Acoustic testing
Acoustic measurement: the noise floor and the anti-alias filter decide everything
Sound pressure spans roughly six orders of magnitude, from 20 µPa to tens of pascals. That range is why 24-bit conversion is not a luxury here, and why an anti-alias filter that tracks the sample rate is not optional.
IEPE as one front end for microphones and accelerometers, the SPL-to-millivolt arithmetic that makes 24-bit necessary, A-weighting and what it hides, the anti-alias constant, AC coupling and the low corner, field calibration, and the sweep-rate trap.
Acoustic measurement is the one application area where the converter's bit count really is a headline specification. Not because more bits sound better, but because the quantity being measured spans a range that a 16-bit converter simply cannot cover — and the arithmetic that shows this takes four lines.
One front end serves microphones and accelerometers
IEPE — also called ICP or CCP depending on who is selling it — is the same two-wire constant-current interface used for industrial accelerometers and for measurement microphones. The instrument supplies a fixed current, typically 2.1 mA, up the same coaxial cable that carries the signal back, and the sensor's internal electronics modulate the resulting voltage.
The practical consequence is that one instrument covers both. A bench that measures the noise of a gearbox and the sound it makes through the air can use the same two modules, the same cable type and the same software path — the only thing that changes is the sensor on the end.
The arithmetic that makes 24 bits necessary
Sound pressure level is defined against a reference of 20 µPa, and 94 dB SPL is exactly 1 Pa. That second fact is the one worth memorising, because it is also the level at which microphones are calibrated and it anchors everything else.
| Level | Pressure | Output of a 50 mV/Pa microphone | What it is |
|---|---|---|---|
| 114 dB | 10 Pa | 500 mV | Front row of a loud PA; the point where hearing damage starts within seconds. |
| 94 dB | 1 Pa | 50 mV | The calibration point, and a busy workshop floor. |
| 74 dB | 0.1 Pa | 5 mV | A quiet office. |
| 54 dB | 10 mPa | 500 µV | A quiet room at night. |
| 40 dB | 2 mPa | 100 µV | The background of a good recording studio. |
Swipe the table sideways to see all columns
Now compare those millivolt figures against the resolution of the input range they would land on. On the ±10 V range, a 16-bit converter has a least significant bit of 305 µV. Divide by the microphone's 50 mV/Pa and that bit corresponds to 6.1 mPa, which is about 50 dB SPL — a normal conversation.
That is the whole argument. A 16-bit converter on the ±10 V range has its least significant bit somewhere around the level of a conversation, so the quiet half of the acoustic range is beneath one bit. A 24-bit converter on the same range has a least significant bit of 1.19 µV, which corresponds to 24 µPa — about 1.5 dB SPL, essentially the bottom of the defined scale.
Weighting: what dBA hides
A-weighting is a filter that approximates the ear's reduced sensitivity at low and high frequencies. By definition it is 0 dB at 1 kHz, and it rolls off to roughly −19 dB at 100 Hz and about −50 dB at 20 Hz. It exists because two noises with identical sound pressure level can be perceived very differently depending on where their energy sits.
- For steady broadband noise — a fan, a motor, traffic — dBA is the right single number and it is what regulations are written in.
- For impulsive content — a hammer blow, a valve slam, a door — dBA under-reports the low-frequency energy that the impulse actually contains, and it does so quietly. Report the linear or Z-weighted level alongside it.
- State the weighting and the time constant explicitly in any result you publish, because a dBA figure without them is not reproducible. The measurement standards that matter here are IEC 61672-1 for the instrument class and IEC 61094-4 for the microphone.
Anti-aliasing, and why it cannot be switched off
Any component above half the sample rate folds back into the band you are measuring, and once it has folded there is no way to tell it apart from a real signal at the folded frequency. The filter that prevents this must be analogue and must sit ahead of the converter, and analyser practice sets its corner at the sample rate divided by 2.56.
The constant is worth remembering because it makes the usable band easy to state. At a 51.2 kS/s sample rate the corner is 20 kHz, so the instrument measures a 20 kHz band and rejects what is above it. Without that filter, a switching supply whining at 40 kHz would fold down to 11.2 kHz — inside the audio band, where it would look exactly like a real 11.2 kHz tone and no amount of averaging would remove it.
AC coupling, and the low corner you accept
An IEPE sensor sits on a DC bias of several volts, and the sound pressure of interest is a small AC signal riding on top of it. AC coupling removes the bias before the converter, which is necessary — but the high-pass corner it introduces is a real limit on the measurement, not a formality.
- The USB-4431 couples with a corner at 0.8 Hz, which is fine for audio and for most machinery noise, and means the measurement is only trustworthy from a few hertz upwards.
- The USB-4432 couples at 0.1 Hz, which is the reason to choose it for infrasound, for large-room pressure changes, or for anything where the question involves the bottom two octaves.
- Below the corner, the reported level falls off and the phase is wrong. A 0.8 Hz corner does not mean 0.8 Hz is measurable; it means the response is 3 dB down there and unusable below.
Calibrate before and after, or the number is not a measurement
A microphone's sensitivity drifts with temperature, humidity and age, and the quoted figure on its calibration certificate is a factory value from some earlier date. The field calibrator exists to close that gap, and IEC 60942 defines it as a device that presents a known sound pressure at a known frequency — conventionally 94 dB at 1 kHz.
- Calibrate immediately before the measurement set and immediately after it. If the two readings differ by more than the tolerance you declared, the data in between is not trustworthy and no post-processing will rescue it.
- Record the calibrator's own tolerance and class with the result. The better classes are specified to a few tenths of a decibel, and that figure is part of the uncertainty budget whether or not it is written down.
- Match the calibrator to the microphone. A calibrator designed for a 1/2 inch microphone on a 1/4 inch capsule gives a level that is wrong by a known but easily forgotten number of decibels.
Sweeping too fast under-reads a sharp resonance
A swept-sine measurement trades time against frequency resolution, and the trade is not forgiving. A resonance has a bandwidth, and to measure its height correctly the sweep has to spend enough time inside that bandwidth for the system to reach steady state.
The arithmetic is short. A resonance at 1 kHz with a quality factor of 100 has a bandwidth of 10 Hz, and the corresponding settling time is on the order of a tenth of a second. A sweep that crosses that 10 Hz window in 20 ms will not reach steady state, and it will report a peak lower than the true one — a mistake that is invisible in the result, because the curve still looks like a clean resonance.
Driving the structure: the 4431 has an output, the 4432 does not
Acoustic and modal testing usually needs an excitation as well as a microphone. The USB-4431 provides one 24-bit analogue output at 96 kS/s with a ±3.5 V range, which is enough for a reference tone, a swept sine or a shaped noise drive into an amplifier. The USB-4432 has no analogue output at all.
| The job | Part | Why this one |
|---|---|---|
| Microphone arrays and modal work, where the excitation and the response must share a clock | USB-4431 | Four simultaneous 24-bit inputs at 102.4 kS/s, per-channel AC or DC coupling, per-channel IEPE, and one 24-bit output to drive the structure from the same time base. |
| Infrasound, high-level pressure or five simultaneous channels with no drive needed | USB-4432 | Five simultaneous 24-bit inputs with four IEPE-capable, a ±40 V range for high-level sensors, and AC coupling that reaches down to 0.1 Hz. |
Swipe the table sideways to see all columns
Errors that quietly invalidate a data set
- Connecting an IEPE microphone to a module with no IEPE excitation, and recording a flat line.
- Leaving the input on ±10 V and losing the entire quiet half of the range.
- Publishing a dBA figure for an impulsive source without also reporting the linear level.
- Assuming the anti-alias filter is present because the module has a high sample rate.
- Calibrating at the start of the week and trusting the calibration for the Friday set.
- Sweeping a high-Q resonance in a fraction of its settling time, and reporting the low peak as the answer.
Every figure in this note is either arithmetic from published resolution and sample-rate specifications, or a value stated by the standard it cites. Where a number depends on a particular module, confirm it against that module datasheet before design freeze.