New Efficient Equalization Algorithms In Low-cost And High-speed NG-PON System

Passive optical network(PON),as a fiber-to-the-home terminal access technology,is one of the best solutions to solve the problem of optical access network.However,due to its low access bandwidth,the current PON system is hard to meet the long-term development requirements of services.Thus,next-generation PON(NG-PON)system urgently needs to be upgraded in terms of functions and performance.Recently,high-rate NG-PON systems based on low-cost optoelectronic devices are the focus of academic and industrial circles at home and abroad.In these systems,intersymbol interference caused by the bandwidth limitation of the optoelectronic device and the cumulative fiber dispersion will seriously affect the system performance,and then reduce the access capability of its PON system.Thus,how to improve the performance of low-costand high-rate NGPON systems is paramoun in current research.Among the existing schemes,the equalization scheme based on digital signal processing is one of the most effective schemes to compensate for the influence of inter-symbol interference.To deal with these issues,the new equalizer schemes based on machine learning algorithm in low-cost and high-rate NG-PON system are the key points focused in this paper.Firstly,wethe models of the researched NG-PON system are conducted and analyzed.Based on these characteristics,several simple machine learning algorithms improved by feature construction and feature engineering are proposed.Then,the feasibility of these new algorithms is analyzed by the theoretical derivation.Finally,25Gb/s NRZ and 50Gb/s PAM-4 transmission experiments based on 10 G low-cost transceiver are constructed to further verify the performance of our proposed algorithms.This paper mainly includes the following parts,(1)In this chaper,the research significance,background,fundamental principles,and key technologies of channel equalization technologies in the high-speed NG-PON are introduced,and then demonstrated the detailed research status at home and abroad of equalization scheme in these systems.(2)The channel characteristics of low-cost high-rate NG-PON system are modeled to design the feature construction scheme suitable for the channel.Then,a novel support vector machine(SVM)scheme based on feature construction for low-cost and high-rate NG-PON system is proposed.By theoretical derivation,the linear separability of the scheme in low-cost and high-rate NG-PON systems is further verified,which reduces the computational and implementation complexity while achieving significant performance improvement.The feasibility of the algorithm is verified by the 25-Gb/s NRZ and 50-Gb/s PAM-4 per wavelength experiments.(3)The equalization algorithm based on single SVM for the high-order modulation NG-PON system is studied.To solve the complexity problem caused by the combination of multiple SVMs to achieve high-order modulation signal equalization task,a novel scheme based on the feature construction scheme for high-order modulation NG-PON system is proposed,which requires only a single SVM model to achieve multi-class high-order modulation equalization tasks.As for the proposed scheme,the computation and implementation complexity of the SVM in high-order modulation systems is greatly reduced with almost no loss of performance.The feasibility of the algorithm is verified by the 50-Gb/s PAM-4 per wavelength signal transmission experiment.(4)The K-nearest neighbor(KNN)algorithm modified by feature engineering for high-rate NGPON system is studied.The proposed feature engineering scheme combines the characteristics of band-limited NG-PON system and KNN algorithm,which is composed of feature construction and feature weighting.The scheme converts the original received data into the data more suitable for system and algorithm characteristics,thus significantly improves the KNN algorithm in the NG-PON system.By the 25-Gb/s/? NRZ and 50-Gb/s/? PAM-4 signal transmission experiment,the feasibility of the algorithm is further verified.