Deep belief network-hidden Markov model based nonlinear equalizer for VCSEL based optical interconnect

Abstract

The data center has developed rapidly over the past few years, leading to the demand for high speed data transmission. Vertical cavity surface emitting lasers (VCSELs) based optical interconnect is evolving to 100 Gb/s and beyond, which makes nonlinear distortions difficult to be compensated or equalized by conventional equalizers. Moreover, the challenge becomes very complicated for conventional equalizers because of the presence of inter-symbol interference (ISI) together with the nonlinear distortions. So many neural network based DSP algorithms such as artificial neural network (ANN) have been proposed to mitigate the distortions. However, ANN has the limitations that sample’s relevant information is not considered, leading to degradation in ANN’s performance and high computational complexity. In this paper, in order to maintain an excellent capability of mitigating nonlinear distortions like other neural network based equalizers while considering the sample’s relevant information to reduce the computational complexity, we propose a deep belief network-hidden Markov model (DBN-HMM) based nonlinear equalizer which is tested in a PAM-4 modulated VCSEL and multimode fiber (MMF) optical interconnect link experimentally. The BER performance can be greatly improved compared with conventional DSP algorithms. In addition, the computational complexity of DBN-HMM based equalizer can be about 41% lower than that of ANN based method with a similar BER performance.