EMI Filter Insertion Loss Calculator
EMI Filter Insertion Loss Calculator
What a built filter really does, once the capacitor’s ESL and the choke’s winding capacitance are in the model and the impedances are yours rather than a 50 ohm jig’s. An ideal LC rises at 40 dB per decade for ever; a real one stops somewhere between one and thirty megahertz and comes back down. This page prints the ideal figure, the real one, the 50 ohm / 50 ohm figure a datasheet would quote, and the difference between them.
Real insertion loss, parasitics and all
the filter the designer page produces — a 1 mH choke with 1% leakage, 50 mohm of winding resistance and 10 pF of winding capacitance, a 4.7 uF X capacitor with 5 mohm of ESR and 1.5 nH of ESL — in differential mode at 10 MHz, between 100 ohm and 100 ohm
The two-port, and the two parasitics that end it
IL = 20·log10 |(A·Zl + B + C·ZsZl + D·Zs) / (Zs + Zl)|
capacitor self-resonance f = 1 / (2π√(C·Lesl)) choke self-resonance f = 1 / (2π√(L·Cw))
reversing the filter ≡ exchanging Zs and Zl
- L esl
- the capacitor’s series inductance, set by its package and its mounting and not by its capacitance. Above the resonance it makes, the capacitor is an inductor and a bigger value makes things worse rather than better
- C w
- the choke’s winding capacitance. Above the resonance it makes, current goes straight past the winding and the choke is not there
- Z s, Z l
- the impedances either side. Insertion loss is not a property of a filter; it is a property of a filter in a circuit, and CISPR 17’s 50 ohm / 50 ohm is one circuit among many
- ABCD
- the two-port’s chain matrix, here a shunt admittance followed by a series impedance. Checked against a nodal solve and against a time-domain simulation of the same network
Worked example
the filter the designer page produces — a 1 mH choke with 1% leakage, 50 mohm of winding resistance and 10 pF of winding capacitance, a 4.7 uF X capacitor with 5 mohm of ESR and 1.5 nH of ESL — in differential mode at 10 MHz, between 100 ohm and 100 ohm
An ideal 10 uH and 4.7 uF between 100 ohm and 100 ohm would give 99.46 dB at 10 MHz. That is the number the 40 dB per decade rule leads you to expect, and it is wrong
The X capacitor's 1.5 nH of package and mounting inductance puts its series self-resonance at 1.896 MHz. Above that it is an inductor: at 10 MHz its impedance is 90.86 mohm and RISING, where an ideal 4.7 uF would have been 3.386 mohm and falling
The choke's 10 pF resonates against the 10 uH of leakage at 15.92 MHz, so at 10 MHz the winding is still inductive — in common mode, against the full 1 mH, the same 10 pF would have resonated at 1.592 MHz and the choke would already be finished
Solving the real network gives 75.2 dB, which is 24.3 dB less than the ideal model promised. That gap is not a safety factor — it is the difference between a filter that works and a retest
In a 50 ohm / 50 ohm jig, which is what CISPR 17 specifies and what a datasheet curve shows, the same filter measures 75.14 dB — 0.03 dB from the figure in your own impedances
Feed these components into the designer page's check mode and it returns the same numbers, because both pages solve the same network with the same expressions
The same 10 uH and 4.7 uF filter, four ways
| Frequency | Ideal | Real, ceramic X cap at 1.5 nH | Real, film X cap at 20 nH |
|---|---|---|---|
| 1 MHz | 64.8 dB | 67.5 dB | 56.1 dB |
| 2 MHz | 73.5 dB | 83.6 dB | 50.8 dB |
| 5 MHz | 87.7 dB | 73.0 dB | 49.3 dB |
| 10 MHz | 99.5 dB | 75.2 dB | 52.4 dB |
| 30 MHz | 118.4 dB | 62.4 dB | 39.9 dB |
Where the insertion loss goes, and what to do about it
| Cause | Symptom on the curve | What actually helps |
|---|---|---|
| Capacitor ESL | The real curve turns over at the capacitor’s self-resonance and comes back down | A lower-inductance package, several smaller capacitors in parallel, a shorter mounting loop. Not more capacitance — that lowers the resonance |
| Choke winding capacitance | The choke stops contributing; the curve flattens | Fewer turns on more core, a sectioned winding, or a separate small inductor or ferrite chosen for the decade you need |
| The impedances either side | A constant offset, and sometimes a peak where the filter resonates against them | Choosing the topology to face the inductor at the low impedance; damping the peak |
| Input-to-output coupling across the filter | Not on this curve at all — it cannot be, it is not in the model | Physical separation, a screen between the two sides, and not running the input and output conductors together |
| A chassis bond made with a wire | Not on this curve either; it is in series with the Y capacitors | A bonded face, or the shortest and widest strap the mechanical design allows |
Eighty decibels on paper, thirty-five in the chamber
An ideal second-order filter attenuates at 40 dB per decade for ever. Plot it and the line goes up and off the top of the chart, and by 10 MHz a modest 10 microhenry and 4.7 microfarad filter is promising a hundred decibels. No filter has ever done that, and the reason is not subtle: the components stop being the components.
The capacitor becomes an inductor. Every capacitor has a series inductance set by its package and its mounting — the path the current takes through the terminations, the lands and the vias. KYOCERA AVX measured 0.87 nH for a 0603 and 1.2 nH for a 1206, and the mounting loop usually adds as much again; a leaded film part on a 15 mm pitch is well over ten nanohenries. That inductance resonates with the capacitance, and above the resonance the impedance rises with frequency instead of falling. A 4.7 microfarad ceramic with 1.5 nH resonates at about 1.9 MHz, which is inside the band a conducted-emissions limit covers. Above it the part is an inductor, and making it a bigger capacitor moves the resonance DOWN and makes the high-frequency behaviour worse.
The choke becomes a capacitor. The same thing in reverse. A winding has capacitance between its turns, a few picofarads to a few tens, and above the resonance it makes with the inductance the current goes straight past the winding. A 1 mH common-mode choke with 10 pF has finished by about 1.6 MHz. In differential mode the same choke resonates far higher, because the differential inductance is only the leakage — which is one of the few places in this subject where the leakage does you a favour.
And the number was never a property of the filter anyway. Insertion loss is defined as the ratio of what arrives at the load without the filter to what arrives with it, so the source and load impedances are inside the definition. CISPR 17 specifies 50 ohm on both sides, which is why every filter datasheet curve is a 50 ohm curve and why they are all comparable with each other. Your circuit is not a 50 ohm jig. A converter’s differential noise looks more like a current source than a voltage source; a vehicle or spacecraft bus is a fraction of an ohm; a regulating converter presents a negative incremental resistance at its input. Move either impedance and the same components give a different number — which is why this page prints both, and the difference between them.
What the model still cannot see. Two things, and above roughly 30 MHz they usually dominate. The first is coupling across the filter: if the input and output sides can see each other — a shared ground trace, conductors running side by side, no screen between them — then energy goes round the filter rather than through it, and no amount of component quality helps. The second is the chassis bond. The Y capacitors’ return path is that bond, and a bond made with a wire is an inductor in series with them. Both are layout, and both are why a filter with an impeccable insertion-loss curve can still fail. Treat this page’s number as a ceiling on what the components can give you, and the layout as what decides how much of it you keep.
The components themselves come from the filter designer, which solves this same network with these same expressions and therefore agrees with this page; the attenuation you need comes from the margin page. Whether the differential stage will destabilise the converter behind it is a different question with its own page: the EMI input filter and Middlebrook check.
Frequently asked questions
Why does my 80 dB filter only give 35 dB?
Almost always one of three things. Above the X capacitor’s self-resonance the capacitor is an inductor and the roll-off stops; above the choke’s self-resonance the winding capacitance carries the current straight past it; and the datasheet figure was measured in a 50 ohm jig, which is not your circuit. This page separates the three: the ideal figure, the real figure in your impedances, and the real figure at 50 ohm.
What is CISPR 17 and why does everyone quote 50 ohm?
CISPR 17 is the IEC standard for measuring the suppression characteristics of passive EMC filtering devices. It defines the asymmetrical, symmetrical and unsymmetrical test circuits with impedances referenced to 50 ohm, which is what a network analyser presents and what makes one manufacturer’s curve comparable with another’s. It was never a claim about your circuit; the worst-case impedance methods of its first edition were dropped in the second because industry had stopped using them.
Can I get the attenuation back by using a bigger capacitor?
Below the self-resonance, yes. Above it, no — and it makes things worse, because for a given package a bigger capacitance resonates lower and the useful band gets shorter. What helps above the resonance is lower inductance: a smaller or reverse-geometry package, several capacitors in parallel, a shorter mounting loop, a feedthrough capacitor mounted in the wall of the box, or a second filter stage designed for the decade you are short in.
Why is the real curve sometimes ABOVE the ideal one?
Because near its series self-resonance a real capacitor has a lower impedance than an ideal one of the same value — the ESL and the capacitance cancel and only the ESR is left. So it shunts better there. The effect is narrow, it depends on an ESR that moves with temperature and part-to-part, and it reverses immediately above the resonance. It is not something to design to.
Which impedances should I actually use?
The ones your circuit has, and if you do not know them, several sets. The honest approach is to bracket: run the page at 50/50 to compare with datasheets, at your best estimate, and at the worst case you can defend. If the answer changes by 20 dB across that range, you have learned something more useful than any single number would have told you.
Does this tell me whether I will pass the test?
No. It estimates insertion loss for a component model in stated impedances. It has no knowledge of your layout, your chassis bonding, coupling across the filter, the measurement bandwidths and detectors the standard specifies, or the operating points your test plan requires. Use it to choose components and to know where the component model stops being the thing that decides.
Related calculators
References
- CISPR 17 Edition 2.0, 2011-06, Methods of measurement of the suppression characteristics of passive EMC filtering devices. Clause 5 covers insertion loss; the asymmetrical (common-mode), symmetrical (differential-mode) and unsymmetrical test circuits are defined in 3.1.14, with impedances referenced to 50 ohm in 3.1.8. The introduction records that the first edition’s worst-case methods were deleted as no longer used in industry. Verified from the standard’s published preview; copyrighted, so no values are reproduced.
- KYOCERA AVX, Parasitic Inductance of Multilayer Ceramic Capacitors, Cain J. Measured parasitic inductance by case size on 0.1 uF samples: 0603 870 pH, 0805 1050 pH, 1206 1200 pH, 1210 980 pH, and reverse-geometry 0612 610 pH and 0508 600 pH. The ESL defaults and the case-size table on this page come from it.
- Ott HW. Electromagnetic Compatibility Engineering. Wiley, 2009. Chapter 4, 4.2.2 Parasitic Effects in Filters; Chapter 5 Passive Components, 5.1 Capacitors (including 5.1.4 Feed-Through Capacitors and 5.1.5 Paralleling Capacitors) and 5.2 Inductors; Chapter 13, 13.3.4 Filter Mounting and 13.3.6 High-Frequency Noise. Section titles checked against the author’s own published contents listing.
- Würth Elektronik, 1-Phase Line Filter Design, application note ANP015, 2024-06-03. The design side of the same network, and the source of the convention that the differential inductance is read off the choke’s own differential-mode impedance curve rather than assumed.
- MIL-STD-461G, Requirements for the Control of Electromagnetic Interference Characteristics of Subsystems and Equipment, 11 December 2015, superseding MIL-STD-461F. A work of the US Government, distributed without charge. Paragraph 5.5.1 makes CE102 applicable from 10 kHz to 10 MHz on all power leads including returns that take power from a source outside the equipment, and Figure CE102-1 gives the limit: 94 dBµV at 10 kHz, falling at 20 dB per decade to 60 dBµV at 500 kHz and flat to 10 MHz. That band is what this page’s chart covers. Read from the standard itself; the curve is implemented on the margin page.