Centralised or distributed: where to put the acquisition units

A geometry question wearing an architecture costume

Two arrangements are possible. Put one acquisition unit somewhere accessible and run every sensor cable to it. Or put small units near each group of sensors, digitise there, and bring back a network. Everything else — module counts, formats, recording — is unchanged.

The reason this looks like a philosophical debate and is not, is that the deciding factors are all measurable: how many channels, how far apart, how much mass is available, and whether a time reference can reach every corner of the aircraft.

Cable mass, calculated once

Take a group of thirty-two analog channels fifteen metres from the central bay. Centralised, that is thirty-two shielded twisted pairs over fifteen metres — some 480 metres of cable, around five kilograms with connectors and protection. Distributed, it is fifteen metres of network cable, fifteen metres of power, and a small acquisition unit: perhaps half a kilogram of cable and a kilogram of box.

The crossover is where the mass of the unit equals the mass of the cable it saves, and for typical airborne harnesses that lands somewhere around thirty channels at ten metres. Below it, centralising is lighter and simpler. Above it, distributing wins on the mass budget before any other argument is made.

Two corrections to that rule of thumb, both in favour of distributing. Four-wire connections double the cable count for bridges and resistance probes. And routing is never the straight-line distance — a cable that goes round a pressure bulkhead can be twice the length you measured on the drawing.

The analog run is where the measurement is lost

Mass is the argument that gets attention; signal quality is the one that should. Every metre of analog cable is an antenna, a resistance and a thermocouple junction waiting to happen.

• A bridge measured over twenty metres needs four-wire connection, or the cable resistance and its temperature enter the reading directly.
• A thermocouple extension over twenty metres collects electromagnetic interference on a signal of forty microvolts per degree, and adds thermal EMF at every junction.
• An IEPE cable has capacitance, and capacitance limits the usable bandwidth for a given supply current — a real constraint at thirty metres.
• And every long run is a bigger loop area for the airframe currents you did not ask to measure.

Digitising within a metre or two of the sensor removes all of it at once. That is the strongest argument for distribution and it has nothing to do with topology fashion: after the converter, the signal is a number and numbers survive thirty metres of Ethernet without an opinion.

What distributing obliges you to solve

Distribution does not remove problems, it exchanges them. Three arrive with it, and pretending otherwise is how distributed installations disappoint.

First, time. One unit has one clock and no alignment question; four units have four, and they are only aligned if you distribute a reference to all of them. That is entirely solvable — PTP over the same network that carries the data, or IRIG-B distributed alongside — but it becomes a designed feature of the installation rather than a property you inherit.

Second, configuration. Four units mean four configurations, four firmware versions and four sets of calibration coefficients, and a mismatch between them does not announce itself: it produces plausible data. The discipline is a single master configuration, pushed to every unit, with a read-back that is checked rather than assumed.

Third, power and installation. Each unit needs a supply, a circuit protection, a mounting provision and a thermal path, and each one is a separate line in the installation dossier. Four small boxes are more installation work than one bigger box, even when they are less mass.

Maintenance and access, which decide more than they should

A unit you can reach between two flights is a unit you will reconfigure, check and trust. A unit behind a panel that takes four hours to open is a unit you will avoid touching, and eventually a unit whose configuration nobody is sure of.

So the physical accessibility of each proposed location deserves a line in the trade study, next to the cable length. Remote configuration over the network removes most of the need to touch a distributed unit, which is precisely why a network-based architecture and remote configuration belong together — one makes the other bearable.

The mixed architecture, which is the real answer

Almost every serious installation ends up mixed, and it is not a compromise: distributed acquisition front ends near the sensor groups, one recorder holding the storage and the download interface, and one time reference feeding everything. The analog stays short, the mass stays low, and there is still only one file at the end of the flight.

A modular platform with a master and slave arrangement over Ethernet is built for exactly that: the same module types in every unit, one of them holding the memory, the others feeding it. The design question stops being "central or distributed" and becomes "how many zones, and where".

Four questions that settle it

1. How many channels, and how far apart are their sensors? This gives you the cable mass, and the cable mass usually gives you the answer.
2. What is the mass budget, and who owns it? An instrumentation engineer who has not asked this will be told the answer late, by someone with authority.
3. Can I distribute a time reference to every proposed location? If the answer is no for one of them, that location is not a candidate.
4. Who maintains this in three years, and can they reach it? The installation that survives is the one somebody can still open.

Answer those four with numbers rather than preferences and the topology stops being a matter of taste. It usually turns out that the aircraft decided it before anyone sat down.