In the case of periodic deterministic signals, the signals may be
described either in terms of periodicity (in the time domain) or in terms of their line spectra (in the frequency domain). No new information is gleaned by moving from one domain to the other. Practical considerations usually determine which representation is the most useful in specific applications. In dealing with stochastic signals it was seen that the appropriate measures by which they may be summarized are the correlation & spectral density functions. Wiener & Khinchin first drew attention to the fact that correlation & spectral density functions form Fourier transform pairs. This observation is observed in the case of autocorrelation function & the power spectral density function as shown below: The definitions of the autocorrelation & the power spectral density functions over samples 2T long gives:
Rxx(λ)= lim T→inf 1/2T -T⌠+T x(t).x(t+ λ)dt
Sxx(ω)= lim T→inf 1/2T {X(ω).X*(ω)}
and taking the Fourier transform (FT) of the autocorrelation function
gives
FT {Rxx(λ)}= -T⌠+T{lim T→inf 1/2T -T⌠+T x(t).x(t+ λ)dt} e-jωλ .dλ
Substituting μ for (t+ λ) & noting that dλ becomes dμ, then
FT {Rxx(λ)}= -T⌠+T{lim T→inf 1/2T -T⌠+T x(t).x(μ)dt} e-jω(μ-t) .dμ
results which, after expanding the exponential term & separating the 2 integrals, becomes
It is recognized that the 2 integrals represent the Fourier transform of
x(t), i.e. X(ω) & its conjugate complex X*(ω), and thus:
FT {Rxx(λ)}= lim T→inf 1/2T {X(ω).X*(ω)} = Sxx(ω)
By identical reasoning it can be shown that the cross-correlation function & the cross spectral density function also form Fourier transform pairs. The Wiener-Khinchin are usually written formally as:
Sxx(ω)= -inf ⌠+inf Rxx(λ). e-jωλ .dλ
Rxx(λ)= 1/2π -inf ⌠+inf Sxx(ω). ejωλ .dω
These 2 equations exhibit the fact that there is no new information
generated by moving from the time domain to the frequency domain or vice-versa. However, in some applications of industrial diagnostics it is easier to interpret the underlying phenomena in one domain or the other. One use of the first of the 2 equations is to use it to calculate the power spectral density from a known estimate of the autocorrelation function. It is now more common to use the Fast Fourier Transform (FFT) to calculate the power spectrum directly from the raw time series associated with the signal under investigation.