Fan Sonar.

This is an acoustic temperature estimate. Speaker and microphone response, blade shape, the grille, reflection angle, and harmonics can affect results. Compare against a thermometer. Try slightly to one side of the fan rather than directly on its axis; keep the fan and device in place while measuring.

Spectrum: 48 kHz or 44.1 kHz; 4096-sample Hann window, hop 256. Carrier f₀ ±4 kHz, excluding ±30 Hz at the center; 2-second average; edge threshold at −20 dB in power from each side’s own peak. Unequal band power or width alone is not an error. The power-weighted centroid differs from Δf max and needs its own reference calibration to estimate temperature.

Envelope: STFT band power is sampled at 187.5 or 172.3 samples/s, with Nyquist limits of 93.75 or 86.13 Hz. For RPM, lower and upper bands are filtered separately, using carrier notches, power detection and an eight-pole 240 Hz filter before downsampling. The envelopes run at 750 or 689.1 samples/s. Observed rhythms and harmonics are scored together in a 10-second FFT. Combined power only corroborates a direction; division cannot invent a missing fundamental. Blade count identifies competing rotation and blade-pass candidates. Repeating pulse shapes are also checked; multiplication cannot invent invisible blade passages. The last three seconds detect speed changes.

Device limit: At 44.1 kHz sampling, only 3.05 kHz is available above a 19 kHz carrier. The unavailable region is hatched on the chart. An edge touching this limit or the ±4 kHz analysis limit cannot be used. If the other side is sufficient, it can be used for reference calibration; absolute calculation needs two valid edges. Try 17–18 kHz.

Stability: The chart uses the last two seconds; RPM and temperature share the same ten-second window. A reference locks to the lower, upper or both bands and never silently substitutes a missing band. A last valid value is labelled and held for at most 1.5 seconds during a brief interruption. Real audio loss refills the window while preserving the reference. CSV and calibration use only current data. Older reference files must be replaced for the new blade-identity method.

Precision limit: Calibration does not prove accuracy. A 1% ratio error is about 6 °C at room conditions. A rotor’s regular spectral lines can obscure small temperature changes. The displayed ± value is a lower bound from spectral resolution and measured variation; reference-free mode adds the entered dimension uncertainty. It is not a validated error interval covering humidity, grilles and every device effect.

Calculation: RPM = BPF × 60 / N. k ref = RPM / selected band measure. c ref = 331.3 + 0.606 × T ref; c = c ref × (RPM / band) / k ref. T = (c − 331.3) / 0.606. In reference-free mode, tip speed = 2πr × RPM / 60 and c = 2 × tip speed × g × f₀ / Δf max. The radial projection g comes from the phone’s axial distance and sideways offset. A near-field circular-tip model is used; g is not assumed to be 1 at every position. The speaker and microphone are approximated as one transceiver at their midpoint. Without radius and position, these two acoustic measurements alone cannot identify temperature. Results outside −40…70 °C are reported as a model/position mismatch.

Audio session: Echo cancellation, noise suppression, and automatic gain control are requested off. Tonal audio uses the music content hint; if browser filters remain on, supported constraints are reapplied. Temperature is unavailable while processing remains enabled. The microphone and sonar tone stop when the tab goes into the background.

Technical sources: Sonar geometry of rotating blades ↗ · Web Audio / AudioWorklet ↗ · NASA / speed of sound and temperature ↗