How it works
A great deal of control design ends up being judged by one picture: the response to a step. How far does the output overshoot, how quickly does it get there, and how long until it stays within a small band around the final value? For the standard second-order system T(s) = ωn²/(s² + 2ζωn·s + ωn²) all of these follow from just two numbers, the damping ratio ζ and the natural frequency ωn, which is why the second-order model is the common reference for loop specifications.
Overshoot depends on ζ alone: exp(−ζπ/√(1 − ζ²)). Peak time is half a period of the damped oscillation, π/ωd with ωd = ωn√(1 − ζ²). These two are exact closed forms. Rise time and settling time are not. The textbook settling formula 4/(ζωn) only says when the exponential envelope enters the 2 % band, and the actual response may have entered earlier, between two peaks. This calculator therefore finds the 10 % and 90 % crossings and the last exit from the band directly from the exact response, and shows the envelope estimate next to it. The exact settling time jumps as ζ changes, because one peak of the oscillation after another moves inside the band. Between ζ = 0.3 and 0.8 it is up to about 25 % shorter than 4/(ζωn), and at critical damping the envelope formula is about 30 % too short, since the response there is (1 + ωn·t)·e^(−ωn·t) rather than a plain exponential.
Power-electronics loops are usually designed in the frequency domain, with a crossover frequency and a phase margin read from a Bode plot. The link to the step response comes through the loop L(s) = ωn²/(s(s + 2ζωn)), an integrator with one more pole, which closes into exactly the second-order T(s) above. For that loop the phase margin and the crossover frequency have closed forms in ζ, so a phase margin can be turned into a damping ratio and an expected overshoot, and a crossover frequency into ωn. A 45° margin gives about ζ = 0.42 and 23 % overshoot; 60° gives ζ = 0.61 and about 9 %.
The same model gives the closed-loop bandwidth, the frequency at which |T| falls to −3 dB, and for ζ below 0.707 the height of the resonant peak in |T|. For damping ratios between 0.3 and 1 the closed-loop bandwidth comes out between 1.3 and 1.6 times the crossover frequency.
Design procedure
- 1Decide what the step response should look like: the overshoot you can accept and the settling time you need.
- 2Convert the overshoot to a damping ratio and a phase margin with the first input, and add 5 to 10° for the poles and zeros the second-order model leaves out.
- 3Choose the crossover frequency from the settling time you need, using the speed input and the band you care about.
- 4Design the compensator for that crossover and margin, for example with the Type II/III compensator calculator.
- 5Measure a reference step or a load step and compare overshoot and settling with the expectation here; a large mismatch usually points to a pole or zero the design did not account for.
Formulas
- Overshoot and peak time
- M_p = exp(−ζπ/√(1 − ζ²)), t_p = π/(ωn·√(1 − ζ²))
- Damping from overshoot
- ζ = −ln M_p / √(π² + ln² M_p)
- Loop crossover and phase margin
- ωc = ωn·√(√(1 + 4ζ⁴) − 2ζ²), PM = atan(2ζ / √(√(1 + 4ζ⁴) − 2ζ²))
- Closed-loop bandwidth
- ω_BW = ωn·√(1 − 2ζ² + √(4ζ⁴ − 4ζ² + 2))
- Resonant peak (ζ < 0.707)
- M_r = 1/(2ζ·√(1 − ζ²))
- Settling time
- exact: last time |y(t) − 1| > band; envelope estimate t_s ≈ −ln(band)/(ζωn) ≈ 4/(ζωn) for 2 %
Worked example
The defaults describe a converter voltage loop with a 45° phase margin and a 5 kHz crossover. That margin corresponds to ζ = 0.420, and the equivalent second-order system has its natural frequency at 5.95 kHz. A reference step then overshoots by 23.3 %, peaking 92.7 µs after the step, with a 10–90 % rise time of 40.0 µs. The response stays within ±2 % after 225 µs, while the envelope formula 4/(ζωn) would have said 249 µs. The closed-loop bandwidth is 8.06 kHz, 1.6 times the crossover, and |T| has a resonant peak of 1.31, or 2.3 dB. Lowering the margin to 30° raises the overshoot to 41.6 % and the resonant peak to 5.7 dB; raising it to 60° brings the overshoot down to 8.8 %, and at 70° it is 1.4 %. The time and frequency values here were checked against an independent Python program that integrates the differential equation numerically and solves the frequency responses by bisection.
Common mistakes
- Expecting a converter loop with 45° margin to overshoot exactly 23 %; the second-order relation is an approximation for real loops.
- Quoting 4/(ζωn) as the settling time; it is an envelope estimate and can be off by tens of percent.
- Mixing f_n, f_d and f_c. They are different frequencies and differ by 10 % or more at typical damping.
- Judging the loop from a load step alone. The load-step response depends on the output impedance and is not the T(s) step response shown here.
- Reading phase margin from a plot that does not include the modulator gain, sampling delay or sensor filter.
Frequently asked questions
Is the 'phase margin ≈ 100·ζ' rule exact?
No, it is a straight-line fit to the exact curve shown in the second chart. It is close for ζ up to about 0.6, which covers phase margins up to roughly 60°. Above that the exact relation flattens, and a 76° margin corresponds to ζ = 1 rather than 0.76.
My converter loop has a 45° margin but overshoots much less than 23 %. Why?
Because a real loop is not exactly the second-order form this page assumes. Compensator zeros, the output-filter double pole, the ESR zero and right-half-plane zeros all change the step response for the same margin. The relation is a useful expectation and a way to compare designs, not a prediction to the percent. A load step also excites the loop through the output impedance rather than the reference path, which gives a different waveform.
Why is the exact settling time different from 4/(ζωn)?
4/(ζωn) is when the decaying envelope e^(−ζωn·t)/√(1 − ζ²) reaches 2 %, rounded. The response itself oscillates inside the envelope and can already be in the band when the envelope is not, or leave the band at a peak just before the envelope gets there. The calculator scans the exact response and refines the last band exit by bisection, so it reports the time after which the output really stays inside.
Should I enter the crossover frequency or the natural frequency?
Enter whichever you have. From a Bode plot of the loop gain you know f_c; from a measured ringing frequency you can estimate f_d, which is close to f_n for light damping. The page converts between them for the chosen ζ; for typical margins the crossover lies 15 to 35 % below f_n.
How were the numbers checked?
A separate Python program integrates y'' + 2ζωn·y' + ωn²·y = ωn² with a fourth-order Runge–Kutta method and measures overshoot, peak, rise and settling times from the samples, and finds crossover, phase margin and bandwidth by bisection on the complex frequency responses. For ζ from 0.2 to 2 all values agree within 0.5 %, the closed-form quantities within 0.001 %. The ζ = 0.5 overshoot of 16.3 % matches the value tabulated by Ogata.
References
- K. Ogata, "Modern Control Engineering", chapter on transient-response analysis (definitions of rise, peak and settling time, overshoot as a function of ζ) and the frequency-response chapter (resonant peak and bandwidth of the second-order system).
- N. S. Nise, "Control Systems Engineering", chapters on time response and on design via frequency response (the phase-margin relation for the loop ωn²/(s(s + 2ζωn))).
- The values on this page were recomputed independently in Python by numerical integration and bisection on the frequency responses.
Results are engineering estimates. Verify your design with datasheets, simulation and measurement.