Towards autonomous photonic reservoir computer based on frequency parallelism of neurons

Reservoir Computing is a bio-inspired computational paradigm particularly well adapted to time-dependent signal processing. The past years have seen the realisation of photonic reservoir computers with performance comparable to digital algorithms. Most of these works are based on delay dynamical systems in which the photonic neurons are treated sequentially. We have recently realised an experimental system based on the concept of frequency multiplexing, in which the neurons are materialised as the amplitude of light at different frequencies. In this system the neurons are processed in parallel, making it in principle much faster than the sequential systems. In most of the works up to now the output of the reservoir is implemented using slow digital offline post-processing. This is also the case of our recent experiment on frequency parallelism. Here we demonstrate, using numerical simulations, the possibility of an analogue electronic readout layer for this system. Our simulations take into account all the experimental limitations of the setup, e.g. sampling rates, device bandwidths, resolutions, and noise. The results obtained are comparable to those produced with the same architecture and a digital output layer. The proposed setup is thus an important step towards analog, low footprint, all-optical information processing. Moreover parallel processing will be necessary for high bandwidth applications.