Este guia está disponível atualmente em inglês.
Technology / Technical guide
Signal Debouncing: Persistence, Sampling & Event Timing
One physical change can produce several short digital changes in its signal representation. Debouncing applies a defined persistence or qualification rule before accepting a change. That rule can reduce repeated event counts in a particular model, but it can also hide a genuine short event. Timing assumptions therefore matter as much as state labels.
An accepted event is the result of a declared rule.
Raw observations, accepted states and event timestamps represent different points in a signal interpretation.
01 / Raw changes
Repeated transitions do not imply repeated intended events.
Microchip describes contact bounce as intermittent making and breaking that can generate multiple digital pulses.
Microchip describes contact bounce as intermittent making and breaking that can generate multiple digital pulses. A persistence rule can represent these pulses differently from an instantaneous threshold. This is a generic signal concept, not evidence of any MARSTON implementation.
Raw input history still contains the short changes. The accepted state is another representation formed from that history. Treating those two signals as interchangeable would obscure which events the rule has delayed, merged or left unrepresented.
02 / Example rule
Consecutive observations define one possible relationship.
Consider an abstract rule that accepts B after three consecutive valid B observations, beginning in accepted state A.
Consider an abstract rule that accepts B after three consecutive valid B observations, beginning in accepted state A. A valid A observation resets this example’s B count. The raw sequence B, B, A, B, B, B leaves the accepted state A until the sixth observation.
The first two B values do not qualify because the third observation interrupts their sequence. This rule is invented to explain persistence; A and B are neutral labels. It specifies no real protection event, input processing behaviour or response of a turbine.
03 / Elapsed time
Observation count is not automatically a duration.
If observations are spaced by Ts, the third qualifying observation occurs 2Ts after the first qualifying observation.
If observations are spaced by Ts, the third qualifying observation occurs 2Ts after the first qualifying observation. With hypothetical Ts = 0.2 s, that span is 0.4 s. It is not automatically three times the interval, because there are two gaps between three observations.
A physical change just after a sampling instant must also wait before its first observation. Irregular timing changes the elapsed span for a given count. The timestamp of acceptance and the timestamp of the underlying physical event can therefore differ even in the ideal example.
04 / Persistence
Continuous dwell and sampled agreement are different claims.
A continuous-time rule can require a condition to persist throughout a duration.
A continuous-time rule can require a condition to persist throughout a duration. A sampled rule only establishes what was represented at its observation instants. Several equal samples do not prove there was no short change between them.
Missing or invalid observations raise another question: a particular rule may reset, pause or invalidate its qualification. None of those responses follows merely from the word debounce. The example’s consecutive valid observations do not define a universal treatment of missing information.
05 / Other filters
Time qualification differs from amplitude smoothing.
An analogue or digital low-pass transformation changes the amplitude history passed onward.
An analogue or digital low-pass transformation changes the amplitude history passed onward. A consecutive-state debounce rule qualifies an event through observations. Debounce can be implemented using filtering, but the two terms do not specify the same mathematical input–output relationship.
Schmitt hysteresis instead makes switching depend on different amplitude boundaries and prior state. Combining these ideas would create a more detailed model. Neither amplitude history nor persistence by itself determines whether a particular short input excursion was physically important.
06 / Scope
Delay cannot establish that a rejected event was noise.
A brief real event and a brief disturbance can have similar observed timing.
A brief real event and a brief disturbance can have similar observed timing. A persistence rule can distinguish duration within its representation, but not necessarily the physical cause. Accepting a later stable state also does not recover an earlier event that was never resolved.
No MARSTON debounce timing, input qualification or protection sequence is asserted. This page provides no safety-delay settings, wiring, fault injection or control changes. It explains the difference between a raw change and the interpretation produced by a stated timing rule.
Related technical guides: Schmitt Hysteresis: Thresholds, History & Retained State · Numerical Derivatives: Signal Slope & Noise Amplification · Model Residuals: Prediction Gaps & Ambiguous Causes.
General background: Microchip AN2805: contact bounce and persistence concepts. These references explain industry concepts; they are separate from the applicable MARSTON product and project documents.
Key concepts and conditions.
Read each quantity with its reference, units and stated conditions. A general relation does not establish a model-specific value.
| Concept / quantity | Meaning / relation | Conditions / limits |
|---|---|---|
| Raw observation | Instantaneous represented input | A timestamp and validity meaning are needed |
| Accepted state | Result after qualification | Distinct from the unprocessed history |
| Consecutive count | Number of uninterrupted qualifying samples | The example resets on a valid opposite sample |
| Elapsed span | Time between qualifying observations | N observations contain N−1 gaps |
| Missed excursion | Change between observations | Equal samples do not prove continuous constancy |
Engineering questions.
Does debounce mean every short event is noise?
No. Duration alone does not establish physical cause.
Do three equal samples span three intervals?
No. Their first-to-third span contains two intervals.
Is acceptance time the physical event time?
Not necessarily; sampling and qualification can delay acceptance.
Does sampled agreement prove continuous persistence?
No. A short change may occur between observations.
Is debounce identical to low-pass filtering?
No. An event qualification rule and an amplitude transformation describe different relationships.
Are turbine safety delays specified here?
No. The example is abstract and gives no operational settings.
Technical enquiry
Discuss the technical question.
One physical change can produce several short digital changes in its signal representation. Share the model and applicable operating or design information for an engineering discussion.