Basis expansion model for tracking and equalizing rapidly varying multimode fiber channels

Abstract

In multi-mode fibers, environmental perturbations induce changes in the channel transfer matrix at rates proportional to the total length of affected segments and the number of modes; potentially reaching levels where adaptive equalizers fail in tracking the channel state. We propose a technique that tracks the channel on a frame-by-frame basis by modeling variations in mode coupling coefficients over the period of a full frame of symbols as a linear combination of discrete prolate spheroidal sequences (DPSS). In simulations of rapidly varying channel scenarios where least-mean square (LMS) and recursive least squares (RLS) fail completely, our proposed approach consistently tracks and equalizes the channel variation; yielding a symbol error rate of 10-4 at 15-dB signal-to-noise ratio. DPSS, being matched to the dynamic channel behaviour, is a key factor in minimizing the pilot resources expended in channel estimation, with losses in spectral efficiency starting from 4% up to 20% in the worst case.