Use of Regularization in Indirect Learning Identification of Predistorter

The scope of this work is regularization approaches aimed to improve accuracy of identification of predistorter for power amplifier (PA) linearization. Use of regularization is motivated by intrinsic losses of indirect learning (IL) architecture and least-squares (LS) criterion of identification against a backdrop of an always present fidelity error of the polynomial expansion model of predistorter. Presented unified view covers regularization effects produced by numerical solver of identification problem, truncation of Volterra series model of predistorter and reduction of sampling frequency ($F_{\mathrm{s}}$) of test signals for identification. The impacts of different approaches on linearization performance are studied separately and then optimized in joint using test bench model of memory nonlinearity. It is shown that with optimal regularization setup and use of appropriate preconditioning of test signal performance of IL/LS approach is very close to reference predistorter derived analytically.