GENERIC CIRCUIT RESPONSE INTERPOLATION FOR ROBUST DESIGN

In electronic circuit simulation in the frequency domain the response is a function of frequency, g(omega). Typically, to estimate g(omega) the circuit is simulated at a large number of frequency points omega(0), and to optimise the circuit a large number of simulations are made. A method for reducing the number of frequency points needed to estimate circuit response is presented in this paper. From an initial base simulation of the circuit an estimate of g(omega) is developed using a parametric statistical model from a fixed subset of omega(0), namely omega(1), carefully selected using the proven statistical method of cross-validation combined with simple thinning out of omega(0). The estimated model is used in conjunction with results from future simulations made at the reduced frequency set omega(1) to estimate g(omega) when the circuit is run under different input conditions, providing an efficient method of circuit optimisation. The method is illustrated with an example which explores the effect of reducing frequency points for simulation on model accuracy. This shows the number of frequency points can be significantly reduced while still estimating the response curves accurately.