Robust Estimation for Digital Predistortion with Nonideal Equalization

This paper proposes a closed loop estimator that adapts a digital predistortion (DPD) module. DPD is used to compensate for nonlinearities in a RF power amplifier (PA). The standard method for a closed loop estimator is a least square approach. It can produce a biased estimate if the observation path is not equalized properly. A prior solution computes the error and basis signals within the estimator in the frequency domain and notches the linear bandwidth so that the coefficient estimate is based on the out-of-band signals, which are due exclusively to the nonlinearity. Unfortunately, this method cannot be used if the out-of-band signals are bandwidth limited within the observation path. The proposed approach also computes the estimator signals in the frequency domain. The linear frequency response from the input to the output of the observation path is estimated using a long-term average. It is used to remove the linear signal spectrum from the output leaving only the nonlinearity portion of the spectrum as an error signal. It is shown that improved DPD performance is achieved even when the frequency response of the observation path is nonideal.