How to Choose Deadband Values by Signal Type

Back to Blog

Measure the Real Noise Band First

Before you assign any number, look at what the signal actually does when the process is steady. Trend the raw value at its fastest scan for a few minutes during stable operation and read the peak-to-peak spread. That spread is the noise floor, and it is the only honest basis for a deadband: set the band below it and you report noise forever; set it just above it and noise stays silent while a real move of even one band gets through. A deadband is a filter, and you cannot size a filter without knowing what you are filtering.

Distinguish measurement noise from real process variation, because they call for opposite responses. Electrical pickup, turbulence at a flow element, and transmitter resolution are noise you want to suppress. A level that genuinely hunts by half an inch because of surface waves is process, and if it matters to the operator you keep it. The deadband should reject the former and pass the latter, which is why the number is a per-point judgment and not a plant-wide constant. The underlying mechanism is the same deadband concept used in control, applied here to reporting rather than actuation.

Express the band in the same units and reference as the point. A deadband can be an absolute value in engineering units or a percentage of span, and the two are not interchangeable across a mixed instrument list. Percentage of span is convenient for templating but punishes a wide-range transmitter operating near the bottom of its range, where a small percent is a large absolute move. When in doubt, set the band in engineering units tied to the physics you measured, not a tidy percentage.

Set the Band Per Signal Family

Flow signals, especially those derived from a differential-pressure element with square-root extraction, are the noisiest points on most sites and need the widest deadbands. The square-root function amplifies noise badly at low flow, so a band that looks generous in flow units is often modest once you account for the DP noise it came from. Level in a large, calm vessel is the opposite: it moves slowly and smoothly, so a tight band captures the real trend without chatter. Pressure sits in between and depends entirely on the process - a gas header is jumpy, a pump discharge on a steady duty is calm.

Temperature is usually the easiest case because thermal mass filters the signal for you; a real temperature rarely steps, so a small deadband rejects the little electrical wiggle while still catching every genuine ramp. The table below gives the relative ranking, not fixed numbers, because the actual value always comes from the noise you measured on that specific point:

Signal type	Typical noise character	Relative deadband
Flow (DP, square-root)	High, worst at low flow	Widest
Pressure (gas header)	Jumpy, process-driven	Wide
Pressure (steady liquid)	Moderate	Medium
Level (large calm vessel)	Slow and smooth	Small
Temperature	Filtered by thermal mass	Smallest

Apply the band per point, not per template, once you leave the easy cases. A template gets you a sensible starting deadband for a family, but the noisy transmitter on a vibrating skid and the quiet one in a calm sump both inherit the same template value and one of them will be wrong. On protocols that carry per-point event thresholds this is exactly the discipline behind DNP3 analog deadbands, where each point earns its own number.

Verifying the Deadband Does Its Job

Prove the band with two tests. First, watch the point at steady state after you set the deadband: it should report essentially nothing, because a correctly sized band swallows the noise floor. If it is still generating a steady stream of updates on a calm process, the band is below the noise and needs widening. Second, force a known small move - bump a setpoint or crack a valve - and confirm the change reports promptly; if a real move of a band or two vanishes, the deadband is too wide and you are losing process fidelity.

Confirm the reporting change downstream, not just at the device. A deadband that suppresses updates at the field end should show up as a cleaner, less dense trend and a lighter data rate in the SCADA system. This is the same mechanism that drives report by exception: fewer updates for the same information. If the field says it is filtering but the historian still stores every scan, the deadband is being applied in the wrong place in the chain and is buying you nothing.

Common Mistakes to Avoid

The classic mistake is one deadband for the whole plant, usually a round percentage someone picked once. It is guaranteed to be too tight for the flows and too loose for the temperatures at the same time, so you get both problems - flooded telemetry and lost detail - from a single setting. Size the band from measured noise per point, or at least per signal family, and the two failure modes both disappear.

The second mistake is confusing a reporting deadband with a control deadband. A reporting deadband decides what data leaves the field or lands in the historian; a control deadband decides when an output moves. Setting one when you meant the other either makes a loop sluggish or floods a database. The third is ignoring square-root amplification on DP flow: a deadband set on the linearized flow value behaves very differently at low flow than at high, so verify it across the operating range, not just at normal load.

Frequently Asked Questions