Most margin habits come from loops that are stable before you close them. Give the plant one right-half-plane pole and two of those habits break: the curve must now go around −1, and lowering the gain, usually the safe move, can destabilise the loop.
Count before you read margins
Take L(s) = K(s+1)/((s² − 4)(0.1s+1)(0.05s+1)): a plant with poles at ±2 rad/s, like a levitated mass or a balancing link linearised about its upright position, with a PD-type zero at −1, an actuator time constant of 0.1 s and a sensor filter of 0.05 s. The open loop has one pole in the right half-plane, so P = 1. With N counted as clockwise encirclements of −1, the closed loop has Z = N + P right-half-plane poles, and stability needs N = −1: one counter-clockwise turn.
The curve starts at L(0) = −K/4 on the negative real axis, crosses that axis once more at 13.0 rad/s, at −K/26.1, and ends at the origin. The counter-clockwise turn exists only when −1 lies between the two crossings: −K/4 < −1 < −K/26.1, that is 4 < K < 26.1.
Three gains, three counts
At K = 3 the curve starts at −0.75, −1 lies outside it, N = 0 and the unstable pole stays: Z = 1, with a closed-loop pole at +0.361. At K = 10 the crossings are −2.5 and −0.383, −1 lies between them, N = −1 and Z = 0. At K = 40 the second crossing has moved to −1.533, beyond −1: the curve now circles −1 clockwise, N = +1 and Z = 2, a complex pair at 1.21 ± j15.61.
| K | Real-axis crossings | N | Z = N + P |
|---|---|---|---|
| 3 | −0.75, −0.115 | 0 | 1 |
| 10 | −2.5, −0.383 | −1 | 0 |
| 40 | −10, −1.533 | +1 | 2 |
Margins come in pairs
At K = 10 the loop tolerates a gain decrease by a factor of 2.5 (7.96 dB, down to K = 4) and an increase by a factor of 2.61 (8.33 dB, up to K = 26.1); the phase margin is 26.7° at 7.17 rad/s. The Bode plot calculator lists both phase crossovers and reports the one nearest 0 dB, −7.96 dB at ω = 0, with a warning that positive margins do not imply stability for this loop. A negative number at ω = 0 is a lower gain limit, not a sign that the loop is unstable.
The lower limit is the practical surprise. A supply that sags under load, a weaker magnet or a worn actuator lowers the loop gain, and here that walks the loop out of its window from below. Specify the margin in both directions, and test the low end as deliberately as the high end.
Try it
In the Nyquist plot calculator, enter Numerator s+1 and Denominator (s^2-4)(0.1s+1)(0.05s+1), then set K to 3, 10 and 40 and read P, N and Z = N + P beside the plot; the closed-loop poles listed with them agree. The root locus calculator with the same G(s) gives the stable range 4 < K < 26.1 and the imaginary-axis crossings at K = 4 (ω = 0) and K = 26.1 (ω = 13.0 rad/s).
The lab's cart-pole shows the lower edge of such a window in the time domain. In the Balance the pole experiment, set the position loop to Angle only and keep Derivative at 3. At Proportional 7 the pole stays up; at 6.5 the run stops when the cart reaches the end of its rail. Linearised, the angle equation needs the force per radian to beat gravity's push, 6.667·P > 45.78, so P > 6.87 N/rad.
Limits
The calculators assume an exact, linear, delay-free model. A dead time in this loop would bend the high-frequency part of the curve and lower the upper limit, and an unstable plant with a bounded actuator can recover only from disturbances small enough for the actuator to catch; beyond that no linear margin applies. The motor's frequency workbench analyses a stable open loop and cannot show this case.
With P open-loop poles in the right half-plane, stability needs P counter-clockwise turns around −1. Read the gain window from both real-axis crossings, and treat a falling gain as a risk.
CHECK YOUR UNDERSTANDING: At K = 10 the actuator loses 70% of its gain. Is the loop still stable?
No. The loop gain becomes 3, below the lower limit of 4: the curve no longer circles −1, so N = 0 and Z = 1.