OSNR and nonlinear noise power estimation for optical fiber communication systems using LSTM based deep learning...

Area of Science:

• Optical Communications
• Signal Processing
• Machine Learning

Background:

• Optical signal-to-noise ratio (OSNR) and fiber nonlinearity are crucial for high-speed optical communication system performance.
• Deep learning methods are emerging for OSNR monitoring in fiber optic systems.

Purpose of the Study:

• To propose a novel method for simultaneously estimating OSNR and nonlinear noise power.
• To address the challenges of nonlinearity in optical fiber communication systems.

Main Methods:

• Utilizing a Long Short-Term Memory (LSTM) network for feature extraction in the frequency domain.
• Training the LSTM model to predict OSNR and nonlinear noise power.
• Simulating a 5-channel long-haul optical transmission system using VPI software.

Main Results:

• Achieved a test error of less than 1.0dB for OSNR across various modulation formats (QPSK, 16QAM, 64QAM) with reference OSNR from 15 to 30dB.
• Demonstrated a test error of less than 1.0dB for nonlinear noise power for QPSK and 16QAM signals under different laser linewidths.
• Validated the method in a simulated multi-channel long-haul optical transmission system.

Conclusions:

• The proposed LSTM-based method provides accurate and reliable estimation of OSNR and nonlinear noise power.
• This approach is a promising solution for nonlinearity-insensitive OSNR monitoring and accurate nonlinear noise power estimation in complex optical systems.