Minimizing EMI in High-Performance Power Conversion Systems

Every switching power converter is a conducted and radiated noise source. That is not a flaw in the design, it is an inherent consequence of rapidly switching current through inductive and capacitive elements. The engineering task is not to eliminate switching noise but to contain it below the thresholds set by regulatory standards and, more immediately, below the levels that degrade system performance.

This article covers the mechanisms that generate EMI in switching converters, the filter architectures that suppress it, and the layout and grounding practices that prevent it from defeating your filters in the first place.

The Two EMI Problems Are Not the Same

Conducted EMI travels back through the power input leads and into the source distribution network. Radiated EMI propagates through the air and couples into nearby circuits or antennas. They share root causes but require different mitigation strategies.

For conducted emissions, regulatory limits in CISPR 22 (information technology equipment) and CISPR 11 (industrial equipment) define allowable voltage noise on the AC mains or DC input lines from 150 kHz to 30 MHz. For radiated emissions, limits apply from 30 MHz to 1 GHz or beyond.

Most EMI problems in DC/DC converters show up first in conducted emissions testing, because the switching frequency and its lower harmonics fall squarely in the 150 kHz, 30 MHz conducted band. Fix conducted emissions well and radiated emissions often follow. The reverse is not always true.

Where the Noise Is Generated

Common-Mode vs. Differential-Mode Noise

Differential-mode (DM) noise appears between the positive and negative supply conductors. It is generated by the switching current that flows in the converter's power loop, from input capacitor, through the switch, through the inductor, back to the capacitor. The size of this loop and its associated inductance determine how much voltage is induced across the loop each switching cycle.

Common-mode (CM) noise appears on both supply conductors simultaneously with respect to chassis or earth ground. It is generated primarily by the dV/dt on switching nodes, drain of a MOSFET, or the switching node of an inductor, capacitively coupling through to chassis through stray capacitances in the layout, transformer interwinding capacitance, and heatsink mounting capacitance. CM noise is typically harder to suppress than DM noise because its coupling paths are parasitic and difficult to characterize precisely.

In a practical converter, you always have both. Most EMI filter designs target DM and CM components separately, with X-capacitors handling DM and common-mode chokes with Y-capacitors handling CM.

EMI Filter Architecture

A standard two-stage EMI filter for a DC/DC converter input includes:

A common-mode choke (high impedance to CM currents, low impedance to differential supply current)
Y-capacitors from each supply rail to chassis (providing a low-impedance return path for CM noise current, keeping it off the supply leads)
X-capacitor across the supply rails (reducing DM ripple voltage)
A second-stage LC filter if the converter's switching noise is particularly aggressive

The common-mode choke is the most sensitive component to get right. Its CM inductance must be high enough at the converter's fundamental switching frequency to present meaningful impedance, but its leakage inductance (which appears in the differential-mode path) must be controlled. Leakage inductance that is too high creates a series resonance with the X-capacitor that amplifies noise at the resonant frequency.

For a 100 kHz switching converter, a CM choke with 1, 4 mH CM inductance is a typical starting point. Verify that the choke's rated current exceeds your maximum DC operating current with margin, CM chokes saturate, and a saturated choke provides no EMI attenuation at all.

Layout: Where Most Filters Fail in Practice

An EMI filter that looks correct on a schematic can be completely ineffective on a PCB. The three most common failure modes are:

Input and Output Traces Routed in Parallel

If the converter's noisy output traces run parallel to and close to the filtered input traces, the noise couples directly from output to input by mutual inductance and capacitance, bypassing the filter entirely. The fix is physical separation and orthogonal routing where separation is not possible.

Y-Capacitors Without a Low-Impedance Chassis Connection

Y-capacitors divert CM noise current to chassis ground. If the chassis connection is made through a long PCB trace or a wire with significant inductance, the connection is not low-impedance at the frequencies of interest. At 10 MHz, 10 nH of inductance has an impedance of 628 mΩ, which is significant compared to a 1 nF Y-capacitor's impedance of 16 Ω at the same frequency. Use multiple vias in parallel to chassis ground connections and keep traces short.

Filter Components Placed After Rather Than Before the Noisy Switching Node

Power converter PCB layout must treat the EMI filter as a clean-to-noisy boundary. Components on the input side of the filter should be physically separated from the switching components of the converter. Any trace that connects across this boundary without going through the filter defeats its purpose. This means the input filter capacitor must be located close to the input pins of the converter module, not close to the external filter.

Spread-Spectrum Clocking: Benefits and Limitations

Spread-spectrum frequency modulation (SSFM) reduces the peak spectral amplitude of switching harmonics by distributing energy across a band rather than concentrating it at a single frequency. A converter that produces a 60 dBµV peak at 100 kHz with a fixed clock might produce a 50 dBµV peak spread across 90, 110 kHz with SSFM, a 10 dB reduction in peak amplitude.

SSFM is genuinely useful for approaching conducted emissions limits when the fundamental or a lower harmonic is failing by a small margin. It is not a substitute for good filtering. If you are 20 dB over the limit, SSFM will not save you. It also complicates synchronization with other clocks in the system, and some applications (medical, precision timing) cannot tolerate a varying switching frequency.

Ferrite Beads as a Supplemental Tool

Ferrite beads are resistive at their target frequency range, not reactive. A bead specified at 600 Ω at 100 MHz provides that impedance through loss, not inductance. This makes them effective for high-frequency conducted noise (above approximately 10 MHz) where they absorb energy rather than reflecting it back to the source.

Do not use a ferrite bead as a primary EMI filter element at the converter's switching frequency. At 100, 500 kHz, most ferrite beads are still predominantly inductive and provide little attenuation. Their value is in suppressing high-order harmonics and fast-edge transients, not the fundamental.

Measurement Before Commitment

Pre-compliance testing with a LISN (line impedance stabilization network) and a spectrum analyzer is strongly recommended before final filter design commitment. Real EMI measurement requires a defined source impedance for the input side, which the LISN provides. Without it, your filter response is dependent on the source impedance of your bench supply, which is not the same as the impedance seen in end use.

A two-day pre-compliance test session before final PCB spin is far cheaper than a failed full compliance test and a board respin.

Closing Thoughts

EMI management in power conversion is a systems discipline. The filter, the layout, the converter topology, the chassis connection, and the grounding architecture all interact. Treating any one of them in isolation produces a design that is full of surprises on the test bench. Get the full picture first.