Signal Integrity

Hypothesis: A validation pipeline that flags rather than silently removes bad samples preserves more usable data and makes later analysis defensible.
Study scope: A single analogue channel through filtering and a three-stage validation pipeline.

02Problem and constraint

A pressure channel shows high-frequency noise, an isolated spike when a connector is disturbed, a short dropout and a slow offset drift. Any of these can be mistaken for physical events if the trace is only smoothed.

Programme constraint

The logger has fixed channel rates and limited bandwidth. Raw data must remain recoverable alongside the validated channel.

03Instrumentation and method

Instrumentation

Strain-gauge pressure transducer with conditioning amplifier
Logger channel at 500 Hz with an anti-alias filter
Reference channel on a second logger to check timing
Event log recording every sample the pipeline touches

Method

1Keep the raw channel. Derive a filtered channel and a validated channel from it, and record exactly what each stage changes.
2Apply range, rate-of-change and dropout checks before any smoothing, so that a spike is flagged and not blurred into its neighbours.
3Report how many samples were flagged and why, alongside every analysis that uses the channel.

04Annotated visualisation

A raw pressure trace with four numbered events. Switch between raw, filtered and validated views.

Fictional · illustrative

FIG 1Trace with event markers

Pressure channel at 500 Hz

Raw

Hover, tap or focus the chart and use ← → to read values.

1 spike · 2 dropout · 3 step offset · 4 noise burst. The filter smears the spike into its neighbours; the validated channel flags it first.

FIG 2What validation changed

Validated minus raw

Validated − raw

Hover, tap or focus the chart and use ← → to read values.

Only two events change samples. The step and the noise burst are reported and left untouched.

05Example data and assumptions

Illustrative event log. The trace is synthetic and generated locally from a fixed seed.
Event	Type	Pipeline action	Samples
1	Spike	Flagged by rate check; interpolated	3
2	Dropout	Flagged by range check; gap kept	18
3	Step offset	Flagged for review; not altered	1
4	Noise burst	Reported; kept in raw channel	40

Illustrative event log. The trace is synthetic and generated locally from a fixed seed.

Assumptions

Noise is broadly white outside the event windows.
Rate-of-change threshold chosen from the physical limits of the hydraulic line.
Both loggers are synchronised to within one sample.

06Trade-offs, limits, verification

Trade-offs

A tight threshold catches more faults but flags more genuine fast events.
Filtering lowers noise but delays the signal; the delay differs from other channels.
Interpolating a short gap is convenient, but the result is no longer measured data and must stay labelled.

Limitations

Cross-channel plausibility checks are not shown here.
Thresholds from one run may not suit another operating condition.
A fault that mimics real behaviour cannot be caught by single-channel checks.

What we would verify in real-world testing

Fault injection at the connector to confirm each check triggers
Comparison against a calibrated reference gauge
Long-duration run to see whether drift is thermal or electrical

07Related

Related capability

04 Electronics & data

Next study

TH-01 Thermal Envelope

Contact

Discuss a programme

Event

Type

Pipeline action

Samples

Spike

Flagged by rate check; interpolated

Dropout

Flagged by range check; gap kept

Step offset

Flagged for review; not altered

Noise burst

Reported; kept in raw channel