Design solutions that pursue the sound quality and experience you need
Stepped sine works by dwelling at each frequency for a few hundred milliseconds, waiting for the system to settle, then reading one amplitude. It is clean and has excellent signal-to-noise ratio, but it is slow: 20 Hz to 20 kHz at 1/24-octave spacing is close to three hundred points, and even at 0.3 s per point that is a minute and a half per unit — impossible on a takt-time line. Automated frequency response testing replaces the whole sequence with a single continuous logarithmic sweep, covering the band in 0.5 to 2 seconds. Deconvolution separates the impulse response from that one sweep, and a single FFT yields the entire curve. Because the sweep spreads each frequency along the time axis, deconvolution isolates the linear response precisely, so the signal-to-noise ratio actually improves on the stepped method. That combination of speed and quality is why the technique took over production.
The impulse response recovered from a sweep contains more than the direct sound. It also carries the first reflection off the fixture walls, the cabinet edge and the room. The fix is a time gate: keep only the window that follows the arrival of the direct sound and discard what comes after, which approximates a free-field measurement without an anechoic chamber. Window length is not a free parameter — frequency resolution is roughly the reciprocal of the window duration, so a 5 ms gate puts the useful lower limit near 200 Hz. Shorten the window and the low end becomes unusable. The usual compromise is a short gate of 3 to 5 ms for the mid and high bands, with near-field measurement or a relaxed tolerance for the bass. Gate placement matters just as much: open it a little before the direct-sound peak. Open too early and low-frequency energy is clipped; open too late and the reflection slips in.
A raw curve is full of fine serrations that come mostly from cavity standing waves and unit-to-unit variation. Gating production on that data would reject good product every day. The industry convention is to judge on a 1/3-octave smoothed curve and to inspect detail at 1/12 or 1/24 octave. Third-octave smoothing roughly matches the critical bandwidth of hearing: it preserves the features that actually change what a listener hears — crossover notches, overall tilt — while erasing meaningless single-bin ripple. One caution: smoothing must average energy in the amplitude domain, never take an arithmetic mean of dB values, or notches get systematically lifted. The mean level of the curve is also the sensitivity figure, so changing the smoothing changes the sensitivity number. Both sides of the specification have to use the same setting.
Judgement means wrapping the curve in an upper and lower limit. Build the band from at least thirty known-good units, compute mean and standard deviation at each frequency, then apply band-dependent tolerances. The bass is dominated by cavity and sealing variation, so ±4 to 5 dB is common. The midrange is where hearing is most sensitive, so tighten to ±2 to 3 dB. The top end is limited by driver consistency and often needs ±4 dB or more. A band built this way catches real defects without rejecting whole batches over a single noisy bin. Build it per model, and re-confirm after any diaphragm, cavity or adhesive change — copying another model's limits is asking for trouble. Operationally, run and archive a golden-sample curve at the start of every shift, and investigate fixture sealing and microphone distance first whenever the curve drifts.
Q: Can frequency response and rub-and-buzz testing share one sweep?
A: Technically yes, but the drive level has to be compromised. Response testing wants a moderate level to avoid compression, while buzz usually only appears at large excursion. The common arrangement is one fixture and one microphone running two sweeps at different levels, buzz first, then response, to save load and unload time.
Q: Why does the same line produce different curves in the morning and at night?
A: Temperature and tooling, mostly. A cold driver has lower compliance and loses a fraction of a dB of bass sensitivity, and an ageing seal or a change in insertion depth shifts the whole curve. Log ambient temperature and trend the golden sample, then attribute drift to the fixture or the environment instead of rejecting product.
Q: Is production response data usable without an anechoic chamber?
A: Yes, provided a gated window and a fixed fixture suppress the reflections, and provided you accept reduced accuracy at low frequencies. Consistency matters more than absolute accuracy: as long as conditions repeat, the limit band still catches relative change. Send samples to a reference environment when absolute sensitivity has to be certified.
Ready to define a response test that holds up on a real line? Request a Quote from Liwei and we will map the measurement setup to the driver you are building.
Shenzhen Liwei Electronic Technology Co., Ltd. focuses on audio headset electronics and complete PCBA design, with over 10 years of experience, 15+ well-known brands and 300+ customers. Our eight solution lines cover hearing protection, shooting and hunting, streaming, gaming, call center, LE Audio and Auracast broadcast speakers, digital active speakers and DSP mixers.
Keywords: frequency response test, speaker frequency response, automated audio test, acoustic measurement
