Experimental study of support vector machine based nonlinear equalizer for VCSEL based optical interconnect

Vertical cavity surface emitting lasers (VCSELs) based optical interconnect is evolving to 100 Gb/s and beyond. Penalties due to relative intensity noise and mode partition noise become significant and cannot be compensated or equalized by conventional equalizer. As one of the machine learning based DSP algorithms, support vector machine based nonlinear equalizer (SVM-NLE) has been proposed for VCSEL based optical interconnect links recently and was proved having an excellent capability of mitigating the modulation nonlinearity. In this paper, we propose a feature enhancement method for SVM-NLE by means of sinc interpolation and the performance of the new SVM-NLE in the VCSEL based optical link transmission experiment is evaluated under different parameter settings and training strategies at 60 Gb/s transmission rate. At least 3.5 dB performance improvements can be derived by our proposal compared with the conventional method. In addition, error free 40 Gb/s transmission is realized with the aid of the new SVM-NLE. Its complexity is also theoretically analyzed from the perspective of AISC design, which is shown controllable and acceptable for an appropriate optical link bandwidth design.