Linear-quadratic system identification with completed frequency domain region of support

In general, aliasing exists in the output of discrete-time nonlinear systems due to the higher order harmonics and intermodulation among inputs. If the input is allowed to cover the full range from DC to half the sample rate, the aliased components in the output must be taken into account for accurate system identification. We have extended a known system identification method to account for these aliased terms by completing the region of support for the quadratic transfer function. We also present a method for preventing a form of time domain aliasing. Together, these approaches can lead to significantly improved system characterization with less data as compared to current techniques.<
>