Power supplies are natural generators of narrow-band noise, present at the fundamental of the switching frequency and its associated harmonic frequencies. To contain the noise, it is critical that designers identify its source and the paths it takes in becoming radiated or conducted emissions. Direct radiation of noise can be stopped with simple measures such as enclosing the supply in a metal box or spray coating the case. Designers also need to pay attention to the internal layout of the supply and the wiring that goes in and out of the supply.

However, interference conducted through the input or output terminals is more difficult to suppress. But with proper transformer design, connection of the heatsink, and filter design, conducted interference also can be reduced, such that the supply can achieve EMI regulatory agency approvals without incurring excessive filter cost. **Fig. 1** shows the typical conducted emissions of an ac power supply depicting the limits on EMI set by the European standard EN55022.

In a typical power-supply design, a high dc voltage is changed to a chopped or a pulsed waveform. This process generates discrete, unwanted narrow-band signals spaced at the repetition frequency that can be either conducted or transmitted through a system with harmonics over a wide range of frequencies.

Most of the conducted interference on the power line is the result of the main switching transistor or output rectifiers. It is very desirable for a power-supply designer to predict signal levels resulting from these types of waveforms to determine the degree of filtering or shielding needed to meet the EMI limits before actually building any power supply.

Let's examine a power supply with a 110-Vac input. The square waveform at the drain of the main switching transistor has three fundamental characteristics in the time domain that have a direct influence on calculating the predicted emission in the frequency domain. These characteristics, which are depicted in **Fig. 2** and **Fig. 3**, are amplitude, rise time and duty cycle.

The fourier envelope of the spectral energy for **Fig. 2** can be estimated by **Fig. 3**, with slope changes indicated by corner frequencies at F1 and F2. The location of each frequency (F1 and F2), which corresponds to a change in the amount of attenuation, can be tailored based on adjusting the duty cycle (D) and rise time (t_{RISE}) of any given waveform.

Assuming 90% of the spectral energy within a signal is contained in the first 10 harmonics, then by placing the location of F2 at a much lower frequency than the tenth harmonic, we can obtain an attenuation of 40 dB/decade with the proper rise-time setting. Harmonic energy content is a function of the waveform's rise time and can be estimated using **Table 1**.

A pulse train such as that of a switching waveform generates discrete, narrow-bandwidth signals spaced at the harmonics of the switching frequency. As **Table 1** suggests, a pulse train with a faster rise time spreads these discrete, narrow-bandwidth noise signals over a wider frequency range, such that there are more, higher-order harmonics. In contrast, a pulse train with a slower rise time exhibits a fundamental with a broader bandwidth, while spreading the discrete, narrow-bandwidth noise signals over a narrower frequency range.

Keep in mind that tailoring the rise time involves a tradeoff of the amount of power dissipation through the main switching transistor versus the level of generated emissions. Fast rise and fall times limit power dissipation in the switching transistor, but produce greater emissions.

By calculating the harmonic content of a pulsed waveform with a given rise, duty cycle and amplitude, we can predict an EMI spectrum over a conducted range (such as 150 kHz to 30 MHz) for any given rise time:

where HRMC is the harmonic content of the pulse and T is the period of the switching waveform.

Using the first equation, the conducted emissions for a power supply with a 110-Vac input, a 150-kHz switching frequency and a 48% duty cycle are calculated and shown in **Fig. 4**. Comparing these results with **Fig. 5** provides a preliminary indication of the levels of attenuation in decibels that will be needed for this power supply to meet the regulatory limits on emissions.

That attenuation of emissions can be achieved through filtering. Additional attenuation can also be obtained by lowering the rise time, but with the penalty of additional temperature rise in the main switching transistor.

To meet the EMI requirements for conducted emissions, the voltage across the line impedance stabilization network (LISN) should be calculated for all frequencies throughout the conducted emissions range. In power-supply products developed for Europe, emissions limits are determined by EN55011 and EN55022, while products designed for North America must meet the limits established by FCC CISPR22. In the case of EN55022, we can translate the emission limits into a voltage level in microvolts (µV) or millivolts (mV), per **Fig. 5**.

Considering the 500 µV needed across the LISN to meet the emission standard with the maximum voltage across the main switch (155 V), we can estimate an overall needed attenuation:

The above calculation does not account for the effects of leakage inductance, which results in a voltage spike at the front end of the drain pulse when the MOSFET turns off. So some additional attenuation will be required to suppress the noise associated with this spike. If we assume 120 dB of total attenuation and take the anti log of (120/20), we find that the signal associated with the drain voltage must be reduced by a factor of at least 10^{6} from drain to line, before it reaches the LISN. Now that we have the predicted emissions, they can be compared to the limits shown in **Fig. 5**, to determine the amount of attenuation required across the frequency range of interest (**Fig. 6**).

In the primary section of the power supply, one problem-generating area is the metal tab of the main switching transistor (**Fig. 7**). The transistor is usually mounted in close thermal contact to a grounded heatsink through an insulator or layer of Kapton tape. In order to maximize the heat transfer from the MOSFET to the heatsink, the Kapton is made as thin as possible, which creates a parasitic capacitance (C_{HS}) from the drain to the chassis. **Fig. 8** illustrates how the presence of this parasitic capacitance causes a noise current to flow through safety capacitors Y1 and Y2.

Choosing the proper values for Y1 and Y2 and selecting the right thickness of the Kapton tape is crucial (**Table 2**). Doing these steps correctly ensures that the noise current (I_{NOISE}) caused by turning on and off the main switch stays in a guided path, so that the expected attenuation is obtained from the filter. If capacitor values and Kapton thickness are poorly chosen, I_{NOISE} will bypass the filter and flow through the chassis.

Both Y1 and Y2 need to have a specific value to meet the ground safety current requirement mandated by Underwriters Laboratories, to protect the user when the safety ground is not properly connected. The value for the line-to-chassis Y capacitors can be calculated as:

where I is the needed leakage current, F is the line frequency (usually 50 Hz, 60 Hz or 400 Hz) and V is the ac line voltage with a typical value such as 110 Vac or 220 Vac (**Table 3**).

When the value of these shunt capacitors is limited to meet some stringent requirement for leakage current, we can increase the inductance value of the input filter to gain more insertion loss. But keep in mind that there is a voltage drop (V_{L} = Ldi/dt) associated with the filter inductor.

The heatsink capacitance C_{HS} and Y capacitances (**Fig. 8**) form a high ratio divider. With a proper selection of Kapton thickness, we can obtain both a low thermal resistance and a specific capacitance value for CHS that could be used to reduce the noise current by as much as 1000 times from the drain before it reaches the input filter. Then, given the overall attenuation requirement calculated previously (10^{6}), the total remaining attenuation from drain to line should be 10^{3}. A typical C_{HS} value is in the range of 10 pF to 100 pF.

At 150 kHz, a second-order line filter can attenuate line-to-ground interference by 40 dB (100 times) with a corner frequency of 15 kHz. This leaves a total remaining attenuation requirement of 10^{1} from drain to line.

Certain inherent elements within the power supply, such as transformer isolation and line inductance and capacitance, will attenuate the signal even before they reach the input filter line. If additional attenuation is needed, lowering the rise time of the main switch is an option given that the added switching losses are acceptable.

Another path for noise current is associated with the interwinding capacitance (C_{WW}) from the primary to secondary winding of the main transformer. This parasitic capacitance couples the switching harmonics into the ground plane along a similar path to that shown in **Fig. 8**.

Many switch-mode power-supply designers are at the mercy of the magnetic supplier and depend on the supplier's expertise for a good transformer design. So, in some cases, the power-supply designers might not fully understand the effects of transformer construction on emissions.

For any real transformer, there is a small capacitance linking the windings together; this capacitance is a function of the spacing and the dielectric used between windings. The size of this interwinding capacitance can be reduced by increasing the separation between the windings and by using low permittivity material to fill the space between the windings.

The number of layers used between the windings, their thicknesses and the material type can be designed to obtain a specific amount of capacitance and leakage inductance to lower emissions down to an acceptable level.

Power-supply designers should also note the potential impact of output rectifiers on the secondary side of the transformer. When these rectifiers change from forward- to reverse-bias states within the time frame of a few nanoseconds, they can produce spikes that extend the EMI envelope beyond the first 10 harmonics. That additional, higher-frequency noise can couple back through the main transformer capacitance, into the primary section, easily pass through the EMI filters and be radiated by the ac line cord.

Several factors have a direct influence on the inter-winding capacitances that provide paths to couple the harmonic noise frequencies to ground. One factor is the transformer winding technique, which could be a single layer, progressive or bank type of winding.

Another issue is the use of a shield, possibly fashioned from copper tape, in the primary section, secondary section or both. All of these factors also have a direct effect on the coupling between each winding and on leakage inductance, which can aggravate the high-frequency ringing during the main transistor's off time.

In general, the key to minimizing the conducted interference is to keep an eye on all interwinding capacitances, as well as capacitances from traces or wiring to ground. Designers should also minimize the inductive coupling from current-carrying conductors by reducing the loop area that the high-frequency noise has to travel through.