Dynamics of the parameter-perturbation process

It is shown how the mean-square error of a linear system varies with time following a small change in any parameter of the system. It has been assumed that the system is in the steady-state condition before the parameter perturbation is applied. The input signal to the system may be deterministic or a random process with known statistical properties. By considering the response to a step parameter change, a linear model is developed relating the change in the mean-square error (ensemble average) to the parameter variation. Further, the principle of superposition is shown to be applicable, so that the change in mean-square error for any arbitrary system input can be found when the behaviour is known for sinusoidal system input signals. Experimental and theoretical work is presented for particular systems.