Training of digital predistortion based on signal-to-distortion-ratio measurements

A novel training technique for digital predistortion (DPD) systems is presented. Previous techniques rely on the time-domain (TD) difference between the ideal transmitter input and the distorted amplifier output as an error metric. This work presents DPD training based solely on frequency-domain signal-to-distortion-ratio (SDR) information. The only a priori information required is the spectral occupancy of the desired signal. The approach is applied to a broadband SiGe BiCMOS power amplifier, including measured results.