Designing a 3rd Order Butterworth Filter: Theory, Math, and Bench Reality

A 3rd order Butterworth filter is a signal conditioning circuit that attenuates frequencies beyond its cutoff at a steep -60 dB per decade (-18 dB/octave) while maintaining a perfectly flat passband with zero amplitude ripple. In a real circuit, it bridges the gap between the weak noise rejection of standard 2nd-order Sallen-Key filters and the passband distortion of higher-order Chebyshev designs, giving you aggressive high-frequency roll-off without altering the amplitude of your target baseband signal.

Think of it like a highway toll plaza with three successive, increasingly strict checkpoints: the first checks basic ID (1st order), the second checks your cargo (2nd order), and the third ensures no contraband slips through. It is highly aggressive against out-of-band noise, but it doesn't damage or delay the legitimate cars (zero passband ripple).

The Anatomy of a Maximally Flat Roll-Off

The defining characteristic of any Butterworth polynomial is "maximal flatness" in the passband. Unlike Chebyshev filters, which trade passband ripple for a steeper transition band, or Bessel filters, which trade roll-off steepness for linear phase response, the Butterworth response is purely optimized for amplitude flatness up to the -3 dB cutoff frequency ($f_c$).

Because the roll-off is -60 dB/decade, a signal at 10 times your cutoff frequency ($10 \times f_c$) will be attenuated by 60 dB (a voltage reduction factor of 1000). This makes it a staple in mixed-signal designs where you need to guarantee that out-of-band noise floors are pushed below the least significant bit (LSB) of your analog-to-digital converter (ADC).

Worked Numeric Example: 1 kHz Low-Pass Design

Let’s design a 3rd order Butterworth LPF with a cutoff frequency ($f_c$) of exactly 1.0 kHz, using a unity-gain Sallen-Key topology for the 2nd-order stage and a simple buffered RC for the 1st-order stage. We will use standard 1% (E96 series) resistors and avoid the trap of using cheap ceramic capacitors.

Stage 1: The 1st-Order RC Buffer

The 1st order section requires a single pole at 1 kHz. We select a standard 10 nF capacitor.

• Formula: $R = \frac{1}{2 \pi f_c C}$
• Calculation: $R = \frac{1}{2 \pi (1000)(10 \times 10^{-9})} = 15,915 \Omega$
• Nearest 1% E96 Value: 15.8 kΩ

Stage 2: The 2nd-Order Sallen-Key (Unity Gain)

For a 3rd order Butterworth, the 2nd-order stage must have a quality factor ($Q$) of exactly 1.0. Using the equal-resistor Sallen-Key design equations for $Q=1$, the capacitor ratio $C_A / C_B$ must be 4.

• Capacitor Selection: Let $C_B = 4.7 \text{ nF}$ and $C_A = 18 \text{ nF}$ (Ratio $\approx 3.83$, close enough for 1% tolerance parts to yield $Q \approx 0.98$).
• Resistor Calculation: $R = \frac{1}{2 \pi f_c \sqrt{C_A C_B}} = \frac{1}{2 \pi (1000) \sqrt{18\text{n} \times 4.7\text{n}}} \approx 17,260 \Omega$
• Nearest 1% E96 Value: 17.4 kΩ for both $R_A$ and $R_B$.

Where You Meet This in Practice

You will rarely see a 3rd order Butterworth filter used for simple LED PWM smoothing (a 1st order RC is fine there). You meet this topology in precision signal chains where the cost of failure is corrupted data or audible distortion:

1. ADC Anti-Aliasing: Placed directly in front of a 12-bit or 16-bit SAR ADC to ensure high-frequency switching noise from nearby DC-DC converters cannot fold back into the baseband during sampling.
2. Audio Crossovers: Used in 3-way active studio monitors. The -18 dB/octave roll-off provides sufficient isolation between the woofer and midrange driver to prevent acoustic lobing and phase cancellation at the crossover point.
3. PLL Loop Filters: In phase-locked loops (like the CD4046 or modern RF synthesizers), a 3rd order passive or active Butterworth filter suppresses the reference frequency spur while maintaining a stable, ripple-free control voltage for the VCO.

Bench War Story: The Buck Converter Noise Disaster

Theory is clean; the bench is messy. Here is a scenario that highlights exactly why a 3rd order filter is sometimes the only way out of a noise floor problem.

The Setup

I was designing a seismic vibration logger using a geophone feeding an ESP32’s internal 12-bit ADC. The target signal was 2 Hz to 50 Hz. The ESP32 was sampling at 500 SPS (samples per second). Power was supplied by a cheap 12V-to-5V buck converter switching at roughly 150 kHz.

The Numbers

The geophone output was tiny: 10 mV to 50 mV peak. I initially used a standard 2nd-order Sallen-Key LPF (-40 dB/decade) with a 100 Hz cutoff to kill the 150 kHz switching hash. At 150 kHz, a 2nd order filter provides about 63 dB of attenuation.

The Outcome

The DSP output showed a mysterious, drifting 3 Hz sine wave that didn't exist in the physical vibration data. The 150 kHz noise, though attenuated by 63 dB, was still large enough to trigger the non-linear sampling switches inside the ESP32's ADC. The ADC aliased the 150 kHz carrier, and intermodulation distortion folded it down into the 0-250 Hz Nyquist band as a low-frequency beat.

What Went Wrong (And The Fix)

I confused "adequate attenuation" with "noise floor elimination." A 2nd order filter simply didn't provide enough stopband rejection for a noisy mixed-signal environment. I swapped the front end to a 3rd order active Butterworth using an OPA340 rail-to-rail op-amp. The extra -20 dB/decade pushed the 150 kHz attenuation past 90 dB, dropping the hash below the ADC's thermal noise floor. The phantom 3 Hz signal vanished instantly. For deep technical context on this exact failure mode, the Analog Devices guide on anti-aliasing is mandatory reading.

Common Confusions: Butterworth vs. Chebyshev vs. Bessel

People commonly confuse the Butterworth response with other standard polynomials, assuming "higher order" or "steeper" is always better. Here is how they actually compare on the bench.

If you are filtering a square wave or a digital pulse train, a Butterworth filter will cause ringing and overshoot due to its non-linear phase response. In that specific case, you want a Bessel filter, even though its roll-off is much weaker. For a comprehensive breakdown of the transfer functions, All About Circuits provides excellent interactive pole-zero plots.

Frequently Asked Questions