Passive vs. active filters: which to build, and the pitfalls by application

What's actually different

A passive filter is built from resistors, capacitors, and inductors alone. It needs no supply, only dissipates signal power, so its gain is always 1 or less. Whatever impedance sits in front of and behind it becomes part of its response.

An active filter is built from an op-amp plus R and C. Because the op-amp isolates (buffers) the source from the load, it's unaffected by what's on either side, and it can provide gain. It can also build a sharp second-order-or-higher response from R and C alone, with no inductor. In exchange, it needs a supply, and the op-amp's bandwidth, slew rate, output swing, and noise become the ceiling.

Put the difference in a table and the choice mostly makes itself.

The passive pitfall: the surrounding circuit changes the response

An RC low-pass's f_c = 1 ÷ (2πRC) assumes zero source impedance and an infinite load. Add a source resistance R_s and it adds to R; add a load R_L and it appears in parallel with C, shifting both the DC gain and f_c. A passive filter's response isn't fixed on its own — it's fixed only together with whatever it's connected to.

This shows up even more starkly in LC filters, where the termination resistance sets Q. Design an LC filter for a 50Ω system and connect it to a high-impedance circuit, and a peak appears in the passband, wrecking the cutoff response. That's why an EMI filter's datasheet says "attenuation in a 50Ω system" — a real power line's impedance (a few Ω to a few hundred Ω, varying with frequency) gives attenuation far from the catalog figure.

The same problem shows up cascading two stages. Stack the same RC twice, and the second stage loads the first, so you don't get −40dB/decade. Either make the second stage's R at least 10x the first, or put a buffer between them — at which point, it's an active filter.

The lack of gain is also easy to overlook. Cascade three passive stages and halve the signal, and a following amplifier needs 2x gain to make it up — at 2x the noise. Whatever a filter throws away in signal, recovering it later brings the noise back up with it.

The active pitfall: the op-amp's limits show up directly

GBW. Building a second-order Sallen-Key with f_c = 100kHz, Q = 0.7 wants an op-amp GBW at least 100x f_c (10MHz+). The higher the Q, and the more loop gain needed at f_c, the more margin required. Rule of thumb: GBW ≥ 100 × f_c × Q. A general-purpose op-amp with GBW = 1MHz is good for roughly f_c = 10kHz and below.

Attenuation floors out at high frequency. A Sallen-Key (VCVS) topology ties its feedback capacitor to the op-amp's output. The op-amp's output impedance rises at high frequency (tens to hundreds of Ω past the GBW), so from around 100x f_c the input starts leaking through that capacitor to the output — attenuation stalls around −40 to −60dB, and comes back up above that. In an ADC anti-aliasing filter, this leakage can let sampling-frequency harmonics through. The fix: add a small RC stage after the op-amp, or use a multiple-feedback (MFB) topology instead — an inverting topology with no such leakage path.

Slew rate and output swing. A large-amplitude signal near f_c needs slew rate SR ≥ 2π × f × V_peak — 3.1V/µs for a 5V peak at 100kHz. The output can also only swing to just inside the supply rails, so a single-supply design needs a mid-rail bias.

Component sensitivity. Build a sharp response like Q = 5, and a 1% error in R or C shifts Q and f_c by a few percent. Splitting the same response into lower-Q cascaded stages is more stable. Use C0G or film capacitors — X7R shifts with DC bias and temperature, dragging the response with it.

Noise and current draw. The op-amp's voltage noise (a few to a few tens of nV/√Hz) adds in. Low-current op-amps for battery use (µA class) only have tens of kHz of GBW, capping the frequencies usable for filtering.

By application: which to choose, and what to watch for

• ADC anti-aliasing. Active (2nd to 4th order). Decide the attenuation needed at half the sampling frequency first, then work backward to the order. Add a trailing RC to fix the Sallen-Key's high-frequency leakage. Pick an op-amp that can drive the ADC's sampling capacitance (able to drive 10–100nF plus tens of Ω at its output). A ΔΣ ADC's oversampling relaxes the requirement enough that a single RC stage is often sufficient.
• Power lines (DC-DC output, IC supply pins). Passive only. LC, or a ferrite bead plus C. An op-amp can't handle the current (the exception is a capacitance multiplier, a transistor circuit that makes a capacitor look bigger). An LC filter can resonate with load transients and worsen ripple, so pair it with a damping resistor or a capacitor with real ESR.
• RF (tens of MHz and up). Passive LC. Op-amp bandwidth isn't enough. Design assuming termination into an impedance (50Ω); in the physical build, an inductor's self-resonance and a capacitor's ESL end up setting the response.
• Sensor signals (a few Hz to tens of kHz, mV level). Active. With a high-impedance source, take it into the non-inverting input and combine gain and filtering in the same stage. Watch for offset and 1/f noise. If DC needs to pass, use a single-supply rail-to-rail part.
• Audio (20Hz–20kHz). Mostly active. Plenty of GBW headroom, and film capacitors to avoid ceramic nonlinearity (distortion). Passive tone circuits exist too, but expect the response to shift with the surrounding impedance.
• EMI suppression (cables, connectors). Passive only. Common-mode chokes, ferrite beads, Y capacitors. Attenuation swings a lot with the surrounding impedance, so measure it rather than trusting the catalog value.
• Low-current, battery-powered circuits. Passive if possible. If active, weigh the quiescent-current-vs-GBW tradeoff and pick the lowest-current op-amp that still hits the needed f_c. Ahead of a microcontroller's ADC, a single RC stage plus the MCU's built-in oversampling is often enough.
• Smoothing digital signals (debounce, edge softening). Passive RC, followed by a Schmitt-trigger input.
• Notch (removing 50/60Hz). Active (twin-T plus op-amp). A passive twin-T has Q = 0.25 — too broad. That said, the notch frequency drifts with temperature and component tolerance, so a digital filter (in the microcontroller) is often the more stable choice.

How to decide

1. Write down the signal's frequency band and the frequency you want to block. If the ratio between them is under 10x, a first-order filter isn't enough.
2. Check the signal's power. If current exceeds a few mA or voltage exceeds the supply rails, passive is your only option.
3. If the frequency is above 1MHz, make passive LC your first choice.
4. Check the impedance on both sides. If the source is high-impedance, or the load varies, go active (or passive plus a buffer).
5. If active, pick an op-amp with GBW at least f_c × Q × 100, and check slew rate, output swing, and quiescent current.
6. In the prototype, measure passband flatness, stopband attenuation (especially above 100x f_c), and distortion at large amplitude. Wherever it doesn't match the calculation, it's almost always one of these three.