Joint equalization-demodulation using deep neural networks with embedded signal-dependent noise variance for...

Area of Science:

• Optical Engineering
• Signal Processing
• Underwater Communications

Background:

• Single-photon avalanche diodes (SPAD) enhance detection sensitivity in underwater wireless optical communication (UWOC).
• SPAD detectors introduce nonlinear distortion and signal-dependent noise (SDN), challenging signal detection in UWOC.
• Underwater channel attenuation further degrades signal quality.

Purpose of the Study:

• To propose a novel signal detection scheme for UWOC systems that mitigates SPAD nonlinear distortion and SDN.
• To improve signal recovery accuracy and extend transmission distances in UWOC.

Main Methods:

• Developed a deep neural network with embedded SDN variance (DNN-SDN) for joint equalization and demodulation.
• Characterized the SPAD array noise model and established a nonlinear relationship between detected photons and signal power.
• Integrated SDN variance as an input feature and used a joint loss function optimized via ablation study.

Main Results:

• The DNN-SDN scheme effectively mitigates SPAD-induced nonlinear distortion and SDN.
• Joint equalization and demodulation prevent information loss and error propagation.
• Performance was validated across different underwater conditions.

Conclusions:

• The proposed DNN-SDN signal detection scheme significantly enhances UWOC performance.
• The method extends UWOC transmission distances by 20%-40% compared to existing schemes.