Skip to content
SeaNergy · Environment-aware acoustic payload · Autonomous underwater vehicles
SeaNergy Adaptive Sonar
Environment-aware acoustic payload · Autonomous underwater vehicles

It's loud, layered,and never the same twice.

Not that the pulse noticed.

We're building sonar that reads the water first.

Run the console System overview
Adaptive Sonar

fixed-frequency sonar goes blind in water that moves

Sediment, temperature, salinity and depth all move the acoustic channel — and they move it during a survey, not between them. A pulse chosen on the deck is already the wrong pulse by the time it reaches the seabed.

solved on the payload itself

No cloud. No pairing. No network call. The optimiser, the physics and the DSP all run on the device, which is the only place they can run fast enough to matter.

How the payload adapts a sonar pulse to the water it is in

Medium state measured on vehicle
Temp
—
°C
TDS
—
g/L
Turbidity
—
NTU
Depth
—
m
Sound speed
—
m/s
— TDS read as a salinity proxy, not a laboratory salinity
Acoustic link · transmit calculated
Mode
—
fC
—
kHz
B
—
kHz
τ
—
ms
Window
—
Resolution
—
mm
Echo SNR two-way
—
dB
—
Acoustic propagation · transmission loss calculated
Range
—
m
Spreading
—
dB
Absorption
—
dB
Scattering
—
dB
One-way TL
—
dB
Echo SNR two-way
—
dB
— Absorption from Thorp; scattering from the payload model
Signal domain simulated
TX
RX
SPEC
SNR before compression
—
dB
Processing gain
—
dB
Traces drawn to their own peak. The magnitude is the number, not the height.
Waveform solver · objective calculated
Sound speed
—
m/s
α at band centre
—
dB/km
Excess scattering
—
dB/km
Margin required
—
dB
Selected fC
—
kHz
— Shaded band clears the detection threshold with margin
Same water · fixed against adaptive calculated
Fixed — solved for the clear water it started in Adaptive — solved for the water it is in
Matched filter · pulse compression simulated
FIXED
ADAPT
PSLR
—
dB
PSLR
—
dB
Detection
—
Detection
—
Correlation against the transmitted replica, in dB. A pulse that is not there does not correlate.
Environmental sensing on-vehicle
  1. DS18B20temperature
  2. TDS probeconductivity
  3. Opticalturbidity
  4. Pressuredepth
  5. Estimatec, α, excess
Transmit chain architecture
  1. DDSphase accum.
  2. DAC12-bit
  3. Recon.filter
  4. Driveranalog
  5. Transducerpressure
01 Medium

The medium is not constant.

Temperature, dissolved solids, suspended load and depth all decide how far sound carries. None of them hold still along a survey line.

02 Platform

One payload, one ping interval.

The payload measures the water, chooses a pulse, transmits it and compresses the return — all of it between one ping and the next, on the vehicle, with no link to the surface.

03 Transmit

A pulse leaves the aperture.

A swept pulse, chosen for water this clear: the widest sweep the transducer will pass, because bandwidth is what buys range resolution.

04 Transition

The water changes.

Warmer, more conductive, and carrying sediment. The sound speed shifts, and the suspended load starts taking energy out of the pulse.

05 Propagation

The loss follows it down.

Spreading is geometry and never changes. Absorption is the water itself. The third term — scattering off suspended matter — is the one that grows, and it grows fastest at the top of the band.

06 Mismatch

The water changed. The waveform did not.

The pulse that was correct an hour ago is still being transmitted. The echo it earns no longer clears the detection threshold at the range the mission asked for.

07 Sense

Measure the medium.

Temperature, conductivity as a salinity proxy, optical turbidity, and depth. Four numbers, read on the vehicle, in the water the pulse is about to cross.

08 Solve

Solve for the pulse.

Every candidate waveform is scored against the sonar equation for the water just measured. The widest sweep that still clears the threshold with margin wins. Nothing here is a lookup table.

09 Synthesise

Build it on the way out.

The chosen parameters go straight into the transmit chain — phase accumulator, converter, reconstruction, driver, transducer — and become a pressure wave.

10 Compare

The same water. A different pulse.

Nothing has been done to the sediment. The range is the range the mission wanted. The only thing that changed is what was transmitted into it.

11 Recover

The medium changes. The sonar adapts.

Matched filtering collapses the returning sweep into a single peak. What arrives buried under the noise leaves the filter as an arrival time.

One payload. Any water. Mission-ready.

The same hardware works a clear reef and a muddy estuary, because the waveform is a decision rather than a setting.

351 kHz
Top centre frequency
drops to 148 kHz as sediment loads
Modelled
470
Candidate waveforms scored
every tick, against the sonar equation
Measured
23.81 dB
Compression gain
time–bandwidth product 240.4
Calculated
13
Stages per ping interval
sample, solve, transmit, correlate
Measured
0
Network calls
no server, no pairing, no cloud
Measured
Engineering proof · 04

Built to be questioned.

The system does not ask you to trust the number. It shows you where the number came from.

Geometry

The model

Three-quarter · Port high — primary presentation angle. Render of the payload housing from the shipped STL.Aft quarter · Starboard — rear crease break. Render of the payload housing from the shipped STL.Low angle · Hull-slot profile — mounting profile reads. Render of the payload housing from the shipped STL.Detail · Vent louvres & part mark — bold cuts and creases. Render of the payload housing from the shipped STL.Underside · Internal cavity — wall section · fastener bosses. Render of the payload housing from the shipped STL.Mesh · 3,934 triangles — manifold shell, facet structure. Render of the payload housing from the shipped STL.Height field · Plan — false colour by Z · contour isolines. Render of the payload housing from the shipped STL.Orthographic set · section — plan, elevations, mid-plane cut. Render of the payload housing from the shipped STL.
Three-quarter · Port high primary presentation angle
Fabrication

The part

Envelope
168 × 210 × 40 mm
Geometry
3,934 triangles, binary STL
Verified
Calipers, against the shipped file
01 Every curve carries its provenance

The number has a source.

Nothing on this site is asserted without a label. Each figure is measured, calculated, modelled or referenced, and each one traces back through the chain that produced it to a piece of published physics or a bench reading. Follow any stage below.

Given
Derived
Decided
02 The adaptation is not a lookup table

The pulse is a decision.

The solver scores every candidate pulse against the current acoustic link budget and takes the widest sweep that still clears its margin. Move the water and watch the candidates re-score. The frequency changes because the scores changed — there is no turbidity-to-frequency table anywhere in this project.

— NTU
51000 NTU

Temperature, salinity and depth are held at the operating point the published sweep was run at, so this reproduces that table exactly.

Current water
Temp
— °C
Salinity
— ppt
Turbidity
— NTU
Depth
— m
Current solution
fC
— kHz
Bandwidth
— kHz
Resolution
— mm
Margin
— dB
Candidate set —/— clear the line
fC kHz B kHz Margin dB Res mm Verdict

The candidate table is computed in the browser.

A candidate must clear the detection threshold plus the margin this water demands — — dB in total. The solver takes the widest surviving sweep, not the strongest echo, because bandwidth is what buys range resolution.

Why
Spreading
— dB
Absorption
— dB
Scattering
— dB

—

Why this candidate Detection margin against candidate centre frequency
What that does everywhere else Selected centre frequency and usable range against turbidity

Frequency falls a long way. Usable range does not — the solver is spending resolution to keep the link closed, which is the trade the whole payload exists to make.

03 It runs with nothing attached

The vehicle carries the loop.

Sensing, estimation, waveform selection, synthesis, transmission and correlation all happen on the payload, between one ping and the next. There is no server in the path, no pairing step and no surface link — which is the only arrangement that works on a deck with no connectivity and a vehicle already under.

  1. 01 Environment the water
  2. 02 Sensors temp · TDS · turbidity · depth
  3. 03 Estimation c, α, excess
  4. 04 Selection the solver
  5. 05 DDS phase accumulator
  6. 06 DAC 12-bit
  7. 07 Conditioning reconstruction filter
  8. 08 Drive analog stage
  9. 09 Transducer pressure wave
  10. 10 Echo the return
  11. 11 Processing matched filter
  12. 12 Telemetry to the deck app
One ping interval Every stage below happens between one transmission and the next, on the payload
  1. Sample
  2. Estimate
  3. Solve
  4. Synthesise
  5. Transmit
  6. Receive
  7. Correlate

Shown in execution order.

04 The loop is closed, and visible

The model can be wrong.

A prediction that cannot be checked is an opinion. The payload compares the echo it predicted against the echo it received and carries the difference into the next decision as a correction term. That term is implemented and you can drive it below — the measurement that will feed it is the next rung on the validation ladder.

  1. 01 Measure the water
  2. 02 Model predict the echo
  3. 03 Transmit the chosen pulse
  4. 04 Receive what came back
  5. 05 Compare predicted against received
  6. 06 Retune carry the error forward
Implemented Correction term · implemented

decide() takes a correction term. Raising it makes the solver assume a worse link than its own physics predicts, and re-solve against that. Which way the answer moves is the model's to decide, not ours — that is running below.

Rung 05 Signal exists in water

The same test in a tank. This is where absorption and scattering start behaving like the model says they should — or do not.

— dB
0 dB12 dB
Model on its own — kHz · —
Next decision, corrected — kHz · —
Range resolution — was —

—