A topic in the Open Knowledge Graph — a free, open map of 15,290 topics and the order to learn them in.
Butterworth filters maximize passband flatness by placing poles on a circle in the s-plane with monotonic magnitude response. The order determines rolloff rate (20 dB/decade per order) and passband ripple (zero). Pole locations follow standard normalized tables, and designs scale easily to any cutoff frequency or implementation topology.
From your work on transfer functions and Bode plots, you know that a filter's frequency response is shaped entirely by its pole and zero locations in the s-plane. The closer a pole is to the imaginary axis, the more it boosts nearby frequencies; poles on or beyond the imaginary axis cause instability. The design problem for a lowpass filter is: given a desired frequency response shape (flat passband, sharp rolloff, attenuated stopband), where should you place the poles? The Butterworth filter answers this with an elegant geometric principle: place all N poles evenly spaced on a circle of radius ω_c in the left-half s-plane.
The resulting magnitude response is |H(jω)|² = 1 / (1 + (ω/ω_c)2N). At ω = 0, the response is exactly 1 (0 dB). At ω = ω_c, the response is 1/√2 (−3 dB) regardless of order. Beyond ω_c, the response rolls off monotonically — no ripple, no bumps, just a smooth decay. This maximally flat property is what the Butterworth optimization achieves: among all polynomial filters of order N with −3 dB at ω_c, the Butterworth has the maximum number of derivatives equal to zero at ω = 0, meaning the passband deviates from unity as slowly as possible. The tradeoff is that this flatness comes at the cost of rolloff sharpness: a Butterworth is less steep in the transition band than a Chebyshev or elliptic filter of the same order.
The rolloff rate follows directly from the mathematics: |H(jω)| ≈ (ω_c/ω)N for ω ≫ ω_c. Each decade of frequency above ω_c attenuates by a factor of 10N, which is 20N dB/decade. A first-order Butterworth gives −20 dB/decade; second order −40 dB/decade; and so on. This is why filter order is a design variable: if your stopband attenuation requirement is 80 dB at one decade above the cutoff, you need at least a fourth-order filter. From your Bode plot knowledge, you can also recognize this as N poles clustered near ω_c all contributing approximately −20 dB/decade each in the rolloff region.
The design procedure is standardized through normalized prototype tables. The normalized Butterworth lowpass prototype has ω_c = 1 rad/s; its pole locations for order N are s_k = e^(jπ(2k+N−1)/2N) for k = 1, 2, ..., N (only the left-half plane poles are used). For N = 2, the poles are at ±j·ejπ/4 = (−1/√2) ± j(1/√2), giving the familiar second-order transfer function with Q = 1/√2 ≈ 0.707 (the Butterworth Q). To design a filter for a different cutoff frequency, you scale the prototype poles by ω_c. To implement in hardware (an op-amp active filter or an LC ladder network), you map the pole locations to component values using standard topologies like Sallen-Key or multiple-feedback. To convert to a digital filter, you apply the bilinear transform — which you'll encounter as a follow-on topic. The prototype-then-scale workflow makes Butterworth filters among the most routinely implemented analog filter designs in instrumentation and audio engineering.