The Complaint That Never Appears in a Datasheet
A power supply that hums at 8 kHz is a commercial problem long before it is an engineering one. The customer hears it in a quiet room, the product gets returned, and the magnetics supplier is asked to “make the noise go away” without anyone having established where the noise comes from. Magnetostriction, fringing forces, loose laminations and a badly potted assembly all produce audible noise, and they are fixed by different means.
Identify the Mechanism First
| Mechanism | Signature | First fix to try |
|---|---|---|
| Magnetostriction | Pure tone at twice the excitation frequency; strongest in grain-oriented silicon steel | Lower flux density, or change material to a lower-magnetostriction grade |
| Gap fringing forces | Tone at the switching frequency, strongest in gapped inductors | Distribute the gap, or clamp the core halves firmly |
| Loose lamination stack | Broadband buzz, worse at higher current | Impregnation or varnish, plus adequate clamping pressure |
| Winding movement | Buzz that changes with load | Varnish or potting to immobilise the winding |
| Bobbin / housing resonance | Tone that changes with mounting torque or enclosure | Change mounting stiffness; add damping; move the resonant frequency |
| Control loop instability | Variable pitch, not a pure tone | Not a magnetics problem: fix the loop or the burst mode |
The diagnostic order matters. Record the spectrum first: a pure tone at exactly twice the line frequency in a 50 Hz design is magnetostriction, while a tone at the switching frequency in a gapped inductor is almost always the gap. Guessing leads to expensive material changes that fix nothing.
What Actually Reduces Noise
- Reduce the flux swing. Magnetostriction is proportional to flux density; a 20 % reduction in Bpk typically gives a clearly audible improvement. The cost is more core volume.
- Immobilise everything. Varnish impregnation or full potting removes the mechanical degrees of freedom that turn a small force into a loud sound. For potted assemblies, the potting compound’s modulus matters, not just its presence.
- Clamp the core properly. Two halves held by a weak clip will sing. A clamp with defined pressure, or a glued joint with a controlled bond line, is the fix.
- Move the resonance. If the assembly resonates near the switching frequency, changing the mounting boss or adding mass shifts the mode away from the excitation.
- Change the material where justified. Amorphous and nanocrystalline cores have different magnetostriction behaviour from grain-oriented steel; for 50/60 Hz transformers, a lower-B design in the same material is usually cheaper than a material change.
What to Ask For in the Specification
- An acoustic requirement, stated as sound pressure level at a defined distance and in a defined mounting. “Quiet” is not a requirement.
- The switching or line frequency, so the supplier can avoid placing a structural resonance on it.
- Whether the part will be potted, varnished, or neither — this changes the design, not just the process.
- Whether the first article must include a noise measurement, and at what load and ambient.
A supplier who has never been asked for an acoustic number will not design for one, and the noise will be discovered in the customer’s living room rather than on the test bench.
Related reading
- Potting and Varnish Impregnation
- Vacuum Pressure Impregnation vs Dip
- 50 Hz vs 60 Hz Operation
- Ferrite Core Gapping
Send us the spectrum recording and the mounting drawing; we will tell you which of the mechanisms above is producing it before quoting a change. sales@sirui-tech.com.