Skip to content Skip to navigation

Connexions

You are here: Home » Content » Orthonormal Filter Banks

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 author

Recently Viewed

Orthonormal Filter Banks

Module by: Ivan Selesnick

How can we choose the two filters h 0 n h 0 n , h 1 n h 1 n so that the two-channel analysis filter bank represents an orthonormal transformation of xn x n (so that QTQ=I Q Q I )? In other words, how do we choose the filters so as to ensure yn=xn yn xn ?

Look at Yz Y z in terms of Xz X z . Using multirate identities, we get:

Yz=12 H 0 1z H 0 zXz+ H 0 -zX-z+12 H 1 1z H 1 zXz+ H 1 -zX-z Y z 1 2 H 0 1 z H 0 z X z H 0 z X z 1 2 H 1 1 z H 1 z X z H 1 z X z (1)
or
Yz=12 H 0 z H 0 1z+ H 1 z H 1 1z+12 H 0 -z H 0 1z+ H 1 -z H 1 1z Y z 1 2 H 0 z H 0 1 z H 1 z H 1 1 z 1 2 H 0 z H 0 1 z H 1 z H 1 1 z (2)
If yn=xn yn xn , then QTQ=I Q Q I and the analysis filter bank represents an orthonormal transform.

We have yn=xn yn xn only if Yz=Xz Yz Xz so we must have:

H 0 z H 0 1z+ H 1 z H 1 1z=2 H 0 z H 0 1 z H 1 z H 1 1 z 2 (3)
H 0 -z H 0 1z+ H 1 -z H 1 1z=0 H 0 z H 0 1 z H 1 z H 1 1 z 0 (4)
The problem is to find H 0 z H 0 z and H 1 z H 1 z that satisfy these two perfect reconstruction conditions.

Suppose we set

H 1 z=z-L H 0 -1z H 1 z z L H 0 1 z (5)
with LL odd, or, equivalently, h 1 n=-1L-n h 0 L-n h 1 n 1 L n h 0 L n Then the left hand side of Equation 4 becomes H 0 -z H 0 1z+-z-L H 0 1zzL H 0 -z H 0 z H 0 1 z z L H 0 1 z z L H 0 z or (because LL is odd) H 0 -z H 0 1z- H 0 1z H 0 -z H 0 z H 0 1 z H 0 1 z H 0 z which is zero. So with the choice for H 1 z H 1 z if Equation 5 the condition Equation 4 is automatically satisfied.

Equation 3 becomes H 0 z H 0 1z+z-L H 0 -1zzL H 0 -z=2 H 0 z H 0 1 z z L H 0 1 z z L H 0 z 2 or

H 0 z H 0 1z+ H 0 -z H 0 -1z=2 H 0 z H 0 1 z H 0 z H 0 1 z 2 (6)
So we need to choose H 0 z H 0 z so as to satisfy this equation. For convenience, define r 0 n r 0 n to be the autocorrelation of h 0 n h 0 n , r 0 n h 0 n* h 0 -n r 0 n h 0 n h 0 n or equivalently
R 0 z H 0 z H 0 1z R 0 z H 0 z H 0 1 z (7)
Then Equation 6 becomes R 0 z+ R 0 -z=2 R 0 z R 0 z 2 or equivalently r 0 n+-1n r 0 n=2δn r 0 n 1 n r 0 n 2 δ n Note that r 0 n+-1n r 0 n=2δn r 0 2n=δn r 0 n 1 n r 0 n 2 δ n r 0 2 n δ n That means, the autocorrelation of the filter h 0 n h 0 n must be a halfband filter,
r 0 2n=δn r 0 2 n δ n (8)
We can therefore obtain h 0 n h 0 n by performing spectral factorization of r 0 n r 0 n . Given r 0 n r 0 n , we can find h 0 n h 0 n so as to satisfy Equation 7 provided that Rω0 R ω 0 for all ωω and provided that r 0 n r 0 n is symmetric ( r 0 n= r 0 -n r 0 n r 0 n ).

In conclusion, the two-channel analysis filter bank can be an orthogonal transform if the filters H 0 z H 0 z and H 1 z H 1 z satisfy the appropriate conditions (the perfect reconstruction condition Equation 6).

Comments, questions, feedback, criticisms?

Send feedback