Pressure Sensor¶

Seaglider knows its depth from one analog pressure sensor in the aft section. It is a strain-gauge bridge: current builds use a Kistler piezoresistive sensor, and older gliders a Paine. The main board amplifies the bridge output and reads it with a 24-bit ADC. Two parameters turn the raw A/D counts into pressure, and from that the depth. Almost everything the glider does in a dive hangs on that number: apogee, flare, the surface decision, $D_TGT, $D_NO_BLEED, the flight model. A pressure fault therefore rarely shows up as "bad data". It shows up as a glider that dives, climbs or aborts at the wrong time.

A second, unrelated sensor on the electronics measures the internal hull pressure (plus humidity and temperature). It is covered at the end of this page.

Source

Paraphrased from 2022–2024 correspondence between Seaglider operators, APL-UW IOP and service providers (Rev E upgrades, Kistler integration, a pressure-cable failure), a Kistler calibration certificate, the manufacturer's support correspondence (2022), and operator notes on the internal pressure sensor. Board revisions and factory configurations vary — defer to APL-UW IOP and your glider's documentation.

From counts to depth¶

The glider computes gauge pressure as

pressure (psig) = counts × $PRESSURE_SLOPE + $PRESSURE_YINT

and depth from pressure.

Parameter	What it is	Where it comes from
`$PRESSURE_SLOPE`	psig per A/D count	The sensor's calibration sheet, the excitation voltage, the amplifier gain and the ADC resolution (below)
`$PRESSURE_YINT`	Offset in psig	The sea-level zero routine, run with the glider at atmospheric pressure

The two are not independent: the Y-intercept is only right once the slope is right. Set the slope first, then zero.

For a sense of scale, one Kistler sensor wired directly to a Rev E board ran with a slope of about 1.09e-4 and a Y-intercept between about −156 and −167 psig. The firmware flags a $PRESSURE_YINT outside −200 to 0 as out of the recommended range. Treat that warning as a sign that the slope, gain or wiring is wrong. It is not something to "fix" by accepting the new value.

Computing the slope from the cal sheet¶

The manufacturer's trim spreadsheet computes the slope as:

$PRESSURE_SLOPE = (sensitivity, psi per mV/V)
                  × (1 / excitation voltage)
                  × (1 / amplifier gain)
                  × (volts per A/D count)
                  × 1000

Sensitivity — full-scale pressure divided by the span on the calibration sheet. A 0–1500 psi Kistler with a span of 9.94 mV/V gives about 151 psi per mV/V.
Excitation — the bridge supply from the board (5 V).
Gain — the total gain between the sensor and the ADC (see below).
Volts per count — the ADC reference over its resolution, about 1.49 × 10⁻⁷ V (2.5 V over 2²⁴ counts).

For that sensor at a gain of 41.16: 151 ÷ 5 ÷ 41.16 × 1.49e-7 × 1000 ≈ 1.093e-4, which is what APL-UW supplied for it.

Get the cal sheet with the sensor

New and spare sensors have arrived without calibration data. Kistler can supply the certificate for a given serial number. Request it before you need the glider, not on the launch day.

Amplifier gain on Rev E¶

On Rev E main boards the pressure amplifier gain is set in the supervisor (hw/super), in steps from 1 to 128. That is only part of the total gain. Some boards also carry fixed hardware gain:

A glider whose pressure sensor goes through an aux board, or a Rev E main board built with the gain in hardware, typically has a fixed gain of about 41.16. Set the supervisor gain to 1 and compute the slope with the hardware value.
A sensor wired straight to a Rev E board with no hardware gain uses the supervisor gain alone.

Stacking the two is a known trap. In one launch, a new Kistler on a factory-built Rev E board had the supervisor gain set to 32 on top of a fixed hardware gain. The sea-level reading was pegged at the ADC maximum and the glider reported itself at ~1467 m on the surface.

To find out which case you have, set the supervisor gain to 1 and repeat the sea-level test. Counts that stay pegged mean a wiring or electrical fault. Counts that become sensible mean the board has hardware gain. Confirm with a one-point check using compressed air or a hand pump on the pressure port. A technician can also look for the gain resistor on the underside of the board.

Paine sensors on Rev E¶

Rev E came after Seaglider moved to Kistler sensors, and a Paine does not work on a Rev E board out of the box. The Rev E amplifier cannot swing all the way to 0 V. The Kistler has a small positive output at zero pressure, so this doesn't matter. The Paine outputs almost nothing at zero pressure, so the ADC sees nothing until about 100 psi. That leaves roughly 60 m of dead band at the shallow end, where the glider needs depth most. Trying different supervisor gains on such a glider did not give readings that agreed with a hand pump.

When a 10/24 V glider with a Paine is upgraded to Rev E, the practical fix is to fit a Kistler as part of the upgrade. Hardware workarounds exist, but they are board-level modifications to be done with APL-UW.

Sea-level zero¶

The sea-level routine (hw/pressure/sealevel in the menus, also runnable from pdoscmds.bat) samples the sensor at atmospheric pressure. It prints the mean, RMS, min, max and peak-to-peak A/D counts, and proposes the Y-intercept that would make the mean read zero.

A healthy result looks like this:

counts well inside the ADC range (neither 0 nor 16,777,215),
peak-to-peak noise of a few hundred to about a thousand counts, which is a few centimetres,
a proposed Y-intercept inside the recommended range, and close to the last good value for that sensor.

At launch the glider runs a similar check ("20 samples mean depth … noise … threshold"). On a healthy glider that shows a mean within a few tens of centimetres of zero and noise of about 3 cm, against a 7 cm threshold.

The zero routine will happily zero garbage

The routine computes a Y-intercept from whatever it reads, and in some contexts (self-test, launch) the new value is accepted automatically. Two real examples:

With counts pegged at maximum, it proposed −2358 psig. The glider then read "0 m" while sitting at the surface with a dead depth channel.
With a disconnected sensor reading zero counts, it set the Y-intercept to essentially 0. Every later boot showed a perfect 0.00 m.

Always look at the A/D counts before believing the metres. If the proposed Y-intercept is far from the last good value, stop and find out why.

After a main board replacement, the pressure calibration does not come across on the SD card. Settings live in the board's non-volatile memory. Re-enter $PRESSURE_SLOPE (from the old parameter dump or the trim sheet), check the gain, then redo the sea-level zero.

When depth goes wrong¶

Symptom	Likely cause
Counts pegged at 16,777,215 at the surface	Too much total gain (supervisor gain on top of hardware gain), or a wiring or electrical fault
Counts at 0; "pressure sensor got zero", "10 consecutive bad pressure reads – sensor dead"	Sensor not connected, a broken conductor or a bad crimp
Depth fine on the bench and on the descent, then wild values on the climb or after apogee	Intermittent connection that moves with pitch or attitude — check the cable and connector
Readings that jump (e.g. between ~0 m and ~100 m) when the cable is touched	Poor joint or crimp in the sensor cable
Hand-pump readings that don't match the gauge, or no response in the first tens of metres	Wrong slope or gain, or a Paine on Rev E (dead band)

Case: a failed deployment. After a refurbishment and a VBD swap, a glider behaved normally in shallow trials but failed a deployment. It flew wrong dive and climb patterns with bad pressure readings, and aborted after three dives. The motor controller has its own ADC for pressure, and its records agreed with the main board's bad values. That pointed away from the main ADC and toward the sensor side. When the glider was opened:

A conductor in the Kistler adapter cable had come loose. The solder joint between the cable shield and a 30 AWG wire had never bonded properly (dull, dirty solder).
Another 30 AWG conductor had a poor crimp, and a pin had a retaining tab bent the wrong way.
A main board had also been damaged while the glider was being opened and inspected.

After re-terminating every pin and adding heat-shrink strain relief where the fine wires leave the shielded cable, the sensor read steadily at sea level while the cable and connector were wiggled.

Lessons:

Run a self-test after every reassembly and read the pressure section critically. In this case the fault was visible in a self-test weeks before the failed launch, but nobody caught it.
The Kistler cable uses 30 AWG conductors, which are fragile. APL-UW crimps them with pins designed for 30 AWG. The 0.1-inch MTA connectors on the main board are insulation-displacement types, so the wire goes in unstripped. Stripping the ends first makes an unreliable joint. Support the point where the fine wires leave the jacket.
Wiggle-test the cable and connector while logging pressure before you close the hull.
Disconnect the battery before unplugging or probing cables on the main board.

Internal pressure sensor¶

The internal sensor (hw/intpress) reports hull pressure in psia, plus relative humidity and temperature. It is how you set and monitor the hull vacuum, and a slow rise during a mission is an early sign of a leak.

The usual sequence at close-up:

With the hull sealed but not yet evacuated, the reading should be atmospheric, about 14.5 psia.
Pull the vacuum down to about 9–9.5 psia (some teams stop at 8.5–9).

New sensors rarely read exactly 14.5 psia at atmospheric. The refurbishment procedure sets a nominal slope and then adjusts $INT_PRESSURE_YINT until the unevacuated reading is 14.5 psia, before pulling vacuum. On one Rev E glider the slope stored on the board was not the manual's value. Keeping the stored slope and adjusting only the Y-intercept gave the right result.

A dropped digit is not a leak

One operator saw readings of about 9.1 psia with an occasional "near-zero" value while pulling vacuum. In the log, those values were printed with the leading digit missing (".12 psia" for 9.12). It was a display glitch, not a sensor or vacuum fault. Read the raw log before chasing a leak.