Opportunistic Routing in Wireless Communications via Learnable State-Augmented Policies

TL;DR

The paper tackles scalable, packet-based routing in large wireless networks using local information by casting routing as a constrained learning problem. It introduces a state-augmented approach that embeds dual variables into the network state and parameterizes routing policies with Graph Neural Networks, trained via an augmented Lagrangian (MoM) framework. The main contributions are a SA-GNN framework for near-optimal, feasible routing, demonstrated to generalize across network sizes and transferred to real-world topologies and testbeds. The approach offers robust, transferable routing policies suitable for dynamic wireless environments, reducing queue instability and improving network utility without centralized control.

Abstract

This paper addresses the challenge of packet-based information routing in large-scale wireless communication networks. The problem is framed as a constrained statistical learning task, where each network node operates using only local information. Opportunistic routing exploits the broadcast nature of wireless communication to dynamically select optimal forwarding nodes, enabling the information to reach the destination through multiple relay nodes simultaneously. To solve this, we propose a State-Augmentation (SA) based distributed optimization approach aimed at maximizing the total information handled by the source nodes in the network. The problem formulation leverages Graph Neural Networks (GNNs), which perform graph convolutions based on the topological connections between network nodes. Using an unsupervised learning paradigm, we extract routing policies from the GNN architecture, enabling optimal decisions for source nodes across various flows. Numerical experiments demonstrate that the proposed method achieves superior performance when training a GNN-parameterized model, particularly when compared to baseline algorithms. Additionally, applying the method to real-world network topologies and wireless ad-hoc network test beds validates its effectiveness, highlighting the robustness and transferability of GNNs.

Figures (13)

• Figure 1: Performance comparison between two unparameterized approaches—Method of Multipliers (MoM) and Dual Descent (DD)—for a network configuration with 10 nodes and 4 flows, evaluated over a single time step.
• Figure 2: Comparison of performance between the parameterized state-augmented approach utilizing GNNs, the unparameterized Dual Descent method and ExOR protocol for networks consisting of 10 nodes and 4 flows, evaluated over $T=100$ time steps.
• Figure 3: Comparison of the performance of state-augmented and dual descent algorithms for networks with 4 flows and varying node counts ($N\in\{10, 20, 50, 75, 100\}$).
• Figure 4: Comparison of state-augmented and dual descent algorithms for a network with 50 nodes and varying flow counts ($K\in\{4, 8, 12, 16\}$).
• Figure 5: Performance of state-augmented algorithm relative to the dual descent for a random network with 50 nodes and 4 flows.