Cepstrum Analysis

Cepstrum analysis transforms a frequency-domain spectrum into the quefrency domain, where periodic patterns in the spectrum (harmonics, sidebands, echoes) appear as distinct peaks. Originally developed for seismic echo detection, it has become an essential tool for machinery diagnostics — particularly gear fault analysis and bearing condition monitoring.

Theory

Definition and Variants

The cepstrum is defined as the inverse transform of the logarithm of a spectrum. V-Listen supports three variants:

Variant	Definition	Properties
Real Cepstrum	c(t) = F⁻¹{ log \|X(f)\| }	Retains only magnitude information. Suitable for detecting harmonic spacing.
Power Cepstrum	c(t) = \| F⁻¹{ log \|X(f)\|² } \|²	Squared magnitude of the real cepstrum. Emphasizes periodicity detection.
Complex Cepstrum	c(t) = F⁻¹{ log X(f) }	Preserves phase. Required for signal reconstruction and echo removal (liftering).

Quefrency Domain

The horizontal axis of the cepstrum has units of time (seconds) but is called quefrency to distinguish it from the original time domain. A quefrency value corresponds to the period of a pattern in the frequency spectrum:

quefrency = 1 / frequency spacing

For example, a gear with 20 teeth on a shaft spinning at 30 Hz produces a harmonic family spaced at 600 Hz. This spacing appears as a cepstral peak at quefrency = 1/600 = 1.667 ms.

Cepstral terminology mirrors spectral terminology with playful letter rearrangements: spectrum → cepstrum, frequency → quefrency, filtering → liftering, harmonics → rahmonics.

Liftering

Liftering is the cepstral equivalent of filtering. By zeroing out specific quefrency ranges and performing the inverse cepstrum, you can:

Short-pass lifter — remove fine harmonic structure, retaining only the spectral envelope (useful for formant estimation in speech).
Long-pass lifter — remove the spectral envelope, retaining only the harmonic/sideband structure (useful for isolating gear mesh periodicity).
Notch lifter — remove a specific echo or periodic component from the signal.

Applications

Echo Detection

An echo in the time domain creates a periodic ripple in the log spectrum. The cepstrum converts this ripple into a single peak at the quefrency equal to the echo delay time. Multiple echoes produce multiple cepstral peaks at their respective delays.

Gear Diagnostics

Healthy gears produce a gear mesh frequency (GMF) with harmonics evenly spaced at GMF = N_teeth × f_shaft. The cepstrum concentrates this entire harmonic family into a single peak, making it easy to monitor. When a tooth develops a defect, sidebands appear around the GMF harmonics, and the cepstral peak amplitude increases.

Condition	Spectrum Pattern	Cepstrum Signature
Healthy gear	Clean GMF harmonics	Moderate peak at 1/GMF
Distributed wear	Elevated sideband levels	Increased peak amplitude
Localized tooth defect	Dense sidebands at shaft speed spacing	Peak at 1/f_shaft with rahmonics
Multiple mesh stages	Overlapping harmonic families	Separate peaks at each 1/GMF

Bearing Diagnostics

While envelope analysis is generally preferred for bearing fault detection, the power cepstrum can confirm bearing defect frequency spacing when the spectrum is too complex for direct identification.

Cepstrum vs Time

V-Listen can compute the cepstrum for successive time blocks and display the result as a quefrency-time map (cepstrogram). This is useful for tracking how cepstral peak amplitudes evolve over time during a run-up, coast-down, or long-term monitoring session.

Parameters

Parameter	Range / Options	Description
Cepstrum Type	Real / Power / Complex	Select the cepstrum variant.
Spectrum Size	1024 – 65536	FFT block size for the initial spectrum computation.
Window	Hanning / Hamming / Rectangular	Window function applied before the FFT.
Quefrency Range	Auto / Manual	Display range on the quefrency axis (ms).
Averaging	1 – unlimited	Number of cepstral blocks to average.
Lifter	None / Short-pass / Long-pass / Notch	Optional liftering for signal editing.

Workflow in V-Listen

Load a time-domain recording from a gearbox vibration sensor or microphone.
Open Analysis → Cepstrum.
Select the cepstrum type (Power Cepstrum is recommended for periodicity detection).
Set the spectrum size for adequate frequency resolution.
Click Calculate.
Identify peaks in the quefrency domain and relate them to known mechanical frequencies.

Practical Tips

The cepstrum is particularly powerful for multi-stage gearboxes where many harmonic families overlap in the spectrum but separate cleanly in the quefrency domain.
Use the power cepstrum for detection and trending; use the complex cepstrum when you need to reconstruct the signal after echo or harmonic removal.
Combine cepstrum analysis with order tracking for variable-speed machinery to maintain consistent quefrency peak positions.
The cepstral peak amplitude is proportional to the regularity of the harmonic family; amplitude growth over time indicates developing faults.

The logarithm in the cepstrum definition compresses the dynamic range. Ensure the input signal has adequate signal-to-noise ratio; noise floor artifacts can produce spurious cepstral peaks at very short quefrencies.