Low Complexity Nonlinear Digital Equalization Technology For High Speed Multi-mode Optical Interconnection

In the recent years,global data-centers(DCs)IP(Internet Protocol)traffic is experiencing an explosive and continuous growth with the massive expansion of Internet of things(Io T),Augmented Reality/Virtual Reality(AR/VR),Ultra High Definition(UHD)display and cloud applications.High speed optical interconnects are highly demanded to provide extensive connectivity between the DCs,switches,servers and even chips.Short reach optical interconnects have evolved to multilevel modulation instead of sustaining traditional NRZ(Non Return to Zero)modulation.In order to meet the needs of information transmission in the DCs,VCSEL(Vertical-cavity surface-emitting laser)-MMF(Multi-mode fiber)intensity modulation and direct detection system based on PAM-4(4-level pulse amplitude modulation)modulation becomes a promising solution.However,with the increase of data rate,the inherent nonlinear effects in the system have become the dominated factor that influences the transmission performance.The traditional linear equalization scheme cannot provide enough performance gain for such transmission case.Therefore,advanced nonlinear equalizations are highly required to introduce powerful nonlinear equalization capability.Besides,driven by the cost and power consumption requirements,the computational complexity of the nonlinear equalizer must be compressed because the huge computational complexity of nonlinear equalization seriously hampers its actual deployment.To sum up,the proposed schemes and related conclusions in this paper are as follows:(1)Volterra polynomial based nonlinear equalization(VE)scheme is proposed and the influence of orders and memory length distribution on VE performance is analyzed.We propose the threshold based pruned retraining Volterra equalization(TRVE)to reduce the computation complexity while maintaining the transmission performance.Multimode 100 Gbps/lane transmission over 100 m OM3 MMF has been experimentally demonstrated with up to 70.7% computation complexity reduction while keeping the BER under the 7% HD FEC limit.In addition,the TRVE with L1-regularization(L1-TRVE)is also proposed and experimentally verified to optimize the threshold selection.(2)Neural network based nonlinear equalization(NNE)scheme is proposed and the effects of multiple NNE parameters on equalization performance are analyzed.The possible overestimation problem of NNE is discussed,and the solution of receiving signal randomization is proposed.in order to solve the problems of the one-shot pruning algorithm,we propose the iterative pruning algorithm and study the impact of threshold and pruning span on the performance of proposed algorithms.112-Gbps VCSELenabled PAM-4 and PAM-8 transmissions are experimentally demonstrated.The results show up to 71% connection compression by use of the iterative pruning algorithms and maximum 28.4% improvement compared with the one-shot pruning algorithm.Finally,the equalization efficiency of VE and NNE and the stability of the pruning network are compared.The experimental results shows that NNE has better equalization performance than VE with the same compulational complexity,while pruned-VE can better adapt to channel changes than pruned-NNE.