Skip to content Skip to navigation

Connexions

You are here: Home » Content » Gauss-Markov Theorem and Wiener Filtering

Navigation

Content Actions
  • Download module PDF
  • Add to ...
    Add the module to:
    • My Favorites
    • A lens
    • An external social bookmarking service
    • My Favorites (What is 'My Favorites'?)
      'My Favorites' is a special kind of lens which you can use to bookmark modules and collections directly in Connexions. 'My Favorites' can only be seen by you, and collections saved in 'My Favorites' can remember the last module you were on. You need a Connexions account to use 'My Favorites'.
    • A lens (What is a lens?)

      Definition of a lens

      Lenses

      A lens is a custom view of Connexions content. You can think of it as a fancy kind of list that will let you see Connexions through the eyes of organizations and people you trust.

      What is in a lens?

      Lens makers point to Connexions materials (modules and collections), creating a guide that includes their own comments and descriptive tags about the content.

      Who can create a lens?

      Any individual Connexions member, a community, or a respected organization.

    • External bookmarks
  • E-mail the authors

Recently Viewed

Gauss-Markov Theorem and Wiener Filtering

Module by: Clayton Scott, Robert Nowak

Let xx and yy be jointly Gaussian distributed: xy∼mxmy R xx R xy R yx R yy x y m x m y R xx R xy R yx R yy Then the conditional distribution of y y given x x is y | x ∼my+ R yx R xx -1x-mxQ y | x m y R yx R xx x m x Q where Q= R yy - R yx R xx -1 R xy Q R yy R yx R xx R xy

We know that the conditional mean y ̂=my+ R yx R xx -1x-mx y m y R yx R xx x m x is the best estimate of yy give xx in a (mean) squared error sense.

Example 1

x=y+W x y W where y∼0 R yy y 0 R yy and W∼0 R WW W 0 R WW and yy and WW are independent. xy∼00 R yy + R WW R yy R yy R yy x y 0 0 R yy R WW R yy R yy R yy y | x ∼ R yx R xx -1x R yy - R yx R xx -1 R xy y | x R yx R xx x R yy R yx R xx R xy y ̂= R yy R yy + R WW -1x=Hx y R yy R yy R WW x H x where H H is the Wiener filter. Minimum MSE estimator of yy given xx. R yx = R yy R yx R yy and R xx = R yy + R WW R xx R yy R WW .

Direct Optimization

x=y+W x y W y ̂=Gx y G x H=argminGEy-GxTy-Gx H G y G x y G x where Ey-GxTy-Gx y G x y G x is the MSE.

MS=Ey-GxTy-Gx=Etry-Gxy-GxT=trEy-Gxy-GxT MS y G x y G x tr y G x y G x tr y G x y G x (1)
Minimizing MSE is equivalent to minimizing
ε2=Ey-Gxy-GxT= R yy -G R xy - R yx GT+G R xx GT ε 2 y G x y G x R yy G R xy R yx G G R xx G (2)
Taking the derivative with repsect to GG ∂∂Gε2=-2 R yx +2G R xx =0 G ε 2 -2 R yx 2 G R xx 0 This implies
H= R yx R xx -1= R yy R yy + R WW -1 H R yx R xx R yy R yy R WW (3)

Figure 1: y= y ̂+y- y ̂ y y y y , where y- y ̂ y y is the error. The error and the estimate are statistically orthogonal to each other.
Geometrical Interpretation
Geometrical Interpretation (geoint.png)

Orthogonality Condition

The optimal Wiener filter H= R yy R yy + R WW -1 H R yy R yy R WW satisfies the following condition Ey- y ̂ y ̂T=0 y y y 0

Ey- y ̂ y ̂T=EyxTHT-HxxTHT= R yx HT-H R xx HT= R yy HT-H R yy + R WW HT= R yy R yy + R WW -1 R yy - R yy R yy + R WW -1 R yy + R WW R yy + R WW -1 R yy =0 y y y y x H H x x H R yx H H R xx H R yy H H R yy R WW H R yy R yy R WW R yy R yy R yy R WW R yy R WW R yy R WW R yy 0 (4)

The Classical Wiener Filter

Figure 2: yt y t is a stationary random signal, wt w t is stationary random noise, and the filter is H H
Figure 2 (block1.png)
We want to find Hω H ω that minimizes the MSE
ε2=E yt ̂-yt2= R ŷ ŷ 0-2 R y ŷ 0+ R yy 0 ε 2 y t y t 2 R ŷ ŷ 0 2 R y ŷ 0 R yy 0 (5)
where R yy τ=Eytyt+τ R yy τ y t y t τ . We can express the MSE in the frequency domain by noting that R yy 0=12π∫-∞∞ S yy ωdω R yy 0 1 2 ω S yy ω where S yy ω S yy ω is the power spectrum of yt y t .

Recall:

R yy τ R yy τ and S yy ω S yy ω are FT pairs: S yy ω=∫-∞∞ R yy τⅇ-ⅈωτdτ S yy ω τ R yy τ ω τ R yy τ=12π∫-∞∞ S yy ωⅇⅈωτdω R yy τ 1 2 ω S yy ω ω τ

Random Signal Response of Linear Systems

Figure 3: xt x t is a stationary random process and H H is a linear time-invariant system.
Figure 3 (block2.png)
yt ̂=∫-∞∞huxt-udu y t u h u x t u
E yt ̂=∫-∞∞huExt-udu= m x ∫-∞∞hudu y t u h u x t u m x u h u (6)
since Ext-u x t u is a constant. In particular, if xt x t is zero-mean then so is yt ̂ y t .

Autocorrelation of Output Process

R ŷ ŷ τ=E yt ̂ yt+τ ̂=E∫hsxt-sds∫huxt+τ-udu=∫∫Ext-sxt+τ-uhshudsdu=∫∫ R xx τ+s-uhshudsdu= R xx τ*h-τ*hτ R ŷ ŷ τ y t y t τ s h s x t s u h u x t τ u u s x t s x t τ u h s h u u s R xx τ s u h s h u R xx τ h τ h τ (7)

Power Spectrum

S ŷ ŷ ω= S xx ωHω¯Hω= S xx ω∥Hω∥2 S ŷ ŷ ω S xx ω H ω H ω S xx ω H ω 2 (8)

Cross-Correlation and Cross-Spectrum

R ŷ x τ=E yt ̂xt+τ=E∫hsxt-sdsxt+τ=∫hs R xx τ+sds= R xx τ*h-τ R ŷ x τ y t x t τ s h s x t s x t τ s h s R xx τ s R xx τ h τ (9)
This implies S ŷ x ω= S xx ωHω¯ S ŷ x ω S xx ω H ω R ŷ ŷ 0=12π∫-∞∞∥Hω∥2 S xx ωdω R ŷ ŷ 0 1 2 ω H ω 2 S xx ω R y ŷ 0=12π∫-∞∞Hω¯ S yx ωdω R y ŷ 0 1 2 ω H ω S yx ω S xx ω= S yy ω+ S ww ω S xx ω S yy ω S ww ω S yx ω= S yy ω S yx ω S yy ω since yt y t and wt w t are independent. Thus, the expression for the MSE becomes ε2=12π∫-∞∞∥Hω∥2 S yy ω+ S ww ω-2Hω¯ S yy ω+ S yy ωdω ε 2 1 2 ω H ω 2 S yy ω S ww ω 2 H ω S yy ω S yy ω ε2 ε 2 is minimized by minimizing the integrand each frequency ωω. This implies Hω S yy ω+ S ww ω= S yy ω H ω S yy ω S ww ω S yy ω Hω= S yy ω S yy ω+ S ww ω H ω S yy ω S yy ω S ww ω

Example 2

Figure 4
Figure 4 (syy.png)
Figure 5
Figure 5 (sww.png)
Figure 6
Figure 6 (h.png)

Comparison

Signal Vector (discrete-time)

H= R yy R yy + R ww -1 H R yy R yy R ww Here H H is a matrix, R yy R yy is the signal "power", and R ww R ww is the noise "power".

Classical (continuous-time)

Hω= S yy ω S yy ω+ S ww ω= S yy ω S yy ω+ S ww ω-1 H ω S yy ω S yy ω S ww ω S yy ω S yy ω S ww ω (10)
Which means that the Wiener Filter is defined as the ratio of signal power to the sum of signal power and noise power.

Comments, questions, feedback, criticisms?

Send feedback