Nyquist, Root Locus and Conditional Stability
Find the loop that passes its phase-margin check and still oscillates, using the Nyquist plot and the closed-loop poles.
You Will Be Able To
- Read the crossings of the negative real axis from an AC run and count encirclements of −1.
- Predict the range of gain over which a conditionally stable loop oscillates.
- Find the closed-loop poles with
pzand link them to the step response. - Explain why a large step can start an oscillation that a small step does not.
- Lesson
- E10.7
- Time
- 35 min
- Runs on
- ngspice
- Check
- 5 questions
A feedback loop passed its review with 62° of phase margin, and on the bench a 50 µV step settled cleanly. A 10 mV step did not: the output now swings from −121 mV to +136 mV at 1.9 kHz without end.
- gain stage output
- loop output
- Time 600 µsgain stage output = -1 V · loop output = 24.98 mV
The step arrives. The error is too large for the gain stage, so its output hits the +1 V limit.
- Time 3 msgain stage output = -1 V · loop output = 134.6 mV
The gain stage now swings between its limits, like an amplifier with much less gain.
- Time 7 msgain stage output = 1 V · loop output = -117.7 mV
The output keeps swinging. One period is 528 µs, which is 1.9 kHz.
Read the plot as a table
| Time | gain stage output | loop output |
|---|---|---|
| 0 s | 0 V | 0 V |
| 600 µs | -1 V | 24.98 mV |
| 1 ms | -1 V | 81.77 mV |
| 2 ms | -1 V | 126 mV |
| 3 ms | -1 V | 134.6 mV |
| 4 ms | -1 V | 85.8 mV |
| 5 ms | 1 V | -59.5 µV |
| 6 ms | 1 V | -77.17 mV |
| 7 ms | 1 V | -117.7 mV |
| 8 ms | 1 V | -109.4 mV |
The same run also applies a 50 µV step, which peaks at 67.6 µV and settles to 50.0 µV.
Phase margin describes one frequency at one gain, and this loop fails at another gain. The Nyquist plot shows that failure before the bench does.
A Loop with 62° of Margin
- Device
- Data
- Three poles
Three stages each add a pole at 1 kHz. Together they can turn the phase by 270°.
- Two zeros
Each lag-lead network also has a zero at 30 kHz. Two zeros turn the phase back by 180°.
- Once around
T(s) = K·(1 + s/z)2/(1 + s/p)3, where p and z are the pole and zero frequencies. The lab measures T directly.
Read the diagram as text
- Subtract: error = in − out
- Gain K: 94300 at nominal
- Pole: RC, 1 kHz
- Lag-lead 1: pole 1 kHz, zero 30 kHz
- Lag-lead 2: pole 1 kHz, zero 30 kHz
- Subtract → Gain K: error
- Gain K → Pole
- Pole → Lag-lead 1
- Lag-lead 1 → Lag-lead 2
- Lag-lead 2 → Subtract: out, fed back
Ideal buffers separate the stages.
- Low frequency
C2 passes no current, so nothing flows in R2. The output equals the input.
- The pole at 1 kHz
C2 starts to conduct through R2 and R3. The pole is at 1/(2π·(R2 + R3)·C2) = 1 kHz.
i
- The zero at 30 kHz
Above 1/(2π·R3·C2) = 30 kHz, C2 is a short. The gain settles at R3/(R2 + R3) = 1/30 and the phase returns to 0°.
i
Read the circuit as text
- v_in voltage source: + →
n1, − →0 - R2 resistor (153.849 kΩ): pin 1 →
n2, pin 2 →n1 - R3 resistor (5.305 kΩ): pin 1 →
n2, pin 2 →n3 - C2 capacitor (1 nF): pin 1 →
n3, pin 2 →0 - out port: terminal →
n2
The network lowers the gain by 30, then adds no phase shift.
The loop gain T is the gain once around the loop, with the minus sign of the subtraction left out. A closed-loop gain is then A/(1 + T). The loop oscillates if T reaches −1, because the denominator becomes zero.
- |T|
- phase of T
- Frequency 1.916 kHz|T| = 79.44 dB · phase of T = -180°
Three poles pull the phase through −180° at 1.92 kHz. Here |T| is still 79.4 dB.
- Frequency 6.76 kHz|T| = 49.84 dB · phase of T = -219.4°
The phase is lowest at 6.76 kHz: −219.4°.
- Frequency 26.82 kHz|T| = 18.87 dB · phase of T = -180°
The phase climbs back through −180° at 26.8 kHz, where |T| is 18.9 dB.
- Frequency 112.2 kHz|T| = -0.00004558 dB · phase of T = -118.4°
|T| reaches 0 dB at 112 kHz. The phase there is −118.4°, so the phase margin is 61.6°.
Read the plot as a table
| Frequency | |T| | phase of T |
|---|---|---|
| 10 Hz | 99.49 dB | -1.681° |
| 100 Hz | 99.36 dB | -16.75° |
| 1 kHz | 90.47 dB | -131.2° |
| 1.916 kHz | 79.44 dB | -180° |
| 6.76 kHz | 49.84 dB | -219.4° |
| 10 kHz | 40.28 dB | -216° |
| 26.82 kHz | 18.87 dB | -180° |
| 100 kHz | 1.152 dB | -121.7° |
| 112.2 kHz | -0.00004558 dB | -118.4° |
| 1 MHz | -19.59 dB | -93.26° |
| 10 MHz | -39.59 dB | -90.33° |
The phase stays below −180° from 1.92 kHz to 26.8 kHz.
The usual check reads the phase where |T| = 1 and finds 61.6° of phase margin. It never looks at 1.92 kHz or 26.8 kHz.
The Nyquist Plot and the −1 Point
A Nyquist plot draws T(jω) as one point for each frequency: the real part across, the imaginary part up. Negative frequencies give the mirror image of the curve.
The Nyquist criterion counts how many times the full curve circles the point −1. This loop has no open-loop pole in the right half-plane. For such a loop, the closed loop is stable only if the curve does not encircle −1.
The curve meets the negative real axis where the phase is −180°. Those two crossings decide whether −1 is inside.
| Gain K | Crossing at 1.92 kHz | Crossing at 26.8 kHz | Where is −1? | Closed loop |
|---|---|---|---|---|
| 94300 | −9383 | −8.777 | Right of both | Stable |
| 943 | −93.83 | −0.08777 | Between them | Unstable |
| 3.143 | −0.313 | −0.00029 | Left of both | Stable |
The lab prints the first two rows. The last row scales the first, because K only stretches the curve.
Read the plot as a table
| Re T | Im T |
|---|---|
| -3.5 | 2.826 |
| -3 | 2.446 |
| -2.5 | 2.049 |
| -2 | 1.636 |
| -1.5 | 1.204 |
| -1 | 0.7564 |
| -0.5 | 0.3061 |
One AC run. Only the branch that decides the count is drawn.
At K = 943 the curve passes above −1 and comes down to the axis at −0.088. Its mirror image passes below −1. Together they close around −1, so the closed loop is unstable.
At K = 94300 both crossings lie left of −1. The curve goes up across the axis at −9383 and comes back down at −8.78. The two cancel, so the net count is zero.
Lower the Gain and the Loop Breaks
A loop is conditionally stable when it is stable at its nominal gain and unstable at some lower gain. This one oscillates when |T| at 26.8 kHz falls below 1 while |T| at 1.92 kHz is still above 1.
The Bode plot gives that range. The gain may drop by up to 18.9 dB, a factor of 8.78. A drop between 8.78 and 9383 times puts −1 inside the curve.
- K = 94300
- K = 3.143
- K = 943
- Time 15 µsK = 94300 = 12.65 mV · K = 3.143 = 2.207 µV · K = 943 = 648.1 µV
At K = 94300 the output peaks at 12.67 mV after 15 µs and then settles to 10.00 mV.
- Time 230 µsK = 94300 = 10 mV · K = 3.143 = 5.162 mV · K = 943 = -7.52 mV
At K = 943 the output rings at 10.1 kHz and each swing is larger than the last.
- Time 458 µsK = 94300 = 10 mV · K = 3.143 = 10.63 mV · K = 943 = 5.855 V
At K = 3.143 the output rises slowly and peaks at 10.64 mV.
- Time 500 µsK = 94300 = 10 mV · K = 3.143 = 10.46 mV · K = 943 = -10.88 V
By 0.5 ms the K = 943 loop swings from −10.9 V to +5.9 V. A real circuit would clip long before.
Read the plot as a table
| Time | K = 94300 | K = 3.143 | K = 943 |
|---|---|---|---|
| 0 s | 0 V | 0 V | 0 V |
| 15 µs | 12.65 mV | 2.207 µV | 648.1 µV |
| 100 µs | 10 mV | 814.1 µV | -15.98 mV |
| 200 µs | 10 mV | 3.969 mV | -108.2 mV |
| 230 µs | 10 mV | 5.162 mV | -7.52 mV |
| 300 µs | 10 mV | 7.82 mV | -525.9 mV |
| 400 µs | 10 mV | 10.26 mV | -2.411 V |
| 458 µs | 10 mV | 10.63 mV | 5.855 V |
| 500 µs | 10 mV | 10.46 mV | -10.88 V |
The lower panel is in volts, the upper in millivolts.
Less gain breaks this loop. The usual gain margin guards against the opposite, too much gain.
Root Locus: Where the Closed-Loop Poles Go
A root locus is the path of the closed-loop poles as K rises from zero. A pole with a negative real part σ decays as eσt. A pole with a positive σ grows.
The pz analysis prints the closed-loop poles in rad/s. The lab runs it at the three gains.
| Gain K | Real part σ of the pole pair | Frequency of the pair | Step response |
|---|---|---|---|
| 94300 | −2.721×105 /s | 50.81 kHz | Decays in a few microseconds |
| 943 | 1.488×104 /s | 10.09 kHz | Grows e times every 67 µs |
| 3.143 | −1.940×103 /s | 1.281 kHz | Decays in about 0.5 ms |
The unstable pair at 10.09 kHz matches the 99.08 µs period in the step figure.
- damping ratio ζ
- pole frequency
- Gain K 3.162damping ratio ζ = 0.2328 · pole frequency = 1.284 kHz
At low gain the pair is damped.
- Gain K 1000damping ratio ζ = -0.2262 · pole frequency = 10.33 kHz
From K = 10.05 to K = 10744 the poles are in the right half-plane.
- Gain K 79430damping ratio ζ = 0.5695 · pole frequency = 50.93 kHz
Above K = 10744 the poles are back on the left. Near the nominal gain ζ is 0.65.
Read the plot as a table
| Gain K | damping ratio ζ | pole frequency |
|---|---|---|
| 1 | 0.5168 | 865.7 Hz |
| 3.162 | 0.2328 | 1.284 kHz |
| 10 | 836.6 µ | 1.912 kHz |
| 100 | -0.2234 | 4.338 kHz |
| 1000 | -0.2262 | 10.33 kHz |
| 10000 | -0.01195 | 26.05 kHz |
| 79430 | 0.5695 | 50.93 kHz |
| 100000 | 0.6773 | 50.39 kHz |
From 51 pz runs. Each boundary gain is 94300 divided by |T| at a −180° crossing.
The poles enter the right half-plane at K = 10.05 and return at K = 10744. At exactly these gains, a crossing sits on −1.
Why a Large Step Finds the Low Gain
The gain stage in the hook cannot swing past ±1 V. A stage that clips gives less output per volt of input. Its effective gain falls as the swing grows.
A large step pushes the effective gain down into the unstable range, and the loop stays at its bottom edge.
A small step keeps the stage linear, so it sees the full gain and the full 61.6° of margin. Start-up and overload do not.
Asked in interviews
What does the Nyquist criterion say for an amplifier that is stable in open loop?
Plot the loop gain T(jω) for all frequencies. The closed loop is stable only if the curve does not encircle −1. Encirclements are counted as a net number, so two opposite crossings to the left of −1 cancel.
What is a conditionally stable loop?
A loop that is stable at its nominal gain and unstable at some lower gain. Its phase passes below −180° while |T| is still above 1, then recovers before the crossover. Here that band runs from 1.92 kHz to 26.8 kHz.
Why can such a loop pass a phase-margin check and still oscillate?
Phase margin reads the phase at one frequency, where |T| = 1. A large signal makes the amplifier clip, which lowers its effective gain. The lower gain puts −1 inside the Nyquist curve, and the loop oscillates.
Try It
Part 1 opens the loop (EF has gain 0) and reads T as v(out). cph() gives the phase without wrapping, and meas finds both −180° crossings and the real part of T at each.
Part 2 closes the loop and runs pz at three gains. The let lines copy the pole pair into the AC plot, because meas reads only analysis vectors.
Part 3 lowers the gain 100 times and reads the crossings again. Each is 100 times smaller, and −1 now lies between them.
Predict before each change. Set the last alter EK gain=943 to 9430: where do the crossings land, and is the loop stable? Change R3 and R5 to 15.9k, which moves the zeros to 10 kHz: does the phase still pass −180°?
Run the netlist to draw its plots here.
fu = 1.122447e+05 ph_fu = -1.183974e+02 f180_lo = 1.915908e+03 f180_hi = 2.682193e+04 re_lo = -9.382675e+03 re_hi = -8.777035e+00 sigma_nom = -2.721055e+05 f_nom = 5.080771e+04 sigma_943 = 1.487948e+04 f_943 = 1.009262e+04 sigma_3 = -1.939680e+03 f_3 = 1.281029e+03 re_lo_943 = -9.382699e+01 re_hi_943 = -8.777014e-02 im_at_m1 = 7.564325e-01Check Yourself
Common Mistakes
Wrong Sign off a loop on its phase margin at the nominal gain alone.
Phase margin reads one frequency, so it cannot see a −180° crossing where |T| is still above 1.
Fix Also check the phase at every frequency below the crossover. If it passes −180°, find the gains that make the loop oscillate.
Wrong Read the phase with vp() or ph() and trust the number.
These functions wrap the phase, so this loop's −219.4° prints as +140.6° and looks harmless.
Fix Use cph(), which keeps the phase continuous through −180°.
let ph = 180/pi*vp(out)
meas ac ph_min MIN phlet ph = 180/pi*cph(v(out))
meas ac ph_min MIN phWrong Call a loop unstable because its Nyquist curve crosses the real axis to the left of −1.
The criterion counts net encirclements, and two opposite crossings on the same side of −1 cancel.
Fix Count the crossings on each side of −1. This lesson's loop crosses at −9383 and −8.78 and is stable.
Wrong Test the closed loop only with a small step.
A small step keeps every stage linear, so it never shows what happens when a stage clips and its gain falls.
Fix Also run a step large enough to saturate the amplifier, and a power-up, and watch whether the loop recovers.
Remember
- A loop that is stable when open is stable when closed only if its Nyquist curve does not encircle −1.
- The curve meets the negative real axis where the phase is −180°. Those crossings, compared with −1, decide stability.
- A conditionally stable loop is stable at nominal gain and oscillates over a range of lower gain. Here that range is 8.78 to 9383 times lower.
- The closed-loop poles cross into the right half-plane at exactly the gains that put a crossing on −1.
- A stage that clips has less effective gain, so a large step can start an oscillation that phase margin never predicts.