Partial pruning strategy for a dual-branch multilayer perceptron-based post-equalizer in underwater visible light...

Area of Science:

• Optical Communications
• Signal Processing
• Machine Learning

Background:

• Underwater visible light communication (UVLC) systems face challenges with signal distortion.
• Post-equalizers (PEs) are crucial for mitigating these distortions.
• Traditional PEs like Volterra PE can be complex and computationally intensive.

Purpose of the Study:

• To introduce a novel partial pruning strategy for a dual-branch multilayer perceptron-based post-equalizer (DBMLP PE).
• To create a sparse DBMLP PE (SDBMLP PE) with reduced complexity and maintained performance.
• To demonstrate the effectiveness and necessity of this pruning strategy in UVLC systems.

Main Methods:

• Implementing a partial pruning strategy for a DBMLP PE.
• The strategy involves protecting output node connections and linear mapping branches during pruning.
• Experimental validation of the SDBMLP PE in a UVLC system.

Main Results:

• The SDBMLP PE achieved a 36.5% reduction in bit error rate (BER).
• The SDBMLP PE uses only 33.8% of the parameters compared to the Volterra PE.
• The SDBMLP PE exhibited performance comparable to the original DBMLP PE but with significantly lower complexity.

Conclusions:

• The partial pruning strategy effectively reduces the complexity of MLPs in UVLC systems.
• The SDBMLP PE offers superior BER performance and lower parameter count than Volterra PE.
• This work significantly enhances the applicability of artificial neural network-based PEs in UVLC.