Martin Ivanov
2014-Oct-28 13:08 UTC
[R] spectral coherence & non-overlapping windows: confidence intervals?
Dear R users,
The confidence intervals for the squared coherence and phase, plotted by
plot.spec.coherence() and plot.spec.phase, respectively,
use the formulae 10.14 and 10.11 from Bloomfield, P. (1976) Fourier Analysis of
Time Series: An Introduction. Wiley:
gg <- 2/x$df
se <- sqrt(gg/2)
?z <- -qnorm((1 - ci)/2)
coh <- sqrt(x$coh);
(tanh(atanh(coh) + z * se))^2; # this is the 95% line for the coherence
cl <- asin(pmin(0.9999, qt(ci, 2/gg - 2) * sqrt(gg *? (coh^{ -2 } - 1)/(2 *
(1 - gg))))); # this is for the phase
Does somebody know how these formulae extend in case of calculating the
cross-spectrum via
K non-overlapping windows, i.e. via the Welch method? I would be very thankful
if you could direct
me to some literature. I tried finding something on the topic myself, but the
information is only
for the case when one smooths with spectral windows.
Best regards,
Martin