MIMO l 1 optimal control problems via the polynomial equations approach (Q4450433)

The authors of this paper present new sub/superoptimization schemes for the general MIMO four-block \(\ell_1\) control problem \[ \begin{cases} {z\choose y}={G_{11}\;G_{12}\brack G_{21}\;G_{22}} {w\choose u}\\ u=C_y\\ G_{11}\in {\mathcal R}{\mathcal A}_{n_z\times n_w}, G_{12}\in {\mathcal R}{\mathcal A}_{n_z\times n_w}\\ G_{21}\in {\mathcal R}{\mathcal A}_{n_y\times n_w},G_{22}\in {\mathcal R}{\mathcal A}_{n_y\times n_w}\end{cases} \tag{1} \] where \(w\in\mathbb{R}^{n_w}\) represents exogenous inputs, \(z\in\mathbb{R}^{n_z}\) regulated outputs, \(y\in\mathbb{R}^{n_y}\) measurable outputs and \(u\in\mathbb{R}^{n_u}\) control inputs. The set of all asymptotically stable closed-loop error maps \(\varepsilon\) from \(w\) to \(z\), which may be achieved by any stabilizing \(C\), are described by the YJBK parametrization \(\varepsilon={\mathcal H}-{\mathcal U}{\mathcal Q}{\mathcal V}\) where \[ {\mathcal H}:=G_{11}+ G_{12}M \widetilde Y_GG_{12},\quad {\mathcal U}: =G_{12}M,\quad {\mathcal V}:=\widetilde MG_{21},\;{\mathcal Q}\in{\mathcal A}_{n_u\times n_y}, \] and \(G_{22}=NM^{-1}=\widetilde M^{-1}\widetilde N\) are expressed via right and left coprime factorizations with \(N,M,\widetilde N\) and \(\widetilde M\) polynomial matrices. \(Z(\cdot)\) denotes the \(Z\)-transform operator defined by \(Z(\widehat H)={\mathcal H}(d):= \sum^\infty_{k=0} \widehat H(k)d^k\), in the complex variable \(d=z^{-1}\), for any matrix sequence \(\widehat H=\{\widehat H(k)\}^\infty_{k=0}\), \({\mathcal A}_{p\times m}\) is the real normed linear space of all \(p\times m\) matrices \({\mathcal H}(d)\) which are \(Z\)-transform of some matrix sequence \(\widehat H\), and \({\mathcal R}{\mathcal A}_{p\times m}\) denotes the subspace of \({\mathcal A}_{p\times m}\) consisting of elements each of whose entries is a stable real rational function. A novel solution is derived by using polynomial techniques and simple algebraic conditions on the free parameter of the YJBK parameterization. In this way some unconstrained \(LP\) formulations are obtained, less affected by redundancy and solvable by standard \(LP\) solvers. A solution scheme based on solving sequences of increasingly larger finite-dimensional sub/super-optimal optimization problems is proposed. A suggestive example is presented in the end of this paper.