Skip to content
Tutorials Center
My learning

Topics

Lesson 18 of 37 ยท Stability

Gain Ranges and Relative Stability

Use Routh-Hurwitz to design stable parameter ranges, measure relative stability by shifting the axis, and see how time delay limits gain, with Python checks.

Knowing that a loop is stable is only the start. In design you ask three practical questions: how much gain can I use?, how far from instability am I?, and what does a delay do to my limit? This lesson uses the tools of the previous three lessons to answer them, and checks each answer numerically.

Finding the stable range of a parameter

Suppose a plant G(s) = 1 / (s (s + 1)(s + a)) is controlled by a proportional gain K with unity feedback. The characteristic equation 1 + K G(s) = 0 is

Text
s^3 + (1 + a) s^2 + a s + K = 0

With the cubic rule from The Routh-Hurwitz Criterion: all coefficients positive (a > 0, K > 0) and a2 a1 > a3 a0, that is (1 + a) a > K. So the loop is stable exactly when

Text
0 < K < a (1 + a)

Two parameters give a stability region: a map of which (a, K) pairs are stable. The boundary K = a(1 + a) is where the loop oscillates at frequency sqrt(a) (from the auxiliary polynomial (1 + a) s^2 + K = 0, using K = a(1+a)).

Python A stability map for two parameters

Press Run (or Ctrl + Enter) to see the result here.

Program input ยท what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

The two methods agree at every grid point. The critical gain increases with a because a plant pole farther to the left leaves more room before the loop goes unstable.

Three stability questions

Gain range How large can K be?
Relative stability How far are the poles from the axis?
Delay limit How much lag can the loop take?

Routh answers the first; shifting answers the second.

1 / 2

Relative stability by shifting the axis

You usually want poles not just in the left half-plane but to the left of some line Re(s) = -sigma, which guarantees a decay rate (and a settling time of about 4/sigma). The trick: substitute s = z - sigma into the characteristic polynomial, giving a new polynomial in z. The original roots are to the left of -sigma exactly when the new polynomial is stable, and you can test it with Routh.

For s^3 + 6 s^2 + 8 s + K and sigma = 0.5, we ask: for what K are all roots to the left of -0.5?

Python Relative stability with a shifted axis

Press Run (or Ctrl + Enter) to see the result here.

Program input ยท what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

For K = 1 the slowest pole is near -0.14: stable but with a slow tail, so the shifted test fails. For moderate K the poles are comfortably left of -0.5, while for K = 20 and K = 40 the oscillatory pair has moved too close to the axis, so the shifted test fails again. Requiring a decay rate narrows the stable range of K from both sides.

What delay does to the stable range

Many real loops contain a time delay T: sensors, networks, transport pipes. A delay adds a factor e^(-sT) to the loop. It has magnitude 1 at every frequency but contributes phase lag -w T, which erodes stability. Because e^(-sT) is not a polynomial, Routh-Hurwitz cannot handle it directly; the usual approach is to replace it by a Pade approximation, e.g. the first-order one

Text
e^(-sT) ~ (1 - sT/2) / (1 + sT/2)

Take L(s) = K e^(-0.5 s) / (s + 1). The exact stability limit is where |L(jw)| = 1 and the phase equals -pi. We compare it with the Pade-based Routh result.

Python Delay limits the gain

Press Run (or Ctrl + Enter) to see the result here.

Program input ยท what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

Without delay, K / (s + 1) stays stable for any gain, because a first-order loop can never lose enough phase. With a half-second delay a finite critical gain appears (about 3.8 exactly; the crude Pade model gives 5), so the approximation is optimistic. The practical lesson: delay makes high gain dangerous, and a coarse approximation can overstate your margin. Chapter 5's frequency-response tools treat delay exactly.

A design checklist

  1. Write the closed-loop characteristic polynomial with the parameter(s) symbolic.
  2. Apply Routh to get the stable range, and confirm numerically with np.roots.
  3. Shift the axis to require a minimum decay rate.
  4. Include sensor lags and delays in the model and recheck.
  5. Choose the operating value well inside the range, not at an edge.

Check your understanding

Quick check ยท 4 questions

of 4 answered You scored / 4 ยท perfect!

  1. 1. For s^3 + 3 s^2 + 2 s + K = 0, what is the stable range of K?

    The cubic rule requires a2 a1 > a3 a0: 3 x 2 > K, so 0 < K < 6.
  2. 2. To check that all roots are to the left of Re(s) = -sigma, you

    The substitution moves the imaginary axis to Re(s) = -sigma, so stability of the new polynomial means the old roots lie left of that line.
  3. 3. How does a pure time delay affect a loop's stability?

    A delay e^(-sT) has unit magnitude but phase -w T, which brings the loop closer to instability.
  4. 4. Why is it risky to rely on a first-order Pade approximation for the stability limit?

    The approximation matches the delay only at low frequencies; at the crossover it can overstate the available margin.

Summary

  • Routh-Hurwitz turns a stability question into inequalities in the parameters, giving ranges like 0 < K < a(1 + a).
  • Shift the axis with s = z - sigma to demand a minimum decay rate (relative stability).
  • Time delay adds phase lag and creates a finite critical gain even for a first-order plant; Pade approximations can be optimistic.
  • Always verify analytic ranges numerically and keep the design well inside the boundary.

Next: Root Locus Basics