|
We exploit the transient dynamics of a nonlinear photonic system to perform useful computation. This is achieved within the framework of reservoir computing. State of the art implementations in photonic hardware are evolving towards simple architectures. With nonlinearities present in either the reservoirs input or output layer, even a linear photonic cavity makes for a potent reservoir. However, when targeting all-optical reservoir computers (coming from opto-electronic systems), commonly used non-linearities in opto-electronic conversion equipment, such as modulators and photodiodes, can no longer be relied on. Therefore, optical nonlinearities must be considered. In this work, we numerically and experimentally investigate a delay-based reservoir implemented in standard single mode optical fibers. Our setup is coherently driven and exploits the optical Kerr nonlinearity, which is present throughout the reservoir’s extent (i.e. the fiber ring cavity), to operate as a state-of-the-art photonic reservoir. A set of systems was considered, with different combinations of linear and nonlinear input and output schemes. And we have been able to quantify the effects of different nonlinearities in the system on its reservoir computing performance. Experimental data shows the positive effects of the distributed Kerr nonlinearity on both the linear memory capacity and nonlinear computational capacity of our reservoir computing system. We find a broad range of power levels where this distributed nonlinear effect improves the reservoirs performance. Moreover, we find that the exploitation of this optical nonlinearity in the reservoirs bulk allows for state-of-the-art reservoir computing performance without relying on opto-electronic nonlinearities elsewhere in the system.
|
