Discretization of continuous linear dynamic systems. Analysis of the methods (Q1058498)

The systems \[ (1)\quad dx(t)/dt=Px(t)+Qu(t),\quad y(t)=Rx(t),\quad t\in [0,\infty), \] \[ (2)\quad x_{k+1}=Ax_ k+Bu_ k,\quad y_ k=Cx_ k,\quad k=0,1,... \] are considered. Here u(t), y(t), \(u_ k\), \(y^ k\) are scalar, x(t), \(x_ k\) are n-vectors, all matrices are constant. System (2) is a simplified model of system (1). It is obtained from (1) by one of the standard discretization methods: the method of standard z- transform (ST) the matched z-transform (MT), the methods of piecewise constant or constant ramp approximation of input (PCA or CRA respectively). If ST is applied, then \(A=\exp (dP)\), \(C=R\), \(B=Q\), and for PCA \(A=\exp (dP)\), \(C=R\), \(B=P^{-1}(A-I)Q\), where I is a unit matrix and d is the sampling period. Analogous formulae are also valid for CRA. Let \(W(\lambda)=R(\lambda I-P)^{-1}Q=r(\lambda -\mu_ 1)...(\lambda -\mu_ m)(\lambda -\lambda_ 1)^{-1}...(\lambda - \lambda_ n)^{-1}\) and \(\chi (\lambda)=C(\lambda I-A)^{-1}B\) be the transfer functions of systems (1), (2) respectively. Define the standard polynomials \(\gamma_ 0(\lambda)=1\), \(\gamma_{\ell}(\lambda)=\gamma_{1,\ell}\lambda^{\ell - 1}+\gamma_{2,\ell}\lambda^{\ell -2}+...+\gamma_{\ell,\ell},\) where \(\ell =1,2,... \), \[ \gamma_{j,\ell}=(1/\ell !)\sum^{j}_{i=1}(- 1)^{i+1}C^{i-1}_{j+1}(j+1-i)^{\ell} \] and \(C^ M_ N\) are binomial coefficients. Denote the zeros of W(\(\lambda)\) by \(\mu_ 1,\mu_ 2,...,\mu_ m\). Then the zeros of \(\chi\) (\(\lambda)\) may be divided into two groups: m zeros have a form of \(\exp (d\mu_ j(d))\), where \(\mu_ j(d)\to \mu_ j\) as \(d\to 0\), and other zeros tend to the roots of the standard polynomial \(\gamma_{\ell}(\lambda)\). Here \(\ell =n-m-1\) for the ST method, \(\ell =n-m\) for PCA and \(\ell =n-m+1\) for CRA. Therefore, in case \(\ell =0\) or 1 the polynomial \(\gamma_{\ell}(\lambda)\equiv 1\) and the used method coincide asymptotically with the MT-method.