Digital predistorter identification based on constrained multi-objective optimization of WLAN standard performance metrics

We present a new approach to identify the parameters of a given digital predistortion (DPD) structure for power amplifier (PA) linearization. Traditional methods optimize a single objective, typically the time-domain mean squared error. We propose to use a multi-objective optimization algorithm to jointly optimize the in-band and out-of-band performance as quantified by the respective metrics defined in the particular communication standard. Our constrained approach allows for checking standard-compliance at the time of DPD identification. Furthermore, the DPD model is not required to be linear in the parameters. We exemplify our approach with a WLAN simulation using a PA model at low back-off. By jointly optimizing the error vector magnitude (EVM) and spectral mask margin, we achieve significantly better results than the widely-used indirect learning architecture for the same memory polynomial DPD structure.