The z-transform & pole-zero design

The Fourier view asks what a system does to pure tones. The z-transform widens the lens to growing and decaying tones too — and in return hands you a map (the z-plane) where a filter's entire character is a constellation of poles and zeros you can place by hand.

Build the intuition

Probe with spirals, not just circles

Fourier probes use e^{iωn} — tones of constant size. The z-transform probes with zⁿ for any complex z = re^{iω}: spirals that rotate while growing (r > 1) or shrinking (r < 1). A difference equation responds to zⁿ with a simple multiplier H(z) — the transfer function, a ratio of polynomials in z. (Its continuous-time twin, the Laplace transform, plays the same role for differential equations.)

H(z) = \frac{B(z)}{A(z)}

Poles and zeros: the constellation

Zeros are z-values where H vanishes — frequencies the filter annihilates. Poles are where H blows up — frequencies it amplifies and rings at; they're the roots of the feedback polynomial, the same r and θ from the difference-equations lesson. The frequency response is read by walking the unit circle: gain rises near poles, dives near zeros. Filter design = sculpting with a constellation.

Stability at a glance

Each pole contributes a response term rⁿ — so the system is stable exactly when every pole sits inside the unit circle. On the map, stability is visible: poles in, safe; poles out, divergence; poles hugging the rim, near-resonance (sharp filters live dangerously). One picture replaces pages of analysis.

\text{stable} \iff \text{all poles } |z| < 1

See it move

InteractivePoles: a system's destiny in two numbers

Radius r0.9

Angle θ0.5

y[n] = 0.9ⁿ · cos(0.5·n) — radius r = 0.9 sets decay, angle θ = 0.5 sets the ringing frequency. Inside the unit circle: it oscillates while dying away. Stable, with character.

You are dragging a pole: r toward the rim sharpens the ring toward resonance; θ tunes its frequency. Cross r = 1 and the design fails — audibly, in real filters.

A worked example

Design a 50 Hz hum-killer

Goal: erase mains hum at 50 Hz from a 1,000 Hz-sampled signal. 50 Hz lives at angle ω = 2π·50/1000 = 0.1π on the unit circle.
Place zeros exactly there: z = e^{±i0.1π} — H is now zero at 50 Hz. The hum dies completely.
Trouble: nearby frequencies also dip. Cure: place poles just inside the circle at the same angle (r ≈ 0.95) to prop the response back up everywhere except the pinpoint.
Result: a notch filter — surgical, stable, four numbers. This exact design runs in medical and audio gear everywhere.

Out in the world

From EQ knobs to flight controllers

Parametric EQ bands are pole-zero pairs whose “frequency” and “Q” knobs move the constellation. Flight and engine controllers are pole-placement exercises with safety margins — keeping poles comfortably inside the circle as conditions vary. The z-plane is where digital behavior gets engineered.

Common confusion, cleared

“The z-transform replaces the Fourier view.”

It contains it: evaluate H(z) on the unit circle (r = 1) and the frequency response appears. The z-plane is the surrounding territory — including the off-circle behavior that explains transients and stability.

“Poles near the unit circle are simply better (sharper) filters.”

Sharper, yes — and longer-ringing, more sensitive to rounding errors, and one coefficient-quantization away from instability. Pole placement is engineering judgment, not maximization.

Recap

z-transform: probe systems with spirals re^{iωn}; H(z) is a polynomial ratio.
Zeros kill frequencies; poles amplify and ring; the unit-circle walk reads the response.
Stability = all poles inside the unit circle — visible at a glance.

Progress saves in this browser.