A topic in the Open Knowledge Graph — a free, open map of 15,290 topics and the order to learn them in.
Filter order determines the steepness of the transition band; an Nth-order filter rolls off at approximately 20N dB/decade for Butterworth designs. Higher order filters have sharper transitions but require more computation and can introduce instability or ringing. The transition band width is bounded by the filter specifications and cannot be made arbitrarily small without increasing order.
Design low-, medium-, and high-order Butterworth filters with identical edge frequencies and measure their rolloff rates. Plot magnitude responses on log scale to see the asymptotic slopes.
When you specify a filter — "pass signals below 1 kHz, reject signals above 2 kHz" — you are defining a transition band: the frequency range from 1 to 2 kHz over which the filter moves from passing to rejecting. No physically realizable filter can make this transition instantaneously; the sharpness of the rolloff is governed by filter order. Understanding the quantitative relationship between order and transition band is the bridge between filter specifications (what the system requires) and filter design (what is physically achievable).
The core relationship for a Butterworth filter is that an Nth-order design rolls off at 20N dB/decade asymptotically in the stopband — equivalently, N × 6 dB/octave. A 4th-order filter rolls off at 80 dB/decade; a 10th-order at 200 dB/decade. The minimum order to meet a pair of specifications — passband ripple A_p dB, stopband attenuation A_s dB, with the ratio of stopband to passband edge frequencies Ω_s/Ω_p — is determined by N ≥ log((10A_s/10 − 1)/(10A_p/10 − 1)) / (2 log(Ω_s/Ω_p)). The key insight from this formula is in the denominator: log(Ω_s/Ω_p). Halving the transition band (moving Ω_s closer to Ω_p) barely changes the numerator but nearly doubles the log in the denominator, roughly doubling the required order. Specifying a very narrow transition band is expensive — an octave-wide transition band might need 4th order, while a tenth-octave transition band might need 40th order for the same attenuation.
Different filter families navigate the order-versus-transition-band tradeoff differently, each representing a different allocation of the same fixed "design budget." Butterworth filters are maximally flat in the passband — no ripple at all — but they roll off gradually near the band edge. Chebyshev Type I filters allow equal-amplitude ripple in the passband and use that tolerance to achieve a sharper transition at the same order; the passband oscillates slightly but the stopband attenuation arrives at a lower frequency. Elliptic (Cauer) filters allow ripple in both passband and stopband and achieve the steepest possible transition for a given order, but introduce transmission zeros (notches) in the stopband and have strongly nonlinear phase. The choice depends on whether passband flatness, stopband monotonicity, or phase linearity matters most for the application.
Beyond frequency response, high filter order introduces serious practical challenges. High-order IIR filters implemented in fixed-point arithmetic are sensitive to coefficient quantization: small rounding errors in the coefficients shift poles and zeros away from their designed locations, potentially moving poles outside the unit circle and causing instability. The standard remedy is to implement the filter as a cascade of second-order sections (biquads) — each biquad has only 5 coefficients and its poles are numerically well-separated, making coefficient quantization far less damaging. In analog hardware, each additional pole requires a reactive element, adding cost and size. The design discipline is straightforward: compute the minimum order that meets specifications, choose the filter family that best fits the application's constraints, and implement it in the most numerically stable form available.