Automatic Optimization with Deep Q-Learning for Digital Predistortion Model of RF Power Amplifiers

In this paper, we propose a method for searching the optimal structure of generalized memory polynomial (GMP), called automatic GMP (Auto GMP), that uses deep reinforcement learning (DRL) to learn the relationship between structure parameters of GMP and the set goal, in order to make a better trade-off between performance and complexity. Based on a comprehensive consideration of implementation difficulty and search efficiency, deep Q-learning (DQN) is chosen to construct the proposed model. Experiments show that it can find a set of structure parameters of GMP with the lowest complexity and optimal performance compared to the greedy algorithm and simulated annealing (SA) algorithm.