Machine learning-assisted design and control for period-one microwave photonic sensing signal

Microwave photonic (MWP) sensing and measurement are envisioned to be a promising alternative to the conventional pure electronic or optical solutions. A semiconductor laser (SL) with external optical feedback (EOF) operating in a period-one (P1) dynamic state contributes a new implementation architecture for MWP systems. However, designing such a SL system to generate frequency-modulated MWP sensing signals through traditional Lang–Kobayashi (L–K) equations requires extensive computational effort to derive the system control parameters (SCP), making real-time adjustment of the SCP impossible in cases where it is needed. This paper proposes an effective design approach based on machine learning. A feedforward neural network (FNN), in conjunction with a gradient descent algorithm, is employed to fast and accurately ascertain the SCP, offering a solution readily applicable in the system design. Both simulation and experiment are conducted to validate the proposed approach.