When Does Spatial Cooperation Help? Contamination Regimes in Decentralized Online Spectrum Anomaly Learning

Alp Berke Ardic; Erdem Koyuncu

cs.LGOct 31, 2026

When Does Spatial Cooperation Help? Contamination Regimes in Decentralized Online Spectrum Anomaly Learning

Authors: Alp Berke Ardic, Erdem Koyuncu

Organizations: Department of Electrical and Computer Engineering, University of Illinois Chicago

Abstract

Learning-based spectrum monitors adapt online to benign propagation changes, but persistent rogue transmissions can gradually become incorporated into the learned background. Spatially distributed receivers provide additional evidence for detecting such anomalies, while decentralized model mixing can also spread contamination to receivers with clean measurements. We study this tradeoff using a decentralized online learner with sample gating and model sharing. We show that, without defenses, contamination propagation is governed by the occupancy rate and is independent of cross-receiver correlation. In contrast, the benefit of spatial fusion depends strongly on correlation and the detector operating point, so cooperation can either improve or degrade detection. We also show that fragmented transmissions can evade gating based on temporally smoothed statistics. Combining smoothed and per-sample gating substantially improves robustness to such intermittent contamination.

Figures & tables

Fig. 1: An

800

-sample rogue dwell, i.e., a continuous interval during which the rogue transmitter remains active. The blindly adapting learner gradually incorporates the rogue into its background model, causing the anomaly score to fall below the alarm threshold while the rogue remains present. Gating prevents this adaptation and keeps the alarm active, but the frozen background estimate becomes stale as the environment drifts. The drift-aware variant continues to track the background while the gate is closed.

Operating point	$\rho_{C}{=}0$	$\rho_{C}{=}0.5$	$\rho_{C}{=}1$
$P_{FA}{=}1\%$ , $\mathbf{a}$	$+0.1$	$+3.8$	$+7.5$
$P_{FA}{=}20\%$ , $\mathbf{a}$	$+6.9$	$+5.5$	$+4.1$
$P_{FA}{=}1\%$ , $3\times\mathbf{a}$	$+16.0$	$+8.0$	$+0.1$
$P_{FA}{=}1\%$ , $4\times\mathbf{a}$	$+25.0$	$+12.5$	$+0.0$

TABLE I: Gain of mean fusion over the best single receiver at

p=0.5

, in percentage points of detection probability, computed by exact exhaustive enumeration.

Fig. 2: Behavior over the occupancy rate and correlation plane. (a) Mean-fusion gain over the strongest single receiver. Lines show exact theory and points show simulation with

200

seeds and

\pm 1

SEM bars. (b) Bias propagated into the clean receiver under unprotected mixing. All five correlation levels land on one line in the rate, as Proposition 1 requires. (c) Contaminated samples admitted. The proposed learner, solid, exceeds local gating, dashed, at low rates when

\rho_{C}=0

and crosses below it as

p

and

\rho_{C}

grow. Panels (b) and (c) use

20

seeds per cell.

Fig. 3: Fragmentation of a fixed transmit budget,

100

contaminated samples of a

400

-sample window with all reachable receivers hit together,

20

seeds,

\pm 1

SEM. Only the burst length

L

varies, and the per-sample clean-reference detection rate stays at

44

46\%

. (a) Alarm performance collapses below the detector memory

1/\alpha=10

, counted per burst (upper) and per contaminated sample (lower). (b) Smoothed gates admit nearly everything at small

L

, the per-sample gate is timing invariant, and the dual gate admits the fewest at every

L

. (c) The blind learner’s absorbed bias barely moves, while the dual gate never absorbs more than

+0.04

dB, at the price of a small stale-model lag at long bursts.

Online Anomaly Detection

Online anomaly detection identifies unusual events in data streams as they arrive, adapting to changing conditions while balancing detection speed, accuracy, and false alarms.

See all 62 papers

Operating Regimes of Decentralized Learning Under Mobility and Bandwidth Constraints
Jun 1, 2026Samuele Sabella, Chiara Boldrini, Lorenzo Valerio +2Wireless Local Area Networks Communication-Efficient Distributed Training
Institute for Informatics and Telematics, National Research Council, Italy
Federated Coreset Subsampling for Distributed Memory Banks
Oct 12, 2026Jason Witzemann, Daniel Pröll, Sebastian HegenbartMemory-Bank-Based Anomaly Detection Coreset Selection
Digital Factory Vorarlberg GmbH, Dornbirn, Austria · University of Salzburg, Salzburg, Austria · Vorarlberg University of Applied Sciences, Dornbirn, Austria
Spatially Distributed Task-Oriented Compression for Multi-Emitter Localization and Characterization with Spectral Overlap
May 31, 2026H. Nazim Bicer, J. Nicholas LanemanModel Compression Wireless Communications
Dept. of Electrical Engineering University of Notre Dame Notre Dame, IN, USA

Jun 1, 2026J/K move · Enter open · S save

Operating Regimes of Decentralized Learning Under Mobility and Bandwidth Constraints

Samuele Sabella, Chiara Boldrini, Lorenzo Valerio, Marco Conti+1

Institute for Informatics and Telematics, National Research Council, Italy

Wireless Local Area Networks Communication-Efficient Distributed Training Decentralized Learning

Open paper AskPDF arXiv

Decentralized learning is a promising paradigm for collaborative training in mobile and pervasive systems, as it avoids a central coordinator and does not require sharing raw data. Yet, most analyses rely on idealized communication assumptions that break down in wireless settings, where connectivity is intermittent, topology changes due to mobility, and bandwidth is limited. We study decentralized averaging under client asynchrony, time-varying contact graphs, and technology-dependent throughput constraints. We implement a fully decentralized protocol that overlaps synchronization with local training and supports partial tensor-level transfers when contacts end early. Using Random Waypoint mobility and multiple wireless technologies (Bluetooth LE, LTE, and Wi-Fi), we quantify how network dynamics and link capacity impact convergence. We identify three operating regimes: (i) inter-contact time largely dictates convergence via mixing, (ii) partial updates are often well tolerated when contacts are frequent, and (iii) very dense contact patterns can trigger contention, reducing effective throughput. These findings provide a practical lens to reason about decentralized learning deployments over realistic wireless systems, highlighting when improving connectivity, increasing bandwidth, or mitigating contention is most impactful.

Abstract

Online Anomaly Detection

Explore similar work

Operating Regimes of Decentralized Learning Under Mobility and Bandwidth Constraints

Federated Coreset Subsampling for Distributed Memory Banks

Spatially Distributed Task-Oriented Compression for Multi-Emitter Localization and Characterization with Spectral Overlap