PowerElecHub

LC Output Filter for a Three-Phase Inverter: Ripple Current, Capacitor and Resonance

By the Power Electronics Hub editor · Published 2026-11-14

A 10 kVA inverter runs from a 700 V DC link, switches at 10 kHz with space-vector PWM and has to deliver a clean 400 V, 50 Hz three-phase output to a resistive or motor load. Between the bridge and the terminals sits an LC filter: one inductor per phase and a star of capacitors. The inductor is set by how much ripple current the switches and the core can take, the capacitor by how much reactive current the inverter can afford to carry at no load, and the resonance that results has to be checked rather than chosen. For this example, 2.2 mH and 10 µF give 4.3 A of peak-to-peak ripple, a capacitor current of 5 % of rated, a resonance at 1.07 kHz and 0.4 % voltage THD at the output.

What sets the inductor

Each bridge leg toggles between the DC rails, so the voltage across the filter inductor is a pulse train whose average is the wanted sine wave. The inductor turns the difference into a triangular ripple current on top of the fundamental. That ripple sets the peak current the switches must turn off, adds core loss and copper loss at the switching frequency, and has to be absorbed by the capacitor. A common target is a ripple of 20 to 40 % of the rated peak current; this example uses 20 % because the load includes motors that should see a quiet voltage.

With the capacitor voltage treated as a smooth sine wave, the ripple in one switching period can be worked out exactly from the switching pattern. With centred SVPWM the worst period of the fundamental cycle is where one phase crosses zero. At that point the phase voltage steps between +Vdc/3 and −Vdc/3, and the peak-to-peak ripple works out to m·Vdc/(4√3·L·f_sw), where m is the modulation index referred to Vdc/2. Since m·Vdc/2 is the peak phase voltage V̂, the DC link drops out.

Rated phase voltage and current
V_ph = V_LL/√3, I = S/(3·V_ph), V̂ = √2·V_ph, Î = √2·I
Modulation index (referred to Vdc/2)
m = V̂ / (Vdc/2), SVPWM linear up to m = 2/√3 ≈ 1.155
Worst-case ripple, SVPWM, m ≥ about 0.6
ΔI_pp,max = V̂ / (2√3 · L · f_sw)
Inductance for a ripple target
L ≥ V̂ / (2√3 · f_sw · ΔI_pp,target)

For 400 V line to line, V̂ is 326.6 V and the rated peak current of 10 kVA is 20.4 A, so a 20 % target is 4.08 A. The formula asks for at least 2.31 mH at 10 kHz. This design takes 2.2 mH, a value that is easier to source, and accepts 4.29 A, which is 21 %.

Why the DC link does not appear

Many design notes size the inductor with ΔI_pp ≤ Vdc/(6·L·f_sw). That expression is the SVPWM ripple at the edge of linear modulation, m = 2/√3, so it is always safe, but it is only reached when the DC link is as low as it can be for the required output. At 700 V this inverter runs at m = 0.93, and the bound gives 5.30 A for 2.2 mH against an actual worst case of 4.29 A. Sizing to the bound for the same 20 % target would mean 2.86 mH, about a quarter more inductance, copper and weight than the converter needs.

The result also says that a higher DC link does not raise the worst-case ripple with SVPWM, as long as m stays above about 0.6. For this filter the ripple stays at 4.29 A for any DC link from 650 V to 1000 V. With plain sine-triangle PWM the same filter sees 4.29 A at 700 V but 5.00 A at 1000 V, because without the common-mode injection the worst period moves to a different point of the cycle once m drops below about 0.85. Below m ≈ 0.6 the simple SVPWM formula slightly underestimates, but the absolute ripple there is smaller anyway.

Modulation index m
0.933
Rated peak current
20.4 A
Worst ripple (SVPWM)
4.29 A (21 %)
Worst ripple (sine-triangle)
4.29 A
Bound Vdc/(6·L·f_sw)
5.30 A
C per phase (star)
9.95 µF
Capacitor current
0.72 A
Resonance f0
1076 Hz
Attenuation at f_sw
-38.6 dB
ωL·I drop
4.3 %

The chart computes the worst switching period over a full fundamental cycle for SVPWM and for sine-triangle PWM at each modulation index, using the exact switching instants of a centred carrier. The dotted line is the Vdc/(6·L·f_sw) bound and the dot is the operating point from the voltage and DC link you enter. Raising the DC link scales both curves up and moves the dot to a lower m, but the dot stays at the same height: with SVPWM the ripple in amperes does not change. Read the sine-triangle curve at the new m and it has risen.

The formulas treat the capacitor voltage as a pure sine wave. A frequency-domain simulation of the complete filter with a 16 Ω per-phase load gives 4.31 A of inductor ripple, 0.7 % above the formula; the difference is the switching ripple on the capacitor itself.

What sets the capacitor

The capacitor draws its current at the fundamental frequency whether or not a load is connected, and the inverter has to supply it. At no load that current is all the bridge carries, so it sets the idle conduction loss, and at full load with a lagging power factor it adds to the current the inductor and switches have to handle. A limit of a few percent of rated reactive power is common; Peña-Alzola and co-authors size the capacitor of a grid filter so that it does not reduce the power factor by more than 5 %. The same 5 % gives C = 0.05·S/(3·ω·V_ph²) per phase in star.

Capacitor per phase, star
C = q·S / (3·ω·V_ph²), q = reactive fraction (0.05 here)
No-load capacitor current
I_C = ω·C·V_ph
Resonance and characteristic impedance
f0 = 1/(2π·√(L·C)), Z0 = √(L/C)
Attenuation of the switching ripple
|v_out / v_bridge| ≈ (f0 / f_sw)² for f_sw ≫ f0

For 10 kVA at 230 V per phase (400 V line to line) and 50 Hz the 5 % capacitor is 9.95 µF, so 10 µF per phase in star, or 3.3 µF per phase if the capacitors are connected in delta. Each draws 0.72 A, which is 5 % of the 14.4 A rated current. With 2.2 mH, f0 is 1.07 kHz and Z0 is 14.8 Ω, and at 10 kHz the bridge voltage ripple is reduced to about 1.2 %, or −38.7 dB. The chart above keeps the unrounded 9.95 µF, so it shows 1076 Hz and −38.6 dB.

Checking the result

The same simulation gives the numbers that matter at the terminals. At rated resistive load the output voltage carries 4.5 V of peak-to-peak switching ripple, about 1.4 % of the 327 V peak, and the voltage THD is 0.40 %. At no load the THD is 0.43 % and the inverter carries only the 0.73 A capacitor current plus the 4.3 A ripple, so its peak current is about 3.2 A.

The inductor also takes a share of the fundamental voltage. At rated current ωL·I is 10.0 V, 4.3 % of the phase voltage. With a resistive load the bridge has to produce 231.4 V per phase, a modulation index of 0.935. With a 0.8 lagging power factor the drop adds more directly to the load voltage and the bridge needs 237.1 V, m = 0.958. SVPWM reaches 1.155 at 700 V, so there is headroom for DC link sag and transients.

Two alternatives show the trade-offs. Doubling the switching frequency to 20 kHz halves the inductance for the same ripple current, to 1.1 mH. f0 rises to 1.52 kHz with the same capacitor, so the output ripple falls by half to 2.3 V instead of to a quarter. Doubling the capacitor to 20 µF also halves the output ripple, to 2.2 V, but doubles the no-load current to 1.45 A, or 10 % of rated, and moves f0 down to 760 Hz.

The resonance still needs damping

Nothing in this filter dissipates energy at f0 except the load, the inductor's DC resistance and the capacitor's ESR. At no load that leaves a resonance that a load step or a harmonic near 1 kHz can excite. In a balanced three-phase system the low-order harmonics left by dead time and device drops sit at 6k ± 1 times the fundamental, so at 50 Hz the 19th and 23rd fall at 950 Hz and 1150 Hz, on either side of this f0. Whether they matter depends on their size and on the control, and it should be checked on the actual design.

Passive damping, such as a resistor in series with an extra capacitor across C, is covered in the LC damping guide. Inverters with output voltage control often damp the resonance in the controller instead, for example by feeding back the capacitor current. Either way, the inductor and capacitor values above are the starting point.

Design steps

  1. 1From V_LL and S, compute V̂ and Î, and check that m = V̂/(Vdc/2) stays below 1.155 at the lowest DC link.
  2. 2Choose the ripple target, usually 20 to 40 % of Î, and compute L ≥ V̂/(2√3·f_sw·ΔI_pp). Use Vdc/(6·L·f_sw) only if the inverter really runs near the edge of linear modulation.
  3. 3Choose the reactive fraction q, usually a few percent, and compute C = q·S/(3·ω·V_ph²).
  4. 4Compute f0 and check that it is well above the fundamental and well below f_sw; the output ripple is roughly (f0/f_sw)² of the bridge ripple.
  5. 5Add the ωL drop to the required bridge voltage at the worst power factor and confirm the modulation index still has headroom.
  6. 6Plan how the resonance will be damped, passively or in the control loop.

Sources

  • M. Liserre, F. Blaabjerg, S. Hansen, "Design and Control of an LCL-Filter-Based Three-Phase Active Rectifier," IEEE Transactions on Industry Applications, vol. 41, no. 5, pp. 1281–1291, 2005 (the widely cited step-by-step procedure for grid-side filters).
  • R. Peña-Alzola, M. Liserre, F. Blaabjerg, R. Sebastián, J. Dannehl, F. W. Fuchs, "Systematic Design of the Lead-Lag Network Method for Active Damping in LCL-Filter Based Three Phase Converters," IEEE Transactions on Industrial Informatics, vol. 10, no. 1, pp. 43–52, 2014 (capacitor sized so that the power factor drops by no more than 5 %).
  • All numbers in this article were computed in Python: the worst-case ripple from the exact switching instants of centred SVPWM and sine-triangle PWM over a full fundamental cycle, and the output ripple and THD from a frequency-domain steady-state solution of the complete LC filter with load. The chart uses the same ripple calculation.

This guide explains general engineering principles. Always verify your design with datasheets, simulation and measurement.

Calculators for this topic

More guides