Calculate the theoretical minimum variance of a localized multitaper spectral estimate and the corresponding optimal weights to apply to each localized spectrum.


var_opt, var_unit, weight_opt = SHMTVarOpt (l, tapers, taper_order, sff, [lwin, kmax, nocross])


var_opt : float, dimension (kmax)
The minimum variance of the multitaper spectral estimate for degree l using 1 through kmax tapers.
var_unit : float, dimension (kmax)
The variance of the multitaper spectral estimate using equal weights for degree l using 1 through kmax tapers.
weight_opt : float, dimension (kmax, kmax)
The optimal weights (in columns) that minimize the multitaper spectral estimate’s variance using 1 through kmax tapers.


l : integer
The spherical harmonic degree used to calculate the theoretical minimum variance and optimal weights.
tapers : float, dimension (lwinin+1, kmaxin)
A matrix of localization functions obtained from SHReturnTapers or SHReturnTapersM.
taper_order : integer, dimension (kmaxin)
The angular order of the windowing coefficients in tapers.
sff : float, dimension (l+lwinin+1)
The global unwindowed power spectrum of the function to be localized.
lwin : optional, integer, default = lwinin
The spherical harmonic bandwidth of the localizing windows.
kmax : optional, integer, default = kmaxin
The maximum number of tapers to be used when calculating the minimum variance and optimal weights.
nocross : optional, integer, default = 0
If 1, only the diagonal terms of the covariance matrix Fij will be computed. If 0, all terms will be computed.


SHMTVarOpt will determine the minimum variance that can be achieved by a weighted multitaper spectral analysis, as is described by Wieczorek and Simons (2007). The minimum variance is output as a function of the number of tapers utilized, from 1 to a maximum of kmax, and the corresponding variance using equal weights is output for comparison. The windowing functions are assumed to be solutions to the spherical-cap concentration problem, as determined by a call to SHReturnTapers or SHReturnTapersM. The minimum variance and weights are dependent upon the form of the global unwindowed power spectrum, Sff.

If the optional argument nocross is set to 1, then only the diagnonal terms of Fij will be computed.


Wieczorek, M. A. and F. J. Simons, Minimum-variance multitaper spectral estimation on the sphere, J. Fourier Anal. Appl., 13, doi:10.1007/s00041-006-6904-1, 665-692, 2007.

Tags: python
Edit me