Experimental reservoir computing using VCSEL polarization dynamics

Jeremy Vatin (1, 2) , Damien Rontani (1, 2) , Marc Sciamanna (1, 2)

1 LMOPS - Laboratoire Matériaux Optiques, Photonique et Systèmes
2 Chaire Photonique

We realize an experimental setup of a time-delay reservoir using a VCSEL with optical feedback and optical injection. The VCSEL is operated in the injection-locking regime. This allows us to solve different information processing tasks, such as chaotic time-series prediction with a NMSE of 1.6 × 10 −2 and nonlinear channel equalization with a SER of 1.5 × 10 −2 , improving state-of-the-art performance. We also demonstrate experimentally, through a careful statistical analysis, the impact of the VCSEL polarization dynamics on the performance of our architecture. More specifically, we confirm recent theoretical prediction stating that a polarization rotated feedback allows for the enhancement of the calculation performance compared to an isotropic feedback. The need for efficient data processing has led to the development of novel computing methods to solve tasks that conventional computers would struggle with [1, 2] : For example, pattern recognition, automatic signal classification, or time-series prediction. Among these methods, the reservoir computing approach, developed within the framework of Machine Learning, allows using photonic systems to form an artificial neural network with a simplified training procedure : only the readout layer is trained with a linear regression [3]. Multiple experimental implementations of reservoir computers have already been proposed and demonstrated good computing ability [4-6]. Because of the technical challenge consisting of interconnecting a large number of photonic units together, another architecture was proposed : the time-delay reservoir [7]. This approach allows using a single photonic node in combination with a large number of virtual nodes spread over different locations in the time-delay line. As a result, the number of nodes in the architecture can be simply increased by changing the length of a delay line. Several implementations have been proposed : the optoelectronic [8, 9] or all-optical configurations [10-13]. The efficiency of these systems strongly depends on the tuning of the different physical parameters, and on the complexity of the dynamics of the nonlinear node. That is why adding other degrees of freedom can help improving their performance. For example, Hicke et al. in [14] added the control of the polarization of an edge-emitting laser along a feedback loop, and concluded theoretically that using the specific configuration of polarization rotated feedback improves the performance of their system in some ranges of operational parameters. We extend this work in several ways, replacing this conventional edge-emitting laser diode by a vertical-cavity laser diode (VCSEL). It eases the possibility to tune the polarization properties of the physical node. This VCSEL-based reservoir computer seems particularly adapted to perform calculation. It exhibits interesting polarization dynamics that allows generating much coupled information in a same period of time and thus improves the computing ability and the memory capacity of the reservoir, as we recently reported theoretically [15]. Moreover, it is well suited to telecommunication tasks as VCSELs are today wide spread in local and metropolitan networks. It also requires less energy as the threshold current of VCSELs is lower than the

Domaines: Optique [physics.optics]; Machine Learning [stat.ML]

Jeremy Vatin, Damien Rontani, Marc Sciamanna. Experimental reservoir computing using VCSEL polarization dynamics. Optics Express, 2019, 27 (13), pp.18579. ⟨10.1364/OE.27.018579⟩. ⟨hal-02169548⟩