A Self-Generating Coefficient List for Machine Learning in RF Power Amplifiers using Adaptive Predistortion

A learning module is proposed that improves the transient response of an adaptive controller used to predistort RF power amplifiers (PA´s). The adaptive controller modifies its predistortion coefficient setting continually as the operating condition changes. The learning module recognizes when the adaptive controller has converged, correlates the coefficient setting with the measured operating condition, and stores both in memory. The learning module also monitors the present operating condition and, in response to abrupt changes, restores past coefficients that were successful under similar conditions. Successful coefficient settings are stored in a list that is indexed using a multi-dimensional attribute vector derived from the measured operating condition. Unlike look-up-tables with array structures, the list generates elements automatically. The size of the list is dynamic, growing as more operating conditions are experienced and contracting as neighboring elements are recognized as redundant