Unbiased bilinear equation error system identification

In a variety of nonlinear system identification applications, a bilinear model form is suitable to model the dynamics of the system. In such situations, parameter estimates are biased when an equation error framework under the least squares (LS) performance criterion is used in the presence of output measurement noise. In this paper an iterative approach to bilinear system adaptation is considered that yields unbiased parameter estimates in certain signal environments. The unbiasedness is achieved by using a mixed LS-total least squares (TLS) criterion and applying an algorithm recently developed by the authors. The performance is illustrated via simulation.