Selecting data acquisition hardware for defence trials

A trial happens once. The acquisition system either captured what was needed or it did not, and no amount of processing afterwards recovers a measurement that was corrupted at the sensor interface.

Start from the measurement, not the channel count

Specifications that begin “32 channels at 100 kS/s” have started in the wrong place. Begin with what physically has to be known: which quantities, over what range, to what accuracy, with what time resolution, and what the result will be compared against.

That chain — sensor, excitation, conditioning, conversion, storage — is a single design problem. The converter is rarely the limiting element. A 24-bit ADC behind a poorly grounded bridge amplifier produces 24 bits of noise, and the datasheet figure on the front page has nothing to do with what the system achieves.

Fix the required accuracy first, then work backwards to what each stage must contribute. It is the only way to know whether a requirement is met or merely hoped for.

Sampling rate and anti-aliasing

Sample rate follows from the highest frequency of interest, with the usual factor of at least two and in practice five to ten for anything whose waveform shape matters rather than just its frequency content.

The part that gets skipped is the anti-alias filter. Energy above half the sample rate does not vanish; it folds back into the band as a signal indistinguishable from real data. On a shock or vibration channel this produces plausible-looking results that are entirely fictional. Filter in hardware ahead of the converter — a digital filter after sampling cannot undo aliasing that has already happened.

Where events are transient rather than continuous — shock, blast, separation — pre-trigger buffering matters more than sustained rate. A system that starts recording when it detects the event has already missed it.

Synchronisation is a specification, not an implementation detail

The moment there is more than one channel or more than one instrument, how the channels align in time becomes part of the requirement. Simultaneous sampling across channels, or a known and characterised skew between them, decides whether phase relationships and time-of-arrival differences mean anything.

Across separate instruments, the timing reference has to be explicit: a shared clock, a distributed trigger, GNSS-disciplined time or IRIG. Stating the accuracy the trial needs — milliseconds, microseconds — early prevents an expensive discovery at analysis.

Timestamp everything, in a single documented convention, and record the convention with the data. A file whose timebase cannot be reconstructed two years later is not a record.

Signal conditioning and grounding

Most acquisition problems are analogue problems. Excitation stability for bridge sensors, correct amplifier configuration, isolation where the sensor sits at a different potential, and a grounding scheme that avoids loops — these determine data quality far more than the converter.

Ground loops are the classic field failure: a system that is clean on the bench and unusable on a vehicle where sensor and logger reference different points. Isolation costs money and channel density and is frequently worth both.

Match the conditioning to the sensor rather than buying a general-purpose card and adapting: thermocouples need cold-junction compensation, strain gauges need stable excitation and completion, IEPE accelerometers need constant-current supply. A card that nominally accepts everything usually does none of it especially well.

Ruggedisation and the trial environment

Trials hardware lives on vehicles, in the field, and in weather. Temperature range, vibration and shock, ingress protection, power quality from an unstable vehicle supply, and EMC in the presence of transmitters all belong in the specification.

Power deserves specific attention. A vehicle electrical system is a hostile supply — sag on cranking, load-dump transients, ripple — and an acquisition system that browns out mid-run has lost the trial. Specify the supply behaviour it must tolerate, and provide holdup where a clean shutdown matters.

Physical access matters too: how the unit is mounted, how cables are strain-relieved and how data is retrieved between runs without dismantling the installation.

Data integrity and recovery

Design for the trial ending badly. Power loss mid-write, a full disk, a disconnected cable — the question is what state the recorded data is in when that happens. Journalling or streaming to a format that tolerates truncation, plus redundant storage where the run cannot be repeated, is cheap compared with a lost trial.

Check the data in the field, before you pack up. A quick integrity and sanity check between runs catches a dead channel while there is still time to fix it; discovering it a week later means the trial did not happen.

Calibration records are part of the data. A measurement without a traceable basis, instrument identity and calibration date is not a result that will survive review, and for defence acceptance that review is certain.

Sources.

• Our own work on data acquisition and logging systems, including an airborne data-logger payload delivered on an Indian Navy quadcopter.
• Sampling and anti-aliasing: the Nyquist–Shannon sampling theorem and common instrumentation practice.

Published 16 August 2026. Last revised 25 September 2026. Corrections to sales@vebixautomation.com.

Instrumenting a trial?

Send the measurement list, the accuracy you need and the environment it has to survive. We will reply with a signal chain and what it will really achieve. Ask for a quote.