Power Line EMI Filters and the LISN: Selection and Placement
How the LISN loads a mains EMI filter, what filter datasheets leave out, where to mount and bond the filter, and how to read conducted scans for filter clues.
When a product fails conducted emission, the usual fix is a power line EMI filter: an X capacitor, a common-mode choke and a pair of Y capacitors, sometimes in a metal can or built into an IEC inlet. Whether that filter delivers the attenuation its datasheet promises depends less on the part number than on two things the datasheet cannot know: the impedances on each side of it, and how it is mounted.
What the LISN presents to your filter
In a CISPR conducted emission test the EUT is powered through a V-network LISN (an artificial mains network in CISPR 16-1-2). Seen from the EUT terminal, each line looks into 50 Ω ∥ (50 µH + 5 Ω) to the reference ground: the 50 Ω is the receiver port, and the 50 µH inductor plus 5 Ω lead back toward the mains, which the LISN decouples with large capacitors. The magnitude of this impedance is about 32.7 Ω at 150 kHz, 47.2 Ω at 500 kHz and 49.3 Ω at 1 MHz, and it is practically 50 Ω above a few megahertz.
The termination is different for the two noise modes. Common-mode (CM) current returns through the line and neutral ports in parallel, so the CM section of the filter works into about 25 Ω. Differential-mode (DM) current flows out through one port and back through the other, so the DM section sees about 100 Ω. These, not the impedance of the building wiring, are the loads the filter works against in the test lab.
What a filter datasheet does and does not tell you
Filter insertion loss is normally measured by the method of CISPR 17 in a 50 Ω / 50 Ω system, with separate curves for common mode and differential mode. That is useful for comparing parts, but neither side of a real installation is 50 Ω. The DM noise source of a switch-mode power supply is its bulk capacitor, a low impedance of a fraction of an ohm. The CM source is the switching node coupling through a few tens of picofarads to a heatsink or chassis, a high impedance. On the other side sits the LISN with about 25 Ω for CM and 100 Ω for DM.
A useful rule is that a shunt capacitor works well when it faces a high impedance, and a series inductor works well when it faces a low impedance. The CM source is a high impedance, so the Y capacitors are most effective on the equipment side of the choke, which is where a typical single-stage mains filter places them. If you rearrange the filter or add a stage, keep this matching in mind instead of simply adding components.
| Datasheet item | What it tells you | What it does not tell you |
|---|---|---|
| CM and DM insertion loss (50 Ω / 50 Ω) | Relative performance of different filters | Attenuation with your power supply as the source and the LISN as the load |
| Rated current and ambient temperature | The current at which the parts stay within their temperature rating | How much DM inductance remains at full load current |
| Choke inductance, X and Y capacitor values | Where the natural frequencies of the filter lie, which you can estimate with an LC calculator | Winding capacitance and capacitor ESL, which limit attenuation in the upper part of the conducted band |
| Leakage current at a stated voltage and frequency | The filter's own contribution to touch current | The total with the other Y capacitors in the product |
Placement: where the filter earns its attenuation
A filter attenuates only the noise that has to pass through it, and two installation faults let noise go around it. The first is a long ground connection. A rule of thumb for a straight wire is about 1 nH per millimetre, so a 5 cm green-yellow lead from the filter case to the chassis adds about 50 nH. That is 3.14 Ω at 10 MHz and 9.42 Ω at 30 MHz, while a 2.2 nF Y capacitor has a reactance of 7.23 Ω at 10 MHz and 2.41 Ω at 30 MHz. The capacitor and the lead form a series resonance at about 15.2 MHz, and above it the CM shunt path becomes inductive and loses its effect. Bonding the metal case directly to bare chassis metal, leaving perhaps 5 nH, moves that resonance to about 48.0 MHz, above the conducted band.
The second fault is coupling between the unfiltered and the filtered wiring. If the input wires run alongside the output wires, past the switching transistor or across the noisy part of the board, capacitive and magnetic coupling carries noise around the filter, and its high-frequency attenuation is then limited by that coupling rather than by its components. The cleanest arrangement is a filter mounted in the enclosure wall, such as a filter with an IEC inlet or a bulkhead-mounted can, so that the unfiltered side stays outside the shield.
- Mount the filter where the power cord enters, not next to the power supply it protects.
- Bond the case to unpainted chassis metal over its full flange, or with screws through conductive surfaces; do not rely on a wire.
- Keep input and output wires on opposite sides of the filter and route them close to the chassis, not across the board.
- For a board-mounted filter, place it at the board edge next to the connector, keep the switching circuit away from it, and do not run other traces under or beside the choke.
Reading a conducted scan for filter clues
Before changing parts, take three reference scans: the EUT with the filter, the EUT with the filter bypassed or removed, and an ambient scan with the EUT switched off. The difference between the first two is the real insertion loss of the filter in your set-up, and the ambient scan confirms that the LISN is not letting mains noise through. Measure both line and neutral, and keep the set-up of the standard, for example 0.4 m to the vertical reference plane and 0.8 m to the LISN for a table-top EUT under CISPR 32.
| Symptom | Likely cause | First check |
|---|---|---|
| Good attenuation below 1 MHz, little effect above 10 MHz | Long ground lead or coupling around the filter | Shorten the bond, separate input and output wires, sweep a near-field probe across the filter |
| A new peak appears just above the natural frequency of the filter | Undamped LC resonance of the filter with the source and the LISN | Estimate the natural frequency; add an RC damping network or use a lossier choke |
| Passes at light load, fails at full load | DM inductance falling with current, or more ripple at higher power | Measure at the worst-case load; check the choke current rating and its DM inductance under bias |
| Line and neutral differ by several dB | Both CM and DM contribute, or the filter is unbalanced | Separate the modes with a CM/DM separator or a current probe |
Safety limits that cap the filter
Y capacitors connect the mains to protective earth, so their 50 or 60 Hz current flows as touch current. A 4.7 nF Y capacitor from line to PE passes about 0.34 mA at 230 V and 50 Hz, and a 2.2 nF capacitor about 0.16 mA. The allowed total depends on the safety standard and the product class, for example IEC 62368-1 for IT and AV equipment or IEC 60601-1 for medical equipment, where patient-connected parts are far stricter. This often forces a smaller Y capacitance than EMC would like, and the missing CM attenuation then has to come from more choke inductance or from a weaker CM source, for example by tying the heatsink to the primary DC return.
Common mistakes
- Choosing a filter by its 50 Ω insertion loss curve alone and expecting the same dB in the product.
- Installing the filter at the power supply end of a long internal cable, so that the cable between the inlet and the filter picks up noise again.
- Adding a second filter stage without checking the new resonance, which can raise emissions at one frequency.
Calculators used in this post
More posts
Results and summaries are for reference. For certification and test reports use the latest official standard text and calibrated instrument data.