Chapter 3: Spectral analysis of discrete time stationary processs Flashcards
State the spectral representation theorem
There ists an orthogonal process {Z(f)} defined on [-1/2,1/2] such that
Xt = int_-1/2^1/2 {exp(i 2π f t △t) dZ(f)} for all t.
By othogonal increments dZ*(f’) dZ(f) = …
d S(I)(f) or S(f) df for f'=f 0 otherwise
What are the properties of {Z(f)}?
- expectation: E{dZ(f)} = 0 for all |f|<1/2
- variance: var{dZ(f)} = E{|dZ(f)|^2} = dS(I)(f) = S(f) df for all |f|<1/2
- uncorrelated: cov{dZ(f’),dZ(f)} = E{dZ(f’)dZ(f)} - E{dZ(f’)} E{dZ(f)} = E{dZ*(f’)dZ(f)} for all |f|<1/2
Write the Fourier transform pair of S(f) and s_tau
S(f) = sum_τ=-∞ ^∞ sτ exp(- i 2π f τ) sτ = int_-1/2^1/2 S(f) exp(i 2π f τ) df = int_-1/2^1/2 exp(i 2π f τ) d S(I)(f)
Define the spectral density funcion S(f) and state its 4 properties
S(f) df is the average contribution (over all realizations) to the power from components with frequencies in a small interval about f.
- S(I)(f) = int_-1/2 ^f S(f’)df’
- 0 <= S(I)(f) <= sigma^2, and S(I)(f) increasing, ie. S(f) >= 0
- S(I)(-1/2) = 0, S(I)(f) = sigma^2, int_-1/2^1/2 S(f) df = sigma^2
- f < f’ => S(I)(f) <= S(I)(f’), S(-f)=S(f)
What is the Nyquist frequency version of the spectral representation theorem (for discrete and continuous time processes) ?
for discrete {Xt}:
Xt = int_-fn ^fn {exp(-2π f t △t) dZ(f)} for fn=1/(2△t) called the Nyquist frequency
for continuous Xt:
Xt = int_-∞ ^∞ {exp(-2 π f t △t) dZ(f)} for △t → 0
Define Linear Time-Invariant digital filter and its spectral density
An LTI is a transformation yt = L{xt} with
- scale-preservation
- superposition
- time-invariance
State the 4 classifications of spectra S(I)(f) an their characteristics
- purely continuous
- purely discrete
- mixed: discrete + non-white continuous
- discrete: discrete + white continuous noise
What is the sdf of a LTI yt,f = L{ξt,f} where ξt,f = exp(i 2πft) ?
Sy(f) = |G(f)|^2 Sx(f) where G(f) = sum_u=-∞ ^∞ g(u) exp(- i 2π f u) for |f|<1/2
for AR(p), MA(q) or ARMA(p,q) replace z by exp(- i 2π f) in G(z) and Sx(f) = |G(f)|^2 Sε(f)
What are the two methods for aliasing?
- method 1: Sxt (f) discrete and Sx(t) (f) continuous
Sxt (f) = sum_k=-∞ ^∞ Sx(t) (f+k/△t) for |f|<1/2 - method 2: folding method using graphs (fold in the tails outside of [-fn,fn])