Since 2019, GPS signals over Europe have been exhibiting a strange anomaly at regular intervals: the signal-to-noise ratio at many monitoring stations drops sharply in the same second, then recovers a few seconds later, as if nothing had happened.
These anomalies last only three to five seconds each time, yet their coverage stretches from the Mediterranean to the Arctic Circle, even crossing the Atlantic to affect Greenland and Canada. No terrestrial source of interference can explain this range. Solar storms can’t explain this spectrum either. The only plausible explanation points to space.
In June 2026, Veritasium released a video that brought this research to public attention (Something is jamming GPS over Europe). The video tells a good story, but its narrative pacing serves a thirty-minute viewing experience. What follows unpacks the paper’s chain of reasoning, making clear how the conclusion was reached step by step: what data, what assumptions, and what boundaries each step relies on.
The paper this article draws on is Clements, Kriezis, and Humphreys’s 2026 “Chasing Lightning: Detecting, Characterizing, and Identifying a Powerful Space-Based GNSS Interference Source” (arXiv:2606.03673), along with the same team’s earlier version presented at the ION GNSS+ conference in Baltimore in 2025, “Transient Space-Based GNSS Interference: Observations and Analysis” (conference paper PDF). The paper has not yet undergone traditional peer review, but independent European teams have partially verified its findings.
GPS signals are extraordinarily weak when they reach the ground. Satellites broadcast at around 50 watts from orbits roughly 20,000 kilometers away, and by the time the signal passes through the atmosphere it has attenuated to well below ambient radio noise. A receiver can fish it out of the noise not because the signal is strong, but because it knows what the signal looks like. Each GPS satellite repeats a specific pseudorandom code; the receiver generates the same code locally, correlates it with the incoming signal, and only counts it as “heard” when it finds a match.
The process is much like listening to someone recite a sentence you already know in a noisy room. You can make it out because you know what they’re going to say next, not because they’re speaking loudly. But if someone else emits anything on the same frequency — even at modest power — this fragile matching process breaks down.
GPS reliability rests on the assumption that no one will transmit a strong signal in that frequency band. And that assumption is precisely what was broken in these events.
The researchers initially faced a simple puzzle: what could simultaneously make GPS receivers in Finland, Italy, and Greenland lose their signal in the same second?
A terrestrial source was ruled out first. Radio waves in the GHz band travel in straight lines and are blocked by the curvature of the Earth. Even if you placed the source on the tallest building in Europe, its direct line of sight would only cover a few hundred kilometers. Yet in these events, the affected stations were thousands of kilometers apart. Aircraft don’t work either — even at an altitude of 10,000 meters, the coverage footprint is still not large enough.
Solar radio bursts were another natural candidate. The Sun occasionally erupts with powerful radio emissions, which can indeed disrupt GPS. But the characteristics of solar bursts don’t match the observed signals. Solar bursts are broadband, affecting multiple GPS frequency bands (L1, L2, L5) simultaneously, whereas the observed interference was narrowband, with a bandwidth of only about 5 MHz, centered near the GPS L1 frequency (approximately 1577.5 MHz). Solar bursts evolve slowly, lasting hundreds of seconds, while these events are short pulses of three to five seconds. Solar bursts affect the entire sunlit hemisphere, yet these events appeared only over Europe.
Having eliminated terrestrial sources and the Sun, the remaining possibility pointed to a clear geometric constraint: the interference source must have been simultaneously above the horizon for all affected stations. The researchers did the calculation for one event on June 9, 2021: even with the lowest possible elevation requirement (0 degrees, meaning the source is right at the horizon), the source must have been at least 1,212 kilometers high. That number locked the answer onto space.
How were these anomalies discovered in the first place? The answer is itself interesting.
There is a global network of several hundred GNSS reference stations called IGS, originally built for geodesy and atmospheric research. These stations record the carrier-to-noise density ratio (C/N₀) of every GPS satellite once per second — in plain terms, “how clearly can I hear this satellite.” The data is publicly archived by NASA, downloadable by anyone.
What the researchers did is conceptually straightforward. For each station, at each moment, calculate how much the current second’s C/N₀ has dropped relative to the average of the preceding seconds. If only one station drops, it could be local interference or equipment trouble. But if dozens of stations, in the same second, drop on the same satellite, the problem is unlikely to be local.
The authors used 165 reference stations and publicly archived data from January 2019 to May 2026, identifying 75 days on which these wide-area transient interference events occurred. The earliest event appeared in October 2019. The temporal pattern of events has one notable feature: they cluster on European working days from Tuesday to Thursday, occurring during business hours, and almost never happen on weekends or at night. Amplifiers don’t pick Wednesdays to break — this pattern suggests human involvement.
With the source confirmed to be in space, the next question was: with tens of thousands of objects in orbit, how do you narrow it down?
The first constraint came from geometry. If a source is simultaneously visible from stations in Finland, Italy, and Greenland, it must be above the horizon for all those stations at the same time. This condition sounds simple, but it is remarkably effective: of the roughly fifteen thousand active objects in orbit, only about 2% could simultaneously satisfy the elevation requirements for all affected stations at any given moment.
The second constraint came from cross-referencing multiple events. The interference occurred on 75 days, not just once. A genuine source must have been in the right position during every event. By overlaying the geographic coverage of multiple events, the candidate set was narrowed further, to approximately 14 orbital objects.
Among these 14 candidates was an Algerian communications satellite whose orbital position matched the timing of several events quite well. But it was eventually eliminated because the researchers discovered that this satellite was itself being jammed during the events. It was not the source — it was another victim.
At this stage, the investigation hit a wall. Fourteen candidates, and no further public data to distinguish among them. At the ION GNSS+ conference in Baltimore in 2025, the researchers publicly presented these interim results and asked other teams for assistance.
There is a key technical bottleneck here, and understanding it is essential to understanding the breakthrough that came next.
Intuitively, you might think: if you know how many decibels the signal dropped, can’t you work backward to the source’s distance and power, and thereby locate it?
The problem is that to infer position from signal strength, you need to know three things: the transmitter’s power, which direction its antenna beam was pointing, and the gain of the receiving antenna at different elevation angles. For an unknown space-based interference source, none of these three are known. Different satellites at different positions, transmitting at different power levels with different beam patterns, could produce nearly identical signal strength patterns on the ground.
Signal strength is a soft constraint: the information it provides is not firm enough to pick a single answer out of a dozen candidates. What the researchers needed was a hard constraint — one that does not depend on transmit power or antenna characteristics. That hard constraint is Time Difference of Arrival.
The turning point came in early 2026. Researchers at the German Aerospace Center (DLR) tried pointing a large tracking antenna at candidate satellites and waiting in real time for an interference signal to appear, without success. The real breakthrough came from an email.
Two European GNSS monitoring stations, in Amsterdam and Trondheim, had captured not only the C/N₀ drop during an interference event on February 11, 2026, but also the raw IQ data. IQ data is the actual radio waveform recorded by the receiver at the signal level — it is the signal itself, not a statistical summary.
The data from these two stations had 2.3 seconds of temporal overlap, and both stations were precisely time-synchronized using GPS satellite signals, with timing alignment accuracy better than 30 nanoseconds. This meant the researchers could compare the time difference between the same signal arriving at two different stations.
The principle of Time Difference of Arrival (TDOA) is simple. Radio waves travel at the speed of light. If Trondheim received the same signal earlier than Amsterdam, the source must be closer to Trondheim. Specifically, a 139-microsecond time difference means the signal source is roughly 42 kilometers closer to Trondheim than to Amsterdam.
This constraint defines a hyperboloid in three-dimensional space: the surface formed by all points whose distance difference to the two stations is 42 kilometers. At the scale of thousands of kilometers, the thickness of this surface is only about 5 meters. Any satellite claiming to be the source must lie on this hyperboloid at that moment.
The researchers used the cross-ambiguity function to estimate TDOA and FDOA (frequency difference, caused by Doppler shift from satellite motion) from the raw IQ data. Then, for every satellite in the candidate set, they took its publicly available orbital data, calculated what the expected time difference and frequency difference would be if the signal were emitted from that position, and compared them to the measured values.
Only one satellite matched: Cosmos 2546. Its position deviated from the hyperboloid by less than 200 meters, well within the uncertainty of the satellite’s publicly available orbital data. Moreover, because the recording spanned 2.5 continuous seconds, the satellite was moving and the hyperboloid was shifting, meaning the true source had to remain aligned throughout the entire interval. Cosmos 2546 passed this test. The probability of a coincidence is indistinguishable from zero.
Cosmos 2546 was launched in May 2020. This means it cannot explain the 2019 events. But it is not alone.
Cosmos 2546 belongs to Russia’s EKS constellation (Unified Space System), an early missile warning system. EKS consists of six satellites operating in Molniya orbits. A Molniya orbit is a highly elliptical orbit with an apogee of about 40,000 kilometers, a perigee of about 1,000 kilometers, and an inclination of roughly 63 degrees. The defining feature of this orbit is that a satellite lingers for an especially long time near apogee — which is over the Northern Hemisphere. The Soviet Union designed this orbit in the 1960s specifically to give communications and early warning satellites prolonged coverage of high-latitude regions.
The researchers looked back at all 75 days of events and found a pattern: on every occasion, at least one EKS satellite was above 35 degrees elevation for all observing stations. The paper’s wording is “highly probable,” not “certain.”
There is an important nuance here. The paper’s identification of Cosmos 2546 strictly applies only to the single 1558.5 MHz event on February 11, 2026 — only that event had raw IQ data enabling TDOA-based localization. The other 74 days of events do not have direct evidence at the same level. The attribution to the EKS constellation is a statistical inference based on observation geometry, not a hard TDOA-level localization on every occasion.
In technical reporting that touches geopolitics, precisely distinguishing “what the paper proved” from “what experts speculate” is not academic pedantry — it is the only line of defense against the narrative being hijacked for political ends.
The paper proved four things. First, since 2019 there has existed a wide-area, transient, narrowband GNSS L1 interference phenomenon covering Europe, Greenland, and Canada. Second, the interference source is in space, at an altitude of at least 1,200 kilometers, ruling out terrestrial sources, aircraft, and solar radio bursts. Third, one 1558.5 MHz event on February 11, 2026, was linked with high confidence to Cosmos 2546 via TDOA localization. Fourth, on every one of the 75 event days, at least one EKS satellite satisfied the observation geometry, and the EKS constellation is highly likely to be the source of these events. The clustering of events on working days and during business hours suggests human involvement.
The paper did not prove three things. First, intent. Whether these signals were weapons tests, covert communications, equipment malfunction, or something else — the body of the paper provides no determinative evidence. The 2025 conference version explicitly states that “intent remains unclear and could be accidental or deliberate.” Second, signal strength. The claim in the video that “the signal is hundreds of times stronger than GPS itself” does not appear directly in the body of the paper and may come from interviews or unelaborated calibration estimates. Third, geographic expansion. The paper’s abstract mentions that a space-based interference source has enormous geographic coverage potential, but the body of the paper does not perform a quantitative simulation of deployment scenarios over, for example, the United States.
In the Veritasium video, Professor Todd Humphreys’s statement “I’m inclined to think this is periodic testing of a capability,” as well as the video’s discussion of covert communications, fall into the category of expert speculation, not paper conclusions. These speculations are themselves reasonable academic discussion, but readers need to know where the boundary lies between them and the paper’s formal conclusions.
Having walked through the entire chain of reasoning, two things matter more than “who did it.”
First, there is an inherent vulnerability in GNSS systems. GPS signals are extraordinarily weak when they reach the ground, relying on the assumption that no one will transmit in that frequency band. Terrestrial interference sources can be blocked by terrain and buildings, but space-based sources beam directly down along the line of sight, hitting everything. The behavior of one particular nation is merely a trigger — the real problem is an intrinsic property of the entire global navigation satellite system architecture. Solution directions include fiber-optic time synchronization, enhanced Loran ground-based navigation systems, and alternative positioning technologies based on magnetic fields or quantum sensing, but none of these are yet at the level of everyday infrastructure.
Second, the unexpected value of open scientific data infrastructure. The IGS network was originally built to measure crustal motion and atmospheric delay, yet it became a global space-based interference monitoring network. Not a single step in the entire chain of deduction relied on classified intelligence: the monitoring data is public, satellite orbits are public, and the raw IQ data was voluntarily provided by peers. What this episode demonstrates is not the technological advantage of any one country, but rather that an open scientific infrastructure generated security value in a completely unanticipated direction.
The paper is currently an arXiv preprint and has not yet undergone traditional peer review. If it passes peer review and is published in a journal, or if an independent team reproduces the TDOA localization results, the credibility of these findings will further increase. Until then, the most appropriate stance is to read it as a rigorous yet bounded research report.
(End of file - total 109 lines)