Magnetic Quantum Navigation - Alternate to GPS Air navigation

A regional jet left Everett, north of Seattle, turned toward the Gulf of Alaska, and flew for four hours and twenty-three minutes with GPS switched off. It came back knowing where it was.
The Defense Innovation Unit announced the result on 4 September 2026 and led with a headline number: 89% better position accuracy than the traditional backups. The write-up is in the DIU news story, with aviation coverage in Aviation International Newsand DefenseScoop.
That number is real, and it is the least interesting thing about the flight. Here is the interesting thing. Every backup navigation system in service today gets worse the longer you use it.

Every backup we have gets worse by the minute

GPS works like a choir of clocks in space. The receiver listens to several satellites at once, compares tiny differences in arrival time, and works out where it is. Good to a few metres on a clear day.

The signals arrive faint. Anyone with modest equipment can drown them in noise, which is jamming, or feed the receiver a convincing lie, which is spoofing. IATA counted a 220% rise in GPS signal-loss events between 2021 and 2024, most of it around Eastern Europe and the Middle East, and said it did not expect the trend to reverse.

When GPS drops, the aircraft falls back on inertial navigation: accelerometers and gyroscopes tracking motion from the last known point. It works for a while. Then small errors compound, and the compounding never stops. Ten minutes is fine. Four hours over open water is a different situation.

Magnetic navigation attacks the compounding directly. It caps the drift down. Every time the system recognises the ground beneath it, the error resets to whatever the map is worth. Fly for one hour or fourteen, and the error at the end looks about the same. The industry term is bounded positioning.

The planet wrote the map a long time ago

Earth is a magnet. That is why a compass points north. The field is not smooth.

Iron-rich rock and old lava flows distort it in ways that change from place to place. Slightly stronger here. Slightly weaker a few kilometres on. Those distortions are magnetic anomalies, and the pattern they make over any stretch of ground is close to unique.

Geologists have been mapping them since the 1940s, when survey aircraft went hunting for ore bodies. The smooth global picture lives in products like NOAA's World Magnetic Model. The wrinkled version, the one that matters here, lives in crustal maps such as the Earth Magnetic Anomaly Grid. A compass reads the smooth field. Magnetic navigation reads the wrinkles sitting on top of it.

A single reading tells you nothing

The sensor returns one number: how strong the field is right here. That number is not a location. Thousands of places on Earth produce the same reading. An aircraft sitting still, taking one measurement, would have no idea where it was.

Movement is what makes it work. Over ten minutes the aircraft traces a run of readings. Up, down, up, a sharp dip, flat. That shape is far rarer than any single value. The software slides the shape across the stored map until it finds the one stretch of crust that could have produced it.

So the inertial system is not a helpful addition at the end of the process. It supplies the shape of the path. Without it there is nothing to match. The two systems are one system, and each covers the other's defect: inertial gives geometry that drifts, magnetic gives an anchor that doesn't.

The quantum part is the thermometer, not the brain

Quantum here describes the sensor. A quantum magnetometer measures field strength by watching how atoms behave, usually a vapour of caesium or rubidium, sometimes defects in a diamond crystal called nitrogen-vacancy centres. These instruments are not new. Optically pumped magnetometers have flown on geological survey aircraft for sixty years and hunted submarines for longer. What changed is size, ruggedness, sampling speed, and price.

Most of them are scalar. They measure total field strength and ignore direction. That sounds like a limitation and it is an advantage: a direction-sensitive sensor inherits every error in the aircraft's own sense of which way it is pointing, and a scalar sensor does not care how the aeroplane is oriented. The DIU programme that funded this work makes the same argument in plainer terms.

The computer doing the matching is ordinary. Clever software, standard hardware.

The aeroplane is louder than the thing you are listening for

Earth's total field runs around 50,000 nanotesla. The crustal anomalies worth navigating by are tens to a few hundred. The aircraft's own magnetic mess, meaning current in the wiring, motors, control surfaces moving, and eddy currents in the skin, is comfortably the same size or larger.

The hard engineering sits in the subtraction. The classic method is Tolles-Lawson compensation, worked out roughly eighty years ago for survey aircraft, which models the aeroplane's magnetic personality as a set of coefficients and removes it. Modern systems do better with machine learning and factor-graph estimation, and the best of them cold-start: they work out the aircraft's signature during the flight instead of demanding a dedicated calibration sortie first. Where you bolt the sensor matters too. Q-CTRL's strongest published result came from a magnetometer hung outside the wing; its harder results came from sensors buried inside the cabin, where the noise lives.

Open ocean is the easy case

Every report on the Pacific flight describes open water as the punishing test. For radar mapping and for a pilot's eyes, it is. There is nothing down there.

For magnetic navigation the Pacific was closer to a home fixture. The seafloor is striped. New crust forms at spreading ridges and locks in the direction of Earth's field at the moment it cools. The field reverses every few hundred thousand years. So the ocean floor holds a record of those reversals as long parallel bands of alternating magnetism — the same stripes that proved plate tectonics in the 1960s. Strong, regular, well surveyed, and running straight across the route that jet flew.

The sensor will commoditize. The map will not.

The public global product, EMAG2, is assembled from surveys whose resolution ranges from about one kilometre to twenty, and coverage is uneven by geography. Worse, those surveys were flown to find minerals rather than to position aircraft. Line spacing, altitude, and processing were all chosen to make geology legible to a geologist. What helps interpret an ore body is not what a navigator needs, a point made carefully in recent work on geophysical data requirements for MagNav.

The good maps, the dense low-altitude current ones, are national assets, classified, or owned by mining companies. You cannot fly this over ground you do not have. Magnetic navigation is therefore a geopolitical capability as much as an engineering one, and the hard bargaining in this field will be over survey data long before it is over sensors.

There is also weather of a different kind. The field varies through the day and jumps during solar storms, by amounts that can swamp the anomaly you are chasing. Standard practice corrects for this using a ground magnetometer station within 100 km of the aircraft. Over the middle of an ocean, there isn't one, which is an open research problem rather than a solved one.

Read the multiples with suspicion

Eighty-nine per cent better than traditional backups. Eleven times better than inertial. Forty-six times. Fifty. Ninety-four. One hundred and eleven.

Those are all published figures from real flights, and none of them is comparable to another. Different aircraft, different maps, different sensor mounting, different baselines, different definitions of error. A multiple is only as meaningful as whatever sits in the denominator.

Absolute numbers are more useful. Q-CTRL's Ironstone Opal, flying a Cessna 208B around Griffith in Australia, logged a best final error of 22 metres after a 365 km leg, about 0.006% of distance flown; that campaign is written up on arXiv and in less technical form by New Scientist. A separate flight in the same campaign finished 112 metres out after 420 km, and that is the one that produced the 46× headline. SandboxAQ's AQNav reports beating unaided inertial on every flight over two hours, with a best observed accuracy under 74 metres.

Set against clear-sky GPS at a few metres, none of that competes. It is not supposed to.

Unjammable is not the same as untrickable

DIU says the magnetic field cannot be spoofed or blocked. Vendors repeat it. Half of that is solid: the system transmits nothing, so there is nothing to jam and nothing for an adversary to hear. Weather and darkness are irrelevant, because it is not looking at anything.

The other half needs care. The attack surface moved rather than disappeared. It now sits in the map, which is a stored file, distributed through a supply chain, and capable of being corrupted or quietly wrong. Repetitive magnetic terrain can also produce a confident match in the wrong place, which is worse than no match at all.

Which points at what this technology is really for. The dangerous GPS attack is not the blackout — that one announces itself. It is the slow spoof that walks your position sideways while the cockpit shows green. An independent reading taken from the planet itself catches that within minutes. Detection, not replacement.

Accuracy is the easy exam

Q-CTRL's system passed RTCA DO-160 environmental qualification in July 2026, covering vibration, interference, and the general physical abuse of flight. A first for quantum navigation. It is also not certification, and not operational approval.

Civil aviation asks a harder question than how accurate a system is. It asks whether the system knows when it is wrong. GPS answers with integrity monitoring: a computed bound the receiver guarantees it sits inside. A navigator that is usually excellent and occasionally confidently wrong fails that exam no matter how good the average looks. Building integrity bounds for map matching, where confidence depends on how distinctive the crust below happens to be that minute, is the open problem. Certification as a sole means of navigation sits behind it.

One bet among several

Honeywell and DIU put the same class of system on a C-17 Globemaster III later this year. DARPA has $24.4 million with Q-CTRL. DIU has Lockheed Martin and Q-CTRL working on a quantum inertial system, which attacks drift from the opposite end. Q-CTRL ran gravity-based navigation on an Australian Navy vessel for 144 hours straight. Same instinct, different planetary signature.

Magnetic navigation is one entry in a wider field. Automated star trackers. Timing borrowed from low-Earth-orbit constellations. Terrain-referenced vision. None of them beats GPS on a calm day. All of them are hard to switch off. The likely answer is several weak, unjammable signals fused together rather than one clean replacement.

The compass was the first instrument that read the planet for position. This is the same instinct with a finer map and a quieter sensor. The crust has been broadcasting the whole time. We only started listening properly once someone made a serious effort to switch the satellites off.