Designing lead and lag compensators on the root locus
Turn overshoot and settling specifications into target poles, compute the angle the loop is missing, place a lead zero and pole to supply it, then add a lag pair near the origin to raise the velocity constant.
THE SHORT ANSWER
Lead–lag design on the root locus turns overshoot and settling specifications into a target dominant pole s_d. A lead zero and pole supply the missing angle φc = −180° − ∠G(s_d) so that the locus passes through s_d, and the magnitude condition gives the gain. A lag pair placed close to the origin then multiplies the velocity constant Kv by its zero-to-pole ratio while barely moving the dominant poles.
Design a lead compensator on the root locus from overshoot and settling specifications, add a lag compensator to meet a velocity-constant requirement, and verify the result with the closed-loop poles and a simulated step response.
The previous lecture showed what a lead or lag network does to the angle and the gain of a loop. This lecture turns that into a procedure: start from written specifications and end with a compensator whose closed-loop poles and step response have been checked.
Review the previous lesson: Lead and lag compensationFirst, picture it
A root locus passes only through points where the angle condition holds. So pick the point where you want the dominant poles, measure the angles from the open-loop poles and zeros, and you know at once whether the locus goes there. If the angles add up to −246.6° instead of −180°, the point misses the locus by 66.6°, and no value of K can fix that, because K does not appear in the angle condition. A lead network is a zero with a pole farther to the left. Seen from the target point the zero makes a larger angle than the pole, so the pair adds positive angle and bends the locus toward the target. A lag network is a pole and a zero huddled near the origin. Seen from the dominant poles, far away, the two are almost on top of each other and nearly cancel. Seen from s = 0 they multiply the low-frequency gain by the ratio of the zero to the pole. Lead decides where the dominant poles sit; lag reduces the steady-state error without moving them much.
Build the idea step by step
The procedure has seven steps. First, translate the specifications into a damping ratio ζ and a real part σ = ζωn: the overshoot fixes ζ, and the 2% settling estimate ts ≈ 4/σ fixes σ. Together they give the target pole s_d = −σ + jωd. Second, check whether gain alone can reach s_d. If the uncompensated locus misses it, compute the angle deficiency φc = −180° − ∠G(s_d). Third, choose where the lead zero goes by a stated rule, and put the lead pole where the zero angle minus the pole angle equals φc. Fourth, find the gain from the magnitude condition. Fifth, find all closed-loop poles, decide whether the pair at s_d really dominates, and simulate the step response. Sixth, compute the velocity constant Kv. If it is too small, add a lag network whose zero-to-pole ratio supplies the missing factor, placed so close to the origin that it changes the angle at s_d by only a degree or less. Seventh, simulate again, because the lag leaves a slow, small tail that must stay inside the settling band.
Read the equations
Target poles. The overshoot fraction Mp fixes the damping ratio and the 2% settling estimate fixes the real part. Both formulas belong to the standard second-order system, which is why the design must later check that the pair at s_d really dominates.
Angle deficiency. The compensator must make the total angle at s_d equal to −180°. Reduce φc to the range from −180° to 180°; a value between 0° and about 60° can be supplied by one lead network.
Lead network and its angle. θz and θp are the angles of the vectors drawn from −z_c and −p_c to s_d. The zero is to the right of the pole, so θz > θp and the network adds positive angle.
Locating a zero or pole from its angle. With the zero directly below the target, θz = 90° and the pole follows from θp = 90° − φc.
Bisector method. ∠s_d is the angle of the target point measured from the positive real axis. Splitting the required angle symmetrically about the bisector gives the smallest pole-to-zero ratio p_c/z_c.
Gain from the magnitude condition, for a plant G(s) written with leading coefficients 1: the product of the distances from all poles divided by the product of the distances from all zeros.
Lag network. It equals z_l/p_l at s = 0 and is close to 1 at the target point, so it raises the error constant without moving the dominant poles much. Keep its angle at s_d to a degree or less.
Slow tail left by the lag, as a fraction of a unit step. To stay inside a 2% settling band, keep z_l − p_l below 0.02 Kv,old.
What the symbols mean
- plant transfer function in unity feedback, written with leading coefficients 1
- target location of the dominant closed-loop pole in the upper half-plane, 1/s
- damping ratio (dimensionless) and undamped natural frequency (rad/s) of the target pair
- real part magnitude ζωn and imaginary part ωn√(1−ζ²) of the target pole, 1/s and rad/s
- overshoot as a fraction of the final value, and 2% settling time in seconds
- angle the compensator must add at s_d, degrees
- lead zero and pole locations (at −z_c and −p_c, 1/s) and the compensator gain
- angles of the vectors from the lead zero and pole to s_d, measured from the positive real direction
- lag zero and pole locations, at −z_l and −p_l, 1/s
- velocity error constant, the limit of sC(s)G(s) as s → 0, 1/s; the steady error to a unit ramp is 1/Kv
These formulas place two closed-loop poles exactly at s_d, but they say nothing about the other closed-loop poles or about the zero that the compensator adds to the closed loop. The overshoot and settling formulas hold only for a pure second-order pair, so the design is finished only after every closed-loop pole has been found and the step response has been simulated.
Upper half of the s-plane for the worked example; the lower half is its mirror image. Grey: gain alone, which meets the ζ = 0.6 ray at −1.22 + j1.63. Blue: with the lead zero at −3 (circle) cancelling the plant pole there and the lead pole at −12.25, the branch passes through the target −3 + j4. The lag pair sits within 0.05 of the origin, too close to see at this scale.
Lead-lag design for G(s) = 1/(s(s+3)(s+10))
Specifications. Unity feedback, overshoot at most 15%, 2% settling time at most 1.5 s, and Kv at least 15 s⁻¹. The overshoot limit needs ζ ≥ 0.517 and the settling limit needs σ ≥ 2.67 s⁻¹. Aim at ζ = 0.6 and σ = 3 to leave room for the overshoot that a compensator zero and a lag tail add.
Gain alone. On the ζ = 0.6 ray the uncompensated locus is at −1.22 ± j1.63, with K = 43.9. The settling estimate is 4/1.22 = 3.3 s, more than twice too slow, and raising K only moves the pair toward lower damping.
Angle deficiency. At s_d = −3 + j4 the vectors from the poles at 0, −3 and −10 have angles 126.87°, 90° and 29.74°.
Zero. Method: put the lead zero directly below the target, at −σ = −3, so that it contributes exactly 90°. In this plant that point holds the plant pole at −3, so the zero also cancels it.
Pole. The pole must contribute 90° − 66.61° = 23.39°. Seen from s_d, a pole at −a makes tan θ = 4/(a − 3).
Gain. Because the zero cancels (s + 3), the magnitude condition needs only three distances: |s_d| = 5, |s_d + 10| = 8.062 and |s_d + 12.25| = 10.078.
Closed-loop poles. The characteristic polynomial factors exactly. The pole at −3 sits on the compensator zero, so it does not appear in the reference-to-output response, and the remaining real pole at −16.25 is more than five times farther left than σ = 3. The target pair dominates.
Simulate. A step simulation of the closed loop gives 8.9% overshoot and a 2% settling time of 1.26 s, close to the second-order predictions of 9.5% and 1.33 s.
Velocity constant. Kv is only 3.32, so it must rise by at least 15/3.32 = 4.52. Choose a lag ratio of 5 with the pole at 0.01 and the zero at 0.05. At s_d the lag adds only −0.37° and changes the magnitude by less than 0.5%.
Tail check. The lag creates a slow closed-loop pole near its zero. Its tail is about 0.04/3.32 = 1.2% of the step, inside the 2% band. With the zero at 0.1 and the pole at 0.02, the ratio would be the same but the tail would be 2.4%, and the simulated settling time would grow to 2.71 s.
Verify. The final closed-loop poles are −2.98 ± j3.98, −3.00, −16.24 and −0.0506. The step simulation gives 10.2% overshoot and a settling time of 1.34 s, with Kv = 16.6 s⁻¹. All three specifications are met.
Reference table
| Step | Formula | Worked example |
|---|---|---|
| 1. Target poles | Mp ≤ 15% needs ζ ≥ 0.517 and ts ≤ 1.5 s needs σ ≥ 2.67; choosing ζ = 0.6 and σ = 3 gives s_d = −3 ± j4 | |
| 2. Angle deficiency | ∠G(s_d) = −(126.87° + 90° + 29.74°) = −246.61°, so φc = 66.61° | |
| 3. Lead zero and pole | zero at −3, directly below s_d (θz = 90°); θp = 23.39° puts the pole at −12.25 | |
| 4. Gain | the zero cancels the pole at −3, so Kc = 5 × 8.062 × 10.078 = 406.25 | |
| 5. All closed-loop poles | −3 ± j4 and −16.25, plus −3, which the lead zero cancels | |
| 6. Velocity constant and lag | Kv = 3.32 must rise by at least 4.52 times; (s + 0.05)/(s + 0.01) gives Kv = 16.6 and adds only −0.37° at s_d | |
| 7. Slow tail and final poles | tail ≈ 1.2% of the step; closed-loop poles −2.98 ± j3.98, −3.00, −16.24 and −0.0506 |
The poles are roots of the characteristic polynomial. Overshoot and settling time must still be confirmed with a step simulation.
More detail and model assumptions
From specifications to a target point
An overshoot limit is a wedge around the negative real axis: poles with ζ above the limit lie inside the rays at ±cos⁻¹ζ from it. A settling limit is a vertical line: poles must lie to the left of σ = 4/ts. Any point inside both regions satisfies the second-order formulas, so the designer still has a choice. Pick a point with some room to spare. The lead zero and the lag tail both tend to add overshoot, and the settling estimate 4/σ is only approximate: in the worked example the target ζ = 0.6 gives 9.5% instead of the allowed 15%, and the final design overshoots 10.2%.
The angle deficiency follows from the angle condition. With the compensator in the loop, the point s_d is on the locus only if ∠C(s_d) + ∠G(s_d) = −180°, so the compensator must supply φc = −180° − ∠G(s_d). If φc is between 0° and about 60°, one lead network can supply it. A larger deficiency calls for two lead stages or a less demanding target. If φc is negative, the target is already reachable with less phase, and gain alone or a small lag may do.
Three ways to place the lead zero
The angle condition fixes only the difference between the zero angle and the pole angle, so one more rule is needed. Zero below the target: put the zero at −σ, directly under s_d. It contributes exactly 90°, the pole follows from θp = 90° − φc, and the arithmetic is the simplest. This lecture's worked example uses it. Bisector: draw the horizontal line from s_d toward the left and the line from s_d to the origin, bisect the angle between them, and place the zero and the pole on rays at ±φc/2 from the bisector. This gives the smallest pole-to-zero ratio, and therefore the smallest increase in high-frequency gain, for the required angle. Cancellation: put the zero on a stable real plant pole, usually the one just to the right of or under the target. The closed loop then behaves more nearly like a second-order system, as long as the plant pole is where the model says it is.
Whatever the rule, the position of a real point −a seen from s_d = −σ + jωd follows from one formula: the vector from −a to s_d has angle θ with tan θ = ωd/(a − σ), so a = σ + ωd/tan θ. An angle above 90° gives a negative tangent and a point to the right of −σ. Avoid a zero much closer to the origin than the target pair: the closed loop then gets a slow real pole near that zero, and the response creeps toward its final value.
Why the lag sits near the origin, and the slow tail it leaves
Write the lag network as (s + z_l)/(s + p_l) with p_l < z_l. At s = 0 it equals z_l/p_l, so it multiplies Kv by that ratio. At the target point, if both numbers are tiny compared with |s_d|, it is close to 1∠0° and the dominant poles barely move. A common choice is to fix the pole at a small number such as 0.01 and put the zero at the required ratio times that. Check the angle it adds at s_d; a degree or less keeps the lead design intact.
The lag also adds a closed-loop pole near −z_l, and it is very slow. Near the origin the lead-compensated loop looks like Kv,old/s. Adding the lag and solving the resulting second-order error equation gives a step-response tail of about (z_l − p_l)/Kv,old that decays roughly like e^(−z_l t). For the tail to stay inside a 2% band, keep z_l − p_l below 0.02 Kv,old. In the worked example the tail is 1.2% with the zero at 0.05; with the zero at 0.1 and the same ratio it would be 2.4%, and the simulated settling time jumps from 1.34 s to 2.71 s. A PI controller is the limit of a lag network whose pole has moved to the origin: it makes Kv infinite for a type-1 plant, at the price of the same kind of slow pole-zero pair.
Lead first, then lag
When both the transient and the steady-state specifications fail, design the lead part first, because it decides where the dominant poles sit and therefore what Kv the loop has before the lag is added. Then compute Kv of the lead-compensated loop, take the ratio to the required value, and design the lag for that ratio. The combined compensator Kc(s + z_c)(s + z_l)/((s + p_c)(s + p_l)) is one network with two zeros and two poles, often called a lag-lead or lead-lag compensator. If the lag shifts the dominant poles noticeably, retune Kc on the ζ ray before simulating. In the worked example the pair only slides 0.6% toward the origin along the ζ = 0.6 ray, and retuning would change Kc by less than 0.01%, so the step is skipped.
Connect the example to the lab
The motor lab's lag lesson uses C(s) = 0.25(s + 8)/(s + 0.5). Its gain at s = 0 is 0.25 × 8/0.5 = 4, sixteen times its high-frequency gain of 0.25. That ratio z/p is exactly the factor by which a lag network multiplies the error constant. Run the lesson and pin the result, then raise only the pole to 1 rad/s and to 2 rad/s, which lowers the gain at s = 0 to 2 and to 1. In the simulator the final speed error grows from about 2.5 rad/s to 4.8 and 9.3 rad/s, while the overshoot falls from 17.5% to 11.8% and 2.6%. The lab's own root-locus view sweeps a proportional gain only, so compare the time responses here. The first part of each run reaches the ±12 V limit, which no linear design accounts for.
Change the lag network's low-frequency gain on the motorMake a prediction before moving a control.
- PREDICTIn the lag lesson the zero is at 8 rad/s and the gain is 0.25. Before running anything, predict how the final speed error changes when the pole moves from 0.5 to 1 and then to 2 rad/s.
- CHANGE ONE THINGChange only the lag pole: 0.5, then 1, then 2 rad/s. Pin each run.
- OBSERVERead the final error and the overshoot of the pinned runs, and note when the applied voltage leaves the ±12 V limit.
- EXPLAINExplain why the final error roughly doubles each time the gain at s = 0 halves, and why the overshoot falls at the same time. Relate both to the lag's zero-to-pole ratio.
Do not trust ζ alone once a compensator is in the loop. The lead zero becomes a closed-loop zero, and a real closed-loop pole often lands near it. If that pole is slower than the target pair and not cancelled, the response is slower and less oscillatory than predicted (exercise 2); if the zero lies close to the pair, the overshoot grows (exercise 5). Cancelling a plant pole with the lead zero is safe only for a well-damped pole you know accurately: the cancelled mode still appears in the response to a load disturbance at the plant input.
Check your understanding+
A lead compensator must supply 75° at the target point. Why might a designer use two lead stages of 37.5° each instead of one?
A single network whose largest phase is 75° needs a pole-to-zero ratio of (1 + sin 75°)/(1 − sin 75°) ≈ 58, and its high-frequency gain rises by the same factor. Two 37.5° stages need about 4.1 each, 16.9 in total, so sensor noise is amplified far less for the same angle.Solve these before opening the solutions
Work each problem on paper first. Open the answer to check yours, then the full solution if you need the steps, and mark how it went.
Unity feedback with the plant below. The dominant closed-loop poles should have ζ = 0.707 and a 2% settling time of about 2 s. (a) Find the target poles. (b) Find the gain K for which the uncompensated loop has ζ = 0.707, and estimate its settling time. (c) Find the angle a compensator must add at the target.
Show answer
Answer(a) s_d = −2 ± j2. (b) K = 25.8 with poles at −1.23 ± j1.23 and a settling time of about 3.3 s. (c) φc = 36.87°.
Show full solution
Solution
Target poles. ζ = 0.707 puts the poles on the 135° ray, where the real and imaginary parts are equal. The settling estimate gives σ = 4/2 = 2.
Uncompensated gain. Put s = r∠135° into s³ + 11s² + 24s + K = 0. With s² = −jr² and s³ = r³(1 + j)/√2, the imaginary part gives r and the real part gives K; the larger root of the quadratic, r = 13.8, gives a negative K and is rejected.
The pair is at −1.228 ± j1.228, so the settling estimate is 3.3 s (a simulation gives 3.55 s). Gain alone cannot reach 2 s.
Angle deficiency. From the poles at 0, −3 and −8, the vectors to −2 + j2 have angles 135°, 63.43° and 18.43°.
Continue exercise 1. Put the lead zero directly below the target pole, at s = −2. Find the lead pole and the gain, then all four closed-loop poles. Can you trust the second-order prediction of 4.3% overshoot and a 2 s settling time?
Show answer
Answerp_c = 3.5 and K = 50. The closed-loop poles are −2 ± j2, −1.37 and −9.13. The real pole at −1.37 is slower than the target pair and is not cancelled, so the pair does not dominate: a simulation shows no overshoot at all and a settling time of 2.17 s.
Show full solution
Solution
The zero contributes 90°, so the pole must contribute 90° − 36.87° = 53.13°.
Magnitude condition, with the distances from 0, −3, −8 and −3.5 over the distance from the zero at −2.
The characteristic polynomial contains the factor s² + 4s + 8 of the target pair. Divide it out to find the other two poles.
Judge dominance. The pole at −1.37 lies to the right of the pair and is only partly offset by the zero at −2. It is the slowest mode, so the step response rises without overshoot and settles in 2.17 s in a simulation, not with the 4.3% the pair alone would give.
Same plant and target as exercise 1. This time let the lead zero cancel the plant pole at −3. Find the lead pole, the gain, the closed-loop poles and Kv.
Show answer
Answerp_c = 6 and K = 80. The closed-loop poles are −2 ± j2, −3 (on the compensator zero) and −10, and Kv = 1.67 s⁻¹. A simulation gives 4.1% overshoot and a 2.21 s settling time.
Show full solution
Solution
With the zero on −3, its angle cancels the angle of the plant pole at −3. The remaining poles at 0 and −8 give 135° and 18.43°, so the new pole must give the rest of the 180°.
Magnitude condition with the three remaining distances.
Closed loop. The cancelled factor stays as a closed-loop pole at −3, and the rest factors exactly.
Velocity constant and check. The third pole at −10 is five times farther left than the pair. A step simulation gives 4.1% overshoot and 2.21 s; the settling time exceeds 4/σ = 2 s because 4/σ is only an estimate, so aim slightly further left if 2 s is a hard limit.
Raise the velocity constant of the design in exercise 3 to Kv = 10 s⁻¹ with a lag network, keeping the slow tail inside a 2% band. First try the pole at 0.01 and check the tail; then choose a zero that keeps the tail below 2%.
Show answer
AnswerA ratio of 6 is needed. With the pole at 0.01 and the zero at 0.06 the tail is about 3.0%, and the simulated settling time grows to 7.97 s. The tail condition requires z_l < 0.040; the zero at 0.03 with the pole at 0.005 gives a tail of 1.5%, Kv = 10, 5.6% overshoot and a 2.60 s settling time.
Show full solution
Solution
Required ratio.
First try: pole 0.01, zero 0.06. Its angle at s_d is only −0.73°, so the root locus near the target barely moves, but the tail estimate is too large.
Simulation of the first try: closed-loop poles −1.98 ± j1.96, −3, −9.99 and −0.062. The output overshoots 7.1% and creeps back so slowly that the 2% settling time is 7.97 s.
Second try. The tail condition z_l(1 − 1/6) < 0.02 × 1.67 gives z_l < 0.040. Take z_l = 0.03 and p_l = 0.005.
Simulation of the second try: poles −1.99 ± j1.98, −3, −10.00 and −0.0305; 5.6% overshoot, 2.60 s settling time and Kv = 10. The price is a lag pole with a 200 s time constant, which in hardware needs an implementation that does not drift.
For the plant below, place the dominant closed-loop poles at −3 ± j3 with a lead compensator designed by the bisector method. Then compare it with the zero placed directly below the target: which design needs the smaller pole-to-zero ratio, and which gives the smaller overshoot?
Show answer
Answerφc = 63.43°. Bisector: zero at 2.51, pole at 7.16 (ratio 2.85), K = 22.6, third pole −3.16, simulated overshoot 9.4%. Zero below: zero at 3, pole at 9 (ratio 3.0), K = 30, third pole −5, simulated overshoot 12.2%. The bisector design has the smaller ratio and here also the smaller overshoot; the zero-below design has the larger Kv, 5.0 against 3.97.
Show full solution
Solution
Deficiency. From the poles at 0 and −2, the vectors to −3 + j3 have angles 135° and 108.43°.
Bisector. The target's angle is 135°, so the bisector makes 67.5° with the real axis, and the zero and pole sit at ±31.72° from it: θz = 99.22° and θp = 35.78°.
Gain and third pole. The closed-loop poles add up to −(2 + 7.16), which fixes the third one.
Zero below the target. θz = 90°, so θp = 26.57° and p_c = 3 + 3/tan 26.57° = 9. The characteristic polynomial factors exactly.
Compare. Ratios 2.85 against 3.0; Kv = K z_c/(2p_c) gives 3.97 against 5.0; simulated overshoots 9.4% against 12.2%, with settling times 1.49 s and 1.40 s. Both overshoots exceed the 4.3% that ζ = 0.707 predicts, because each design leaves a closed-loop zero near the dominant pair.
Mark each exercise after you check your answer. Marks stay in this browser.
Terms this lecture teaches
- lag compensator지상 보상기
- compensator보상기
- lead compensator진상 보상기
- angle condition각도 조건
- dominant poles지배극점
- magnitude condition크기 조건
- pole–zero cancellation극점·영점 상쇄
- static error constants오차 상수
Frequently asked questions
How do I find the angle a lead compensator must add?
Add the angles of the vectors from every open-loop zero to the target s_d and subtract those from every pole; the deficiency is φc = −180° − ∠G(s_d). For G = 1/(s(s + 3)(s + 10)) and s_d = −3 + j4 the pole angles are 126.87°, 90° and 29.74°, so φc = 66.61°.
Why can't gain alone put the closed-loop poles at the target point?
K appears only in the magnitude condition, not in the angle condition, so changing it moves the poles along the existing locus. If the angles at s_d do not add up to −180°, no gain reaches that point; only added zeros and poles change the shape of the locus.
Where should the lead zero go?
The angle condition fixes only the difference between the zero and pole angles, so one more rule is needed: the zero directly below the target for the simplest arithmetic, the bisector method for the smallest pole-to-zero ratio, or cancellation of a stable plant pole that is known accurately. Avoid a zero much closer to the origin than the target pair, which leaves a slow closed-loop pole near it.
Why is the lag compensator placed so close to the origin?
At s = 0 the lag multiplies Kv by z_l/p_l, but seen from the distant dominant poles its zero and pole almost coincide; in the worked design (s + 0.05)/(s + 0.01) adds only −0.37° at s_d. It does leave a slow tail of about (z_l − p_l)/Kv,old of the step, so keep z_l − p_l below 0.02 Kv,old for a 2% settling band.
Why must I still simulate after placing the dominant poles?
The design formulas put two closed-loop poles exactly at s_d but say nothing about the other poles or about the closed-loop zero the compensator adds. The overshoot and settling formulas hold only for a pure second-order pair, so find every closed-loop pole and simulate the step response before calling the design finished.
Why the next lesson follows
The same two networks can also be designed on the Bode plot, where the specifications are an error constant and a phase margin instead of target poles. The next lecture repeats the design in the frequency domain.
Further reading
Finished the reading and the exercises? Mark the lecture as done to tick it on the syllabus. Progress stays in this browser.
EIGENROOM · https://eigenroom.com/theory/lead-lag-root-locus · Updated 2026-11-15