How is an EMI filter designed?

First measure or compute unfiltered conducted noise and set the needed attenuation from the margin to the standard's limit (CISPR etc.). Then choose L and C to provide that attenuation over the required band (150 kHz–30 MHz for conducted noise). A single LC stage adds about 40 dB of attenuation per decade above its corner frequency.

Why do filters stop working at high frequency?

Because components have parasitics: inductors have inter-winding capacitance and capacitors have lead and electrode inductance (ESL). A 100 μH inductor with 20 pF self-resonates at 3.6 MHz and a 1 μF capacitor with 15 nH at 1.3 MHz; above those, the inductor acts as a capacitor and the capacitor as an inductor. Calculated attenuation at 30 MHz fell from an ideal 125 dB to 34 dB.

What if attenuation is insufficient?

Adding stages is often more efficient than enlarging L and C. One stage needed about 4.7 times the L and C (473 μH, 4.7 μF) for 60 dB at 150 kHz, while two stages of 100 μH and 1 μF gave 73 dB. At high frequency, add low-ESL capacitors in parallel and lay out parts to prevent input–output coupling.

EMI Filter Design

Category: Electromagnetic Analysis – EMC | Revised 2026-10-01

EMI Filter Design

Theory: insertion loss and parasitics

Overview

🙋

Our power supply's conducted noise exceeds the limit. How should I choose a filter?

🎓

First measure how many dB you're short at which frequencies. Conducted noise covers 150 kHz–30 MHz; in switching supplies, differential-mode noise tends to dominate at low frequency and common-mode at high. An LC filter attenuates about 40 dB per decade above its corner, so L and C follow from the needed attenuation. But that applies to ideal parts; real inductors have inter-winding capacitance and capacitors have lead and electrode inductance. These make the filter nearly ineffective at high frequency, so design for both low-frequency attenuation and high-frequency parasitics.

Insertion loss

$$ IL = 20\log_{10}\left|\frac{V_{L,\,without}}{V_{L,\,with}}\right|,\qquad f_0 = \frac{1}{2\pi\sqrt{LC}} $$

Ratio of load voltage without and with the filter. Source and load are both 50 Ω (the LISN value). A single stage with series L and shunt C on the load side: L = 100 μH, C = 1 μF.

Parasitics and self-resonance

$$ Z_L = \left(\frac{1}{j\omega L} + j\omega C_{p}\right)^{-1},\qquad Z_C = \frac{1}{j\omega C} + j\omega L_{ESL} + R_{ESR},\qquad f_{SRF} = \frac{1}{2\pi\sqrt{L C_p}},\ \frac{1}{2\pi\sqrt{C L_{ESL}}} $$

C_p is inductor winding capacitance (20 pF); L_ESL and R_ESR are the capacitor's equivalent series inductance (15 nH) and resistance (10 mΩ).

Coffee Break Trivia

Standards born to protect radio

International efforts to limit electrical-equipment noise began with CISPR, founded in 1934 to protect the then-spreading radio broadcasts from motor and lighting interference. The LISN inserted in power lines for conducted-noise measurement provides a 50 Ω reference so mains impedance doesn't vary from test to test. Filter performance is often compared on this 50 Ω basis.

Worked examples

Example 1: insertion loss, ideal vs with parasitics (L = 100 μH, C = 1 μF, corner 15.9 kHz)

Frequency	Ideal	With parasitics	Winding capacitance only	ESL only
150 kHz	34.0 dB	34.1 dB	34.0 dB	34.1 dB
1 MHz	65.9 dB	74.3 dB	66.6 dB	73.6 dB
5 MHz	93.9 dB	71.3 dB	94.1 dB	71.1 dB
10 MHz	105.9 dB	53.8 dB	89.1 dB	70.6 dB
30 MHz	125.0 dB	33.6 dB	88.2 dB	70.5 dB

The inductor self-resonates at 3.56 MHz and the capacitor at 1.30 MHz. Near 1 MHz the capacitor approaches series resonance and attenuation actually increases, but above that both parasitics combine and attenuation drops sharply.

Example 2: getting 60 dB at 150 kHz

Configuration	Components	IL at 150 kHz
1 stage (baseline)	100 μH + 1 μF	34.0 dB
1 stage (enlarged)	473 μH + 4.73 μF	60.0 dB
2 stages	two sets of 100 μH + 1 μF	72.7 dB

🙋

Two stages of the same parts beat one stage of big parts.

🎓

One stage's attenuation rises with frequency squared but only in proportion to the LC product, so gaining 20 dB needs ten times the product. Two stages rise with frequency to the fourth power, and identical parts multiply their effect per stage. Large inductors need more turns and gain winding capacitance, hurting high-frequency performance, another reason to prefer more stages. Inter-stage resonance can occur, though, so damping resistors suppress the peaks. And however good the parts, if input and output wiring sit close and couple magnetically or electrically, noise jumps the filter. At high frequency, layout and grounding determine performance.

Design and analysis workflow

Measure unfiltered noise separated into differential and common mode.
Set required insertion loss per frequency from the margin to the limit plus headroom.
Confirm insertion loss with circuit simulation including parasitics.
Damp inter-stage resonances with resistors.
Minimise input–output coupling and ground impedance by layout, then verify by measurement.

Coffee Break Trivia

“A bigger choke made it worse”

A power supply exceeded the limit near 20 MHz. Swapping in a higher-inductance common-mode choke improved low frequencies but worsened 20 MHz. The bigger choke's many turns gave large winding capacitance, lowering self-resonance to a few MHz, so at 20 MHz it acted as a capacitor and passed noise. Adding a small high-frequency choke as a second stage brought 20 MHz within limits.

Common mistakes

Mistakes and fixes

Mistake	Effect	Fix
Designing with ideal parts	Ineffective at high frequency	Include parasitics
Just enlarging L	Self-resonance drops	Add stages
Not separating modes	Wrong countermeasure	Separate DM and CM
Ignoring inter-stage resonance	Worse at certain frequencies	Add damping
Input and output wiring close	Noise bypasses filter	Separate the layout

🙋

I'd like to learn related topics.

🎓

Rate this article

Thank you for your feedback!

Helpful

More details

Report error

Helpful

More details

Report error

Written by the NovaSolver Editorial Team
Built with engineers and AI (editorial policy) — Sitemap

About the Authors

EMI Filter Design

Theory: insertion loss and parasitics

Overview

Insertion loss

Parasitics and self-resonance

Standards born to protect radio

Worked examples

Example 1: insertion loss, ideal vs with parasitics (L = 100 μH, C = 1 μF, corner 15.9 kHz)

Example 2: getting 60 dB at 150 kHz

Design and analysis workflow

“A bigger choke made it worse”

Common mistakes

Mistakes and fixes

Related Topics

Related fields