Stability intuition
Keep the loop gain complex and ask when negative feedback stops behaving as a correction mechanism. Start from the characteristic denominator 1+L(s), interpret the critical point −1, define gain crossover, phase margin and gain margin, connect poor margin to peaking/ringing, and show why extra poles, capacitive loading and frequency-dependent feedback networks can erode stability. Use Bode/Nyquist intuition without turning 45° or 60° into universal design laws.
- Explain why the poles of the closed-loop system are determined by 1+L(s)=0 under the chapter-04 sign convention.
- Interpret the Nyquist critical point −1 as the condition L=−1 that makes the closed-loop denominator vanish.
- Define gain-crossover frequency ωc by |L(jωc)|=1 and phase margin PM=180°+∠L(jωc).
- Define the −180° phase-crossing frequency and gain margin, in ratio and dB form, without confusing it with closed-loop gain margin.
- Use Bode intuition to explain why additional poles and delay reduce phase margin near unity loop gain.
- Connect low phase margin qualitatively to closed-loop peaking and underdamped ringing while avoiding a universal one-to-one PM↔damping formula.
- Explain why a single dominant pole is benign and why pushing crossover toward additional poles increases risk.
- Relate op-amp noise gain to crossover and stability, distinguishing unity-gain-stable and decompensated devices conceptually.
- Explain how capacitive load interacts with nonzero output impedance/internal dynamics to add phase lag and why isolation/compensation may be required.
- Use a stability workflow: establish operating point → obtain L(s) with correct loop definition → inspect crossover and margins → test closed-loop frequency/step response → verify across loading/tolerance conditions.
Negative feedback is defined by what the returned signal does around the operating point. Stability asks a harder question: after the loop accumulates frequency-dependent phase lag, does the return still oppose a perturbation strongly enough, or can the loop reproduce and reinforce it?
05.1The characteristic equation is 1+L(s)=0
Chapter 04 gave the canonical closed-loop relation
With the subtractive sign convention already built into the loop, closed-loop poles are roots of the characteristic equation
Therefore stability is not controlled by |L| alone. Both magnitude and phase of the complex loop gain matter.
05.2Why −1 is the critical point
If at some complex frequency condition the loop transmission becomes L=−1, then
and the closed-loop denominator vanishes. In the Nyquist plane this makes −1 the critical point for our convention. A trajectory that approaches −1 indicates shrinking robustness; encirclement must be interpreted with the number of right-half-plane poles of the open-loop transfer.
For an open-loop-stable L(s) with no right-half-plane poles, avoiding an encirclement of −1 is the familiar Nyquist stability condition. If L(s) itself has unstable poles, simple “do not cross −1” intuition is insufficient and the full Nyquist count is required.
05.3Gain crossover and phase margin
Figure 5.1 — Reading margins from L(jω)
The one question is: how much distance remains before the critical condition L=−1?
Phase margin is read where loop-gain magnitude reaches unity (0 dB). Gain margin is read where phase reaches −180°. These margins summarize local distance from the critical condition; they do not replace Nyquist when there are multiple crossings or unstable open-loop poles.
Define a gain-crossover frequency ωc at which
For a regular single crossover, the phase margin is
Example: if ∠L(jωc)=−125°, then PM=55°. Phase margin measures how much additional phase lag at that crossover would bring the loop to −180° while its magnitude is still unity.
A positive phase margin is useful local evidence, but multiple 0 dB crossings, nonminimum-phase dynamics or unstable open-loop poles require more careful Nyquist reasoning.
05.4Phase crossover and gain margin
At a phase-crossing frequency ω180 where
the gain margin is the factor by which loop magnitude could increase before reaching unity:
If |L|=0.25 at −180°, GM=4 or about 12.04 dB. This is a margin on loop gain, not a statement that the closed-loop amplifier may safely have four times more signal gain.
05.5One dominant pole is benign; extra poles consume phase
A single-pole loop
adds at most −90° of phase lag, so a conventional negative-feedback loop with no other problematic dynamics remains far from the −180° condition at unity gain. A second pole adds another phase lag:
If crossover moves near or beyond ωp2, the accumulated phase can approach −180°. Dominant-pole compensation deliberately lowers crossover so that unity loop gain is reached before later poles have contributed too much phase.
05.6Low phase margin often appears as peaking and ringing
For common low-order minimum-phase loops, shrinking phase margin usually corresponds to a more resonant closed-loop response: frequency-domain peaking and time-domain overshoot/ringing become more pronounced.
Do not turn this into an exact universal conversion between PM and damping ratio ζ. Such formulas are model-dependent. Extra zeros, higher-order poles, delay and nonminimum-phase behavior can break a simple one-to-one mapping.
05.7Noise gain tells an op-amp where it meets its open-loop curve
For an op-amp feedback circuit, β=1/Gnoise. Hence
The loop crossover occurs approximately where the open-loop gain magnitude |A| falls to the noise gain. A higher constant noise gain intersects a falling open-loop curve earlier, often before additional internal poles add as much phase lag.
A unity-gain-stable op-amp is internally compensated to tolerate noise gain near 1 under specified loading. A decompensated device can offer higher bandwidth but may require a minimum stable noise gain greater than 1. The datasheet, not the schematic symbol, determines that constraint.
05.8Capacitive load changes the real loop
Figure 5.2 — Real follower with capacitive load
The one question is: why can CL erode margin even when the follower is wired correctly?
Unity feedback returns vo directly to the − input. In a real op-amp, nonzero output impedance and internal dynamics interact with CL and can add another pole/phase lag. The capacitor is not “unstable by itself”; its interaction with the real dynamic model changes L(s).
With an ideal zero-output-resistance voltage source, CL would not create the same output pole. Real op-amps have nonzero, frequency-dependent output impedance and internal poles. Their interaction with load capacitance can introduce additional lag and reduce phase margin.
Possible remedies include choosing an op-amp specified for the load, reducing capacitive loading, or adding a datasheet-recommended isolation/compensation network. A series output resistor is a common technique, but its value and effect are device- and load-dependent and must be analyzed rather than copied blindly.
05.9Delay is phase lag even without another simple pole
A pure delay contributes
so its phase lag grows without bound with frequency while its magnitude remains 1. This is why layout, sampling, propagation or other delayed paths can damage margin even when they do not resemble a single lumped RC pole.
05.10A stability workflow that survives real circuits
- Establish the intended DC operating point and verify the amplifier is not saturated.
- Define the feedback loop and obtain L(s) without destroying the operating point/loading.
- Identify all gain crossovers and phase crossings, not only the first visually convenient one.
- Read PM/GM where meaningful and use Nyquist reasoning when open-loop instability or multiple crossings make margin summaries ambiguous.
- Inspect the resulting closed-loop magnitude and transient response for peaking, overshoot or ringing.
- Repeat across expected load capacitance, component tolerance, device spread, supply and temperature conditions when they materially change dynamics.
05.11Mastery exercises
- At ωc, L has phase −132°. Find PM.
- At the −180° phase crossing, |L|=0.2. Find gain margin as a ratio and in dB.
- A loop crosses 0 dB before its second pole becomes active. Explain why this is generally safer than crossing a decade above that second pole.
- An inverting op-amp has noise gain 10. Explain why its loop crossover differs from a unity-gain follower using the same op-amp.
- Why does “PM=60°” not guarantee identical overshoot for all amplifiers?
- A follower is stable unloaded but rings with a long cable/capacitive sensor. Give a loop-level explanation and two classes of remedy.
- Why must a loop with an open-loop right-half-plane pole be analyzed with the full Nyquist criterion rather than only checking a positive phase margin?
Solutions
1. PM=180°−132°=48°.
2. GM=1/0.2=5; GMdB=20log10(5)=13.98 dB.
3. Before the second pole, accumulated phase lag is smaller; above it, the second pole contributes substantial additional lag and reduces margin.
4. Since β=1/Gnoise, noise gain 10 gives β=0.1, so |Aβ| reaches unity where |A|≈10 rather than |A|≈1. With a falling A curve this occurs at a lower frequency.
5. PM is only a local loop metric; higher-order poles, zeros, delay and nonminimum-phase effects alter the relation between margin and time-domain damping.
6. The load capacitance interacts with real output impedance/internal dynamics, adding phase lag. Remedies include a device rated for that load or a properly designed isolation/compensation network.
7. Nyquist stability depends on both encirclements of −1 and the number of unstable open-loop poles; local crossover margins alone do not contain that information.
AE-03 is now conceptually complete: differential pair → op-amp model → negative feedback → loop gain → stability intuition. AE-04 can now descend into transistor capacitances, Miller effect, cascode bandwidth and noise with a feedback framework already in place.