Abstract

We study the impact of amplitude and phase differences between the parallel power amplifier (PA) branches in a phased array and their impact on the performance of the digital predistortion (DPD). The DPD coefficients are estimated from the array response in the far-field. The DPD coefficients need to be updated for changes in the nonlinear behavior of the PAs due to amplitude and phase variations. We present a training mechanism which makes the DPD robust to branch specific amplitude and phase weights and can tolerate these variations without the need of adapting to individual changes in the nonlinear behavior of the PA branches. The DPD is trained for a set of random amplitude and phase weights following normal distribution and the resultant mean DPD coefficients are used for predistortion. The simulation results show that the mean DPD can achieve the same average linearity performance as the continuously trained reference DPD for 32 elements uniform array.