Abstract
Three different techniques for linear SISO system model reduction are presented. Each method is based on an integral error criterion in the frequency domain which is derived from an equivalent time-domain criterion. Two of the methods are of the linear least-squares class and are therefore easily solved. The third method falls into the broad class of non-linear least-squares problems and demands the use of a powerful optimization technique. The efficiency of each of the methods is compared to previous curve-fitting least-squares techniques and recommendations for their use are made.