Frequency-domain subspace identification using FRF data from arbitrary signals

Recently an extension of the frequency-domain subspace identification methods was developed that can cope with arbitrary excitation signals and transient effects. In this paper the implications of this extended model are investigated for the H1H1 estimator. It is shown that an extended model makes it possible to start from frequency response functions (FRFs) without suffering from leakage, by modelling the initial conditions of each data block used in the averaging procedure of the H1H1 estimator. In this way the measurements can be averaged in order to reduce both the data set and the noise levels without introducing bias errors caused by leakage phenomena. It is shown by both simulations and real-life measurements that the presented approach improves the modal parameter estimates and in particular the damping estimates.