Filter Design

Butterworth Filter

/but-er-werth fil-ter/ — Maximally Flat
An analog filter with maximally flat magnitude response in the passband. Transfer function: |H(jω)|² = 1/(1 + (ω/ωc)2N). Roll-off: 20N dB/decade (6N dB/octave). Zero passband ripple with all derivatives zero at DC. Poles equally spaced on the s-plane unit circle. Prototype g-values: gk = 2 sin(π(2k-1)/(2N)) define LC ladder element values. Used for anti-aliasing, IF filtering, and interstage matching where flatness and group delay are priorities.
Ripple: 0 dB (flat)
Roll-off: 20N dB/dec
@ ωc: -3 dB

Understanding Butterworth Filters

Stephen Butterworth published his filter design in 1930, seeking the flattest possible frequency response. The Butterworth approximation achieves this by forcing all derivatives of the magnitude-squared function to zero at DC. This "maximally flat" property means the filter responds as uniformly as possible to signals within the passband, introducing no amplitude ripple that could distort wideband signals. The tradeoff is a more gradual transition from passband to stopband compared to Chebyshev or elliptic designs of the same order.

In RF engineering, Butterworth filters are the starting point for filter synthesis. The normalized prototype g-values provide a universal set of element values that can be scaled to any cutoff frequency and impedance level. For a low-pass to bandpass transformation, the g-values determine the coupling coefficients and resonator Q requirements. The Butterworth response is also the foundation for distributed filter design: coupled-line, hairpin, and interdigital bandpass filters can all be synthesized starting from Butterworth (or Chebyshev) prototypes using impedance inverter theory.

Butterworth Design Equations

Magnitude response:
|H(jω)|² = 1 / (1 + (ω/ωc)2N)
@ ω = ωc: |H| = -3 dB
@ ω = 2ωc: |H| = -10 log(1+22N) dB

Order selection:
N = ⌈log(10As/10 − 1) / (2 log(ωsc))⌉
As = required stopband attenuation (dB)

Prototype g-values:
gk = 2 sin(π(2k−1)/(2N))
g0 = gN+1 = 1
N=3: g = {1, 1, 2, 1, 1}
N=5: g = {1, 0.618, 1.618, 2, 1.618, 0.618, 1}

Impedance/frequency scaling:
Lk = gk × Z0 / (2πfc)
Ck = gk / (Z0 × 2πfc)

Filter Approximation Comparison

Filter TypePassbandTransitionGroup DelayOrder for 40 dB @ 2×fc
ButterworthMaximally flatGradualGoodN = 7
Chebyshev I (0.5 dB)EquirippleSharperModerateN = 5
Chebyshev IIFlatSharpModerateN = 5
Elliptic (Cauer)EquirippleSharpestPoorN = 4
BesselNon-flatVery gradualBest (linear)N = 10+
Common Questions

Frequently Asked Questions

What makes Butterworth "maximally flat"?

The first 2N-1 derivatives of |H(jω)|² are zero at DC, making the response as flat as mathematically possible for order N. Poles are equally spaced on the left half of the s-plane unit circle. No passband ripple, monotonically decreasing magnitude. At ω_c: exactly -3 dB. The smoothest transition from passband to stopband for any given order.

How do you determine filter order?

From attenuation specification: N = ceil(log(10^(A_s/10) - 1) / (2 log(ω_s/ω_c))). Example: 40 dB at 2×f_c requires N = 7. Then g-values g_k = 2 sin(π(2k-1)/(2N)) define the LC ladder network. Scale to target impedance and frequency: L = g×Z_0/(2πf_c), C = g/(Z_0×2πf_c).

When to use Butterworth vs. Chebyshev?

Butterworth: flattest passband, best group delay, simplest design. Use for anti-aliasing, IF filters, wideband matching where flatness matters. Chebyshev I: allows passband ripple (0.01-3 dB) for sharper transition, saves 2+ elements for same stopband rejection. Elliptic: sharpest cutoff but worst group delay. For narrowband RF bandpass (duplexers, channel filters), Chebyshev or elliptic dominate.

RF Filter Design

Request a Quote

Need custom filter design, prototype synthesis, or filter integration services? Contact our team.

Get in Touch