Open AccessRobust weighted K-means clustering algorithm for a probabilistic-shaped 64QAM coherent optical communication system
Author(s) -
Xishuo Wang,
Qi Zhang,
Xiangjun Xin,
Ran Gao,
Qinghua Tian,
Feng Tian,
Xiaolong Pan,
Yongjun Wang,
Leijing Yang
Publication year - 2019
Publication title -
optics express
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.394
H-Index - 271
ISSN - 1094-4087
DOI - 10.1364/oe.27.037601
Subject(s) - quadrature amplitude modulation , cluster analysis , algorithm , qam , computer science , robustness (evolution) , probabilistic logic , weighting , bit error rate , artificial intelligence , physics , biochemistry , decoding methods , chemistry , acoustics , gene
A novel weighted K-means scheme for a probabilistic-shaped (PS) 64 quadrature amplitude modulation (QAM) signal is proposed in order to locate the decision points more accurately and enhance the robustness of clustering algorithm. By using a weighting factor following the reciprocal of Maxwell-Boltzmann distribution, the proposed algorithm can combine the advantages of PS and K-means robustly while reducing the overall computational complexity of the clustering process. Experimental verification of the proposed clustering technique was demonstrated in a 120-Gb/s probabilistic-shaped 64QAM coherent optical communication system. The results show that the proposed algorithm has outperformed K-means with respect to bit error rate (BER), clustering robustness and iteration times in both back-to-back and 375km transmission scenarios. For the back-to-back situation, the proposed algorithm is capable of achieving about 0.6dB and 1.8dB OSNR gain over K-means clustered signals and unclustered signals. For the case of transmission, the proposed clustering procedure can robustly locate the optimal decision points with launched signal power ranging from -5dBm to 5dBm, while the working range for K-means procedure is only -4dBm to 2dBm. In addition, the proposed weighted algorithm takes less iteration times than K-means to converge, especially when the signal impairments caused by fiber Kerr nonlinearity is severe.