Second-Order System: ωn, ζ (Damping Ratio), ωd

Why second-order systems matter in control design

A first-order system has one energy-storage element and one pole; a second-order system has two, and that second pole is what makes overshoot and ringing possible at all. Most real control loops — a motor with inertia and a current loop, a buck converter with an inductor and output capacitor, a thermal plant with a sensor time lag stacked on a thermal mass — behave like second-order systems either natively or because a dominant-pole approximation reduces their higher-order dynamics down to two poles that matter.

How poles move from first-order to second-order behavior

A first-order system's characteristic equation is τs + 1 = 0, giving a single real pole at s = -1/τ and a response that can only decay monotonically — no overshoot is mathematically possible. Add a second energy-storage element and the characteristic equation becomes quadratic: s² + 2ζωn·s + ωn² = 0. The roots of that quadratic can now be a complex-conjugate pair, and a complex pole pair is exactly what produces oscillation in the time domain. Whether those roots are real or complex, and how far they sit from the imaginary axis, is governed entirely by two numbers: natural frequency ωn and damping ratio ζ.

Real-world systems requiring second-order models

A DC motor driving an inertial load with viscous friction is a textbook second-order plant: armature inductance and inertia are the two energy stores. A switching converter's LC output filter is second-order by construction. A mechanical structure — an antenna gimbal, a 3D-printer gantry — behaves as a mass-spring-damper, which is the mechanical archetype the electrical model is borrowed from. Even plants that are genuinely fourth- or fifth-order are routinely approximated as second-order for controller design, because one dominant pole pair usually sets the response shape while the rest are far enough left in the s-plane to be ignored.

Connection to feedback loop dynamics

Closing a feedback loop around almost any plant produces a closed-loop transfer function whose denominator can be forced into the standard form s² + 2ζωn·s + ωn², regardless of what the open-loop plant looked like. That is why ωn and ζ are not just properties of masses and springs — they are the parameters a control engineer tunes by choosing gains, and the resulting closed-loop pole locations are what the entire design exercise is actually aiming at.

Natural frequency ωn: the undamped oscillation rate

Natural frequency ωn is the rate, in rad/s, at which the system would oscillate if all damping were removed. It is a property of the energy-storage elements alone — inertia and stiffness, inductance and capacitance — and it sets the overall speed scale of the response before damping is even considered.

How ωn emerges from characteristic equation coefficients

In the standard second-order characteristic equation s² + 2ζωn·s + ωn² = 0, ωn² is the constant (s⁰) coefficient once the s² coefficient has been normalized to 1. For a mass-spring-damper system m·ẍ + b·ẋ + k·x = 0, dividing through by mass m gives ẍ + (b/m)ẋ + (k/m)x = 0, so ωn = √(k/m) — stiffness over mass, square-rooted. For an LC circuit, the analogous result is ωn = 1/√(LC).

Units and practical interpretation (rad/s vs Hz)

ωn is expressed in rad/s, not Hz, because it comes directly from the Laplace variable s, which has units of rad/s. To convert to a frequency a bench instrument would report, divide by 2π: f = ωn / (2π) Hz. A servo loop with ωn = 200 rad/s corresponds to roughly 31.8 Hz — a distinction worth checking explicitly before comparing a calculated ωn against a scope measurement in Hz.

Impact on settling time and bandwidth

Larger ωn means a faster response and wider closed-loop bandwidth, roughly proportional to ωn for a fixed ζ. Settling time scales approximately as Ts ≈ 4/(ζωn) for a 2% band, so doubling ωn while holding ζ fixed halves the settling time. This is also where ωn trades directly against noise sensitivity and actuator effort — pushing ωn higher demands more control bandwidth and more gain, which is the same bandwidth that lets sensor noise through, a trade-off covered in detail under the complementary sensitivity function.

Damping ratio ζ: controlling overshoot and ringing

Damping ratio ζ is a dimensionless number that determines how much a second-order system oscillates before settling, independent of how fast it is. It is computed directly from the characteristic equation's roots, and it alone decides whether the step response rings, creeps, or arrives in one smooth motion.

Definition from characteristic equation roots

Solving s² + 2ζωn·s + ωn² = 0 with the quadratic formula gives s = -ζωn ± ωn√(ζ² - 1). The term under the square root is the switch: when ζ² - 1 is negative, the roots are complex and the system rings; when it is positive, the roots are real and the system does not oscillate at all.

Three regimes: underdamped (ζ < 1), critically damped (ζ = 1), overdamped (ζ > 1)

Most servo and regulatory loops target ζ between 0.6 and 0.8 — enough margin to avoid visible ringing while not sacrificing speed the way an overdamped design does.

How ζ determines pole locations in the s-plane

For the underdamped case, the complex poles sit at s = -ζωn ± jωn√(1-ζ²), and geometrically they lie on a circle of radius ωn centered at the origin, at an angle θ = cos⁻¹(ζ) from the negative real axis. This is the basis of constant-ζ design lines drawn radiating from the origin on root-locus plots: picking ζ fixes the angle, picking ωn fixes the radius, and the intersection is the pole location a compensator must place.

Trade-off between speed and stability

Pushing ζ down moves the poles closer to the imaginary axis angularly, increasing overshoot and ringing for the same ωn; pushing ζ up toward or past 1 removes oscillation but slows the response because the dominant real pole moves toward the origin. There is no value of ζ that is simultaneously fastest and free of overshoot — every design is a compromise point on that line, and matching it to the actual requirement (not just picking 0.707 by habit) is worth doing deliberately.

Damped frequency ωd: the actual oscillation you observe

Damped frequency ωd is the frequency at which an underdamped system actually rings, and it is always lower than ωn. The relationship is ωd = ωn√(1 - ζ²), derived directly from the imaginary part of the complex pole pair.

Derivation: ωd = ωn√(1 − ζ²)

Starting from the pole location s = -ζωn ± jωn√(1-ζ²), the imaginary part is by definition the frequency of the decaying sinusoid in the time response, so ωd = ωn√(1-ζ²) falls directly out of the pole algebra rather than requiring a separate derivation from the differential equation.

Only exists when underdamped (ζ < 1)

For ζ ≥ 1 the poles are real, there is no oscillatory component, and ωd is undefined — asking "what is the damped frequency of an overdamped system" is a category error, not a calculation with an answer of zero.

Relationship to measured ringing in step response

ωd is what a scope or logger actually shows: measure the time between two successive peaks (or zero crossings) of the ringing step response, call it Td, and ωd = 2π/Td. This is the only one of the three parameters you can read directly off a waveform without any computation beyond a stopwatch and a calculator.

How ωd differs from ωn in practical measurements

At ζ = 0.1 the difference is small (ωd ≈ 0.995·ωn), but at ζ = 0.6 — a typical servo target — ωd ≈ 0.8·ωn, a 20% gap that matters if someone mistakes the measured ringing frequency for ωn when checking a design against spec.

Step response: connecting ωn, ζ, ωd to time-domain behavior

The standard second-order step response to a unit step is:

c(t) = 1 - (e^(-ζωn·t) / √(1-ζ²)) · sin(ωd·t + φ),   φ = cos⁻¹(ζ)

Every visible feature of that waveform — overshoot, settling time, ring period — is a function of only ωn, ζ, and the derived ωd.

Peak overshoot as a function of ζ alone

Percent overshoot is %OS = 100·exp(-ζπ/√(1-ζ²)), and critically, ωn does not appear in that formula at all. This is why overshoot is purely a damping-ratio spec: a system with ζ = 0.5 overshoots by about 16.3% whether ωn is 10 rad/s or 10,000 rad/s. Full derivations of overshoot alongside the other time-domain specs — rise time, peak time, settling time, steady-state error — are laid out in Performance Specs: %OS, Ts, Tr, Tp, Ess.

Settling time dependence on ωn and ζ

The 2% settling time approximation Ts ≈ 4/(ζωn) shows both parameters acting together: the product ζωn is the real part magnitude of the pole, i.e., the decay rate of the envelope, so settling time depends on where the pole sits horizontally in the s-plane, not on ωn or ζ individually.

Period of oscillation from ωd

The time between successive peaks in the ringing response is Td = 2π/ωd, directly measurable on a scope and the quickest cross-check that a step response matches a design's predicted ωd before trusting any other number pulled from the same trace.

Design procedure: selecting ωn and ζ from performance requirements

Pick ζ to satisfy the overshoot spec first, then pick ωn to satisfy the settling-time or rise-time spec, since overshoot depends only on ζ while speed depends on both.

Translating spec limits into pole placement

Given a maximum allowable %OS, invert %OS = 100·exp(-ζπ/√(1-ζ²)) to solve for the minimum ζ. Given a maximum allowable Ts, use ωn ≥ 4/(ζ·Ts) with that ζ already fixed. The pair (ζ, ωn) then places the dominant closed-loop pole at s = -ζωn ± jωn√(1-ζ²), which becomes the target for root-locus or pole-placement compensator design.

Common ζ values for industrial servo and regulatory loops

Position servos commonly target ζ ≈ 0.6–0.7 for a crisp but not ringy response; process-control loops (temperature, flow) often use ζ ≈ 0.8–1.0 because overshoot past a setpoint can be operationally undesirable even at the cost of some speed. Always verify the actual target against the governing requirements document for the application rather than defaulting to ζ = 0.707 out of habit.

Bandwidth vs robustness: balancing ωn choice

Raising ωn to hit an aggressive settling-time spec increases required loop bandwidth and gain, which narrows phase margin and raises sensitivity to unmodeled high-frequency dynamics — actuator resonances, sensor delay, PWM sampling effects. A reasonable starting rule is keeping ωn at least 5–10x below any known unmodeled resonance frequency in the plant, then verifying with a full frequency-response check rather than assuming the two-pole model extends arbitrarily high.

Checking pole dominance after design

A compensator rarely produces exactly two closed-loop poles — it produces however many poles the compensator plus plant order dictates, and other poles or zeros can corrupt the "ideal" second-order shape. After design, verify the dominant pole pair is at least 5x closer to the imaginary axis than any other closed-loop pole; if not, the ωn/ζ-based overshoot and settling-time formulas above will not match a full simulation of the actual closed loop.

Measurement and verification from test data

Measuring ζ and ωn from a captured step response takes two readings off the waveform and two algebra steps, no curve-fitting software required.

Extracting ωn and ζ from step response waveforms

Measure percent overshoot %OS directly from the trace, then solve for ζ:

ζ = -ln(%OS/100) / √(π² + ln²(%OS/100))

Worked example: a measured overshoot of 20% gives ln(0.20) = -1.609, so ζ = 1.609 / √(π² + 1.609²) = 1.609 / √(9.87 + 2.59) = 1.609/3.53 ≈ 0.455. Then measure the ring period Td between the first two peaks — say Td = 15.7 ms — giving ωd = 2π/0.0157 ≈ 400 rad/s. Finally recover ωn = ωd / √(1-ζ²) = 400/√(1-0.207) = 400/0.890 ≈ 449 rad/s.

Using peak overshoot and period to back-calculate parameters

That three-step procedure — overshoot to ζ, ring period to ωd, ωd and ζ to ωn — is the standard commissioning method, and it requires nothing beyond a scope capture with at least two visible overshoot peaks; a response that settles before a second peak appears needs a finer timebase or a deliberately larger test step to get a usable reading.

Frequency response signatures of ζ at different operating points

On a Bode plot, ζ shows up as the height and sharpness of the resonant peak near ωn: a lightly damped system (ζ < 0.3) shows a pronounced peak above 0 dB, while ζ > 0.707 shows no peak at all, just a monotonic rolloff. Checking this signature at multiple operating points (different load, different setpoint) catches plants whose effective ζ changes with operating condition — common in motor drives where friction and backlash are load-dependent.

Commissioning checks for control loop tuning

During commissioning, capture a step response at the actual operating load, compute ζ and ωn by the method above, and compare against the design targets before accepting a tune. A measured ζ far lower than designed usually points to unmodeled compliance or backlash adding an extra lightly-damped pole pair; a measured ωn far lower than designed usually points to actuator saturation limiting effective loop gain during the test step.

Related reading

• First-Order Step Response: Rise Time, Settling Time — comparing response metrics.
• Noise Amplification: T(s) at High Frequency — understanding the high-frequency feedback trade-off.