Information processing using a photonic reservoir computer based on a coherently driven passive cavity with an analog readout layer

Reservoir computing is a technique based on recurrent neural networks to process time-dependent signals [1]. Recently, we have successfully demonstrated, both experimentally and in simulation, a linear passive fiber cavity used as a hardware based reservoir computer [2]. On a wide range of benchmark tasks, this setup has better performances than the best digital algorithms and all previous experimental reservoir computers, see e.g. [3,4]. In the present work, we numerically investigate the use of an analog readout layer on this system, inspired by the one shown in [5]. The aim is to directly generate the output signal without any off-line post-processing. Our simulations include all the experimental limitations of the setup, e.g., device bandwidths, sampling rates, noise, resolutions, etc.