Dear R community,
I would like to conduct a simulation study that involves the
generation and recovery of times series spectra. Spectrum analysis
is a new area for me so I am very much in the crawling stage at this point.
I am having difficulty understanding the output of the spec.pgram (or
spectrum)
function. Specifically, I do not understand the metric of the results. In
my
simulation the recovered coefficients correlate .99 with the generating
coefficients, so everything appears in order -- accept for my
confusion regarding the metric of the recovered coefficients.
I have written a function (makeWave) to generate the series. I then call
this
function with the required arguments. Next I analyze the resulting series
with
spec.pgram and compare the results with the generating coefficients
(at the appropriate frequencies). In a small simulation study the
generating
and recovered coefficients correlated .99 -- however the metric of the two
sets of coefficients differ by several orders of magnitude.
I would be very happy to send the actual code of this simulation
(less than 2 pages of code) to anyone who could help me understand how
to scale my recovered coefficients to the metric of the generating
parameters.
Specifically, I am looking for a general solution to the scaling problem (if
one exists).
Thank you in advance for any and all help.
Niels Waller
Vanderbilt University
R 1.9.1
Windows XP
##------------------------------------------------------------------##
##FUNCTION: makeWave
##Purpose: to generate periodic time series (no white noise)
##Arguments
## c0 :: coefficient for frequency 0
## c.n :: vector of (complex) coefficients for frequencies 1...+n
## cminus.n :: vector of (complex) coefficients for frequencies 1...-n
## N :: number of time points in generated wave
## f :: fundamental frequency of wave
makeWave<-function(c0, c.n, cminus.n, N, f){
k<-1:N #k = time point
x<-rep(0,N)
w <- 2*pi*f
for(t.i in 1:N){ ## over time t.i
x[t.i] <- c0
temp<-0
for(j in 1:length(c.n)){ ## over frequency j
temp<-temp + c.n[j] * exp(1i*w*j*t.i) +
cminus.n[j] * exp(1i*w*j*t.i)
}
x[t.i]<-x[t.i]+temp
}
x ## composite wave
}