Design of Software-Based Optimal Signals for System Identification

Two software-based optimal signal designs, namely, the model-based optimal signal excitation 2 (MOOSE2) signal from the MOOSE2 program and the optexcit signal from the Frequency Domain System Identification Toolbox, are compared, along with the flat spectrum signal as a benchmark. The work is motivated by the lack of existing comparison and the need to provide recommendations to aid industry practitioners and applied researchers in selecting the most suitable signal for system identification. The effectiveness of the signals in combination with the choice of model structure is evaluated based on the determinant of the covariance matrix of the estimated parameters and the mean and maximum errors in the frequency response. Based on the probabilities of each combination of signal and model giving the highest estimation accuracy for lowpass and bandpass systems of various orders, recommendations are given on their selection. The feasibility of the proposed recommendations is illustrated through an application example on a modular tray oven. Results from this work are significant in bridging the current gap between theory and practice.