Cascaded Deep Q-Network for Optimization of Substrate Integrated Waveguide Couplers

In this brief, a method to automatically optimize substrate integrated waveguide (SIW) couplers with different power-dividing ratios is proposed. In the proposed method, a cascade model is presented for the first time. In this cascade model, four-port SIW network is classified into two typical elements: coupling and phase difference. By cascading these typical elements in a certain order, SIW coupler with different power-dividing ratios can be achieved. The order of the basic elements can be automatically decided by using deep Q-network (DQN). A library is built to store the optimization results. From the library, SIW couplers with different power-dividing ratios can be obtained in seconds. To validate the proposed method, multiple four-port directional couplers with different power-dividing ratios are designed. Three directional couplers are fabricated and experimentally verified. Furthermore, this method is also promising for the design and optimization of other circuits.