Laplace, Transfer Functions, Poles and Zeros

From jω to s: widening the window

In rung 5 you learned the frequency response: feed a circuit a pure sine at frequency ω and measure how it scales the amplitude and shifts the phase. That trick works beautifully — but only for sinusoids that have been running forever and will run forever. It says nothing about the *first second* after you flip a switch, when capacitors are still charging and the system is finding its feet. That opening transient is exactly where circuits ring, overshoot, and occasionally tear themselves apart.

The fix is a single, audacious generalization. The Fourier world lives on the imaginary axis, evaluating signals at s = jω — pure oscillations that neither grow nor decay. The Laplace transform replaces that imaginary frequency with a full complex frequency s = σ + jω. The new real part σ controls growth and decay; the imaginary part jω still controls oscillation. Where Fourier saw a single vertical line, Laplace sees an entire two-dimensional plane — the s-plane — and a signal like e^(σt)·cos(ωt) is just a single point on it.

             jω  (imaginary axis = oscillation)
              ^
              |        x   <- decaying ringing
   growing    |       /         (σ<0, ω≠0)
   (unstable) |      /
   ───────────+─────────────────> σ  (real axis = decay)
      σ>0     |    LEFT half-plane = STABLE
              |    (everything decays)
              x
              |
     pure sine sits ON the jω axis (σ=0): never decays

Differential equations become algebra

Here is the magic that made Laplace indispensable to engineers. A capacitor's current is i = C·dv/dt; an inductor's voltage is v = L·di/dt. Any circuit built from R, L, and C obeys a differential equation, and solving differential equations by hand is genuinely painful. The Laplace transform has one property that demolishes the whole problem: differentiation in time becomes multiplication by s. The operation d/dt — a calculus headache — turns into a plain factor of s you can multiply and divide like any number.

So every familiar component gets a complex impedance in the s-domain: a resistor stays R, an inductor becomes sL, and a capacitor becomes 1/(sC). Suddenly Kirchhoff's laws and the voltage-divider rule apply *exactly as they did with resistors* — except now the 'resistances' are functions of s. You write algebra, not calculus. Solve for the variable you want, and you have a tidy ratio of polynomials in s.

Series RLC, output across the capacitor:

   Vin --[ R ]--[ sL ]--+-- Vout
                        |
                     [1/sC]
                        |
                       GND

Voltage divider in s-domain:

   Vout      1/(sC)                 1
   ----  = ───────────────  = ───────────────────
   Vin     R + sL + 1/(sC)     LC·s² + RC·s + 1

=> a ratio of polynomials in s  =  the TRANSFER FUNCTION H(s)

Poles and zeros: the DNA of a system

A transfer function is a ratio of two polynomials, H(s) = N(s) / D(s). Factor each polynomial and the whole behavior of the system collapses into a short list of special numbers. The roots of the numerator N(s) — the values of s that make H(s) zero — are the zeros. The roots of the denominator D(s) — the values of s that make H(s) blow up to infinity — are the poles. Plot them on the s-plane (zeros as circles ○, poles as crosses ✕) and you have a pole-zero map: a compact portrait that an experienced engineer reads the way a doctor reads an ECG.

Why do poles matter so much? Because the impulse response — how the system reacts to an infinitely sharp 'kick' — is built entirely from the poles. Each pole at location s = σ + jω contributes a term e^(σt)·cos(ωt) to the time-domain response. Read that off directly: the pole's real part σ becomes the decay rate of an exponential envelope, and the pole's imaginary part ω becomes the ringing frequency. The poles *are* the natural modes of the circuit — the frequencies it 'wants' to vibrate at when disturbed, just like a struck bell.

1. Pole far left (σ very negative): e^(σt) decays fast → that mode dies out almost instantly. Fast, well-behaved.
2. Pole near the jω axis (σ small & negative): slow decay → the response lingers and ringing takes a long time to fade. Sluggish, oscillatory.
3. Pole exactly on the jω axis (σ = 0): undamped oscillation that never decays — a perfect, forever-ringing tone (marginally stable).
4. Pole in the right half-plane (σ > 0): e^(σt) grows without bound → the output explodes. Unstable — the circuit saturates, oscillates wildly, or destroys itself.

The one rule that keeps systems alive

Everything above condenses into a single, non-negotiable law of stability. A linear time-invariant system is stable if and only if every one of its poles lies in the left half of the s-plane (every pole has σ < 0). One pole in the right half-plane is fatal: that mode's e^(σt) term grows forever, and the slightest noise — always present — gets amplified until the output rails against the supply voltage or the op-amp clips. There is no 'small' instability; an unstable pole eventually dominates everything.

There is also a beautiful bridge back to rung 5. Stability is what *lets you* talk about frequency response in the first place: only when all poles sit safely in the left half-plane does the transient die away, leaving a clean steady-state sinusoid for the frequency response to describe. To evaluate that response you simply slide along the jω axis — set s = jω — and read |H(jω)| and ∠H(jω). Poles near the axis create resonant peaks; zeros near the axis create dips. The s-plane portrait and the Bode plot are two views of the same object.

Worked example: the second-order system

The second-order system is the workhorse of analog and control engineering — the RLC circuit you built above, a car's suspension, a motor with inertia. Its transfer function is written in a canonical form that every engineer memorizes, because two numbers in it set the *entire* behavior:

                 ωn²
   H(s) = ──────────────────────
           s²  +  2ζωn·s  +  ωn²

   ωn = natural frequency  (how fast it wants to oscillate)
   ζ  = damping ratio      (how quickly oscillation dies)

   Poles:  s = -ζωn  ±  ωn·√(ζ² - 1)

   ζ < 1 (underdamped):  s = -ζωn ± jωn√(1-ζ²)
         |                   \______/  \__________/
         |                    decay      ringing
         real part σ          rate       frequency ωd

Watch the poles march as you turn the damping knob ζ. This single parameter is the whole story of how the system responds to a step input — like suddenly commanding a motor to a new position:

1. ζ = 0 (undamped): poles sit on the jω axis at ±jωn. The step response oscillates forever at ωn — a bell that never stops. Marginally stable.
2. 0 < ζ < 1 (underdamped): complex pole pair in the left half-plane. The step response overshoots the target then rings with decaying oscillation. Smaller ζ → bigger overshoot, more ringing. This is the lively-but-bouncy regime.
3. ζ = 1 (critically damped): two real poles meet at -ωn. The fastest possible rise with *no* overshoot — the gold standard for a clean step.
4. ζ > 1 (overdamped): two real poles spread apart on the negative-real axis. No overshoot, but the slow pole near the origin dominates and the response crawls sluggishly to target.

Step response vs damping (output approaching target = 1.0):

  ζ=0.2  ____   __                      <- big overshoot, long ringing
        /    \ /  \____ ~~~ . . . . . . . . 1.0
       /      v
      /
  ____/

  ζ=0.7  ______________________________  <- small overshoot, settles fast
        /        \____________________ 1.0   (the engineer's sweet spot)
       /
  ____/

  ζ=1.0  ______________________________  <- no overshoot, monotonic
        /     ________________________ 1.0
       /  ___/
  ____/__/

  ζ>1   ________________________________ <- sluggish, no overshoot
        /              __________________ 1.0
       /         _____/
  ____/_________/

Why this unlocks the rest of the track

You have now met the central abstraction of analog, signal, and control engineering: take a real circuit full of capacitors and inductors, transform it into the s-plane, factor out its poles and zeros, and read its future off a chart. This one move replaces calculus with algebra, replaces guesswork about stability with a hard geometric rule, and connects the steady-state frequency world of rung 5 to the transient, switch-on world of real hardware.

From here the track branches naturally. The same H(s) machinery powers the Laplace-based convolution you will use to predict output for any input. It feeds straight into control theory — designing feedback loops and watching the closed-loop poles migrate as you crank the gain (the root-locus method is literally a movie of poles sliding across the s-plane). And it sets up the digital cousin — sampled systems live in the z-plane, where the left-half-plane stability rule becomes the unit circle. Every one of those topics is, at heart, the same conversation about where the poles sit.