← Back to the chapter

Technical detail · Time-Series Compression Audit Limits

Under the Hood: FFT reduction and its evidence limit

A vibration node sampling at 4 kHz with 2-byte samples produces 8 kB each second. Continuously transmitting that raw stream is unrealistic for a battery node. If an FFT window retains ten frequency bins with frequency, magnitude, and phase, the payload may be roughly 120 to 160 bytes plus metadata: a 50× class reduction before protocol overhead. The MCU-versus-radio energy trade-off depends on the chosen node.

Those peaks cannot reconstruct the waveform exactly. Keep the raw window locally, or sample it on demand, when full evidence is required. Reduced bytes answer a narrower question, even when they are useful for diagnosis.

Emit evidence about the reduction itself: algorithm name, version, window size, original sample count, retained feature count, error bound or lossless flag, and saturation, overflow, or missing-sample markers. These fields let cloud analytics distinguish an audit-grade answer from a trend answer or an exploratory visualization. A restored block must retain its meaning as well as its numbers.

← Back to the chapter