Automatic discovery of optimal microwave photonic architectures in a complete topological space

Microwave photonics (MWPs) has become a ground-breaking driving force for the evolution of radar, wireless communication, and computing systems, as it merges the technical advantages of microwave and photonics. However, the conventional design strategies of MWP systems are based on experts' empirical knowledge and physical intuitions, therefore, it is challenging to explore the theoretically optimal system architectures among all possibilities. Here, we present an MWP system design strategy that can discover the optimal architecture by automatic iteration and optimization. In the proposed strategy, MWP system architectures are abstracted as graphs represented by basic signal processing units and topological connections of these units. Then, we create the topology generation and parameter search algorithms to find the theoretically optimal architectures in a complete topological space. Results demonstrate that the proposed strategy can not only find the known-good architectures, but also discover non-intuitive optimal architectures with better performance and lower system costs. This strategy breaks through the empirical limitations of conventional design strategies and provides a promising framework for finding high-performance and low-cost MWP systems for future radar, wireless communication, and computing systems.