The discrete Fourier transform represents a sampled signal using complex frequency components. FFT algorithms compute this transform efficiently.
Mathematical principle
x[n] is the sampled signal; N is its length; k identifies a frequency bin.
Input definitions
| Item | Meaning |
|---|---|
| Samples | Values in a defined order and unit. |
| Sampling rate | Samples per second, used to label frequency. |
| Transform length | For an N-point transform, bin spacing is fₛ/N. |
Analysis choices
| Item | Meaning |
|---|---|
| Window | Changes leakage and amplitude scaling. |
| Normalization | Specify amplitude, power, or power spectral density. |
| Spectrum range | Distinguish one-sided and two-sided representations. |
Worked mathematical figure
Interpretation
A peak identifies a frequency component in the analyzed segment. Zero padding samples the spectrum more densely; it does not add information to the original recording. Preserve the window and normalization when comparing spectra.
Scientific references
Related major models
From the principle to a scoped workflow
Before using a spectrum, identify its input domain, normalization and units. The same frequency axis can accompany different measured quantities; a visually similar peak does not make amplitude and density interchangeable.
| Learning path | What to inspect and what it establishes |
|---|---|
| Full-record FFT amplitude and segmented Welch density | The CWRU guide separates the full-record amplitude spectrum from the Welch density table, and retains the sampling rate, window, detrending and complete exported rows. Evidence scope: One named bearing-recording workflow; it does not establish every spectral estimator, other sensors or fault diagnosis. |
| Complex FID, frequency coordinates and ppm | The spectroscopy guide follows a selected complex FID through FFT and ppm coordinates, distinguishing plotted axis direction from saved numeric order. Evidence scope: One synthetic four-voxel input with one selected voxel; it is not chemical identification, metabolite quantification or general spectral validation. |
Watch the scientific explanation
Read the video transcript
0:00 Inspect a two-second waveform, sampled at 100 samples per second. It is the sum of a 5 Hz component with amplitude 0.7 and a 12 Hz component with amplitude 0.3.
0:08 The discrete Fourier transform separates the frequency components. The one-sided amplitude spectrum uses the same 200 samples, with interior coefficients scaled by 2/N.
0:16 Two peaks appear at 5 Hz and 12 Hz. Their amplitudes match the known inputs because both components complete an integer number of cycles in the record.