The Effect of Power Amplifier Characteristic Variations on DPD-Based Linearization Performance

In this study, we investigated the effect of power amplifier (PA) characteristic variations on the performance of digital predistorter (DPD)-based PA linearization. The proposed DPD adopts a deep neural network (DNN) architecture that integrates a one-dimensional convolution layer, thereby enabling block-based processing with efficient memory utilization. Variations in the PA characteristics may arise from long-term changes in the operating environment or fluctuations in the input signal level due to measurement errors. To examine these effects, we utilized PA output data collected over a one-month period along with additional output data measured under varying input signal levels to train the DNN model. The linearization performance of the PA was evaluated in terms of the error vector magnitude (EVM) and adjacent channel power ratio (ACPR), and the performance degradation was analyzed under various conditions considering temporal variations and input signal level changes.