Digital pre-distortion using a Gauss-Newton-based direct learning architecture for coherent optical transmitters.

Area of Science:

• Optical Communications
• Signal Processing
• Nonlinear Distortion Mitigation

Background:

• Transmitter nonlinear distortion degrades optical transmission quality.
• Digital pre-distortion (DPD) is a key technique for compensation.
• Existing methods often require complex training or auxiliary models.

Purpose of the Study:

• To introduce and evaluate a novel Digital Pre-Distortion (DPD) coefficient identification method.
• To apply the Direct Learning Architecture (DLA) with the Gauss-Newton (GN) method in optical communications for the first time.
• To compare the performance of the GN-based DLA against the traditional Least-Square (LS)-based Indirect Learning Architecture (ILA).

Main Methods:

• Implementation of the Direct Learning Architecture (DLA) for DPD coefficient identification.
• Utilizing the Gauss-Newton (GN) optimization method for coefficient estimation.
• Comparison with the Indirect Learning Architecture (ILA) employing the Least-Square (LS) method.
• Validation through extensive numerical simulations and experimental results.

Main Results:

• The Gauss-Newton (GN)-based Direct Learning Architecture (DLA) was successfully applied for DPD in optical communications.
• The DLA was realized without the need for an auxiliary neural network.
• The GN-based DLA demonstrated superior performance compared to the LS-based ILA.
• This superiority was particularly evident in low signal-to-noise ratio (SNR) scenarios.

Conclusions:

• The proposed GN-based DLA offers an effective and efficient approach for DPD in optical transmissions.
• The DLA provides a significant advantage over ILA, especially under challenging low SNR conditions.
• This work represents a novel advancement in DPD techniques for optical communication systems.