Feedforward Control Estimator

Compute the feedforward gain that cancels a measurable disturbance before it reaches the output, and see the peak-error reduction over feedback control alone.

➕ Feedforward Control Estimator
Kp
τp s
Kd
τd s
Feedback tuning time constant (λ)2
s
0.120
Peak error reduction
—
Feedforward gain Kff
—
Feedback Kc
—
Feedback Ti
—
Peak error, feedback only
—
Peak error, with feedforward
—
Step-by-step working

➕ What is Feedforward Control?

Feedforward control is a technique that measures a disturbance directly and applies a corrective action to cancel its effect on the output before that effect ever appears, rather than waiting for feedback control to notice an error after the disturbance has already thrown the process off. It is layered on top of ordinary feedback control, never used as a replacement for it, since feedforward can only compensate for the specific disturbance it measures and can never correct for anything else, including its own model errors.

Feedforward shows up wherever a disturbance is both measurable and disruptive enough that waiting for feedback to react is too slow. A distillation column with a measurable feed-flow disturbance uses feedforward to adjust reboiler duty before the composition drifts. A paper machine measures incoming stock consistency and feeds forward a correction to the dilution valve before basis weight varies. An HVAC system measures outdoor temperature and feeds forward a correction to heating output before indoor temperature has time to sag.

A common misconception is that feedforward control is a form of prediction or forecasting. It is not, it reacts to a disturbance the instant it is measured, using a known model of how that disturbance and the manipulated variable each separately affect the output, it does not anticipate disturbances before they occur. Another misconception is that a well-designed feedforward compensator eliminates the need for feedback entirely, in practice no process model is perfect, so feedback remains essential to correct whatever the feedforward path's model mismatch leaves behind.

This calculator computes the ideal feedforward gain for simple first-order process and disturbance models, along with a feedback-only PI baseline, then runs a time-domain simulation of a unit disturbance step to show exactly how much peak output error feedforward removes compared to feedback control acting alone.

📐 Formula

Gff(s)  =  −Gd(s)/Gp(s)  =  −(Kd/Kp)×(τps+1)/(τds+1)
Gp(s) = Kp/(τps+1), how the manipulated variable affects the output
Gd(s) = Kd/(τds+1), how the measured disturbance affects the output
Kff = −Kd/Kp, the feedforward compensator's steady-state gain; τp is its lead time constant, τd is its lag time constant
Feedback baseline (direct-synthesis PI on the process alone): Kc = τp/(Kpλ), Ti = τp
Example: Kp=2, τp=5 s, Kd=1.5, τd=8 s → Kff = −1.5/2 = −0.7500

📖 How to Use This Calculator

Steps

1
Enter the process model. Type Kp and taup, the gain and time constant of how the manipulated variable affects the output.
2
Enter the disturbance model. Type Kd and taud, the gain and time constant of how the measured disturbance affects the output.
3
Read the feedforward gain and peak-error reduction. See the feedforward gain Kff, the feedback-only baseline gains, and how much feedforward reduces peak output error for a unit disturbance step.

💡 Example Calculations

Example 1 — Disturbance Slower Than the Process

Kp=2, τp=5 s, Kd=1.5, τd=8 s, λ=2 s

1
Feedforward gain: Kff = −1.5/2 = −0.7500
2
Feedback baseline: Kc = 5/(2×2) = 1.2500, Ti = 5.0000 s
3
Peak error to a unit disturbance step: feedback only = 0.2363, with feedforward = 0.000000
Peak error reduction = 99.99% (near-total cancellation under a perfectly matched model)
Try this example →

Example 2 — Large Disturbance Gain, Slow Process

Kp=1, τp=10 s, Kd=3, τd=4 s, λ=3 s

1
Feedforward gain: Kff = −3/1 = −3.0000
2
Feedback baseline: Kc = 10/(1×3) = 3.3333, Ti = 10.0000 s
3
Peak error to a unit disturbance step: feedback only = 0.9520, with feedforward = 0.000000
Peak error reduction = 99.99%
Try this example →

Example 3 — Matched Process and Disturbance Time Constants

Kp=0.5, τp=6 s, Kd=0.8, τd=6 s, λ=1.5 s

1
Feedforward gain: Kff = −0.8/0.5 = −1.6000
2
Feedback baseline: Kc = 6/(0.5×1.5) = 8.0000, Ti = 6.0000 s
3
Peak error to a unit disturbance step: feedback only = 0.1261, with feedforward = 0.000000
Peak error reduction = 99.99%
Try this example →

❓ Frequently Asked Questions

What is feedforward control?+
Feedforward control measures a disturbance directly and applies a corrective action to the manipulated variable before that disturbance has a chance to affect the controlled output, instead of waiting for feedback to detect an error after the fact. It is added on top of, not instead of, ordinary feedback control.
What is the formula for the feedforward gain?+
The ideal feedforward compensator is Gff(s) = -Gd(s)/Gp(s), the negative of the disturbance transfer function divided by the process transfer function. For simple first-order Gd(s)=Kd/(taud*s+1) and Gp(s)=Kp/(taup*s+1), this works out to a lead-lag filter with static gain Kff=-Kd/Kp, lead time constant taup, and lag time constant taud.
Why is the feedforward gain usually negative?+
Because feedforward must move the manipulated variable in the opposite direction to whatever effect the disturbance has on the output, to cancel it. If both the disturbance gain Kd and the process gain Kp are positive (increasing either one increases the output), the feedforward gain -Kd/Kp comes out negative by construction.
Why isn't the feedforward gain just a constant multiplier?+
Because the disturbance and the manipulated variable generally reach the output through different dynamics (different time constants), matching only the steady-state gain would cancel the disturbance's final effect but leave a large, uncancelled transient error while the two paths settle at different rates. The lead-lag filter (taup*s+1)/(taud*s+1) matches the dynamic shape, not just the final value.
How much does feedforward actually reduce peak error compared to feedback alone?+
With a perfectly matched process and disturbance model and no dead time, feedforward can reduce peak error by close to 100%, essentially canceling the disturbance's effect on the output before feedback ever needs to react. Real installations see a smaller, still often substantial, reduction because process models are never perfectly known.
What feedback controller does this calculator use as the baseline?+
A direct-synthesis PI controller tuned on the process alone: Kc = taup/(Kp*lambda), Ti = taup, where lambda is the desired closed-loop time constant. This is the same exact pole-cancellation PI design used for a delay-free first-order process, giving a clean, well-defined feedback-only baseline to compare feedforward's improvement against.
Does feedforward control require dead time compensation like a Smith Predictor?+
Not in this simplified first-order scope, but in general, yes, if the process's dead time exceeds the disturbance path's dead time, the ideal feedforward compensator would require predicting the future, which is not physically realizable. This calculator deliberately scopes to simple first-order process and disturbance models without dead time to keep the feedforward compensator always realizable.
Can feedforward control be used without feedback control?+
It is possible but strongly discouraged in practice. Feedforward alone has no way to correct for model error, sensor drift, or any disturbance it cannot measure, so a feedforward-only scheme drifts uncorrected over time. Virtually all real feedforward implementations run alongside a feedback loop that handles everything feedforward's model does not capture.
What is the difference between feedforward control and cascade control?+
Feedforward control cancels a specific, measured disturbance before it affects the output at all. Cascade control (see the Cascade Control Loop Tuning Calculator on this site) instead speeds up how quickly a nested secondary loop rejects disturbances entering inside it, regardless of whether those disturbances are directly measured. The two techniques solve different problems and are frequently combined.
What units does this calculator use?+
Process gain Kp and disturbance gain Kd are entered as plain dimensionless ratios (in matching input/output units), process time constant taup, disturbance time constant taud, and the feedback tuning parameter lambda are all entered in seconds. The peak error values are in the same units as the process output.
Tags: Feedforward Control Calculator Feedforward Gain Calculator Disturbance Rejection Calculator Feedforward Feedback Control Control Systems Calculator Calculator Formula Free Online

🔗 Related Calculators

🔁
Cascade Control Loop Tuning Calculator
Engineering - Control Systems
⏳
Dead Time and Smith Predictor Calculator
Engineering - Control Systems
🎯
PID Tuning Calculator (IMC)
Engineering - Control Systems
📉
Steady-State Error Calculator
Engineering - Control Systems
⏱️
Settling Time and Rise Time Calculator
Engineering - Control Systems

What is feedforward control?

Feedforward control measures a disturbance directly and applies a corrective action to the manipulated variable before that disturbance has a chance to affect the controlled output, instead of waiting for feedback to detect an error after the fact. It is added on top of, not instead of, ordinary feedback control.

What is the formula for the feedforward gain?

The ideal feedforward compensator is Gff(s) = -Gd(s)/Gp(s), the negative of the disturbance transfer function divided by the process transfer function. For simple first-order Gd(s)=Kd/(taud*s+1) and Gp(s)=Kp/(taup*s+1), this works out to a lead-lag filter with static gain Kff=-Kd/Kp, lead time constant taup, and lag time constant taud.

Why is the feedforward gain usually negative?

Because feedforward must move the manipulated variable in the opposite direction to whatever effect the disturbance has on the output, to cancel it. If both the disturbance gain Kd and the process gain Kp are positive (increasing either one increases the output), the feedforward gain -Kd/Kp comes out negative by construction.

Why isn't the feedforward gain just a constant multiplier?

Because the disturbance and the manipulated variable generally reach the output through different dynamics (different time constants), matching only the steady-state gain would cancel the disturbance's final effect but leave a large, uncancelled transient error while the two paths settle at different rates. The lead-lag filter (taup*s+1)/(taud*s+1) matches the dynamic shape, not just the final value.

How much does feedforward actually reduce peak error compared to feedback alone?

With a perfectly matched process and disturbance model and no dead time, feedforward can reduce peak error by close to 100%, essentially canceling the disturbance's effect on the output before feedback ever needs to react. Real installations see a smaller, still often substantial, reduction because process models are never perfectly known.

What feedback controller does this calculator use as the baseline?

A direct-synthesis PI controller tuned on the process alone: Kc = taup/(Kp*lambda), Ti = taup, where lambda is the desired closed-loop time constant. This is the same exact pole-cancellation PI design used for a delay-free first-order process, giving a clean, well-defined feedback-only baseline to compare feedforward's improvement against.

Does feedforward control require dead time compensation like a Smith Predictor?

Not in this simplified first-order scope, but in general, yes, if the process's dead time exceeds the disturbance path's dead time, the ideal feedforward compensator would require predicting the future, which is not physically realizable. This calculator deliberately scopes to simple first-order process and disturbance models without dead time to keep the feedforward compensator always realizable.

Can feedforward control be used without feedback control?

It is possible but strongly discouraged in practice. Feedforward alone has no way to correct for model error, sensor drift, or any disturbance it cannot measure, so a feedforward-only scheme drifts uncorrected over time. Virtually all real feedforward implementations run alongside a feedback loop that handles everything feedforward's model does not capture.

What is the difference between feedforward control and cascade control?

Feedforward control cancels a specific, measured disturbance before it affects the output at all. Cascade control (see the Cascade Control Loop Tuning Calculator on this site) instead speeds up how quickly a nested secondary loop rejects disturbances entering inside it, regardless of whether those disturbances are directly measured. The two techniques solve different problems and are frequently combined.

What units does this calculator use?

Process gain Kp and disturbance gain Kd are entered as plain dimensionless ratios (in matching input/output units), process time constant taup, disturbance time constant taud, and the feedback tuning parameter lambda are all entered in seconds. The peak error values are in the same units as the process output.

📌 Quick Tips

💡Feedforward control needs a measurable disturbance, it cannot compensate for a disturbance the control system cannot see. Feedback control, by contrast, reacts to any disturbance's effect on the output, measured or not, but only after it has already shown up as an error.
💡The feedforward gain's sign is opposite to the ratio of disturbance gain to process gain, if both Kd and Kp are positive, Gff is negative. That negative sign is exactly what drives the manipulated variable the opposite way the disturbance would otherwise push the output.
💡Feedforward is never used alone in practice, it is always paired with feedback, since no process model is perfect. Feedback cleans up whatever the feedforward path's model mismatch leaves behind.
💡This calculator's near-total peak-error reduction assumes a perfectly matched process and disturbance model, with no additional dead time on either path. Real installations typically see partial, not complete, disturbance rejection because of model uncertainty, measurement noise, and unmodeled dynamics.
💡The feedforward compensator here is a lead-lag filter, not just a static gain, matching the dynamic shape of the disturbance path to the process path is what lets it cancel the transient, not just the final steady-state effect.