Design solutions that pursue the sound quality and experience you need
Rub and buzz describe two families of defects. Rub is mechanical: the voice coil, lead wires, adhesive beads or stray particles scrape against each other as the diaphragm moves. Buzz is electrical or structural: an intermittent contact or a loose joint chatters and radiates a high-frequency rattle. Both share the same signature — the energy is tiny, the event is brief, and it only appears in a narrow band and above a certain excursion. Total harmonic distortion sums the energy of every harmonic across the whole sweep, so a 0.1% transient burst disappears inside normal distortion. Worse, buzz components are often not integer multiples of the drive frequency at all, so they never land in the harmonic series that THD is defined on. That is why a unit that a trained listener rejects on the first pass can still show a perfectly clean THD curve.
If the defect hides in the non-harmonic and high-order region, detection starts by subtraction. The standard approach feeds a pure sine into the driver, captures the output with a measurement microphone, then attenuates the fundamental by at least 40 dB and removes the first several harmonics — typically up to the tenth order. What remains is the rub and buzz residual. In practice this is done with a tracking notch filter, or by synchronous averaging in the log domain followed by spectral excision. Two details decide whether the system works: the stop-band depth must be deep enough, and the phase response must not be disturbed. Any fundamental leakage lifts the entire residual curve and the sensitivity collapses. The residual is plotted in dB against frequency, and a defect appears as a sharp spike that stands clear of the surrounding floor.
Rub and buzz are events, not steady states, so excitation matters more than the algorithm. Stepped sine holds each point for tens of milliseconds and gives the best stability at the cost of test time. A continuous logarithmic sweep with deconvolution separates the impulse response quickly and suits production takt time. Sweeping too fast will skip short intermittent buzz; keep the rate at 0.5 s per octave or slower, and go beyond 1 s per octave for critical models. Drive level deserves equal attention: coil rub usually only appears at large excursion, so test at 80–100% of rated power. Measuring at low level is self-defeating. Keep the start frequency above twice the resonance frequency as well, otherwise normal large-displacement nonlinearity will be flagged as a defect.
A curve alone cannot gate a production line; it has to become a number. Three quantities are used in practice: the maximum deviation of the residual against a golden sample within the band of interest (dB), the number of spikes crossing the threshold, and the cumulative width of the offending bands. Peak alone is easily skewed by a single noisy bin, so pairing it with a count and a width filter removes most false calls. Thresholds still come from data — measure at least 30 known-good units, take the mean plus three sigma, then add 3–6 dB of margin, because rub and buzz testing is extremely sensitive to fixture sealing and clamp pressure. Operationally, run a golden-sample baseline curve at the start of every shift and archive it. Any drift is a fixture or microphone problem first, a product problem second.
Q: Can rub and buzz testing replace human listening?
A: Not entirely, but it covers most quantifiable defects. The instrument wins on consistency and traceability, the listener wins on judging whether something actually sounds wrong. The practical split is 100% instrument screening with listener sampling for confirmation, each checking the other.
Q: Why do two measurements of the same unit look so different?
A: Usually too fast a sweep, inconsistent insertion depth or clamp pressure, or ambient noise leaking into the fixture. Lock the sweep and dwell parameters first, then check fixture repeatability with a GR&R study before blaming the product.
Q: Can we reuse a threshold from another model or a customer's file?
A: Not recommended. Residual floors differ substantially with diameter, sensitivity and cavity, so a copied limit either misses defects or rejects good units. Build a baseline per model from known-good distribution, and re-confirm whenever material, line or fixture changes.
Ready to tighten your incoming and in-line acoustic screening? Request a Quote from Liwei and we will review your test setup against the driver you are building.
Shenzhen Liwei Electronic Technology Co., Ltd. designs audio and headphone electronics with full PCBA solutions, backed by over 10 years of experience serving 15+ well-known brands and 300+ customers. Our portfolio covers hearing protection, shooting and hunting, live streaming, gaming, call center, LE Audio broadcast speakers and more. Contact Liwei for a custom audio solution.
Keywords: rub and buzz, rub buzz detection, speaker abnormal noise, acoustic testing, THD limitations
