Gaussian kernel-aided deep neural network equalizer utilized in underwater PAM8 visible light communication system.

Area of Science:

• Optical Communications
• Machine Learning
• Signal Processing

Background:

• Underwater visible light communication (VLC) systems face significant nonlinear distortion, limiting data transmission rates.
• Traditional deep neural network (DNN) equalizers require extensive training and computational resources.

Purpose of the Study:

• To introduce a novel Gaussian kernel-aided deep neural network (GK-DNN) equalizer for mitigating nonlinear distortion in underwater PAM8 VLC channels.
• To enhance the efficiency and performance of VLC systems through optimized equalizer design.

Main Methods:

• Development of a GK-DNN equalizer incorporating a Gaussian kernel to improve distortion compensation.
• Proposal of a novel structural design for the GK-DNN equalizer to reduce computational load and training data requirements.
• Experimental demonstration of a 1.5 Gbps PAM8 VLC system over a 1.2-meter underwater channel using the GK-DNN equalizer.

Main Results:

• The GK-DNN equalizer achieved a 47.06% reduction in necessary training iterations compared to traditional DNN equalizers.
• The proposed GK-DNN equalizer design effectively saved computing resources and reduced the required training data volume.
• Successful demonstration of a 1.5 Gbps PAM8 VLC system with effective nonlinear distortion compensation over 1.2 meters of underwater transmission.

Conclusions:

• The GK-DNN equalizer presents a significant advancement in compensating for nonlinear distortion in underwater VLC.
• The novel design strategy optimizes resource utilization and training efficiency, making it suitable for practical VLC applications.
• This technology enables higher data rates for underwater optical communication systems.