CONVERGENT AUTOMATIC-SYNTHESIS PROCEDURES FOR SAMPLED-DATA NETWORKS WITH FEEDBACK.

摘要

The purpose of the research described herein was to develop convergent automatic-synthesis procedures for sampled-data networks with feedback,when the input and desired response sequences for the network are given and the network is of a fixed order. The capability of iterative synthesis using performance feedback has been obtained. The conventional sum-squared-error performance criterion was found to result in error surfaces with multiple minima when applied to the synthesis of linear sampled-data networks with feedback. A modified quadratic error criterion was developed,consisting of the sum of the squares of the difference between the desired output of the network and the actual network output during synthesis,with the modification that the desired output,rather than the network output,was fed back. Minimization of this criterion is shown to converge to a unique solution. For the order of network chosen,if a solution exists that can yield exactly the desired response for the given input sequence. It is shown that minimization of the modified performance criterion will achieve this solution. (Author)