An Adaptive Purification Controller for Quantum Networks: Dynamic Protocol Selection and Multipartite Distillation

Efficient entanglement distribution is a cornerstone of the Quantum Internet. However, physical link parameters such as photon loss, memory coherence time, and gate error rates fluctuate dynamically, rendering static purification strategies suboptimal. In this paper, we propose an Adaptive Purification Controller (APC) that automatically optimizes the entanglement distillation sequence to maximize the goodput, i.e., the rate of delivered pairs meeting a strict fidelity threshold. By treating protocol selection as a resource allocation problem, the APC dynamically switches between purification depths and protocols (BBPSSW vs. DEJMPS) to navigate the trade-off between generation rate and state quality. Using a dynamic programming planner with Pareto pruning, simulation results show that our approach mitigates the “fidelity cliffs” inherent in static protocols and reduces resource wastage in high-noise regimes. Furthermore, we extend the controller to heterogeneous scenarios, and evaluate it for both multipartite GHZ state generation and continuous-variable systems using effective noiseless linear amplification models. We benchmark its computational overhead, showing decision latencies in the millisecond range per link in our implementation.

💡 Research Summary

The paper addresses a fundamental challenge for the emerging Quantum Internet: how to distribute high‑fidelity entanglement over noisy, time‑varying links while keeping resource consumption low. Classical approaches in existing quantum network simulators (NetSquid, SimQN, SeQUeNCe, QuNetSim) typically fix the number of purification rounds or select a single protocol per link, ignoring the dynamic nature of loss, memory decoherence, and gate errors. To overcome this limitation, the authors introduce the Adaptive Purification Controller (APC), a modular component integrated into the KOSMOS quantum‑network simulation framework.

APC takes as input a pre‑computed routing path, a target end‑to‑end fidelity F*, and per‑link physical parameters (channel length, initial fidelity, Bell‑state measurement success probability, classical communication delay, memory coherence time, and gate error rates). It then produces a detailed plan specifying, for each hop, which purification protocol to use (BBPSSW or DEJMPS), how many recurrence rounds to apply, and where to perform entanglement swapping. The controller also supports optional post‑processing stages: multipartite GHZ stabilizer purification after the bipartite arms are established, and continuous‑variable (CV) distillation based on noiseless linear amplification (NLA) models.

The technical core consists of three tightly coupled layers:

1. Physics Backend – Implements closed‑form fidelity update formulas for BBPSSW (Eqs. 2‑3) and DEJMPS (Eqs. 5‑7), as well as conversion between Werner parameters and Bell‑diagonal vectors. It also models local depolarizing noise on gates and measurements (Eq. 17) and includes effective models for GHZ stabilizer passes and CV NLA stages (Eq. 12). Multi‑round success probabilities and raw‑pair consumption are derived from Eq. 9‑10, providing a quantitative “entanglement cost” metric.

2. Dynamic‑Programming Planner with Pareto Pruning – For each link the planner enumerates feasible (protocol, round‑count) options, propagates cumulative time, cost, and success probability, and discards any partial plan that is dominated in the three‑dimensional objective space. This Pareto pruning dramatically reduces the combinatorial explosion: although the theoretical search space scales as O(N·R·P) (N hops, R max rounds, P protocols), the actual number of surviving candidates remains in the low hundreds, enabling near‑real‑time decision making.

3. Policy Module – Uses heuristic rules based on link loss, initial fidelity, and error rates to bias the planner toward the more suitable protocol. For example, when error asymmetry is high, DEJMPS is favored because it handles Bell‑diagonal noise more robustly; when loss is modest and initial fidelity already above ~0.8, BBPSSW’s higher per‑round success probability makes shallow purification preferable.

The authors evaluate APC on a series of simulated repeater chains ranging from 5 to 10 hops, with channel losses between 0.1 and 0.4 dB/km, memory coherence times from 10 ms to 1 s, and gate error probabilities from 10⁻⁴ to 10⁻². Three baselines are compared: (i) fixed‑depth BBPSSW, (ii) fixed‑depth DEJMPS, and (iii) a naïve static scheduler. The results demonstrate several key benefits:

• Avoidance of “fidelity cliffs.” Static schemes exhibit a sharp drop in overall success probability once the number of purification rounds exceeds a critical value, because the per‑round success probability falls faster than fidelity improves. APC dynamically selects shallower depths in high‑noise regimes, thereby staying on the favorable side of the cliff and achieving up to a 2.3× increase in goodput (the rate of delivered pairs meeting the fidelity target).

• Resource efficiency. In scenarios with severe loss, APC reduces unnecessary deep purification and instead performs more frequent swapping, cutting overall latency by ~15 % while maintaining the fidelity constraint.

• Multipartite extension. By adding a GHZ stabilizer purification stage, APC raises multipartite goodput by a factor of 1.8 relative to bipartite‑only strategies, provided the GHZ success probability exceeds ~0.6. The planner treats the GHZ stage as an additional cost‑benefit trade‑off, integrating it seamlessly into the path‑level optimization.

• Continuous‑variable capability. Using the effective NLA model, APC can optimize the number of NLA stages K and gain g to improve logarithmic negativity from 0.45 to 0.62, illustrating the flexibility of the framework beyond discrete‑variable qubits.

• Computational overhead. Decision latency per link ranges from 1.2 ms to 4.8 ms, with total planning time under 5 ms for a 10‑hop path. This meets the sub‑10 ms real‑time control budget anticipated for 6G‑era quantum hardware.

The paper’s contributions are summarized as follows:

1. A modular APC that cleanly separates physics modeling from network‑level planning, allowing future extensions (e.g., new noise models, alternative distillation protocols) without altering the planner.
2. A fast, Pareto‑based dynamic‑programming algorithm that simultaneously optimizes time, entanglement cost, and end‑to‑end success probability.
3. Unified support for bipartite recurrence (BBPSSW/DEJMPS), multipartite GHZ stabilizer purification, and CV NLA‑based distillation within a single planning framework.
4. Empirical evidence that adaptive depth and protocol selection significantly improve goodput and reduce wasted resources compared with static baselines.

In conclusion, the Adaptive Purification Controller demonstrates that intelligent, link‑aware selection of purification depth and protocol can substantially enhance the performance of quantum networks operating under realistic, fluctuating conditions. The authors suggest future work on hardware‑in‑the‑loop experiments, scaling to more complex topologies (meshes, random graphs), and incorporating machine‑learning‑driven policy learning to further refine adaptation.