TInterpolationMethod Enumeration

type TInterpolationMethod = ( imNone, imNumeric, imQuadratic, imBarycentric, imQuinnFirst, imQuinnSecond, imJain );

Defines a set of frequency spectrum peak interpolators.

Selects how the true frequency of a peak that falls between the discrete bins f_k = k HzRes is estimated. imNone returns the bin centre, so the error is up to 1/2 HzRes. imNumeric works with any window and zero padding and is the most accurate, also returning the complex spectral value so amplitude and phase are recovered. imQuinnFirst and imQuinnSecond use the complex spectrum at the three central bins; imQuadratic, imBarycentric and imJain use the three amplitudes. All interpolators except imNumeric require a rectangular window with no zero padding and an amplitude spectrum, otherwise the analyzer falls back to imNone. For a true off-bin frequency f_0 the recovered estimate f satisfies | f - f_0| << HzRes, with imNumeric and imQuinnSecond reaching about 10^(-3) HzRes.

Values

NameDescription
imBarycentricRequires rectangular window, no zero padding, but works with amplitude spectrum only and is fast. Accuracy is not very good.
imJainRequires rectangular window, no zero padding, but works with amplitude spectrum only and is very fast. Accuracy is not very good.
imNoneNo frequency interpolation will be performed.
imNumericNumerical peak interpolation is the most accurate and can be used with any window and any zero padding setting, but it is also the slowest. Absolute accuracy of the numerical interpolator can be increased dramatically by using a Kaiser window with high attenuation factors.
imQuadraticRequires rectangular window, no zero padding, but works with amplitude spectrum only and is fast. Accuracy is not very good.
imQuinnFirstWorks with rectangular window and no zero padding. Requires complex frequency spectrum and is very fast.
imQuinnSecondWorks with rectangular window and no zero padding. Requires complex frequency spectrum and is very fast. The second Quinn estimator is very accurate and fast, if your frequency spectrum is not based on zero padded data and you can afford to use the rectangular window.