GNSS-INS for Subsea: When Tactical-Grade IMUs Earn Their Premium

The GNSS-INS Brief, And Why It Travels Subsea

Inside GNSS, together with Hexagon | NovAtel and Inertial Sense, ran a webinar on tightly integrated GNSS-INS for autonomous platforms. James Chan, business unit lead for INS in Hexagon’s Aerospace & Defence Division, gave the system-level perspective. Walt Johnson, founder and CTO of Inertial Sense, focused on low-SWaP-C tactical-grade MEMS implementations. The framing was familiar to anyone who works with positioning: GNSS remains the backbone, but it is vulnerable to multipath, spoofing and jamming, and it fails outright in urban canyons, tunnels and on indoor transitions. Inertial navigation is offered as the natural complement, propagating a continuous high-rate solution through those outages.

Chan walked through the building blocks. An IMU pairs accelerometers, which measure linear acceleration, with gyroscopes, which measure rotational rate, three axes each for six degrees of freedom. Many units now add a three-axis magnetometer for heading, taking them to nine degrees of freedom, and a barometer for altitude, taking them to ten. Magnetometers need calibration for local interference and magnetic declination. The point Chan returned to is the one that governs everything downstream: every IMU drifts, and those errors accumulate without an external correction. Drift rate depends on sensor stability, and that stability is what separates one grade of IMU from the next.

The integration architecture was described in standard terms. Nearly all inertial systems run a filter, typically an Extended Kalman Filter, with the INS solution propagating and GNSS updates folded in to compensate IMU errors. Between updates the inertial solution bridges the gap and keeps delivering position, velocity and attitude when GNSS is unavailable. None of this is new to our community, but the webinar laid out the cost-accuracy ladder cleanly, and that ladder is exactly the decision our subsea clients face every mobilisation.

Why The Grade Decision Has Money Attached

The webinar speaks to airborne and ground autonomy, but the engineering transfers directly to subsea work, with one substitution. Below the surface there is no GNSS to aid the filter. The inertial core is identical, the EKF is identical, the drift physics are identical, but the aiding sensors change to a Doppler velocity log for velocity and to USBL or LBL acoustics for position and, sometimes, heading. Everything the speakers said about drift, sensor stability and the price of accuracy carries straight into the ROV and AUV positioning brief.

That matters because grade selection is a recurring client decision with real money attached. The jump from an industrial MEMS to a tactical-grade MEMS, and again from tactical MEMS to a fibre-optic gyro (FOG) system, is a step change in cost, and often in size and power. Clients routinely ask whether the cheaper unit will do, and the honest answer is that it depends on the aiding environment and the survey specification, not on the IMU in isolation. The webinar’s central message, that integration quality and aiding determine the delivered solution, is the right starting point for that conversation.

What Changes Once The Aiding Goes Acoustic

The cost-accuracy ladder underwater turns on one parameter that surface GNSS users rarely think about: whether the IMU can gyrocompass. Gyrocompassing means resolving true north from the Earth’s rotation rate, roughly 15 degrees per hour. A FOG or ring-laser system with in-run gyro bias stability of hundredths of a degree per hour sees Earth rate clearly and finds north from a static alignment. A tactical-grade MEMS, with gyro bias of the order of single to low tens of degrees per hour, cannot reliably separate Earth rate from its own bias. It therefore needs an external heading source: a dual-antenna GNSS heading on the surface, a magnetometer, or a USBL-derived heading once submerged. This single distinction explains why most precise subsea survey spreads still carry FOG-grade INS.

The second reality is what bounds drift once you are deep. With continuous DVL bottom-lock, a calibrated bottom-locked Doppler velocity log typically holds velocity error to roughly 0.1 to 0.4 percent of speed over ground, so the dominant residual is heading error projected onto distance travelled. The relation is simple enough to put a number on: cross-track error is approximately the heading error in radians multiplied by the distance run, so a heading good to 0.1 degree (about 0.0017 radian) walks off by roughly 1.7 metres over one kilometre, while a 1 degree heading error gives about 17 metres over the same run. Get the heading wrong by a fraction of a degree and the along-track position walks off in proportion to how far the vehicle runs. A good DVL keeps an aided MEMS honest while the vehicle is slow and close to the seabed and the acoustics are talking. Lose bottom-lock over a soft or steeply sloping seabed, or in heavy suspended sediment, and the system reverts to water-track or pure coasting, where the cheaper IMU’s drift reasserts itself quickly.

Third is the integration topology. Loosely-coupled aiding feeds the EKF a finished position or velocity from GNSS, USBL or DVL. Tightly-coupled aiding feeds the filter raw observables: individual satellite pseudoranges on the surface, individual acoustic ranges or DVL beam velocities below it. The tightly-coupled benefit is that the filter can still use partial information, three acoustic returns instead of a full fix, or two DVL beams instead of four, when a loosely-coupled scheme would reject the epoch entirely. That resilience is real, but it is not free of conditions. It demands correct error modelling of each observable, accurate lever arms and clean time synchronisation, or the filter ingests biased measurements and produces a confident, wrong answer.

Five Recurring Specification Errors

1. Buying grade on the gyro bias number alone

The datasheet line that buyers fixate on is in-run gyro bias stability. It matters, but on its own it does not tell you whether the system meets your job. A tactical MEMS with an excellent bias figure still cannot gyrocompass, and if your operation has long stretches without USBL or magnetometer heading, that limitation, not the bias number, sets your error budget. Specify against the heading-determination method and the aiding availability, then read the bias figure in that context.

2. Treating tightly-coupled as automatically better

Tightly-coupled integration is the right architecture for sparse or intermittent aiding, but the marketing word does not survey the seabed. The delivered accuracy depends on how well the acoustic and Doppler observables are calibrated and modelled inside the filter. A loosely-coupled system with well-conditioned USBL and a properly calibrated DVL often beats a tightly-coupled system with sloppy lever arms and unmodelled latencies. Ask the vendor what observables are used, how their uncertainties are set, and how outliers are detected and weighted.

3. Assuming DVL bottom-lock is always there

The whole case for a cheaper IMU underwater rests on continuous velocity aiding. Bottom-lock is not guaranteed. It drops on altitude excursions, over very soft sediment, on steep slopes, and in turbid water near active intervention. Every dropout is a window of pure inertial coasting, and the cheaper IMU’s drift fills that window. Map your operation’s likely bottom-lock gaps before you choose grade, not after.

4. Ignoring lever arms and time synchronisation

The error that defeats more INS integrations than sensor grade is geometry and timing. The IMU, the DVL transducer, the USBL transceiver and the vehicle reference point sit at different offsets and must be tied together to the centimetre, and the measurements must share a common, latency-corrected timebase. A few centimetres of lever-arm error or tens of milliseconds of unaccounted latency at survey speed produces a bias the filter cannot distinguish from a real position. No grade of IMU recovers an installation that has not been measured and synchronised properly.

5. Writing the spec to surface GNSS accuracy

Clients sometimes carry a surface positioning expectation into a subsea scope, then are surprised by the acoustic error budget. USBL slant-range and bearing uncertainty grows with depth and with sound-velocity error through the water column. A tight INS does not erase that; it smooths and bridges it. Build the specification from the subsea error sources up, with the inertial system as the tool that bounds drift between acoustic fixes, not as a substitute for a sound positioning geometry.

Matching IMU Grade to the Survey Specification

The decision becomes tractable once you start from the deliverable rather than the sensor. IHO S-44 (Edition 6) is the yardstick for hydrographic deliverables, with its total horizontal and total vertical uncertainty limits at 95 percent confidence by order, from the tighter Exclusive and Special Orders down through Order 1a, Order 1b and Order 2. It is the wrong yardstick for everything else: pipeline and structure as-built work, cable route survey and subsea metrology are not S-44-governed, and their tolerances come from the project specification and from IOGP and IMCA survey guidance, not from an S-44 order. Do not carry an S-44 order into a metrology scope; metrology is judged against a stated dimensional tolerance, often at the centimetre level on a relative measurement, which is a different acceptance test entirely. Fix the order or the tolerance first; the IMU grade then follows from the aiding environment that the operation actually provides.

It is worth connecting a heading grade to the uncertainty budget at least once, because that link is what turns a datasheet into a specification. Take a vehicle running 500 metres of line between USBL fixes. A FOG-grade system holding heading to about 0.05 degree contributes roughly 0.4 metre of cross-track position error over that run (0.05 degree is about 0.00087 radian, times 500 metres); a tactical-grade spread relying on a magnetometer heading good to perhaps 0.5 degree contributes about 4.4 metres over the same distance. Carried into the horizontal budget in quadrature with the acoustic and DVL terms, 0.4 metre sits comfortably inside an S-44 Special Order total horizontal uncertainty of 2 metres at 95 percent, whereas 4.4 metres breaches it on the heading term alone. The same arithmetic, run for your line lengths and aiding interval, tells you directly whether a grade closes the budget before you ever mobilise.

Where the vehicle stays slow, near the seabed, with reliable DVL bottom-lock and regular USBL updates, a well-integrated tactical-grade MEMS can carry routine ROV positioning and inspection inside the looser S-44 orders. The aiding does the heavy lifting and the IMU’s job is to interpolate smoothly between fixes. This is precisely the low-SWaP-C case the webinar made for, and on small electric ROVs and observation-class vehicles it is increasingly the sensible economic choice.

Where the operation runs long inertial coasts, an AUV transit between acoustic networks, a corridor with frequent bottom-lock loss, or a precise metrology where heading must be trusted without continuous external aid, the FOG-grade system earns its premium. Its ability to gyrocompass and to hold heading through extended acoustic gaps is the difference between meeting Special Order and missing it. The cost step is justified by the error budget, not by preference.

Recommendations

• Derive the IMU grade from the spec, not the other way round. Fix the S-44 order and the horizontal and vertical uncertainty budget, characterise expected DVL bottom-lock and acoustic-update availability, then choose the grade that closes that budget with margin.
• Test the coasting case before mobilisation. During the dockside or in-water trial, force a USBL dropout of a defined duration with DVL maintained, then a combined DVL-and-acoustic outage, and measure the position drift on reacquisition. That number, not a datasheet, tells you what the system delivers on your job.
• Verify heading determination explicitly. If the unit cannot gyrocompass, confirm the external heading source and its availability across the full operation, and record the heading accuracy that source provides at working depth.
• Measure lever arms to the centimetre and prove time synchronisation. Tie IMU, DVL, USBL and reference point into one frame, drive timing from a hardware PPS or equivalent hardware timestamping, and quantify residual latency. Treat this as an acceptance item, not a formality.
• Run a figure-of-eight or equivalent dynamic calibration and a static alignment as part of mobilisation, following IMCA and IOGP survey practice, and log the alignment quality before accepting the spread.
• Interrogate the integration, not just the sensor. Ask whether aiding is tightly or loosely coupled, which observables feed the filter, how observation uncertainties are configured, and how the filter detects and rejects acoustic and DVL outliers. The answers separate a positioning system from a box of good sensors.

The webinar’s underlying argument holds offshore as firmly as it does in the air: the delivered navigation solution is set by the integration and the aiding, with the IMU grade as one term in the budget rather than the whole story. For subsea clients weighing a tactical MEMS against an FOG system, that is the right frame. Buy the grade your error budget needs, prove it with a coasting test, and put the saved cost into geometry, calibration and synchronisation, where it returns more accuracy per unit of spend than the next rung up the IMU ladder.

---

Based on: Integrating GNSS and Inertial: Tactical-Grade Performance for Modern Autonomous Applications