Balanced parametrizations of stable SISO all-pass systems in discrete time (Q1589790)

The authors consider discrete-time single-input/single-output (SISO) linear systems \(x(t+1)=Ax(t)+bu(t)\), \(y(t)=cx(t)+du(t)\), where \(A\in {\mathbb C}^{n\times n}\), \(b\in {\mathbb C}^{n\times 1}\), \(c\in {\mathbb C}^{1\times n}\), \(d\in {\mathbb C}\), and where \(x(t)\in {\mathbb C}^n\), \(u(t)\in {\mathbb C}\), \(y(t)\in {\mathbb C}\) denote the state, the input signal, and the output signal, respectively, at time \(t\in {\mathbb Z}\). The transfer function of such a system is given by \(g(z)=d+c(zI_n-A)^{-1}b\), which is a proper rational function of McMillan degree at most \(n\). The system \((A,b,c,d)\) with state space \({\mathbb C}^n\) is called minimal if all other systems with transfer function \(g(z)\) (realizations of \(g(z)\)) have state-space dimensions (orders) not less than \(n\). In this case \(n\) is equal to the McMillan degree of \(g(z)\). Minimal systems \((A_1,b_1,c_1,d_1)\), \((A_2,b_2,c_2,d_2)\) are i/o equivalent, i.e., have the same transfer function if and only if they are related by \((A_2,b_2,c_2,d_2)=(TA_1T^{-1},Tb_1,c_1T^{-1},d_1)\), where \(T\) is an \(n\times n\) nonsingular matrix. A discrete-time SISO system of a finite order \(n\) is called all-pass if its transfer function has modulus \(|g(z)|=1\) on the unit circle \(\{ z\in {\mathbb C}: |z|=1\}\). If, moreover, such a system is stable, i.e., the eigenvalues of matrix \(A\) belong to the closed unit disk \(\{ z\in {\mathbb C}: |z|\leq 1\}\), then \(g(1/z)\) is a rational inner function, i.e., has no poles on the closed unit disk and has modulus one on the unit circle. A minimal stable realization \((A,b,c,d)\) is called balanced if its controllability Grammian \(W_c\) -- the solution to the discrete-time Lyapunov equation \(W_c-AW_cA^H=bb^H\) -- coincides with its observability Grammian \(W_o\) -- the solution to the discrete-time Lyapunov equation \(W_o-A^HW_oA=c^Hc\) (here \(T^H\) stands for the Hermitean transpose of matrix \(T\)). The main result of the paper under review is the description of a parametrization of i/o equivalence classes of discrete-time minimal stable all-pass SISO systems of order \(n\). The uniquely-determined representative of each class is a system \((A,b,c,d)\) which is balanced and which has the property that the corresponding (square) reachability matrix \([b,Ab,A^2b,\ldots,A^{n-1}b]\) is upper-triangular and has positive entries on the main diagonal. This is a discrete analogue of the continuous-time balanced canonical form of \textit{R. J. Ober} [Int. J. Control 46, 643-670 (1987; Zbl 0634.93015)]. It is shown that the balanced canonical form in the discrete case can be parametrized by Schur parameters [see, e.g., the book \textit{N. I. Akhiezer}, The Classical Moment Problem, Olivier and Boyd, Edinburgh, Scotland (1965; Zbl 0173.41001) for the definition and properties of Schur parameters]. Let us remark that the only part of Proposition 2.4, that the authors consider as new, is part (ii)(a), which asserts that any discrete-time linear SISO system \((A,b,c,d)\) of order \(n\) having unitary matrix \(V=\left[\begin{smallmatrix} d & c \\ b & A \end{smallmatrix}\right]\) is a stable all-pass system, regardless of whether \((A,b,c,d)\) is minimal or not. However, this result is also well known, even in a more general formulation. Namely, if the input space, the output space, and the state space of system \((A,b,c,d)\) are finite or infinite dimensional Hilbert spaces, and \(V\) is a unitary operator, and the completely non-unitary part \(A_{cu}\) of the (contractive) operator \(A\) belongs to the class \(C_0\) (i.e., \(A_{cu}^k\rightarrow 0\) as \(k\rightarrow\infty\) in the strong operator topology), then the characteristic function \(g(1/z)\) of \(A\) (related to the transfer function \(g(z)\) of system \((A,b,c,d)\)) is inner, i.e., analytic on the open unit disk and having isometric operator boundary values at almost every point of the unit circle [\textit{B. Sz.-Nagy} and \textit{C. Foiaş}, Harmonic Analysis of Operators on Hilbert Spaces, North Holland, Amsterdam (1970; Zbl 0201.45003)]. Let us also remark that Lemma 5.1 of the paper under review asserts that for the set of inner functions the topology of pointwise convergence of Taylor coefficients and the topology of convergence in the \(H_2\)-norm on the unit circle coincide. The authors' proof is too complicated. It takes more than four pages of the text, and uses their parametrization of i/o equivalence classes of stable all-pass systems. However, the assertion of Lemma 5.1 easily follows from the well-known general statement: if a sequence \(\{ x_k\}\) in some Hilbert space converges weakly to the element \(x\) of the same space, and \(\|x_k\|\rightarrow\|x\|\) as \(k\rightarrow\infty\), then \(x_k\rightarrow x\) strongly, i.e., in the norm of the space. Indeed, the pointwise convergence of Taylor coefficients coincides with a weak convergence in the Hilbert space \(H_2\), and the \(H_2\)-norms of all inner functions are equal to 1, hence the pointwise convergence of Taylor coefficients of the sequence of inner functions implies its convergence in \(H_2\)-norm. Clearly, the strong convergence also implies the weak convergence of the same sequence.