EMC Challenges in Electric Vehicle Automotive Systems

Introduction

The propulsion systems of electric vehicles pose a number of EMC risks. The one that has received the most public attention is probably the interference it can cause to AM radio reception; however, the interference can spread to control systems on the vehicle itself and to other nearby RF systems. EVs can be electromagnetically loud, even while they’re being so audibly quiet that companies have to add in substitute engine noise so unobservant pedestrians will notice them.

The electromagnetic noise in a moving EV is dominated by the inverter, and to a lesser extent the DC/DC converter of the system. Figure 1 shows a typical architecture of the high voltage (HV) side of an EV. The HV system only deals with power; control signals are handled on the low voltage (LV) side. In this system, the HV battery will be supplying DC power at several hundred volts. The DC/DC converter will step that high voltage down to 12 Vdc to power the typical loads in a vehicle (power steering, anti-lock brakes, infotainment, etc.) in place of an alternator. The inverter takes the HV DC from the battery and converts it to power the motor — often converting from DC to three-phase AC, but that’s not universal. Then, depending on the vehicle, at times the EV motor can be free spinning (if the vehicle is, for example, rolling downhill or from regenerative braking) and can generate power that can charge the battery. That power will also go through the inverter. The HV battery may power other HV loads, such as certain cooling systems, but this covers the basics.

Figure 1. Generic HV architecture.

In order to perform the necessary power conversions, the inverter typically uses a switch mode power supply. The transistors used are traditionally insulated-gate bipolar transistors, but wide band gap devices such as SiC and GaN transistors are becoming more popular.

EVs are notorious for creating electromagnetic interference in the 50 kHz – 50 MHz range, even when their fundamental switching frequencies are 20 kHz or below. Their combination of fast switching operations with very high currents means that even their high harmonics can have substantial effects. The higher order harmonics of a square wave are supposed to fall off as per Figure 2a and Figure 2b.

Figure 2. a) First through one-hundred-and-first harmonics of a trapezoidal waveform and b) first through twenty-first harmonics showing the dramatic fall-off expected.

If one has a system switching 50 mA, the fact that the third harmonic (60 kHz for a 20 kHz switching operation) is down to 20% of that, or 10 mA, means that the effects will likely be small. By the time one reaches the fifty-first harmonic (1.02 MHz, right in the AM band), they are likely down to 1/1000 of the original amplitude, 50 uA. However, if starting from a system switching 20A, then the third harmonic may yield an amplitude of 4 A, and the fifty-first may still have an amplitude of 20 mA, more than enough to interfere with neighboring antennas or CAN transmission lines. These higher frequency currents must be carefully controlled.

EV Challenges

Figure 3 shows a generic EV inverter architecture, with the HV battery, filtering at the input of the inverter, and three switching banks to feed a three-phase AC motor.

Figure 3. Generic EV inverter architecture.

It is important to remember all the other capacitive paths available to higher frequency noise — the “hidden schematic” as Dr. Clayton Paul used to put it. Some potential parasitic paths are notionally illustrated in Figure 4.

Figure 4. EV inverter with chassis-to-enclosure included and parasitic capacitance highlighted.

One would tend to say that this parasitic capacitance, directing high frequency noise onto the enclosures and chassis, is unfortunate. The goal is to avoid this noise to escaping the electronics and getting onto structure, where it can easily cause a radiated emission or cross-talk issue.

However, this effect may be compounded by the choice of filter topologies. Many app notes and other design guides default to including Y-caps in filters, and these may also be present in the filters in EV modules. Two ways of showing this topology are in Figure 5a and Figure 5b.

Figure 5. a) Inverter with Y-caps as typically depicted on a schematic. b) Inverter Y-caps shown with more explicit connection to chassis.

Looking at Figure 5b, it is hard to see why, if the parasitic capacitance depicted in yellow is bad, that the intentionally placed capacitance highlighted in red is good — unless the intentional Y-caps in red are helping re-direct the noise following the parasitic paths in yellow. Depending on the application, placing Y-caps can help or hurt. When Y-caps are called for by an app note, one may want to consider placing pads for the caps but not placing them, and only adding them in if there’s an issue and they’re shown to improve performance.

This is especially important since, once the high frequency noise from the switching harmonics arrives on the housing, it is almost impossible to predict where it will go from there. Every harmonic, being at a different frequency, will see a different set of impedances looking at the structures and cabling of the vehicle. Even a “minor” higher-order harmonic may see a low impedance path that allows it to become a more severe radiated emissions threat than one would expect.

Unpredictable Noise Illustration

Figure 6 shows a simple test setup featuring a signal generator producing a 20 MHz square wave. It outputs the signal onto a cable terminated with a passive 50 Ω resistor. There is a current probe hooked to a spectrum analyzer looking at the RF currents on the cable (common mode), and also a TinySA hooked to a rod antenna 1 m away monitoring the radiated emissions from the cable. Both spectrum analyzers are set to look at 10 MHz – 87 MHz (to avoid distracting the reader with the FM radio energy passing through the author’s garage).

Figure 6. a) Test setup with signal generator, spectrum analyzer hooked to current probe, and TinySA hooked to rod antenna 1 m distant. b) Spectrum measured by current probe. c) Spectrum measured by TinySA.

In this frequency range we are seeing the fundamental switching frequency and the second, third, and fourth harmonics. If we look back to Figure 2b, the second and fourth harmonics should not be visible (if this were a perfect square wave), and the third harmonic should be less than half the amplitude of the fundamental. Instead, Table 1 shows the values with both spectrum analyzers set to max hold.

Table 1. Values Measured from the 20 MHz Signal Generator Test Setup

Frequency (MHz)	CM Conducted (dBμV)	Radiated (dBμV)
20	74	61
40	54	56
60	69	65
80	56	48

What can be observed here is the effect of uncontrolled and unpredictable impedances on the emissions of a system. Inside the signal generator, at the point where this switching noise is produced, the voltages likely follow a profile much closer to the ideal shown in Figure 2b. However, there is a lot of non-ideal behavior even in such a simple system:

The even harmonics are much more prominent in both measurements than one would expect — possibly from jitter in the square wave generation

On the conducted side the third harmonic is 5 dB down from the fundamental

On the radiated side, the third harmonic is 4 dB up from the fundamental

On the conducted side, the second harmonic is 20 dB down from the fundamental

On the radiated side, the second harmonic is only 5 dB down from the fundamental

(Note: the conducted and radiated values are not directly comparable, due to the different measurement methods and the different RBW settings on each. The spectrum analyzer was using a 1 MHz RBW, and the maximum RBW of the TinySA used was 621 kHz.)

What this demonstrates is that the 60 MHz noise likely saw a much lower impedance in the cable path than the 20 MHz signal did, allowing the 60 MHz noise to generate more current in the cable than would be expected. Likewise, the cable is almost certainly a better radiator at 40 and 60 MHz than at 20 MHz, allowing it to produce higher radiated emissions at 40 and 60 MHz than predicted either from mathematically normal harmonic fall-off or from the common mode conducted emissions on the cable.

Potential Mitigations (and Their Pitfalls)

What is an entertaining demonstration with a 5 V square wave can be a massive and unpredicted regulatory test failure in an EV pulling 20 A at 700 V, leading to schedule delays and cost overruns. The best place to fix this issue is on the board, with careful attention to layout and filtering. Once noise escapes onto housing and cabling, the options are limited and unreliable. Consider the exterior noise mitigation techniques attempted in Figure 7 and the results in Table 2.

Slowing down rise/fall times (triangle waveform)

Re-routing cable

Extending cable (doubling length)

Re-routing the extended cable

Adding a ferrite

Figure 7. Mitigation techniques when board/box interior cannot be altered. a) Slowing rise/fall times, b) re-routing cable, c) changing cable length, d) re-routing extended cable, and e) adding ferrite.

Table 2. Results of Different Mitigation Techniques

Current Probe

Antenna

Mitigation	20 MHz	60 MHz	20 MHz	60 MHz
Triangle Wave	-4 dB	-12 dB	-10 dB	-9 dB
Folded Cable	+2 dB	+6 dB	-26 dB	+6 dB
Extended Cable	-1 dB	-1 dB	-4 dB	+8 dB
Extended and Serpentine	-2 dB	+4 dB	-17 dB	+16 dB
Ferrite	0 dB	-3 dB	-1 dB	-12 dB

With these mitigations, results are as much a function of luck as anything else. The ferrite selected had an advertised frequency range of 1 MHz to 300 MHz, but in this application, it had almost no effect on the 20 MHz spike, while providing a good reduction to the 60 MHz radiated emissions. Changing the length and routing of the cable often had different effects at different frequencies, sometimes improving the 20 MHz region but making 60 MHz worse. The only technique that had a salutary effect across the board was slowing down the rise/fall times of the waveform, and of course that is usually off the table for power and thermal efficiency reasons. Any of these modifications could be further tweaked to “optimize” their performance — changing the position or number of ferrites, playing with cable lengths and arrangements — but they don’t address the root cause of the problem.

In a high-power system such as the HV side of EVs, even small imperfections in the topology and layout inside the box can have major consequences for success or failure in EMC testing (primarily CISPR 12, 25, and 36). Parasitic pathways that would be inconsequential in a LV design can allow limit-busting amounts of noise to escape in HV system. Close attention to inside-the-box design, through best practices, simulation/analysis, and expert design reviews is critical for EV compliance and on-time deliveries.

EMC Challenges in EV Systems

Related Resources

Related Articles

Report Abusive Comment