An Enhanced Binary Particle Swarm Optimization for Pruning Digital Predistortion Models

Abstract

In this letter, we propose an enhanced binary particle swarm optimization (PSO) algorithm with symmetrical uncertainty (EBPSO-SU) to reduce the complexity of the digital predistortion (DPD) model. In millimeter-wave (mm-wave) communication systems, the power consumption issue is notable due to the considerable number of redundant terms in the DPD models. To prune these terms, the correlation between the label (output signal) and features (basic function terms) is first leveraged for swarm initialization. Subsequently, the EBPSO algorithm, incorporating a modified velocity-to-position mapping formula, is employed to identify key terms of the model. Measurement results from a 28 GHz power amplifier operating with a 200 MHz input signal illustrate that the proposed pruning algorithm can reduce the complexity of the full generalized memory polynomial (GMP) model by 90% while ensuring equivalent performance.