Error propagation properties of recursive least-squares adaptation algorithms (Q1064310)

Many algorithms giving effective estimations of the coefficients of the equation \(y(t)+a_ 1y(t-1)+...+a_ ny(t-n)=e(t),\) \(t=0,1,..\). for observation data y(t) are known. Here e(t) is, for example, white noise. It is shown that some of the known algorithms are stable and some are unstable (exponentially) with respect to roundoff error. Results have various applications, in particular, in the theory of adaptive systems.