Clustering algorithms for Stokes space modulation format recognition

Area of Science:

• Optical Communications
• Signal Processing
• Machine Learning

Background:

• Stokes space modulation format recognition (Stokes MFR) is essential for blind digital coherent receivers.
• Clustering algorithms are critical for Stokes MFR performance, particularly at low signal-to-noise ratios (SNRs).

Purpose of the Study:

• To extensively study and compare the performance of six distinct clustering algorithms for Stokes MFR.
• To evaluate these algorithms in discriminating between various dual-polarization modulation formats.

Main Methods:

• Investigated k-means, expectation maximization, DBSCAN, OPTICS, spectral clustering, and maximum likelihood clustering.
• Tested algorithms on dual-polarization BPSK, QPSK, 8-PSK, 8-QAM, and 16-QAM signals.
• Assessed performance metrics including minimum SNR, detection accuracy, and algorithm complexity.

Main Results:

• Determined the minimum required SNR for each clustering algorithm and modulation format.
• Quantified the detection accuracy achieved by each algorithm.
• Analyzed the computational complexity of the evaluated clustering methods.

Conclusions:

• Identified the most effective clustering algorithms for Stokes MFR across different modulation formats and SNR levels.
• Provided a comprehensive performance benchmark for selecting clustering algorithms in optical coherent receivers.