Least-squares \(\nu\) th-order polynomial estimation of signals from observations affected by non-independent uncertainty (Q2495996)

Let \(z(k)\) and \(y(k)\) be \(n\times1\) vectors describing the signal and its observation at time \(k\). Suppose that the observation equation is given by \[ y(k)=U(k)z(k)+v(k), \] where the signal \(\{z(k),k\geq0\}\) is a zero-mean stochastic process, the additive noise \(\{v(k),k\geq0\}\) is a zero-mean white sequence and the multiplicative noise \(\{U(k),k\geq0\}\) is a sequence of Bernoulli random variables with \(P[U(k)=1]=p(k)\neq0\), satisfying that the conditional probability \(P[U(k)=1\mid U(j)=1]\) is independent of \(j\) for \(j<k\). The aim of the paper is to obtain a least-squares \(\nu\)-th order polynomial estimator of the signal \(z(k)\) based on the observations \(\{y(1),\ldots,y(L)\}\), \(\nu\) being an arbitrary integer. More precisely, the authors are interested in obtaining a least squares \(\nu\)-th order polynomial filter \((L=k)\) and a fixed-point smoother \((L>k)\) of the signal. At the end of the paper an application how to estimate a signal transmitted in a multichannel system is presented.