Photonic Reservoir Computing Using Optoelectronic Oscillators With Direct Laser Modulation

Reservoir computing (RC) is a simple yet efficient machine learning algorithm that has reached state-of-the-art performance on several benchmarks. Incorporating ultra-fast photonic devices, implementations of RC based on optoelectronic oscillator (OEO) with a fiber delay-line has shown the ability to process both the optical and RF signals at an unprecedented high speed. However, previously reported OEOs used for reservoir computing require an external modulator, typically a bulky LiNb$\rm O_{3}$ Mach-Zehnder Modulator (MZM), not only making it hard for such systems to be integrated onto a silicon chip but also increasing the overall cost. In this article, we present a simpler implementation of OEO-based RC without the need to use external modulators. Contrary to previous implementations of OEO-based RCs that inject the signal into the delay line via external modulators to excite VNs, we show that the input signal can be injected into the delay line by internally modulating the laser. The intensity modulation of the laser in this architecture introduces a piecewise-linear power intensity transfer function, leading to a rectified linear unit (ReLU) non-linearity inside the reservoir layer. We first derive the equations that govern the dynamics of such an OEO-based RC. Then, we numerically simulate its dynamics and successfully train and test this direct laser modulated OEO-based RC on several datasets, namely the Nonlinear Auto Regressive Moving Average (NARMA10) dataset, Nonlinear-Channel Equalization (NCE) dataset, Radar Signal Forecasting dataset, and Santa Fe Laser Series Prediction dataset, reaching a performance that is equivalent to or better than the traditional OEO-based RCs with external modulators.