General solution of certain matrix equations arising in filter design applications (Q848589)

An explicit expression is obtained for all rectangular Toeplitz matrices \(\mathbf{B}\) and \(\mathbf{C}\) which verify the equation \({\mathbf{BB}}^H+{\mathbf{CC}}^H=a\mathbf{I}\) for some \(a>0\). This equation is rewritten as the polynomial Bézout equation on the unit torus \(|B(z)|^2+|C(z)|^2=a\), \(|z|=1\), where \(B(z)\) and \(C(z)\) are complex polynomials of degree at most \(L-1\), \(L\in\mathbb{N}\). The polynomial solutions are obtained by using the filter approach. An orthogonal complex filter of length \(2L\) is a complex vector \({\mathbf h}=(h_1,h_2,\dots,h_{2L})\) with \(h_1h_{2L}\neq0\) which is orthogonal to its even shifts, i.e. \(\sum_{n=1}^{2L-2k}h_k\overline{h}_{n+2k}=0\), \(k=1,\dots,L-1\). If \(\mathbf{h}\) has unit Euclidian norm, \(\mathbf{h}\) is called paraunitary complex filter. A general explicit parametrization of all complex orthogonal filters is provided, which is similar to the one obtained by the author in two previous papers for real filters. The simplest general expression for paraunitary complex filters is derived, as well as a procedure to design the general solution of the polynomial equation. It is emphasized that in this approach the solutions \(B(z)\) and \(C(z)\) are built directly, without need of spectral factorization or root finding procedure. Some examples of paraunitary filters design (with \(L=2\)) are presented.