Principles of System Identification

224 Principles of System Identification: Theory and Practice

reason is that auto-regressive models result in linear-in-parameter predictors, whereas MA models

produce predictors that are non-linear in parameters. Linear estimators are advantageous because

they lead to unique solutions (especially when combined with least squares methods, see Chapter

14) and are computationally simple, whereas non-linear estimators require numerical optimization

algorithms, which most often deliver locally optimal solutions.

Listing 9.2 MATLAB code for Example 9.9

% Generate the random signal

ek = randn (2000 ,1) ;

dvec = [1 -1.2 0.35];

vk = filter(1 ,dvec , ek );

% Pl