nufft_best_gauss.m

 function [sig, kernel, kernel_ft] = nufft_best_gauss(J, K_N, sn_type)
%function [sig, kernel, kernel_ft] = nufft_best_gauss(J, K_N, sn_type)
%|
%| Return "sigma" of best (truncated) gaussian for NUFFT
%| with previously numerically-optimized width
%|
%| in
%|	K_N		K/N
%|	J		# of neighbors used per frequency location
%|	sn_type		'zn' or 'ft' (latter recommended)
%|
%| out
%|	sig		best sigma
%|	kernel		string for inline kernel function, args (k,J)
%|	kernel_ft	string for Fourier transform function, arg: (t)
%|
%| Copyright 2002-4-11, Jeff Fessler, University of Michigan

if nargin < 1, help(mfilename), error(mfilename), end

if nargin == 1 && streq(J, 'test')
	[sig, kernel, kernel_ft] = nufft_best_gauss(6)
	clear sig kernel kernel_ft
return
end

if nargin < 2, K_N = 2; end
if nargin < 3, sn_type = 'ft'; end

if K_N ~= 2, error 'only K/N=2 done', end

%if ir_is_octave % avoid annoying load warning
%	warning('off', 'Octave:load-file-in-path', 'local')
%end

nufft_dir = path_find_dir('nufft');
s = load([nufft_dir filesep 'param-data/nufft_gauss2.mat']);

ii = find(J == s.Jgauss2);
if length(ii) ~= 1
	disp(s.Jgauss2(:)')
	error 'only above J values done'
end
if streq(sn_type, 'ft')
	sig = s.Sgauss2.ft(find(J == s.Jgauss2));
elseif streq(sn_type, 'zn')
	sig = s.Sgauss2.zn(find(J == s.Jgauss2));
else
	error 'bad sn_type'
end
[kernel, kernel_ft] = nufft_gauss('string', J, sig);