Fast Inverse Nonlinear Fourier Transforms for Continuous Spectra of Zakharov-Shabat Type

TL;DR

This paper introduces a fast inverse nonlinear Fourier transform algorithm for signals with only a continuous spectrum, significantly improving computational efficiency for applications like fiber-optic communications.

Contribution

It presents the first fast inverse NFT algorithm for signals without a discrete spectrum, reducing complexity from quadratic to near-linear logarithmic time.

Findings

Achieves $ ext{O}(D ext{log}^2 D)$ complexity, nearly ten times faster than previous methods.

Quantifies the effect of time-domain truncation on the NFT.

Applicable to fiber Bragg grating design and nonlinear inverse synthesis.

Abstract

The nonlinear Schr\"odinger equation (NSE) is well-known to model an ideal fiber-optic communication channel. Even though the NSE is a nonlinear evolution equation, it can be solved analytically using a nonlinear Fourier transform (NFT). Recently, there has been much interest in transceiver concepts that utilize this NFT and its inverse to (de-)modulate data. Fast algorithms for the (inverse) NFT are a key requirement for the simulation and real-time implementation of fiber-optic communication systems based on NFTs. While much progress has already been made for accelerating the forward NFT, less is known on its inverse. The nonlinear Fourier spectrum comprises a continuous and a discrete part in general, but so far only fast inverse NFTs for signals whose continuous spectrum is null have been reported in the literature. In this paper, we investigate the complementary case and propose the first fast inverse NFT for signals whose discrete spectrum is empty. This is the case required by transmitters in the recently proposed nonlinear inverse synthesis scheme, but the problem also occurs in different application areas such as fiber Bragg grating design. Our algorithms require only $\mathcal{O}(D\log^{2}D)$ floating point operations to generate $D$ samples of the desired signal, which is almost an order of magnitude faster than the current state of the art, $\mathcal{O}(D^{2})$. This paper also quantifies, apparently for the first time, the impact that truncating a signal in the time-domain has on the NFT.