Functions / Requirements of Direct-Off-Line SMPS -- OUTPUT FILTERS

Home | Articles | Forum | Glossary | Books

AMAZON multi-meters discounts AMAZON oscilloscope discounts



1 INTRODUCTION

Undoubtedly, one of the most objectionable properties of switchmode supplies is their predilection for high-frequency radiated and conducted ripple and noise (RF interference).

To keep this interference within reasonable bounds, there must be strict attention to noise reduction techniques throughout the electrical and mechanical design. Faraday screens can be used in transformers and between high-frequency high-voltage components and the ground plane. (These screening methods are more fully covered.) In addition, to reduce conducted-mode noise, low-pass input and output filters will be required.

2 BASIC REQUIREMENTS

Output Low-Pass Filters

The following section on output-filter design assumes that normal good design practice has already been applied to minimize conducted-mode noise and that RFI filters have been fitted to the input supply lines.

To provide a steady DC output, and reduce ripple and noise, LC low-pass filters (as shown in FIG. 1a) will normally be provided on switching supply outputs.

In forward converters, these filters carry out two main functions. The prime requirement is one of energy storage, so as to maintain a nearly steady DC output voltage throughout the power switching cycle. A second, and perhaps less obvious, function is to reduce high-frequency conducted series and common-mode output interference to acceptable limits.

Unfortunately, these two requirements are not compatible. To maintain a nearly constant DC output voltage, the current in the output capacitor must also be nearly constant; hence a considerable inductance will be required in the output inductor. Since the inductor must also carry the DC output current, it is often large and may have many turns. This results in a large interwinding capacitance, giving a relatively low self-resonant frequency.

Such inductors will have a low impedance at frequencies above self-resonance and will not provide very effective attenuation of the high-frequency components of the conducted interference currents.


FIG. 1 (a) Power output filter, showing parasitic interwinding capacitance CC and series resistance Rs for L1, and series inductance ESL and resistance ESR for C1.


(b), (c) Output filter equivalent circuits at low frequency (b) and high frequency (c).

3 PARASITIC EFFECTS IN SWITCHMODE OUTPUT FILTERS

FIG. 1a shows a single-stage LC output filter (such as might be found in a typical forward converter. It includes the parasitic elements CC, Rs, ESL, and ESR.

The series inductor arm L1 shows an ideal inductor L in series with the inevitable winding resistance Rs. The parasitic distributed interwinding capacitance is included as lumped equivalent capacitor CC.

The shunt capacitor C1 includes the effective series inductance ESL and the effective series resistance ESR.

The equivalent circuit of this network at low and medium frequencies is shown in FIG. 1b. The effects of CC, ESL and ESR are small at low frequencies and may be neglected. From this equivalent circuit, it is clear that the filter will be effective as a low pass filter for the low and medium end of the frequency range.

A second equivalent circuit for high frequencies is shown in FIG. 1c. At high frequencies, the ideal inductance tends to high impedance, taking out the L-Rs arm, and the ideal capacitor C tends to zero, taking out C. Thus, the parasitic components become predominant, effectively changing the single-stage low-pass LC filter to a high-pass filter. This occurs at some high frequency, where the interwinding capacitance CC and effective series inductance ESL become predominant. Hence, this type of power output filter is not very effective in attenuating high-frequency conducted mode noise.

4 TWO-STAGE FILTERS

As shown above, attempts to satisfy all the voltage averaging and noise rejection requirements in a single LC filter would require the selection of expensive components, particularly in flyback converters. Even then, only mediocre high-frequency performance would be obtained.

FIG. 2 shows how a far more cost-effective wideband filter can be produced, using a second-stage, much smaller, LC filter to reject the high-frequency noise. The second stage (L2, C2) may be quite small and inexpensive because only small inductance and capacitance values are required in this second stage. At the same time, much lower cost standard electrolytic capacitors and inductors may be used in the first stage (L1, C1), thus reducing the overall cost and improving the performance.


FIG. 2 Two-stage output filter.

In FIG. 2, the first capacitor C1 is selected for the required ripple current rating and energy storage needs. (This depends on the load current and the operating frequency.) C1 will often be quite large, but does not need to be a low-ESR type when a two-stage filter is used.

The first inductor L1 is designed to carry the maximum load current with mini mum loss and without saturation. To obtain the maximum inductance and minimum resistance in the smallest size, L1 will have a multiple-turn multilayer winding.

Although this gives the maximum inductance, it results in a relatively large inter winding capacitance and low self-resonant frequency. Suitable core materials for L1 include gapped ferrites, Permalloy, iron-dust toroids, or gapped silicon iron in "E-I" shapes. L1 will have the majority of the inductance required for energy storage considerations.

The second inductor L2 is designed to have the maximum impedance at high frequency, and requires a low interwinding capacitance. This will provide a high self resonant frequency. L2 may take the form of a small ferrite rod, a ferrite bobbin, small iron-dust toroids, or even an air-cored coil. Since the AC voltage across L2 is small (of the order of 500 mV), the magnetic radiation from a high-reluctance magnetic path will be quite small and should not present an EMI problem. Normal ferrite materials may be used for a ferrite rod inductor, as the large air gap will prevent DC saturation of the core.

The second capacitor C2 is much smaller than C1. It is selected for low impedance at the switching and noise frequencies (rather than for its energy storage ability). In many cases C2 will consist of a small electrolytic shunted by a low-inductance foil or ceramic capacitor. Since L1 and L2 conduct a large DC current component, the term "choke" is more correctly applied to these items. A design example follows.

5 HIGH-FREQUENCY CHOKE EXAMPLE

To get the best performance from the high-frequency choke L2, the interwinding capacitance should be minimized.

FIG. 3a shows a 1-in-long ferrite rod choke with a 5/16-in diameter, wound with 15 turns of closely packed #17 AWG wire. FIG. 3b shows a plot of phase shift and impedance as a function of frequency for this choke. The phase shift is zero at the self resonant frequency, which in this case is 4.5 MHz.

The impedance plot in FIG. 3c shows the improvement obtained by reducing the interwinding capacitance. This plot was obtained from the same choke after spacing the windings and insulating them from the rod with 10-mil Mylar tape.

In this second example, 15 turns of 20-gauge wire are used, with a space between each turn. The plot shows that the reduction in interwinding capacitance has increased the impedance and shifted the self-resonant frequency to 6.5 MHz. This will result in a reduction in high-frequency noise in the final filter.

A small proportion of the high-frequency interference will bypass the filter by inductive and capacitive coupling in the pcb or supply leads. The effect of this will be reduced by fitting the smaller capacitor C2 as close as possible to the output terminals of the supply.





FIG. 3 (a) Ferrite rod choke. (b) Impedance and phase shift of ferrite rod choke with tight winding, as a function of frequency. Note self-resonant frequency at 4 MHz. (c) Impedance and phase shift of spaced winding (low inter winding capacitance) ferrite rod choke. Note self-resonant frequency at 6 MHz.

6 RESONANT FILTERS

By selecting capacitors such that their self-resonant frequency is near the switching frequency, the best performance will be obtained.

Many of the small, low-ESR electrolytic capacitors have a series self-resonant frequency near the typical operating frequencies of switchmode converters. At the self resonant frequency, the parasitic internal inductance of the capacitor resonates with the effective capacitance to form a series resonant circuit. At this frequency, the capacitor impedance tends to the residual ESR.

FIG. 4 shows the impedance plot of a typical 470-MF low-ESR capacitor as a function of frequency. This capacitor has a minimum impedance of 19 m7 at 30 kHz. Very good ripple rejection can be obtained at 30 kHz by taking advantage of this self-resonant effect.


FIG. 4 Impedance and phase shift of a typical commercial-grade 470-MF electrolytic capacitor as a function of frequency. Note self-resonant frequency and minimum impedance at 29 kHz.

7 RESONANT FILTER EXAMPLE

FIG. 5 shows a typical output stage of a small 30-kHz, 5-V, 10-A flyback converter with a two-stage output filter. (In flyback converters, the transformer inductance and C1 form the first stage of the LC power filter.) A second stage high-frequency filter L2, C2 has been added.

For this example, the same 1-in long, 5b16-in diameter ferrite rod inductor used to obtain plot c in FIG. 3 is used for L2. The 15 spaced turns on this rod give an inductance of 10 MH and a low interwinding capacitance. The 470-MF low-ESR capacitor used for the impedance plot in FIG. 4 is fitted in position C2.

Note: The minimum impedance of this capacitor occurs at 30 kHz, where the phase shift is zero. This is the series self-resonant frequency for this capacitor. Its impedance will be predominately resistive with a value of 19 m7, as shown in FIG. 4.

The attenuation provided by this LC network at 30 kHz (the switching frequency) may now be very easily calculated, since the capacitor C2 looks predominately resistive and forms a simple divider network with the series impedance of inductor L2. (The small phase shift can be neglected, as XL2 q ESR of C2.) The ratio of the output voltage ripple (Vout ) to the ripple voltage across the first capacitor C1 is…

As XL2 q ESR, the attenuation ratio A, tends to…

Where inductive reactance XL _ 2PfL

ESR _ effective series resistance of capacitor at resonance…

From Fig., the ESR of C2 at 30 kHz is 0.019 7. Hence, the attenuation ratio A, will be…

This gives a ripple rejection ratio of 100:1 at the switching frequency.

The switching frequency ripple is normally the predominant ripple component in fly back converters. By making use of the self-resonant properties of the electrolytic capacitor, an extremely good ripple rejection of 40 dB is obtained with very small, low-cost components. Further, the improved high-frequency noise rejection is obtained without compromise to the medium-frequency transient response, because the series inductance has not been increased significantly.


FIG. 5 Example of resonant output filter applied to a flyback converter secondary.

8 COMMON-MODE NOISE FILTERS

The discussion so far has been confined to series-mode conducted noise. The filters described so far will not be effective for common-mode noise; that is, noise voltages appearing between the output lines and the ground plane.

The common-mode noise component is caused by capacitive or inductive coupling between the power circuits and the ground plane within the power supply. Initially this must be reduced to a minimum by correct screening and layout at the design stage.

Further reduction of the common-mode output noise may be obtained by splitting inductor L1 or L2 into two parts to form a balanced filter, as shown in FIG. 6. Additional capacitors C3 and C4 are then required between each output line and the ground plane to provide a return path for the residual common-mode noise current. In effect, L1(a) and C3 form a low-pass filter from the positive output, and L1(b) and C4 form the filter for the negative output, with the ground plane as the return path.


FIG. 6 Common-mode output filter.

Because of the decoupling provided by the much larger capacitor C2, acceptable results will often be obtained by fitting a single common-mode decoupling capacitor in position C3 or C4.

9 SELECTING COMPONENT VALUES FOR OUTPUT FILTERS

The size and value of the main output inductor L1 and storage capacitor C1 (FIG. 1a) depends on a number of factors:

Type of converter -- Operating frequency -- Maximum load current -- Minimum load current

Mark space ratio (duty cycle)

Ripple current, Ripple voltage, Transient response, Output voltage

The requirements of L1 will now be considered in terms of the requirements for this type of converter.

10 MAIN OUTPUT INDUCTOR VALUES (BUCK REGULATORS)

In general, the main inductance L1 in the output of a buck regulator filter circuit should be as small as possible to give the best transient response and minimum cost. If a large inductance is used, then the power supply cannot respond rapidly to changes in load cur rent. At the other extreme, too low an inductance will result in very large ripple currents in the output components and converter circuits, which will degrade the efficiency. Further, discontinuous operation will occur at light loads.

One approach is to select L1 such that the inductor will remain in continuous conduction for the minimum load current (often specified as 10% of Imax).

Keeping the inductance in continuous conduction has two advantages. First, the control circuit is only required to make small changes in pulse width to control the output voltage as the load changes (provided the inductor remains in conduction throughout the operating cycle). Second, the output ripple voltage will remain small over this range of load changes.

The main disadvantage of this approach is that the inductance can be quite large; more over, the rule cannot be used if the load current must be controlled right down to zero.

A second, more universal rule is to choose the inductance value such that the ripple cur rent has an acceptable peak-to-peak limit, say 10% to 30% of the maximum load current at nominal input voltages.

Note: In flyback converters, the main inductance L1 is integral to the transformer, and its value is defined by the power transfer requirements. In this type of converter, high ripple currents must be accommodated in the filter components, particularly for complete energy transfer systems.

11 DESIGN EXAMPLE

Assume that a design is required for the main output inductor L1 for a single-ended for ward converter and filter, as shown in FIG. 1a. The specification for the converter is as follows:

Output power _ 100 W

Output voltage _ 5 V

Output current _ 20 A

Output frequency _ 30 kHz

Minimum load _ 20%

The design approach will assume that the output ripple current must not exceed 30% of I_load (6 A p-p in this example).

Also, to allow for a range of control, the pulse width at nominal input will be 30% of the total period (that is, 10 Ms).

To provide an output of 5 V at a pulse width of 30%, the transformer secondary voltage will be…

...where tp _ total period (at 30KHZ), ms

to _ "on" time ms

Vs _ secondary voltage

The voltage VL across the inductor L1 during the forward "on" period is the secondary voltage less the output voltage, assuming that the output capacitor C1 is large and the volt age change during the "on" period is negligible.

Then…

For steady-state conditions, the current change for the "on" period must equal the current change during the "off " period (in this example, 6 A). Neglecting second-order effects, the inductance may be calculated as follows:

...where L _ required inductance, MH

$t _ "on" time, Ms

$i _ current change during "on" time

VL _ voltage across inductor

Therefore...

Note: A simple linear equation can be used, as the voltage across the inductance is assumed not to change during the "on" time and di/dt is constant.

In this example, the inductance is large because sufficient energy must be stored during the "on" period to maintain the current during the "off " period. In push-pull forward converters the "off" period is much smaller, so that the secondary voltage and hence the inductance value would also be smaller.

12 OUTPUT CAPACITOR VALUE

It is normally assumed that the output capacitor size will be determined by the ripple cur rent and ripple voltage specifications only. However, if a second-stage output filter L2, C2 is used, a much higher ripple voltage could be tolerated at the terminals of C1 without compromising the output ripple specification. Hence if ripple voltage were the only criterion, a much smaller capacitor could be used.

For example, assume that the ripple voltage at the terminals of C1 can be 500 mV. The current change in L1 during the "on" period will mainly flow into C1, and hence the capacitance value required to give a voltage change of 500 mV can be calculated as follows (the following equation assumes a perfect capacitor with zero ESR):

...where C _ output capacitance value, MF

$I _ current change in L1 during "on" period, A

$ton _ "on" time Ms

$Vo _ ripple voltage, V p-p

Therefore...

Hence, just to meet the ripple voltage requirements, a very small capacitor of only 120 MF would be required. However, in applications in which the load current can change rapidly over a large range (transient load variations), a second transient load variation criterion may define the minimum output capacitor size.

Consider the condition when the load suddenly falls to zero after a period of maximum load. Even if the control circuit responds immediately, the energy stored in the series inductor (½ LI 2 ) must be transferred to the output capacitor, increasing its terminal voltage. In the above example, with an output capacitor of only 120 MF, a series inductance of 19.4 MH, and a full-load current of 20 A, the voltage overshoot on load removal would be nearly 100%. This would probably be unacceptable, and hence the maximum acceptable voltage overshoot on load removal may become the controlling factor.

The minimum output capacitor value to meet the voltage overshoot requirements using the transferred energy criteria can be calculated as follows:

Energy in output inductor when full load is suddenly removed:

The energy change in the output capacitor after the event will be...

...where Vp _ maximum output voltage _ 6 V

Vo _ normal output voltage _ 5 V Hence

If the maximum voltage in this example is not to exceed 6 V, then the minimum value of output capacitance will be

Further, the ripple current requirements may demand that a larger capacitor be used. Some allowance should also be made for the effects of the capacitor ESR, which will increase the ripple voltage by about 20% typically, depending on the ESR and ESL of the capacitor and the size, shape, and frequency of the ripple current.

In conclusion, it has been shown that very effective series- and common-mode con ducted ripple rejection can be obtained by the addition of a relatively small additional LC output filter network. This relatively simple change allows good ripple and noise rejection to be obtained using lower-cost medium-grade electrolytic capacitors and conventional inductor designs.

13 QUIZ

1. Discuss the major disadvantage of switch mode power supplies compared with the older linear regulator types.

2. Is the design of the output filter the only most important factor in reducing output ripple noise?

3. Explain the meaning of the term "choke" as applied to output filters.

4. Why are power output filters often relatively ineffective in dealing with high frequency noise?

5. Why are two-stage filters sometimes used in output filter applications?

6. What is the difference between common-mode and differential-mode noise filters?

7. In what way does the design of a common-mode choke differ from that of a series mode choke?

Also see: Our other Switching Power Supply Guide

Top of Page

PREV. | Next | Guide Index | HOME