How to work it 01 Pick family and kind IIR or FIR, then lowpass, highpass, bandpass or bandstop; band kinds reveal F2. 02 Enter the numbers. For IIR the order field counts biquads (total order = 2×); FIR counts taps−1. Then Fs and Fc — or F1 and F2. 03 Read the spec line and table characteristic, kind, section count and frequency mapping, then b0, b1, b2, a1, a2 per section (or the taps). 04 Check the order sweep , which redesigns orders 1–12 in place and shows one metric per section count; failed orders show a dash. 05 Copy the C or Python block , or send H(z) onward with the sample rate attached.
Worked readouts Every figure is the exact displayed value; the same inputs reproduce them.
Design
Readout
IIR butterworth bandpass, 2 biquads, Fs 48000, F1 300, F2 3400
spec IIR butterworth bandpass, 2 biquad(s), Fs=48000, F1=300, F2=3400 (Fc=1009.95 Hz, BW=3.503 oct) ; both sections 0.168856, 0, -0.168856, -1.64778, 0.662288; sweep -3 dB bandwidth (Hz): order 1 3050.9 , order 2 1976.41 , order 12 747.446
the same edges as bandstop
sections 0.831144, -1.64778, 0.831144, -1.64778, 0.662288; sweep Max stopband atten. (dB): 37.3411 , 74.6821 , 448.093 at orders 1, 2, 12
IIR butterworth highpass, 2 biquads, Fs 48000, Fc 300
spec IIR butterworth highpass, 2 biquad(s), Fs=48000, Fc=300 ; sweep -3 dB cutoff (Hz): 308.141 , 314.435 , 354.394 at orders 1, 2, 12
How the sweep decides Each sweep point rebuilds the filter at that section count and reads its 256-point response: −3 dB crossings by linear interpolation, the deepest stopband point for bandstop, and a failed mark when the response never crosses. The metric label follows the kind, and fixing everything else is what makes the column comparable across orders.
Assumptions and limits
Tables are rounded; the sweep is interpolated. Coefficients show 6 significant digits and crossings come from a 256-point response, so treat the last digit as orientation.
FIR parity is a hard rule. Highpass, bandpass and bandstop FIR orders must be even — the spectral inversion needs a centre tap — and an odd order stops with an error rather than returning the wrong shape.
The band edges you type are not the band width you get Band designs are converted before any coefficient is produced: the centre becomes the geometric mean √(F1·F2) and the width becomes log2(F2/F1) octaves. F1 = 300 and F2 = 3400 at 48 kHz store Fc=1009.95 Hz, BW=3.503 oct in the spec line — a geometric centre, not an arithmetic midpoint — and the sweep measures what the cascade really does: 1976.41 Hz of realised −3 dB bandwidth at two biquads, 3050.9 at one and 747.446 at twelve.
Where it fits Prototyping a speech bandpass The telephone band at 48 kHz is the table's example: two biquads, five coefficients each, ready for C or Python. The sweep prices the trade across one and twelve biquads.
Choosing between kinds and families Switch the same edges to bandstop and the metric becomes attenuation: 74.6821 at two biquads, 448.093 at twelve. The control panel accepts the resulting H(z) for loop analysis, and the phase portrait panel handles dynamics the filter view does not.
Privacy Filter design and the sweep run entirely in your browser; the specification you enter is never uploaded.
References
R. Bristow-Johnson, Cookbook formulae for audio EQ biquad filter coefficients , webaudio.github.io (访问日期:2026-10-01)— biquad formulas and the octave-bandwidth parameter.
Wikipedia, Butterworth filter , en.wikipedia.org (访问日期:2026-10-01)— maximally flat passband and −3 dB cutoff.
Wikipedia, Digital biquad filter , en.wikipedia.org (访问日期:2026-10-01)— second-order sections and cascade structure.
Wikipedia, Finite impulse response , en.wikipedia.org (访问日期:2026-10-01)— windowed-sinc design and tap symmetry.
Calculators in this hub
Hand-picked tools, one click away. The mini versions compute live and carry your values into the full calculator.
Sources & review
Reviewed by CalcX Editorial Team
Updated 2026-10-01