Neural network-based dynamic nonlinear MIMO equalization in 3-D polarization multiplexed direct detection systems.

Area of Science:

• Optical Communications
• Signal Processing
• Machine Learning in Photonics

Background:

• Direct detection systems are crucial for high-capacity optical communication.
• Polarization multiplexing increases data rates but requires advanced equalization.
• Nonlinear impairments challenge signal quality in direct detection systems.

Purpose of the Study:

• To develop and validate a dynamic nonlinear MIMO equalizer for 3-D polarization multiplexed direct detection systems.
• To compare the performance of different neural network architectures for multi-dimensional equalization.
• To provide insights for adaptive equalization in high-capacity optical systems.

Main Methods:

• Integration of a dynamic MIMO feedforward equalizer (FFE) and a static MIMO neural network (NN).
• Experimental validation in a 10-km, 150-Gb/s 3-D Stokes vector direct detection (SVDD) system.
• Comparison of single multi-output NN (joint NN) versus multiple single-output NNs (separate NN) architectures.

Main Results:

• The proposed equalizer achieved a 0.1056 normalized generalized mutual information (NGMI) improvement over the Volterra equalizer.
• The separate NN architecture demonstrated superior performance when signal quality varied across dimensions.
• The dynamic nonlinear MIMO equalizer effectively compensated for impairments in the SVDD system.

Conclusions:

• The dynamic nonlinear MIMO equalizer offers significant performance gains for 3-D polarization multiplexed direct detection systems.
• Separate NN architectures are advantageous for equalization when channel conditions differ across signal dimensions.
• This work contributes to the design of efficient adaptive equalization techniques for future high-capacity optical networks.