Global Optimization of Design Parameters of Analog Pre-Distortion Linearizers for RF Power Amplifiers

In this work, we compare a few widely used nature-inspired global optimization algorithms in designing an analog predistortion (APD) linearizer to correct a nonlinear GaN power amplifier (PA). A typical dual-branch APD is utilized for this purpose, with the nonlinear branch composed of a driver and nonlinear amplifier, and the linear branch realized by a phase shifter. The relevant design parameters, i.e. driver gain, non-linear amplifier bias resistor, gate voltage, and phase offset, are optimized. Four stochastically based, global optimization algorithms are compared: Particle Swarm (PS), Differential Evolution (DE), advanced Genetic Algorithm (GE), and Random (R) optimization. After performing 10 different and independent optimization trials, each randomly initialized, DE outperforms the other algorithms and converges to the global minimum after about 919 iterations. Using the optimized parameters, the APD+PA characteristics are then used to calculate the APD+PA output spectrum. A reduction of more than 15 dB in spectral regrowth is observed using the optimized APD.