Power conversion topology and EMI noise

All electronic devices are powered by direct current, usually after AC rectification. The voltage is then converted by a DC-DC converter to the voltage required by the load. Currently, most DC-DC converters are based on high-frequency switching technology. Effective high-frequency switching has always been regarded as the key to module power density and performance. The higher the switching frequency, the smaller the magnetic components and capacitors used , the faster the response time, the lower the noise, and the smaller the required filter.

However, all DC-DC converters still generate electromagnetic interference (EMI) or noise, and the noise level generated, whether common mode, differential mode or radiated noise, can vary greatly from one manufacturer to another or from one conversion technology to another. The root cause of these differences lies in how the noise is generated.

Although no power conversion topology is perfect, some topologies are particularly suitable for certain application requirements. There are hundreds of DC-DC converters on the market, each with different designs and topologies, which can be roughly divided into two categories: pulse width modulation (PWM) and quasi-resonant zero current switching (ZCS).

It is a very daunting task to fully understand the number of topologies, so this article focuses on analyzing the noise performance of two mainstream topologies, specifically comparing the performance of fixed frequency DC-DC converters (PWM) and variable frequency quasi-resonant DC-DC converters (zero current ZCS).

Comparison between pulse width modulation and quasi-resonant zero current switching

The power density of pulse width modulation (PWM) modules is limited because it requires a trade-off between operating efficiency and switching frequency. The core of the problem is "switching losses". The switching element causes discontinuities in the inductor current when it is turned on and off instantaneously, thus generating heat. The power consumption caused by switching losses will increase directly with the switching frequency of the PWM module until it becomes a significant loss contributor. At that point, efficiency will drop rapidly and the thermal and electrical stress on the switching element will become unmanageable. This non-zero current switch module has the property of switching losses, which becomes a switching frequency barrier and limits its ability to increase power density.

The quasi-resonant zero-current switching converter uses a forward switching topology, which switches only when the current passes through zero, overcoming the switching frequency barrier. Each switching cycle transmits an equal amount of "energy packets" to the output of the module. Each "on" and "off" is performed at the moment of zero current, forming a switch with almost no power consumption. The operating frequency of the zero-current switching converter can exceed 1 MHz. It avoids the discontinuous current characteristics of the traditional topology; it realizes "no power consumption" to transfer energy from input to output, greatly reducing conduction and radiation noise.

There is a big difference between the noise generated by PWM and ZCS converters. Figure 1 compares the conducted noise of PWM and ZCS converters. It is obvious that the waveform of the ZCS converter is a sine wave instead of a square wave. In addition, since the current waveform does not have the sharp parts of the rise and fall that are almost vertical, and the harmonic content is lower, the stress of parasitic components is reduced, so the noise is lower. In contrast, the input voltage of PWM is switched at a fixed frequency (usually hundreds of kHz) to form a series of pulses. The width of the pulse is adjusted to provide the correct output voltage and sufficient current for the load. At full load, the current waveform looks like a square wave (Figure 2).

Figure 1 – Conducted input noise spectra of a zero-current switching converter with a common-mode choke (left) and a pulse-width modulation converter with a filter (right).

Figure 2 - Current waveforms for zero current switching and PWM topologies

Many power engineers think that it is easier to filter out noise from fixed frequency converters than from variable frequency converters, but the opposite is true1. This is simply an illusion created by the term "fixed frequency". It is basically a "misnomer". Both architectures have both roughly fixed frequency elements and variable frequency elements that change depending on the operating point.

Converter specifications: 48 V input, 5 V output, 30A.

Figure 2 compares the waveform of the current flowing to the main switch. The bandwidth or on-time T1 of the quasi-resonant converter is fixed, while the switching frequency T2 is variable. In contrast, the switching period of the PWM converter is fixed, while the bandwidth is variable. Figure 3 shows the noise spectrum produced by these two topologies.

Figure 3 – Current waveforms and spectra for PWM (top) and zero current switching (bottom). Note: Waveforms are not drawn to scale.

However, in the design of variable frequency, because it is basically a half-wave rectified sine wave, there is no high-frequency component involving the rising and falling steep edges of the current waveform. Therefore, the waveform spectrum amplitude of the variable frequency converter is lower and the bandwidth is narrower.

In a PWM converter, most of the energy is at the fixed frequency and its odd multiples (harmonics). A 100 kHz PWM converter has conducted noise mainly at 100 kHz, with some at 300 and 500 kHz. Because it is a square wave, there are significant harmonics between 10 – 30 MHz, which means that the high di/dt excites the parasitic elements in the converter. Adequate input filtering is required to filter out the 100 kHz noise at full load. The waveforms of these converters have higher spectral noise levels and a wider range of harmonic distribution.

Obviously, if you want to minimize the noise of the DC-DC converter, the first step should be to choose a suitable topology, such as zero-current switching, which has low inherent common-mode noise. In addition, in noise-sensitive applications, converters with the following characteristics should be avoided: If the control device is mounted on a copper board, this will cause parasitic capacitance between the primary control element and the secondary control element through the copper board, thus forming a higher amplitude common-mode noise.

Passive EMI Filters

Although power modules usually have internal input and output filters, external filters are required to meet system requirements or meet recognized specifications such as FCC and EU standards for noise conducted from power systems to the power grid. Many power engineers will design their own solutions. Most DC-DC converter manufacturers will provide detailed application notes and send knowledgeable and experienced application engineers to help solve these problems. In addition, there are some DC-DC converter suppliers that provide AC front-end and EMI filter modules. Using these filter modules not only saves time, but also guarantees quality and performance. These EMI filters are designed specifically for the supplier's converter modules. As long as the wiring is proper and the converter module and filter are used together, it is guaranteed to meet specific EMC specifications.

In the United States and Europe, conducted noise is strictly regulated by FCC and VDE Class A and B limits. In the United States, conducted noise from industrial equipment should meet FCC Class A requirements, and conducted noise from home appliances should meet the more stringent FCC Class B requirements. In Europe, all countries require industrial equipment and home appliances to meet EN55022 (or VDE) Class B standards.

The switching frequency of most current switching power supplies is between 100 kHz and 1 MHz. The main peaks on the conducted noise spectrum reflected to the power grid usually come from the fundamental frequency of the switching frequency and its harmonic components.

These conducted noise standards, such as EN55011 and EN55022, specify that the conducted noise reflected from the converter to the grid must not exceed a specified upper limit in the frequency band from 150 KHz to 30 MHz. To meet these requirements, all conducted noise, i.e. the sharp points in the frequency spectrum, must be below the specified limits.

These EMI filters are usually made into a single device, (configuration similar to Figure 4). It is a device with through-hole pins, with common mode chokes and Y-capacitors (line to ground), plus two inductors and an X-capacitor (line to line). Transient protection is provided by Z1, and such a filter has enough attenuation to meet the Class B conducted noise limits.

Figure 4 – EMI input filter compliant with EN55022, Class B

In addition, capacitors, inductors, and filters (active and passive) are often used to attenuate conducted noise (whether common-mode or differential-mode noise). The following will discuss the filtering performance of various components after adding them one by one, and propose a new EMI solution.

Figure 5a on the left shows a 48 V DC-DC converter with a differential mode capacitor C1 connected to the input. This single 120 F, 100 V electrolytic capacitor is used to keep the input impedance low, stabilize the voltage and ensure good transient response. It stores energy for the module and should be as close to the input as possible for best results.

Taking this configuration as a starting point, Figure 5a on the right shows the harmonic levels of a 48 V input, 150 W full load DC-DC converter with a differential mode capacitor connected, as well as the EMI and harmonic standards required by Class A and B. Obviously, adding only a differential mode capacitor will not meet the requirements.

Figure 5b shows the case where bypass capacitors and differential mode capacitors are added. Although the noise level still does not meet the standard, it has been significantly improved. Note that each bypass capacitor connected to the input and output terminals is grounded to the substrate. These capacitors are 4700 pF, 100 V Y-capacitors commonly used in the industry. Y-capacitors are very effective in attenuating the noise derived from the converter.

The noise generated by a 48 V converter at full load is higher than that of a 3.3 V converter at half load. However, a clear improvement can be seen in Figure 5b.

Adding a 27 µH differential mode inductor L1, we can see in Figure 5c that the 48 V converter still does not meet the standard in the low frequency part and the noise level is still higher than the Class B limit.

Figure 5d shows that a common-mode choke is used to replace a differential-mode choke. The common-mode choke itself also has differential-mode inductance and can replace the differential-mode choke. The common-mode choke can increase the attenuation capability of the Y-capacitor because the common-mode choke forms a high impedance to the common-mode noise generated by the converter, allowing the noise to be transmitted to the ground through the Y-capacitor along a lower impedance path.

Figure 5 - Noise spectrum of a 48 V, 150 W DC-DC converter connected to different components.
a. Differential mode capacitor b. Bypass capacitor c. Differential mode inductor d. Common mode filter (without differential mode choke)

The noise of the 48 V converter is now just slightly above the Class B standard and requires a little more filtering. With the addition of the common-mode filter, the 3.3 V converter is now fully compliant with the Class B standard at both full and half load.

Active EMI Filters. The electronics industry is constantly demanding smaller products with more features, and this trend is irreversible. The size of the system is constantly shrinking, and more functions need to be squeezed onto the board or in the rack, which greatly increases the chances of interference between instruments. As the frequency increases and the voltage level decreases, electromagnetic interference control becomes a very important design task. It is very complicated to control electromagnetic interference well. The entire design is affected by many factors, and a variety of filters (active or passive) are used to manage the conducted noise.

Compared to passive filtering solutions, active filters can reduce the space occupied by the common mode choke, making the entire component volume only 1" x 1" x 0.2", which is a very thin surface mount component. Overall, this solution saves circuit board space, and the component is thin, allowing air to flow over it to help dissipate heat.

The active EMI filter (see QPI in Figure 6) can attenuate common-mode and differential-mode noise between 150 KHz and 30 MHz, meeting EN55022 (CISPR22) requirements.

Figure 6 - Connection diagram of active filter (QP1) and DC-DC converter.
The values ​​of Cin, C1, C2, C3 and C4 should be provided by the DC-DC converter manufacturer.

Figure 7 is a noise test diagram with and without an active filter connected. The test conditions are in accordance with the CISPR22 standard. The results show that the total noise of the loaded DC-DC converter is lower than the EN55022 Class B quasi-peak detection level, indicating that the active filter effectively filters out the conducted noise.

Figure 7 - Conducted EMI noise from a DC-DC converter.
With active filter (lower picture); without filter (upper picture)

When selecting and evaluating EMI filters, designers should be aware that they must test the performance of the filter used in their products, and the test equipment and conditions must comply with the EMI standards that their products must comply with. When selecting filters and appropriate designs, reference should be made to the amplitude and spectrum without the filter.

The conducted noise of a product should include differential mode and common mode noise, and may also include radiated noise, which depends on the measurement device of the EUT shielding and wiring shielding. CISPR 16-2-1 of the IEC International Electrotechnical Commission lists the method for measuring conducted interference.

The performance of a filter is very dependent on the input bus and load impedance. It cannot be inferred from zero bias, 50  insertion loss data alone. The filter components, instrument grounding, and noise source impedance will affect the final noise performance and change the amplitude and phase of the associated spectrum.

Active EMI filter can attenuate common mode and differential mode noise between 150 KHz and 30 MHz, meeting EN55022 requirements. It generates low impedance on the shield plate by inducing common mode current flowing to the busbar, directing the noise to the source of the noise. When the active EMI filter is properly connected as shown in Figure 6, the control circuit will actively drive the shield pin to reduce the common mode current in the busbar until the common mode current value is attenuated to the level shown in Figure 7.