Simplified SVM Equalization Algorithm Based on Single Hyperplane Training Enabled 50Gb/s PAM-4/8 With 10-G Optics in NG-PON System

• Published in: IEEE photonics journal Vol. 13; no. 4; pp. 1 - 7
• Main Authors: Xi, Yu, Bi, Meihua, Xin, Miao, Wang, Chenyu, Yu, Jiasheng, Hu, Weisheng
• Format: Journal Article
• Language: English
• Published: Piscataway IEEE 01-08-2021; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Algorithms; Bandwidth; channel equalization; Classification; Complexity; Distortion; Equalization; Equalizers; Hyperplanes; Intensity modulation direct detection; Optical communication; Pulse amplitude modulation; support vector machine; Support vector machines; Task analysis; Training; Training data
• ISSN: 1943-0655; 1943-0647
• DOI: 10.1109/JPHOT.2021.3103195

In this paper, we firstly present a novel equalizer only with a single support vector machine (SVM) training hyperplane to compensate the distortion of multi-class PAM-M signals in low-cost limited-bandwidth next generation passive optical network (NG-PON) system. For this scheme, only single SVM-based hyperplane is trained, and then it is extracted and used to realize the multi-classification task of PAM-M by adjusting the corresponding offset, which would greatly reduce the complexity of equalization algorithm and enhance its application in future high-speed PON system. Moreover, the 50-Gb/s PAM-4/8 per wavelength transmission experiment with 10-G optics in 20km standard single mode fiber (SSMF) PON system are set up and utilized to manifest the feasibility and effectiveness of our scheme. The results show that, compared with the common SVM scheme with multi-classification hyperplane training, our method can reduce M-1 times computational complexity caused by hyperplane training for multiple-level pulse amplitude modulation (PAM-M) signals. Except this, the propose scheme can also maintain almost the same bit error ratio (BER) performance as the common SVM classifier with multi-hyperplane case.