Reducing computation complexity by using elastic net regularization based pruned Volterra equalization in a 80 Gbps...

Area of Science:

• Optical communications
• Signal processing
• Machine learning applied to signal equalization

Background:

• High-speed optical signals require advanced equalization techniques like Volterra equalization (VE) for performance enhancement.
• The high computational complexity of conventional VE hinders its practical implementation in physical systems.

Purpose of the Study:

• To propose and experimentally demonstrate an elastic net regularization-based pruned Volterra equalization (ENPVE) method.
• To significantly reduce the computational complexity of Volterra equalization while preserving system performance.

Main Methods:

• A three-phase pruning strategy involving pre-training with an adaptive elastic net regularizer, pruning insignificant weights, and retraining the equalizer.
• Experimental demonstration using an O-band 80-Gbps Pulse Amplitude Modulation with 4 levels (PAM4) signal transmitted over 40 km of single-mode fiber (SMF).

Main Results:

• The proposed ENPVE achieved a 97.4% complexity reduction compared to conventional VE.
• A 20.2% complexity reduction was observed compared to L1 regularization-based Volterra equalizer (L1VE).
• The ENPVE method maintained system performance at a received optical power of -4 dBm.

Conclusions:

• ENPVE offers a computationally efficient solution for high-speed optical signal equalization.
• The elastic net regularization effectively prunes redundant coefficients, leading to substantial complexity reduction without performance degradation.
• This method enables more practical physical implementations of advanced equalization techniques in optical communication systems.