Cloudy Skies: Early Detection of Jamming Against Iridium LEO Satellite-IoT Links
summary
The gist
Low Earth Orbit (LEO) satellite-IoT systems increasingly provide connectivity to mobile assets in remote areas…
In short
The study investigated if jamming of Iridium LEO satellite-IoT links can be detected from decoded messages before the link fails. Researchers compared autoencoders to traditional signal quality estimators and found that jamming is detectable when it raises the receiver noise floor by at least 9 dB. This suggests that physical layer distortions, not just power or distance, reveal an attack.
Key concepts
- Physical Layer Information
- This refers to the raw data received from a satellite link before it is fully processed into usable information. The paper focuses on analyzing these raw IQ samples to find subtle distortions caused by jamming that traditional metrics might miss.
- Autoencoder-based Anomaly Detectors
- These are machine learning models trained on normal, unjammed data. They learn the 'normal' pattern of the physical layer signals. When jamming occurs, the resulting distorted signals deviate significantly from this learned normal pattern, triggering an alert.
- Receiver Noise Floor
- This is a measure of the background noise level present at the satellite receiver. The study found that jamming's effect on this floor is the primary factor determining whether an attack can be detected, rather than just how powerful the jammer is.
Terminology used across episodes
This episode discusses
The paper
Cloudy Skies: Early Detection of Jamming Against Iridium LEO Satellite-IoT Links · Read on arXiv
Ingrid Huso, Giulia Fliescher, Savio Sciancalepore, Gabriele Oligeri
Hamad Bin Khalifa University · Eindhoven University of Technology
Low Earth Orbit (LEO) satellite-IoT systems increasingly provide connectivity to mobile assets in remote areas, which makes their resilience to radio-frequency jamming a growing concern. Timely jamming detection, i.e., before complete link failure, has been studied for terrestrial networks and Global Navigation Satellite Systems (GNSS), but it remains unclear whether these approaches are effective on LEO links, where propagation varies rapidly, the received signal is weak, and jamming should be detected from the messages that are still correctly received. This paper investigates the extent to which jamming of Iridium LEO satellite-IoT links can be detected from the physical layer of correctly decoded messages. To this aim, we collect and publicly release a real-world dataset of Iridium messages under 47 jamming conditions, spanning two propagation environments, static and mobile receivers, and different jammer distances, powers, and signal types. We use this dataset to analyze three one-class autoencoder-based anomaly detectors operating on physical-layer information and compare them against two signal-quality estimators commonly used as jamming indicators in GNSS-based applications. Our results show that detectability is governed by the rise of the receiver noise floor, through which jammer power, distance, and receiver mobility act. With 1000-sample chunks, i.e., 40 ms of received signal at 25,000 samples per second (about four correctly decoded messages), the autoencoders reach an AUC of at least 0.84 whenever a noise jammer raises the noise floor by at least 9 dB, whereas the C/N 0 and SNR estimators fail, since received messages, although correctly decoded, retain a high SNR under jamming. Therefore, timely jamming detection for LEO satellite-IoT links like Iridium should rely on physical-layer information beyond the signal-quality metrics adopted in the GNSS domain.
Transcript
Introduction to the show: ident: Security Radio. Generated commentary on the latest security and cryptography papers.
Nadia: We're looking at the paper: "Cloudy Skies: Early Detection of Jamming Against Iridium LEO Satellite-IoT Links".
Elias: Here's the gist of the paper. Low Earth Orbit (LEO) satellite-IoT systems increasingly provide connectivity to mobile assets in remote areas, which makes their resilience to radio-frequency jamming a growing concern Timely jamming detection, i.e., before complete link failure, has been studied for terrestrial networks and Global Navigation Satellite Systems (GNSS), but it remains unclear whether these approaches are effective on LEO links, where propagation varies rapidly, the received signal is weak, and jamming should be detected from the messages that are still correctly received, before the link is lost This paper investigates the extent to which jamming of Iridium LEO satelliteIoT links can be detected from the physical layer of correctly decoded messages Core Investigation The primary goal is to investigate whether jamming of real-world Iridium satellite-IoT links can be identified from the PHY-layer information of correctly decoded messages before link disruption The study compares three one-class autoencoder-based anomaly detectors operating on physical-layer information against two signal-quality baseline estimators commonly used as jamming indicators in GNSS-based applications The authors show that detectability is governed by the rise of the receiver noise floor, through which jammer power, distance, and receiver mobility act Detection Performance and Limitations The results demonstrate that jamming leaves detectable physical-layer distortions not identifiable by traditional signal-quality metrics Specifically, the SNR- and C/N0-based estimators often achieve AUC values below 0.7 even under strong interference conditions because received messages, although surviving ongoing jamming, frequently retain favorable signal-quality metrics The authors find that with 1000-samples chunks, the investigated autoencoderbased anomaly detectors achieve an AUC of at least 0.84 whenever noise jamming increases the receiver noise floor by at least 9 dB Physical Factors Governing Detectability The analysis reveals that detection performance cannot be explained solely by jammer power, distance, waveform, or mobility The key evidence appears in Fig.
Nadia: 8(c): while the measured JSR increases from approximately-20 dB to-8 dB, the SNR of the decoded bursts remains close to 20 dB throughout the experiment This behavior occurs because increasing the jammer power primarily increases the interference observed at the receiver, which directly affects the physical-layer signal processed by the autoencoders Ultimately, what determines the visibility of the attack is the amount of interference reaching the receiver front-end, which manifests as an increase of the measured noise floor Model Selection and Deployment Among the evaluated approaches, the Raw AE offers the most favorable trade-off between detection capability, transferability, and implementation cost The Raw AE contains only 1,072 trainable parameters, compared with 62,850 for the LSTM-AE and 1,655,824 for the IB AE This result suggests that the Raw AE offers the most favorable trade-off between detection capability, transferability, and implementation cost Overall, our results provide three main insights First, jamming detectability is governed primarily by the increase of the receiver noise floor rather than by individual attack parameters such as jammer power, distance, or receiver mobility Second, conventional GNSSinspired indicators based on SNR and C/N0 provide limited detection capability because correctly decoded messages often retain favorable signal-quality measurements despite ongoing interference Third, anomaly detectors operating directly on IQ samples remain effective even within a single SNR class, showing that jamming leaves characteristic physicallayer distortions that are not captured by traditional signalquality metrics These findings demonstrate that early jamming detection in satellite-IoT systems is feasible already from chunks of 1000 samples (i.e.
Paper summary: Nadia: So, we're looking at the paper called "Cloudy Skies: Early Detection of Jamming Against Iridium LEO Satellite-IoT Links." The basic idea here is that because LEO satellite systems are becoming crucial for connecting mobile assets in remote spots, they're getting hit with more radio frequency jamming.
Elias: And the core question they're tackling is whether we can actually detect this jamming before the entire link completely breaks. It’s interesting because on LEO links, you have these specific challenges: fast propagation changes, weak received signals, and the ability to detect it from messages that are still being correctly received.
Priya: What I find important right away is the focus on detecting jamming from physical layer information of messages that survive the interference. That means we're not just looking at a simple signal strength drop; we're digging into what the signal actually looks like after it gets corrupted but still decoded.
Nadia: Exactly. The paper claims they investigated how much jamming of real-world Iridium satellite-IoT links can be identified from that physical layer information of correctly decoded messages. It sets up a real test using a dataset they collected under forty-seven different jamming conditions, covering various receiver types and jammer scenarios <ref:2611.08118#pg1>.
Elias: And the thesis seems to hinge on comparing different anomaly detection techniques against established methods used for things like GNSS jamming indicators. They are specifically testing autoencoder based anomaly detectors against signal quality estimators that rely on metrics like the signal to noise ratio or C over N zero.
Priya: That comparison is key because they found that those traditional SNR and C over N zero based estimators often don't perform well, getting AUC values below zero point seven even when jamming is strong <ref:2611.08118#pg2>. This suggests those standard metrics miss something important about what happens to the message data itself when interference is present.
Nadia: It means the paper argues that because the received messages can still maintain favorable signal quality metrics even while being jammed, those conventional indicators are limited in their ability to spot the trouble early on. The investigation shows that detectability isn't just about how powerful the jammer is or how far away it is.
Paper summary: Elias: And what they found governing this detection capability seems to be more related to how much the receiver noise floor rises, and that rise depends on things like jammer power, distance, and receiver mobility. They point to a specific way this manifests in their data analysis.
Priya: That's where the results get concrete for me. They show that when they used one thousand sample chunks with their investigated autoencoder based anomaly detectors, they achieved an AUC of at least zero point eight four whenever the noise jamming increased the receiver noise floor by at least nine decibels <ref:2611.08118#pg2,the receiver noise floor by at least>. That’s a measurable threshold for detection under those specific testing conditions.
Nadia: So, to put it plainly, this paper demonstrates that you can detect jamming from physical layer distortions using these autoencoder detectors as soon as the receiver noise floor is raised by nine decibels or more <ref:2611.08118#pg2>. That’s a practical threshold for early warning in this context.
Elias: But there's a bit more nuance here regarding what actually causes that noise floor rise, which they explain by looking at how the jamming power, distance, waveform, and mobility all play into it. The key finding seems to be that the visibility of the attack is ultimately determined by how much interference reaches the receiver front-end.
Priya: That leads me to my question about what this means for us: if it’s primarily about the noise floor increase rather than just jammer power or distance, then traditional jamming indicators based on those individual attack parameters are less useful when we're trying to set up early detection systems for satellite IoT.
Nadia: That’s right. And looking at the evaluation methods, they found that the Raw AE approach offered a better trade-off between performance, how easily it can be applied elsewhere, and how much implementation cost there is compared to other models they tested like the LSTM autoencoder or the IB autoencoder.
Paper summary: Elias: The Raw AE was favored because it only had one thousand seventy-two trainable parameters, which is much less than sixty-two thousand eight hundred fifty for the LSTM autoencoder or one million six hundred fifty-five thousand eight hundred twenty-four for the IB autoencoder. It suggests a practical choice for deployment where resources might be limited.
Priya: From a measurement researcher's view, it’s interesting that this model works even when operating within a single signal quality class, meaning it finds distortions that aren't captured by the traditional signal quality metrics they mentioned earlier. That’s a big deal for building more robust detection methods.
Nadia: So, the overall implication is that conventional GNSS inspired indicators based on SNR and C over N zero are limited because messages can still look good even under interference, but these direct anomaly detectors on IQ samples remain effective at spotting those characteristic physical layer distortions. This paper gives us a way to set up early jamming detection in satellite IoT systems based on one thousand sample chunks <ref:2611.08118#pg2>.
Elias: And the authors have released a real-world dataset of Iridium messages collected under forty-seven distinct jamming conditions, which they intend to support future research into the security of these satellite IoT communication systems <ref:2611.08118#pg1,a real-world dataset of Iridium messages>. That publicly available data is important for other researchers to build upon their work.
Priya: So, in short, the paper shows that early detection isn't just about looking at a single parameter; it's about monitoring the physical layer distortions that occur when noise increases the receiver noise floor, and they provide a specific method using raw autoencoders that hits an AUC of zero point eight four under those conditions <ref:2611.08118#pg2,increases the receiver noise floor>.
Nadia: That’s what it boils down to for us: we can start thinking about how to build detection systems that focus on the physical layer characteristics of the messages themselves, rather than just relying on traditional signal quality metrics.
Elias: And for the world, it means that setting up timely jamming detection in LEO satellite systems is feasible even when dealing with the complex propagation and weak signals inherent in that environment, provided you focus on these specific physical layer anomalies.
Conclusion: Nadia: So we've seen how this research looks at detecting jamming from messages that are still mostly okay even when things are jammed on these LEO satellite IoT links.
Elias: Yeah, it’s about figuring out if you can catch the interference before the whole connection goes down, which is a big problem when you're dealing with those fast-moving orbits and weak signals.
Priya: What this paper really focuses on is using these physical layer details—the raw data—to find anomalies that traditional signal quality checks miss.
Nadia: Exactly, and the authors are testing some autoencoder detectors against the usual SNR or C over N zero metrics used for GNSS jamming indicators.
Elias: And they found that those traditional indicators often fail because even when a message is surviving interference, its signal quality metrics can look fine to them.
Priya: That’s what caught my attention, because it means the authors are showing that these direct anomaly detectors on the IQ samples can still spot those physical layer distortions.
Nadia: They found that if the receiver noise floor rises by at least nine decibels, their autoencoder detector hits an AUC of at least zero point eight four across one thousand sample chunks.
Elias: That's a concrete number that shows detection is possible under those specific conditions, but the authors also pointed out it’s not just about the jammer power or how far away it is.
Priya: They showed that what actually determines if an attack is visible is how much interference gets to the receiver front end, which shows up as that noise floor increase.
Nadia: So, they're suggesting you don't need to track every single attack parameter; you just need to monitor that noise floor rise.
Elias: And they chose a specific model called the Raw AE because it had a good balance of performance and implementation cost compared to bigger models like the LSTM or IB autoencoders.
Priya: It’s interesting that they found this approach works even when the signal quality is still technically in one class, which suggests these physical layer distortions are real and detectable.
Nadia: Overall, this paper gives us a way to set up early jamming detection in satellite IoT systems based on one thousand sample chunks.
Elias: It proves that timely detection isn't just about looking at a single parameter; it’s about monitoring the physical layer characteristics of the messages themselves.
Priya: The implication is that we might be able to build more robust detection methods by focusing on those specific physical layer anomalies instead of just relying on standard signal quality metrics.
Nadia: So, we have this method using raw autoencoders that works when the noise floor jumps a certain amount, which is a practical starting point for anyone building satellite IoT security systems.
More episodes
- 2608.27496-ROPE: Routed Origin Policy Enforcement against Indirect Prompt Injection
- Daily Summary for 2026-11-19
- 2607.07085-Is Randomness Necessary for Adaptive Data Analysis?
- The day in conversation for 2026-11-19
- Daily Summary for 2026-11-19
- 2609.10601-Threshold Choice, Not Sample Size, Bounds Trustless Verification of Nondeterministic Compound AI Workflows
- 2610.38786-Adversarial Process Disturbances at a Deployable Alarm Budget: Detector Evaluation on a Simulator and a Real Plant
- 2511.08660-Binary and Multiclass Cyberattack Classification on GeNIS Dataset
- 2604.27302-Static Attribution of Android Residential Proxy Malware Using Graph Kernels
- Daily Summary for 2026-11-18