100nF

Smoothing capacitor after a rectifier: sizing and ripple

Size a bridge rectifier's reservoir capacitor from load power, line frequency and valley voltage: 70.6 W at 85 VAC needs 107.6 µF, not I/(2fC)'s 152.7 µF.

The smoothing capacitor after a bridge rectifier is sized by an energy balance: it must hold the load on its own from one peak of the rectified sine until the next half-cycle climbs back to meet it. For a 70.6 W converter running from 85 VAC at 50 Hz, with the capacitor allowed to sag to 72.14 V (60 % of the peak), that takes 107.6 µF, about 1.5 µF per watt. The familiar estimate ΔV ≈ I/(2fC) asks for 152.7 µF for the same ripple, 42 % more, because it assumes the capacitor supplies the load for the whole 10 ms between peaks when it does so for only 7.05 ms.

The exact form is Equation 11 of TI’s UCC28740 datasheet (SLUSBF3D), and TI works it with numbers in the TIDUB51 reference design, a 60 W, 24 V flyback. Every derived number below comes from the module behind the smoothing capacitor calculator, so the page and the calculator agree. TIDUB51’s worked example has three places where its own figures do not line up, and each is set out where it arises.

How a reservoir capacitor sets the ripple voltage

A rectifier alone delivers a string of half-sines. With a capacitor across its output, each half-cycle charges the capacitor along the rising sine to the peak, √2 × Vin less the diode drops. Past the peak the sine falls away faster than the load can discharge the capacitor, the diodes stop conducting, and the capacitor carries the load alone. Its voltage falls until the next half-sine rises to meet it. That meeting point is the valley, Vmin, and the difference between peak and valley is the peak-to-peak ripple.

Two stacked plots against time over 30 milliseconds for a bridge rectifier on 85 volts AC at 50 hertz feeding 107.6 microfarads and a 70.6 watt constant-power load. Top: the rectified sine peaking at 120.2 volts, and the capacitor voltage, which follows the sine up to each peak, then falls on its own for 7.05 milliseconds to a 72.14 volt valley, where the next half-sine catches it and recharges it in the remaining 2.95 milliseconds of the 10 millisecond ripple period. Bottom: the capacitor current, a short positive pulse that steps to 3.25 amps as the diodes turn on and falls to zero at the peak, then a negative discharge current of between 0.59 and 0.98 amps for the rest of the period.
Fig 1 — The reservoir capacitor carries the load alone for the discharge window t_d and gets all of that charge back in the conduction time t_c. Computed by the calculator's module for TI's example: 85 VAC, 50 Hz, 70.6 W, 107.6 µF, ideal diodes that stop at the peak. The capacitor current in this model has an RMS value of 1.25 A against an average load current of 0.72 A.

Two intervals make up each ripple period. The discharge window td runs from the peak to the valley, and the conduction time tc is what remains. With the line phase measured from a zero crossing, the peak is at 90° and the next half-cycle of a full-wave rectifier reaches Vmin at 180° + arcsin(Vmin/Vpk), so for a line frequency f:

td=1f(14+12πarcsin⁡Vmin2 Vin),tc=12f−td=12πfarccos⁡Vmin2 Vint_d = \frac{1}{f}\left(\frac{1}{4} + \frac{1}{2\pi}\arcsin\frac{V_{min}}{\sqrt{2}\,V_{in}}\right), \qquad t_c = \frac{1}{2f} - t_d = \frac{1}{2\pi f}\arccos\frac{V_{min}}{\sqrt{2}\,V_{in}}

The lower the valley is allowed to go, the earlier the next half-cycle catches the capacitor and the shorter the window. In the figure the capacitor is alone for 7.05 ms and recharges in 2.95 ms. All the charge the load took in the first interval goes back in the second, which is why the current in the lower trace is a tall, narrow pulse.

The simple ripple formula: ΔV ≈ I/(2fC)

The textbook estimate treats the load as a constant current and has the capacitor supply it for the entire ripple period, 1/(2f) after a bridge or a centre-tapped transformer:

ΔV≈Iload2fC(full-wave),ΔV≈IloadfC(half-wave)\Delta V \approx \frac{I_{load}}{2 f C} \quad\text{(full-wave)}, \qquad \Delta V \approx \frac{I_{load}}{f C} \quad\text{(half-wave)}

It is the limit of zero conduction time. Since the real window is shorter than the ripple period, the formula overstates the ripple a given capacitor leaves, and the capacitance it returns for a target ripple is too large by exactly the ratio of the ripple period to td. That makes it a safe estimate. It is also a quick one, good for a first pass, and it is close when the ripple is a small fraction of the peak.

A switching converter is not a constant-current load. It draws constant power, so its input current rises as the capacitor voltage falls. The energy the capacitor gives up, ½C(Vpk² − Vmin²), factors into C × ΔV × Vmid, where Vmid is halfway between peak and valley, so a constant-power discharge moves exactly the charge a constant current of P/Vmid would. Feeding the simple formula that current is the fairest comparison, and it is what the calculator shows beside the exact result.

The conduction-angle form: TI’s Equation 11

The UCC28740 datasheet states what the bulk capacitor is for (§8.2.2.3, p. 20):

Determine the minimum voltage on the input bulk capacitance, CB1 and CB2 total, in order to determine the maximum Np-to-Ns turns-ratio of the transformer. The input power of the converter based on target full-load efficiency, the minimum input RMS voltage, and the minimum AC input frequency determine the input capacitance requirement.

and then gives the equation (p. 21):

CBULK=2PIN×(0.25+12π×arcsin⁡(VBULK(min)2×VIN(min)))(2VIN(min)2−VBULK(min)2)×fLINEC_{BULK} = \frac{2P_{IN} \times \left(0.25 + \dfrac{1}{2\pi} \times \arcsin\left(\dfrac{V_{BULK(min)}}{\sqrt{2} \times V_{IN(min)}}\right)\right)}{\left(2V_{IN(min)}^{2} - V_{BULK(min)}^{2}\right) \times f_{LINE}}

Read it as energy. The bracket divided by fLINE is the discharge window td above. The numerator is then 2 × P × td, twice the energy the load takes while the diodes are off. In the denominator, 2VIN(min)² is the square of the peak, so the difference is Vpk² − Vmin². The whole expression is ½C(Vpk² − Vmin²) = P × td solved for C.

The valley appears on both sides, inside the arcsine and in the denominator, so the equation cannot be turned round algebraically to give the valley for a chosen capacitor. The datasheet says as much (p. 21):

Equation 11 provides an accurate solution for the total input capacitance based on a target minimum bulk-capacitor voltage. Alternatively, to target a given input capacitance value, iterate the minimum capacitor voltage to achieve the target capacitance value.

The calculator does that iteration by bisection. Equation 11 has no diode drop in it, which costs little on rectified mains, where two drops of about a volt sit against a 120 V peak.

How far off is the I/(2fC) estimate?

Plot of bulk capacitance per watt of constant-power load, on a log axis from 0.8 to 20 microfarads per watt, against the valley voltage as a share of the peak from 40 to 95 percent, for a bridge rectifier on 85 volts AC at 50 hertz. Two rising curves: TI's Equation 11 with the true discharge window, and the textbook current over twice the line frequency times capacitance estimate, which assumes the capacitor carries the load for the whole ripple period and is always higher. At a 60 percent valley Equation 11 gives 1.52 microfarads per watt and the textbook 2.16, 42 percent more; at 90 percent the gap has closed to 17 percent.
Fig 2 — Microfarads per watt at 85 VAC, 50 Hz, from the calculator's module: Equation 11 against the textbook I/(2fC) fed the equivalent current P/V_mid. The textbook figure is larger by the ratio of the ripple period to the discharge window, 42 % at a 60 % valley and 17 % at 90 %.

The gap depends only on how deep the ripple is. At 85 VAC and 50 Hz, with the valley at 60 % of the peak, Equation 11 gives 1.52 µF per watt and the simple formula 2.16 µF per watt, 42 % more. With the valley at 80 % the excess is 26 %, and at 90 % it is 17 %. The shallower the ripple, the closer the discharge window comes to the full ripple period and the better the simple formula does.

Line voltage and frequency move both curves together. Capacitance per watt goes as the inverse square of the line voltage: for a 60 % valley, 1.52 µF/W at 85 VAC becomes 0.83 µF/W at 115 VAC and 0.21 µF/W at 230 VAC, which is why universal-input supplies carry bulk capacitors that look oversized on a 230 V bench. Frequency is a straight inverse: 1.27 µF/W at 60 Hz, 1.62 µF/W at 47 Hz.

Worked example: 60 W flyback from 85 VAC

TIDUB51’s design targets (Table 2, p. 9) are 85 to 265 VAC, 47 to 63 Hz, 60 W out at a targeted efficiency of 85 %, so PIN = 60/0.85 = 70.6 W. For the valley it says (p. 10):

The minimum recommended valley voltage on the input bulk capacitors is taken as 60% of the peak of the minimum AC voltage.

and prints 85 × √2 × 0.6 = 72.14 V. The product is 72.12 V; the difference is immaterial, and the steps below use TI’s printed figure.

peak      √2 × 85 V                                  = 120.21 V
arcsine   72.14 / 120.21                             = 0.6001
window    t_d = (0.25 + arcsin(0.6001)/2π) / 50 Hz   = 7.05 ms
energy    C = 2 × 70.6 W × t_d / (2 × 85² − 72.14²)  = 107.6 µF   TI: ≥ 107.7 µF
simple    I = P/V_mid = 70.6 / 96.17                 = 0.734 A
          C = I / (2 × 50 × 48.07 V)                 = 152.7 µF

TI prints 107.7 µF, within 0.05 % of the module’s 107.6 µF and small enough to be intermediate rounding.

The design then fits more than the minimum (p. 10):

To meet the needs of hold up time, bulk capacitance is selected higher than this calculated value. The bulk capacitor selected is 82 µF × 2 = 164 µF.

followed by “Using Equation 3 and CBULK = 164 µF, VBULK(min) = 86.7 V.” Equation 3 is the datasheet’s Equation 11 under another number. Inverting it for 164 µF at 50 Hz, the frequency TI used one line earlier for the 107.7 µF, gives 88.7 V. At 47 Hz it gives 86.7 V. So the two results on the same page of TIDUB51 are worked at different line frequencies, without saying so: 107.7 µF at 50 Hz and 86.7 V at the design’s 47 Hz minimum. Neither is wrong for its frequency, but the minimum capacitance for a 72.14 V valley at 47 Hz is 114.5 µF, and that is the number the datasheet’s “minimum AC input frequency” calls for.

The minimum bulk voltage a converter has to work from

Plot of the minimum bulk voltage, from 50 to 125 volts, against bulk capacitance from 60 to 330 microfarads, for a bridge rectifier on 85 volts AC feeding a 70.6 watt constant-power load, at line frequencies of 47, 50 and 60 hertz. Each curve rises steeply at first and flattens towards the 120.2 volt peak. Marked points: 107.6 microfarads holds 72.14 volts at 50 hertz; 164 microfarads holds 88.7 volts at 50 hertz and 86.7 volts at 47 hertz. Below about 52 microfarads the capacitor empties before the next half-cycle arrives.
Fig 3 — The valley a bulk capacitor holds for TI's 70.6 W at 85 VAC, by the calculator's inversion of Equation 11. TI's computed minimum of 107.7 µF sits on the 50 Hz curve at 72.14 V; the 164 µF fitted holds 88.7 V at 50 Hz and 86.7 V at the design's 47 Hz minimum, the figure TI prints.

The valley is the lowest input the converter ever regulates from, and it is an input to the next design step. TIDUB51 opens its section with it (p. 9):

The value of the bulk capacitor used determines the minimum input voltage for the flyback converter. This in turn determines the primary-to-secondary turns ratio of the transformer.

Its Equation 13 (p. 12) puts VBULK(min) in the numerator of the maximum turns ratio: NPS(max) = DMAX × VBULK(min) / (DMAGCC × (VOCV + VF + VOCBC)). With DMAX = 0.51, DMAGCC = 0.425 and a secondary voltage of 24 V + 0.4 V + 0.01688 V, TI gets 4.261 from 86.7 V. The same expression at the 72.14 V the minimum capacitor would leave gives 3.545. TI’s comment on what that costs (p. 12): “A higher turns-ratio generally improves efficiency, but may limit operation at low input voltage.” The article on how a flyback converter works follows the turns ratio through to duty cycle and switch stress, and the flyback transformer calculator takes the minimum input voltage as its starting point.

The curve in the figure is steep at the bottom and flattens towards the peak. At 47 Hz, doubling 164 µF to 328 µF raises the valley from 86.7 V to 103.2 V. Going the other way costs more for less: 114.5 µF holds 72.1 V, 80 µF holds only 50.0 V, and below 52.0 µF the capacitor is empty before the next half-cycle arrives. A part sized near the knee loses many volts of valley to tolerance and ageing. Size at the lowest line voltage and the lowest line frequency the product is specified for, with the capacitance the part will have at the end of its life.

Ripple current in the smoothing capacitor

The capacitor’s ripple-current rating matters as much as its capacitance, because every coulomb the load draws passes through the capacitor’s ESR twice: once slowly on the way out and once quickly on the way back. In the ideal model the current is easy to write down. While the diodes conduct, the capacitor rides the sine and its current is C × dv/dt = C × 2πf × Vpk × cos φ; while they are off, it is the load current, P/v. For TI’s 164 µF at 47 Hz and 70.6 W, integrating the square of both over one ripple period gives:

discharge window / conduction time     8.05 ms / 2.59 ms of 10.64 ms
average load current                   0.68 A
capacitor current as the diodes start  4.03 A   (diodes: 4.85 A with the load)
RMS, charging pulse alone              1.20 A
RMS, discharge alone                   0.60 A
RMS capacitor ripple current           1.34 A   (2.0 × the average load current)

A capacitor carrying twice the DC load current in RMS ripple is normal for a capacitor-input rectifier, and it is the figure to hold against the capacitor’s rating at the ripple frequency, 94 to 120 Hz here. The capacitor ESR calculator turns an RMS current and a dissipation factor into watts.

TIDUB51 reaches almost the same number by a different route, and the route does not hold together. Its Equation 5 (p. 10) for the capacitor charge time is printed as:

tCH=1−sin⁡−1[2×VIN(min)−VBULK(min)2×VIN(min)]4×50 Hzt_{CH} = \frac{1 - \sin^{-1}\left[\dfrac{\sqrt{2} \times V_{IN(min)} - V_{BULK(min)}}{\sqrt{2} \times V_{IN(min)}}\right]}{4 \times 50\ \text{Hz}}

and evaluated to 3.6 ms. The arithmetic is right: the arcsine of (120.21 − 86.7)/120.21 is 0.282 rad and (1 − 0.282)/200 is 3.59 ms. The expression is not. It subtracts an angle in radians from 1 as if it were a fraction of a quarter cycle, and its argument is the ripple, not the valley. With no ripple at all it returns 5 ms, a full quarter cycle of conduction, where the conduction time should shrink to zero. The geometry gives arccos(86.7/120.21)/(2πf): 2.44 ms at 50 Hz and 2.59 ms at 47 Hz, about a millisecond shorter than TI’s figure. Equation 5 also uses 50 Hz beside a VBULK(min) and a tRL worked at 47 Hz.

Equation 7 follows, introduced with “The minimum input capacitor RMS ripple current rating is given by”, as CBULK × (√2 × VIN(min) − VBULK(min)) / (tRL − tCH) × √3 = 1359.4 mA, with tRL = 10.6 ms from Equation 6. That lands within 2 % of the 1.34 A integrated above, so the conclusion survives. The next equation (Equation 8, p. 11, peak AC input current) divides by √(3.6 ms / 21.27 ms) to get 3.96 A. Its symbol is tRL, which Equation 6 has just set to 10.6 ms; 21.27 ms is the full 47 Hz line period. With 10.6 ms the same expression gives 2.79 A. The document does not say which is intended.

One caution applies to every figure in this section. The model has a source with no impedance, and a real line, fuse, filter and inrush element widen the charging pulse and lower its peak. For the same charge a wider pulse has a lower RMS value, so the ideal figure errs high. The converter’s own switching-frequency input current also flows in the bulk capacitor and is not in these numbers.

Low-voltage supplies: diode drops and constant current

A transformer, bridge and linear regulator is the constant-current case, and here the simple formula does much better. Take a 12 VAC secondary at 50 Hz, a 1 A load and 2 V of ripple, through a Vishay DF-S bridge, whose datasheet (p. 2) gives the “Maximum instantaneous forward voltage drop per diode” as 1.1 V at 1.0 A. Two diodes conduct at a time.

peak      √2 × 12 V − 2 × 1.1 V                    = 14.77 V
valley    14.77 V − 2 V                            = 12.77 V
window    t_d, with the drops inside the arcsine   = 8.44 ms
exact     C = 1 A × 8.44 ms / 2 V                  = 4220 µF
simple    C = 1 A / (2 × 50 Hz × 2 V)              = 5000 µF   (18 % more)

Leaving the two diode drops out moves the whole waveform up by 2.2 V on paper, 13 % of the 16.97 V source peak. A regulator that needs its output plus its dropout at the valley sees 12.77 V, not 14.97 V; the LDO dropout calculator checks what is left.

The pulse is harsher than the offline one. The diodes conduct for 1.56 ms of each 10 ms and carry a mean of 6.4 A while they do, and in the ideal model the capacitor’s RMS ripple current is 2.6 A for a 1 A load. Transformer winding resistance softens that considerably in practice, but the direction is clear: the smaller the ripple, the shorter the conduction time and the higher the ratio of ripple current to load current. The bridge’s 1.1 V is specified at 1.0 A, and in the pulse it is carrying several times that.

Voltage rating, tolerance and switch-on

The capacitor’s voltage rating comes from the highest line at no load, when nothing pulls the peak down. TIDUB51’s input range tops out at 265 VAC, where √2 gives 375 V, but it rates the parts for more (p. 10):

For this design, two capacitors (C2 and C3) with a 450-V rating need to be used to meet the maximum AC voltage rating of 300-V AC. To ensure high temperature long life product, a capacitor load life of 10,000 to 12,000 hours at 105°C operation is recommended.

√2 × 300 V is 424 V, which TI rounds to “up to 425-V DC” on the next page.

The capacitance the equations return is what the part must still have at the low end of its tolerance, cold, at the end of its life. An electrolytic bought at the computed value will not meet it.

At switch-on an empty capacitor is a short circuit across the rectifier for the first half-cycle. The DF-S bridge is rated for a “Peak forward surge current single half sine-wave superimposed on rated load” of 50 A (p. 1), and a larger reservoir makes the surge longer. The article on inrush current limiters covers sizing the element that keeps the surge inside that kind of rating.

What the ideal-rectifier model leaves out

  • Diodes that conduct past the peak. Equation 11 has the capacitor leave the sine exactly at its peak. It in fact follows the sine down until the sine falls faster than the load discharges it, then departs later and lower with less time to go. The calculator solves that case as well (a derivation that is not in TI’s documents): for 164 µF at 47 Hz it puts the valley at 87.4 V in place of 86.7 V. Equation 11 errs on the safe side.
  • Source impedance. Transformer resistance and leakage, mains wiring and the EMI filter lower the peak under load and widen the pulse. The valley is lower than the ideal one; the ripple current is too. For a transformer supply, use the loaded secondary voltage.
  • ESR. The capacitor’s series resistance adds a step to the ripple at each charging pulse, on top of the ripple the capacitance leaves. The article on measuring supply ripple covers probing a rail without the probe adding ripple of its own.
  • Half-wave rectifiers. The window is half a line period longer, (0.75 + arcsin(x)/2π)/f, and the ripple is at the line frequency. The full-wave formulas do not apply.

Sources