Investigating a Filtered Feedback Weight in the Output Signal to Simulate a Self-learning Layer in an Optical Neural Network

Abstract

An edge-emitting semiconductor laser (SL) source with a wavelength of 1310 nm is used experimentally to follow a transmitted signal that is subject to selected modifications. SL maximum measured optical power was -11 dBm at a temperature of 24oC, which is configured as a chaotic influencer semiconductor laser (ISL) in a unit of an optical neural network. The weight of the sent signal, after division, is controlled by two uniform Fiber Bragg gratings (FBGs) with slightly different periods. In addition, the feedback strength is changed by two radio frequency attenuators. The passing signal via these two filters is detected and summed again before being directly modulated and re-injected to the ISL again via the bias current. The results confirm a variable chaotic FWHM and bandwidth as a function of applied parameters. Emission for selected strengths is controlled by radio frequency attenuators after the power division. For each opto-electronic feedback (OEFB) branch, a noticeable difference in Fourier space is noted in addition to time and phase spaces.
Results indicate a large variation in the calculated number of spikes (4268 to 71330) and also in the calculated FWHM (0.80991-10.5928) MHz This indicates the possibility of deriving the SL in a different number of quantum states by introducing filtered chaotic dynamics. Generalizations for the effect of optical weights, achieved in this study, open the door to increasing the processing speed in the processing units of computers and self-learning terms in robots.