React FFT / Spectrogram

Copy Page

A time-domain trace tells you a machine is vibrating. The spectrum tells you at what frequency, which tells you which part. This ships a real radix-2 Cooley–Tukey FFT with selectable windows — switch to the waterfall and a drifting defect tone shows up as a bend rather than a number that changed.

Installation

Props

It is a real FFT

Not a decorative approximation: an in-place iterative radix-2 Cooley–Tukey transform with bit-reversal permutation, verified against known tones — a full-scale 500 Hz sine at 12.8 kSa/s with a 256-point FFT lands in bin 10 at 0.00 dB, exactly where the arithmetic says it should.

Two details that make the numbers trustworthy:

• Window normalisation. Magnitudes are divided by the window's coherent gain, so applying a Hann window doesn't silently cost you 6 dB of amplitude. Switch between Hann, Hamming, and rectangular and a tone's reported level stays put while the skirts change — which is the actual trade you're making.
• Bin resolution is stated, not hidden. The header prints Hz/bin. With a 256-point FFT at 12.8 kSa/s that is 50 Hz, so the demo's 480 Hz shaft tone reports as 500 Hz — correct behaviour for the resolution, and better shown than smoothed over. Raise fftSize for finer bins at the cost of time resolution.

The waterfall computes overlapping frames at 50% hop and stacks them newest-last, coloured on the same CVD-safe ramp as Heatmap / Matrix. For the time-domain view of the same capture, pair with Measurement Cursors.