Data acquisition, timing and measurement synchronisation

Name the instant represented by a value

Sampling selects observations at discrete times; quantisation assigns finite digital values to them. The time when the physical input is sampled can differ from the time when a converted result becomes available or reaches a file. A timestamp needs a declared meaning and a relation to the physical event being studied.

A multiplexed system can share one converter across channels, acquiring them in sequence. A simultaneous-sampling arrangement can reduce this interchannel offset, but clock paths and channel delays still require characterisation. The advertised sample rate may refer to each channel or to an aggregate conversion rate; the distinction changes the attainable record.

A common start does not ensure a common timebase

Channel skew is a relative timing difference. Latency is elapsed time between an input event and its available result. Jitter is variation in timing from one event to another. A constant delay may be corrected if known, while uncertain or variable delay leaves uncertainty. Filtering and sensor response can add frequency-dependent delay beyond the converter clock.

Two independent clocks can start together and gradually separate. A supplied relative rate mismatch of 20 parts per million accumulates 20×10⁻⁶×100 s=.002 s over 100 s. A common trigger can align a start while leaving this drift. Synchronisation evidence must address the duration and the precision needed for the actual comparison.

Keep the input spectrum and the record intact

A finite sampling rate cannot distinguish every possible continuous waveform. For ordinary low-frequency-band acquisition, the input needs a justified band limit and adequate sampling margin. An analogue anti-alias filter suppresses unwanted components before sampling; a later digital filter cannot generally identify and remove a component already folded onto the wanted signal.

Record channel order, actual timing, filter settings, units and configuration. Buffering can decouple acquisition from file writing, but it can also conceal growing delay or overflow unless monitored. Missing samples, repeated stale values and reordered records should remain identifiable. A smooth line drawn through them is a presentation choice, not replacement evidence.

Check an apparent phase difference

The original common signal is sin(2π50t) V. Channel A samples at 0, 5, 10 and 15 ms; channel B at 2, 7, 12 and 17 ms. The channels observe the same waveform, but pairing each row as simultaneous introduces a 2 ms timing error. Its phase equivalent is 360×50×.002=36°.

At the first row, A sees 0 V while B sees sin(.2π)≈.588 V. The figure places both at their actual times so their common origin remains visible. The waveform is supplied exactly for this example; sparse points alone would not establish that it is a unique 50 Hz sine. A timing correction also cannot repair an unknown sensor lag without further evidence.

Try the reasoning

1. Does the file-write timestamp necessarily identify sample time?

   Show the answer and reasoning

   No. Conversion, transport, buffering and software can separate those instants.

2. Why can one common trigger leave a long record misaligned?

   Show the answer and reasoning

   Independent sample clocks can drift after the shared start.

3. Can digital smoothing reliably undo aliasing after acquisition?

   Show the answer and reasoning

   No. Aliased and wanted components may already be indistinguishable in the sampled record.

What to carry forward

Attach every changing measurement to its actual acquisition time and bandwidth.