Two numbers on a Bode plot tell you most of what you need to know about whether a feedback loop will behave itself. Gain margin and phase margin measure how much extra gain or extra lag the loop can absorb before it starts to ring, and then to oscillate. They are also easy to misread: plenty of loops that looked comfortable on paper misbehave the first time a valve sticks or a motor gets hot.
The reassuring part is that the targets are not secret. Most process, motion and audio loops sit inside fairly well-worn ranges, and knowing those ranges makes tuning far less of a guessing game.
What the two margins actually measure
Treat the whole loop as a single transfer function: controller, actuator, plant and sensor multiplied together. Its open-loop magnitude and phase, plotted against frequency, describe how the closed loop will behave — at least while the system stays linear and time-invariant, which is a bigger assumption than it sounds.
Gain margin is read at the phase crossover frequency, where the open-loop phase has fallen to exactly −180°. It tells you how much you could multiply the loop gain by before the magnitude at that frequency reaches 1. In decibels, it is simply minus the open-loop gain at that point. A gain margin of 6 dB means you can double every gain in the loop before it becomes marginally stable.
Phase margin is read at the gain crossover frequency, where the open-loop magnitude is 1, or 0 dB. It is the gap between the actual phase there and −180°. A 45° phase margin means the loop can absorb another 45° of lag — a slower sensor, more dead time, a filter someone adds next year — before it tips over.
If the phase never reaches −180° at all, the gain margin is infinite. That is common for well-damped, roughly first-order loops, and it is a good place to be.
Reading them off a real Bode plot
The slope at crossover matters as much as the numbers. A loop crossing 0 dB on a −20 dB per decade slope has roughly 90° of phase available and will usually feel calm. Cross on a −40 dB per decade slope and you are closer to −180° already, so the phase margin will be thin. The practical rule is to cross 0 dB on a −20 dB per decade segment and keep any −40 dB per decade territory at least an octave away from the crossover frequency.
Dead time is the usual culprit when margins vanish. It adds phase lag linearly with frequency: about 57° per second of delay at 1 rad/s, and twice that at 2 rad/s. Half a second of transport delay on a flow loop is harmless at low frequency and fatal near crossover. Filters behave similarly — every sensor filter, anti-alias filter and averaging window spends phase you did not budget for.
Why the two margins travel together
Phase margin maps quite neatly onto closed-loop damping. As a rough guide, damping ratio is about phase margin divided by 100, so 60° gives a damping ratio near 0.6 and around 10% overshoot on a step. At 45° you are nearer 0.45 and 16–20% overshoot. Drop to 30° and you get 35% overshoot, a long ringing tail and audible or visible wear on the mechanics.
Gain margin covers a different failure. It protects against gain errors you cannot see at design time: a sensor whose sensitivity drifts, an amplifier that runs hotter and louder, a mechanical resonance that peaks higher than the model suggested. It is not a substitute for phase margin, and in a conditionally stable loop — one where reducing gain also destabilises — gain margin alone tells you very little.
Rough targets by application
Process control
- General rule: 6–10 dB gain margin and 45–60° phase margin. Below 6 dB or 45° is where most plants start to feel brittle.
- Flow and pressure loops: noisy and fast, so detune deliberately — 60–70° phase margin with a low crossover costs little and stops the valve chattering.
- Temperature loops: slow but nonlinear, with actuator stiction and long lags. 8–10 dB gain margin is not excessive.
- Level loops: integrating behaviour means small controller gains, so proportional-only control with a deliberately modest 6 dB of margin is often the honest answer.
Motion and servo
- Velocity and position loops usually land at 40–60° phase margin and 8–12 dB gain margin. Below 30° feels twitchy and shows up as overshoot, current spikes and mechanical noise.
- Current loops sit inside the velocity loop and are usually tuned for 45–60° phase margin with a much wider bandwidth.
- Mechanical resonances are the classic gain-margin problem. A notch filter helps, but the notch has to stay accurate as load and temperature change, which is exactly why you want headroom.
Audio
- Power amplifiers and analogue active filters are normally designed for at least 45° phase margin, and often 60° or more, checked against the worst-case load — including a long capacitive cable.
- Electroacoustic loops such as feedback suppression and active noise control are dominated by latency. A single millisecond of delay is a full 360° of phase at 1 kHz, so crossover frequencies stay low.
- Because people move microphones and loudspeakers, keep 6–10 dB of gain margin in reserve and lean on limiting rather than on tuning to the edge.
Where the margins lie to you
Margins are a local, linear measure. Saturation, rate limits, deadband and stiction all consume stability that the Bode plot never showed you. So does operating point: a loop tuned at full flow may be marginal at 10%, and a servo tuned cold behaves differently once the gearbox warms up.
A useful second opinion is the sensitivity peak, the maximum of |1/(1+L)| across frequency. Values between about 1.2 and 2 are typical; above 2 generally means a phase margin below roughly 30° and a loop that will amplify disturbances it should be rejecting.
How to check the margins on a live loop
- Build the loop as the controller sees it — sensor filter, anti-alias filter, sampling delay, actuator dynamics, dead time — not as the datasheet describes it.
- Measure the response in place with a small step, chirp or PRBS injection rather than trusting a model alone.
- Find the gain crossover and read the phase there; find the phase crossover and read the gain there.
- Repeat at the worst realistic operating point: light load, low flow, cold start, worn actuator.
- Confirm with a step test. If the overshoot and settling time match what the margins predicted, you understand the loop.
- Leave two or three decibels of your margin unspent. You will need it.
How much margin is enough?
Six decibels and 45° is the traditional floor, and it is a floor for a reason: below it, ordinary drift and wear can push a stable loop into oscillation. Ten decibels and 60° is comfortable for most process work, while motion and audio loops generally want 45–60° of phase margin with 8–12 dB of gain. Tighter margins buy bandwidth and speed; they also buy a plant that demands attention.
The real test is not the number on the plot but whether the worst realistic operating point still behaves. If you find yourself running at 3 dB of gain margin to get the performance you need, the loop is probably telling you something useful about the plant — an oversized valve, a sluggish sensor, an unmapped resonance. Fix that, and the margins look after themselves.
Photo: RyanMcGuire / Pixabay
