In the previous article I described how LCL topology, active damping and modulation strategies together shape the EMC budget of a hybrid inverter. On the schematic, filter and common-mode choke are each a single symbol with two or three values. In reality, the magnetic components are the filter. Anyone who treats their design as a procurement task and specifies only inductance, current and saturation limit gets exactly the kind of surprises in the test lab that the clean topology decision was meant to avoid.

Six design decisions in magnetics that make the difference between a filter that works on paper and one that works in the field.

1. Core Material Selection: No Default Path

Four material families compete in a 15 kW hybrid inverter for three very different positions: the boost inductor on the DC side, the LCL filter inductors on the AC side, and the common-mode choke ahead of them.

MnZn ferrites become particularly attractive at switching frequencies above roughly 50 kHz, where their loss scaling works in their favor. They remain usable in the 16 to 32 kHz range typical for grid-tied inverters, but already saturate around 0.4 T and lose further saturation flux density as temperature rises. For a 15 kW hybrid inverter at these frequencies, a ferrite-based design grows large in volume quickly.

Powder cores made from iron-silicon-aluminium or iron-nickel-molybdenum alloys carry their air gap distributed through the material and tolerate high DC bias. This makes them ideal for boost and PFC inductors. Their specific losses are higher than for ferrite, but with correct material selection the trade-off works out.

Nanocrystalline cores with permeabilities typically between 30,000 and well over 100,000, low core losses and saturation flux densities around 1.2 T are practically without alternative for common-mode chokes, because high CM inductance in a compact form factor can only be achieved through extreme permeability.

Amorphous ribbon cores sit between these and occasionally show up in boost inductors when powder cores reach their thermal limits.

Anyone who delegates this choice to a supplier who “just picks what works” gives away on average five to fifteen percent in volume or efficiency, often both.

2. Steinmetz Is Not Enough: Loss Calculation Under PWM Excitation

The classic Steinmetz equation P_v = k · f^α · B^β describes core losses under sinusoidal excitation. In an inverter, however, the flux waveform is triangular with a superimposed fundamental, so fundamentally non-sinusoidal. For real designs, the simple Steinmetz formula typically underestimates losses by ten to fifty percent, depending on waveform shape and modulation depth.

The Improved Generalized Steinmetz Equation (iGSE) integrates the loss density over the instantaneous dB/dt waveform and delivers significantly more accurate results for PWM excitation. Prerequisite: reliable material data in the relevant frequency and flux density range. Datasheets often provide only one loss curve at sinusoidal excitation, which is not enough.

In practice this means: every new material choice requires validation of the loss parameters through dedicated measurements, or a very serious discussion with the core manufacturer. Underestimated losses lead to thermal hotspots that do not appear in the datasheet but slowly degrade the component in the field.

3. Winding Technique: Skin and Proximity Losses Are Not Optional

The skin depth in copper is roughly 0.52 mm at 16 kHz, and only 0.30 mm at 50 kHz (both at room temperature, roughly ten percent higher at 80 °C operating temperature). Solid round wires above this skin depth no longer make full use of their cross-section. The ohmic winding loss rises disproportionately with frequency.

The answer is litz wire with individual strand diameters below twice the skin depth at the highest relevant current frequency. For 16 kHz that means individual strands below 1 mm, in practice 0.1 to 0.2 mm to also keep proximity losses between adjacent layers under control.

Proximity effects are often the actual driver. In poorly dimensioned multilayer windings, adjacent layers induce eddy currents that can drive the effective resistance to five to ten times the DC value. In well-designed litz wire windings, the AC-to-DC resistance ratio typically stays between 1.2 and 2. Dowell’s equation provides an analytical estimate that can be parameterized in MATLAB and iterated over the winding geometry. Final verification is done by measurement on the prototype.

4. Air Gap: Every Strategy Has Its Quirks

In ferrite cores with a concentrated air gap, stray flux emerges that induces eddy current losses in adjacent winding layers (fringing loss). At high switching frequencies this loss component can exceed the core losses themselves.

Three counter-strategies:

• Distributed gap: several small air gaps instead of one large gap reduce the local stray flux density

• Powder cores: the air gap is distributed throughout the material, stray flux is negligible, but core losses are higher

• Winding distance from the air gap: setting back the first layers several millimeters from the gap, at the cost of winding window area

The choice depends on DC bias and permitted overall height. For a boost inductor with high DC content, a powder core is often the most robust choice; for the LCL inductors on the AC side, a ferrite with distributed gap can be more efficient.

5. Common-Mode Choke: A Design Problem in Its Own Right

The common-mode choke follows different design rules than the differential-mode inductor. The goal is high impedance for common-mode currents combined with negligible differential-mode inductance. This only works with symmetric winding on a high-permeability core.

Saturation here is not caused by the load current itself, but by its asymmetric components. A nominally symmetric three-phase current with small asymmetry can drive the CM core into saturation unnoticed, because small differential components on the CM path integrate up over time. The design must account for this worst case, not for the ideal symmetric scenario.

Material choice: nanocrystalline ahead of amorphous ahead of MnZn ferrite. Permeabilities above 30,000 are required to realize CM inductances of several millihenries in a reasonable form factor.

Winding strategy: bifilar or trifilar winding with minimum leakage inductance. Every microhenry of leakage inductance translates into differential-mode inductance and shifts the LCL filter resonance in an uncontrolled way.

6. From Calculation to Production Component

Anyone not using a commercial FEM suite leans more heavily on analytical models and measurement instead of simulation. That is not necessarily worse, provided the analytical depth is sound and the measurement effort is honestly planned.

The workflow that works for us:

1. Analytical first design in MATLAB. Saturation condition, Steinmetz losses with iGSE extension for PWM excitation, Dowell winding losses across number of layers and frequency, simple thermal model with thermal resistances.

2. Specification document to the magnetics supplier with unambiguous measurement criteria for first samples: small-signal impedance over frequency, AC resistance at switching frequency, thermal behavior under full load.

3. First samples in our own lab: impedance analysis across the relevant frequency range, thermographic measurement under load, EMC pre-screening on the inverter.

4. Iteration based on measured values, not on adjusted assumptions.

5. Series production sampling against a fixed list of agreed measurement criteria, continuously.

The strength of this approach is that every iteration rests on a measurement, not on a simulation whose material parameters are often questionable anyway. The weakness: you cannot afford a cheap first design, the first samples must already be close to target. That shifts diligence to the front end, into the analytical phase, where it costs less than a late iteration round on the prototype.

7. What This Means for System Design

Magnetics design is the point at which the topology decisions discussed in the first article become physical. A SiC decision delivers the promised efficiency gain only if the inductors can handle the higher edge rates and frequency content. Active filter damping works only if the leakage inductances of the inductor do not shift the calculated resonance. A carefully chosen modulation strategy reduces common-mode steps effectively only if the CM choke does not saturate from unrecognized asymmetries.

Anyone who treats magnetics as a procurement topic pushes the risk into the validation phase. Anyone who treats magnetics as a design discipline and iterates jointly with the supplier closes the gap between schematic and test lab.

In the ampareq Gen3 program at awb-it, magnetics design is an integral part of hardware development, not a downstream specification. That moves effort to the front end and reduces the iterations that would otherwise inevitably arrive under time pressure.

If you are working on magnetic design questions in your own development or have gathered experience with alternative material choices, I would be glad to exchange views in the comments or via direct message.