The optimization process of RF circuits usually needs to address the complexity of high-dimensional nonlinear problems. To cope with this challenge, this paper applies reinforcement learning techniques to it and proposes a DQNN-based collaborative optimization algorithm for RF circuit parameters. The algorithm establishes a mathematical model for the parameter optimization problem of RF circuits, searches for the optimal circuit parameter design in the circuit parameter design space using reinforcement learning algorithms, and designs a DQNN neural network structure to realize the parameter optimization search. Take the charge pump phase-locked loop as experimental example, the method that this paper puts forward shows very good performance on both convergence speed and relative loss. After the optimization work was completed, the phase noise of the voltage-controlled oscillator and the charge pump phase-locked loop has been decreased by 2.14dBc/Hz@1MHz and 4.05dBC/Hz@1MHz, respectively. The tuning range of the voltage-controlled oscillator and the working bandwidth of the charge pump phase-locked loop have respectively seen an increase of 0.122 GHz and 0.034 GHz. Furthermore, the peak vibration frequency of the voltage-controlled oscillator has been increased by 0.095GHz@4V. The results show that the method possesses efficient and stable parameter optimization capability, which verifies its applicability in RF circuit parameter co-optimization.
