SPECTRAL ANALYSIS TECHNIQUES FOR SYSTEM IDENTIFICATION 



J. S. Bendat 



1 . Introduction 



In the past few years, new analytical procedures have been developed 

 for optimum linear system identification using spectral analysis techniques 

 I1-3J. These techniques apply to: 



(a) single input/single output problems. 



(b) single input/multiple output problems. 



(c) multiple input/single output problems. 



(d) multiple input/multiple output problems. 



The key to carrying out this work is the implementation of new practical 

 computational algorithms showing how to decompose output records from 

 input records. Data are allowed to be realistic stationary random or tran- 

 sient random records with arbitrary correlation properties between the 

 records. 



This problem was previously treated in UJ and in other books where 

 the general solution is derived by involved matrix computations or equiva- 

 lent algebraic operations that are difficult to carry out and interpret. 

 Complicated formulas were given for desired multiple coherence functions 

 and partial coherence functions which did not provide significant engineering 

 insight to inherent relationships of interest. Straightforward engineering 

 interpretations are now being obtained by the new procedures. 



The purpose of this paper is to outline some special features in these 

 recent developments to help engineers conduct this analysis. Many 



110 



