A blind approach to the Hammerstein-Wiener model identification (Q1614327)

The identification task of a sampled Hammerstein-Wiener model from input-output measurements is considered. A blind approach exploiting over-sampled output data is applied. The main idea of the approach is to recover the unavailable interaction signals based only on the polynomial inverse of the output nonlinearity and the output measurements. No prior knowledge about the structure of the input nonlinearity is assumed, however the output nonlinearity is approximated by a polynomial model. The system input is not assumed to be a white noise process. Under these conditions the parametric identification problem of the polynomial output nonlinearity and the sampled linear system dynamics are solved separately by using the standard parametric LMS algorithm. Next, the input nonlinearity is reconstructed as a graph from appropriate input-output data points, with the output data computed from the estimated models and the observed outputs of the whole system. The effectiveness of the approach is illustrated by means of numerical simulations.