From a gentle slope to a sharp corner
Guide 1 left us with a working but gentle filter: a single resistor and capacitor make a first-order low-pass that rolls off at only 20 dB per decade. That is fine for trimming hiss, but if you need to pass a 1 kHz signal and firmly reject a 2 kHz neighbour sitting just one octave away, a slope that takes a whole tenfold change in frequency just to drop the signal by a factor of ten is hopelessly slack. We need a steeper cliff, and the cheapest honest way to get one is to go from first- to second-order.
Doubling the order does two things at once. First, it doubles the asymptotic roll-off to 40 dB per decade — past the corner the signal now falls a hundredfold for every tenfold rise in frequency instead of tenfold. Second, and this is the genuinely new part, it unlocks a brand-new control over how the response behaves right at the corner itself. A first-order filter has no say in that shape; a second-order one does, and that freedom is everything.
Where does a second order even come from? Nature's own version is the LC resonant loop you met in the AC rung: an inductor and a capacitor trading energy back and forth give a true second-order resonance with a peak and a 40 dB/decade skirt. The trouble is inductors — at audio frequencies a useful one is bulky, expensive, lossy, and happy to pick up hum from every transformer in the room. The beautiful trick of the active filter is to throw the inductor away and let an op-amp fake the same second-order behaviour using nothing but resistors and capacitors. The most popular way to do that is the Sallen-Key.
Inside a Sallen-Key
A Sallen-Key section is almost embarrassingly simple: two resistors, two capacitors, and one op-amp wired as a buffer or a modest amplifier. Draw the low-pass version and you see what looks like two RC low-passes in a row — but with one wire that changes everything. Instead of grounding the first capacitor, you tie it to the op-amp's output. That single connection is positive feedback, and it is the whole secret of how a couple of capacitors learn to resonate like an inductor.
Sallen-Key LOW-PASS (unity-gain buffer)
node A node B
Vin --[R1]--+------[R2]------+------(+)\
| | \
| [C2] op-amp >--+--- Vout
| | / |
| GND (-)---/ |
[C1] | |
| +---------+ (-in tied to Vout: gain = 1)
| |
+-------------------------------+ (C1: top=node A, bottom=Vout)Trace it at the extremes. Far below the corner both capacitors are effectively open circuits — no current flows through them — so the signal strolls straight through the two resistors to the op-amp's input, and the op-amp (a unity-gain buffer in the simplest version) copies it to the output at full size. Far above the corner the capacitors are nearly short circuits, dumping the signal to ground and to the low-impedance output, so almost nothing survives: a clean low-pass. The interesting action is in between.
Right around the corner, the magic wire earns its keep. As the input starts to lag, the op-amp's output — a faithful copy of the input — drives current back through that first capacitor into the network, topping up the node just as it begins to droop. It is the electrical twin of pushing a child on a swing: a small shove perfectly in time with the motion builds a resonant peak that mimics exactly the energy-trading resonance of an LC loop, with no inductor in sight. And because the op-amp's output is stiff and low-impedance, the section does not sag when you hang the next stage on it — so you can cascade Sallen-Key blocks back-to-back without them fighting, the very loading problem that plagues passive sections.
Q and damping: the dial second-order unlocks
That resonant peak needs a name and a number, and you already met it for the RLC loop: the quality factor, Q. Q measures how lightly the section is damped — how vigorously it rings. Its mirror image is the damping factor, with the simple link damping = 1 / (2 times Q): high Q means low damping. At low Q (say 0.5) the response rounds off gently and stays well-behaved on transients. At high Q (say 5 or 10) the curve rears up into a tall peak just before the corner, and the filter rings like a struck bell when you hit it with a step. The single most important value sits in between: Q = 0.707 (that is 1 / sqrt 2) gives the maximally flat Butterworth response — the flattest possible passband with no peak at all — which is why it is the default choice for a plain low-pass.
In the equal-component Sallen-Key (both resistors equal, both capacitors equal) the corner and the Q split cleanly between two knobs. The R and C values set where the corner sits, while the op-amp's gain K sets how sharp it is, through the tidy relation Q = 1 / (3 - K). A unity-gain buffer (K = 1) gives Q = 0.5; nudge the gain up and Q climbs. Here is the honest edge of the trick, though: as K approaches 3, Q shoots toward infinity. At exactly K = 3 the positive feedback is no longer a controlled push but a runaway — the loop satisfies the Barkhausen condition and the 'filter' becomes an oscillator. A Sallen-Key lives its whole useful life with its gain held safely below 3.
Designing a Sallen-Key, step by step
Designing a second-order low-pass is now a short, honest recipe. Pick the corner, pick the Q (which guide 4 will turn into a named response), and let the equations hand you the parts. Let us build a 1 kHz Butterworth low-pass to keep it concrete.
- Set the target: corner f0 = 1 kHz, and Q = 0.707 for a Butterworth (flat) response.
- Choose equal capacitors in a sensible range, say C = 16 nF, and make both resistors equal at R. The corner obeys the same formula as the first-order one: f0 = 1 / (2 times pi times R times C).
- Solve for R: R = 1 / (2 times pi times f0 times C) = 1 / (2 times pi times 1000 times 16 times 10^-9), which is about 9950 ohm — reach for a standard 10 k.
- Set Q with the gain: Q = 1 / (3 - K) = 0.707 needs K = 1.586. Build that non-inverting gain with two more resistors, K = 1 + Rf / Rin, so Rf / Rin = 0.586 — for example Rin = 10 k and Rf = 5.9 k (a standard 5.6 k lands at K = 1.56, Q = 0.69, close enough for most jobs).
- Check the op-amp can keep up (next paragraph), and you are done: one chip, four resistors, two capacitors, and a clean 40 dB/decade low-pass.
One check separates a design that works from one that quietly misbehaves: the op-amp must be fast enough. Its gain-bandwidth product has to dwarf the corner — a common rule of thumb is GBW at least 100 times the corner frequency times the gain — or the op-amp's own falling response will drag your corner down and slump your Q. For our 1 kHz filter even a humble op-amp with 1 MHz GBW has a thousandfold margin, but push the corner to 100 kHz and you must choose more carefully. And remember the op-amp's slew rate: a fast, large-amplitude signal can slam into the slew limit and distort no matter how pretty the small-signal Bode plot looks. The op-amp's golden rules only hold inside its bandwidth and slew budget — outside them the filter is on its own.
Other shapes: high-pass, band-pass, and the notch
The same skeleton gives every response; you just move the parts. Swap each resistor for a capacitor and each capacitor for a resistor, and the low-pass becomes a second-order high-pass — lows now blocked, highs passed, with the same Q machinery. For a tuned band-pass the favourite is a different topology, the multiple-feedback (MFB) section, which wraps the signal around an inverting op-amp through two feedback paths. MFB inverts the signal and is a touch harder to picture, but it is less twitchy than Sallen-Key at high Q and is the standard choice when you want a sharp single-band peak.
When you need real tunability or a very high Q, reach for the state-variable filter: it spends two or three op-amps but hands you low-pass, high-pass, and band-pass outputs simultaneously from the same circuit, with the corner and Q adjustable almost independently. And the notch? You can build a deep one by subtracting a state-variable's band-pass output from the original signal, or with the classic twin-T network — exactly the surgical tool for killing 50/60 Hz mains hum that guide 1 promised. Each topology is a different trade of part count, tuning ease, and sensitivity; none escapes the underlying second-order physics.
Honest limits and stacking higher
What if 40 dB/decade still is not steep enough? Cascade. Put two second-order sections in series for a fourth-order filter at 80 dB/decade, three for sixth-order at 120 — the op-amp's stiff output lets each section drive the next without loading it. But here is the catch that guide 4 exists to resolve: the sections are not identical copies. To get a clean overall Butterworth (or Chebyshev, or Bessel) shape, each stage needs its own specific corner frequency and its own Q, read from a design table. Stack two naive Q = 0.707 sections and you do not get a fourth-order Butterworth — you get a lumpy response with a dip.
Step back and the picture is simple. A second-order section is the fundamental building block of every serious filter: it sets a corner with R and C, shapes that corner with a single dial called Q, and — in the Sallen-Key — conjures the resonance of an inductor from an op-amp and two capacitors. Master this one block and higher orders are just a stack of them. All that is left is choosing the flavour of Q and corner for each stage, which is exactly where guide 4 takes you.