A real-time analogue computer for the estimation of system dynamics
It is possible to obtain, the best (in the minimum mean-square error sense) linear model of a control system by solving a convolution-type integral for the impulse response of the system. This thesis presents a method for obtaining an approximate solution for the impulse response by solving a system of linear equations which is statistically equivalent to the convolution integral. An analogue computer which can solve the system of equations is described. The computer samples the sign of the
... the sign of the input signal at an adjustable rate and stores this information in a shift register. The output signals from the shift register are then used to compute functions statistically related to the correlation functions of the system signals. A set of linear equations relating these functions is solved using an arrangement similar to the Gauss-Seidel Iteration method. The computer utilizes a time-sharing technique and the step response of the system can be generated as a repetitive waveform. The overall operation of the computer is described in block diagram form. The individual circuits are described and the results of a computational test are given.