Telemetry
Decommutation and calibration: from raw counts to engineering units
Between the bits on the medium and the curve on the screen, every parameter goes through two operations: decommutation finds it, calibration gives it a unit. Most wrong values in a test report come from one or the other, and both can be checked.
Two steps, two questions
A recording does not contain "airspeed" or "strain on frame 24". It contains words: positions in a PCM frame, labels on an ARINC 429 bus, messages on a MIL-STD-1553 bus, samples from an acquisition module. Turning them into physical values takes two steps that are often confused.
- Decommutation finds the bits of a parameter: which word, which bits in it, at what rate, put together how.
- Calibration turns those bits into a value with a unit: counts into volts, volts into microstrain, degrees or bar.
scroll
Finding the parameter
In a PCM frame, a parameter is a position: word 37 of every minor frame; word 12 of every fourth minor frame if it is subcommutated; words 5, 133, 261 and 389 if it is supercommutated four times in a 512-word frame. The decommutator locks on the synchronisation pattern, counts the words, uses the subframe counter for the subcommutated ones, and hands each parameter its words in order.
On a bus, a parameter is an address: a label on ARINC 429, a subaddress and a word position on MIL-STD-1553, a key and a parameter number in an Ethernet stream. The principle is the same — an address, then bits — and so is the risk: the address comes from a document, and the document must match the configuration that flew.
Some parameters do not fit in one word. A value of more than twelve bits spreads over two words of an ARINC 717 frame; a 32-bit counter over two 16-bit words. The frame map says which part is the most significant, and getting it backwards gives values that jump at every carry.
Reading the bits: the binary format
Once the bits are found, they must be read the way they were written. The same sixteen bits stand for different numbers depending on the format:
scroll
| Format | What it is | The trap |
|---|---|---|
| Unsigned binary | A count from 0 upwards | A small negative value shows up as a very large one |
| Two's complement | The usual signed integer | Read as unsigned, small negative values land near full scale |
| Offset binary | Zero at mid-scale, common at converter outputs | Read as two's complement, the sign flips around zero |
| BCD | One decimal digit every four bits | Read as binary, the values are wrong but stay plausible |
| Discrete bits | One bit, one state | The polarity: active high or active low |
| Floating point | IEEE 754 on 32 or 64 bits | The byte order between two systems |
Bit order adds a second layer: a word sent most significant bit first or last, an ARINC 429 label that travels reversed, bytes swapped between a big-endian network and a little-endian computer. Laying the word out bit by bit settles most of these questions in a minute; our online ARINC 429 and MIL-STD-1553 decoders do exactly that.
From counts to volts
For an analog channel, the raw value is a converter output. With an N-bit converter over a range of ±V, one count is worth 2V / 2^N: about 0.3 mV with 16 bits over ±10 V. The acquisition setup holds the range and the conditioning gain; without them, the counts are just numbers.
Many acquisition units take this step themselves and record values already in volts, or already in engineering units. That is convenient, provided everyone knows it. A ground station that applies the converter law a second time gives values wrong by orders of magnitude — or wrong by a few percent, which is worse, because nobody notices.
From volts to engineering units: the calibration law
- Linear: value = a × x + b, from a sensitivity and an offset. Most pressure transducers, accelerometers and displacement sensors.
- Polynomial: a few coefficients, for sensors whose response bends, or a linear sensor corrected over its range.
- Table: pairs of points, interpolated in between, when the response is measured rather than modelled.
- Reference curves: thermocouples and resistance thermometers follow published tables, with cold-junction compensation for thermocouples.
- Derived: a value computed from several parameters — principal strains from a rosette, a pressure difference, a rate from two positions.
The law comes from a calibration: the sensor measured against a reference at known points, on a known date, and signed. That document belongs to the recording as much as the data does. Without it, a value has a unit but no traceability.
Where the definition lives
Everything above is configuration: the frame map, the formats, the converter ranges, the calibration laws. In IRIG 106 recordings, TMATS can carry it with the data: its data conversion group describes, measurement by measurement, the binary format, the unit and the conversion — pairs of points, polynomial coefficients, derived or discrete. Elsewhere it lives in a test database and in the setup files of each unit.
The rule is the same everywhere: the configuration archived with a flight must be the one that flew. A calibration updated on Tuesday and applied to Monday's flight is a classic, and so is the reverse.
Checking the result
- Known inputs before the first flight: a calibrated pressure source on a pressure line, a known angle on a position sensor, and the expected value read at the end of the chain.
- A shunt calibration on strain-gauge bridges: a known resistor across one arm gives a known output, which checks everything behind the gauge.
- Cross-checks between redundant measurements: two sources of the same quantity must agree within their tolerances.
- Plausibility on the ground: an aircraft at rest, at a known altitude, temperature and attitude, gives a reference for many parameters before engine start.
- And a line in the report saying which calibrations and which configuration version were applied.
The usual mistakes
- The wrong binary format: values that look plausible and are not.
- A calibration applied twice, once on board and once on the ground.
- A sensor replaced, its calibration not.
- A unit mismatch hidden by a plausible scale: psi for kilopascals, millivolts per volt for millivolts.
- A subcommutated parameter read on the wrong subframe: the right format, the right law, the wrong parameter.