Same order, three personalities
Guide 3 handed you the second-order Sallen-Key section and its one expressive dial, Q. But it left a question hanging: what value of Q should you choose, and once you cascade several sections for a higher order, what Q does each one get? That is exactly the question these three famous names answer. Butterworth, Chebyshev, and Bessel are not three different circuits — they are three recipes for the resistor and capacitor values (and therefore the per-section Q) that shape the same hardware into a different overall response.
Here is the deep idea underneath all three. A filter is forever trying to approximate an impossible ideal — the brick wall that passes everything up to the corner perfectly and nothing beyond it. You are given a fixed budget (the order) and three things you would love to have at once: a dead-flat passband, a steep transition, and clean timing (no smearing of the waveform's shape). You cannot maximize all three. Each family is a different decision about which to favour and which to sacrifice — the same compromise the guide-1 callout warned was unavoidable, now given three named answers.
Butterworth: maximally flat, the safe default
Butterworth spends its whole budget on one virtue: a passband that is as flat as mathematically possible. Its amplitude follows 1 / sqrt(1 + (f / fc)^(2n)), where n is the order — a curve that stays glued near full strength, never rising above it and never rippling, then bends down through exactly -3 dB at the corner before settling into its roll-off. People call it maximally flat, and it is the one you reach for when you have no special reason to do otherwise. A neutral, well-mannered all-rounder.
For a single second-order section, the Butterworth recipe is the famously round value Q = 0.707 (a damping of 1.414) — just gentle enough that the response neither peaks up at the corner nor sags early, the critically-flat setting you met as the well-behaved middle of the Q dial. Its roll-off is medium, its phase is moderate, and a step or square wave through it overshoots only slightly before settling. Nothing it does is best-in-class, and that is precisely why it is the default everyone starts from.
Chebyshev: trade ripple for a steeper cliff
Chebyshev makes a deal: it lets the passband ripple up and down by a small amount you choose in advance — typically 0.5 dB or 1 dB of wobble — and in exchange it bends downward harder, giving a noticeably steeper transition than a Butterworth of the same order. If you must reject a frequency that sits uncomfortably close to your passband edge and you cannot afford the extra stages a steeper Butterworth would need, this is the trade. The deeper you allow the ripple, the steeper the cliff you earn.
The bill comes in two parts. First, that passband is no longer flat — every signal in it is nudged up or down by the ripple, which matters if you care about amplitude accuracy. Second, the higher per-section Q (a 0.5 dB-ripple second-order section runs near Q = 0.86, and steeper designs push higher still) means more overshoot and more ringing on a fast transient, plus more phase distortion. The familiar 'standard' Chebyshev ripples in the passband and is smooth in the stopband; an inverse (type II) variant flips that, keeping a flat passband and rippling in the stop region instead.
Bessel: keep the waveform's shape intact
Bessel optimizes a quantity the other two ignore: phase. It aims for linear phase, which is the same as a constant group delay — every frequency in the passband is delayed by the same number of microseconds. Why care? Because if all the frequency components of a signal are held back by the same amount, they stay lined up, and a complex waveform comes out the far end with its shape preserved. A square wave through a Bessel emerges as a clean square wave with rounded corners and almost no overshoot or ringing — whereas the same square wave through a high-Q Chebyshev comes out wobbling.
Nothing is free, so Bessel pays with the laziest roll-off of the three: for the same order, its skirt is the gentlest and its corner the softest. Its second-order section runs at a low Q of about 0.58 — well damped, no peak, very forgiving. Reach for Bessel whenever the shape in time matters more than razor selectivity: passing data pulses or square waves without distorting them, audio paths where transient crispness counts, and any place where ringing would be mistaken for real signal.
Cascading to higher order — and a side-by-side
To go beyond second order you cascade sections: a fourth-order filter is two Sallen-Key stages in a row (an odd order adds one first-order section). The crucial subtlety — and a classic beginner trap — is that the stages do not all use the same cutoff and the same Q. The recipe deliberately staggers them. For a fourth-order Butterworth low-pass, both stages sit at the same corner frequency but one runs at Q = 0.54 (heavily damped) and the other at Q = 1.31 (peaked); the peaked stage lifts the response right where the damped stage is starting to sag, and together they sum to that famous flat passband with an 80 dB/decade skirt.
RESPONSE passband roll-off near step response reach for it when ...
shape the corner (square wave)
------------ ------------ -------------- --------------- -------------------------
Butterworth maximally medium slight you want a flat, neutral
flat overshoot default
Chebyshev ripples steepest rings the most must reject a nearby
(e.g. 0.5dB) tone; ripple is OK
Bessel gently gentlest almost none waveform shape / timing
drooping matters (pulses, audio)
Note: far down in the stopband ALL THREE roll off at the same 20 x n dB/decade.
The family only changes the behaviour NEAR the corner.
4th-order Butterworth low-pass = two Sallen-Key sections at the SAME cutoff:
section 1: Q = 0.54 (well damped, no peak)
section 2: Q = 1.31 (peaks up to fill in the flat top)
cascaded -> flat passband, 80 dB/decade beyond the cornerIn practice you never hand-derive those staggered Q values — you read them from a filter table or let a design tool spit them out, then size each section's parts to match. Two warnings come with high order. The peaked, high-Q stages are touchy: their corner and gain shift with component tolerance, so a sharp Chebyshev demands tighter resistors and capacitors than a lazy Bessel does. And each stage leans on its op-amp having enough gain-bandwidth left at the cutoff — a tired op-amp quietly drags the real response away from the textbook curve.
Choosing for a real job — honestly
Now map the families onto the jobs from guide 1. An anti-alias in front of a converter, or the reconstruction low-pass smoothing a DAC's staircase, usually wants a sharp cut just below half the sample rate, so Butterworth or Chebyshev earns its keep — unless you are passing pulses, where Bessel's clean timing wins. Audio crossovers lean Butterworth-ish for a smooth blend. Data and pulse paths lean Bessel. And to kill one fixed pest like 50/60 Hz mains hum you do not pick a family at all — you build a notch, often as a state-variable or twin-T section tuned to that single frequency.
So the honest decision procedure is short. First fix the order from how much rejection you need at how far past the corner — that sets your budget. Then choose the family by what you are willing to spend it on: flat amplitude (Butterworth), steepest cut for the order (Chebyshev, paying ripple and ringing), or faithful waveform shape (Bessel, paying a lazy skirt). The order buys the rejection; the family decides where the compromise lands.