An analog input that will not sit still is one of the most common field complaints in an instrumentation loop. The reading jitters around its true value, spikes to nonsense for a scan or two, or ripples in time with a nearby motor, and the trend in SCADA looks like a fuzzy band instead of a clean line. Almost always the sensor itself is fine and the problem is electrical noise getting into the 4-20mA loop somewhere between the transmitter and the input card. This guide walks the field procedure to find and kill that noise in order, from breaking ground loops to fixing shield grounding, separating the signal from motor cabling, and deciding where damping and spike filters belong.
Troubleshoot a noisy 4-20mA signal in one line: A noisy or jumpy 4-20mA signal is caused by electrical interference coupling into the loop, not by a bad sensor, so the fix is to find where the noise enters and stop it. Work in order: find and break any ground loop, verify the cable shield is grounded at one end only, separate the signal wiring from VFD and motor cables, then add or tune damping at the transmitter and, if fast spikes remain, a median or low-pass filter. Cleaning the wiring first and filtering last gives a stable reading without hiding a real process change.
The symptom is an analog value that moves when the process is not moving. It might jitter by a small amount continuously, showing up as a fuzzy trend, or it might sit clean and then jump to a wild value for one or two scans before settling. Sometimes the noise is periodic and tracks something in the plant, most often ramping up whenever a nearby variable frequency drive or large motor starts. The first diagnostic instinct should be to decide whether the movement is real process variation or induced noise, because a genuinely swinging process needs a control answer, not a filter.
The likely causes fall into a short list. A ground loop, where the loop finds two paths to earth at slightly different potentials, injects a circulating current that rides on top of the measurement and often shows up as a steady offset or a slow wander. Poor or double-ended shield grounding lets the shield act as an antenna instead of a drain, so capacitively coupled hum leaks into the signal pair. Running the signal cable in the same tray or conduit as VFD output leads or motor power couples switching noise directly into the loop, which is the classic cause of a reading that gets noisier the harder the drive works. Finally, a genuinely fast electrical transient can produce the one-scan spikes that no amount of averaging cleanly removes.
Reading the pattern narrows the cause before you touch anything. A constant fuzz points at continuous coupling from nearby power or a marginal shield. Movement that appears only when a specific drive or pump runs points squarely at VFD interference and cable routing. A steady offset that changes when you disconnect an earth connection points at a ground loop. Isolated spikes that break the trend for a scan or two point at transients that a spike-rejecting filter, rather than damping, is designed to handle.
Start at the grounding, because a ground loop and a bad shield are the faults that filtering can only mask, never cure. Confirm the loop has a single, deliberate ground reference and hunt for the second, accidental one. A common test is to measure the voltage between the two ground points you suspect; a real potential difference confirms a loop. Where the transmitter and the input card each try to reference the same signal to a different earth, an isolated input or a signal isolator breaks the loop cleanly and often makes the noise vanish outright. This is the point to consult the ground-loop concept page for the underlying mechanism before you start lifting earths.
Next verify the cable shield. The rule is one shield, grounded at one end only, usually at the control panel, with the far end left cut back and insulated so it cannot touch anything. A shield grounded at both ends creates its own ground loop through the shield itself, and a shield grounded at neither end floats and picks up interference. Physically confirm which end is landed, that it is a solid connection to the panel ground bar, and that no drain wire is shorting to a gland or tray somewhere along the run. Fixing shield grounding is frequently the single change that turns a fuzzy trend clean.
Then attack routing and separation. Trace the signal cable and see where it shares a tray, conduit, or bundle with VFD output cables, motor leads, contactor wiring, or other power runs, because those are the sources injecting switching noise. Separate the signal into its own tray or maintain physical spacing, cross power runs at right angles rather than running parallel, and use the shielded, twisted-pair cable the loop should have had in the first place. On a drive-fed motor the VFD output is especially aggressive, so keeping analog signals well away from it, and using the drive's recommended output filtering, does more than any amount of software smoothing.
Only after the wiring is clean should you reach for damping and filters, and it helps to know which tool fits which problem. Transmitter damping is a time constant set in the instrument that smooths continuous jitter by slowing how fast the reported value can change; it is ideal for a reading that ripples around the true value but useless against isolated spikes, which it merely spreads out. A first-order low-pass filter in the PLC does much the same job downstream and is the right choice when you cannot change the transmitter. For one-scan spikes, a median filter is the correct tool, because it rejects a single outlier sample outright while passing genuine step changes, where an average would drag the spike into the trend. The median-filter and low-pass-filter concept pages cover how each behaves so you can pick deliberately rather than stacking filters blindly.
The question of whether the filter belongs at the transmitter or at the PLC comes down to what you are protecting and what you need to keep. Damping at the transmitter cleans the signal for every consumer of that loop and is best for steady mechanical or electrical ripple that no one downstream needs to see. A filter in the PLC or SCADA layer is better when only certain tags need smoothing, when you want to keep the raw value available for diagnostics, or when you are rejecting transients that arise after the transmitter. A good rule is to apply the least filtering that gives a usable reading, because heavy damping and long filters add lag that can hide a real, fast process excursion an operator needs to catch.
This matters for cloud SCADA because a noisy analog input creates two operational problems beyond an ugly trend. It causes nuisance alarms when the jitter crosses a threshold, waking on-call staff for a value that is not really there, and it corrupts the historized data that operators later trend for troubleshooting. A cloud SCADA platform such as Merobix will faithfully record whatever the field sends, so the discipline is to fix the noise at the source and use damping or a median filter to protect alarms and history, rather than to widen alarm deadbands until the alarm is meaningless. Cleaning the signal in the field and filtering deliberately keeps the remote view trustworthy, which is the whole point of monitoring a site you are not standing at.
Fix the wiring first. A ground loop, a badly grounded shield, or a signal cable run alongside VFD cabling injects noise that a filter can only mask, and heavy filtering adds lag that can hide a real process change. Break ground loops, ground the shield at one end only, and separate the signal from power cabling, then add the least damping or median filtering needed to keep alarms and history clean.
It depends on the noise. Steady ripple that no one downstream needs to see is best smoothed with damping at the transmitter, which cleans the loop for every consumer. Isolated one-scan spikes are better handled by a median filter in the PLC or SCADA layer, because a median rejects a single outlier while passing genuine step changes, and keeping the raw value available at the PLC helps later diagnostics.
That pattern is the signature of VFD interference coupling into the loop. A variable frequency drive switches its output at high frequency, and when the analog signal cable shares a tray, conduit, or bundle with the drive output or motor leads, that switching noise couples into the signal pair. The fix is physical separation, crossing power runs at right angles, using shielded twisted-pair cable grounded at one end, and applying the drive's recommended output filtering.
Merobix reads your field devices into a cloud SCADA - the real thing behind these terms, live in days from any browser.