Chebyshev Filters

The Chebyshev filter makes a deliberate trade: it accepts a controlled amount of ripple in the passband in exchange for a dramatically sharper rolloff. Named after the Russian mathematician Pafnuty Chebyshev, who studied equiripple approximation polynomials in the 1850s, this filter type allows the designer to specify exactly how much the passband can deviate from flat — and uses that allowance to push the transition band as steep as possible. The result is that a 5th-order Chebyshev with 0.5 dB ripple can attenuate harmonics 6–10 dB more than a 5th-order Butterworth with the same component count.

This makes the Chebyshev the usual choice for harmonic filters and anti-aliasing filters where the frequencies to be rejected sit close to the wanted ones. A 0.5 dB ripple introduces only minimal signal variation across the passband, yet it buys enough extra stopband rejection to meet a specification with far fewer components than a Butterworth. A filter that needs eight or nine Butterworth poles might need only five Chebyshev poles — fewer components, lower insertion loss, and a smaller board.

The Chebyshev Equiripple Response

The Chebyshev low-pass filter is defined by its magnitude-squared response:

The Chebyshev polynomials T_n(x) are defined recursively:

• T₀(x) = 1
• T₁(x) = x
• T_n(x) = 2x·T_n−1(x) − T_n−2(x)

For |x| ≤ 1 (inside the passband), T_n(x) oscillates between −1 and +1. This is the mathematical source of the equiripple behavior: within the passband, the response ripples up and down with constant amplitude, reaching exactly 0 dB and −R_dB alternately (where R_dB is the ripple). Every Chebyshev filter of order n has exactly n peaks and valleys in the passband. At the passband edge (x = 1), T_n(1) = 1 always, so the response always equals −R_dB at f = f_c.

For |x| > 1 (above the passband), T_n(x) grows rapidly — much faster than xⁿ. This rapid growth is why Chebyshev filters roll off so much more sharply than Butterworth filters of the same order.

The ripple factor ε is related to the passband ripple in decibels:

Chebyshev low-pass filter response (dashed, 0.5 dB ripple) compared to Butterworth (solid) of the same 5th order. The Chebyshev exhibits equiripple oscillations in the passband but achieves significantly greater attenuation above the passband edge frequency f_c.

View Larger

Choosing Passband Ripple

The ripple specification is the key design choice for a Chebyshev filter. More ripple means sharper rolloff but more amplitude variation in the passband. The table below summarizes the practical implications:

For harmonic filters, 0.5 dB ripple is the most common choice. The 0.5 dB amplitude variation is negligible in practice — a signal passing through will vary by at most ±0.5 dB (about ±6% in voltage) depending on exactly where it sits in the passband, and for a single-frequency signal that simply means a slightly different fixed level. Yet this small ripple buys enough extra stopband attenuation to significantly reduce the required filter order.

Normalized Component Value Tables

As with the Butterworth, Chebyshev component values are presented as normalized prototypes referenced to 1 ohm and ω_c = 1 rad/s. The values depend on both the filter order n and the chosen passband ripple. For equal source and load impedances (50 Ω), odd-order Chebyshev filters have equal terminations automatically; even-order filters have unequal terminations and require a transformer or resistive pad unless modified.

This is a critical practical difference from Butterworth: even-order Chebyshev filters (n = 2, 4, 6) have different source and load impedances in the normalized prototype. For this reason, odd-order designs (n = 3, 5, 7) are universally preferred for filters with matched 50 Ω source and load.

Chebyshev prototype values — 0.1 dB ripple (ε = 0.153):

Chebyshev prototype values — 0.5 dB ripple (ε = 0.349):

Chebyshev prototype values — 1.0 dB ripple (ε = 0.509):

Notice how the component values change with ripple. Higher ripple gives higher g values overall — which translates to larger component values, or alternatively, more reactive impedance variation across the passband (consistent with the ripple). The odd-order symmetry (g₁ = g_n) holds for all Chebyshev odd orders with equal terminations, just as with Butterworth.

Stopband Attenuation Formula

For a Chebyshev low-pass filter with passband ripple ε, the attenuation at any frequency f above the passband edge f_c is:

For frequencies well above the passband (f ≫ f_c), this simplifies to an asymptotic approximation — but for engineering calculations at specific harmonic frequencies, the full formula is needed. Many scientific calculators and filter design programs can evaluate cosh and arccosh directly. For hand calculation, use the identity: arccosh(x) = ln(x + √(x² − 1)) for x ≥ 1.

Quick comparison at f = 2·f_c (frequency ratio = 2.0):

The Chebyshev advantage grows with filter order. A 5th-order Chebyshev at 0.5 dB ripple provides 45 dB at twice the cutoff, in a 5-pole filter. The equivalent Butterworth needs 7 poles for 43 dB at the same frequency ratio. Two fewer components, a smaller board, and lower insertion loss.

Worked Design Example: 7 MHz Harmonic Filter

Let's design a Chebyshev low-pass filter to clean up the output of a 7 MHz RF signal source — an oscillator followed by a buffer amplifier delivering up to a few watts into 50 Ω, whose frequency can be set anywhere from 7.0 to 7.3 MHz. We need at least 43 dB of attenuation at the second harmonic. This time the tighter rolloff works in our favor — we can place the cutoff close to the passband while still achieving adequate harmonic suppression. We start by trying a 5th-order design.

Recommended physical layout for a 7-pole Chebyshev low-pass filter. The toroids alternate in orientation by 90° to minimize coupling, the shunt capacitors mount directly to the ground plane, and input and output are at opposite ends.

View Larger

Chebyshev vs. Butterworth: When to Use Each

Now that we have seen both filters in detail, the decision is straightforward:

For harmonic filters — one of the most common LC filter projects — the Chebyshev at 0.5 dB ripple is the standard choice. For any application involving digital signals, pulse waveforms, or precise group delay requirements, the Butterworth or Bessel is preferred.

Group Delay and Phase Distortion

The equiripple passband of the Chebyshev filter comes with a cost: its phase response is not linear, and its group delay varies across the passband. Group delay is defined as the negative derivative of phase with respect to frequency. When group delay is constant (flat), all frequency components of a signal experience the same time delay — the signal's shape is preserved. When group delay varies, different frequencies are delayed by different amounts, distorting the signal's envelope.

For a single-frequency signal such as an oscillator output, the group delay variation of a 7th-order Chebyshev is irrelevant — there is only one frequency to delay. For audio passing through a Chebyshev low-pass filter whose cutoff is far above the audio band, the group delay is essentially flat and the effect is inaudible. The Chebyshev is entirely suitable for harmonic filtering and for most audio and sensor filtering.

Where group delay becomes critical is in filters for digital data and pulses, and in measurement chains, where the filter operates at or near its cutoff with complex waveforms. A Chebyshev low-pass filter rings noticeably on a fast step — you can see it on an oscilloscope as overshoot and damped oscillation after each edge. In those cases, a Bessel filter (covered in the Filter Types Overview lesson) is appropriate.

Practical Construction Considerations

The Chebyshev filter has one notable difference from the Butterworth in construction: the outer inductors (L₁ and Lₙ) have relatively high g values (around 1.7 for 0.5 dB ripple), while the inner inductors have even higher values. This means the Chebyshev filter uses more inductance overall than the equivalent Butterworth, which translates to more turns on the toroid cores — but also means higher Q requirements to keep insertion loss low.

Below about 30 MHz, a well-wound T68-2 toroid in a well-designed ladder filter should have insertion loss under 0.3 dB across the passband for filter orders up to 7. The main insertion loss culprit is low-Q inductors with high wire resistance. For filters carrying more than a watt or so, use 22 AWG or heavier enameled copper wire to keep winding resistance low. Measure the Q of your inductors with an LCR meter or Q meter if possible; a Q above 150 at the operating frequency keeps each inductor's contribution to insertion loss below about 0.1 dB.