Deep Learning Assisted Wide-range Microwave Photonic Sensing

Abstract

We present deep learning (DL) assisted widerange microwave photonic (MWP) sensing using optical microresonators. The measurement range is extended by using a tunable laser to switch the optical carrier wavelength among a group of points, dividing the wide optical transmission spectrum of interest into segments with bandwidths suitable for the radio frequency operational bandwidth of the sensor. By adopting DL techniques to process the combined interrogation output of each segment, the laser wavelength instability effect can be mitigated, enabling accurate wide-range MWP sensing. As a proof-of-concept, a MWP sensor operating at two carrier wavelengths and adopting principal component analysis-assisted deep neural networks is demonstrated experimentally for glucose concentration measurement. The system operation range is doubled to 67.4 GHz. The estimation root-mean-square error in the presence of both thermal interference and laser wavelength instability is achieved to be 3.4-fold better than that using linear fitting.