Covariance resetting in recursive least squares estimation

Describes several methods for the conditional modification of S(t) to improve the tracking properties of the recursive extended least squares estimator. Simulation studies show that they do provide marked improvements in estimator performances. The results also indicate that using trS(t) as a condition for modification of S(t) appears to provide the best overall performances. Guidelines are also given on how the conditional limits may be automatically specified. It should be noted that the results presented pertain to an open loop situation. In the case of closed loop adaptive control, then a further requirement may be that parameter tracking should be `smooth´ so that the calculated changes in control would not be excessive