Sometimes the most important information hides not in a single frequency line, but in the pattern of several lines. A harmonic series (50, 100, 150, 200 Hz…) says something different from isolated lines. Cepstral analysis makes such patterns visible.

What is a cepstrum?

c(n) = IFFT [ log |X(k)|² ]

Periodic structure in the spectrum becomes a peak in the cepstrum. The x-axis is called quefrency (deliberately twisted from "frequency"). Unit: seconds, but representing period in the spectrum.

Use cases

Bearing diagnosis: roller bearing damage produces impulses at characteristic frequency, modulating the carrier and creating a harmonic series – which becomes a single cepstral peak. Much clearer than 30 scattered lines.

Echo detection: echoes appear as isolated peaks in the cepstrum – core to speech analysis and sonar.

Pitch: fundamental frequency of human voice. Standard in every modern speech system.

Why the logarithm?

log(A · B) = log(A) + log(B)

Separates multiplicative components into additive ones. If a signal is excitation × transfer function, both components separate cleanly.

Rahmonics instead of harmonics

Cepstrum researchers playfully twisted terms: frequency → quefrency, harmonics → rahmonics, filter → lifter.

Practical example: bearing damage

An outer-race defect produces impulses every 6.7 ms (roller pass frequency). The spectrum shows a harmonic series at 149 Hz spacing, hard to read among other components. The cepstrum shows one peak at 6.7 ms – instantly visible.

What to remember

  1. Cepstrum = FFT of log-spectrum.
  2. Reveals periodic patterns in the spectrum as single peaks.
  3. Standard for bearing diagnosis, echo detection, speech.
  4. A specialist tool in inline acoustic setups, not a universal hammer.