates well only when its element impedances mismatch the source and load. Separating conducted emissions into common-mode and differential-mode components with a mode-separating network is therefore the first diagnostic step, since attacking the wrong mode wastes both money and volume. Designers must also verify that the filter's output impedance stays well below the converter's negative input impedance across frequency, the Middlebrook criterion, or the filter and the regulator will form an unstable pair. Damping networks and lossy ferrites address both the stability and the self-resonance problems, since every real filter component ceases to behave ideally above its own resonance.

Layout and Grounding

Layout determines how much field a given current produces, and no filter recovers what a bad layout gives away. The single most valuable rule is to minimize the physical area of high di/dt loops, particularly the commutation loop formed by the input capacitor, the switching devices, and their interconnect, because both the stray inductance and the radiated magnetic field scale with that area. Placing the return path directly beneath the outgoing path on an adjacent layer lets return current flow under the trace at high frequency, shrinking the loop to the dielectric thickness.

Minimizing the area of high dv/dt copper is equally important, since the switch node's own surface acts as a capacitively coupled antenna. Uninterrupted return planes provide low-impedance return paths and reduce coupling between circuits; a slot cut through a plane forces return current to detour and can raise loop inductance dramatically. Sensitive analog and gate-drive circuitry belongs away from switching nodes and magnetic components, with the return current of each isolated gate driver referenced deliberately to its own source terminal. Grounding strategy follows frequency: single-point schemes work at low frequency where connection inductance is negligible, while high-frequency designs need a low-impedance reference structure with short, distributed connections.

Shielding Techniques

Shielding attenuates fields that neither source reduction nor filtering has removed. Total shielding effectiveness is conventionally decomposed into a reflection term, an absorption term, and a correction for multiple internal reflections, all expressed in decibels. Reflection dominates for electric fields and for good conductors at low frequency; absorption dominates at high frequency and grows with thickness measured in skin depths. Low-frequency magnetic fields are the hardest case, because reflection loss is small and a high-permeability alloy such as mu-metal, which diverts flux rather than reflecting it, is usually required.

In practice, apertures and seams govern real enclosure performance far more than material choice. A slot radiates efficiently when its longest dimension approaches half a wavelength, so a long, narrow seam leaks far more than a round hole of equal area; replacing one long slot with a row of small perforations preserves ventilation while raising effectiveness considerably. Conductive gaskets, overlapping seams, and closely spaced fasteners maintain electrical continuity where panels meet. Cable penetrations deserve particular attention, because an unfiltered wire passing through a shield conducts interference straight past it and turns the cable into the structure's most effective antenna. Localized shields, such as an electrostatic screen between transformer windings, intercept displacement current at the source and often outperform a larger enclosure.

EMC Standards Overview

Power electronic products must meet EMC regulations before they may be placed on the market. The International Special Committee on Radio Interference, known by its French acronym CISPR, publishes the emission standards on which most national rules are based. CISPR 11 covers industrial, scientific, and medical equipment; CISPR 32 covers multimedia equipment and replaced the earlier CISPR 22 for information technology equipment; CISPR 14-1 covers household appliances and power tools; and CISPR 25 sets component-level limits for vehicles across 150 kHz to 2.5 GHz, protecting the receivers installed on board. Each standard divides products into classes: Class A applies to commercial and industrial environments, while the Class B limits that govern residential products are stricter, typically by about 10 decibels for radiated emissions and by more than that across parts of the conducted range.

The IEC 61000 series addresses the remaining requirements. IEC 61000-3-2 limits harmonic current for equipment drawing up to and including 16 amperes per phase, IEC 61000-3-12 extends coverage above 16 amperes and up to 75 amperes, and IEC 61000-3-3 limits voltage fluctuation and flicker. The IEC 61000-4 subseries defines immunity tests, including electrostatic discharge in part 4-2, radiated radio-frequency fields in part 4-3, electrical fast transients in part 4-4, surge in part 4-5, and conducted radio-frequency disturbances in part 4-6. In the European Union, conformity is assessed against harmonized standards under the EMC Directive 2014/30/EU; in the United States, unintentional radiators fall under Part 15 Subpart B of the Federal Communications Commission rules. Sector-specific requirements add further obligations, such as IEC 60601-1-2 for medical devices and dedicated railway, marine, and aerospace standards. Identifying the applicable set at the concept stage is far cheaper than discovering it after tooling is committed.

Measurement Fundamentals

Meaningful comparison against a limit requires the prescribed measurement conditions, which CISPR 16 defines. Conducted emissions on the mains port are measured from 150 kHz to 30 MHz through an artificial mains network, commonly called a LISN, that presents a defined and repeatable impedance to the equipment under test while isolating it from the supply. Automotive work uses a different network with much smaller inductance, reflecting the impedance of a vehicle harness rather than a building installation. Radiated emissions are measured from 30 MHz upward on an open area test site or in a semi-anechoic chamber, with the receive antenna scanned in height and the equipment rotated to find the worst case.

Detector choice and resolution bandwidth are as much a part of the limit as the decibel figure. The quasi-peak detector applies charge and discharge time constants that weight a disturbance by its repetition rate, so infrequent impulses read lower than continuous carriers; average and peak detectors serve alongside it, and a result must always be quoted with its detector. Resolution bandwidth is fixed by band: 9 kHz across the conducted range from 150 kHz to 30 MHz and 120 kHz from 30 MHz to 1 GHz. Because a converter's emissions are usually broadband, a measurement taken with the wrong bandwidth cannot be compared with the published limit at all.

Pre-Compliance Testing

Pre-compliance work finds problems while they are still cheap to fix. Near-field probes locate hot spots on a board and distinguish a leaking transformer from a radiating cable; current probes measure the common-mode current on a harness, a quantity that correlates strongly with radiated emissions; and a spectrum analyzer with a LISN gives a conducted scan whose shape, if not its absolute accuracy, matches what an accredited laboratory will report. Comparative measurements are the real value: whether an added component improved a resonance by 6 decibels is a question a bench setup answers reliably even when its absolute uncertainty is several decibels.

Prudent practice targets a margin of roughly 6 decibels below the limit at pre-compliance, absorbing measurement uncertainty, unit-to-unit variation, and the difference between a bench setup and an accredited site. Designers should also leave physical provision for remediation, such as unpopulated filter footprints, ferrite space on cables, and gasket land on an enclosure, because a design with nowhere to add a component leaves only expensive options when a scan comes back marginal.

Certification Process

Formal demonstration of compliance takes place at a laboratory accredited to ISO/IEC 17025, using calibrated instruments, documented test methods, and a defined configuration of the equipment under test. The choice of operating mode, cable arrangement, and load matters enormously, because a product tested in a benign mode may fail in a mode a customer actually uses; test plans should therefore exercise the worst case deliberately.

The route to market differs by jurisdiction and product class. In the European Union the manufacturer issues a Declaration of Conformity supported by a technical file, taking responsibility without mandatory third-party approval for most products. In the United States, unintentional radiators such as switching power supplies follow the Supplier's Declaration of Conformity procedure, while intentional radiators require certification through a Telecommunication Certification Body. Retaining the technical file, test reports, and a record of the tested configuration is a legal obligation, not a formality, and a design change that alters layout, shielding, or filtering can invalidate an existing report and require retest.

Simulation-Driven Design

Analysis pays for itself only when its fidelity matches the question being asked at that stage. Concept work uses analytical models and reluctance networks to size components and compare topologies in minutes. Detailed design brings in two-dimensional and axisymmetric field analysis for winding loss and gap fringing, then three-dimensional analysis for busbars, modules, and enclosures where no symmetry exists. Late-stage work extracts parasitics from the final layout and feeds them back into circuit simulation to predict switching waveforms and the emissions spectrum before the first board is built. After release, the same models serve as diagnostic tools when a field failure or a production variation must be explained.

Model Validation

A simulation result carries no authority until it has been compared with measurement. Impedance analyzers verify extracted inductance and capacitance across frequency; calorimetric or electrical loss measurement verifies predicted core and winding loss; thermocouples and infrared imaging verify predicted temperature rise; and a LISN scan verifies the predicted emissions envelope. Disagreement is informative, since a consistent offset usually indicates a material property or an omitted parasitic rather than a numerical error.

Organizations that maintain a library of validated material models, meshing conventions, and modeling practices reach useful accuracy far faster on subsequent projects than those that rebuild each model from scratch. Equally important is a documented sense of each model's limits: knowing that a given core-loss fit is valid only below a certain flux density, or that an extracted network is trustworthy only below its first unmodeled resonance, prevents the quiet extrapolation that produces confident and wrong answers.

Design Optimization

Once a model is trusted, optimization becomes practical at a scale no prototype campaign could match. Parametric sweeps expose which variables actually drive performance and which merely add complexity, and sensitivity analysis distinguishes a nominal optimum from a design that tolerates manufacturing spread. Multi-objective methods trade efficiency, volume, cost, and emissions margin against one another and return a Pareto front rather than a single answer, which is usually the more honest representation of an engineering choice.

Because full field solutions are expensive to evaluate repeatedly, optimization commonly runs against a surrogate model fitted to a designed set of simulations, with the field solver invoked only to confirm the candidates that survive. The discipline this imposes matters as much as the algorithm: an optimizer will exploit any part of the model that is more optimistic than reality, so constraints must include saturation margin, thermal limits, insulation coordination, and manufacturability rather than efficiency alone.