摘要： Due to the inherently nonlinear nature of optical fibres, the increased demand for transmission capacity means that fibre optic communication systems will reach a limit, known as the Linear Capacity Limit. A radically new solution has received significant attention in the past few years, which is based on Nonlinear Fourier Transform (NFT). Under NFT a signal q(t) in time domain transforms into a continuous, qc(λ), and a discrete, qd(λk), complex spectral part, with continuous and discrete eigenvalues λ, and λk, respectively. Considering only multisolitons (a class of optical signals that have discrete NFT eigenvalues), it is well-known that 1) complex eigenvalues λk are invariant, and 2) the spectral part propagates as: (cid:3031)((cid:1878)) = |(cid:1869)(cid:3038) (cid:1869)(cid:3038) (cid:3031)(0)|(cid:1857)(cid:3037)(cid:2957)((cid:3053)); Φ((cid:1852)) = ∠(cid:1869)(cid:3038) (cid:3031)(0) ? 4(cid:2019)(cid:3038) (cid:2870) (cid:1878), where |(cid:1869)(cid:3038) (cid:3031)(0)| and ∠(cid:1869)(cid:3038) (cid:3031)(0) are the initial spectral amplitude and phase associated with the eigenvalue λk, respectively. In the discrete NFT domain, QPSK modulation has been proposed to encode data on the phase of the NFT spectral components qd(λk). For efficient encoding/decoding of the proposed modulation schemes, it is crucial to have knowledge of NFT noise perturbation on spectral phases and eigenvalues. There are limited reports on correlations among eigenvalues in multi-soliton systems and correlation between eigenvalues and their corresponding phases. However, there is no report with detailed analysis of correlation properties among NFT parameters of multi-soliton signals as a function of initial signal parameters. Here, we investigate correlations of the discrete spectra components in a two-eigenvalue system related to their relative NFT phase differences and distinguish a correlation pattern, which reaches extreme values at a NFT phase difference of π. We show this pattern to be unrelated to the choice of eigenvalues nor their spectral amplitude or constant phase offsets. We show that not only the eigenvalues but also the phases of signals in NFT domain are correlated, approaching a maxima at a NFT phase differences of π. We consider second-order solitons with different eigenvalue pairs, where for each of them, 129 symbols have been defined in the NFT domain with varying initial NFT phase differences Φ1,2=∠qd 2 between 0 and 2π, representing Φ1,2 on discrete positions of propagation over z. For every symbol, 1500 copies have been produced and exposed to white Gaussian noise. Using a combination of the fast inverse NFT algorithm and bidirectional NFT algorithm, each signal is translated into time domain, introduced to Gaussian noise, and translated back into NFT domain. The resulting scattering data has been used to estimate the eigenvalues λ1,2 and corresponding spectral components qd 1,2. The ellipticity (defined as the relative difference of principal axes of the scattering cloud) as a function of phase difference for a range of eigenvalue pairs, Fig.1 a,b, shows almost an ellipse at a phase difference of π and a circle at π/2. While both (Im((cid:79)1),Im((cid:79)2)) and (∠qd 2) are highly correlated at the phase difference of π, |Correlation coefficient|≈1, they are negative of each other, where its reason is under investigation. These correlation properties help to gain deeper physical insight of noise properties of signals in NFT domain, which is crucial for developing efficient NFT communication schemes.