Stokes Space Modulation Format Identification for Optical Signals Using Probabilistic Neural Network

Abstract

A Stokes space modulation format identification (MFI) method using probabilistic neural network (PNN) is proposed for coherent optical receivers. According to amplitude histograms obtained by the distribution of Stokes vectors on the s 1 axis, the incoming signals are first classified into PDM-mPSK, PDM-16QAM, and PDM-64 QAM signals based on PNN. To further identify PDM-mPSK signals, the constellation feature of Stokes vectors on s 2s 3 plane is extracted by image processing techniques and then processed by PNN. The high identification accuracy is demonstrated via numerical simulations with 28-Gbaud PDM-QPSK, PDM-8PSK, PDM-16QAM, and PDM-64QAM signals over a wide optical signal-tonoise ratio range. Owing to the characteristic of PNN, the proposed MFI method has simple training process, small training data size and a small number of required symbols. Proofof-concept experiments have been implemented to verify the effectiveness of the proposed MFI method among 28-Gbaud PDM-QPSK, PDM-8PSK, and PDM-16QAM signals.

Machine Learning Techniques for Optical Performance Monitoring and Modulation Format Identification: A Survey

“…There are many types of ANN, which have the same concept but differ in the architecture. For example, Probabilistic Neural Network (PNN) is a kind of feed-forward artificial neural network, which can approach a Bayes-optimal solution [49]. This solution chooses the class that has the maximum a posteriori probability of occurrence.…”

“…• MFI-based Stokes space representation with supervised ML Instead of exploiting unsupervised ML in conjunction with Stokes space, the authors in [49] and [179] proposed using supervised ML algorithms. In [49], Stokes space in conjunction with two consecutive PNN algorithms was proposed. In this technique, the AH of projected parameters on (s 1 , s 3 ) plane are generated first.…”

“…In contrast, in coherent detection acquisition systems, MFI becomes more reliable because DSP algorithms for CD and some other impairments compensation are often employed. Features that have been considered from such acquisition systems include AH [39], [49], [95], [181], [196], [200], [203], CDF [197], and power distribution [130]. These features are insensitive to PN and frequency offset.…”

“…In dual polarization modulations, the signals are prone to polarization rotation as well as PN and frequency offset. The received signal that is acquired coherently is often mapped into Stokes space constellation to produce features that are completely independent of these impairments [49], [166]- [172], [175], [177]- [179], [198]. However, Stokes space constellation features are sensitive to ASE noise, because the mapping distorts the ASE noise probability density function.…”

Performance Investigation of Modulation Format Identification in Super-Channel Optical Networks

“…Instead of using clustering algorithms in Stokes space, the authors in [27] exploited the constellation diagrams of the different modulations and used them as input images to CNN. Meanwhile, the authors in [28] used both AH and constellation images with two consecutive probabilistic neural network (PNN) algorithms to identify different modulation formats. In [29], the authors proposed an MFI technique based on deep neural networks (DNNs) in conjunction with features extracted from received signals' density distributions in Stokes axes.…”