Variable step-size MLMS algorithm for digital predistortion in wideband OFDM systems

As one of the most promising linearization techniques, adaptive Digital Predistortion (PD) has been widely utilized in modern wireless communication systems for improving the efficiency of Power Amplifier (PA). In view of the non-stationary signal environment for the wideband PAs, an efficient indirect learning adaptive PD is proposed in the paper based on the Memory Polynomial Model (MPM). The coefficients of the proposed PD can be effectively identified by the Modified Least Mean Square (MLMS) learning algorithm. In addition, more stable convergence and lower steady-state error can be achieved simultaneously for the PAs with deep memory effects by adopting the variable step-size parameter. Theoretical analysis results regarding the learning stability, convergence behavior, and selection criteria of initial settings are derived. Simulations demonstrate that MLMS outperforms traditional LMS, Normalized LMS (NLMS), and Generalized Normalized Gradient Descent (GNGD) algorithms in terms of the Normalized Mean Square Error (NMSE) and out-of-band Power Spectral Density (PSD) under the noisy feedback condition for the wideband PAs¹.