ASTRA: ADMM-Accelerated Topology Reconfiguration for Dynamic Satellite Constellations

João Norberto; Ricardo Ferreira; Cláudia Soares

cs.LGSep 27, 2026

ASTRA: ADMM-Accelerated Topology Reconfiguration for Dynamic Satellite Constellations

Authors: João Norberto, Ricardo Ferreira, Cláudia Soares

Organizations: Department of Computer Science NOVA School of Science and Technology Caparica, Portugal

Abstract

Dynamic topology reconfiguration is central to the reliability and efficiency of large satellite constellations, yet many existing approaches rely on idealized assumptions such as full constellation deployment or uniform orbital spacing. We present Adaptive Satellite Topology via Regret-Aware learning (ASTRA), a theoretically-grounded framework for dynamic satellite topology reconfiguration that builds on an online learning formulation and makes it computationally practical. ASTRA combines an ADMM-based offline solver with efficient online updates for both online gradient descent and online conditional gradient, yielding markedly cheaper constrained updates than generic optimization pipelines. On the theory side, we show that for a relevant class of entry-wise nonzero utility matrices, the objective is strongly convex, which yields logarithmic static regret for online gradient descent, and we further instantiate known dynamic-regret guarantees under inexact ADMM inner loops. Empirically, ASTRA matches or improves topology quality, presenting a good trade-off with computational time on synthetic constellations, and it remains effective on real Starlink data under partial deployment and non-uniform spacing, where idealized structural assumptions break down. These results position ASTRA as an efficient and theoretically grounded approach to topology reconfiguration in realistic Low Earth Orbit networks.

Figures & tables

Metrics	E $\uparrow$	A-Deg $\uparrow$	CC $\downarrow$	A-SP $\downarrow$	Time (s) $\downarrow$
+Grid	130	3.94	1	4 .036	0.001
$\times$ Grid	124	3.76	1	4.978	0.001
DoTD	1 31	3 .97	1	4.412	0.023
Motif	132	4.00	1	4.292	1.326
GUTO	128	3.88	1	3.767	0.091
ASTRA	128	3.88	1	3.767	0.071

Table 1: Methods applied on Iridium-like constellation, with 6 orbital planes and 11 uniformly-spaced satellites per plane. ASTRA is capable of achieving the same results as GUTO in a fraction of the time, while closely matching with state-of-the-art performance.

		Metrics
Method	Approximation	E $\uparrow$	A-Deg $\uparrow$	CC $\downarrow$	A-SP* $\downarrow$
+Grid	—	$2869.14_{\pm 14.123}$	$3.608_{\pm 0.017}$	$9.038_{\pm 2.019}$	—
$\times$ Grid		$2468.77_{\pm 19.325}$	$3.205_{\pm 0.026}$	$3.346_{\pm 1.314}$	—
DoTD		$\mathbf{3183.16_{\pm 0.586}}$	$\mathbf{3.998_{\pm 0.001}}$	$4.211_{\pm 1.436}$	—
GUTO	H	2900.95 ± 13.033	3.650 ± 0.015	$\mathbf{1.233_{\pm 0.491}}$	$15.742_{\pm 0.388}$
ASTRA	H	$2900.24_{\pm 13.058}$	$3.649_{\pm 0.015}$	1.264 ± 0.533	$15.760_{\pm 0.381}$

Table 2: Comparison on Starlink data. Only GUTO and ASTRA attain runs with a single connected component; therefore, A-SP* is reported only for topologies satisfying this condition. ASTRA preserves the connectivity regime of GUTO while improving efficiency.

Algorithm:	+Grid	$\times$ Grid	DoTD	GUTO	ASTRA
Time (s)	$0.021_{\pm 0.001}$	$0.026_{\pm 0.001}$	$3262.927_{\pm 48.097}$	$665.027_{\pm 2.093}$	$567.949_{\pm 5.615}$

Table 3: Average time and respective std each method takes to propose a topology (Starlink), after 10 runs.+Grid and

\times

Grid are the fastest methods.

Appendix figures & tables6 assets

Supplementary material from the paper’s appendix.

Appendix

Parameter:	Description:
$\lambda$	Weight given to the utilities being considered.
$\rho$	Penalty parameter from the Augmented Lagrangian.
$\epsilon$	Stopping criteria value.
$\text{proj}_{its}$	Number of iterations for the Projected Gradient (Alg. 1 ) that deals with the update over the matrix $X$ .
T	Maximum number of iterations of the ASTRA algorithm.
$\Lambda$	Weight to be given to the connections between satellites.

Appendix

Table 4: List of the parameters for the ASTRA offline algorithm.

$\lambda$	Approximation	E $\uparrow$	A-Deg $\uparrow$	CC $\downarrow$	A-SP $\downarrow$	Time (s)
		Metrics
0	H	121	3.667	1	4.678	0.047
0.15		122	3.697	1	4.648	0.055
0.30		123	3.727	1	4.650	0.071
0.45		122	3.697	1	4.726	0.104
0.60		124	3.758	1	4.482	0.133

Appendix

Table 5: Effect of the parameter

\lambda

and the approximation of the FOV of a satellite on the results of the ADMM-based algorithm, for an Iridium-like constellation.

$\rho$	Approximation	E $\uparrow$	A-Deg $\uparrow$	CC $\downarrow$	A-SP $\downarrow$	Time (s)
		Metrics
0.001	H	128	3.879	1	4.082	0.204
0.01		124	3.758	1	4.482	0.212
0.1		124	3.758	1	4.482	0.178
0.25		124	3.758	1	4.482	0.078
0.5		124	3.758	1	4.482	0.086

Appendix

Table 6: Effect of the parameter

\rho

and the approximation of the FOV of a satellite on the results of the ADMM-based algorithm, for an Iridium-like constellation.

$\text{proj}_{its}$	E $\uparrow$	A-Deg $\uparrow$	CC $\downarrow$	A-SP $\downarrow$	Time (s)
	Metrics
5	124	3.758	1	4.556	0.075
10	124	3.758	1	4.556	0.108
20	124	3.758	1	4.556	0.207
30	124	3.758	1	4.556	0.323
40	124	3.758	1	4.556	0.412

Appendix

Table 7: Effect of the parameter

\text{proj}_{its}

on the results of the ADMM-based algorithm, for both synthetic and real data.

Algorithm:	E $\uparrow$	A-Deg $\uparrow$	CC $\downarrow$	SP* $\downarrow$	$\#\{\neg P\}$ $\downarrow$	A- $\#\{\neg P\}$ $\downarrow$
	Metrics
GUTO	$\mathbf{2898.71_{\pm 13.993}}$	$\mathbf{3.647_{\pm 0.017}}$	$\mathbf{1.129_{\pm 0.335}}$	$\mathbf{15.668_{\pm 0.372}}$	0	$0_{\pm 0.0}$
ASTRA	$2897.87_{\pm 14.232}$	$3.646_{\pm 0.017}$	$1.171_{\pm 0.377}$	$15.704_{\pm 0.355}$	0	$0_{\pm 0.0}$
ASTRA-OGD	$2746.68_{\pm 19.604}$	$3.456_{\pm 0.024}$	$1.855_{\pm 0.728}$	$14.303_{\pm 0.431}$	6153	$\mathbf{89.174_{\pm 9.902}}$
ASTRA-OCG	$\mathbf{3169.20_{\pm 2.511}}$	$\mathbf{3.984_{\pm 0.003}}$	$\mathbf{1.029_{\pm 0.168}}$	$\mathbf{11.103_{\pm 0.473}}$	15199	$220.275_{\pm 48.962}$

Appendix

Table 8: Graph metric comparison between offline and online algorithms for 70 consecutive Starlink topologies. In addition to graph metrics, we also consider the total number of connections that do not respect matrix

P

, “

\#\{\neg P\}

”, and its average value per topology “A-

\#\{\neg P\}

”.

Number of iterations:	20K	30k	40k
Time (s)	$330.870_{\pm 3.462}$	$430.103_{\pm 0.983}$	$658.597_{\pm 3.747}$

Appendix

Table 9: Wall-clock time needed for ASTRA-OCG to achieve a certain number of iterations for a fixed topology. The mean values and respective std are presented for 10 runs.

Explore similar work

Online Learning for Dynamic Constellation Topologies
Mar 26, 2026João Norberto, Ricardo Ferreira, Cláudia SoaresLow Earth Orbit Satellite Networks Communication Topologies
NOVA School of Science and Technology, Caparica, Portugal
FedRings: A Scalable and Topology-Aware Federated Learning Framework for LEO Satellite Constellations
Aug 4, 2026Ziwu Liu, Inês Pinto Gouveia, Rehana Yasmin +2Low Earth Orbit Satellite Networks Federated Learning
CEMSE Division, King Abdullah University of Science and Technology (KAUST) Thuwal 23955-6900, Kingdom of Saudi Arabia
Differentiable Satellite Constellation Configuration via Relaxed Coverage and Revisit Objectives
Apr 21, 2026Shreeyam Kacker, Kerri CahoyMultiple Satellite Systems Near-Earth Objects
Department of Aeronautics and Astronautics MIT

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

Online Learning for Dynamic Constellation Topologies

João Norberto, Ricardo Ferreira, Cláudia Soares

NOVA School of Science and Technology, Caparica, Portugal

Low Earth Orbit Satellite Networks Communication Topologies

Open paper AskPDF arXiv

The use of satellite networks has increased significantly in recent years due to their advantages over purely terrestrial systems, such as higher availability and coverage. However, to effectively provide these services, satellite networks must cope with the continuous orbital movement and maneuvering of their nodes and the impact on the network's topology. In this work, we address the problem of (dynamic) network topology configuration under the online learning framework. As a byproduct, our approach does not assume structure about the network, such as known orbital planes (that could be violated by maneuvering satellites). We empirically demonstrate that our problem formulation matches the performance of state-of-the-art offline methods. Importantly, we demonstrate that our approach is amenable to constrained online learning, exhibiting a trade-off between computational complexity per iteration and convergence to a final strategy.

Abstract

Explore similar work

Online Learning for Dynamic Constellation Topologies

FedRings: A Scalable and Topology-Aware Federated Learning Framework for LEO Satellite Constellations

Differentiable Satellite Constellation Configuration via Relaxed Coverage and Revisit Objectives