In the past years, nonlinear frequency division multiplexing (NFDM) has been investigated as a potentially revolutionary technique for nonlinear optical fiber communication. However, while NFDM is able to exploit the Kerr nonlinearity, its performance lags behind that of conventional systems. In this work, we first highlight that current implementations of NFDM are strongly suboptimal, and, consequently, oversensitive to noise: the modulation does not ensure a large minimum distance between waveforms, while the detection is not tailored to the statistics of noise. Next, we discuss improved detections strategies and modulation techniques, proposing some effective approaches able to improve NFDM. Different flavors of NFDM are compared through simulations, showing that (i) the NFDM performance can be significantly improved by employing more effective detection strategies, with a 5.6 dB gain in Q-factor obtained with the best strategy compared to the standard strategy; (ii) an additional gain of 2.7 dB is obtained by means of a simple power-tilt modulation strategy, bringing the total gain with respect to standard NFDM to 8.3 dB; and (iii) under some parameters range (rate efficiency η≤30%), the combination of improved modulation and detection allows NFDM to outperform conventional systems using electronic dispersion compensation.
Robust neural network receiver for multiple-eigenvalue modulated nonlinear frequency division multiplexing system