Navigating with no GPS, by the rocks below

I have written in this publication about the two ways to attack a satellite navigation signal: jamming, which denies you a position, and spoofing, which hands you a false one, and about how you go find the transmitter doing it. Both of those pieces end in the same uncomfortable place. GPS is a whisper from twelve thousand miles up, and anything that faint can be shouted over by someone standing much closer. That is not a flaw anyone is going to fix. It is the geometry.

So the serious work has moved to the other question. If the satellites are going to be unreliable, what else do you navigate by?

The map was always down there

Here is the idea, and it is old. The Earth's crust is not uniformly magnetized. Different rocks, laid down at different times with different iron content, leave a fixed, wrinkled pattern of small magnetic variations across the planet. That pattern is not random and it is not repeating. Over any decent stretch of ground it is effectively a fingerprint. Survey it once, store the map, and you have a reference that no one can turn off, because nobody is transmitting it. It has been sitting under everyone's feet since before there were navigators.

Magnetic-anomaly navigation, MagNav, uses exactly that. You fly along measuring the local magnetic field, you compare what you are measuring against the stored anomaly map, and where the measured sequence lines up with the map, you get a position fix. The recent field-trial paper describes the principle in one clean sentence: it provides "passive, non-jammable navigation through periodic position fixes obtained by comparing local measurements of Earth's crustal field against known anomaly maps."

The word doing the work there is passive. This is the part my old job trained me to notice before anything else. A radar finds things by flooding the sky with energy, which means an adversary with a cheap receiver detects the radar long before the radar detects the adversary. I made that argument at length when I wrote about drone sensors that listen instead of scan, and about what your own transmitters cost you. A magnetometer is a receiver and only a receiver. It puts nothing into the air. There is no emission to jam, no signal to spoof, and no beam to home in on. So "undetectable" here is not marketing. It is a straightforward consequence of the sensor being passive, and it is the single most valuable property on this list, because in a contested spectrum the thing that gives you away is usually your own transmitter.

What the trials actually reported

The reason this is a 2026 story and not a 1970s story is that the measurement finally got good enough. In a preprint posted in April 2025, a nineteen-author team reported field trials of a MagNav system built around what they call proprietary quantum magnetometers, paired with new denoising and map-matching algorithms, in a package small enough to fit "on fixed-wing drones or in the avionics bay of a commercial airliner." They flew it at altitudes up to 19,000 feet and compared it against a strategic-grade inertial navigation system.

Their headline results, in their words: the system delivered "up to 46x better positioning error than the velocity-aided INS," and "the best final positioning accuracy we achieve is 22m or 0.006% of the flight distance." Across varying conditions, altitudes and flight patterns, airborne trials "consistently achieve at least 11x advantage over the INS." They also ran what they describe as the first successful MagNav in a ground vehicle using publicly-available anomaly maps, with bounded error 7x lower than the INS.

In July 2026 the company behind that work, Q-CTRL, announced that its Ironstone Opal system had become "the first quantum navigation system to achieve safety-of-flight qualification under the stringent RTCA DO-160 airworthiness standard," and described it as "an unjammable, unspoofable, undetectable quantum navigation system that works when GPS is unavailable," with deployments underway with Airbus on commercial aircraft and Lockheed Martin in defense. The performance figure in that announcement is a different one: positioning accuracy "better than 0.3 nautical miles through 95% of a flight."

Read those two numbers side by side

Twenty-two meters and 0.3 nautical miles are both real, and they are not in conflict, but a reader who takes away only the first one has been misled by their own attention. Twenty-two meters is stated in the paper as the best result achieved. Three-tenths of a nautical mile is about 550 meters, and it comes with the qualifier that actually matters for buying decisions: 95% of a flight. That second number is the one describing a system rather than a good day.

The bigger reframe is what the comparison is against. Every figure above is measured against an inertial navigation system, not against GPS. An INS is dead reckoning: it knows where it started and integrates its own motion, so its error grows without bound the longer it runs. MagNav's contribution is that it periodically re-anchors that drift to a fixed map. So the honest description is not "it replaces GPS." GPS, when it is working, puts you within meters anywhere on Earth. This puts you within a few hundred meters most of the time and, crucially, keeps you there rather than letting the error walk away over the next hour. It does not restore the satellites. It slows the bleeding, which in a jammed environment is the whole game.

One more piece of provenance, because this publication's rule is that I tell you what the source is. That trial paper is an arXiv preprint with a single version and no journal reference, filed under quantum physics, and Q-CTRL's chief executive and founder is among its authors. It is the company's own report of its own product. That does not make the numbers wrong, and posting the full method publicly is more than most vendors do. It does mean nobody independent has refereed them yet, and you should hold them accordingly.

The two dependencies nobody puts on a slide

Independent work points at what a brochure will not. The first constraint is the map, and more precisely the shape of the map. In a peer-reviewed paper in NAVIGATION: Journal of the Institute of Navigation in December 2025, Prasenjit Sengupta pointed out that the accuracy metrics civil aviation already uses may not transfer here, because those metrics assume "the uncertainty distribution in the horizontal plane is typically radially symmetric." Magnetic navigation, in his phrase, "challenges these assumptions." The reason is physical: the anomaly gradient has a direction. Where the field changes sharply as you cross it, the fix is strong. Where it is flat, or where you are flying along a feature instead of across it, there is much less information to match. Performance is better in regions with stronger gradients, and the error ellipse is not a circle.

That means a single accuracy figure describes MagNav far less completely than it describes GPS. GPS is roughly as good over Kansas as over Norway. This is not. Its performance is a property of the terrain you happen to be over, your heading across it, and whether anyone surveyed it well.

The second constraint is that your own vehicle is the loudest magnetic object in the experiment. The crustal signal you are trying to read is small, and it is buried under the field generated by the aircraft itself. A 2024 paper from a Stanford and SandboxAQ team states the problem plainly: "external magnetic fields induced by aircraft electronics and Earth's large-scale magnetic fields disrupt the weaker signal of interest." Their contribution is a better compensation method layered on the classical Tolles-Lawson model, and they report up to a 64% reduction in compensation error against conventional approaches. Take that as the tell. When the active research frontier is still how to subtract your own airframe, calibration is the bottleneck, and calibration is exactly the thing that degrades when you change the payload, swap the aircraft, or fly somewhere new.

To their credit, the trial team claims progress on precisely this: their system "learns model parameters online without special vehicle maneuvers providing robustness to various configuration changes (e.g. changing payload or latitude)." If that holds up under independent testing, it is a more important result than the 22 meters.

The signal

I once wrote that "unjammable" is a marketing word, because the fiber-optic drones being sold that way had only moved their vulnerability somewhere else. I want to be fair to this case, because it is genuinely different. A passive magnetometer reading a fixed geological pattern really cannot be jammed or spoofed by an adversary in the field, and it really does not announce itself. That is not a word game. That is physics, and it is the right direction to be building in.

But passive is a trade, not a free lunch, and the trade is always the same one. An active sensor carries its own source of information and pays for it by being visible. A passive sensor is invisible and pays for it by depending entirely on what the environment gives it. Here that dependency has a name and an owner: the anomaly map. Somebody has to survey it, keep it current, and hold it, and where the map is thin or the gradient is flat, the sensor is quietly worse and does not necessarily tell you so.

So when the pitch says unjammable, unspoofable, undetectable, believe all three, then ask the two questions that follow. Over what ground, and against what baseline? The gear here is real and the physics is honest. The number on the slide is still only a measurement taken under conditions, and the conditions are the product.