Ferrite Beads and CM Chokes: Impedance, DC Bias and Placement
How to choose ferrite beads and common-mode chokes: reading R, X and Z curves, the impedance divider, DC bias derating, saturation and where to place them.
Ferrite beads and common-mode chokes are the most common parts in an EMC fix, and the most often misused. A bead chosen only by its headline value, such as 600 Ω at 100 MHz, can give almost no attenuation in the real circuit, and a choke can fail because of the wrong core material or winding style. This post explains how to read the data and decide what to use and where.
What a ferrite actually does
A ferrite bead is not a simple inductor. Its impedance Z has a reactive part X, from the inductance of the conductor through the ferrite, and a resistive part R, from losses in the ferrite material. At low frequencies X dominates and the bead behaves like a small inductor that stores energy. Above a crossover frequency R dominates and the bead turns noise current into heat instead of reflecting it. Above its self-resonance, set by parasitic capacitance between the terminals, the bead becomes capacitive and its impedance falls again.
The resistive region is the useful one for EMC: an inductive element can resonate with nearby capacitance and move the noise to a new frequency, while a resistive element dissipates it. Check that R, not only |Z|, is high at the frequency of the problem.
Reading the impedance curve
A good datasheet plots |Z|, R and X against frequency. The single value on the front page is |Z| at one frequency, usually 100 MHz, measured with no DC current. Check where the R and X curves cross, how wide the region with high R is, and where |Z| peaks and starts to fall. Two beads with the same 600 Ω rating can differ widely at 30 MHz or at 1 GHz.
The impedance divider: why the circuit matters
A bead or choke is a series element, so its effect depends on the source and load impedances around it. With purely resistive values, the insertion loss is approximately IL = 20 · log10((Zs + Zb + Zl) / (Zs + Zl)), where Zs is the source impedance, Zb the bead impedance and Zl the load impedance.
In a 50 Ω system with a 50 Ω source and a 50 Ω load, a 600 Ω bead gives 20 · log10(700 / 100) = 16.9 dB. Between a 1 Ω source and a 1 Ω load, typical of a power rail with decoupling, a bead of only 100 Ω gives 34.2 dB. Between a 5 kΩ source and a 5 kΩ load, the same 100 Ω bead gives 0.1 dB, practically nothing. Series elements work in low-impedance circuits and shunt capacitors in high-impedance circuits, so a bead on a high-impedance signal line does little unless it is combined with a capacitor to ground.
The same logic applies to a clamp-on ferrite on a cable. If the common-mode loop formed by the cable has an impedance of about 150 Ω (an example value; it varies widely with cable length and frequency), a core adding 200 Ω gives 20 · log10(350 / 150) = 7.4 dB. Passing the cable twice through the core raises the impedance roughly four times well below the core's resonance; with 800 Ω the estimate rises to 16.0 dB. At higher frequencies the capacitance between turns limits this gain.
DC bias, saturation and core material
DC current through a bead magnetises the ferrite and lowers its permeability, so the impedance falls as the current rises. Many small chip beads lose a large part of their impedance well below their rated current, because the rating is usually a temperature-rise limit. Use the impedance-versus-DC-bias curve at the actual load current. In the 50 Ω example, if the bead drops from 600 Ω to 200 Ω at the operating current, the insertion loss falls from 16.9 dB to 9.5 dB, which a zero-bias LCR meter reading will not show.
A common-mode choke avoids most of this. The operating current flows out on one winding and back on the other, so its flux cancels in the core, while common-mode current adds. This allows a high-permeability core and a large common-mode inductance without saturation from the load current. Leakage inductance, unbalanced windings and large common-mode currents, such as surge currents, can still saturate the core.
Core material sets the frequency range. Manganese-zinc (MnZn) ferrite has high permeability and suits mains filters from tens of kHz to a few MHz. Nickel-zinc (NiZn) ferrite has lower permeability but keeps working up to hundreds of MHz, so it is used in clamp-on cores, beads and chokes aimed at radiated emissions.
Bead and capacitor resonance
Below its crossover a bead is inductive and forms an LC low-pass filter with the decoupling capacitor after it, with a resonance f0 = 1 / (2π · √(LC)). An example bead with 1 µH effective inductance and a 10 µF ceramic capacitor resonates near 50.3 kHz, with a characteristic impedance √(L/C) of 0.32 Ω. Low-ESR ceramics leave this resonance almost undamped, so noise near 50 kHz can be amplified instead of attenuated. Damp it with a capacitor that has some ESR, a small series resistor or a bead with more loss at low frequencies.
Common-mode chokes: what to check
For a mains choke, check the common-mode impedance or inductance at the failing frequencies, the rated current, the DC resistance at temperature and the leakage inductance. Sectional winding, with the two windings on opposite halves of a toroid, gives low capacitance between windings, good insulation between line and neutral and leakage inductance that acts as a differential-mode filter. Bifilar winding gives tighter coupling but more capacitance between windings, which reduces the high-frequency common-mode impedance.
For a signal-line choke on USB, Ethernet, HDMI or CAN, the differential-mode impedance at the data frequencies must stay low, or the choke distorts the eye diagram. Use parts specified for that interface, check the differential insertion loss and the mode conversion, and keep the two lines symmetrical up to the choke. An unbalanced choke converts common-mode noise into differential noise and the reverse.
| Part | Acts on | Typical use | Key data to read | Main pitfall |
|---|---|---|---|---|
| Chip ferrite bead | Mainly differential-mode noise on one conductor | Power rail isolation, filtering of slow I/O lines | R and X curves, impedance at DC bias, DC resistance | Impedance collapses at load current; resonance with decoupling capacitors |
| Signal-line common-mode choke | Common-mode noise on a pair | USB, Ethernet, HDMI, CAN ports | Common-mode impedance, differential insertion loss, mode conversion | Damages signal integrity if not specified for the data rate |
| Mains common-mode choke | Common-mode noise on line and neutral | AC input filters, with X and Y capacitors | Inductance, rated current, leakage inductance, winding capacitance | Wrong core material for the failing frequency; heating at full load |
| Clamp-on ferrite core | Common-mode current on a whole cable | Fixes in test, cables of external units | Impedance versus frequency, inner diameter, number of turns | Little effect if the cable's common-mode loop impedance is already high |
Where to place them
- Put cable filters right at the connector, with nothing unfiltered between the part and the cable.
- Keep the input and output sides apart; parallel traces or a shared ground pour let noise couple around the part.
- On a power rail, place the bead between the supply and the local decoupling, with the capacitors on the protected side.
- Never put a bead in a ground return or a ground plane connection; the return current then takes a larger loop.
- Fit clamp-on ferrites within a few centimetres of the product end of the cable.
Common mistakes
- Selecting by the 100 MHz value when the failure is at 30 MHz or 900 MHz.
- Testing a bead on an LCR meter at zero current and assuming the same impedance at full load.
- Adding a ferrite to a cable whose shield is poorly terminated. It may pass at one frequency, but the margin is lost when the cable changes.
- Stacking several beads in series on a rail without checking for new resonances.
- Using a mains choke with a MnZn core to fix a radiated emission at 200 MHz, where its impedance has already fallen.
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.