Disclosed in the present invention is an
optical fiber nonlinearity
equalization method used in a 64-
QAM coherent light
communications system,
processing received 64-
QAM data, and comprising: configuring a received
data set to be a first-
level data set, calculating a density parameter of each
data point, setting a threshold, and selecting data having a density parameter exceeding the designated threshold as a second-
level data set; demodulating the second-
level data set, dividing into 64 clusters, and acquiring 64 centroids; according to the acquired centroids, categorizing data in the second-level
data set into corresponding clusters according to a nearest
Euclidean distance, and updating using centroids of 64 acquired new clusters; allocating data in the first-level
data set into corresponding clusters according to a nearest
Euclidean distance, acquiring 64 clusters after
categorization, and calculating an optimum
centroid of an acquired cluster. The present invention can rapidly and precisely select a globally optimal K-means cluster
centroid without requiring iteration, greatly decreasing the effects of Kerr nonlinearity in
optical fiber, and enabling an obtained
bit error rate performance to be half an
order of magnitude higher than previous un-processed performance.