Computational Precision Requirements for Optimal Weights in Adaptive Processing

The required accuracy for computing the estimated optimum weights of an adaptive processor has been analyzed by investigating the effects of errors in computing the inverse matrix. It is shown that the required precision depends upon the matrix. An equation for the general case is derived. Several special cases are considered in detail. It is shown that the case of a single interference source requires the highest precision. The least stressing case is identifi\´ed and compared to the worst case. The requirements for a "typical" case are also considered. A comparison of the requirements for the covariance matrix estimation technique and for adaptive weight implementation using gradient descent techniques is given. It is shown that there is a dichotomy in that cases that do not stress one technique tend to stress the other.