Real-time channel conditional distribution tracking for intelligent decoding of optical IMDD signals

Area of Science:

• Optical Communications
• Machine Learning
• Signal Processing

Background:

• Intensity-Modulation and Direct-Detection (IMDD) systems are crucial for optical communications.
• Modulator bias point shifts can degrade signal quality and decoding performance.
• Adaptive decoding schemes are needed to maintain high communication reliability.

Purpose of the Study:

• To propose a real-time machine learning scheme for tracking the conditional distribution of optical IMDD signals.
• To enable adaptive decoding that is robust to modulator bias point shifts.
• To demonstrate the practical implementation and effectiveness of the proposed scheme.

Main Methods:

• A real-time machine learning scheme utilizing linear optical sampling and inline Gaussian Mixture Modeling (GMM) programming.
• End-to-end conditional distribution tracking for optical signals.
• Experimental validation on a 20-Gbits/s optical Pulse Amplitude Modulation-4 (PAM-4) system.

Main Results:

• Optical PAM-4 signals were down-sampled to 250 Msa/s for statistical characterization using inline GMM.
• The real-time learned distribution allowed intelligent decoding, perfectly adapting to changing Mach-Zehnder intensity modulator bias points.
• Achieved a bit error rate (BER) below 3.8⋅10^-3, significantly enhancing communication reliability.

Conclusions:

• The proposed real-time machine learning scheme effectively tracks optical signal distributions for adaptive IMDD system decoding.
• The method demonstrates robustness against modulator bias shifts, improving communication reliability.
• The scheme offers potential for practical implementation in other machine learning-based signal decoding applications.