Wideband digital predistortion via synergistic bias extension and nonlinear feature clustering

This study proposes a new biased extended memory polynomial model and a new clustering enhanced recursive least squares algorithm, which jointly innovate the digital pre-distortion algorithm. Compared with the conventional digital pre-distortion algorithm, it has higher accuracy without increasing complexity. The proposed biased extended memory polynomial model introduces a constant term to explain the DC offset and system error, and shows special effectiveness under low amplitude excitation. At the same time, the clustering enhanced recursive least squares algorithm uses K-means clustering to partition the nonlinear feature space, thereby realizing coefficient propagation between clusters, realizing local parameter updates, and significantly accelerating the convergence speed, improving convergence stability and accuracy.