Professional Documents
Culture Documents
23, 2002
Strict Stationarity
o All statistical properties are invariant with time
Weak Stationarity
o Mean µ x and autocorrelation ρ x (τ ) are invariant with time
Ergodic Process
o Stationary process where the statistical properties of one sample
are identical to the ensemble statistics
Auto-correlation
T
1
•
T →∞ T ∫
R x (τ ) = lim (x(t ) − x )(x(t − τ ) − x )dt
0
dR (τ = 0)
• =0
dτ
• R(τ = ∞ ) = 0
• If x(t ) = y 1 (t ) + y 2 (t ) + y 3 (t ) + ..... + y n (t ) where the y ‘s are independent, then
R x (τ ) = R y 1 (τ ) + R y 2 (τ ) + ..... + R yn (τ ) and since R x (τ = 0) = σ x2 , then
σ 2 x (τ ) = σ 2 y 1 + σ 2 y 2 + ..... + σ 2 yn
Page 1 of 2
CEE490b Jan. 23, 2002
Power Spectrum
∞
• Two-sided spectrum G(f ) = ∫ R(τ )e −i 2πfτ dτ
−∞
∞
S(f ) = 4 ∫ R(τ ) cos( 2πfτ )dτ
0
∞
This is a “Cosine Fourier Transform Pair”
R(τ ) = ∫ S(f ) cos( 2πfτ )df
0
• The area under the power spectrum equals the variance, σ 2
• Dimensions of power spectrum are (dimension of x 2 )/frequency, for example
if “x” is displacement in metres, then the unsits of the power spectrum are:
m 2 / Hz
• Power spectra are often normalized by the variance, σ 2 , so the area is 1.0
• Examples of Random Processes:
Page 2 of 2