This tells me it takes about 2.7 seconds to calculate 63661 intercepts. (I didn’t run a comparison with interp1, so I can’t compare the times.) It seems to me to produce the result you want:
an = 1:0.2:2e5;
clean = sin(an);
noise = -0.03 + (0.06).*rand(size(clean));
data = clean + noise;
% Find zero crossings:
T0 = clock;
cnv = data .* circshift(data,[0 1]);
zx = find(cnv < 0);
% Interpolate zeros:
for k1 = 2:size(zx,2)-1
ixrng = zx(k1)-2:zx(k1)+2;
X = an(ixrng);
Y = data(ixrng);
b = [ones(size(X)); X]'\Y';
Xi(k1) = -b(1)/b(2);
end
T1 = clock;
DT = etime(T1,T0)
figure(1)
plot(an, data)
hold on
% plot(an(zx), zeros(size(zx)), 'b*')
plot(Xi, zeros(size(Xi)), 'pr')
hold off
axis([0 25 ylim])
grid
It implements a simple linear least-squares regression, then solves for the X-intercepts.
