Project Overview

This project introduces QuRA, a Deep Q-Reinforcement Learning (DQRL)–based routing framework for quantum networks.
Quantum routing is challenging due to fidelity degradation, probabilistic links, and resource contention. Existing shortest-path and ILP-based solutions either perform poorly or fail to scale.

QuRA addresses these issues by learning adaptive, per-hop routing decisions that dynamically balance fidelity, congestion, and resource usage, achieving high performance with significantly lower computation cost.

Key Ideas

• Fidelity-aware routing: Explicitly accounts for fidelity loss from entanglement swapping
• Hop-by-hop decision making: Reduces complexity while maintaining high routing quality
• Multi-request support: Jointly schedules and routes concurrent requests
• Topology generalization: Works across different networks without retraining

QuRA Routing Framework

QuRA operates via a centralized controller with a DQRL agent that selects a (request, next-hop) action at each step based on the global network state, incrementally constructing end-to-end entangled paths.

Fidelity-Aware Motivation

Hop-count–based shortest paths may fail to meet fidelity thresholds, while QuRA learns to select longer but higher-fidelity paths, enabling successful entanglement delivery under strict quality constraints.

Performance Highlights

• Up to 90% higher success rate under fidelity constraints
• ~79% runtime reduction compared to ILP-based routing
• Scales efficiently with network size and traffic load
• Adapts to dynamic conditions without retraining

Publication

Tasdiqul Islam, Engin Arslan, Md Arifuzzaman

QuRA: Reinforcement Learning–Based Routing for Quantum Networks
IEEE Consumer Communications and Networking Conference (CCNC’26). IEEE, Las Vegas, NV, USA. (Accepted for publication)