Our former . eight EMC articles were aimed at familiarizing unaware readers with the fundamentals of EMI/EMC, justifying the EMC norms and testing, and explaining in simple terms the five basics interference coupling mechanisms, with the essential guidelines for controlling them. The present article goes deeper into one of the simplest, most compact and economical piece of the entire EMC arsenal: the filter. With current handling ranging from tens of Amp for signal filters up to more than hundred Amps for power line filters, they exist in all sorts of size, volume and packaging. They can be optimized against Common Mode (CM) or Differential Mode (DM) interference, or both.
No matter if you plan to buy a commercial, off-the shelf (COTS) or make your own, you must understand how they work and how to determine the most adequate arrangement of L,C elements. A filter should not be picked-up off an EMC guru’s tool case: it must be designed. Finally, although regarded as a cure against conducted EMI, filters can also reduce radiated interference, for Emission or Susceptibility.
2. Selecting the right filter
Given that an EMI filter is a low-pass element, the following parameters dictate its choice :
a) the required attenuation (DM/CM) for a given frequency, rigorously termed Insertion Loss (IL) *
b) the cut-off frequency
c) the number of poles, which itself depends on a) and b)
d) the impedances on the source and load sides of the filter, that in turn will guide the filter scheme
e) type of filter: simple capacitor, simple inductor, L-C, Tee or Pi
f) the normal service voltage/ current, to be considered for the filter capacitors and inductors
* Remark: Attenuation vs Insertion Loss
Strictly speaking, ”attenuation” is the ratio of the EMI voltage (or current) at filter input, to EMI voltage (or current) at its output. Insertion Loss (IL) being the ratio of the voltages (or currents) at the load w/o filter and with the filter in place., is a more exact definition for a filter performance. This eliminates the possible errors due to the various elements of the circuit : wiring, connectors, source and load impedances etc… However, since attenuation or rejection are most commonly used, we will keep ”attenuation” along this article.
3. Influence of actual source and load impedances
A same filter will exhibit different performances, depending on the values of Zs, ZL, the impedances on the source and load sides. Because of this, one must beware of filter manufacturer’s data. They are generally measured in a 50Ω/50Ω set-up (per Mil Standard 220), and real in-situ attenuation may differ significantly, to such extent that the performance shown on the data sheet is probably the only one you will never get! Table Fig.3 shows the preferred choice for the 4 possible impedances configurations. The approximate border for LOW / HIGH impedance designation is 100Ω.
For answering the question: ”why a C, a L, a L,C,a Pi or a T ?” there is a simple, flawless rule for choosing L or C elements, that suffers no exception: ”With filters, capacitors should look toward High impedances, both sides, inductors should look toward Low impedances, both sides”.
4. Power Line Filters
Practically no modern equipment, with its fast digital circuits, switch-mode power suply regulators and eventually RF devices, could meet EMC requirements without an efficient filtering, Switchers are the main cause of conducted emissions, violating the Military and Civilian RFI limits by as much as 40 or 50dB for unfiltered items. Reciprocally, the power mains are carrying a multiplicity of EMI noise (see our Article in March 2017 EE Magazine) whose causes are out of our control, and that equipments must tolerate without malfunctions or errors. In both cases, the power line filter is a key element for EMC compliance.
For EMI immunity, the power filter attenuation must be such as the most susceptible circuits inside the equipment do not receive on their power input a noise voltage greater than their threshold of sensitivity:
Attenuation: A(dB) = 20log [Vemi (w/o filter) / Vload (with filter)] This after having accounted eventually for the built-in attenuation of the power-supply regulator.
An equipment must not exhibit malfunction when submitted to bursts of 1kV(CM) 300kHz ringing transients on its 230V AC input. The noise immunity of internal digital circuits is 0.5V on their dc input. The existing filtering on the regulator output provide already 20dB attenuation. What is the required filter attenuation?
A(dB) = 20 Log (1000/ 0.5) – 20dB = 66 – 20 = 46dB @ 300kHz
For Emission control, the filter attenuation must be such as the HF noise generated by the equipment on its power input (CM and DM) does not violate conducted EMI emissions limits for AC or DC power mains:
Attenuation: A(dB) = Vemi (dBμV(w/o filter) – VdBμV(spec. limit)
4.1 Cut-Off Frequency and Number of elements in a power line filter
The number of poles in the filter (i.e the number of L,C elements that are cascaded) is determined by the desired attenuation in a given EMI frequency range, and by Fco, the cut-off frequency (-3dB point) below which the filter has no attenuation. For instance, at a first glance, it would seem, that a power line filter should :
- exhibit no attenuation at all for 50/60Hz
- start attenuating any undesired frequency above ≈100 -150Hz.
This would be a perfect, but huge and expensive filter, because of the physical size of its capacitors and inductors. These are only found near large loads (greater than tens or hundreds of kWatts) for correcting the power factor of heavy inductive loads, or reducing harmonic distorsion on the utility side, upstream.
So the cut-off frequency of line filters for individual equipments are usually in the 10-30kHz range. Fig2 gave the attenuation of any filter, given its cut-off (Fco) and numer of poles (n).
4.2 Optimizing a Filter against EMI Emissions on Power Line
One earliest cause of user-created noise to the power mains has been the ac-dc rectification and gate-controlled rectifiers used in many equipments. Increasing use of high frequency converters like switch-mode power supplies and inverters, variable speed drives, light dimmers, fluorescent lights, has shifted the noise emissions towards higher frequencies, causing both conducted and radiated EMI issues. In this respect, EMC filters not only can reduce conducted interference up to ≈ 30 MHz, but they are also acting against radiated emissions from the power cord at frequencies above 30 MHz. Several mechanisms, internal to the switcher, are contributing to these emissions (Fig 5).
CM emissions caused by switch-mode power supplies
- The CM conducted emission path is generally the major violator of limits, for equipments supplied in 115v or 230V ac. The CM leakage current is caused by the stray capacitance of the switching transistor, IGBT or MOSFET to their heatsink or nearby chassis. If the power supply is an isolated-type, a second stray capacitance may exist through the primary-to-secondary barrier of the transformer. Leakage current (Icm) closes by the chassis ground, returning via the power mains impedance then back to the equipment Ph and Neutral input wires. For testing purposes, the noise level is measured at the artificial network (LISN) socket ( Fig 5).
Numerical Example 2
A 150Watt switch-mode power supply operates at a frequency of 50kHz. Influent parameters are:
- Stray capacitance between the switching transistor and chassis: 100pF.
- ESR of primary buffer capacitor on rectified 230V input: 0.15Ω up to 1MHz
- Primary current, for full load, estimated: 1 Amp with 230V ac, or 2 Amp with 115V
- Corresponding worst-case peak switched current (115V case) : 3 Amp.
- European EN 55022 class B limits:
– 150kHz: 56dBμV
– 1MHz: 46dBμV
For compliance with safety regulations, the CM filter capacitors (AC line- to-ground) should not exceed 5nF each, keeping the ground leakage current < 0,5mA.
For the first harmonics of the switcher up to a few MHz, the CM current driven by the dV/dt of the switches into the stray capacitance Cp behaves as a high impedance source (current source). Considering the CM filtering action alone, the simplified switching circuit of Fig.5 results in the spectrum of CM current in the Ph/N wires (line # 4 on calculation table 1). Up to a few MHz, with the load being the two 50Ω LISNs in parallel, the source impedance is dominated by that of the 100pF stray capacitance Cp (1600 Ω at 1MHz).
The Fourier analysis of the 400V switched voltage gives a 50kHz fundamental of 180Vrms, with a 3rd harmonic of 60Vrms. Beyond 3MHz, the spectral amplitude decreases sharply like 1/F^2 thanks to the 100ns rise time. The bottom line on Table 1 shows worst case Class B limit violation of 44dB and 58dB at 150kHz and 1MHz respectively. Notice that we did not consider interference at the fundamental 50kHz frequency because there is no civilian regulatory limit below 150kHz. Entering a default value Fco = 10kHz for the power line filter, we see on graph of Fig.2 that, for a Femi / Fco ratio = 150/10 :
- no single element (n = 1) filter can match the 44dB need at 150kHz
- a two-element( n = 2) filter with L looking toward the LISN and C looking toward the high impedance (according to Fig. 3) will be all right (47dB).
Thus, according to Fig.3, the filter should have at least line-to-chassis (CM) capacitors looking toward the high impedance side. The value C is imposed by the safety constraint of 2 x 5nF maximum. The value of L is determined as follow, looking at the simplified CM path ( Fig.5): Looking at the 2 branches (Icm1) and (Icm2) in parallel, the CM filter capacitors (C1 + C2) form a divider bridge with the leakage capacitor Cp.
The corresponding voltage reduction factor is:
100pF / (2 x 5,000pF) = 0.01
The 150kHz harmonic of the switch. transistor voltage being 60Vrms (line 1 on Table), the reduced voltage Va-b across C1 or C2 filter capacitor is:
Va-b = 60 V(rms) x 0.01 = 0.6V or 600mV
This reduced voltage is in turn applied to the network of filter inductance Lf in series with the two LISNs in // (approximately 30 Ω /2 at 150kHz). Given the class B limit @ 150kHz is 56 dBuV or 0,6mV, we find that the series impedance of filter inductance should be:
Lω = 2 π F.L ≥ 15Ω x 600mV/0,6mV, hence L ≥ 16 mH
Taking a factor 2 (6dB) design margin: L = 32 mH
Note: we did not recalculate the filter elements for 1MHz: although limit violation is greater, we know that a 2-stage filter will provide 32 dB more attenuation at 1MHz than for 0,15 MHz.
DM emissions caused by switch-mode power supplies
- Combining with the CM current, the DM current that circulates back and forth in the power input wires is due to the input dc storage capacitor. Since this capacitor has parasitic series resistance and self-inductance (ESR, ESL) the flow of the main switching current creates a differential voltage (line-to-line) across the primary input. By constructing (Fig.6) the switched current spectrum, we can determine the DM voltage across the input DC storage capacitor, that in turn appears across the two LISNs. Notice that the set-up of the norm is giving the voltage at the RF output of one LISN. (Fig 6)
Contrasting with CM emission, the DM voltage across the low impedance ESR, typ. < 1Ω below a few MHz, behaves as an almost perfect voltage source. Therefore, using the Fig.3, the filter should have at least an inductor looking toward the primary capacitor and a line-to-line (DM) capacitor looking toward the two LISNs in series.
Calculation Table 2 shows a needed attenuation at 150kHz of 34dB, significantly less than for the CM case. Since the CM mode filtering capacitors C1, C2 are useless against this DM noise, an inductance value of 300uH, teamed with a 0.47uF DM capacitor will provide 43 dB attenuation at 150kHz. This DM inductance can be split into 2 windings, one for Phase and one for Neutral, avoiding a primary unbalance on the AC input. Or a single winding can be inserted on the dc high voltage side, after the rectifier bridge. In many cases, the double DM inductor can be obtained at no cost, using the leakage flux of the CM bifilar filter choke, typically 1% of the nominal CM value.
As a result, an appropriate filter structure is shown in Fig.7, optimized against CM and DM emission. The DM inductance is provided by the leakage flux of the double-wound CM choke, resulting in substantial space and cost savings. CM capacitors are often termed Cy, because of their safety class (Fig 7 and 8). Some manufacturers of ”Off The Shelf” (OTS) Switch-Mode PowerSupplies incorporate an EMI filter in their modules, some others do not.
So, designers using OTS regulators should check if the EMI conducted emissions are documented by the vendor. In any case, whatever they design their own filter or they plan for a commercial item, they should look at the most appropriate filter schematic, as described above.
4.3 Quick, coarse approximation for 50/60Hz filter components
By default of accurate calculations, the following rule of thumb can be used to define quickly conservative, maximum values for the filter elements. The rationale is that, for an ac power mains, we do not want the filter capacitor to derive uselessly too much 50Hz/60Hz current from the power mains. All the same, the filter inductor should not cause unacceptable 50Hz or 60Hz voltage drop.
If we decide 1% as a tolerable impact on ac current consumption and/ or input voltage, we get:
Xc ≥ 100 ZL and XL ≤ 0.01 ZL (Eq1)
with Xc, XL being the impedances of filter capacitor and inductor at the power mains frequency, and ZL the equivalent load impedance. For a 50Hz ac input, using more practical units for components:
C(μF) ≤ 32 I/V and L(mH) ≤ 0.032 V/I (Eq2)
If there are several capacitors like in a ”Pi” filter, the formula applies to all the capacitors in parallel.
What are the maximum value of DM filtering capacitor (Line-to-Line) and inductor we can accept for a 230V equipment with 10A of normal line current?
Answer: C(μF) ≤ 32 x 10/ 230 = 1.4 μF , L(mH) ≤ 0.032 x 230/10 = 0.73mH
4.4 Filter action against Power Line Overvoltage Pulses.
Although seen as ”clean-it-all” components, filters, being low-pass elements are unefficient against energetic pulses with duration exceeding a few μsec. This is because long pulses have a large energy content in the low frequency range, where the filter has no attenuation. For instance, a short 1kV / 100ns pulse will be efficiently damped down to 10V by a typical power line filter. But the same pulse with 50μs duration (a lightning- induced transient for inst.) is lasting long enough for the filter to get charged up to the peak value. So the 1kV pulse will have its rise time stretched, but its crest value almost unchanged. What is needed against energetic pulses is a component that is not frequency-selective but amplitude selective, like TVS or MOV.
4.5 Self-Resonance of power Line filters, and its effects.
At the exact resonance of its inductor with its capacitor, a filter has no attenuation and may exhibit a gain instead, that is an overshoot when the power is switched on or off. Typically, this overshoot manifests as a brief increase of the input voltage by 40 to 100% of its peak value, with a possible stress or damage to the post-filter power supply components: rectifiers, electrolytic capacitors etc ….This can be prevented by adding a low value damping resistor to the dc storage capacitors, or a transient voltage suppressor (TVS) after the filter, since this protection is often needed anyway for lightning and other AC mains transients.
4.6 Recommendations for Filter mounting.
As much as its performances, the way a power line filter is mounted is crucial to its effectiveness. It should be installed as close as possible to the equipment wall opening where the power cord passes, the best being a through-wall mounting (Fig. 9). Metallic filter case is preferrable, making a tight metal-to-metal contact with the equipment box, or at least with a metal barrier. Wires on the line side of the filter should never be bundled with those on the load side. PCB-mounted ”home-made” power line filters are more prone to parasitc effects that deteriorate the expected performance. Without a true metallic separation, crossing of input traces with output traces, and capacitive coupling with other nearby traces are a frequent risk.
5. EMC filters for Control/Command and low-level Signal lines.
Filtering a signal line is a different challenge: signal filters must keep the desired signal unaffected in amplitude and phase, from dc up to the highest useful bandwidth necessary for signal integrity. If not given by the datasheet of the signal technology, this useful bandwidth can be determined as follows:
Given the clock frequency or the transition time of digital pulse, the useful bandwidth, hence the EMC filter cut-off is found as follows:
- if only the clock frequency is known : Fco = 5 x F(clock) (Eq. 3.a)
- if only the transition time (tr) is known: Fco (Hz) = 0,4 /tr(sec), or using more practical units: Fco (MHz) = 400/ tr(ns) (Eq.3.b)
These formulas for practionners takes into account a 20% margin for making sure that the useful signal suffers no objectionable distorsion in waveform and phase. From this, the filter can be entirely defined through the calculation steps described in Sect.2. One must also define, from the actual interface (Balanced or Unbalanced), if the filter elements are arranged for simple DM attenuation, or for CM attenuation. Whatever the filter structure, the following conditiond must be satisfied for an optimal filtering. Given the cut-off frequency Fco:
Example 3: Filter for reducing unwanted emissions from I/O signal (Fig. 10)
Fastest Digital signal to be filtered: Clock 20MHz, tr = 4ns Zs = ZL ≈ 100Ω. It was found that the 300MHz spurious noise riding over the digital pulse train must be attenuated by 24dB. Define the proper filter.
Fco, from formulas (3.a or 3.b) = 100MHz The filter should have no attenuation below 100MHz, but must provide ≥ 24dB at 300 MHz.
Fig.2 shows that for a ratio Femi / Fco = 300 /100, a single pole filter (n =1) is not sufficient, neither is a 2 pole filter. It takes a n =3 filter, that will produce a 28.5 dB attenuation. With 100Ω being the transition for the ”Low impedance/ High impedance” criteria of Fig.3, we can select either a ”Pi” or a ”T” structure. In any case, criteria of Eq.(4) and Eq.(5) are used:
With a ”Pi” structure, C will be split into 2 x 17pF capacitors
With a ”T” structure, L will be split into 2 x 160 nH inductors
5.1 One-pole, L or C filtering for moderate attenuation
When the needed attenuation is less than 30dB, it is possible to get this figure with a single capacitor or inductor, bearing in mind that:
a) this filter will be a one-pole type, with only 20dB/decade attenuation above its cut-off frequency.
b) it will work well only if Zs and ZL are in the same range of values: both > 100Ω for a capacitive filter, both < 100Ω for an inductive filter.
For instance, if one knows that ZL = 1000Ω, one common mistake is to assume that it is enough if the filter capacitor impedance Xc is << 1000Ω to get a filtering action. In fact, no definition of the filter can me made without knowing at least the module of Zs and ZL . Cut-off frequency and attenuation of one-pole filters can be easily found in the following table:
A part of inductive filtering, ferrites toroids, beads and sleeves deserve a special mention. Often regarded as the miracle, last-minute EMC remedies, they can be integrated in the design as part of the filtering line of defense. One interesting feature is that they are generally ”lossy” ferrites, whereas some of the EMI energy is wasted in heat. They can be used as DM as well as CM blocking (2 conductors in the same ferrite hole). The relevant parameter of a ferrite is its impedance/frequency curve (Fig.11), which allow to predict its attenuation by calculating:
5.3 HF limitations of filter components
Filters made of discrete components have their own parasitic problems. As frequency increases, inductors tend to behave as capacitors because of their internal winding capacitance and the stray capacitance between their leads. Capacitors tend to behave as inductors because of their terminal leads parasitic inductance. In a sense, above a certain frequency each L or C element start doing exactly the reverse of what is expected, spoiling the filter attenuation down to 0dB. Careful filter component mounting may shift the problem to higher frequencies, but it always exist, as shown on Fig.11.A. The only filter that never has parasitic response is the feed-thru capacitor, as seen next.
5.4 Feed-through filters
Feed-through filters are based on a coaxial configuration of capacitor that suffers no parasitic inductance at all. The dielectric material is filled- in between the center pin electrode and the outer cylindrical electrode. This outer armature can be soldered, screwed or press-fitted into the chassis or PCB ground plane. Thanks to this design, the filtering is that of a perfect capacitor, with practically no frequency limitation. The coaxial capacitor can be teamed with ferrite beads to form a Pi or T filter (Fig.12.A). Since their mounting requires some mechanical work, they do not lend to cheap, mass-production items. A low-cost version exist in form of 3-terminal, surface-mount capacitor known as ”semi-feedthru”.
Although a no-match to the real feed-thru, it offers decent performance, provided the center terminal pad is soldered to the PCB ground plane by a direct via, and preferrably two vias in parallel for even less parasitic inductance (Fig.12.B).
5.5 Filtered Connectors
For filtering a significant number of identical signal lines (like a parallel bus), an interesting alternative is the filtered connector, where each contact pin is a miniature filter. A single filtered connector can replace a set of one standard n-contact receptacle + ”n” discrete filters + associated mounting space and extra wiring. In the most expensive types, each contact is constructed as a tubular ceramic feed-thru, combined or not with a tubular ferrite to form a 2 or 3-stage filter (Fig.13). Less expensive, yet efficient ”space-saver” versions are using a planar array or flexible membrane with embedded ceramic capacitors.
5.6 Mounting precautions with signal filters
More than any others, filters for high speed ( > 30MHz bandwidth) require careful precautions for PCB mounting. Input-to-ouput traces crosstalk should be avoided, and must be checked both horizontally (on a same layer) and vertically (layer-to-layer). When changing layer after leaving a filter, the filtered trace should not run close to an internal, un-filtered trace. This applies to the inter-layers vias as well.
Michel Mardiguian EMC Consultant, France firstname.lastname@example.org