My name is Perry Tsao and I'm a Systems and Applications Manager for Texas Instruments. Today I'm going to talk to you about designing low-EMI power converters for automotive systems. And this is part three, where I'll go over input filter design and debugging EMI issues.We'll start with input EMI filter design.
So there's two things to keep in mind when you're designing an input filter. The one is you need to design it so that it attenuates enough so that you can meet the standard you're shooting for. Typically for automotive, this is CISPR 25 Class 5. The second is to make sure that it doesn't interfere with the normal operation of the Buck converter.
Mil-Std-461 EMI Filter Installation 3 In Figure 2, the input terminals are shielded from the output terminals with a connector cover and shield of the wires from the connector to the filter. FIGURE 2 MIL-STD-461 is not a difficult spec to meet if care is taken in the early stages of a design to contain the conducted and radiated emissions.
So you don't want to put a filter in and make the Buck converter unstable.So on the right, I have an example of Buck converter where we're measuring the conducted EMI, then there's no input filter. And it clearly fails the CISPR 25 Class 5 standards, which are shown in red over here in this region, and then also this region.
So we know it fails, but then how do we estimate how much filter attenuation to add?So this case is pretty easy. We could just say, oh, I need this peak to go down by this amount. And then that tells us the dB microvolts level that you need to attenuate. But sometimes it's easier to calculate it because you don't have a measurement, or based on just an input voltage measurement. And so you can just measure the input ripple. And then from that, calculate the desired dB microvolt level that you want, and then get an attenuation.Or if you, for example, don't even have a design yet but you still want to do a filter, you can estimate it based on just what the input curve would be given a switching frequency, the input capacitor, and the duty cycle. And so you run this calculation and you calculate this.
You know this is the goal, the measurement that you're trying to get to. And then from that, you can get your attenuation.So once you have an idea for how much attenuation you want, basically, you can calculate the filter.
So you're almost always going to use a CLC filter, the CF, LF, and the CD. The calculation for this filter, the equations are detailed in this app note, AN-2162. And there's a spreadsheet also that's associated with it. And the spreadsheet will basically make it easier for you to just input the information you want and then a calculate a filter. And then it will even estimate the performance of the filter as well.So I'll go through one quick design example.
And so we saw the EMI plot for the LM53603 without the input filter. So based on that measurement, we know we want to reduce the noise by 45 dB microvolts. And so we enter this information into the spreadsheet and then it estimates that we could get a filter transfer function like this that basically is saying at- this is 5 megahertz. We're down like 80 dB.And so we're basically getting more than 45 dB microvolts that's necessary for this frequency region, from maybe 2 megahertz to 100 megahertz. So there's clearly parasitics in here, so the filter will ring. The input capacitor will have a soft resonant frequency.
The inductor will have a soft resonant frequency.So this filter eventually stops being effective. Attenuation?
will disappear. So this basically is going to- you can see based on this transfer function, you get the idea for where the filter will be effective and attenuate.
So it looks like, in this case, up to about 70 megahertz, and then it's going to start to get worse.So in this design example, what we did is we took that input filter and we put it on this Buck converter. And so this is the conducted EMI plot on the left without the input filter, and on the right with the input EMI filter.
And you clearly see there's a large reduction in all these peaks. And so now we can pass the CISPR 25 Class 5 limits.So this is a 30 megahertz to 108 megahertz plot.
And on this plot, it's basically 30 megahertz to 108 megahertz. And so that's this full frequency range of the conducted EMI test for the CISPR 25 setup.Now we're going to talk about debugging EMI issues. Inevitably, in some of your designs, they will fail EMI initially, right? So what should you do? So I guess the first thing is to blame the test setup. So that's very tempting, but often it's quite true that the setup is to blame.Second thing is to turn power on and off to the board. So this can help you figure out where the issue is coming from.
You can cycle the power, like if you have multiple power systems on your board. Sometimes the EMI issue comes from the output, not the input. So you can exclude other noise sources, turn sections on and off.The other one is just kind of moving the input wires around.
Sometimes it's due to the way the input wires are routed. This is an example of where the setup was actually to blame. So in this case, we had a lab of power supply turned on, but the unit under test was actually off and not doing anything. But you can see that there was these big spikes that actually caused failures. So in this case, the test setup was the problem. And it was easy enough to just change this power supply off and use a battery.Another question that often comes up is because the CISPR 25 standard, it has some guidelines of how you should set up the test for conducted emissions, for example, but they're not rigid, right?
And so there is some flexibility. And some of it depends on the product that you're trying to test.Oftentimes, in our case, we're not testing a full product. We're just testing Buck converter on an evaluation board. And so on that evaluation board, there is oftentimes not a load. And so the question is, what's better? Should you put the load right on top of the board, or should you maybe put it off to the side? In this case, we took two wires and twisted them together and put a load resistor off to the side.And so which scenario is more realistic kind of depends on what application you're shooting for.
In some applications, the power is consumed entirely on one PCB. In other cases, there is some power being transmitted over wires. But in general, what setup is better kind of depends.
And it depends on what you're trying to do, obviously, in terms of your application.But it also- here, these two graphs kind of illustrate how it affects the shape of the conducted EMI. So this is a measurement from 30 megahertz to 200 megahertz. You can see when the load resistors are right on top of the board, you have this resonance that's kind of over here, in the 110 megahertz region. And when you have the wires going a long distance and then to a remote load, you tend to get this resonance, like this peak, now it happens here.And so if you're just measuring conducted EMI, then you would most likely prefer this one because the conducted EMI measurement stops at 108 megahertz. But there are cases where you should do a measurement like this because that's actually closer to what the actual application is. And maybe you need to add some additional filtering on these output wires to get rid of this resonance. Because what this really is is the capacitive coupling between these output wires and this copper plate on the bottom of the test stand.So this is actually just the measurement data from a Buck converter without an input filter.
But clearly, if you're failing because of low harmonics of the switching frequency, changing the layout's not going to help, really. What you need to do is have a better input filter. So this is why you have differential input filters, to get rid of this first, second, third harmonics of the switching frequency.If you're failing at the high harmonics, and then this can often be due to board layout, possibly because the input capacitor to the Buck converter is too far away. Or maybe it's not a good capacitor. So this is the frequency region where yeah, you would tend to look at the board layout and the input capacitor placement. So this is kind of around 30 megahertz or so.And then another place where people often fail is for conducted or radiated noise, you'll see that these peaks get larger right around 100 megahertz.
And this is often where the switch node ringing becomes an issue. So this is where you might want to try some slew rate control to help reduce this peak. And this region is also where sometimes the parasitic capacitances of where the switch node couples into the ground plane cause an issue. And this can often be fixed with a shield, or a common mode choke, or sometimes you just need to do a better layout, or slew rate control can also help.OK. So just some overall high level tips. So number one is pass conducted emissions first because it's easier to measure.
And generally, if you don't pass conducted EMI, you won't pass radiated EMI. Find the root cause before you start your redesign. I think this is very critical because otherwise, you're just shooting in the dark.And I think the other thing is just to keep in mind that just because it passes today doesn't mean it will pass future requirements. The requirements are getting tougher and tougher. And the designs are also becoming more complex. So as you try to cram more stuff into the same board, you tend to have more issues.So EMC tests are just models of the vehicle at the system level. They're not 100% accurate and they're always evolving.
So just keep that in mind that this is a changing area and you need to stay on top of things to stay ahead of competition.OK. Thanks for listening. If you're interested in finding out more information, go to ti.com/autodcdc where we have lots of links to other information about automotive DC-DC converters. Arrow-top close delete download menu search sortingArrows zoom-in zoom-out arrow-down arrow-up arrowCircle-left arrowCircle-right blockDiagram calculator calendar chatBubble-double chatBubble-person chatBubble-single checkmark-circle chevron-down chevron-left chevron-right chevron-up chip clipboard close-circle crossReference dash document-generic document-pdfAcrobat document-web evaluationModule globe historyClock info-circle list lock mail myTI onlineDataSheet person phone question-circle referenceDesign shoppingCart star tools videos warning wiki.