13.4 Stability: Routh–Hurwitz & Root Locus ended with a gap. Padé approximations of a delay moved the critical gain from to against the exact , and the exact answer came from a phase condition on that needed no polynomial at all. This lesson takes that route seriously. Instead of asking where the poles are, we ask what the loop does to a sinusoid, one frequency at a time, and read stability, speed and robustness off two curves. It works on a model or on measured data: a wind-tunnel fan can be shaken at one frequency and its response measured, with no model at all.
Transcript
- Feed the cruise plant e to the i omega t and look for an output of the same form. That gives H equal to G of i omega, the transfer function at s equal i omega. The transient dies like e to the minus 0.1 t, and the settled output is the gain times a sine, shifted by the phase. At omega 0.1 and K equal 1 the lab integrates the ODE and measures gain 0.7036 and phase minus 50.7106 degrees: a lag of 8.851 seconds. At K equal 4 the gain is 2.8144.
- A Bode plot draws decibels against frequency on a log axis, so a product becomes a sum. The cruise plant is two first-order factors with breaks at 0.1 and 1 radians per second: the asymptote is flat, then falls 20, then 40 decibels per decade. A gain of K equal 4 lifts the whole curve to 12.04 decibels without touching the phase. At omega equal 1 the asymptote reads minus 7.96, and the true gain is minus 11.01. At a break one factor is down 3.010 decibels and 45 degrees.
- A second-order factor falls 40 decibels per decade above omega n. At omega n the gain is exactly 1 over 2 zeta, with phase minus 90 degrees. The peak rises to 1 over 2 zeta root 1 minus zeta squared. At zeta 0.1 that is 14.023 decibels, at 0.3 it is 4.847, and at 0.5 only 1.249. The gain at omega n is 13.979, 4.437 and 0.
- Now the loop. The gain crossover is where the loop gain is 1, and the phase margin is 180 degrees plus the phase there. The cruise loop at K p equal 4 crosses at 0.3625 radians per second with phase margin 85.495 degrees. At K p equal 10 the crossover is 0.7815 and the margin 59.282. At 19 it is 1.2077 and 44.360. This phase never reaches minus 180, so the gain margin is infinite: the proportional loop is stable at every gain, as in 13.4.
- Add an integrator: the PI loop at K p equal 4, gain K i. The phase crosses minus 180 at 0.7071 radians per second whatever K i is, where the loop gain is K i over 5.5. So the gain margin is 5.5 over K i: 11 at K i equal 0.5, 2.75 at 2, 1.1 at 5, exactly 1 at 5.5. The phase margin shrinks with it, to zero at 5.5 and minus 3.147 at 6, where the gain margin is 0.9167. That edge is the Routh edge of 13.4: one boundary, two methods.
- A delay drops the phase at the gain crossover by omega c times theta, so the delay the loop can take is the phase margin in radians over omega c. The cruise loop with K p equal 4 can take 4.116 seconds, K p equal 10 only 1.324, and K p equal 19 just 0.6411: more gain, less budget. For 1 over 10 s plus 1 at K equal 16.3506 the margin is exactly 1 second: 13.1's critical gain. With the engine lag and a 1 second delay, the critical gain is 12.759.
- The bandwidth is where the closed loop has fallen to 1 over root 2. For the cruise plant it grows with the gain, 0.2161 at K p equal 1 up to 1.9579 at 19, while the rise time falls from 10.112 seconds to 1.023. The product stays near 2, so the rise time is about 2.2 over the bandwidth, good to about ten percent. One pole gives ln 9, 2.197 exactly. The lightly damped loop, 1.601, is further off.
- The sensitivity S is 1 over 1 plus L, and its size is 1 over the distance from L of i omega to the point minus 1. Its peak M s bounds both margins: the gain margin is at least M s over M s minus 1, and the phase margin is at least 2 arcsine of 1 over 2 M s. Cruise at K p equal 4 has M s 1.1767; the 13.1 loop with a delay has 1.3958, and 2.8631 at K equal 10. For the PI loop at K i equal 2 the bounds are 1.925 and 27.8 degrees, below the actual 2.75 and 42.1.
- Margins look at one point of the curve. Take the loop K over s times s plus 1 times a lightly damped mode at 3 radians per second, K equal 7.2. Phase margin 53.833 degrees at 0.6942, gain margin 2.2855 at 2.6312: rules of thumb accept both. But the curve passes close to minus 1 between them: M s is 2.2109 at 2.8081 and the closed-loop poles are minus 0.1118 plus or minus 2.8577 i, damping 0.039. M s catches it, and guarantees only gain margin 1.826 and phase margin 26.1 degrees.
- A sine through a linear system returns a sine scaled by G of i omega. Bode plots add the factors, and the margins read the distance to minus 1: the PI gain margin is the Routh edge, 5.5 over K i, and the phase margin is a delay budget, 4.116 seconds at K p equal 4. Margins can mislead; M s does not. This lesson chose the controller structure in advance. Lesson 13.6 is not yet written; it is meant to take the state-space route: feed back every state, and place the closed-loop eigenvalues.
A sinusoid in, a sinusoid out
Take the cruise plant of the last four lessons, the engine lag and the car , so . Feed it the complex input and look for an output of the same form, with a constant. Then gives , and gives
So the answer is the transfer function with . The homogeneous part of the solution decays like (the pole at ), so after a few time constants only this particular solution is left. Taking the imaginary part, for a real input :
The gain says how much the amplitude is scaled and , normally negative, says how far the output lags. This is the forced-oscillation result of 5.6 Forced Oscillations & Resonance for any plant: a sine is an eigenfunction of every linear time-invariant system (6.5 Convolution & Transfer Functions), and is the eigenvalue.
Worked example 1. For the cruise plant at rad/s (unit gain): and , so ( dB). The phases are and , so , a lag of rad s. In the lab (cruise plant, , ) the inset simulates the ODE with RK4 and measures the settled output by projecting the last whole periods on and : gain , phase , the formula to the digits printed. At the gain is and the phase is unchanged.
Bode plots: decibels and a log axis
A Bode plot draws (decibels) and in degrees against on a log axis. A product becomes a sum: for the dB values add, and so do the phases.
A first-order factor has . Below the break frequency it is about dB, above it falls dB per decade. The asymptotes meet at , where the true gain is dB. The phase is a decade below the break, at it and a decade above.
A second-order factor has break frequency and slope dB per decade above it. At the gain is exactly with phase . From the resonance lesson (recomputed by the lab on ) the peak is at with height :
| peak at | peak height | gain at | |
|---|---|---|---|
| dB | dB | ||
| dB | dB | ||
| dB | dB |
For there is no peak. The short period of 12.5 Trim, Linearization & Longitudinal Modes (, ) as a unit-gain block (a stand-in, not the elevator-to-pitch transfer function; the lab’s short-period preset) peaks at rad/s by only dB.
Building a product. The cruise plant is two first-order factors with breaks at and rad/s, so the asymptote is flat, then , then dB per decade. A gain lifts the curve by and leaves the phase alone. For the flat part sits at dB; at the asymptote is , dB, and the true gain is , dB. The lab’s asymptote toggle draws the straight lines under the curve.
A delay has and phase . It leaves the magnitude plot alone and eats phase without limit as grows: 13.1’s warning as a curve that falls through .
The Nyquist picture, briefly
Plot in the complex plane as runs up from . A closed-loop pole has , so is critical. The Nyquist criterion (from memory, not proved here) says a loop that is itself stable gives a stable closed loop when the curve does not circle . For the curve leaves and spirals into the origin through the lower half plane, never reaching the negative real axis: 13.4’s result that the proportional cruise loop is stable at every gain.
For the PI loop with the denominator at is . It is real when , and there it equals . So : the curve crosses the negative real axis at at the same frequency whatever the gain. At it passes through ; at it crosses at , circles the critical point, and 13.4 counted two unstable roots. (A pole at the origin needs a small detour, from memory.)
Gain margin and phase margin
The phase crossover is where the phase is ; the gain margin is the factor by which the gain can grow before the curve reaches . The gain crossover is where ; the phase margin is the extra lag that takes the curve to .
The PI loop, all from the lab. Cruise plant with a PI controller at , so with as the gain:
| PM | GM | GM in dB | |||
|---|---|---|---|---|---|
The gain margin is in every row, since . The Routh edge of 13.4, , is the gain at which the margin is , and is the frequency of the zero row’s auxiliary polynomial: one boundary, two methods. A margin below is a loop past its edge. The phase margin shrinks with the damping: 13.4’s pair had , and at , , .
Worked example 2: the cruise loop at . Here gives, with , , so and . The phase there is , so , and the lab reports rad/s and . This plant’s phase never reaches : no phase crossover, infinite gain margin.
For the relation (from memory; the lab agrees at four values) is : , , , at , , , , hence the rule (from memory). The cruise loop has closed-loop damping , , at , , (13.2), against , , : a rough indicator, ten to fourteen percent high here.
Phase margin is a delay margin
A delay leaves the gain crossover where it is and drops the phase there by . The loop reaches the edge when (radians), so
is the extra delay the loop can take. For the cruise plant with no delay (lab values): has rad at rad/s, s; gives s and gives only s. More gain raises and the budget falls.
Worked example 3: the 13.1 plant. For , , the gain crossover is and with no delay: s. With the s delay already in the loop the lab shows and a delay margin of s, and agrees. Setting leaves a phase margin of .
Without delay, gives s, and at the margin is s: a s delay is exactly what 13.1 found critical at . Gain and delay margin are two views of one boundary: at s the gain margin is ( at , at ); at , 13.1’s critical gain for s, the delay margin is s.
Now the cruise plant with the engine lag and a s delay. At the lab gives (an extra margin of s, so s in total, matching the no-delay of s above) and a gain margin of , so the critical gain with the delay is . At the budget of s is less than the s present: , , unstable.
Bandwidth and rise time
The bandwidth is where the closed loop has fallen to of its low-frequency value, the highest frequency the loop follows. For at (13.1), , so rad/s and the to rise time is s, product . For the cruise plant (rise time measured by the lab on the step response of ):
| rise time | product | PM | ||
|---|---|---|---|---|
| s | none | |||
| s | ||||
| s | ||||
| s | ||||
| s |
The product stays near , so (from memory, exact for one pole) is good to about ten percent here. A resonant loop is further off: the lightly damped loop below gives .
Sensitivity from the frequency response
is the distance from to , so : where the curve passes close to , disturbances and plant errors near that frequency are amplified. The peak is the reciprocal of the closest approach.
The peak also bounds the margins. If everywhere, then at the phase crossover gives , so , and at gain crossover has , so . The lab prints both bounds.
Lab values:
| loop | at | |
|---|---|---|
| cruise, | ||
| cruise, | ||
| cruise PI, , | ||
| cruise PI, , | ||
| , s delay, | ||
| same, |
The delay row is 13.1’s peak near rad/s, in a loop with and . For the bounds from are and , below the actual and .
Honest limits of margins
Each margin looks at one point of the curve and asks what happens if one thing changes, the gain or the phase. Real plants change in many ways at once.
A good-looking pair with a bad loop. Take , : an integrator, a lag and a lightly damped mode at rad/s with (the lab’s lightly damped preset). It has at rad/s and ( dB) at rad/s, which rules of thumb accept, but at rad/s and the closed-loop poles are , : a ring at rad/s. The curve passes close to between the two crossover points, where neither number looks. catches it (and guarantees only , ).
Delay and models. A pure delay gives infinitely many phase crossovers (the lab prints three), and a loop with several crossings needs the whole curve. Margins are computed from a model: they say how far the model is from the edge, not the plant.
Concept checks
Playground and exercise
The first cell scans a grid for the margins and of the cruise loop (agreeing with the lab to about three digits), the second integrates the cruise ODE under a sine and measures the settled gain and phase, and the exercise asks you to write the crossover and margin functions.
Problems
A first-order system is driven by . Find the settled output and the time lag between input and output peaks.
Show worked solution
and . The gain is and the phase is rad, so . The output lags by s.
Let . (a) Give the DC gain in dB, the breaks and the asymptote slopes. (b) Find the exact gain crossover and the phase margin of .
Show worked solution
(a) : DC gain dB; breaks at and rad/s; slopes , then from to , then dB per decade.
(b) with gives , so and rad/s. (This is the cruise plant sped up tenfold at : same PM, frequencies times ten.) The phase is , so (the lab agrees, ).
For the cruise plant under PI control with , . Find the phase crossover frequency and the gain margin at , and say how it relates to 13.4’s edge .
Show worked solution
At the denominator is . It is real and negative when , where it equals . So rad/s and , giving at ( dB). The edge is where the margin is . (Lab, custom plant num , den : at rad/s, at .)
For with gain , show that the gain crossover is and that the delay margin is . Evaluate it at , and find the gain for which s.
Show worked solution
gives . The phase is , so and . At : , , , rad, and s. Setting has no closed form; a few steps of bisection give , which is 13.1’s critical gain at a s delay (). A larger gain always tolerates less delay.
What comes next
Frequency response described a loop by how it treats a sinusoid, for one input and one output and for a controller structure chosen in advance. It did not say how to choose the dynamics. An aircraft has many states, and the yaw damper of 12.6 fed back one of them. Lesson 13.6 (13.6 State-Space Control & Pole Placement) takes the state-space route: feed back every state, , and place the closed-loop eigenvalues, with the margins of this lesson as the check.
Key Takeaways
- A stable system driven by settles to : for the cruise plant at (unit gain), gain and phase .
- Bode plots use dB and so that factors add. A first-order corner is dB and ; a second-order factor has gain at and a peak ( dB at ).
- A delay costs phase and no gain. The cruise loop at has .
- The gain margin of the PI cruise loop at is , the Routh edge over the gain, at .
- : s at , s at ; for it is s at , 13.1’s critical gain.
- Bandwidth times rise time is near ( at , for one pole). gives and , not the converse: GM , PM coexisted with and .