Advanced DSP - FFT windows: spectral leakage

The measured frequency spectrum is not only affected by broadband noise, but also by narrow band frequency components either of noise or of the signal:

A frequency componentof noise or signal at one frequency will contribute to the measured value at an FFT bin. The amount of the contribution is proportional to the amplitude of the window function's kernel, centred at the FFT bin, measured at the interfering frequency. This amount is what is meant when we talk about spectral leakage.

Spectral leakage changes both the amplitude and position of a frequency measurement. The effects of spectral leakage are worst when detecting small signals in the presence of nearby large signals.

To minimise the effect of spectral leakage, a window function's kernel should have low amplitude sidelobes away from the centre: and the rate of fall off to the low sidelobes should be rapid. The peak sidelobe level is a useful indicator of how well a window function suppresses spectral leakage: so is the rate of fall off to the sidelobes.