Spectral data in the form of an
-point DFT is quantized to frequency bins spaced apart by
Spectral peaks do not generally fall exactly at bin frequencies, thus it is common to interpolate frequency peaks to better estimate their center frequencies and amplitudes.
A common method for interpolating spectral magnitude peaks is to use quadratic interpolation, which involves fitting a parabola to the three most significant values around a peak.
Given the three spectral magnitude values around a peak as
, the interpolated peak location (in bins) is given by
in the range -1/2 to +1/2.
The magnitude at the peak is given as
©2004-2017 McGill University. All Rights Reserved.
Gary P. Scavone