Modulation Format Identification Based on an Improved RBF Neural Network Trained With Asynchronous Amplitude Histogram

Abstract

This paper proposed a method of modulation format identification using Radial Basis Function Artificial Neural Network (RBF-ANN) trained with Asynchronous Amplitude Histograms (AAHs). Compare with the traditional RBF-ANN, the proposed method is improved by applying Expectation Maximization (EM), which takes advantage of the statistical feature of AAHs, to select center vector for radial basis function. Assuming distribution of each bin in AAH as Gaussian mixture model (GMM), the mean values of the model can be exploited as the center vector which obtained using EM. This approach ensures that the center vector is unbiased and optimal. The center vector is implemented to RBF-ANN to identify different modulation formats. Numerical simulation results demonstrated that identification accuracy was about 99% for three commonly-used modulation formats within the OSNR between 40 ~ 10dB. And the CD tolerance was 1000ps/nm. In comparison, former center vector selection approaches include K-means and random selection were applied. The result showed that the EM method improved the identification accuracy by 2% to 4% when OSNR = 10dB and CD = 100ps/nm. Owing to its excellent performance, this method can be employed in the next generation optical transport network for auto-adaption modulation format identification.