Contested airspace
Jamming is routine over active fronts and increasingly common around airports and ports. Aircraft are cheap and numerous; the navigation hardware can't be.
Satellite fix available12%
Jammed or spoofed for most of the sortie
Positioning · Navigation · Timing
Hold positionwhen thesatellitesdon't.
Deadwater is a navigation module for low-cost drones. It learns the error in the MEMS sensors already on board, so a £20 chip holds position through jamming that used to need a £4,000 gyroscope.
Schematic track · error growth to scale in the readout below
ElapsedT+0.0 s
Satellite fixJammed
Raw MEMS error0.00 m
Deadwater error0.00 m
SensorTDK ICM-42688-P · £19
The failure mode
A cheap sensor loses the aircraft in seconds.
Every drone carries an inertial measurement unit — the same accelerometer and gyroscope pair that tells your phone which way up it is. When the satellites go quiet, that chip is all that’s left.
Its bias error doesn’t stay still. Accelerometer bias integrates twice into position, which grows with t². Gyroscope bias tilts the estimated attitude, leaking gravity into the horizontal channel and growing with t³. That cubic term is the one that ends missions.
The industry answer is to buy the error down with hardware: fibre-optic and ring-laser gyroscopes at thousands of pounds a unit, on airframes that are meant to be expendable. We go the other way.
| Since jam | Raw MEMS£20 part, uncorrected | Deadwaterthe same £20 part | Tactical grade£4,000 part |
|---|---|---|---|
| 5 s | 0.00 m | 0.00 m | 0.00 m |
| 15 s | 0.00 m | 0.00 m | 0.00 m |
| 30 s | 0.00 m | 0.00 m | 0.00 m |
| 1 min | 0.00 m | 0.00 m | 0.00 m |
| 5 min | 0.00 m | 0.00 m | 0.00 m |
Modelled free-inertial horizontal error, no aiding, no satellite fix. Figures are representative of a consumer-grade MEMS IMU rather than measured on a specific airframe. Bench results are available under NDA.
Five minutes into a jam, corrected output on a £19 chip lands within about twice the error of a unit costing two hundred times as much.
How it works
Learn the error, not the signal.
Deadwater doesn't replace your navigation filter or your hardware. It sits one step upstream and cleans the measurements going in.
The result is a change in the shape of the error curve, not just its size. Correcting the dominant bias terms pulls the aircraft’s estimate back from a cubic climb into something a mission can absorb.
Integration
Runs on the flight controller you already ship.
No new silicon, no supply chain change, no second sourcing exercise. If your airframe has an IMU and an EKF, it has everything Deadwater needs.
We sell to the people building the aircraft, not to the people flying it. Integration is a library and a calibration profile for your part number — typically a few weeks from first call to a flying build.
We price against the hardware you no longer have to fit and the airframes you stop losing, not against the chip we run on. A sensor costs £19. A sortie that ends in the wrong field costs the aircraft, the payload and the mission.
- Targets
- ARM Cortex-M4F and above, Cortex-A, NVIDIA Jetson
- Footprint
- 180 kB flash, 48 kB RAM
- Update rate
- 400 Hz, under 1.4 ms per step on a Cortex-M7
- Interfaces
- MAVLink, ROS 2, plain C API
- Sensors
- Consumer and industrial MEMS IMUs, calibrated per part number
- Network
- None. Runs entirely on the aircraft, no telemetry off-board
- Delivery
- Static library with source escrow, plus an engineer for the first airframe
- Licensing
- Per-airframe royalty or fixed programme licence
Target figures for the reference build. Confirm against your own board before designing them into a programme.
Why now
Three things changed at once.
This problem has existed for as long as inertial sensors have. What's new is that it now lands on aircraft that can't afford the traditional answer.
Jamming
It stopped being an exception.
GNSS denial is a permanent condition over active fronts, and interference is spreading into civil airspace around airports and ports. Aircraft now have to be designed for it rather than around it.
Volume
The aircraft became disposable.
Drone production runs to millions of airframes a year, built in the expectation of losing them. Nobody fits a £4,000 gyroscope to something on a one-way mission.
Silicon
The growth is at the cheap end.
MEMS is the fastest-growing segment of inertial navigation, and consumer grade is outgrowing navigation grade. The volume is arriving exactly where the hardware is weakest.
Where it runs
Anywhere the sky is gone.
The same problem shows up wherever satellites can't reach. Defence is the sharpest version of it, not the only one — which is why this is a dual-use business.
Underground mining
Survey and haulage drones work hundreds of metres below the last fix, in galleries where nothing gets through the rock.
Satellite fix available0%
No fix below the portal
Tunnels and utilities
Inspection runs lose lock at the entrance and don't get it back until they come out the other end, if there is another end.
Satellite fix available3%
Fix at the portals only
Warehouse and indoor logistics
Steel racking and roof structure kill reception well before the aircraft is inside the building.
Satellite fix available15%
Intermittent, unusable for control
Why us
The same problem, pointed at an airframe.
Getting a trustworthy signal out of a sensor that lies is a narrow, specific skill. It is not drone engineering, and it is not aerospace. It is the one thing this problem actually turns on.
We have spent years building machine learning that pulls reliable movement data out of low-cost MEMS sensors, in human movement rather than flight. Same class of part, same noise floor, same drift, and the same discipline: learn the error of the specific sensor in front of you instead of specifying a better one.
Deadwater points that at an aircraft instead of a body. The physics around the sensor changes. The work of squeezing signal out of a cheap chip does not.
It is why we think the answer here is software rather than better hardware — and why we were most of the way there before we started.
Markets
The money is moving toward the cheap end.
Inertial navigation is an old, large market built on expensive hardware. The part of it growing fastest is the part that can't afford expensive hardware.
$9.4bn – $0.0bnInertial navigation, 2026
The spread is real. Different houses size this differently.
5 – 0%Growth of the whole market
Heading for the high teens to low forties of billions by the mid-2030s.
0.0%Growth of MEMS
The cheap chip type, and the fastest-growing technology segment.
0.0%Growth of consumer grade
Outpacing navigation grade, and the reason is the drone build-out.
Market sizes and growth rates from published industry research, including MarketsandMarkets. Ranges reflect genuine disagreement between sources.
Our slice, built from the bottom up.
Most of the headline number is missiles, crewed aircraft and ships running hardware we don’t compete with. What we can actually sell is narrower: navigation software licensed to the people building low-cost drones. So here is the whole calculation.
This is our model, not a report’s — nobody sizes this sub-slice yet, because it barely exists. Every line of it is an assumption and every one is arguable. That’s the point: argue with the inputs and the output moves. It’s a more useful conversation than a number nobody can defend.
Low-cost airframes built a yearDefence, industrial and inspection platforms. Excludes consumer toys.8 – 12 M
Flying where GNSS is deniedContested airspace, underground, indoors.~1 in 3
Makers who license rather than buildThe rest roll their own, or ship without and accept the drift.~1 in 3
Navigation licence per airframeDefence and industrial platforms, not hobby builds.$150 – 400
Serviceable market, today$130m – $530m
Growing faster than the market around it. The ceiling is set by one thing — how many drone makers license a navigation module rather than building their own.
Get in touch
Tell us what you're flying.
Evaluation builds go out to drone manufacturers and integrators under NDA. The first call is technical, not commercial.
- We reply within two working days.
- First call covers your IMU part number, flight controller and mission profile.
- Evaluation build and calibration profile follow, under NDA.
Or write to contact@deadwater.tech