PaperlayerView exact revision on arXiv

Source-linked AI summary

Information Transmission using the Nonlinear Fourier Transform, Part III: Spectrum Modulation

Mansoor I. Yousefi, Frank R. Kschischang

arXiv:1302.2875v2cs.IT

TL;DR

High-power WDM transmission is limited by nonlinear inter-channel interference, motivating a nonlinear alternative for fiber-optic communication. The paper develops NFDM by modulating discrete nonlinear-Fourier degrees of freedom and analyzes noise perturbations, examples, and implementation constraints. It demonstrates deterministic interference removal in principle, while noting that larger simulations are needed to establish high spectral efficiencies.

  • Problem

    High-power WDM networks face rate degradation from nonlinear inter-channel interference, while existing evidence does not yet establish scalable high-spectral-efficiency nonlinear transmission.

  • Method

    The paper uses NFDM to modulate discrete nonlinear-Fourier spectral data, with inverse-NFT transmission based on the Darboux transform and perturbation analysis of noise.

  • Results

    The paper provides first-order noise results, including zero-mean complex Gaussian eigenvalue perturbations, and examples illustrating NFT-based data transmission.

  • Takeaways & Limitations

    Integrability-based nonlinear-Fourier signaling can provide a transmission framework in which deterministic interference is removed for the cubic NLS channel.

Abstract

from arXiv · show

Motivated by the looming "capacity crunch" in fiber-optic networks, information transmission over such systems is revisited. Among numerous distortions, inter-channel interference in multiuser wavelength-division multiplexing (WDM) is identified as the seemingly intractable factor limiting the achievable rate at high launch power. However, this distortion and similar ones arising from nonlinearity are primarily due to the use of methods suited for linear systems, namely WDM and linear pulse-train transmission, for the nonlinear optical channel. Exploiting the integrability of the nonlinear Schrödinger (NLS) equation, a nonlinear frequency-division multiplexing (NFDM) scheme is presented, which directly modulates non-interacting signal degrees-of-freedom under NLS propagation. The main distinction between this and previous methods is that NFDM is able to cope with the nonlinearity, and thus, as the the signal power or transmission distance is increased, the new method does not suffer from the deterministic cross-talk between signal components which has degraded the performance of previous approaches. In this paper, emphasis is placed on modulation of the discrete component of the nonlinear Fourier transform of the signal and some simple examples of achievable spectral efficiencies are provided.

I. INTRODUCTION

The paper develops NFDM for nonlinear optical channels, focusing on discrete-spectrum modulation and inverse-NFT transmission methods. It motivates this approach by WDM’s nonlinear interference and outlines noise analysis, implementation assumptions, and computational constraints.

  • I. INTRODUCTION: The paper extends earlier NFT work by developing inverse-NFT transmitter methods, analyzing noise effects on received spectra, and illustrating achievable spectral efficiencies.The proposed implementation uses the Darboux transform and examples of transmission schemes.
  • I. INTRODUCTION: NFDM generalizes OFDM to integrable nonlinear dispersive channels and targets deterministic inter-channel and inter-symbol interference removal.Its discrete spectral carriers are treated as stable features of NLS propagation.
  • I. INTRODUCTION: Conventional soliton systems have low spectral efficiency, while pulse-shaping with digital backpropagation saturates around ρ ∼5 −9 bits/s/Hz at SNR ∼20 −30 dB.The paper attributes this saturation to WDM’s incompatibility with nonlinear fiber propagation.
  • I. INTRODUCTION: The paper models fiber propagation with a stochastic NLS equation under bandwidth and power constraints, including both lumped and distributed noise.The system assumes bandlimited Gaussian noise and an equality power constraint.
  • I. INTRODUCTION: WDM is a nonlinear multiuser interference channel with memory, where XPM and FWM create inter-channel interference and pulse trains also experience intra-channel interactions.The simplified model uses M = 1 and pure sinusoids to expose the essential frequency-domain effects.

1) Single-user Memoryless Channels:

Capacity behavior in nonlinear channels depends on memory, receiver processing, and interference assumptions. In WDM networks, inter-channel interference differs from intra-channel effects because users generally cannot cooperate or jointly detect.

  • 1) Single-user Memoryless Channels:: Nonlinearity can cause capacity to saturate when signal-dependent spectral broadening exceeds a receiver’s bandwidth.The precise behavior depends on the definitions of bandwidth and transmission duration.
  • 1) Single-user Memoryless Channels:: A saturating capacity is a serious data-communication limitation because increasing average cost may stop improving rates or restrict admissible input distributions.Peak-power and finite-symbol-cost constraints can also prevent arbitrary increases in average cost.
  • 2) Single-user Channels with Memory:: For finite-memory channels, intra-channel nonlinear effects need not make capacity vanish when the receiver performs joint detection over a long data block.The user’s SPM component can be available for deterministic compensation and coding-based treatment.
  • 2) Single-user Channels with Memory:: Sub-optimal receiver processing, including naive channel inversion, can leave signal-dependent residual terms and cause achievable rates to saturate with input power.The linear-channel analogy contrasts complete ISI removal with nonlinear inversion that retains signal-dependent distortion.
  • 2) Single-user Channels with Memory:: In WDM networks, unknown or non-cooperating interference users make cancellation and alignment difficult, while FWM and XPM drive high-power rate degradation.FWM is cubic in signal amplitude and can overwhelm the desired signal as power or user count grows.

C. Inter-channel Interference as the Capacity Bottleneck in WDM Optical Fiber Networks

The paper identifies WDM’s linear multiplexing as the source of severe nonlinear inter-channel interference and motivates nonlinear multiplexing over NLS-supported modes. NFDM aims to eliminate deterministic cross-talk and inter-symbol interference.

  • C. Inter-channel Interference as the Capacity Bottleneck in WDM Optical Fiber Networks: Linear time- and frequency-domain multiplexing loses orthogonality under nonlinear fiber propagation, producing inter-channel interference that limits WDM rates.The paper proposes multiplexing nonlinear modes instead, including N-soliton functions in the focusing regime.
  • C. Inter-channel Interference as the Capacity Bottleneck in WDM Optical Fiber Networks: A 5-channel, 2000 km WDM simulation shows residual channel-of-interest mismatch after backpropagation because neighboring signals are added and dropped between spans.Backpropagation applied only to the channel of interest cannot remove interference from other users.
  • C. Inter-channel Interference as the Capacity Bottleneck in WDM Optical Fiber Networks: As average transmitted power increases, WDM’s channel-of-interest SNR and information rate vanish to zero in the simulated network scenario.The corresponding interference-limited behavior is also described analytically for high SNR.
  • C. Inter-channel Interference as the Capacity Bottleneck in WDM Optical Fiber Networks: NFDM is designed to induce a multiuser channel in which deterministic inter-channel and inter-symbol interference are simultaneously zero for all users.Here deterministic interference means terms present even without noise.
  • C. Inter-channel Interference as the Capacity Bottleneck in WDM Optical Fiber Networks: This deterministic orthogonalization follows from integrability of the cubic NLS equation and holds for arbitrary dispersion, nonlinearity, signal power, or transmission distance.The claim is specific to the integrable cubic NLS setting.

2) Discrete Spectrum Modulation via the Hirota Bilinearization Scheme:

The Hirota method provides an analytically insightful way to construct N-solitons by converting the nonlinear NLS problem into bilinear equations and summing interacting exponential terms. Its computational complexity grows rapidly with N.

  • 2) Discrete Spectrum Modulation via the Hirota Bilinearization Scheme:: The Hirota method transforms the nonlinear equation into homogeneous bilinear PDEs whose solutions are expressed as sums of exponentials.The transformation separates or cancels the nonlinearity in the resulting bilinear formulation.
  • 2) Discrete Spectrum Modulation via the Hirota Bilinearization Scheme:: The transformed signal q = G/F is built from exponential terms whose interaction structure is organized through binary-vector sums and block-partitioned coefficients.The block structure captures interactions among eigenvalue-related variables and their conjugates.
  • 2) Discrete Spectrum Modulation via the Hirota Bilinearization Scheme:: The method separates amplitude and phase roles: F is real-valued and contributes to amplitude through ∂tt log F, while F alone does not contribute to phase.Both F and G contribute to the signal amplitude.
  • 2) Discrete Spectrum Modulation via the Hirota Bilinearization Scheme:: Hirota’s construction includes two-way, three-way, and higher-order interaction terms among soliton components, removing their interference in the generated signal.The method provides analytical insight into nonlinear signal degrees-of-freedom and N-soliton structure.
  • 2) Discrete Spectrum Modulation via the Hirota Bilinearization Scheme:: ∼22N terms in F and G make Hirota computation rapidly more complex, making N-soliton generation difficult for N > 10 without truncation.This limits the method’s practicality for numerical generation at larger N.

3) Recursive Discrete Spectrum Modulation Using Darboux Transformation:

The Darboux transformation recursively constructs N-solitons from a known solution while preserving existing eigenvalues and adding a selected eigenvalue. This makes it suitable for numerical inverse-NFT generation and discrete-spectrum modulation.

  • Darboux transformation is particularly well suited for numerical implementation, whereas Hirota provides analytical insight and Riemann-Hilbert methods also capture the continuous spectrum.The paper primarily uses Darboux for numerical N-soliton generation and discrete-spectrum simulations.
  • The Darboux transformation constructs a new integrable solution from a known solution and its associated eigenvector.The transformed signal and eigenvector continue to satisfy the underlying integrable system.
  • Recursive Darboux steps preserve the existing eigenvalues and add the selected eigenvalue to construct higher-order N-solitons.Starting from q = 0, repeated transformations generate higher-order solutions.
  • The method updates both the signal and eigenvector at each iteration, as illustrated by the Darboux construction procedure.The paper presents the recursion as a two-step iterative algorithm.

C. Evolution of the Discrete Spectrum

The discrete spectrum evolves through soliton propagation and noise-induced perturbations. In the small-noise regime, eigenvalue fluctuations are approximately zero-mean Gaussian, while distributed-noise variance remains signal-dependent.

  • C. Evolution of the Discrete Spectrum: The real and imaginary parts of an eigenvalue determine soliton frequency and amplitude, respectively, while nonzero real parts produce motion in retarded time.Distinct real parts cause components to travel at different speeds and eventually separate as z →∞.
  • C. Evolution of the Discrete Spectrum: Pulse broadening and component shifts depend on fiber length, mass points, dispersion, and dispersion management, requiring careful consideration when dispersion is unmanaged.The cited discussion treats this as a propagation boundary for multi-soliton signals.
  • C. Evolution of the Discrete Spectrum: The analysis does not comprehensively treat noise because inter-channel interference is regarded as the stronger distortion limiting conventional WDM rates.The model also omits fiber loss under an ideal distributed Raman-amplification assumption.
  • C. Evolution of the Discrete Spectrum: Noise perturbs the discrete eigenvalues and spectral amplitudes from their nominal values, motivating perturbative analysis of the nonlinear Fourier transform.The perturbation is modeled by expanding eigenvectors and eigenvalues in powers of a small noise parameter.
  • C. Evolution of the Discrete Spectrum: The first-order eigenvalue perturbation is distributed as a zero-mean complex Gaussian under the small-noise approximation.The result follows from the perturbation expression involving left and right eigenvectors of the non-self-adjoint operator.

2) Distributed Noise:

For distributed noise, the paper models the stochastic NLS channel through small perturbations and derives an approximate eigenvalue evolution. The resulting deviations are conditionally Gaussian with signal-dependent variance.

  • 2) Distributed Noise: Distributed noise is represented by a normalized noise process scaled by a small parameter associated with noise power.The model combines signal loss and distributed noise in n(t,z).
  • 2) Distributed Noise: The first-order evolution of each eigenvalue is obtained by expanding the eigenvector and eigenvalue in the noise level and taking an inner product with the corresponding adjoint eigenvector.The zeroth-order eigenvalue remains constant while the first-order term varies along propagation.
  • 2) Distributed Noise: The eigenvalue deviation is approximately a zero-mean conditionally Gaussian random variable whose variance depends on the signal.For N ≥2, the relevant eigenvectors are best evaluated numerically.
  • 2) Distributed Noise: The perturbation expansion mainly describes the bulk of the distribution in the small-noise limit because higher-order terms become important as noise variance grows with signal.The first-order term nevertheless accounts for signal dependence in the noise effect.

B. Perturbation of Spectral Amplitudes

The paper uses perturbation analysis to characterize noise-induced fluctuations in nonlinear spectral quantities and then evaluates their implications for soliton transmission and spectral efficiency. The analysis is informative but approximate, with scope limited by small-noise and related modeling assumptions.

  • Noise perturbation analysis: Perturbation analysis models spectral-amplitude fluctuations under noise using an expansion around the noiseless solution.The continuous spectral amplitude is obtained from a Riccati equation, and the state is expanded in powers of the noise parameter.
  • Noise perturbation analysis: If E|q̂1(λ)|^2 is unbounded, the perturbation expansion fails and requires a slow-scale variable T = ϵt.This limitation follows under the stated delta-correlated-noise assumption.
  • Noise perturbation analysis: The first-order perturbation analysis is inaccurate but still provides insight into nonlinear spectral statistics.The paper uses this analysis to motivate subsequent evaluation of achievable spectral efficiencies.
  • Spectral-efficiency evaluation: The analysis assumes small noise and finite propagation conditions so that continuous-spectrum growth, noise-generated solitons, and spectral collapse remain negligible.The stated condition is 2Wzσ2 ≪ E(0), with propagation distance not exceedingly long.
  • Spectral-efficiency evaluation: The baseline on-off-keying soliton system achieves about ρ ≈ 0.15 bits/s/Hz at average power P0 = 0.16 mW.The reported spectral-efficiency values in Fig. 7 are lower bounds because the optimization problem was not solved globally.

B. Spectral Efficiency of 2-Soliton Systems

The paper constructs simple 2-soliton constellations by modulating nonlinear spectral parameters and compares them with conventional on-off-keying soliton transmission. In the example, adding a 2-soliton signal increases spectral efficiency at approximately the same average power, while NFT processing supports decoding.

  • Reference system: The reference OOK system provides about ρ0 = 0.33 bits/s/Hz at P0 = 0.1876 mW and R0 = 7.42 Gbits/s.At the stated noise level, the scheme essentially achieves 2 bits/symbol.
  • Spectral-efficiency comparison: The full constellation reaches ρ = 1.2121 × ρ0 and R = 1.2121 × R0 at approximately the same average power.Its average power is P = 0.46P0, corresponding to P = 0.1748 mW, and its average duration is T = 1.65T1.
  • Spectral-efficiency comparison: The added 2-soliton signal increases the constellation size without much cost in the time × maximum-bandwidth product.The paper identifies this signal as going beyond conventional pulse shapes and notes that such signals are best decoded with the NFT.
  • Receiver considerations: Receiver processing must estimate pulse duration, and fixing the symbol duration can change the balance between signal-set cardinality and spectral efficiency.The paper describes duration checks using NFT computations and gives a maximum duration of 2.58T1 for the example.
  • Propagation behavior: Purely imaginary eigenvalues preserve spectral efficiency through the fiber because they do not undergo major temporal or spectral broadening.The paper therefore reports essentially equal spectral efficiencies at the fiber input and output for these solitons.

2) Modulating Eigenvalues and Spectral Amplitudes:

The paper explores modulating discrete and continuous nonlinear spectral components, reports achievable efficiencies, and discusses NFDM’s potential for multiuser transmission with reduced cross-talk. It also identifies practical design and computational limitations.

  • Modulating Eigenvalues and Spectral Amplitudes:: 1.79×ρ0 bits/s/Hz is achieved at about the same average power, while fixing symbol durations yields 2.2ρ0 = 0.73 bits/s/Hz at 80% power.The constellation uses 16 elements from a 3-ary spectral-amplitude alphabet.
  • Modulating Eigenvalues and Spectral Amplitudes:: Modulating discrete spectral parameters can avoid major temporal or spectral broadening when the real parts of eigenvalues remain unmodulated.
  • Modulating Eigenvalues and Spectral Amplitudes:: Spectral-parameter choices can produce undesirable peak-to-average power ratio, bandwidth, or time duration, requiring signal-set expurgation and better design criteria.
  • Modulating Eigenvalues and Spectral Amplitudes:: A 1.5 bits/s/Hz spectral efficiency is achieved after pruning multi-solitons with undesirable bandwidth or duration.The simulation begins with 30 imaginary-axis eigenvalue points and all N-solitons for 1 ≤ N ≤ 6.
  • Modulating Eigenvalues and Spectral Amplitudes:: The reported spectral efficiency uses a simplistic design, while more sophisticated modulation of amplitudes, phase, and off-axis eigenvalues may yield higher efficiencies.
  • Modulating Eigenvalues and Spectral Amplitudes:: Continuous-spectrum modulation is also considered using raised-cosine pulses propagated through an optical fiber channel.
  • Modulating Eigenvalues and Spectral Amplitudes:: NFT and FFT-based backpropagation achieve approximately the same rates in both single-channel and five-channel WDM simulations.At low SNRs, the two methods achieve about the same rates; at higher SNRs, backpropagation degrades from ISI or inter-channel interference, while NFT may be higher when users are appropriately multiplexed.
  • Modulating Eigenvalues and Spectral Amplitudes:: Simulations do not extend beyond 25–30 dB SNR because of high numerical complexity, coinciding with the onset of significant nonlinearity and discrete spectral mass points.

B. Noise in the Spectral Coordinates

NFDM separates deterministic nonlinear distortions from noise-induced effects by modulating nonlinear spectral degrees of freedom. Its benefits depend on channel integrability, while noise and implementation complexity remain important constraints.

  • Deterministic interference arises from other users even without noise, whereas stochastic interference results from noise coupling with their signals.The paper notes that deterministic interference is typically stronger than stochastic interference.
  • Noise breaks the NLS integrability structure, so NFDM can still experience weak stochastic interference despite avoiding strong deterministic interference.Time-domain white noise may also have correlated spectral coordinates that detectors must account for.
  • NFDM requires an integrable or near-integrable channel; loss, higher-order dispersion, filters, and communication equipment can cause deviations from this assumption.Soliton transmission through practical equipment is presented as evidence that the overall channel may remain nearly integrable.
  • The NFT is computationally demanding: an n-point implementation is currently O(n^2), compared with O(n log n) for the FFT.The transmitter may be even more complex, motivating faster algorithms.
  • In noise-free simulations, NFDM achievable rates are unbounded while WDM rates vanish or saturate at high power because of deterministic distortions.The paper has not simulated NFDM capacity at high SNR, where stochastic interference may eventually cause a finite-rate peak.
  • NFDM provides deterministic orthogonalization: users can overlap in time and frequency while remaining separated in the nonlinear Fourier domain.This removes deterministic inter-channel and intra-channel interactions in the ideal integrable channel.

APPENDIX A SOLUTION OF HIROTA EQUATIONS

The appendix develops Hirota-equation solutions through perturbative expansions and recursive coefficient equations. For integrable NLS equations, the recursion truncates to produce exact finite-order multi-soliton solutions.

  • Exponential functions are used as candidate solutions because the Hirota equations are homogeneous in derivative order.The expansion introduces exponential phases and associated dispersion relations.
  • The functions F and G are expanded in powers of a small parameter, producing recursive equations for successive coefficients.The recursion begins with f0 = 1 and g0 = 0.
  • For integrable NLS equations, the recursive series truncates, yielding exact finite-order solutions rather than an indefinitely continuing expansion.The appendix contrasts this finite termination with the behavior of a general nonlinear system.
  • The construction starts with the one-soliton solution and adds exponential terms to obtain higher-order N-soliton solutions inductively.The appendix explicitly demonstrates the N = 1 and N = 2 cases before stating the inductive construction.
  • For the demonstrated four-term construction, coefficients beyond the fourth iteration vanish, so the expansion terminates.The appendix states that fk = gk = 0 for k ≥ 5.

C. General Formula

The general construction extends the recursive Hirota procedure to N-soliton solutions and relates the spectral parameters used in different solution methods.

  • N-soliton solutions are obtained inductively by adding an additional exp(XN) term and updating F and G through similar calculations.The resulting structure of F and G becomes apparent from this repeated construction.
  • Hirota eigenvalues λi = (αi + jωi)/2 correspond to ζi through ζi = −2jλi = ωi −jαi.The eigenvalues agree across the Riemann-Hilbert, Hirota, and Darboux methods, although other spectral parameters differ.

APPENDIX B PROOF OF THE DARBOUX THEOREM

The appendix proves the Darboux transformation by constructing transformed eigenfunctions and showing that they satisfy the corresponding transformed Lax equations. The proof uses matrix identities involving the Darboux matrix and its commutator terms.

  • A known eigenvector and its adjoint are assembled into a matrix S, with Γ and Σ defining the auxiliary Darboux quantities.The construction uses Γ = diag(λ, λ∗) and Σ = SΓS−1.
  • The matrix Σ satisfies a commutator evolution equation derived from the original Lax system.The appendix gives Σt = [JΣ + Q, Σ].
  • The Darboux transformation maps eigenfunction pairs for q to transformed eigenfunction pairs for q̃.The transformed vectors are represented by V and U, with Λ = diag(µ, µ∗).
  • The proof verifies the transformed evolution by substituting the Darboux relations and simplifying the resulting matrix expression.The final algebra uses commutator identities and the relation between VΛ and ΣV.
  • The transformed functions satisfy both the P-equations and M-equations associated with the transformed potential q̃.This establishes that the transformation preserves the relevant Lax-pair structure.
Loading 1302.2875v2…