Acoustic diagnostics

Eight-microphone acoustic sensing on RK3308

A microphone array is not a general-purpose ear; it is a bandpass instrument. This note derives the usable band of an 8-element array from its geometry, puts 22 field scenarios on one specification grid, and states plainly which decisions can stay on the device.

Figures marked [chip] come from the Rockchip RK3308 product page and datasheet. [physics] values are derived in this note from the speed of sound and array geometry. [industry] bands are the ranges in common use across vibration monitoring, acoustic emission and structural health monitoring. [engineering] statements are inferences from the above, not measurements.

Why an array

One microphone tells you that something happened. An array tells you where, and to whom.

Three capabilities an 8-element array buys that a single capsule cannot.

Spatial filtering

Eight channels let you steer a beam at the source and place a null on the dominant interferer. In a machine room that is the difference between listening to a bearing and listening to the fan beside it.

Redundancy

A capsule that is blocked, wet or dead still leaves seven channels. You can detect and exclude the bad channel instead of writing off the whole monitoring point.

Spatial separation

Beamforming separates “the source moved” from “the source changed” — two conditions with identical single-microphone signatures and completely different maintenance responses.

Hardware VAD keeps the pre-trigger buffer: at the instant of the event the onset is already recorded.

Aperture

8 elements give you 1.8 octaves — and spacing them further apart cannot widen it

Two critical frequencies bracket an array’s useful band: below c/D there is no directional gain, above c/2d grating lobes appear. Their ratio is log₂((N−1)/2), which is independent of spacing. Spacing only slides the same 1.8-octave window up or down the spectrum.

Element spacing d	Array length (N−1)d	Usable from c/D	Aliasing above c/2d	Where it fits
20 mm	140 mm	2.4 kHz	8.6 kHz	In-module array. High-frequency transients only — glass break, impact.
40 mm	280 mm	1.2 kHz	4.3 kHz	Panel-mounted array. Hand tools, small gearboxes, hob and pipe noise.
100 mm	700 mm	490 Hz	1.7 kHz	Cabinet or room array. Mid band — ventilation fans, pumps, rail.
300 mm	2100 mm	163 Hz	572 Hz	Machine- or mast-mounted. Low-frequency machinery and water noise.
1000 mm	7000 mm	49 Hz	172 Hz	Outdoor structure or tower. Structural modes, sway, cable force.

Read that as a constraint, not a defect. A 20 mm module array is excellent at glass break and rockfall; it is blind to a wind-turbine main bearing at 15 Hz. Choosing the array is choosing which questions you are allowed to ask.

Figure 1 — Usable beamforming band per aperture, for an 8-element array. Lower limit c/D, upper limit c/2d, both derived from c = 343 m/s.

The practical consequence: most projects need two apertures, not one — a dense array for transients and a sparse, long array for structural low frequencies, on the same board. A single aperture cannot be widened, but the element count can: 16 elements would give 2.9 octaves instead of 1.8.

Scenario matrix

22 scenarios on one specification grid

Every row is fixed by the same four numbers: the band you must be sensitive to, the sampling rate that follows from it, how fast the judgement has to be made, and where that judgement can live.

Group	Scenario	Band of interest	Sampling	Response	Decision
Rotating and reciprocating machinery	Motor / pump — unbalance, misalignment, bearing	10 Hz – 10 kHz	≥ 24 kHz	hours	On device
Rotating and reciprocating machinery	Gearbox / reducer — mesh, tooth breakage	100 Hz – 20 kHz	≥ 48 kHz	hours	On device
Rotating and reciprocating machinery	Press / machine tool — impact, tool wear	200 Hz – 20 kHz	≥ 48 kHz	milliseconds	On device
Energy assets	Wind turbine — main bearing, blade, yaw	0.1 Hz – 10 kHz	≥ 24 kHz + slow channel	hours	Trigger + review
Energy assets	Tower / blade — modal and low-frequency response	0.1 – 20 Hz	≥ 100 Hz	minutes	Trigger + review
Energy assets	Transformer / reactor — abnormal hum	50 Hz – 5 kHz	≥ 12 kHz	hours	On device
Livestock houses	Ventilation fan — bearing, belt, rotor lock	20 Hz – 10 kHz	≥ 24 kHz	minutes	On device
Livestock houses	Animal vocalisation — cough, alarm, farrowing	50 Hz – 8 kHz	≥ 24 kHz	seconds	On device
Livestock houses	Feed line / scraper — jam and abnormal impact	20 Hz – 8 kHz	≥ 24 kHz	minutes	On device
Hydrology and slope hazards	River acoustics — discharge and turbulence shift	20 Hz – 5 kHz	≥ 12 kHz	minutes	On device
Hydrology and slope hazards	Debris flow / flash flood — low rumble plus sustained scour	10 Hz – 2 kHz	≥ 6 kHz	minutes	Device + cloud
Hydrology and slope hazards	Rockfall / slope collapse — impact transient and rolling rhythm	100 Hz – 20 kHz	≥ 48 kHz	milliseconds	On device
Transport and structures	Bridge — vehicle passage response, expansion-joint impact	1 Hz – 2 kHz	≥ 6 kHz	seconds	Device + cloud
Transport and structures	Pavement — tyre noise as a proxy for damage and ponding	200 Hz – 5 kHz	≥ 12 kHz	seconds	On device
Transport and structures	Rail track — wheel/rail noise, joints, corrugation, switches	100 Hz – 20 kHz	≥ 48 kHz	milliseconds	On device
Transport and structures	Rope and stay cable — tension, wire break, rain-wind vibration	0.5 Hz – 5 kHz	≥ 12 kHz	minutes	Trigger + review
Transport and structures	Building — curtain wall, structural noise, lifts, plumbing	50 Hz – 10 kHz	≥ 24 kHz	seconds	On device
Indoor and home	Glass breakage	2 – 8 kHz	≥ 24 kHz	milliseconds	On device
Indoor and home	Fall — impact plus vibration	20 Hz – 2 kHz	≥ 8 kHz	milliseconds	On device
Indoor and home	Hob left dry-burning — whistle, vaporisation, empty pan	1 – 16 kHz	≥ 40 kHz	seconds	On device
Indoor and home	Pipe — leak, cavitation, blockage	100 Hz – 50 kHz	≥ 100 kHz	minutes	On device
Security	Drone intrusion — blade-pass frequency and harmonics	100 Hz – 8 kHz	≥ 24 kHz	seconds	On device

Bands are [industry] practice, not measured figures. Sampling rates are [engineering]: the band doubled with margin, then snapped to a rate the RK3308 audio paths can actually produce.

Acquisition

Three capture paths, three jobs

The RK3308 reaches microphones three different ways. The band you need decides which one you use.

PDM — up to 192 kHz per channel

8 channels on a 5-wire interface, 16–24 bit. This is the only path that reaches the band the pipe-leak and impact rows need. Use it when the answer lives above 20 kHz.

I2S / TDM — 2 × 8 channels at 48 kHz

The workhorse for standard digital microphones and mid-band work. The two buses are independent, so a 48 kHz dense array and a slow structural channel set can run at the same time.

On-chip CODEC — 8 × ADC, 24 bit

Analog capsules wired straight to the SoC: differential or single-ended input, programmable gain, and two microphone-bias rails at up to 0.85 × AVDD3V3. Fewest parts, and the only path that works with an ordinary analog measurement microphone.

93 GBper node per day at 48 kHz, 24 bit, 8 channels

371 GBper node per day at 192 kHz — why capture cannot be continuous

3.86 GB/hone hour of raw 8-channel audio

The hardware VAD matters more than it looks. It accepts analog, I2S and PDM microphones, and for arrays it keeps both the pre / post buffers — so at the instant of the trigger the recording already contains the samples from before it. For impact events the onset is the diagnostic information, and a system that starts recording after the threshold has already discarded it.

Decision split

On-device judgement, cloud explanation

The RK3308 has no NPU. That is a specification, not a defect — it simply fixes which half of the problem stays local.

Layer	Runs on	What it decides	Output
Feature extraction	RK3308	One-third-octave and mel spectra, envelope, crest factor, kurtosis, cross-channel coherence and TDOA	One feature vector per frame
Baseline and novelty	RK3308	Deviation from the commissioned fingerprint of this machine or structure	A score, plus which band caused it
Classification and grade	RK3308, small model	Known event classes with a coarse severity grade; classical DSP thresholds for the remainder	An event with a class and a grade
Explanation and action	Cloud	What the deviation probably means, which part to inspect, which work order to raise, how this asset compares with its fleet	Language, a work order, a trend narrative

The split is deliberate. A judgement that must fire in milliseconds cannot wait for a round trip; a narrative that needs fleet context cannot live on a 1.3 GHz core with no NPU. Reporting feature vectors and events instead of audio also means the raw recording never leaves the site — which is usually the compliance requirement anyway.

Field pitfalls

Five places where the installation quietly changes the signal

Pitfall	Symptom	What to do
Wind and rain on the diaphragm	A broadband low-frequency rise that swamps the machine signature	Windscreen plus a physical high-pass; do not try to filter it out in software after the ADC has already clipped
Uncalibrated channels	The beam points slightly wrong and the null lands on the wrong side	Commission against a single known source and store per-channel gain and delay offsets; re-check after any capsule change
Clock skew between nodes	Cross-node TDOA drifts and distributed localisation stops working	One clock domain per array; sync nodes to a common time base and timestamp events at the capture point, not at upload
Mounting resonance	A spectral peak that belongs to the bracket, not to the machine	Measure the empty structure first and fold that spectrum into the commissioned fingerprint
Continuous raw capture	Storage fills in days and the interesting hour is overwritten	Use the VAD pre/post buffers, keep raw audio only around events, and retain features continuously

None of these are exotic. They are why acoustic monitoring pilots fail after the demo: the algorithm was fine, and the installation silently changed the signal.

—

Basis of figures

Chip capability [chip]: Rockchip RK3308 product page and datasheet — 8-channel PDM at up to 192 kHz, 16–24 bit, on a 5-wire interface; 2 × 8-channel I2S/TDM at 48 kHz; on-chip CODEC with 8 × 24-bit ADC; hardware VAD accepting analog, I2S and PDM microphones with multi-microphone pre/post buffering. Aperture figures [physics]: derived here from c = 343 m/s and an array of 8 elements with total length (N−1)d. Data rates [physics]: 8 channels × 24 bit at the stated rate, expressed in binary gigabytes. Scenario bands [industry]: ranges in common use across vibration monitoring, acoustic emission and structural health monitoring, offered as engineering guidance rather than measurements. No field data from a deployed RK3308 array is presented here.