UBC Theses and Dissertations
A special purpose analog computer for statistical system identification Fieguth, Werner
An iterative method of system identification based on solving the integral equation [Formula omitted] for h(σ) at ten equally spaced points (σ[subscript i] = 1,2,..,10) is described. Replacing the above integral by a finite sum at ten different values of τ results in a set of ten equations in the ten unknowns h(σ[subscript i]). A statistically identical and much more easily handled set of equations, obtained by using one-sample products in place of the actual correlation functions, is then solved by a Gauss-Seidel-like iteration method, the convergence properties of which show this approach to the identification problem to be a useful one for a large class of system input signals. A simple computer to realize the above identification method is described in some detail. The use of a simple quantization form of correlation allows shift registers to carry out the required delay operations. Storage for the computer's estimates of the h(σ[subscript i]) is in the form of step motor driven potentiometers, which also carry out one of the multiplication operations. The very encouraging results of a number of relatively realistic identification tests using the computer are given.
Item Citations and Data