Optimal deconvolution filter design under parameters perturbation in transmission channels (Q1911854)

The problem considered is that of filtering and smoothing a signal masked by additive noise, when some parameters are randomly perturbed. The choices for the different entries in the model are as follows: 1) Transmission channel: \(H(q^{-1},\alpha+\delta\alpha)\) with \(q^{-1}:=\) backward shift operator, \(\alpha:=\) vector parameter (non random), \(\delta\alpha:=\) random perturbation of the vector parameter with known, small variance, independent of the signals in the system. 2) Signal: \(u(k)=G(q^{-1})v(k)\), where \(v\) is a white noise. 3) Received signal: \(y(k)= H(q^{-1},\alpha+\delta\alpha)u(k)+ w(k)\), where \(w(k)=D(q^{-1},\beta+\delta\beta)n(k)\), with \(n\) a white noise independent of \(v\), and \(\beta\) and \(\delta\beta\) are to be interpreted analogously to \(\alpha\) and \(\delta\alpha\). 4) Criterion: \(J= E[(u(k-l)-\widehat u(k-l\mid k))^2]\) with \(\widehat u(k-l\mid k)= F(q^{-1})y(k)\). 5) Objective: find \(F\) that minimizes \(J\). The paper exhibits the minimizer \(F\). An extension requiring only known bounds on the covariance matrices of the parameters' perturbations is provided, and the results of a simulation study are reported.