Improved Detection Strategies for Nonlinear Frequency-Division Multiplexing

Abstract

Two novel detection strategies for nonlinear Fourier transform-based transmission schemes are proposed. We show, through numerical simulations, that both strategies achieve a good performance improvement (up to 3 dB and 5 dB) with respect to conventional detection, respectively without or only moderately increasing the computational complexity of the receiver.

Mitigating the Impact of Noise on Nonlinear Frequency Division Multiplexing

“…The NFDM setup considered in this work is sketched in the upper part of Figure 1 [13,24] and assumes that only the continuous part of the nonlinear spectrum is modulated. The encoder generates a 16-QAM signal s(t) encoding N b symbols with symbol rate R s = 1/T s = 50 GBd, and adds a guard time of N z null symbols to account for burst transmission and group velocity dispersion (GVD).…”

“…These modified statistics result in an unfavorable dependence of performance on the signal duration, which makes the FNFT strategy highly suboptimal, and practically limits the achievable spectral efficiency [9]. The importance of this issue is now widely acknowledged, so that some alternative detection strategies have been proposed [11,13,[22][23][24][25].…”

“…A possible way to circumvent this difficulty is to resort to the causality property of the NFT demonstrated in [13]. Indeed, the combination of this property with the sequential structure of some NFT algorithms allows us to develop some improved detection techniques which, though suboptimal, yield an improved performance compared to the standard FNFT detection without an exponential increase of complexity [13,23,24]. In the remainder of this section, we illustrate the NFT causality property and its implications in terms of detection, we describe three improved detection strategies, and we discuss their complexity.…”

“…The incremental FNFT (I-FNFT) detection-sketched in Figure 1 in green-uses the NFT causality property to reduce the noise entering the standard FNFT detection, applying an iterative detection [24]. When detecting the k-th symbol, the I-FNFT strategy processes the signal samples up to the time instant t k , such that the noise located after t k does not enter the FNFT, resulting in a reduction of noise in the nonlinear frequency domain, where symbols are detected.…”

“…A possible explanation of this negative result comes from the use of detection strategies borrowed from conventional communication systems, and optimized for the AWGN channel rather than for the actual noise statistics. Indeed, a number of improved detection strategies have been proposed to address this issue [11,13,[22][23][24][25] but, unfortunately, without fully solving it. Another possibility is that the linear modulations (e.g., quadrature amplitude modulation (QAM)) that are employed to encode the information on the nonlinear spectrum, while assuring a certain minimum Euclidean distance between the waveforms in the nonlinear frequency domain, resulting in a too-small distance in the time domain, making the system oversensitive to the ASE noise.…”