Learning Quantum Entanglement Distillation with Noisy Classical Communications

Hari Hara Suthan Chittoor, Osvaldo Simeone

Abstract

Entanglement distillation is a critical process for establishing high-fidelity quantum links in a future quantum internet. However, traditional distillation protocols rely on the assumption of noiseless classical communication for coordination. This paper introduces NA-LOCCNet (Noise-Aware LOCCNet), a machine learning-based approach that utilizes Parameterized Quantum Circuits (PQCs) to design distillation protocols resilient to classical channel errors. By optimizing a noise-aware objective function, NA-LOCCNet maintains high output fidelity even when classical coordination signals are corrupted, outperforming established protocols like DEJMPS and standard LOCCNet.

Introduction

The realization of a global "Quantum Internet" hinges on the ability to distribute high-quality entangled qubits (ebits) across vast distances. Due to environmental decoherence and hardware imperfections, the entangled pairs initially generated are often "noisy" mixed states. Entanglement distillation addresses this by using Local Operations and Classical Communication (LOCC) to extract a single high-fidelity pair from multiple lower-quality ones.

A significant gap in current research is the reliance on perfect classical communication. In practical scenarios, the classical bits Alice and Bob exchange to coordinate their distillation steps are subject to noise. A single bit-flip can lead to the erroneous retention of a low-quality pair or the discard of a high-quality one. This paper addresses this vulnerability by proposing a "noise-aware" framework for learning distillation protocols.

Methodology: NA-LOCCNet

The authors frame the design of distillation protocols as a supervised learning task using Parameterized Quantum Circuits (PQCs). In this setup, Alice and Bob employ quantum circuits where gate parameters (rotation angles $\theta$) are optimized to maximize distillation performance.

Novel Circuit Architecture

A key contribution is the introduction of a specific two-qubit gate ansatz: the Pauli $ZY$-rotation. The researchers found that this architecture provides greater flexibility and robustness compared to standard CNOT-based designs, allowing the model to find optimal solutions in the presence of noise.

The NA-LOCCNet circuit architecture featuring the Pauli $ZY$-rotation gate

Noise-Aware Optimization

Unlike standard methods that maximize fidelity assuming perfect communication, NA-LOCCNet optimizes for average conditional fidelity. This objective function explicitly incorporates a Binary Symmetric Channel model to account for bit-flip probabilities ($p$). By considering the likelihood of classical errors during training, the resulting protocol is inherently robust to such noise.

Experimental Evaluation

The study utilizes simulated S-states, a common model for noisy entanglement, and evaluates performance across a range of classical bit-flip probabilities ($p \in [0, 0.5]$).

Key Results

The primary finding is the superior robustness of NA-LOCCNet. As shown in the performance comparison, while traditional protocols like DEJMPS and standard LOCCNet see a sharp decline in output fidelity as classical noise increases, NA-LOCCNet maintains significantly higher performance.

Average output fidelity vs. bit flip probability $p$

For instance, at a bit-flip probability of $p=0.25$ and an input fidelity of 0.6, NA-LOCCNet achieves an output fidelity of approximately 0.85, whereas DEJMPS drops to ~0.7. Remarkably, even at maximum entropy ($p=0.5$), where classical signals are purely random, NA-LOCCNet prevents the output fidelity from falling below the input level—a "do no harm" principle that standard protocols fail to uphold.

Discussion and Insights

The success of NA-LOCCNet highlights that the choice of quantum gate architecture (the "ansatz") is as critical as the optimization strategy. The $ZY$-rotation gate enabled the discovery of robust protocols that simpler architectures could not reach. Furthermore, the results demonstrate that robustness does not necessarily come at the cost of efficiency; NA-LOCCNet achieves high fidelity without a significant reduction in the probability of success compared to other machine-learning approaches.

Conclusion and Future Work

NA-LOCCNet represents a significant step toward practical quantum networking by accounting for the realities of noisy classical infrastructure. Future research will focus on scaling these protocols to distill more than two pairs ($k > 2$) and exploring multi-party entanglement scenarios. Additionally, testing these optimized PQCs on Noisy Intermediate-Scale Quantum (NISQ) hardware will be essential to validate their performance against real-world gate errors and decoherence.

The authors have open-sourced their implementation, which is available on GitHub.

Summary updated

BibTeX
@inproceedings{chittoor2023learning,
  author = {Hari Hara Suthan Chittoor and Osvaldo Simeone},
  title = {{Learning Quantum Entanglement Distillation with Noisy Classical Communications}},
  booktitle = {IEEE International Conference on Acoustics, Speech and Signal Processing},
  year = {2023},
  doi = {10.1109/icassp49357.2023.10097202},
}
Figure