All right. Hey, Pradeep. 

Yes? 

OK, fantastic. We'll go ahead and get started. So welcome to the automotive TI Tech Day. This is the "Tips and Tricks for Addressing EMI Issues in Power Supplies" session being presented by Pradeep, Brian, and Bob. My name is Jared Becker. And I'll be the moderator for this session. 

All participants are going to be muted. But please feel free to use the chat function to ask questions. Or if you're having any trouble hearing or seeing the presentation, you can let us know there and we'll help address that for you. So with that said, I'll go ahead and hand it off to Pradeep to get started. 

Thanks, Jared. Glad to be with you all today. EMI is definitely a hot topic for us. It's very challenging at times to figure out. And we'd love to share with you some of the tips and tricks that we've learned along the way. 

The general flow for today, we're going to break this presentation up into three parts. First of all just talking about the background and introduction to EMI, what are some of the sources of them in power supplies; how do you test for it; things like that. Then the next section, Brian King, my colleague is going to take over and talk about how we go about debugging EMI problems, and some of the tips and tricks you'd used there. 

And then in this last section, Bob Sheehan, another teammate of mine will go through a design example. He'll show a couple of designs that he did and the path that he set to passing CISPR 25. Now, before we dive in, I'm going to try to open up a poll and see if you guys can give some feedback. 

If you were to rank the top design challenges that you face, where would you put EMI and EMC? Is it your top design challenge? Is it maybe somewhere in your top three, in the middle? Maybe it's the least of your concerns. Maybe you've never even heard of EMI and EMC. I'd love to get your feedback on that. 

I see some of the responses coming in right now. So far this is what I'm seeing. I don't know if you guys can see this. I'm still kind of new to this WebX system. It seems like most people are saying it's in their top three design challenges. That is probably why you're here. 

So I'm glad to hear that it's relevant to you. And hopefully for the one person who says it's their top design challenge, hopefully you'll walk away with some good insight and tips for solving the challenges that you see. Well, let's move on for sake of time. Thanks for that feedback. 

For today what we're going to look at, like I said, is the overall system when it comes to EMI, EMC, and designing to pass electromagnet compatibility requirements. And I thought I'd start with just a general overview diagram of a system that you'd typically see in a vehicle where you'll have your power input. Maybe it's a 12 volt right off the battery or some power coming from your alternator. 

And you're trying to power some loads and there's various power converters and various loads that you could have in your ECU. And the key thing I wanted to show in this diagram here is that your power is flowing from left to right; where your power is DC and it's flowing from your power input to your load. But the noise that's being generated over here by your power converters is propagating from right to left. 

So the AC portion of that signal is going from the right-hand side to the left-hand side. And usually what we have in these systems up front is some various protection and filtering circuits for suppressing transients and reverse polarity protection. And then what we'll spend probably more time talking out, though, is some of the filtering-- like differential mode pi filters, common mode filters, things like that are often used to address EMI problems. And this specifically is looking at conducted EMI in this example. 

Now, where do the sources of these emissions, these noise, this EMI, where does it come from? Well, with switching power converters, if you're familiar with them-- and hopefully have some basic familiarity with them. There are some switches that turn on and off. In this picture, I have a basic buck converter. 

And whenever you turn these switches on and off, you have this voltage here, which we call the switch node. As these switch turn off in a complementary fashion, that switch node voltage goes up and goes down, a high and low. And these edges, when the switches turn on and off, that voltage has a very high dv/dt. It's clear it's very high. 

Similarly what you have going on here is these pulses of currents, these slugs of current that occur when, let's say, the high-side switch here is turned on. That inductor current is flowing into the high-side switch. And it has high di/dt edges on these pulses of current. So that, as we'll see later, also causes some problems. 

But one thing that I'll just start off with here, when you have these pulses of current that are being pulled by your power converter-- in this case a buck converter, you'll have some amount of ripple on your input capacitors. And this ripple voltage on here is on the order of, let's say. 50 millivolts. But if you convert that to dv microvolts, that's maybe 90-plus dv microvolts. And that's well beyond the spec for what we want to limit our ripple or our emissions to in that frequency range. 

So if this slide will advance-- give me a moment. Sorry. I'm just getting the blue spinning arrow. OK, there we go. Just as a background, what you have here is a time domain wave form. It's just the ripple on that input capacitor like I showed before. 

What we're really going to be looking at is the frequency domain representation of that time domain signal. You can use the Fourier Transform to do that. Any periodic signal you can express has the sum of sinusoidal waveforms. That's hopefully some stuff you remember learning back when you were in school. 

And the fundamental, which is usually the highest component, is in this case 400 kilohertz-- because that's what this converter was switching at. And that's what you see represented here. But then there's also higher order harmonics as you go up in frequency. 

And so if we look at these magnitude versus frequency plots during the rest of the presentation, what we're trying to do is make sure from a conducted EMI perspective that we keep that noise or the frequency spectrum that's measured lower than some limit. So if you are in an IT industrial setting, you'll use probably CISPR 32 or 22. 

And as you see it here, the Class B level is shown in red. And it's a relatively continuous line from 150 kilohertz up to about 30 megahertz. Now in automotive, CISPR 25 is a common EMI standard. And it goes also from 150 kilohertz, but in this case up to 108 megahertz. 

Another interesting thing about CISPR 25 is it has much lower limit levels, as you can see, 20-plus dv microvolts low. And so a lot of people consider the automotive EMI requirements more challenging. You might say, though, that there are some gaps in the frequency as opposed to this continuous line. But a lot of car OEMs will have other lines that fill in the gaps and maybe even add some different limit levels. But CISPR 25 is a common standard that we often use for testing. It's a good starting spot. 

Now, it's important to understand the difference between differential mode noise and common mode noise when we talk about EMI because you solve these different problems in different ways. So differential mode noise is what I like to think of as the noise that you'd expect. That the ripple on the input that we looked at earlier. 

And it's basically due to current that's flowing out on the plus wire and back on the minus wire. And it's relatively proportional or increases with load current. And you usually attenuate that with a differential mode pi filter. 

Common load noise, on the other hand, can be a little bit more tricky because parasitics come into play quite heavily there. So common mode noise is where you'll have the current of the signal flowing out on both the plus and minus wires of your supply. And that current will go through typically maybe some parasitic capacitance to the chassis ground in returning to the source. 

And this is mostly independent of load currents. And it is often filtered with a common mode choke. But it's even better if you can just avoid it in the first place and avoid the common mode choke if you can. 

So let's quickly talk about differential mode filter. What we're trying to do here is filter out the pulses of current that are being pulled by your switching converter. Let's say it's a buck converter in this case. And here's the input ripple on that buck converter. And what we want to do is filter it out so we have a nice, clean signal on the other side. 

The pi filter basically operates like a second-order low-pass filter where you have an inductor and a capacitor. And this CD and RD that's shown here, those are just there to damp the output impedance of this filter. You want to make sure that it doesn't peak above the input impedance of your power converter for stability purposes. 

But Basically what you do is you'll pick L and your C values such that you get the desired attenuation at the switching frequency. So here's a -10 dB per decade downslope. And you'd pick your corner frequency here based on your L and C such as to get the attenuation that you want at this fundamental switching frequency that we see here. Because the 45 millivolts peak to peak is just way too high, and it's not going to pass the requirements. 

Now, life would be great and everything would be peachy and easy if there weren't parasitics. So anytime you have a capacitor and inductor, there's some parasitic elements to it. You have equivalent series inductance. You have equivalent series resistance, capacitance, et cetera. 

And these parasitics will degrade the performance of your pi filter, especially at higher frequencies. You can see it here in this part on the right. So the dotted gray curve here is what ideally you'd want with that nice -40 dB per decade downslope. But a real converter is going to have additional resonances that show up and really degrade the performance of the attenuation of that filter at the higher frequencies. 

So sometimes what you'll see people is they'll have maybe a two-stage pi filter with maybe a ferrite bead or something like that. That will help lower that gain or improve the attenuation of the filter, especially at higher frequencies. 

Common mode noise is often caused by the switch node. We often see the switch node parasitically capacitively coupling to the chassis or other parts of the system. And that's one of the main sources of common node noise. 

Now, you can see here on the right a really simplified diagram of what's happening here. Your switch node is going up and it's going down. When the high-side switch is turned on, it goes up. When the low-side switch is turned on, it goes down. 

And this parasitic capacitance is at the same time kind of charged and discharged. And so you have this parasitic current that's flowing through this parasitic capacitance and around through the chassis ground and through the wiring harness. And this capacitance is somewhat sensitive or dependent on the height that this system or this equipment is relative to the chassis ground. 

It's kind of like two plates of a capacitor-- what's that distance of separation? That's one of the reasons why CISPR 25, in the standard, it sets that the height that this has to be for testing to be 5 centimeters. If you lowered it or raised it, it would adjust that parasitic capacitance. 

This here is an example test setup for conducted EMI testing using CISPR 25. And you can see that low permativity support that I was just talking about, that 5 centimeters. This is your equipment under test, whatever board it is that you're testing. You have some node resistors there to mimic the load. 

But all of this is sitting on this big copper tabletop that's tied to the back wall and grounded. You have your battery or some low power, low noise power source, I should say. And these artificial networks here are what you are passing your power source through to get to your equipment under test. 

It's also where you're measuring the measurement ports for measuring the spectrum of your conducted EMI. That's what you'd attach your spectrum analyzer or EMI test receiver to. And that's the basic setup there for conducted EMI testing for CISPR 25. 

Also since we're going to be talking a little bit about this in this talk, CISPR 32/22, it's a similar test setup for a conducted EMI, where you'll have some equipment under test. The difference here is a LISN is usually placed on the ground and you have some horizontal and vertical ground plane. It's not necessarily in a screened room. But you can have it in a chamber of some sort. 

And you similarly use, just like the artificial network before, there's a line impedance stabilization network, or LISN, that you use for measuring the emissions from. And this setup is typically used for equipment that's connected into the power grid that you plug into the wall. And we have a few more of those pieces of equipment these days in automotive. But some of the same principles that our team uses to solve EMI challenges for this equipment apply directly to other automotive systems as well. And so I'm going to hand it off to Brian King now to talk us through some of the techniques that he's developed for debugging EMI problems. 

OK. Thank you, Pradeep. So let's go ahead and dive right in. So this first set of slides here we're going to look at, deal with separating out the common mode and differential mode of noise. So when you first test your system, you're going to get some EMI plot. And chances are it's not going to pass. 

So this particular example, like Pradeep was saying, is from the industrial side of things and is with regards to CISPR 32. But again, all these concepts that I'm going to talk about here can be applied equally to CISPR 25 as well. But in this particular case, we built our prototype, tested it for the first time, and we failed miserably. 

So what are you going to do next? From looking at this, there's no way to tell whether it's differential mode or common mode. So you don't know where your problem is and how to tackle that. Next slide, Pradeep, please. 

So luckily there are some very cheap tools that you can get that can help you to separate out the common mode noise from the differential mode noise. So these little splitters and combiners, which you can see in the middle picture of this slide, they have three terminals on them. And so there's two different units that you can get from minicircuits.com. They're about $60 each. 

And one you can use to actually look at only the differential mode signal. And the other one you can use for looking at only the common mode signal. So you can separate out your noise into those two different components and then tackle them individually, which will make your debug time much, much quicker. And so the way it works is you need two LISN's, one for each line-- the positive and negative in your automotive systems. 

One line would go into pin number one. And the other one would go into terminal number two, if you were looking at that center picture there. And then the S-terminal in the middle goes out to your EMI analyzer. And so depending on which one you're looking at, you'll see just the differential mode or just the common mode. 

So if you want to know more about what's actually inside those splitters and combiners, there's a really good paper that's written on this, that's given at the bottom of the screen there, that you can read about. It goes through the math about the filters that's in there and how it works and evaluates different units from minicircuits.com. But these two ones here are good from a pretty low frequency up to about 60 megahertz. So they're pretty useful for debugging, not just industrial systems for CISPR 32, but also CISPR 25. 

And they have some other ones at a higher frequency for the automotive space as well. 

OK. 

Hey, Brian. This is Jared. One question came in on the chat. And I wanted to see if maybe this is the time to address it. So the question is, how do you estimate the magnitude of the flicker noise that is generated due to the switch to the power supply? And I think that's in reference to what Pradeep was showing earlier. But I don't know if this piece of equipment can do that or if you have an answer to that question. 

"Wicker noise," I'm not sure what that refers to. Do you know, Pradeep? 

Maybe that's the ripple. Maybe he means the ripple noise due to the switching of the power converter. There are some tools that you can use. I mean, TI has on its web bench online design tool, it'll help estimate the first maybe 10 or 20 harmonics of the switching ripple noise. We've done just basic circuit simulations as well. 

I think it's difficult to estimate some of the higher frequency noises because parasitics really come into play there. And it's hard to get accurate models of all the parasitic elements. But if you want to-- and I'll just quickly go back to that. 

What you can do if you wanted to simulate it is you can model your-- this is like a set of current pulses. And then just have the capacitor, and then you can-- I'd stay on your simulation software. You can just take an FFT of this node here, and an FFT on the other side of your filter and get an estimate of what that looks like. 

And for at least the first maybe 10 minutes or so, you might be able to get some reasonable information. But like I said, the challenge really comes down to the higher frequency stuff where the parasitics really come into play a lot. Hopefully that answered the question. 

I think that's good. I think you can do that with differential mode noise, or what we're referring to as flicker here. But it's almost impossible to do for common mode noise because you don't really know what you're parasitic elements are that are creating that common mode noise. So it's next possible-- I would say pretty close to impossible to simulate common mode noise ahead of time. 

I mean, I think that from our experience, you can simulate it. But the question is, how accurate is your simulation? 

Right. 

You can generate a model of it. You can estimate it. But then when you actually get to the bench and built the hardware, does it match or not? 

My money would be on it doesn't. 

Yeah. OK. Good. Good question. 

I saw one more question come in. And it was, how does the second stage pi filter corner frequency-- how is that decided? That one-- it's usually much higher. And sometimes it may not even be a second stage. 

In that example, it was a second stage. Sometimes they even leave out one of the capacitors, maybe this one here. And they'll just have their two inductive elements in series. So it could just be combined. 

Usually you're choosing your ferrite bead more for the higher frequency performance and the attenuation that you get, not so much based on the corner frequency. So usually what you do is you'd look up on the ferrite bead what sort of impedance it has at that higher frequency range. OK, for sake of time, I suggest we move on. 

Yeah. So we can go through these next few slides pretty quickly because it just shows using that spliter and combiner. So here's what you would see, from that original plot we looked at, if you actually split it up into the two different components. 

And you can see that at the lower frequencies, which is pretty typical, we have a big differential mode problem. At the higher frequencies, it's more common mode. So that helps you focus your efforts. OK? Next slide. 

So then you can go through and tackle each one individually here. We fixed the differential mode. We still have the common mode to fix. Next slide. And then we can tackle that one individually and get the common mode to pass. And then when we have both of them done, we can go back and look at the overall picture and see that we have total victory in our battle against EMI here and have won both battles to win the war; common mode and differential mode. OK. 

So here's a slide that just gives some basic tips for differential mode filtering. I do want to point out that a lot of this material is coming from our Power Supply Design Seminar topics. This year, for 2020, we have one on automotive EMI. 

And we have another topic on low-power AC to DC supplies and the EMI there. Those are coming up in the next couple of months. So if you haven't signed up for that, and you want to dive a lot deeper than what we're talking about today, you can go to TI.com/psds and sign up for those and get a lot more information. 

But on this slide, we are showing some tips for differential mode. So one thing that you have to be careful of is saturation of your input filter. So you may do a good job of simulating and picking the correct inductance for your inductor. But when you actually go to select the inductor you're going to use in your system, you need be careful that it doesn't actually saturate at your max load conditions. 

And this is a even bigger concern in AC to DC supplies. With the AC current, at the peaks of the sine waves, it can be a lot higher than the average current. So you need to be aware of that. 

The other thing is beware of where you're placing the filter on your board. Just because you have the right filtering component, if you do a bad job of your layout and get those components close to your switching waveforms, you could couple noise straight to those components. And so even though they're filtering stuff, they may be picking stuff up at the same time. 

And the other thing that Pradeep hit it on earlier, which is key for all EMI problems, is parasitics. They are usually the cause of most EMI problems. And they can also limit the effectiveness of your filtering components as well. So an inductor is not an inductor at high frequencies. 

At some point, it becomes a capacitor to the parasitic capacitance. And the capacitor is not a capacitor at high frequencies. At some point, the ESL makes it look like an inductor at really high frequencies. So be aware of the parasitics of your components, especially when you're trying to model your filtering. All right. Next slide. 

So common mode noise-- how is common mode noise generated? I think Pradeep talked about this a little bit earlier too. Common mode noise is those stray currents that find their way to earth ground or in our automotive systems' chassis ground. 

So the currents come in down our input lines. But they lose their way and stray off the course and somehow have to find a way back home to their source. And that ends up being through earth ground or chassis ground, which is what we don't want. So that's the common mode currents. 

When we have isolated power supplies, it becomes an even bigger problem, whether that's AC to DC isolated power supplies or even in some of the automotive systems where we use those isolated power supplies. That's because inside the transformers, if you think about it, these magnetic components have windings in them. And they have high voltage on them switching at high frequencies in close proximity to each other. 

And so they form a capacitive plate across our isolation boundary, and inject a lot of current across that isolation boundary. If we have a floating load or high impedance to earth or chassis, those currents have a difficult time finding their way back, and eventually show up as common mode noise. 

And this is really common if you have a bias supply for a traction inverter or something like that. You'll often want to make this parasitic capacitance as small as you can to reduce that come mode noise. 

So there's some techniques you can use. And we go into depth in this in the seminar series. But it really comes down to the construction of your transformer. And what we're trying to show here is that if you pay attention to the construction of your transformer, there are techniques you can use to actually put new windings inside your transformer to cancel out the common mode currents from the primary secondary windings that are already there. 

So you if you can inject a current that's equal in magnitude but opposite in phase to the common mode currents that your other windings are generating, you can essentially null them out and greatly reduce your common mode currents. And again, we can't go into all the details here. But check out that seminar topic if you want to see more details about that. 

So once you've constructed your transformer with those cancellation windings in there. it's difficult to get it exactly, perfectly canceled out or nulled out. So what you'll end up having to do is play around with the number of turns on your cancellation winding to make sure that you get the common mode currents minimized. 

And this slide just shows some techniques for actually evaluating the transformer design by itself, outside of the system, to make sure that the transformer is common mode balanced. And basically what we're doing is you inject a sine wave using a function generic. It could be a low voltage sine wave across the primary winding of your transformer. 

And then you look with a scope probe at how much voltage is induced between the primary windings and the secondary windings. And so you look not just at the magnitude of the signal that you're picking up, which in this top plot is the signal that we're injecting-- a 1 volt peak to peak. And then the purple plot on the bottom is what we're measuring between, the primary and the secondary. 

And that's our common mode noise we're measuring. And we see it's about 98 millivolts peak to peak. So our job is to try to get that down as close to zero as we can by changing or playing with those number of turns on our cancellation watt. 

So you can see here we started with 98 millivolts peak to peak. If we add two cancellation turns to our transformer, we get that down to 26 millivolts. So that helps to reduce it. And then you notice if we add one more turn and go from two cancellation turns to three cancellation turns, we're at 21 millivolts. 

But notice what else happened. If you look closely, you'll see that the blue waveforms on the bottom are actually now our of phase when we have three extra cancellation turns. The phase slips. So that means we added too many cancellation turns. And now we're injecting excess common mode current. 

And so if we keep adding more turns it's going to get worse. So somewhere between two and three it's perfectly canceled. In this case, we would pick those extra three turns because that's a little bit closer to being canceled. All right. 

There are also techniques you can use to measure, to actually look at the common mode noise signal in the time domain, as opposed to measuring it in the frequency domain, which is what we're doing with our network analyzers when we're measuring the EMI. So to use this technique, basically what you want to do is you want to disconnect any Y-capacitors you may have between your input or output. 

So basically, leave the output of your isolated supply floating. And then on the wires that come from between your power supply to your load, you want to wrap a piece of bus wire, which is what you can see in the picture in the top right there. The black and the red wires are going between the power supply and the load. 

And that's just a floating piece of bus wire wrapped around both wires equally. And so that forms like a capacitor that picks up the common load field and couples it to that piece of bus wire. And then with our scope, we look at what we can pick up in the time domain between chassis ground and that bus wire. 

And what we will see when we look at that is the actual common mode signal in the time domain. And so it can really give you some insight as to what's going on, as opposed to just looking at the frequency domain. Next slide. And next. So these pictures on the right will show you what you're going to see. 

So in this case it was an AC to DC power supply. But it could have been a DC supply for automotive. And what you'll see is you'll pick up what looks like switch node waveform. And so the larger that signal is, then the more common mode problem you have. So similar to the way we optimized the transformer by itself, you can optimize it in the system this way by trying to minimize that signal that you're picking up with your scope probe. 

So you can see the signal's fairly large there in the top right. And then after doing various techniques through balancing our transformer and a few other techniques, this particular example you can see a more better balanced, lower common mode noise result on the bottom right. 

And so what's typical too is that it's easier to cancel out, minimize those lower frequency components. But you'll notice in the bottom picture, there's still some high frequency spikes, which is a lot easier to deal with. It requires smaller filter components to filter that out and to get it to pass. OK? 

OK. Thanks, Brian. 

Sure. 

Now I'm going to hand it off to Bob to walk us through a design example showing a path to passing CISPR 25. Bob, you want to take it away? 

You're on mute, Bob. 

Yeah, you're on mute, Bob. 

Can you hear me now? 

Yes, we can hear you now, great. 

OK, great. I was just unmuting my phone. So I worked on a couple of designs for automotive. And the first design, which was PMP21417 did not pass. We built the design. We sent it to the customer. And even though our internal conducted emissions tests looked good, the customer reported that we were failing the radiated emissions. 

So we had an opportunity to go back and redesign that. So we generated a new design and a new layout. I modified the previous design and layout and redesigned two of the rails. So our internal conducted emissions did look better. And the customer reported that we were passing the radiated emissions. So that was great. So let's move to the next slide here. 

Let's take a look at the architecture, what we had to start with and what we did to get some improved performance. So here we can see the architecture where we have our 12-volt battery coming in with a reverse protection circuit with a MOSFET, and then EMI filter. And for the 21417 we had an LMS36355, with just a synchronous buck controller, running at 400 kilohertz. And that produced a 5-volt 1-amp output. 

And from the 5-volt, we used a dual converter 2amp buck running at 2.2 megahertz, which use a fixed frequency part. And that gave us outputs of 3.3 and 1.2 volt at 1/2 an amp. For the improved design, we made some changes to that. And the two main things we did was we included some better filtering and some higher frequency switching. 

So we went to 2.1 megahertz. for the 5-volt converter. And then we used to single TPS62067, which were 3 megahertz 2-amp buck inverters. The key difference here being that they are spread spectrum parts. So with frequency dithering, we should and did get some better results. 

Let's take a look at the next slide here. So in the reverse protection and filtering section, what we see is this is the front end now where we're coming in from the battery on the left. We've got the reverse protection in the middle and then the EMI filter on the right. 

And basically it's a two-stage differential-type filter where we've got just a discrete inductor 4.7 microhenry backup capacitors and then a ferrite bead. What we did to make some improvements here is we removed the ferrite bead on the converter side of the reverse battery protection. And we redistributed the filter capacitors to get some better split. 

And the reverse battery protection, actually the MOSFET is resistive. So that already gives us some attenuation right there. And then we added a common mode inductor to the front end. And we'll have a chance to look at the final design, both with and without the common mode inductor, to see what contribution that played. 

So for our 5-volt supply, what we see here is the original design was 400 kilohertz. We did change to 2.1 megahertz. Now, both are spread spectrum. But we got a lot better performance from the higher frequency switch on the final design. 

So moving on to the next slide, the 3.3-volt and 1.2-volt supplies were originally in one chip with a 2.2 megahertz fixed frequency. We went to two separate converters. Both of them were the same part 3 megahertz spread spectrum. And that made a significant difference in the overall performance. So next slide, please. 

Taking a look at the printed circuit board layout-- and this is where a number of improvements were made. So in the left side we see the original design where the input is on the same edge of the board as the outputs. And so there is opportunity there for some noise to couple through. 

We moved that input and the filter to the opposite edge of the board. And we also did a better job of isolating the filter from the surrounding circuit to make sure that there's no noise getting in through other copper traces, internal layers, or whatever-- back through that input filter. 

And then a couple of things that we did with the new part, we were able to place ceramic capacitors on both sides of the converter to make sure that we're splitting those currents and providing the lowest, shortest path to those input decoupling capacitors. Next slide, please. 

So going through the circuit board layout-- and this is a four-layer board. This is showing a mid-layer one, where for the input common mode filter, we kept the area under the filter clear. And we did that on each layer of the board. We also removed copper under the 5-volt inductor to keep it from coupling back into any adjacent circuits. Next slide. 

Mid-layer two, we see the same type of strategy; pretty much a copy of mid-layer one for those two critical areas where the common mode filter is and the 5-volt inductor there. And then looking at the bottom layer, again, we've kept that common mode filter. And of course that really does make a big difference in improving the overall performance of the circuit. 

So here we've got a comparison of the conducted emissions at low frequency. On the left you can see a pretty big fundamental from our 400 kilohertz switcher for the 5-volt. And then we can also see that 2.2 megahertz fundamental for the dual output 33 and 1.2. 

Moving to higher frequency switching with spread spectrum, we can see that the peaking and the harmonics are greatly reduced. The fundamental of 2.1 megahertz and 3 megahertz is shown. But by the time we get to the end of the frequency on the low frequency spectrum, most of those spikes are gone. Go ahead. 

Real quick, we got a question in about the layout. The question was what was the logic for removing the 5-volt power plane from the third layer in the second layout. So I guess it's the third layer. What was the logic for removing the 5-volt power plane from the third layer? 

Because we did it on one internal layer, we did it on the second. And the capacitance is reduced by removing that copper. So basically if we're looking at what is going to couple from the switch node into the ground plane-- and that's a common mode issue. By removing that, we're preventing it from coupling. 

And we do have a solid ground plane on the bottom of the board. But the capacitance is reduced from the top. So it hopefully made some improvements. We didn't do incremental spins to really measure the difference from one to the other. But that was the thinking. 

You're kind of separating the distance between the two planes of a capacitor I think. 

Yes. 

Hey, Bob. I had one more question come in on the chat. And it's related to measurement. It's sort of a high-level question. Can you use, I think it's a high frequency current probe, over both wires to measure the common mode turns? Is that a technique you could use? 

Could you say that once more? 

I don't know that you can measure-- go ahead, Pradeep. You wanted to restate that? 

Yes. 

Was that related to the technique that Brian talked about or what Bob was talking about? 

This was back when Brian was talking about the technique. [AUDIO OUT] 

How about we circle back to that once we get done with this one? 

OK. That sounds good. 

If that's OK? 

Sure. 

Yeah. 

And the last question that you got Bob was about the cutout. Not under the inductor, but the plane attached to the 5-volt output was the comment here. You see this plane on the 5-volt here on the original design? 

Oh. 

And you don't have that plane here. Was there any reasoning for that? 

No. The original design was copied from some other stuff. And I don't recall exactly. It just seemed like it was going places where it didn't needed to go and just seemed like it was spread out. So we ended up just simplifying it. I don't know that it made any difference with EMI or not. 

It seemed extraneous. And we wanted to maintain better ground plane throughout. So instead of separating that, we just-- we didn't need the copper for conducting the current. So we just kept the 5-volt connection as short and direct as possible and maximized the area for ground plane. 

OK. 

So we were looking at the low frequency. And I was about to go on to the next slide to look at conducted emissions on high frequency. And really, as you can see from the original design, almost all the harmonics are gone on the final design. The peaking is almost not evident. 

And one of the things that we see here, and I made a note here, is that on the right we see that the line is flat. And on the left, we see that it's kind of headed back up. And the common mode inductor really did reduce that upward trend of the noise floor. So next slide, please. 

So here's a comparison of the final design showing without and with the common mode inductor. So we can see that peaking and harmonics are reduced above the 10 megahertz on the low frequency plot. And then let's look at the high frequency plot next. As I mentioned previously, that upward trend of the noise floor is also reduced on the high frequency plot. 

So it's possible that the design could work without the common mode inductor. But we were not able to actually de-radiate it with our internal setup. So we left that to the customer. And the final design with common mode inductor did pass. So next slide, please. 

I think this one's just the overall conclusion slide for the presentation. We know that meeting EMC requirements can be really challenging. And hopefully in this presentation you got a greater understanding of what are some of the sources and causes of EMI. 

And hopefully you also walked away with some new techniques and tricks to identify differential mode noise and common mode noise and how to help solve it. And Bob just walked us through one example. I know there was still some more questions. And we're happy to stay later to answer some additional questions that you may have. Feel free to enter them into the chat. 

This might be a good time to go back to the question that was posed earlier. Can we use a high frequency current probe over both wires to measure common mode currents? I'm guessing that was in reference to what Brian showed and talked to earlier. So I'll go back to his side. And maybe he can comment on that. 

So, yeah, you can definitely do that if you have one. But for those of us that don't have a high frequency current probe, this is just an alternative way of doing that. But yeah, that's a good point. But if you clamp across both wires, you could see the common mode currents with a current program. 

You can make yourself a little sniffer probe by just winding some wire around an really tiny open know U-core where the ends of the core are open. And just terminate it with a 1-K and resistor. And maybe use a probe socket to slip it over the end of your scope probe. And then you can take that little sniffer and go around. 

The problem is usually, on the circuit board, your traces are not necessarily like the two wires shown in this picture. So trying to pick up common mode signals with a sniffer probe is difficult. You can find currents in any one wire. But trying to pick up the common mode might be difficult in that respect. 

Are there more questions? If you do have any more, please enter them into the chat. We'd be happy to talk to them. I'm not seeing anymore on my chat. Jared, do you see any more questions coming into the chat on your side? 

I got one that just came in. It said, explain the cancellation winding versus shielding winding in the isolation transformer. Repeat the procedure. 

Yeah, there is a difference between just simple shielding and cancellation winding. So a lot of people, when they are constructing their transformer and they try to put some sort of mitigation technique in there, they'll just simply put a shield. 

So a lot of times that may be just a sheet of copper, like a single turn of copper foil between the primary and secondary winding. And then tie that back to primary ground so that basically you're forming capacitor between your primary windings and that shield winding, and shunting all the current to that shield winding and then stretching that back to primary ground. That's a shield. 

What we're doing different with the cancellation winding is that this is not a single turn of foil. This is multiple turns of stranded wire. And we're calculating out how many turns we want to put in there based on the average number of turns between the primary and secondary windings and the voltages between them, so that we inject a current that's opposite the currents that are already in the transformer. 

So we're basically nulling out the currents that are there, as opposed to trying to shunt them to ground. Because the problem with just doing a shield is that you'll never get it perfectly balanced. Because you may shunt with the shield. You'll shunt the currents from the primary windings. 

But you're still going to have currents from the secondary windings injecting current to your shield and coming across your isolation boundary from secondary to primary. So you're still going to have common mode currents with the shield. Whereas with the cancellation winding, we can try to null out the currents to get as close to zero as we can get them. Hopefully that makes sense. 

Yeah, this presentation doesn't have all of the waveforms that I think explain it in detail. I'd really encourage the person who asked the question to go check out the 2020 Power Supply Design seminar topic. This presentation here goes into much greater detail that walks you through the waveforms. It helps explain how this works in more detail. 

Yeah. 

But that's a great question. It took me a little while to figure it out as well. Other questions? 

No more on the chat. But we'll just give it maybe 30 seconds in somebody's currently writing a question. 

OK. 

While we're waiting, I'm going to do a magic trick. I going to have an invisible cup come in. It's going to disappear. OK. That was your magic trick for today, everyone. 

Nice. All right we got another question. This one says, what is the cancellation winding connected to? How is the current injecting? 

Yeah, that's a good question. So the cancellation winding-- one end of the winding will be tied to an AC quiet node. So that's either primary ground or it could be the positive terminal of your input capacitor. So something that doesn't have any switching waveform on it at all. So it's a DC quiet node on your primary side. 

The other end of the cancellation winding is actually going to be buried inside your transformer. It doesn't come out of your transformer. So basically what it is is just a capacitor that has a certain amount of average voltage across it. So it forms a capacitance across the isolation boundary. 

And so by controlling the voltage in the capacitance, we control how much current is inducted across that boundary. And that's the whole point of it. So if you're looking at this diagram that you see here, the green windings are the primary windings. The blue is the secondary. And then that purple winding is our common mode balance. 

So if you can imagine, if we didn't have that purple winding there, there'd be two interfaces between primary and secondary. There's an interface between the lower, or inner half primary and the secondary. And then there's another interface between the outer half primary and the secondary. 

And so there's some average voltage built up across each one of those windings, both half primaries and both in the secondary. And there's some capacitance based on that surface area and distance between those windings. And it's injecting currents across those boundaries. 

So what we're doing is when we put that purple winding in there, we're actually doing a couple of things. So the first thing we're doing is we created a shield. Basically we put a barrier between that secondary and the inner primary, which is at the bottom. So there's no capacitance between those anymore. We essentially blocked that off. 

And so what you really want to do is you want put your highest voltage half primary on the bottom. Because we want to block that altogether. Now, the next thing we're going to do is that we're going to calculate how many turns we want on that purple winding so that the average voltage across that winding is equal to the average voltage between the outer half primary and the secondary-- so that we can cancel out those currents and get the currents to be the same. 

And then we flip the phase of that winding too. Flip the dots if you would like, so that the current's opposite in magnitude. So hopefully that explains it in a little more detail. I know it's complicated, which is why I didn't try to pack all the information in this presentation. Because we'd need probably closer to hour and a half than 45 minutes to do the presentation. 

Sure. We have another question that came in. Any power efficiency loss in using the cancellation? 

Yeah. So that's the thing about engineering right? I mean, nothing comes for free. Yeah, there are a couple effects that we are going to have from putting that in there. So one thing is obviously it's going to have higher leakage because we just put more space between our primary and secondary windings. 

The other thing that's going to happen is that we're going to have more total parasitic capacitance on the switch node. And every time we add copper in there, we are increasing total-- even though we're nuling out the common mode currents across isolation boundary, there's still turn to turn capacitance every time we add a turn in there. And so we're going to increase switching losses because of that. 

Another thing is if we're switching at a really high frequency, you have to be worried about proximity effect, circulating currents. Even though there's no net current in that winding, because it's just an open winding, there's still proximity effect and still circulating eddy currents in there that can create losses. 

So, yeah, it's going to hurt your efficiency a little bit. But the trade-off is that now we can use a much, much smaller common mode choke outside of the system, or hopefully not have to use one at all if you're so lucky. But regardless, you should be able get by with a much smaller one, which may end up saving you efficiency in the long run because it will have less resistance. 

Or maybe it's going to make your product smaller. But the trade-off is that higher leakage, higher total switch node capacitance, and the possibility of proximity effects. But in the end, you usually come out ahead-- usually. 

OK. And then we had another Question. Come in here. Can a transformer common mode balance coil be beneficial also in a push-pull isolated DC? Such as using the SN6505 driver? 

Yeah. I mean, it applies to any topology. How you do your wire stack-up might be a little bit different in a push-pull than a flyback obviously. Because the number of windings is different. But you can put a cancellation winding into any transformer to try to cancel out those currents. 

But it starts with understanding your layer stack-up and understanding what parasitic capacitance you've got between the windings that are adjacent across the isolation boundary. So you have to identify where your isolation boundary is, how many you have within that stack up. Figure out what the average number turns is on either side of that boundary, and then insert your cancellation winding in there to cancel that out. 

OK. Very good. I think we're at time here. 

OK. 

But I want to thank Pradeep, Brian, and Bob for the time today. And thank you everybody who attended outside of TI for joining as well. All the session recordings and presentation will be available to view and download next week. 

And in the meantime, you'll be receiving an email with links and a post-event survey. If you get a chance to fill out that survey, we really appreciate the feedback and want to continue to improve our content for future tech days and other training events. So if you could fill that out, that would be very much appreciated. But other than that, thank you for attending and everybody have a great rest of your day. 

Thanks, Jared. 

Thank you. 

Thank you. 

Thanks, everybody. 

You bet. Bye-bye.