D2D Power Allocation via Quantum Graph Neural Network

AI summary

Overview

Research area: quantum machine learning and wireless resource management, specifically D2D power allocation using a fully quantum Graph Neural Network (QGNN).
Technical level: Advanced.
One-sentence scope: The paper proposes and evaluates a fully quantum GNN that implements message passing with parameterized quantum circuits (PQCs) for SINR maximization in D2D wireless networks.

What This Paper Is About

D2D wireless networks need to allocate transmit power across many pairs so that signal-to-interference-plus-noise ratio (SINR) is maximized, but classical methods like WMMSE and classical GNNs become costly or suffer memory bottlenecks and over-smoothing on large graphs. The paper builds a fully quantum GNN where message passing is performed by PQCs, then tests it on D2D power control. The goal is to see whether a quantum-native graph model can match or beat classical benchmarks with fewer parameters and better scalability.

Key Contributions

1. Presents a fully quantum Graph Neural Network (QGNN) that implements message passing end-to-end via Parameterized Quantum Circuits (PQCs).
2. Introduces Quantum Graph Convolutional Layers (QGCLs) that encode features into quantum states, process graphs with NISQ-compatible unitaries, and retrieve embeddings through measurement.
3. Proposes a stochastic graph decomposition into k-neighbor star subgraphs to handle qubit limitations, turning a graph with N nodes into N overlapping stars with randomly selected neighbors as leaves.
4. Demonstrates the approach on D2D power allocation for SINR maximization, reporting competitive or better performance versus a classical GCN benchmark and WMMSE, with fewer parameters and inherent parallelism.

Main Findings

• Performance versus WMMSE: The QGNN rapidly surpasses the WMMSE power allocation baseline by epoch 4.
• Final sum-rate: The QGNN reaches approximately 2.67 bps/Hz, about a 6% gain over the classical GCN, while the GCN plateaus at the WMMSE level.
• Training data efficiency: Both models are trained in an unsupervised fashion using only 300 channel realizations, with the Adam optimizer and an initial learning rate of η = 5 × 10^-2.
• Generalization: The QGNN’s training and testing curves align closely, indicating faster convergence and superior generalization compared to the classical GCN.
• Parameter and hardware profile: The abstract reports that the QGNN matches classical performance with fewer parameters and inherent parallelism, using NISQ-compatible unitaries.
• Architecture scale example: The paper illustrates a graph with 4 nodes and 6 edges, each node represented by a 2D feature vector and encoded into a 7-qubit PQC.

Methodology in Plain English

The researchers replace the weight matrices of a classical GNN with parameterized quantum

Authors’ abstract

Increasing wireless network complexity demands scalable resource management. Classical GNNs excel at graph learning but incur high computational costs in large-scale settings. We present a fully quantum Graph Neural Network (QGNN) that implements message passing via Parameterized Quantum Circuits (PQCs). Our Quantum Graph Convolutional Layers (QGCLs) encode features into quantum states, process graphs with NISQ-compatible unitaries, and retrieve embeddings through measurement. Applied to D2D power control for SINR maximization, our QGNN matches classical performance with fewer parameters and inherent parallelism. This end-to-end PQC-based GNN marks a step toward quantum-accelerated wireless optimization.