Glossary term
Fast Fourier Transform
An efficient family of algorithms for computing the discrete Fourier transform and its inverse.
Definition
methodAn efficient family of algorithms for computing the discrete Fourier transform and its inverse.
The Fast Fourier Transform reduces the cost of transforming sampled data between time and frequency representations. It is essential in vibration analysis, spectral estimation, communications, image processing, audio, radar, controls, and numerical simulation.
The Fast Fourier Transform is an algorithmic way to compute the discrete Fourier transform much faster than direct summation. For a sequence of length N, direct computation requires on the order of N^2 operations, while common FFT algorithms require on the order of N \log N operations.
Engineering role
The FFT is used to find frequency content in sampled data. Engineers use it for vibration diagnostics, modal testing, acoustic analysis, power-quality monitoring, communications, radar, image processing, spectrum analysis, filter implementation, and control-system identification. It turns a time record into frequency bins with amplitude and phase information.
Sampling requirements
An FFT can only describe what is present in the sampled record. The sampling rate determines the Nyquist frequency. The record length determines bin spacing. If the original signal contains frequency content above the anti-aliasing limit, aliasing can create false peaks that no FFT can correct after the fact.
Windowing and leakage
Real measurements rarely contain an integer number of cycles in the record. This causes spectral leakage, where energy spreads into adjacent bins. Window functions reduce leakage at the cost of amplitude and resolution changes. The chosen window, averaging method, scaling convention, and overlap should be stated when comparing spectra.
Amplitude interpretation
FFT output can be scaled as peak amplitude, RMS amplitude, power spectral density, one-sided spectrum, or two-sided spectrum. These forms are not interchangeable. For engineering use, the units and scaling must match the measurement question: displacement, velocity, acceleration, pressure, voltage, current, or dimensionless response.
Common mistakes
Common mistakes include trusting a spectral peak without checking sample rate, windowing, record length, sensor bandwidth, and noise floor. Another frequent error is treating frequency-bin spacing as the same as true frequency accuracy. For rotating machinery and transient events, stationarity and speed variation must also be considered.