How it works
An LC low-pass filter puts an inductor in series with the signal and a capacitor across the output. Below the resonant frequency f0 = 1/(2π√(LC)) it passes the signal almost unchanged; above f0 it attenuates at 40 dB per decade, twice as fast as an RC filter, and it does so with almost no loss because neither L nor C dissipates power. That is why it is the standard output filter of buck converters, inverters and class-D amplifiers.
f0 alone does not describe the filter. The second number is the characteristic impedance Z0 = √(L/C), and what matters is how it compares with the load. Driven from a low-impedance source, the filter's quality factor is Q = R_load / Z0. A Q of 0.707 gives the flattest passband with no peak (Butterworth); 0.577 gives the best step response (Bessel); 0.5 is the fastest response without overshoot (critically damped). A light load or no load at all makes Q large, and the response peaks at f0. With Q = 5 the output is 14 dB higher at resonance than at DC.
The Design tab works backwards: give it the load resistance, the −3 dB frequency you want and the response shape, and it returns L and C, the nearest E12 parts and what those parts actually give you. The Analyze tab takes L and C you already have and returns f0, Z0, Q, the real −3 dB point, the peaking and the attenuation at any frequency, for example your switching frequency.
The detailed panel below models the parts as they really are: the capacitor ESR and the inductor DCR add damping, and an R_d–C_d leg across C can flatten the peak without burning power at DC. It also shows the Bode plot and the step response.
The terms are used loosely. An "LCR" or "RLC" filter calculator usually means this same circuit with its load or series resistance included, which is exactly what sets Q. If R is in series with L instead of across C (a series RLC), the damping is Q = Z0 / R instead, so a larger resistor damps more.
Design procedure
- 1Decide what you need to remove and what must pass, for example 400 kHz switching ripple and a 20 kHz audio band.
- 2Place f0 well below the frequency you want to remove. Every decade between f0 and that frequency gives about 40 dB.
- 3Pick the response shape. Butterworth (Q = 0.707) is the usual choice; Bessel or critical damping if overshoot matters more than flatness.
- 4Compute Z0 = R_load / Q, then L and C, and round to standard values. Check f0, Q and f_−3dB again with the rounded parts.
- 5Check the lightest load you expect, including no load. Q grows as the load gets lighter, so this is where the peak is largest.
- 6Choose real parts: inductor saturation current above the peak current, capacitor voltage rating and ripple current, and for ceramics the capacitance left at the DC bias.
Formulas
- Resonant (cutoff) frequency
- f0 = 1 / (2π · √(L · C))
- Characteristic impedance
- Z0 = √(L / C)
- Transfer function with load R
- H(s) = 1 / (s²LC + sL/R + 1), Q = R / Z0 = R · √(C / L)
- Magnitude
- |H| = 1 / √((1 − x²)² + (x / Q)²), x = f / f0
- −3 dB frequency
- f_−3dB = f0 · √(a + √(a² + 1)), a = 1 − 1/(2Q²) (= f0 for Q = 0.707)
- Peaking (Q > 0.707)
- |H|max = Q / √(1 − 1/(4Q²)) at f = f0 · √(1 − 1/(2Q²))
- Design from a load
- f0 = f_c / (f_−3dB/f0), Z0 = R / Q, L = Z0 / (2πf0), C = 1 / (2πf0 · Z0)
- Series RLC (R in series with L)
- Q = Z0 / R
Worked example
Class-D amplifier into an 8 Ω speaker with a bridge-tied load: each half of the output filter sees half the speaker, 4 Ω. For a 40 kHz Butterworth response Z0 = 4 Ω / 0.707 = 5.66 Ω, so L = 5.66 Ω / (2π · 40 kHz) = 22.5 µH and C = 1 / (2π · 40 kHz · 5.66 Ω) = 703 nF. The nearest E12 parts are 22 µH and 680 nF, which give f0 = 41.1 kHz, Q = 0.703 and f_−3dB = 40.9 kHz: close enough. At a 400 kHz switching frequency the filter attenuates by about 40 dB. If the speaker is unplugged, Q rises without limit and the response peaks sharply near 41 kHz, so check that case too.
Butterworth L and C for common loads (class-D output filters)
L and C for a flat (Q = 0.707) response into a resistive load, the nearest E12 parts, and what those parts give. For a bridge-tied load, use half the speaker impedance for each half of the filter: 2 Ω for a 4 Ω speaker, 4 Ω for 8 Ω.
| Load R | f_c | L | C | E12 L / C | f_−3dB (E12) | Q (E12) |
|---|---|---|---|---|---|---|
| 2 Ω | 30 kHz | 15 µH | 1.88 µF | 15 µH / 1.8 µF | 30 kHz | 0.693 |
| 2 Ω | 40 kHz | 11.3 µH | 1.41 µF | 12 µH / 1.5 µF | 37.5 kHz | 0.707 |
| 2 Ω | 50 kHz | 9 µH | 1.13 µF | 8.2 µH / 1.2 µF | 54.6 kHz | 0.765 |
| 4 Ω | 30 kHz | 30 µH | 938 nF | 33 µH / 1 µF | 27.3 kHz | 0.696 |
| 4 Ω | 40 kHz | 22.5 µH | 703 nF | 22 µH / 680 nF | 40.9 kHz | 0.703 |
| 4 Ω | 50 kHz | 18 µH | 563 nF | 18 µH / 560 nF | 50 kHz | 0.706 |
| 8 Ω | 30 kHz | 60 µH | 469 nF | 56 µH / 470 nF | 32.1 kHz | 0.733 |
| 8 Ω | 40 kHz | 45 µH | 352 nF | 47 µH / 330 nF | 38.2 kHz | 0.670 |
| 8 Ω | 50 kHz | 36 µH | 281 nF | 39 µH / 270 nF | 46 kHz | 0.666 |
A loudspeaker is not a resistor: its impedance rises at high frequency, which raises Q. Check the response with the real load, or at least with no load.
Second-order LC response vs Q (dB)
Gain in dB of an LC low-pass at multiples of f0 for several values of Q. The column for Q = 0.707 is Butterworth; Q = 2 and 5 show a lightly loaded filter.
| f / f0 | Q = 0.5 | 0.577 | 0.707 | 1 | 2 | 5 |
|---|---|---|---|---|---|---|
| 0.5 | −1.9 | −1.2 | −0.3 | +0.9 | +2.0 | +2.4 |
| 0.8 | −4.3 | −3.1 | −1.5 | +1.1 | +5.4 | +8.1 |
| 1 | −6.0 | −4.8 | −3.0 | 0.0 | +6.0 | +14.0 |
| 1.25 | −8.2 | −7.0 | −5.4 | −2.7 | +1.5 | +4.2 |
| 2 | −14.0 | −13.2 | −12.3 | −11.1 | −10.0 | −9.6 |
| 3 | −20.0 | −19.6 | −19.1 | −18.6 | −18.2 | −18.1 |
| 5 | −28.3 | −28.1 | −28.0 | −27.8 | −27.7 | −27.6 |
| 10 | −40.1 | −40.0 | −40.0 | −40.0 | −39.9 | −39.9 |
| 20 | −52.1 | −52.1 | −52.0 | −52.0 | −52.0 | −52.0 |
Far above f0 the curves converge on 40 dB per decade regardless of Q; Q only changes the region around f0.
Common mistakes
- Designing for f0 only. The same f0 can be a flat filter or a 20 dB peak, depending on L/C and the load.
- Assuming the load is a resistor. A loudspeaker's impedance rises well above its nominal value at high frequencies, and a converter load can be anything from open circuit to a step, so the real Q differs from the design Q.
- Forgetting that ceramic capacitors lose a large part of their capacitance at DC bias, which moves f0 up.
- Using RF filter tables (doubly terminated) for a filter driven by a switching stage or op-amp, or the other way around.
- Ignoring the inductor's saturation. Near saturation L drops, f0 moves up and the ripple current grows.
Frequently asked questions
What is the cutoff frequency of an LC filter?
The resonant frequency f0 = 1/(2π√(LC)) is usually quoted as the cutoff. It equals the −3 dB point only when Q = 0.707 (Butterworth). With a heavier load (lower Q) the −3 dB point falls below f0; with a lighter load it moves above f0 and the response peaks.
How do I choose L and C for a given cutoff frequency?
f0 fixes only the product LC. The ratio comes from the load: pick Z0 = √(L/C) = R_load / Q, with Q = 0.707 for a flat response. Then L = Z0/(2πf0) and C = 1/(2πf0·Z0). The Design tab does this for you.
Why does my LC filter peak or ring?
At resonance the only things that absorb energy are the load, the capacitor ESR and the inductor DCR. A light load and low-ESR ceramic capacitors give a high Q, which shows up as a peak in the frequency response and ringing on a step. Add a damping leg across C (see the detailed panel) or accept some loss with a series resistor.
How much does an LC filter attenuate?
Well above f0 the attenuation is about 40·log10(f/f0) dB: 40 dB one decade above f0 and 80 dB two decades above. In practice the inductor's winding capacitance and the capacitor's ESL limit this at high frequencies.
Is an RLC or LCR filter calculator different?
Usually not. Those names refer to the same second-order filter with the resistance included. What matters is where R sits: across the capacitor (the load), Q = R/Z0; in series with the inductor, Q = Z0/R.
Does this work for RF filters with 50 Ω on both sides?
Not directly. The formulas here assume a low-impedance source. A doubly terminated filter (50 Ω source and load) uses different element values from filter tables, because the source resistance also damps the circuit.
References
- R. W. Erickson, D. Maksimović, "Fundamentals of Power Electronics," 2nd ed., Kluwer, 2001, ch. 8 (second-order responses, Q and peaking) and ch. 10 (input filter damping).
- A. B. Williams, F. J. Taylor, "Electronic Filter Design Handbook," 4th ed., McGraw-Hill, 2006 (Butterworth and Bessel responses, element values).
Results are engineering estimates. Verify your design with datasheets, simulation and measurement.