MANET-GNN: Learned Decentralized Optimization of Power Allocation in Multi-Channel MANETs

Tomer Alter; Nir Shlezinger; Michael Segal

cs.LGSep 30, 2026

MANET-GNN: Learned Decentralized Optimization of Power Allocation in Multi-Channel MANETs

Authors: Tomer Alter, Nir Shlezinger, Michael Segal

Organizations: ECE School, Ben-Gurion University of the Negev, Israel

Abstract

MANETs enable flexible infrastructure-less wireless connectivity in dynamic and resource-constrained environments. As modern MANETs exploit multiple frequency channels and support heterogeneous traffic patterns, decentralized transmit-power allocation becomes increasingly challenging. We develop a unified learned optimization framework for decentralized power allocation in dynamic multi-hop, multi-channel MANETs. We formulate a constrained end-to-end throughput maximization problem covering unicast, multicast, multicommodity, convergecast, and many-to-many communication. Although centralized and non-convex, this problem serves as an unsupervised training objective for MANET-GNN, a message-passing GNN that operates as a distributed learned optimizer. MANET-GNN uses only local, possibly noisy, CSI and a prescribed number of neighbor message exchanges, enabling low-latency decentralized inference while generalizing across topologies and network sizes. Numerical results show that MANET-GNN achieves centralized-competitive performance across communication frameworks, remains robust to channel uncertainty, and scales effectively across MANET configurations.

Figures & tables

Fig. 1 : Illustration of a multi-channel manet with $B=3$ channels, $|\mathcal{V}|=5$ nodes, and $|\mathcal{E}|=6$ links.

Fig. 2 : Illustration of the considered multi-channel manet communication frameworks.

Symbol	Definition
$w_{{j}}^{(b)}(t)$	AWGN noise received at node $j$ on channel $b$
$h_{i\rightarrow j}^{(b)}(t)$	Channel coefficient between nodes $i$ and $j$ on channel $b$
$p_{i\rightarrow j}^{(b)}(t)$	Power allocated by node $i$ to node $j$ on channel $b$
$s_{i}^{(b)}(t)$	Transmitted signal from node $i$ on channel $b$
$\mathcal{N}_{j}(t)$	Set of neighboring nodes of node $j$

TABLE I : Key variables and parameters

Panel of Figure 4 — Fig. 3 : Gated message-passing layer of MANET-GNN.

Panel of Figure 5 — Fig. 4 : Mean rate versus snr for unicast ( F1 ) communication over QuaDRiGa channels.

Method	Computational Complexity	Total Communication	Node $i$ Communication	Remarks
MANET-GNN (Decentralized)	$\mathcal{O}\!\left(L\|{\bm{\theta}}\|\right)$	$\mathcal{O}\!\left(L\|\mathcal{E}\|BK\right)$	$\mathcal{O}\!\left(L{\rm deg}(i)BK\right)$	$L$ fixed rounds (latency-bounded learned optimizer)
Centralized Optimizer B1	$\mathcal{O}\!\left(I\big(\|\mathcal{E}\|BK+C(\mathcal{G})\big)\right)$	$-$	$-$	Iterative optimization with objective evaluation each iteration
Equal Split B2 (Each Node Locally)	$\mathcal{O}\!\left(\|\mathcal{V}\|\right)$	$-$	$-$	One-shot local allocation (no routing required)
Greedy-Split B3 (Centralized)	$\mathcal{O}\!\left(C(\mathcal{G})+\|\mathcal{V}\|\right)$	$-$	$-$	Shortest-path/smallest-subgraph selection, equal power on chosen route
Greedy-Split B4 (Decentralized)	$\mathcal{O}\!\left(\|\mathcal{V}\|^{2}\|\mathcal{E}\|\right)$	$\mathcal{O}\!\left(\|\mathcal{V}\|\|\mathcal{E}\|\right)$	$\mathcal{O}\!\left(\|\mathcal{V}\|\right)$	Distributed Bellman–Ford-style path discovery [ 58 ]
Best Single Channel B5 (Centralized)	$\mathcal{O}\!\left(C(\mathcal{G})+\|\mathcal{V}\|\right)$	$-$	$-$	Max–min link evaluation per band, full power on the strongest bottleneck

TABLE II : Online complexity of considered algorithms in floating point operations and in communications

Panel of Figure 7 — Fig. 5 : Mean rate versus snr for multicast communication over QuaDRiGa channels.

Panel of Figure 8 — Fig. 6 : Mean rate versus snr for multicommodity ( F3 ) communication over QuaDRiGa channels.

Panel of Figure 9 — Fig. 7 : Mean rate versus snr for convergecast ( F4 ) communication over QuaDRiGa channels.

Panel of Figure 10 — Fig. 8 : Mean rate versus snr for Many-to-Many ( F5 ) communication over QuaDRiGa channels.

Panel of Figure 11 — Fig. 9 : Generalization and depth analysis for MANET-GNN in the many-to-many framework.

Explore similar work

Learning to Optimize: Joint Routing and Flow Allocation on Sparse Non-Euclidean Networks
Jul 26, 2026Haomiao Sun, Fang He, Congyuan Ji +1Vehicle Routing Problem Bringing Network Coding
Department of Industrial Engineering, Tsinghua University, Beijing 100084, P.R. China · School of Management Science and Engineering, Central University of Finance and Economics, Beijing 100081, P.R. China
Operating Regimes of Decentralized Learning Under Mobility and Bandwidth Constraints
Jun 1, 2026Samuele Sabella, Chiara Boldrini, Lorenzo Valerio +2Decentralized Learning Power Allocation
Institute for Informatics and Telematics, National Research Council, Italy
Learning to Cover Locally: Graph Neural Combinatorial Optimization under a Hard Information Horizon
Sep 30, 2026Johannes F. Loevenich, Thies Moehlenhof, Laurin Holz +3Neural Combinatorial Optimization Combinatorial Optimization
cortAIx Labs, Thales Deutschland, Ditzingen, Germany · Department of Mathematics/Computer Science, University of Osnabrück, Osnabrück, Germany

Jul 26, 2026, cs.LGJ/K move · Enter open · S save

Learning to Optimize: Joint Routing and Flow Allocation on Sparse Non-Euclidean Networks

Haomiao Sun, Fang He, Congyuan Ji, Xindi Tang

Department of Industrial Engineering, Tsinghua University, Beijing 100084, P.R. China · School of Management Science and Engineering, Central University of Finance and Economics, Beijing 100081, P.R. China

Vehicle Routing Problem Bringing Network Coding Interleaved Graph Attention Offline Reinforcement Learning Prompt Optimization

Open paper AskPDF arXiv

We study an integrated pickup-and-delivery problem on sparse, non-Euclidean networks that jointly optimizes cyclic routing, cargo flow allocation, and cross-cycle service. The tight coupling of these operational constraints creates a complex discrete-continuous decision space with highly restricted feasible regions. To overcome these computational challenges, we propose Double-Channel Graph Attention (DCGA), an end-to-end reinforcement learning framework. DCGA isolates network reachability and demand-service logic into separate graph channels and constructs valid routes using a simulator-coupled, constraint-informed decoder. Experiments on LinerLib benchmarks demonstrate that DCGA achieves seconds-level inference and delivers state-of-the-art solution quality on instances beyond a specific scale, with its advantage over existing baselines widening significantly as problem size increases. Supported by extensive stability and ablation analyses, our results demonstrate that this structure-aware learning approach provides an effective, low-latency engine for realistic routing-and-flow optimization.

Abstract

Explore similar work

Learning to Optimize: Joint Routing and Flow Allocation on Sparse Non-Euclidean Networks

Operating Regimes of Decentralized Learning Under Mobility and Bandwidth Constraints

Learning to Cover Locally: Graph Neural Combinatorial Optimization under a Hard Information Horizon