Herein, we propose and experimentally validate a fiber FP sensor system based on a silicon-on-insulator (SOI) chip. The demodulation principle is on the basis of mapping environmentally induced spectral shifts into speckle patterns through multimode interferences. In this paper, we demonstrate for the first time a purely optical, chip-scale AI solution for high-mode isolation, speed-of-light demultiplexing of MMF modes using a three-dimensional diffractive neural network (DNN). To do so, we propose to use the MUSIC algorithm. It is shown that the use of traditional methods for estimating the number of signal components leads to poor operation of the MUSIC. This paper addresses the issue of low demod-ulation accuracy in interferometric signals caused by sig-nificant errors in direct peak finding and positioning dur-ing multi-peak demodulation of fiber-optic MEMS Fabry Perot Sensors.
