Orthogonal filters have applications in many areas of signal processing. In this paper, the author presents a simple and explicit expression for real orthogonal filters.
If the dimension of the filter is L, then the new parametrization is optimal; namely, it depends on L/2+1 real numbers, which is exactly the degree of freedom of the problem.
The steps of the clever construction are as follows:
Denote h1, a1, a2, ... , aL/2 the parameters. With a1, a2, ... , aL/2 a lower triangular matrix is defined. Put a = (a2, ... , aL/2)t, b =(I+AAt)-1a and c = -Atb. Then, by Theorem 2, h = (h1, ... , hL) is an orthogonal filter if and only if h2=h1a1, heven = h1 b and hodd = h1c, where heven = (h4, h6, ... , hL-2, hL)t and hodd = (h3, h5, ... , hL-1)t. By normalization of the general formula, one gets an expression for paraunitary filters as well.
The results are made completely explicit for four-taps low-pass paraunitary filters.