The FFT operation is performed for sampled numerical data within a certain interval (for example, 1,024 or 2,048 points). The operation to cut out a certain portion of a waveform in this manner is referred to as cutting out a waveform with a window (time window) or applying a window. The Fourier transform itself is defined in terms of infinite data. This also applies to discrete Fourier transform (DFT). The FFT analyzer analyzes signals on the premise that, when the waveform is cut out with a window, the waveform in this section is repeated infinite times.At this time, if the analysis data length (time length) is integer times the period of each frequency, the waveform presumed by the FFT analyzer is identical to the actual input waveform, making it possible to obtain a single line spectrum.

However, if the analysis data length is not integer times the period (the frequency resolution not applied, and the front end of a frame does not connect to the tail end), the waveform obtained by FFT is distorted and side lobes appear because the power is not concentrated in the spectrum. Theoretically, data with infinite length is required to obtain a single line spectrum. Since the FFT analyzer processes signals using data within a limited section, a leakage error occurs. The processing for minimizing this leakage error is referred to as window processing. When a peak-shaped function with which both ends of the frame reaches zero is applied, the front and tail ends of the frame are connected, resulting in smaller error. This kind of function is referred to as the window function, and the processing for synchronization of the analysis signal using the window function is referred to as window processing. As a result, the shape of the spectrum becomes similar to that of the line spectrum.

A typical window is the Hanning window; other types of windows are used that best suit the analysis signal.