Low complexity exponential pruning learned digital back-propagation method for fiber nonlinearity mitigation.

Abstract

Learned digital back-propagation (LDBP) is emerging as a promising solution to mitigate the nonlinear fiber distortions that limit the capacity of optical communications. However, the computational complexity of LDBP, which is affected by dispersion compensation, will increase significantly as optical communications move toward higher baud rates and longer transmission distances. Herein, we propose what we believe to be a novel method called exponential pruning LDBP (EP-LDBP), which is achieved by pruning the redundant LDBP taps with adjustable parameters. Experimental results show that EP-LDBP achieves a 61.5% reduction in computational complexity compared to LDBP in frequency domain without sacrificing compensation performance in a 21-channel wavelength division multiplexing (WDM) transmission over 1600 km fiber using 60 Gbaud 16-ary quadrature amplitude modulation (16QAM). Furthermore, our analysis of EP-LDBP under varying baud rates (30-60 G) and transmission distances (400-1600 km) demonstrates its superior potential in reducing complexity, thus aligning more effectively with the evolving landscape of optical fiber communications.