Practical EMI Line Filters

When we start designing EMI filters, we will find that safety issues, thermal issues, and even loop stability concerns are intricately linked to the central issue of EMI. In particular, we must look closely at the aspect of safety, because even though we can attempt to sell equipment that doesn't function satisfactorily, product safety is a legal requirement without which we just can't sell. So particularly in an off-line application, where the voltages are high enough to cause injury, safety becomes a major concern, even if we are just designing its EMI filter.

In this section, we will focus mainly on filters for (single-phase) off-line power supplies.

However, tips for dc-dc converters will also be provided along the way.

Safety Issues in EMI Filter Design

The concept of safety and how it impacts the filter section is summed up as follows:

+ Any exposed metal (conducting) part (e.g. the chassis or output cables) is capable of causing an electrical shock to the user. To prevent a shock, such parts must be earthed and /or isolated from the high voltage parts of the power supply in some way.

+ No single point failure anywhere in the equipment should lead the user to be exposed to an electrical shock. There should be two levels of protection, so that if one gives way, there is still some protection available.

+ Levels of protection that are considered essentially "equivalent" are a) earthing of any exposed metal surface, b) physical separation (typically 4 mm) between any exposed metal and parts of the circuit containing high voltage, and c) a layer of approved insulator between any exposed metal and the high voltage. Note that the insulator must have a minimum dielectric withstand capability of 1500 V ac or 2121 V dc.

+ To qualify the preceding slightly - connecting the metal enclosure of the equipment to earth can sometimes be considered as an acceptable. level of safety protection -- there are exceptions to this, as we will soon learn. However, assuming for now that earthing is acceptable, we know that to protect the user in case the earth connection fails (maybe due to something as simple as a loose contact), we need to provide one more level of protection. So this could simply be the "4 mm" of separation.

But consider the case of a high-voltage mosfet (switch) mounted on the (earthed) metal enclosure (for better heatsinking). Clearly, we can't provide any level of protection through physical separation. So in this case, we need to place one layer of approved insulator between the mosfet and the enclosure. Note that in this position, the insulator serves as 'basic insulation.'

+ What if we have an exposed conductor that's not connected to earth (such as for equipment with a two-wire ac cord), or if earthing is itself not an acceptable. level of protection for that particular type of equipment as per safety regulations? Then, besides the layer of basic insulation, we need another insulating layer (with identical dielectric withstand capability). This is called 'supplementary insulation.' Together these two layers (basic + supplementary) are said to constitute 'double insulation.' We could also use a single layer of insulation, with dielectric withstand properties equivalent to double insulation (i.e. 3000 V ac or 4242 V dc). That would then be called reinforced insulation. So for example, if the equipment is by design, meant only for a two-wire ac cord, we would need two layers of approved insulators (or a single equivalent layer) from primary side to any exposed metal (e.g. output).

+ Why do we even bother to connect the enclosure to earth in the first place? In some cases, that's not even considered an acceptable. level of protection. And besides, we could achieve two-level protection simply by double (or reinforced) insulation. The main reason for using an earthed metal enclosure is that we want to prevent radiation from inside the equipment from spilling out. Without a metal enclosure, whether connected to earth or not, there is very little chance that a typical off-line switching power supply can ever hope to comply with radiated (and possibly conducted) emission limits. That is especially true when switch transition speeds are dropping to a few tens of nanoseconds. Earthing further improves the shielding effect.

+ The metal enclosure is rather expectedly eyed by engineers as an excellent and fortuitous heatsink. So in practice, power semiconductors are often going to be mounted on the enclosure (with insulation). However by doing this, we also create leakage paths (resistive/capacitive) from the internal subsystems/circuitry to the metal chassis. And though these leakage currents are small enough not to constitute a safety hazard, they can present a major EMI problem. If these leakage currents are not 'drained out' in some way, the enclosure will charge up to some unpredictable/ indeterminate voltage, and will ultimately start radiating (electric fields). That would clearly be contrary to the very purpose of using a metal enclosure. So we then need to connect the enclosure to earth (other than for safety reasons!). We note that even if we didn't have power devices mounted on the enclosure, there could be other leakage paths present to the enclosure. And besides that, an unearthed enclosure would also inductively pick up and reradiate the strong internal electric/magnetic fields.

+ Therefore, a) providing a good metal enclosure, and b) properly connecting it to earth is the most effective method of preventing radiated EMI. However, by creating this galvanic connection (to earth), we also now provide a "multi-lane freeway" for the conducted (common mode) noise to flow "merrily" into the wiring of the building. So now, to be able to stay within the applicable conducted emission limits, we need to provide a common mode filter somewhere.

+ Generally speaking, if the equipment is designed not to have any earth connection at all (e.g. a two wire ac cord), there will usually be no metal enclosure present either. Ignoring the problem of meeting radiation limits for now, the good news here is that no significant common mode (CM) noise can be created either - simply because CM noise needs an Earth connection by definition. Therefore a CM filter need not be present in this case. However, we must remember that conducted noise limits include not only common mode noise, but differential mode (DM) noise too.

So irrespective of the type of enclosure and earthing scheme, DM filters are always required.

+ One of the simplest ways of suppressing noise is to provide decoupling (between the nodes involved). For CM noise this could mean just placing high-frequency ceramic capacitors between the L and E wires, and also between the N and E wires, possibly at several points along the PCB. But the problem is that each of these CM line filter capacitors also unintentionally passes some ac line current into the chassis (besides the CM noise). The ac component is not considered to be "noise," but it can certainly cause an electrical shock to the user. Therefore, safety regulations restrict the total amount of current that's injected into the earth/enclosure. And this in turn means that in any common mode filter stage we need to place an upper limit on the net CM filter capacitance. However, we know that if the "C" of an LC filter is made smaller, then to maintain the desired attenuation level (resonant frequency), we need to correspondingly increase the L. Therefore, it's not surprising that the inductance used for the CM filter stage (in off-line applications) is usually fairly large (several mH).

Practical Line Filters

Ill. 1: Practical Line Filter and the CM and DM Equivalent Circuits

We now look at a typical power supply line filter, as shown in Ill. 1. Its ultimate purpose is to control conducted emissions in general, and therefore it has two stages (as highlighted) - one for differential mode and one for common mode. Let us make some relevant observations.

+ Both the CM and DM stages are symmetrical (balanced). From the viewpoint of the noise emerging from the bridge rectifier and flowing toward the LISN, there are in effect two LC filters in cascade (both for DM and CM noise). This filter configuration can provide good high-frequency attenuation (roll-off).

+ Occasionally, unbalanced filters may be tolerable - for example a single DM choke (i.e. on one line only). Or sometimes, in very low-power applications, just a plain decoupling capacitor (e.g. C1) may suffice. Sometimes tuned filter stages are seen in commercial off-line power supplies (e.g. from Weir Lambda, UK). But there are some anecdotal industry experiences that suggest that under severe line transients or under input surge waveforms, as those typically used for immunity testing, tuned filters can display unexpected oscillations (resonances), ultimately provoking failure of the power supply itself. Therefore, tuned filters are generally avoided in most commercial designs.

+ Note that the filter is usually placed before the input bridge (i.e. toward the incoming ac line input) - especially because in that position it also suppresses the noise originating from the bridge diodes. Diodes are known to produce a significant amount of medium- to high-frequency noise, especially at the moment they are just turning OFF. Small RC snubbers (or sometimes just a "C") are therefore often placed across each diode of the input bridge. Though sometimes, we can get away simply by choosing diodes with softer recovery characteristics.

+ Note that input bridge packs using ultrafast diodes are often peddled as offering a significant reduction in EMI. In practice they don't really make much difference - at least not enough to justify their steep cost. In fact typically, the faster a diode, the greater are the reverse current and forward voltage spikes that it produces at turn-off and turn-on. So very fast bridges may in fact produce worse EMI scans.

+ Typical practical values for the inductance of a CM choke in medium-power converters range from 10 to 50 mH (per leg). The DM choke is always much smaller (in inductance, but not in size as we will see). Typical values for the DM choke are 500 µHto1mH.

+ In Ill. 1, we have shown both the CM and DM filter stages as being symmetrical (balanced). So for example, we have placed identical DM chokes on each of the L and N lines. In Fig. 1 we see that in fact the DM choke is also a part of the CM equivalent circuit (and vice versa). And since line impedance imbalance can cause CM noise to get converted into DM noise, it's always advisable to keep both the CM and DM stages symmetrical (balanced).

+ One obvious way to maintain equal CM inductances in both lines is to wind them on the same core (e.g. a toroid). That automatically assures a good inductance match (assuming of course that there are an equal number of windings per leg).

Note that if we are winding the CM choke ourselves (as during prototyping), we must note the relative direction of the windings, as indicated in Ill. 1 (see third sample choke picture). With such a winding arrangement, the magnetic field inside the core will cancel out completely (in principle) for DM noise. Similarly, the flux due to the operating ac line current will also cancel out (that too being differential in nature). Therefore the choke will present an impedance only to the CM noise component.

Note: The reader is cautioned that there are several widely used but confusing symbols for the CM choke found in schematics in related literature. But whatever the symbol, as long as it's meant to serve as a common mode choke, the direction of the windings must be as shown for the toroid in Ill. 1.

+ If we reverse the current direction in one of the windings of a CM choke, then it becomes a DM choke (for both lines). However, now it's also subject to the flux produced by the ac line input current (no cancellation occurs). DM chokes, in general, should always be put through a "saturation check" - because of the impedance they present to the line current.

+ We see that DM chokes may need to be quite large, just to avoid core saturation - despite the fact that their inductance is usually much less than that of CM chokes.

But in fact, a CM choke can also be very large. That, however, is primarily necessitated not by the typically higher inductance required, but more so by the desire to provide the required inductance with the minimum amount of copper losses.

So, a core with a high AL value is sought, and that usually spells "bigger core." We should also keep in mind that we don't want the core to "topple over" and saturate, on account of small imbalances in symmetry of the windings. So we may ultimately need to oversize the CM choke for various such reasons.

+ Theoretically, there is no need for any air gap in a common mode choke, because the flux due to the line current is expected to cancel out completely. In practice, it doesn't fully, mainly due to slight differences in the individual winding arrangement (despite the equal number of turns). At a minimum, this causes the core to get dc-biased in one direction, and thereby cause an imbalance in the inductance it presents to the two lines. This would expectedly degrade the EMI performance, but in extreme cases, the core may even saturate. Note that core saturation in the filter is clearly not a catastrophic event (like the saturation of the main inductor/transformer of the converter can be), but since it's accompanied by severely worsening EMI-suppression efficacy, we need to prevent that too. Therefore, as in a forward converter transformer, a small air gap is usually present, even in a CM choke.

This may be an actual air gap (between split core halves), or it may be a distributed gap, as in powdered iron cores. Though this lowers the inductance index ( AL) somewhat, the resulting solution is much more immune to production variations, and is also more stable over time. In general, whenever we introduce an air gap, the core starts partially acquiring the properties of the interposing air - and since air never saturates, the air-gapped core too has a much softer saturation characteristic.

+ We can consider spending some more money and avail ourselves of magnetic materials like "amorphous" cores or 'Kool Mu' if we want to achieve higher inductance (with higher saturation flux densities), in a smaller size.

+ Toroidal CM chokes in particular, when used in off-line applications (i.e. with both windings on the same core) must meet safety requirements relating to the separation distances between the windings ('clearance' and 'creepage,' as discussed later). So for example, we can't simply wind the two windings carelessly overlapping each other - we need to maintain a specified physical separation. Nor can we just use a bare toroid core to wind them on - we need an approved coating and /or a suitable. bobbin.

+ A bare ferrite can be a rather good electrical conductor, especially if it's the more commonly used manganese-zinc ferrite (as opposed to nickel-zinc formulations).

This can be confirmed by simply pressing the tips of an ohmmeter at two points on the surface of any bare ferrite lying around in the lab. Further, if we are trying to rely on the enamel coating of a typical copper magnet wire to protect from shorts, we should know that the coating is considered to be just operational/functional insulation, and is not considered to be even basic insulation.

+ Note that L_cm in Ill. 1 is the inductance of each leg of the CM choke.

Therefore, it's the inductance measured across either winding, with the other winding open. Now, if we repeat this measurement, but instead of keeping the other winding open, we short its ends together, what we measure is the leakage inductance Lk. By definition, the two leakage inductances in each leg are uncoupled, and therefore they can't be sharing any magnetic path. Therefore the leakage inductance of a CM choke behaves differently from the rest of the choke - differential currents no longer cancel out for this inductance. In effect, Lk presents an inductive impedance to DM noise. This "hidden" inductance of a CM choke has been successfully exploited by filter designers, to serve as an "unintentional" DM choke. Therefore, in low-power converters, we usually won't see any separate DM chokes - just CM chokes. The good news here is that the leakage inductance is effectively an air-cored inductor, so it never saturates - even if for some reason, its "parent" CM choke saturates completely. Thus the efficacy of a leakage-based DM choke is maintained at any supply current level.

Note: In any transformer, if we measure its leakage inductance (by shorting the secondary winding), the reading remains virtually unchanged even if we remove the core completely from the bobbin. That is because leakage, by definition, is uncoupled and does not pass through the magnetic core - if it did, it would be "coupled."

+ The inter-winding capacitance of a choke affects its characteristics significantly at high frequencies. This can be intuitively visualized as providing an easy detour for noise to simply flow past the windings. To minimize the end-to-end capacitance of a toroidal winding, it's recommended that the winding be single layer. Also, in Ill. 1, the sample CM choke picture in the middle is better than the one to its left, in terms of minimizing the end-to-end capacitance. That is because of the split introduced in each winding section by the special bobbin used. The split also helps increase the leakage inductance (which helps reduce DM noise). Bobbins with several such splits are also available at a price.

+ Line to line capacitors are called 'X-capacitors' ("X-caps"). X-caps when used in off-line applications before the input bridge must be safety approved. But after the bridge (on the rectified side), it's basically a 'don't care' situation from the safety point of view. Note that since it's essentially a front-end component, approved X-caps are typically impulse-tested up to 2.5 kV peak.

+ Line-to-earth capacitors are called 'Y-capacitors.' Since Y-caps are critical in terms of having the potential to cause electrocution if they fail, approved Y-caps are typically impulse-tested up to 5 kV peak. Note that Y-caps used anywhere on the primary side (in off-line applications) must always be safety approved. Depending on the location in the power supply, we may even need two Y-caps in series (basically corresponding to double insulation). However, sometimes we can also find Y-caps placed between the secondary ground and earth/enclosure (for EMI suppression purposes). In this position, it's usually acceptable. to use any ordinary 500 V ac rated capacitor (unapproved).

+ Traditionally, off-line X-caps were of special metallized ? lm + paper construction whereas Y-caps were a specially constructed disc ceramic type. However we can also find X-caps that are ceramic, as we can find Y-caps that are ?lm type. It's a choice dictated by cost, performance, and stability concerns. Film capacitors are known to always provide much better stability over temperature, voltage, time, and so on - than most ceramics. In addition, if they are of 'metallized' construction, they also possess self-healing properties. Note that ceramic capacitors don't have any inherent self-healing property. However, ceramic Y-caps are specifically designed never to fail shorted under any condition, as this would pose a serious safety hazard.

+ If for any reason (e.g. filter bandwidth or cost) ceramic is preferred for the Y-cap positions, then we need to carefully account for its basic tolerance, its variation with respect to temperature and applied voltage, and all other long-term variations and drifts. That is because we need a certain filtering efficacy, but at the same time we can't increase the leakage current into the chassis. In this regard, we should keep in mind that the capacitance stated in the datasheet is not just a nominal (or typical) value, but in fact it happens to be a fairly misleading value. For example, the ?ne print may reveal that the test voltage at which the capacitance is stated is close to, or equal to zero Volts! So the actual capacitance it presents in a working circuit may be very different from its declared value. This is, in general, especially true for ceramic capacitors that use a high dielectric constant ("high-k") material (e.g. Z5U, Y5V, and so on). We should also know that ceramic capacitors age, except for COG/NPO types. A typical X7R capacitor ages 1% for every decade of time (in hours). So its capacitance after 1000 hours will be 1% less than what it was after 100 hours, and so on. Higher dielectric constant ceramics like Z5U can age 4 to 6% for every decade of time. So in effect our filter stage, too, gets less efficacious with time. And we need to account for this in the initial design.

+ Theoretical filter performance is based on the assumption that we are using "ideal" components. However, real-life inductors are always accompanied by some winding resistance (DCR) and some inter-winding capacitance. Similarly, real capacitors have an equivalent series resistance (ESR) and an equivalent series inductance (ESL).

At high frequencies, the inductance will start to dominate, and so a capacitor will basically no longer be functioning as one (from the signal point of view). However, capacitors with smaller capacitances generally remain capacitive up to much higher frequencies than do larger capacitances. See Tbl. 1 for some typical self-resonant frequencies (the point above which, capacitors start becoming inductive). Therefore, quite often, a smaller Y-cap may help, where a large Y-cap is not yielding results.

We can also consider paralleling a larger value Y-cap with a small Y-cap.

Tbl. 1: Practical limitations in selecting components and materials for EMI filters

+ Surface mount ("SMD") versions of off-line safety capacitors are also now appearing - for example from Wima, Germany (wima.com) and Syfer, UK (syfer.com). But we must realize that it's not enough that the capacitor merely 'complies' with a certain safety standard - the capacitor should actually be approved (tested by various safety agencies, and carrying their respective certification marks). From the electrical point of view, one of the great advantages of SMD components is their virtually non-existent ESL. This improves their high-frequency performance in any filter application. On the flipside, some ESR or dc winding resistance ("DCR") is often useful in helping damp out oscillations.

Without any resistance altogether, oscillations would last forever. That is one of the reasons why engineers sometimes pass one or both of the leads of a standard through-hole Y-cap through a small ferrite bead (preferably of a material with lossy characteristics, like Ni-Zn). This can often help suppress a particular high-frequency resonance involving the Y-cap, which is showing up in the EMI scan. But we must be careful that in doing so, we are not ending up with a radiation problem instead.

+ Designers of low-voltage, low-power dc-dc converters may find the "X2Y" patented product range available from Syfer (and from the company X2Y itself - at http://www.x2y.com) very useful if they need to miniaturize and lower the component count. This is a three-terminal integrated SMD capacitor-based EMI filter that simultaneously provides line-to-line and also line-to-ground decoupling. Picor (a subsidiary of Vicor) at http://www.picorpower.com is also now selling what is billed as the industry's first active input EMI filter stage for standard 48 V bricks.

It may be a viable choice if board space is at a premium, despite its roughly $20 cost.

+ We note that a Y-cap is always tested to higher safety standards than an X-cap.

So we can always use a Y-cap at an X-cap position, but not vice versa. For example, we can consider placing a ceramic Y-cap in parallel with a ?lm X-cap, so as to improve the DM filter bandwidth.

+ Generally, we try to maximize filter performance by increasing its 'LC' product as much as practically possible (thus lowering its resonant frequency). Further, given a choice, we would prefer to harness that improvement by using larger capacitances, instead of impractically-sized inductors. But as we know, the maximum Y-capacitance we can use is limited by safety considerations. X-caps too were limited for many years to a maximum value of 0.22 µF (though occasionally 0.47 µF was also seen). But that was simply availability and component technology limitations. Nowadays, we can get X-caps up to 10 µF. We should be conscious, however, that large input capacitances can cause undesirably high inrush surge currents at power-up. This may also cause eventual failure of the X-cap, especially if it's the very first component after the ac input inlet. Film caps can self-heal from such an event every time it occurs, but eventually the capacitance gets degraded slowly over time with each successive event. Therefore, despite EMI concerns, we should try and place X-caps after any input surge protection element - for example, the NTC (negative temperature coefficient) thermistor, or wirewound resistor, and perhaps even after a front-end choke.

Note: What were traditionally called X and Y capacitors are now more accurately called X2 and Y2 capacitors respectively. From the viewpoint of safety regulations (like impulse voltage rating etc.), the X1 and Y1 are considered virtually equivalent to two X2 and Y2 capacitors in series, respectively. For example Y1 caps are impulse tested to 8 kV. Also, the original terms 'X-caps' and 'Y-caps' have recently started getting defaulted to refer to the more uncommonly used (higher voltage) X1 and Y1 capacitors instead.

Note: In off-line power supplies, for better EMI suppression, we may decide to place Y-caps from the rectified dc rails (either one or both) to earth. So sometimes we place Y-caps from primary ground to secondary ground (usually connected to earth ground), or from the HVDC (high voltage dc) rail to secondary ground. In either of these positions, a Y1 capacitor (or two Y2 capacitors in series) may be required.

Note: Safety regulations for Nordic regions (and Switzerland) may require each Y-cap shown in Ill. 1 to be actually two Y2 capacitors in series (or a single Y1 capacitor). Historically, this has been necessitated by the fact that earthing is poor in those geographical regions. In fact, it used to be pointed out that even the main conference room of the Norwegian safety agency NEMKO (literally Norwegian Electric Material Control) did not have any earth connection available in the wall outlets. Therefore, practically speaking, a lack of earth is not considered a fault condition in many parts of the world, but is just a normal condition (this actually also includes about one-third of homes in the United Sates). Therefore, very often, whether the equipment is supposed to be earthed or not, it's expected to have reinforced insulation anyway. Earthing, if present, is then just for helping out with EMI. We see that Y1 caps will often find use even in single-phase equipment.

However, X1 caps are basically meant only for 3-phase equipment, since there is no pressing safety need for such a high voltage rating between the L and N wires in single-phase equipment.

Safety Restrictions on the Total Y-capacitance

Y-caps don't just bypass high-frequency noise, but also conduct some of the low-frequency line current. That is what the X-caps do too, the difference being that the Y-caps carry this current into the protective earth/chassis. To prevent a fatal electric shock from occurring, international safety agencies limit the total RMS current introduced into the Earth by the equipment to a maximum of typically 0.25 mA, 0.5 mA, 0.75 mA, or 3.5 mA (depending on the type of equipment and its 'installation category'- i.e. its enclosure, its earthing and its internal isolation scheme). But note that somehow, 0.5 mA seems to have become the industry default design value, even in cases where 0.75 mA or 3.5 mA may have been allowed by safety agencies. It is important to know how high one can actually go in terms of ground leakage current, as this dramatically impacts the size and cost of the line filter, in particular the choke.

Keeping the discussion here at a theoretical level, we can easily calculate that we get 79 µA per nF at 250 VAC/50 Hz. This gives us a maximum paralleled capacitance of 6.4 nF for 0.5 mA, or 44.6 nF for 3.5 mA, and so on. So, a typical configuration in off-line power supplies consists of four Y-caps, each being 1nF or 1.2 nF or 1.5 nF. Or only two Y-capacitors, each of value 2.2 nF. Note that there may be other parasitic capacitances and /or filter capacitances present, which should be accounted for in computing the total ground leakage current, and thereby correctly selecting the Y-caps of the line filter. However we must keep in mind that if for better EMI performance/CM noise rejection, a Y-cap is connected from the rectified dc rails to earth (or from the output rails to earth), there is no ground leakage current through these capacitors in principle. Therefore there is no limit on their capacitance either.

Equivalent DM and CM Circuits

The filter in the upper half of Ill. 1 reduces to the CM and DM equivalent schematics shown in the lower half of the same ill.. The equivalent schematics are from the viewpoint of the noise coming out via the bridge, heading toward the mains wiring (LISN).

Some observations are:

+ We see that the DM choke acts as a CM filter element too.

+ The leakage inductance of the CM choke appears as a DM filter element too.

+ Both the Y-caps also appear in the DM equivalent circuit (though arguably they will not add much to the heftier X-capacitance).

+ Considering that a very small value for L_dm is usually enough (because of the much larger X-capacitance possible), no "intentional" DM choke may be required. The leakage inductance of a common mode choke is roughly 1 to 3% of L_cm, depending on its construction. That is usually enough to serve as an unintentional, but effective DM choke.

+ Though CM chokes usually have a high inductance (and that's certainly needed - particularly for complying with CISPR 22 limits below 500 kHz), a good part of the CM noise is usually found in the frequency range of 10 to 30 MHz. So we must consider the fact that not all ferrites have sufficient bandwidth to be able to maintain their inductance ( AL) at such high frequencies. In fact, materials with a high permeability tend to have a lower bandwidth, and vice versa ("Snoek's law").

Therefore a "high-inductance" CM filter may look good on paper, but may not be as effective as we had thought, at high frequencies. See Tbl. 1 for typical values of initial permeability vs. bandwidth (bandwidth being defined here asa6dB fall in permeability).

+ A DM noise generator is more like a voltage source. So putting in an LC filter works well for a DM source, as it simply presents a "wall" of impedance that serves to block the DM emissions from entering the mains lines. But this strategy by itself is not going to be very effective for CM noise, because a CM noise source behaves more like a current source. And we know that current sources demand to keep current flowing, and can therefore surmount any "wall" of impedance we may place in their path (by increasing the corresponding voltage). However, if, besides placing a "wall" of impedance, we also present an alternative route for the current to keep flowing, we would be successful in preventing the CM noise from entering the mains wiring of the building. Thereafter we could "kill" the noise by dissipating the associated energy. This places a rather unusual-sounding demand on the CM choke -- not only do we need high bandwidth, but we should actually lower its quality factor 'Q', especially at high frequencies. One way to achieve that's to increase the DCR. But that will impede the line current too, and thereby lower the efficiency of the entire power supply. A better solution is to use a "lossy" ferrite material for the CM choke. The usual ferrite used for power transformers and inductors is predominantly of manganese-zinc composition. But lossy ferrites of nickel-zinc composition are actually more helpful in "killing" high-frequency CM noise components. Unfortunately, they also have such low initial permeabilities that it's impossible to get the desired high inductance (at lower frequencies). Therefore the lossy CM choke is usually an add-on to the normal CM filter stage. It could be just a small bead/toroid/sleeve made of similar lossy material, with both the L and N wires passing through its aperture.

+ Engineers are often mystified to find that making the DM choke out of (low permeability) powdered iron or lossy ferrite helps too, when all else has failed - despite all the talk about DM noise being essentially a "low-frequency emission." The reason seems to be as follows - the CM noise in a power supply is actually a non-symmetric mode, at its point of creation. Though ultimately, by cross-coupling, it does tend to spread into both the lines equally. It has been shown that non symmetric noise can be considered as a mix of CM and DM components. Therefore in practice, we do get a fair amount of high-frequency DM noise too - arising out of the non-symmetric CM noise. That is why high bandwidth/low permeability/lossy materials can help in DM noise suppression too.

+ The DM and CM filters are usually laid out in the order shown in Ill. 1. The basic idea seems to be that the last stage the noise encounters (as it travels from the power supply into the mains) should be a common mode filter. Because, if the last stage was a DM stage for example, it may not be very well balanced from the viewpoint of the noise emerging from the CM filter. And so the CM noise could get converted into DM noise, as previously explained. However, we do have a DM stage now, to hopefully take care of these additional DM noise components! Therefore, many successful commercial designs have reversed the order as drawn in Ill. 1, with the DM stage being placed closer to the power inlet. In brief, there seems to be no hard and fast rule for which stage should come before which other stage.

+ A possible location for an additional X-cap is directly on the prongs of the inlet socket (at the entrance to the power supply). We remember that in this position any line-to-line capacitor will be exposed to a huge current surge at power-up, and could degrade, if not fail immediately. So if this X-cap position seems to be the last resort, it should at least be made as small as possible (typically 0.047 µF to 0.1 µF). Or we can try ceramic capacitors in this position (approved ceramic X-caps or Y-caps should be tried here).

+ Similarly, the two front-end Y-caps ("C4" in Ill. 1), or two additional Y-caps, can also be connected directly on to the prongs of the ac inlet socket, rather than on the PCB. This can help a great deal if the wires going from the PCB to the mains inlet socket are themselves picking up stray fields (and are therefore beyond assistance from the main filter stage, which unfortunately lies on the PCB - before the point of noise injection).

+ Sealed chassis mountable. line filters (sometimes with integrated standard IEC 320 line inlets) are available from several companies like Corcom (now part of Tyco Electronics) and Schaffner, Germany (at www.schaffner.com). Such filters perform excellently but are less flexible to subsequent tweaking, and also far more expensive than board-mounted solutions.

Note: Incidentally, Schaffner also makes some of the most widely used, standard test equipment for immunity surge-testing.

+ Note that the performance of most commercially available line filters is specified with 60 or 50 hz at both ends of the filter. Therefore its actual performance in a real power supply may be quite different from what its datasheet says.

+ In general, the traces on the PCB corresponding to the filter section should be thick and wide for low inductance. CM noise suppression also usually requires a very good high-frequency connection to the enclosure. So, the relevant traces of the PCB should be connected to the chassis through several metal standoffs if possible.

However, if standoffs are not feasible, the connection (to the enclosure) should be made via thick braids of fine insulated wire. A "good" connection is usually also helpful between the enclosure and the earth wire (middle prong of the IEC inlet).

In that case braided wire can also be used. In the past, major power supply manufacturers had their own special custom-made metal brackets to connect the earth prong of the IEC inlet socket to the enclosure. But nowadays, standard IEC 320 inlets with built-in metal brackets are directly available, such as that from Methode Electronics Inc. at www.methode.com.

Some Notable Industry Experiences in EMI

One of the most stubborn cases of conducted EMI failure encountered by the author was ultimately (and rather mysteriously) solved by simply reversing the orientation of the CM choke (turning it by 180 deg. on the PCB). It was later deduced that the leakage from the core was being picked up by a nearby trace or component, and so the phase of the coupling had somehow become an issue (interference pattern). But since most inductors/chokes are symmetrically built, and also don't carry any marking to distinguish one side from the other, implementing such a fix was not easy in production. However nowadays, with so many similar "orientation-sensitive" cases being reported (even relating to the main inductor of the converter itself), some key inductor manufacturers have taken the step of placing a 'polarity mark' on their inductors/chokes.

In another well-documented EMI problem at a leading power supply manufacturing house, it was discovered that the CM choke had to be rotated by 90 deg (not 180 deg) to comply. That clearly spells "bad news" if the unit is already in production, because it means the PCB layout has to be redesigned (and perhaps the power supply needs to be re-qualified too).

Next: DM and CM Noise in Switching Power Supplies

Prev: EMI: Maxwell to CISPR

top of page   Home