A residual-based selective window for robust recursive least squares estimation
S.F. Hsieh, K.J.R. Liu · 1991
An algorithm performing recursive least squares (RLS) estimation is proposed. It is based on selectively rejecting outliers arising from noise spikes; therefore, this method can avoid the bias of parameters estimation due to some large noise perturbations. Unlike a sliding fixed-window scheme, this windowing scheme can be noncontinuous. It depends on the estimated level of observed errors (residual). By monitoring the residuals in a recursive manner, one can effectively remove those spurious observed data by downdating them. The proposed scheme is useful, especially when some short-time large interferences perturb the system occasionally. In this respect, it outperforms existing schemes, either exponentially growing or sliding window.>