Conducted vs Radiated Emissions: What Differs, What Fixes Each
Conducted vs radiated emissions compared: frequency ranges, LISN vs antenna, CM and DM noise, and which filters, ferrites and layout fixes work for each test.
Almost every electronic product sold in the EU, the United States, Korea or Japan has to pass two emission tests: conducted emission on its power port (and often on its wired network ports), and radiated emission from the whole product with its cables. Engineers often treat a failure in either one simply as EMI and reach for the same ferrite. That wastes time, because the two tests measure different quantities over different frequency ranges through different coupling paths, and a fix that removes a conducted failure can do nothing for a radiated one. This article compares the two tests, explains why the boundary sits at 30 MHz, and sorts the common countermeasures by the test they actually help.
What each test measures
Conducted emission measures the noise voltage that the product pushes back onto its supply lines. The equipment under test (EUT) is powered through a line impedance stabilization network (LISN), called an artificial mains network in CISPR 16-1-2. The LISN presents a defined 50 Ω / 50 µH impedance to the EUT across the band, so the result does not depend on the building wiring, and it blocks noise arriving from the mains. A receiver connected to the LISN measurement port reads the voltage on each line, live and neutral, in dBµV.
Radiated emission measures the electric field produced by the EUT and its cables at a defined distance, usually 3 m or 10 m, in a semi-anechoic chamber or on an open-area test site. The receiver reads the antenna output in dBµV and adds the antenna factor and cable loss to obtain dBµV/m. To find the maximum, the EUT is rotated and the antenna is scanned in height, typically from 1 m to 4 m, in both horizontal and vertical polarization.
| Aspect | Conducted emission | Radiated emission |
|---|---|---|
| Frequency range (CISPR 32) | 150 kHz to 30 MHz | 30 MHz to 1 GHz, and above 1 GHz up to 6 GHz depending on the highest internal frequency |
| Measured quantity | Voltage at the LISN port, dBµV | Field strength, dBµV/m |
| Receiver bandwidth | 9 kHz | 120 kHz below 1 GHz; 1 MHz above 1 GHz |
| Detectors | Quasi-peak and average | Quasi-peak below 1 GHz; peak and average above 1 GHz |
| Main path | Noise current flowing out on the supply wires | Cables acting as antennas (common mode); PCB loops and enclosure slots at higher frequencies |
| Class B limit example | Mains port: 66 to 56 dBµV QP and 56 to 46 dBµV AV (0.15 to 0.5 MHz), 56/46 (0.5 to 5 MHz), 60/50 (5 to 30 MHz) | 30 dBµV/m QP (30 to 230 MHz) and 37 dBµV/m QP (230 MHz to 1 GHz) at 10 m |
The limits look abstract until you convert them to current. The LISN measurement port is a 50 Ω load, so the 46 dBµV average limit in the 0.5 to 5 MHz band equals about 200 µV, or about 4.0 µA of noise current into 50 Ω. The 56 dBµV quasi-peak limit corresponds to about 12.6 µA. Radiated testing is just as sensitive: a few microamps of common-mode current on a 1 m cable at 100 MHz can exceed the Class B radiated limit, as our article on cable common-mode current works out. Both are microamp tests, which is why small layout details decide the result.
Why the boundary sits at 30 MHz
The split is physical, not administrative. At 30 MHz the wavelength is 10 m. A power cord or a 1 m signal cable is a small fraction of a wavelength and radiates poorly, and a 3 m or 10 m test distance would place the antenna in the near field, where readings depend strongly on position. Measuring the voltage or current that drives the cable is more repeatable. Above 30 MHz the same cables approach resonant lengths; a 1 m cable is half a wavelength long at 150 MHz, so a few microamps of common-mode current produce a measurable field. Below 30 MHz the cable is the conductor that delivers the noise to the victim; above it, the cable is the antenna.
Differential mode and common mode in each test
Differential-mode (DM) noise flows out on one supply line and back on the other. Common-mode (CM) noise flows out on all lines in the same direction and returns through stray capacitance to the reference ground, for example from a switching transistor's heatsink or a transformer's interwinding capacitance. In a conducted measurement the LISN sees both. As a rule of thumb, DM tends to dominate the lower part of the band, up to a few MHz, while CM dominates the upper part. Radiated emission from cables is almost always CM: DM currents in a closely spaced pair cancel at a distance, while CM currents add. This difference is the key to choosing fixes.
Which fixes work for which test
| Countermeasure | Effect on conducted | Effect on radiated | Watch out for |
|---|---|---|---|
| X capacitor across the line (DM) | Strong on DM, especially below a few MHz | Little | Needs safety-rated X-class parts; mounting inductance limits it at high frequency |
| Y capacitors line to earth (CM) | Strong on CM in the upper conducted band | Some, if leads are short and the earth bond has low inductance | Touch-current limits in safety standards cap the capacitance |
| Mains CM choke | Strong on CM from 150 kHz to a few MHz | Often weak above 30 MHz, because winding capacitance bypasses the choke | Check the impedance curve over the whole range you need |
| Clamp-on ferrite on a cable | Small; ferrite impedance is low below a few MHz | Useful from about 30 to 300 MHz, where its impedance is highest | Works only where the circuit impedance is low compared with the ferrite; place it near the connector |
| Slower edges, snubbers, gate resistors | Small on the switching fundamental | Strong on high harmonics and ringing | Costs efficiency or timing margin; check temperatures |
| Spread-spectrum clock or dithered switching | Lowers readings if the spread exceeds the 9 kHz RBW | Same principle with the 120 kHz RBW | Total energy is unchanged; timing-sensitive interfaces may object |
| Shielded enclosure | None by itself; noise leaves on the cable | Strong against board-level radiation | An unfiltered cable penetration defeats it |
A practical diagnosis sequence
- Conducted: record live and neutral separately. Use a CM/DM separator network, or a current probe clamped around both supply wires (reads CM) and around one wire (reads CM plus DM), to find which mode dominates at each failing frequency before choosing an X capacitor, a Y capacitor or a choke.
- Conducted: read the shape. Peaks at multiples of the switching frequency point to the converter; a broad hump in the upper MHz region often comes from CM current through heatsink or transformer capacitance.
- Radiated: clamp a current probe on each cable at the failing frequency and convert the reading to dBµA with the probe's transfer impedance. The cable with the highest CM current is the antenna to fix first.
- Radiated: unplug, or add a ferrite to, one cable at a time. If the emission drops by several dB, that cable is radiating; if nothing changes, examine the enclosure, its slots and the PCB itself with a near-field probe.
- Radiated: read the harmonic spacing. Peaks spaced 25 MHz apart point to a 25 MHz clock or a circuit derived from it; a broadband rise that follows the converter load points to switching edges and ringing.
Common mistakes
- Treating 30 MHz as a wall. A conducted spectrum that is still rising at 30 MHz usually means a radiated problem just above it.
- Ignoring the set-up. In CISPR 32 conducted testing a table-top EUT sits 0.4 m from a vertical reference ground plane and 0.8 m from the LISN. Changing these distances changes the CM stray capacitance and therefore the result, and so does the routing of the cables.
- Forgetting the other ports. CISPR 32 also limits conducted emission on wired network ports such as Ethernet, measured with an impedance stabilization network (ISN) or a current probe, and a product that passes on the mains can still fail there.
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.