DVL vs INS vs USBL: What an AUV Mission Actually Needs

Three references, three failure modes, and the error arithmetic that sizes an AUV navigation suite.

Most AUV navigation failures are not sensor failures. They are reference failures: a vehicle is navigating confidently against the wrong quantity, and nobody notices until the survey is processed. DVL, INS and USBL are usually discussed as competing options, yet they measure three different things, in three different frames, and they fail in three different ways.

This article puts the three side by side with published numbers from our own datasheets, derives the error arithmetic that actually sizes a suite, and shows where each sensor stops earning its power budget.

Three sensors, three different references

A DVL measures velocity. Bottom tracking references the seabed, so the velocity is over ground and independent of current. Above its maximum altitude it falls back to water-mass tracking, and then the velocity is relative to moving water, which is a different quantity with a different error.

An INS measures attitude and integrates it. Gyroscopes give heading, accelerometers give specific force, and the two are combined into position by dead reckoning. Accuracy therefore depends on time and distance, not on a single ping: a good INS that is left unaided drifts, and the drift is dominated by heading error.

A USBL or LBL measures position. It resolves range and bearing from a transceiver array to a transponder, giving an absolute fix in the array's frame. It is accurate at the moment of the fix and does not drift, but it updates slowly and it needs acoustic line of sight.

That gives the useful division of labour: DVL supplies a smooth, high-rate velocity that keeps dead reckoning short-term accurate; INS supplies attitude, which is what turns velocity into a trajectory; USBL or LBL supplies the absolute anchor that stops the trajectory from wandering. Any two of the three cover some missions. All three cover the ones where the answer has to be defensible after the vehicle is recovered.

What a DVL contributes, and its altitude envelope

The single number that decides whether a DVL is useful on a given mission is maximum bottom-track altitude, and it is a frequency trade, not a quality grade. Our published range covers seven models:

ModelFrequencyMax altitudeMin altitudeBottom-track accuracy
DVL-600K Phased (Tingseek)600 kHz110 m1 m±0.2% ±1 mm/s
DVL-300K Phased (Tingseek)300 kHz200 m—±0.3% ±1 mm/s
DVL-150K Phased (Tingseek)150 kHz500 m—0.3–0.5% ±2 mm/s
DVL-600K Piston (Tingseek)600 kHz100 m0.5 m±0.3% ±0.3 cm/s
DVL-300K Piston (Tingseek)300 kHz250 m1 m±0.5% ±0.5 cm/s
VAM-600K (Tingseek)600 kHz100 m0.5 m±0.3% ±0.3 cm/s
DVL-Hover-H1000 (Tingseek)1000 kHz45 m0.05 m±0.4% ±5 mm/s

DVL-600K phased-array Doppler velocity log

DVL-600K Phased: four-beam phased array, 2.3° beam width, 30° nominal angle, 110 m maximum bottom-track altitude and a 1 mm/s velocity resolution.

Read the table as a ladder of missions rather than a ranking. A 150 kHz unit reaches 500 m of altitude, which is what a deep-water transit needs, but its accuracy is quoted at 0.3–0.5% ±2 mm/s. A 600 kHz phased array is twice as accurate at ±0.2% ±1 mm/s but only sees bottom out to 110 m. The H1000 exists for a different problem again: 0.05 m minimum altitude and a blind zone under 5 cm, for a vehicle hovering close enough to inspect a weld.

Two details in the fine print matter more than buyers expect. First, the 600K Phased unit's 4-beam phased array with 2.3° beam width gives a narrower footprint than a conventional piston array, which matters over sloping or patchy seabed where wide beams average incompatible ranges. Second, its automatic water-layer mode is not a spare feature: when bottom lock is lost, the DVL switches to a 10 m fixed water layer and keeps producing a velocity, which is exactly the moment a naive navigation filter will accept a wrong reference.

What an INS contributes, and why heading dominates

An INS is judged by three numbers: heading accuracy, attitude accuracy and position error as a percentage of distance travelled. Our integrated units publish 0.25° heading RMSE with 0.03° pitch and roll on the NAVS-3000, and ≤0.15°secL heading on the PRO variant of the NAVS-P600K, with ≤0.5% of distance in a straight line and ≤0.3% typical.

The secL in those figures is secant latitude, and it is the most misread symbol in navigation datasheets. Heading error grows with latitude: secant of 30° is 1.155, of 45° is 1.414, of 60° is 2. So a 2°secL heading specification is 2.31° at 30° latitude and 2.83° at 45°, and a buyer who tests at 30° and deploys at 60° will see twice the heading error the datasheet promised.

Heading error then converts into cross-track position error by simple geometry: cross-track equals distance times the sine of the heading error. At 2.31°, that is 40 m per kilometre and 202 m over 5 km. At 0.173° — the PRO figure at 30° latitude — it is 3 m per kilometre and 15 m over 5 km. That single line explains why a mission that tolerates 10 m of error cannot run on a pure-inertial solution regardless of how good the accelerometers are, and why a two-degree heading specification and a 0.15-degree one are not comparable products.

Alignment also belongs in the budget. Our units align in ≤5 min static and ≤15 min dynamic. A vehicle that leaves the ship before alignment converges, or aligns in a seaway, starts the mission with an unresolved heading offset that no amount of DVL aiding will remove.

What USBL and LBL contribute, and what they cost

USBL gives an absolute fix from a single surface array. Our P300 publishes a 300 m positioning and communication range at 30–40 kHz, ranging accuracy of 1%D ±0.3 m, direction finding of ±1.5°, a 4 s position refresh, and tracking for up to five targets. LBL trades coverage for accuracy: 3 m positioning over 2,000 m with medium-frequency transponders, 5 m over 8,000 m with low frequency.

The 4 s refresh is the number that decides how you use it. A DVL at 200 Hz produces 800 updates in the interval between two USBL fixes. So USBL cannot smooth a trajectory; it can only anchor it. When the fix arrives, it typically differs from the dead-reckoned position, and the difference has to be blended in over several seconds rather than applied as a step, or the vehicle's controller sees a position jump it will try to correct.

That also reframes the accuracy comparison. At 300 m slant range, a 1%D ±0.3 m USBL fix is 3.3 m. Over the same 4 s, an INS/DVL solution at 2 m/s and 0.5% of distance drifts about 4 cm. USBL is therefore worse than dead reckoning in the short term and better in the long term, which is exactly why the two belong in one filter.

The error budget, in one table

Every navigation architecture question reduces to how much error you will accept at the end of a leg. The arithmetic below uses published heading and position figures, converted with the formulas in the previous sections:

SuiteAt 1 kmAt 5 kmWhat dominates
INS only, 2°secL at 30° lat40 m202 mheading (2.31°)
INS only, 0.15°secL at 30° lat3.0 m15 mheading (0.17°)
INS + DVL, ≤0.5% of distance5 m25 mDVL scale factor and heading
INS + DVL, Z-pattern ≤0.1%1 m5 mfilter tuning, not sensor
INS + DVL + USBL resetsbounded by fixbounded by fixUSBL absolute accuracy

The Z-pattern row is worth pausing on. The same hardware publishes ≤0.5% of distance for a straight line and ≤0.1% for a Z-shaped survey pattern, because a line that crosses its own track lets the filter observe and correct its scale factor. A three-line difference in specification is a survey-design decision, not a product difference.

Four mission profiles, four suites

Near-bottom survey AUV, 30–80 m altitude. INS plus a 600 kHz phased DVL covers it: bottom stays within 110 m, accuracy is ±0.2% ±1 mm/s, and no acoustic positioning is needed. Add USBL only if the survey must be georeferenced to an absolute frame.

Long transit at 150–500 m altitude. The 300 kHz and 150 kHz phased units extend bottom tracking to 200 m and 500 m. Past 500 m, or over soft mud where bottom lock fails, the suite becomes INS plus sparse USBL or LBL fixes, and the mission has to be planned with those fixes in mind.

ROV station keeping at 0.2–1 m altitude. The envelope inverts: the constraint is the blind zone, so the H1000 with 0.05 m minimum altitude and 45 m maximum is the right sensor, and the 150 kHz unit is unusable no matter how good its accuracy.

Long-duration AUV with infrequent surfacing. Power and volume decide. A compact INS/DVL unit at under 2 kg and 300 m depth rating, with GNSS on the surface and USBL when available, gives a self-contained solution that does not need a support vessel holding station.

NAVS-3000 integrated underwater INS and depth sensor

NAVS-3000: 0.25° heading RMSE, 0.03° pitch and roll, ≤0.5% of distance straight-line and ≤0.1% on a Z-pattern, with a 3,000 m depth rating.

Five questions that size the suite

Run these before comparing product names, because each one eliminates entire families:

  1. What is the operating altitude, and does the bottom stay inside it? Compare against the maximum bottom-track altitude of the candidate DVL, not its accuracy headline.
  2. How far does the mission travel between absolute fixes? Convert that distance into error using the table above, at the latitude actually planned.
  3. Is the final product relative or absolute? Photogrammetry and inspection can live in a local frame; charting and change detection cannot.
  4. What does the seabed do to bottom lock? Soft mud, dense weed and steep slopes all reduce usable altitude, and the mitigation is a different sensor, not a setting.
  5. What are the power, mass and volume budgets? A 14 kg INS with a 3,000 m housing and a 2 kg integrated unit solve the same arithmetic problem at very different vehicle cost.

Asked in that order, the answers usually converge on one of three suites rather than on a component. That convergence is the point: the DVL series is a family of altitude envelopes, the integrated navigation systems are pre-fused INS/DVL packages, and the USBL P300 with its LBL transponder array covers the absolute end of the same problem.

FAQ

Do I need USBL if the vehicle already has a DVL and INS?

Only if the survey has to be tied to an absolute frame, or if the mission runs long enough for accumulated error to exceed the tolerance. A near-bottom survey inside the DVL's altitude envelope, with a well-aligned INS, often needs no acoustic positioning at all. A transit survey that must be compared with last year's data does.

What happens when the DVL loses bottom lock?

Two things change at once, and both are dangerous if unflagged: the reference stops being the seabed, and the filter keeps receiving plausible velocity. Above maximum altitude the unit switches to water-layer tracking, which reports velocity relative to moving water; over soft mud or dense weed it may drop out entirely, leaving pure inertial navigation with the drift shown in the table. Log the lock status and treat any period without bottom lock as a different error regime.

Is the DVL or the INS the primary sensor?

Neither, and that framing causes most of the bad specifications written into tenders. The INS supplies attitude, which converts velocity into a trajectory; the DVL supplies velocity, which keeps that trajectory honest over the mission duration. Specifying one without the other's accuracy, alignment time and aiding scheme leaves the system-level error undefined.

What does an integrated unit actually save?

Mass, volume, cabling and the interface work of fusing two sensors that were designed separately. It does not remove the need to plan aiding: an integrated INS/DVL package still needs a heading reference at the start, GNSS on the surface and, for absolute work, an acoustic fix. The datasheet figure of ≤0.5% of distance assumes bottom lock is maintained.

DVL vs INS vs USBL: What an AUV Mission Actually Needs
China Sonar 2026年11月16日
存档