🔬 optics

Perturbation-based Compensation with EEPN-free Phase Recovery as Back Propagation

The paper proposes a feed-forward perturbation-based compensation method that utilizes noisy received signals to outperform conventional decision-based approaches and, when combined with EEPN-free carrier phase recovery, achieves additional performance gains through a fully symmetrical propagation-backpropagation structure.

Original authors: Chuang Xu, Alan Pak Tao Lau

Published 2026-05-11

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Chuang Xu, Alan Pak Tao Lau

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Big Picture: Fixing a Blurry Message

Imagine you are trying to send a clear, high-definition video message across a very long, bumpy road (the fiber optic cable). As the message travels, two main things mess it up:

The Bumps (Nonlinearity): The road itself distorts the shape of the video.
The Shaky Camera (Phase Noise): The camera at the start and the camera at the end are both shaking slightly, making the image wobble.

The goal of this paper is to fix the video at the destination so it looks exactly like the original, without needing to ask the sender for a "perfect copy" to compare against.

The Old Way: Guessing and Checking

Traditionally, to fix the "bumps" (nonlinearity), the receiver tries to guess what the original message was.

The Analogy: Imagine you receive a muddy footprint. To clean it, you try to guess what the shoe looked like, then try to "un-muddy" it based on that guess.
The Problem: If your guess is wrong (which happens often if the signal is bad), you make the footprint muddier. This is called "error propagation." It's like trying to fix a typo by guessing the word, but if you guess the wrong word, you introduce a new error.

The New Way: The "Backwards Walk" (Perturbation-Based Compensation)

The authors propose a smarter way called Rx-PC (Receiver-based Perturbation Compensation).

The Analogy: Instead of guessing what the shoe looked like, you simply take the muddy footprint and walk the path backwards.
How it works: The math used to describe how the signal gets messed up is like a recipe. If you follow the recipe forward, you get a mess. If you follow the exact same recipe in reverse, you undo the mess.
The Magic: You don't need to know what the original "clean" signal was. You just take the noisy, messy signal you received and run it through a "reverse machine." Because the machine is a perfect mirror image of the road, it naturally cleans up the distortion. This is called a "feed-forward" method because it moves in one direction without needing to stop and ask, "Did I get that right?"

The "Shaky Camera" Problem (Phase Noise & EEPN)

There is a second problem: the cameras at both ends are shaking (Laser Phase Noise).

The Old Setup: Usually, the system fixes the road bumps first, then tries to stop the camera shake. But because of how the road and the camera shake interact, fixing the road after the camera has already started shaking creates a new kind of blur called EEPN (Equalization-Enhanced Phase Noise). It's like trying to stabilize a shaky video after you've already zoomed in too much; the shake gets amplified.
The New Setup (EEPN-free): The authors use a special trick with "pilot tones" (like little lighthouses sent along with the message). These lighthouses help separate the camera shake at the start from the shake at the end.
The Result: They can stop the camera shake before they fix the road bumps. This prevents the "EEPN" blur from ever happening.

The Perfect Symmetry: The "Reverse Link"

The paper's biggest breakthrough is combining the two ideas above.

The Analogy: Imagine the transmission of the signal is a movie played forward.
- Old Method: The receiver tries to edit the movie to fix the glitches, but they are working with a broken script (guessing the symbols).
- New Method: The receiver builds a perfect mirror image of the entire journey.
  1. They undo the camera shake first (using the pilot tones).
  2. Then, they run the signal through the "reverse road" (the Rx-PC) to undo the bumps.
Why it wins: Because the receiver's "reverse machine" is an exact mirror of the sender's "forward machine," every distortion is canceled out perfectly. It creates a "fully symmetrical" link. The paper shows that this symmetry works best when you use the noisy signal before you try to clean up the camera shake, rather than after.

What the Numbers Say

The authors tested this with simulations (computer models of the system):

Less Guessing: Their new method works better than the old "guess-and-check" method, especially when the signal is weak or the lasers are very shaky.
Better Picture Quality: When they combined the "reverse road" fix with the "EEPN-free" camera fix, the signal quality improved significantly (up to 2 dB gain in some cases).
High Shaking Tolerance: Even when the lasers were shaking a lot (simulating older, cheaper lasers), this new method kept the picture clear, whereas the old methods got very blurry.

Summary

The paper introduces a system that fixes distorted internet signals by reversing the journey rather than guessing the answer. By perfectly mirroring the transmission process and fixing the "shaky camera" issue before fixing the "bumpy road," they create a cleaner, more reliable connection without needing to make risky guesses about what the original message was.

Technical Summary: Perturbation-based Compensation with EEPN-free Phase Recovery as Back Propagation

Problem Statement
Fiber nonlinearity (NL) induces significant distortion in coherent optical communication systems, limiting transmission performance. Conventional perturbation-based (PB) post-compensation (PC) methods rely on symbol decisions to reconstruct NL distortion for subtraction. This decision-based approach introduces two primary issues: increased computational complexity and error propagation, where incorrect symbol decisions degrade the NL estimation and compensation efficacy. Furthermore, in systems with laser phase noise (PN), the interaction between chromatic dispersion compensation (CDC) and receiver PN creates Equalization-Enhanced Phase Noise (EEPN), which saturates the Signal-to-Noise Ratio (SNR) at high launch powers. Existing solutions to EPN often involve pilot tones or residual carriers but have not been fully integrated with nonlinearity compensation in a symmetrical framework.

Methodology
The authors propose a feed-forward, perturbation-based compensation scheme that utilizes the noisy received signal directly, eliminating the need for symbol decisions. The methodology is built upon the following principles:

Additive-Multiplicative (AM) Model: The study utilizes the AM-model, which approximates the NL fiber channel more accurately than the standard additive (A) or multiplicative-additive (MA) models. In this model, NL distortion is represented as a combination of a "circular" symmetric component and a "tangent" (phasal) component.
Rx-based Perturbation Compensation (Rx-PC): Unlike conventional decision-based PC (Dx-PC), which subtracts estimated distortion based on decided symbols, the proposed Rx-PC treats the compensation process as a "back propagation" in a perturbative sense. The noisy received sequence ( $\mathbf{r}$ ) is used directly as the input to calculate the NL distortion terms ( $\mathbf{A}$ and $\mathbf{B}$ ), which are then reversed. This avoids the error propagation inherent in decision-feedback loops.
Integration with EEPN-free CPR: The authors combine Rx-PC with an EEPN-free Carrier Phase Recovery (CPR) scheme. This scheme uses pilot tones or a residual carrier at different frequencies to separate and mitigate Transmitter (Tx) and Receiver (Rx) PN before the CDC stage. By isolating and mitigating Rx PN prior to dispersion compensation, the EEPN effect is eliminated.
Symmetrical Link Structure: The core innovation lies in the sequential reversal of channel impairments. By applying EEPN-free CPR followed by Rx-PC (specifically using the signal after CDC but before final CPR, denoted as $C_{DC}C_{DC}$ ), the system forms a fully symmetrical reverse link of the transmission path, including fiber nonlinearity and phase noise.

Key Contributions

Decision-Free Compensation: The paper demonstrates that a feed-forward PC scheme using the raw received signal (Rx-PC) outperforms conventional decision-based methods (Dx-PC) and avoids the complexity of re-mapping post-FEC bits.
Back Propagation Interpretation: The work establishes that Rx-PC is effectively a back propagation operation within the AM-model channel, where the noisy signal serves as the precise input required for reversal, rather than an approximation of the transmitted signal.
Optimal Impairment Sequencing: The study identifies that the order of compensation is critical due to the non-commutativity of impairments. The proposed flow—reversing Rx PN, then CDC, then NL distortion—yields the best performance.
Synergistic Gain: The combination of EEPN-free CPR and Rx-PC creates a fully symmetrical propagation-backpropagation structure, which is not achievable with Pre-distortion (PD) because Tx PN cannot be incorporated into PD calculations performed on clean symbols.

Simulation Results
Simulations were conducted using 32,768 symbols of 45 GBaud DP-16QAM over 15 × 100 km of G.652 fiber with 100–1300 kHz laser linewidths. Key findings include:

SNR Saturation: Conventional CPR exhibits SNR saturation (approx. 20.3 dB) due to EEPN at high powers. EEPN-free CPR improves this by approximately 6 dB, translating to a 1.1 dB gain at optimal launch power.
Rx-PC vs. Dx-PC: The proposed Rx-PC scheme using the signal after CDC ( $C_{DC}C_{DC}$ ) outperforms the version using signal after CPR ( $C_{PR}C_{PR}$ ) and significantly outperforms conventional Dx-PC. The $C_{DC}C_{DC}$ approach preserves Tx PN and NL PN from the forward link, allowing the "back propagation" nature of Rx-PC to effectively reverse them, whereas early CPR mitigates these components prematurely, reducing the efficacy of the subsequent PC.
Performance Gains: With a 1200 kHz linewidth and 25 GHz pilot separation, the combination of EEPN-free CPR and Rx-PC ( $C_{DC}C_{DC}$ ) achieves a peak SNR improvement of 1.75 dB compared to CDC-only, and a Q-factor gain of 2 dB over CDC-only.
Linewidth Dependence: The gains are most pronounced with wider linewidths (e.g., 1200 kHz). At 200 kHz, the gains are smaller (0.25–0.37 dB) due to the reduced impact of Rx PN asymmetry.

Significance and Claims
The paper claims that the proposed method offers a robust solution for compensating intra-channel nonlinearities without the penalty of error propagation associated with decision-feedback. By integrating this with EEPN-free CPR, the system achieves a "fully symmetrical link structure" that reverses the transmission impairments in the correct order. The authors assert that this approach provides significant performance gains, particularly in systems with high laser phase noise, and represents a practical advancement over conventional decision-based compensation and standard CPR techniques. The work highlights that the "back propagation" nature of the algorithm allows for effective compensation even when using noisy, undecided signals.