2026-08-27 · Technical Q&A

Magnetostrictive Displacement Sensor: Why the Last Digit Keeps Flickering

Once a magnetostrictive displacement sensor is installed, one thing tends to draw the eye: the last digit of the position reading keeps flickering, moving two or three counts back and forth. The conclusion first — a flickering last digit is usually not a failure. It is what quantisation steps plus the noise floor look like. The real question is whether it eats into control quality.

Magnetostrictive displacement sensor rod, flange and M12 connector — From the echo in the waveguide to the output signal: every stage on that chain adds a little noise to the last digit

Split the flicker into three kinds first

All three look like "the number is jumping", but the causes are different and the fixes point in opposite directions:

Fast last-digit jitter: one or two least-significant steps, fast and even. Quantisation plus noise floor — normal, and the cure is filtering or a dead band.
Slow drift: one step every few minutes to tens of minutes, mostly in one direction. This is temperature, and the figure to read is the temperature coefficient — see Operating vs Storage Temperature.
Sudden jumps: tens of millimetres and back, or tied to a machine action. Not noise — treat it as interference or a connection problem and follow the fault flow in Signal-Jump Troubleshooting Flow.

Resolution is not stability: steps and noise are two things

Resolution is the smallest step that can be distinguished, commonly in 1 / 2 / 5 / 10 / 20 / 50 / 100 μm classes. The noise floor is how much a static reading moves. Their ratio decides whether the last digit sits still:

Noise well below one quantisation step: the last digit is pinned, and the reading looks very steady.
Noise at the same order as one step: the reading flips between two adjacent counts — the fast jitter described above.

The same symptom therefore has two explanations: noise that is genuinely high (an installation issue), or resolution ordered higher than the application needs, which magnifies movement that could safely have been ignored. How resolution differs from the other accuracy terms is in The Three Accuracy Terms.

Three fixes on the controller, each with a penalty

There are three usual routes, and they cost different things:

Method	What it does	What it buys	What it costs
Moving average	Averages the last N cycles	Random noise falls with √N	Lag of about N/2 cycles
First-order low-pass	Follows the input by a time constant	High-frequency jitter is smoothed	Phase lag; loop gain must come down with it
Dead band (hysteresis)	Updates only past a threshold	Display and output stop flipping	Real small movements are swallowed

Which one to use depends on whether the loop needs to be steady or to keep up. There is a common trap here: set the filter time constant longer than the control cycle and the position loop looks steady, while the response is in fact slower and the cycle time drops — the operator notices first. How refresh rate should match the control cycle is in Matching Response Time and Refresh Rate.

When to stop tuning and look at the installation

Pattern is the clue. In these cases, move attention from the controller to the machine:

Flicker appears with a particular action (a drive starting, a proportional valve shifting, a welder nearby) — check earthing and routing, see Shield: One End or Both.
Flicker grows with time and with cabinet or cylinder temperature — deal with the heat rather than increasing the filter.
The analog output flickers while the same unit's bus reading does not — the problem is in the output loop or supply, see Reading at Zero: End of Stroke or Broken Wire?
Two sensors on the same machine start flickering together — suspect a shared earth or supply first, see On-Site EMC Fixes: Breaking Source, Path and Victim.

The criterion: is the flicker inside the accuracy budget?

Put the rolling peak-to-peak against the process tolerance. If the flicker peak-to-peak is under one third of the permitted process error and shows no clear pattern, leave it alone. If it exceeds one third, or appears with a machine action, work through the table above. The reference is the process tolerance, not "the last digit must never move" — the sensor's own resolution makes that impossible. How to convert parameters into a tolerance in millimetres is in Displacement Sensor Parameters Explained.

Frequently Asked Questions

Q: The last digit keeps flickering. Is the sensor faulty?

Usually not. Fast last-digit jitter is what quantisation steps plus the noise floor look like, and two or three counts moving back and forth is common. The real question is whether it eats into the accuracy budget: if the flicker peak-to-peak is under one third of the permitted process error and shows no clear pattern, leave it alone. It only needs handling when it exceeds one third, or appears together with a machine action.

Q: All three kinds look like a jumping number. How do I tell them apart?

Separate them by amplitude and rhythm. One or two least-significant steps, fast and even, is quantisation plus the noise floor and is normal. One step every few minutes to tens of minutes, mostly in one direction, is temperature drift and the figure to read is the temperature coefficient. Tens of millimetres and back, or tied to a machine action, is not noise: treat it as interference or a connection problem. The three point in opposite directions, so classify before acting.

Q: What controller-side fixes exist for flicker, and what does each cost?

Three. Moving average: averages the last N cycles, cutting random noise with the square root of N, at the cost of a lag of about N/2 cycles. First-order low-pass: follows the input by a time constant and smooths high-frequency jitter, at the cost of phase lag, so loop gain must come down with it. Dead band or hysteresis: updates only past a threshold so display and output stop flipping, at the cost of swallowing real small movements.

Q: Is a longer filter time constant always steadier?

No. Set the filter time constant longer than the control cycle and the position loop looks steady, while the response is in fact slower and the cycle time drops; the operator usually notices first. Which method to use and how strongly to set it depends on whether the loop needs to be steady or to keep up. More is not better.

Q: When should I stop tuning and look at the installation instead?

Pattern is the clue. If flicker appears with a particular action such as a drive starting, a proportional valve shifting or a welder nearby, check earthing and routing first. If it grows with time and with cabinet or cylinder temperature, deal with the heat rather than increasing the filter. If the analog output flickers while the same unit's bus reading does not, the problem is in the output loop or supply. If two sensors on the same machine start flickering together, suspect a shared earth or supply first.