Abstract
Characterizing noise in distributed quantum systems is a prerequisite for a functional quantum internet. This paper introduces a comprehensive framework for Quantum Network Tomography (QNT) that extends beyond simple star topologies to arbitrary network graphs. By leveraging "Mergecast" operations and a "progressive etching" algorithm, the proposed method uniquely identifies Pauli channel parameters while rigorously accounting for State Preparation and Measurement (SPAM) errors. Simulations demonstrate high accuracy and robustness to significant photon loss.
Introduction
To realize a large-scale quantum internet, precise knowledge of the noise characteristics of internal links—Quantum Network Tomography (QNT)—is essential. Historically, QNT protocols have been restricted to simple star topologies and specific noise models, such as bit-flip errors. Two primary challenges have hindered broader application: identifiability and SPAM errors. Identifiability refers to the difficulty in uniquely determining individual link parameters from end-to-end measurements due to mathematical ambiguities (e.g., knowing the product of noise but not individual values). SPAM errors, inherent in real-world hardware, often conflate link noise with errors at the network's edges.
Methodology
The authors propose a multi-layered strategy to characterize networks from the periphery to the core.
Mergecast and Progressive Etching
The core innovation is Mergecast, a protocol where two nodes send quantum states to an intermediate node that performs a CNOT gate. This "merging" of information transforms multiplicative noise relationships into a solvable system, resolving the sign ambiguity that plagues Unicast-based methods.
Complementing this is Progressive Etching, an algorithmic "peeling" process. The protocol first identifies the noise parameters of the outermost peripheral channels. Once characterized, these channels are effectively "etched" away, allowing the virtual monitors to move deeper into the network topology. This iterative process continues until the entire network is mapped.
SPAM Error Estimation
Unlike previous works that assume perfect hardware, this framework introduces dedicated circuits to decouple and estimate state preparation error ($s$) and measurement error ($m$) independently. By isolating these hardware-specific errors, the QNT results can be accurately "cleaned" to reflect the true noise of the internal links.
BypassUnicast
For networks characterized by "bypassable Pauli channels"—where specific states pass through unaffected—the authors provide BypassUnicast. This simplified protocol offers higher efficiency and lower Mean Squared Error (MSE) by exploiting the specific symmetries of the noise.
Results and Discussion
The framework was validated using NetSquid, simulating realistic conditions including depolarizing channels and up to 70% photon loss over 10km fiber links.
- Universal Identifiability: The Mergecast + Etching framework was proven to uniquely identify all Pauli channel parameters in arbitrary topologies.
- Robustness: High accuracy was maintained even under significant loss, provided quantum memories were utilized for synchronization.
- Efficiency Gains: The BypassUnicast protocol significantly outperformed general methods for compatible noise types, proving that noise-type awareness allows for leaner tomography.
- Error Analysis: A "V-shape" MSE profile was observed, indicating that balancing sample sizes across different protocol stages is critical for experimental accuracy.
Conclusion
This work provides a foundational architecture for mapping the "quantum void" of internal network links. While Mergecast requires intermediate nodes capable of CNOT operations—a higher hardware requirement than simple forwarding—it offers a rigorous path to full network characterization. Future work will explore non-Pauli noise estimation and the theoretical convergence rates for massive-scale networks.
Summary updated