WEBVTT 00:00:00.000 --> 00:00:02.460 align:middle line:90% [SQUEAKING] 00:00:02.460 --> 00:00:04.920 align:middle line:90% [RUSTLING] 00:00:04.920 --> 00:00:06.396 align:middle line:90% [CLICKING] 00:00:06.396 --> 00:00:09.650 align:middle line:90% 00:00:09.650 --> 00:00:11.990 align:middle line:84% DAVID PERREAULT: OK, why don't we get started? 00:00:11.990 --> 00:00:15.950 align:middle line:84% So what I wanted to do in the next few lectures 00:00:15.950 --> 00:00:18.050 align:middle line:90% is talk about filtering. 00:00:18.050 --> 00:00:22.160 align:middle line:84% And this is ripple filtering or EMI, Electromagnetic 00:00:22.160 --> 00:00:24.020 align:middle line:90% Interference filtering. 00:00:24.020 --> 00:00:28.140 align:middle line:84% And why am I going to spend so much time talking about filters? 00:00:28.140 --> 00:00:32.090 align:middle line:84% And the reason is because in most power supplies, 00:00:32.090 --> 00:00:35.300 align:middle line:84% and in fact in a lot of power electronics circuits generally, 00:00:35.300 --> 00:00:38.830 align:middle line:84% the filters take up a large fraction of the overall volume 00:00:38.830 --> 00:00:39.330 align:middle line:90% and cost. 00:00:39.330 --> 00:00:42.170 align:middle line:84% So if you look at a typical offline power supply, 00:00:42.170 --> 00:00:47.520 align:middle line:84% 25% of the volume might be devoted to just filters 00:00:47.520 --> 00:00:49.880 align:middle line:84% because we're switching, we're transferring energy. 00:00:49.880 --> 00:00:52.920 align:middle line:84% And the ripple we generate, we really need to suppress. 00:00:52.920 --> 00:00:59.138 align:middle line:84% And we want to suppress it maybe to meet some application desire, 00:00:59.138 --> 00:01:00.680 align:middle line:84% but we might want to suppress it also 00:01:00.680 --> 00:01:02.630 align:middle line:84% because we're not allowed to sell something 00:01:02.630 --> 00:01:06.296 align:middle line:84% that you plug into the wall that has too much ripple. 00:01:06.296 --> 00:01:10.080 align:middle line:84% So if we think about it, there's kind 00:01:10.080 --> 00:01:13.620 align:middle line:84% of two kinds or two flavors of specifications 00:01:13.620 --> 00:01:15.880 align:middle line:90% we often have to meet. 00:01:15.880 --> 00:01:17.625 align:middle line:84% One of them is in the time domain. 00:01:17.625 --> 00:01:20.700 align:middle line:90% 00:01:20.700 --> 00:01:23.370 align:middle line:84% If I'm developing a power supply, 00:01:23.370 --> 00:01:25.002 align:middle line:84% somebody might tell me, oh, look, 00:01:25.002 --> 00:01:26.460 align:middle line:84% I'm only going to allow you to have 00:01:26.460 --> 00:01:30.690 align:middle line:84% 20 millivolts, either peak or RMS, of output ripple 00:01:30.690 --> 00:01:34.690 align:middle line:84% because we want such and such a level of noise in our output. 00:01:34.690 --> 00:01:37.200 align:middle line:84% And that's one kind of specification 00:01:37.200 --> 00:01:39.870 align:middle line:90% you might need to meet. 00:01:39.870 --> 00:01:42.750 align:middle line:84% More typically, however, and almost universally, 00:01:42.750 --> 00:01:47.010 align:middle line:84% if it's something like you're talking about an input filter, 00:01:47.010 --> 00:01:50.040 align:middle line:84% you're going to plug it into the wall, 00:01:50.040 --> 00:01:52.160 align:middle line:90% is in the frequency domain. 00:01:52.160 --> 00:01:59.700 align:middle line:90% 00:01:59.700 --> 00:02:06.110 align:middle line:84% That is, how much energy can you inject in a given frequency 00:02:06.110 --> 00:02:06.610 align:middle line:90% bandwidth? 00:02:06.610 --> 00:02:10.050 align:middle line:84% So a typical thing was you might detect this with a 9-kilohertz 00:02:10.050 --> 00:02:12.820 align:middle line:84% resolution bandwidth for where you're picking up energy 00:02:12.820 --> 00:02:16.800 align:middle line:84% and look in the span, say, between for conducted emissions, 00:02:16.800 --> 00:02:20.310 align:middle line:84% look in the span between, like, 150 kilohertz and 30 megahertz 00:02:20.310 --> 00:02:24.420 align:middle line:84% is most commercial standards for conducted EMI. 00:02:24.420 --> 00:02:28.230 align:middle line:84% Above 30 megahertz, you typically use antennas and stuff 00:02:28.230 --> 00:02:28.840 align:middle line:90% to pick it up. 00:02:28.840 --> 00:02:31.858 align:middle line:84% And we're not going to talk about radiated electromagnetic 00:02:31.858 --> 00:02:32.400 align:middle line:90% interference. 00:02:32.400 --> 00:02:36.180 align:middle line:84% But in fact, if you don't pass conducted interference, 00:02:36.180 --> 00:02:39.170 align:middle line:84% you won't pass radiated either because basically when 00:02:39.170 --> 00:02:41.670 align:middle line:84% you conduct energy into your input leads, your output leads, 00:02:41.670 --> 00:02:44.320 align:middle line:84% that's an antenna, and it broadcasts off noise. 00:02:44.320 --> 00:02:47.382 align:middle line:84% So it's important to meet these things 00:02:47.382 --> 00:02:49.590 align:middle line:84% if you're going to sell commercial products that plug 00:02:49.590 --> 00:02:52.410 align:middle line:90% into a wall or into a vehicle. 00:02:52.410 --> 00:02:54.368 align:middle line:84% Into the wall, you've got to deal with the FCC. 00:02:54.368 --> 00:02:56.202 align:middle line:84% If you're going to plug it into the vehicle, 00:02:56.202 --> 00:02:58.320 align:middle line:84% you've got to deal with the vehicle manufacturer. 00:02:58.320 --> 00:03:01.710 align:middle line:84% And you're going to have to pass these specifications in order 00:03:01.710 --> 00:03:03.450 align:middle line:90% to sell something. 00:03:03.450 --> 00:03:04.810 align:middle line:90% And it's quite important. 00:03:04.810 --> 00:03:08.040 align:middle line:90% Let me just give you an example. 00:03:08.040 --> 00:03:11.280 align:middle line:84% Once upon a time, I consulted for-- 00:03:11.280 --> 00:03:14.040 align:middle line:84% I'd call it a garage startup, but really, they 00:03:14.040 --> 00:03:17.550 align:middle line:84% were in the guy's basement, so it was a basement startup. 00:03:17.550 --> 00:03:19.840 align:middle line:84% And what they were developing-- this was early days. 00:03:19.840 --> 00:03:22.060 align:middle line:84% They were developing a solar inverter to sell. 00:03:22.060 --> 00:03:26.100 align:middle line:84% And so they were developing the power stage 00:03:26.100 --> 00:03:30.325 align:middle line:84% to convert energy into the grid and getting the efficiency up 00:03:30.325 --> 00:03:31.450 align:middle line:90% and doing all these things. 00:03:31.450 --> 00:03:36.690 align:middle line:84% But they hadn't gotten around to designing the EMI filter yet. 00:03:36.690 --> 00:03:39.690 align:middle line:84% So the guy's neighbor came up to him and said, 00:03:39.690 --> 00:03:41.430 align:middle line:90% are you having trouble? 00:03:41.430 --> 00:03:43.690 align:middle line:84% Around every day around 3 o'clock, 00:03:43.690 --> 00:03:46.650 align:middle line:84% I can't watch television because, I don't know, 00:03:46.650 --> 00:03:48.188 align:middle line:84% everything just fuzzes out on me. 00:03:48.188 --> 00:03:50.730 align:middle line:84% And what the guy in the startup didn't tell him was 3 o'clock 00:03:50.730 --> 00:03:52.570 align:middle line:84% was when he started testing every day. 00:03:52.570 --> 00:03:55.230 align:middle line:84% He was dumping too much ripple into the grid 00:03:55.230 --> 00:03:57.760 align:middle line:84% and the next house over suddenly couldn't watch television. 00:03:57.760 --> 00:04:00.750 align:middle line:90% So these are real problems. 00:04:00.750 --> 00:04:03.540 align:middle line:84% It's not just you got to pass a spec. 00:04:03.540 --> 00:04:05.620 align:middle line:84% It's you will interfere with things. 00:04:05.620 --> 00:04:08.490 align:middle line:84% If you think about it in a car, if you're plugging something 00:04:08.490 --> 00:04:14.566 align:middle line:84% into the USB outlet or into the cigarette lighter or something, 00:04:14.566 --> 00:04:17.140 align:middle line:84% any ripple you conduct is running around 00:04:17.140 --> 00:04:18.160 align:middle line:90% on the wires in the car. 00:04:18.160 --> 00:04:20.570 align:middle line:90% And it's right next to a radio. 00:04:20.570 --> 00:04:23.830 align:middle line:84% So your radio is a very sensitive detector 00:04:23.830 --> 00:04:26.470 align:middle line:84% of very, very small quantities of noise. 00:04:26.470 --> 00:04:28.750 align:middle line:84% And you will mess with that if you do not 00:04:28.750 --> 00:04:30.630 align:middle line:90% have very, very good filtering. 00:04:30.630 --> 00:04:33.930 align:middle line:84% So what we're going to find is that you've got to do filtering, 00:04:33.930 --> 00:04:36.430 align:middle line:84% but it's not just, oh, I need to make my ripple a little bit 00:04:36.430 --> 00:04:36.930 align:middle line:90% small. 00:04:36.930 --> 00:04:40.270 align:middle line:84% You've got to make it really small in radio frequency bands 00:04:40.270 --> 00:04:43.630 align:middle line:84% in order to not interfere with things. 00:04:43.630 --> 00:04:47.300 align:middle line:84% So we're going to talk about conducted emissions. 00:04:47.300 --> 00:04:49.030 align:middle line:84% And what I'm going to mostly focus-- 00:04:49.030 --> 00:04:52.780 align:middle line:84% most of what I'm going to say about input filters-- 00:04:52.780 --> 00:04:55.000 align:middle line:84% like, if I'm plugging my power converter in the wall, 00:04:55.000 --> 00:04:58.700 align:middle line:84% it's between the power stage and my power converter in the wall. 00:04:58.700 --> 00:05:00.560 align:middle line:90% So that's the input side filter. 00:05:00.560 --> 00:05:04.600 align:middle line:84% Most of it applies to output side filters too. 00:05:04.600 --> 00:05:07.670 align:middle line:84% But we'll focus on input filters. 00:05:07.670 --> 00:05:13.250 align:middle line:84% So let's just think, to get a sense of the challenges 00:05:13.250 --> 00:05:15.710 align:middle line:84% or issues, I want to start by talking about, 00:05:15.710 --> 00:05:18.860 align:middle line:84% how do you measure these things and what drives that? 00:05:18.860 --> 00:05:21.840 align:middle line:84% So let's just think about a very simple buck converter. 00:05:21.840 --> 00:05:26.150 align:middle line:84% So suppose I go off and build a buck converter. 00:05:26.150 --> 00:05:29.210 align:middle line:90% So here's my buck converter. 00:05:29.210 --> 00:05:31.920 align:middle line:84% And I'm not going to worry about the output filter. 00:05:31.920 --> 00:05:34.350 align:middle line:84% Let's just assume he's connected to a resistor. 00:05:34.350 --> 00:05:37.550 align:middle line:84% But here, I have the input filter. 00:05:37.550 --> 00:05:40.872 align:middle line:84% And maybe I can consider the capacitor for the buck converter 00:05:40.872 --> 00:05:42.330 align:middle line:84% the first part of the input filter. 00:05:42.330 --> 00:05:46.130 align:middle line:84% And here I have I'm going to draw some current ix. 00:05:46.130 --> 00:05:49.460 align:middle line:84% I'm going to put some input filter here. 00:05:49.460 --> 00:05:51.680 align:middle line:84% But for the moment, let's just assume my capacitor 00:05:51.680 --> 00:05:54.320 align:middle line:90% is the input filter I've got. 00:05:54.320 --> 00:05:57.526 align:middle line:84% So this thing right here is right now my filter. 00:05:57.526 --> 00:05:59.693 align:middle line:84% And now I'm going to connect this up to some source. 00:05:59.693 --> 00:06:03.410 align:middle line:90% 00:06:03.410 --> 00:06:07.280 align:middle line:84% And maybe this source is the battery in my car as vs. 00:06:07.280 --> 00:06:10.160 align:middle line:84% And then there's some source impedance, zs. 00:06:10.160 --> 00:06:13.580 align:middle line:84% And that may be-- at DC, that might be very small. 00:06:13.580 --> 00:06:15.660 align:middle line:90% But at AC, maybe it's not. 00:06:15.660 --> 00:06:19.535 align:middle line:84% So there's some source impedance that I'm driving in. 00:06:19.535 --> 00:06:22.250 align:middle line:90% 00:06:22.250 --> 00:06:24.176 align:middle line:90% And here's the challenge. 00:06:24.176 --> 00:06:27.080 align:middle line:90% 00:06:27.080 --> 00:06:32.930 align:middle line:84% What would be a model for this as far as conducting noise 00:06:32.930 --> 00:06:33.720 align:middle line:90% into the input? 00:06:33.720 --> 00:06:35.780 align:middle line:84% And maybe I would consider my noise 00:06:35.780 --> 00:06:38.750 align:middle line:90% to be whatever ripple here is-- 00:06:38.750 --> 00:06:40.470 align:middle line:90% I'll call this iy. 00:06:40.470 --> 00:06:45.770 align:middle line:84% So if I'm driving ripple current back into this input, 00:06:45.770 --> 00:06:47.720 align:middle line:84% then I'm generating noise in the input. 00:06:47.720 --> 00:06:50.870 align:middle line:84% And maybe that's what's interfering with my radio. 00:06:50.870 --> 00:06:52.770 align:middle line:90% So how would I model this? 00:06:52.770 --> 00:06:54.560 align:middle line:84% Maybe the easiest thing for me to do 00:06:54.560 --> 00:07:00.568 align:middle line:84% is say let me pretend this is a constant current inductor. 00:07:00.568 --> 00:07:02.610 align:middle line:84% I'm not going to worry about the inductor ripple. 00:07:02.610 --> 00:07:06.060 align:middle line:84% And a simplified model for this would be something like this. 00:07:06.060 --> 00:07:14.800 align:middle line:84% I'd have some current ix, which is some square wave. 00:07:14.800 --> 00:07:19.980 align:middle line:84% And then maybe I would only care about ix tilde, 00:07:19.980 --> 00:07:22.532 align:middle line:90% the AC component of that. 00:07:22.532 --> 00:07:24.420 align:middle line:84% So maybe I'll just worry about the AC 00:07:24.420 --> 00:07:27.120 align:middle line:90% component of this current. 00:07:27.120 --> 00:07:28.800 align:middle line:84% It's coming into the filter capacitor 00:07:28.800 --> 00:07:30.760 align:middle line:90% and into the source impedance. 00:07:30.760 --> 00:07:34.050 align:middle line:90% 00:07:34.050 --> 00:07:40.260 align:middle line:90% And maybe I will worry about iy. 00:07:40.260 --> 00:07:43.740 align:middle line:84% Or maybe I will worry about the voltage here, vx. 00:07:43.740 --> 00:07:46.700 align:middle line:90% 00:07:46.700 --> 00:07:52.246 align:middle line:84% I could specify either thing to worry about the filter. 00:07:52.246 --> 00:07:55.830 align:middle line:84% Well, let's just think about the effect of the source impedance 00:07:55.830 --> 00:07:58.165 align:middle line:90% on my measurements. 00:07:58.165 --> 00:08:08.010 align:middle line:84% If zs goes to 0, if I had a really good source, then 00:08:08.010 --> 00:08:08.850 align:middle line:90% what would happen? 00:08:08.850 --> 00:08:15.160 align:middle line:90% If zs goes to 0, iy goes to ix. 00:08:15.160 --> 00:08:17.730 align:middle line:84% If there's no source impedance, this capacitor does nothing, 00:08:17.730 --> 00:08:19.610 align:middle line:84% and all the ripple goes back into my input. 00:08:19.610 --> 00:08:23.262 align:middle line:90% 00:08:23.262 --> 00:08:29.495 align:middle line:84% Of course, if zs goes to 0, then v sub x tilde goes to 0. 00:08:29.495 --> 00:08:32.039 align:middle line:90% 00:08:32.039 --> 00:08:34.890 align:middle line:84% So if I had no source impedance, I'd get-- 00:08:34.890 --> 00:08:36.929 align:middle line:84% all of my input would see the ripple current 00:08:36.929 --> 00:08:40.289 align:middle line:84% from the converter, but I wouldn't have any voltage ripple 00:08:40.289 --> 00:08:42.510 align:middle line:90% at the input of my converter. 00:08:42.510 --> 00:08:46.320 align:middle line:84% On the other hand, if z sub s at ripple frequencies 00:08:46.320 --> 00:08:49.960 align:middle line:84% went to infinity, what would happen in that case? 00:08:49.960 --> 00:08:52.390 align:middle line:84% Well, then I'd have a really high impedance here. 00:08:52.390 --> 00:08:55.410 align:middle line:84% The ripple current that I'd measure would go to 0 00:08:55.410 --> 00:08:57.690 align:middle line:84% because this is just a very high source impedance, 00:08:57.690 --> 00:09:00.600 align:middle line:84% and I wouldn't-- all the current would go through the capacitor. 00:09:00.600 --> 00:09:04.620 align:middle line:90% And v sub x would go to-- 00:09:04.620 --> 00:09:09.030 align:middle line:84% it would go to i sub x divided by j omega c. 00:09:09.030 --> 00:09:11.720 align:middle line:90% 00:09:11.720 --> 00:09:13.460 align:middle line:90% So my point is this. 00:09:13.460 --> 00:09:15.470 align:middle line:84% Depending upon what I care about, 00:09:15.470 --> 00:09:21.180 align:middle line:84% whether I'm measuring voltage here or current here, 00:09:21.180 --> 00:09:26.630 align:middle line:84% the source impedance affects both of those measurements. 00:09:26.630 --> 00:09:32.000 align:middle line:84% And that's a problem because here's the way EMI works. 00:09:32.000 --> 00:09:34.550 align:middle line:90% So if I want to go-- 00:09:34.550 --> 00:09:36.140 align:middle line:84% if I want to go sell a power supply, 00:09:36.140 --> 00:09:38.960 align:middle line:84% a commercial power supply that I'm going to plug into the wall, 00:09:38.960 --> 00:09:42.440 align:middle line:84% I have to go test it and make it meet standards. 00:09:42.440 --> 00:09:44.250 align:middle line:90% And then I don't do anything. 00:09:44.250 --> 00:09:46.710 align:middle line:84% I just take that result and I put it in a desk drawer. 00:09:46.710 --> 00:09:51.230 align:middle line:84% And as long as nobody complains, great. 00:09:51.230 --> 00:09:53.780 align:middle line:84% But if somebody comes back and says, 00:09:53.780 --> 00:09:56.230 align:middle line:84% you're emitting too much EMI, then I've 00:09:56.230 --> 00:09:57.980 align:middle line:84% got to produce this test result, and we've 00:09:57.980 --> 00:09:59.230 align:middle line:90% got to come to some agreement. 00:09:59.230 --> 00:10:03.440 align:middle line:84% Or they've got to be-- they've got to be able to ascertain, 00:10:03.440 --> 00:10:06.830 align:middle line:84% did my power supply pass the test or not? 00:10:06.830 --> 00:10:10.550 align:middle line:84% And what I'm telling you here is that if you 00:10:10.550 --> 00:10:14.840 align:middle line:84% try to just plug a power supply into a source, the result 00:10:14.840 --> 00:10:17.480 align:middle line:84% you get depends very, very heavily 00:10:17.480 --> 00:10:19.730 align:middle line:90% on this source impedance. 00:10:19.730 --> 00:10:23.330 align:middle line:84% So if I'm designing a power converter for a car, 00:10:23.330 --> 00:10:26.580 align:middle line:84% that source impedance depends upon what model car, what 00:10:26.580 --> 00:10:29.500 align:middle line:84% week it is, what else they have plugged into the bus. 00:10:29.500 --> 00:10:32.340 align:middle line:84% It depends on all kinds of stuff that I don't control. 00:10:32.340 --> 00:10:36.000 align:middle line:84% So if I'm going to sell a power supply to plug into the BMW 00:10:36.000 --> 00:10:39.180 align:middle line:84% or into some other brand car, they 00:10:39.180 --> 00:10:42.990 align:middle line:84% have no way of coming to an agreement with me 00:10:42.990 --> 00:10:45.960 align:middle line:84% as to whether my power supply passes 00:10:45.960 --> 00:10:49.560 align:middle line:84% because on a different model or with different stuff plugged in, 00:10:49.560 --> 00:10:52.470 align:middle line:84% you might get a completely different result. 00:10:52.470 --> 00:10:55.910 align:middle line:84% Does that make sense to everybody? 00:10:55.910 --> 00:10:58.170 align:middle line:90% So what do you do about this? 00:10:58.170 --> 00:11:02.490 align:middle line:84% And this is what underpins EMI test procedures. 00:11:02.490 --> 00:11:05.810 align:middle line:84% And I will say that for any procedure 00:11:05.810 --> 00:11:09.500 align:middle line:84% you want to test, like SAE has a set of specifications 00:11:09.500 --> 00:11:11.820 align:middle line:84% that they recommend to test procedures. 00:11:11.820 --> 00:11:15.090 align:middle line:84% If you want to qualify for something into the wall, 00:11:15.090 --> 00:11:20.210 align:middle line:84% FCC Title 17 has a whole set of procedures that you follow. 00:11:20.210 --> 00:11:22.190 align:middle line:84% And these procedures are designed 00:11:22.190 --> 00:11:25.220 align:middle line:84% to deal with this issue, that you've 00:11:25.220 --> 00:11:29.450 align:middle line:84% got to frame a test that if I test it this week 00:11:29.450 --> 00:11:32.030 align:middle line:84% and I test it next week and I test it the week after that 00:11:32.030 --> 00:11:35.030 align:middle line:84% in some other place, I'm going to get more 00:11:35.030 --> 00:11:38.270 align:middle line:84% or less the same result. So it's not that it necessarily 00:11:38.270 --> 00:11:40.580 align:middle line:84% reflects the real environment, but it 00:11:40.580 --> 00:11:46.910 align:middle line:84% reflects a standardized environment that I can rely on. 00:11:46.910 --> 00:11:50.180 align:middle line:90% So here's how they do it. 00:11:50.180 --> 00:11:52.603 align:middle line:84% What we would really do is the following. 00:11:52.603 --> 00:11:54.020 align:middle line:84% Or actually, I'm going to give you 00:11:54.020 --> 00:11:56.710 align:middle line:84% a slightly simplified version of what you would really do. 00:11:56.710 --> 00:12:00.560 align:middle line:90% 00:12:00.560 --> 00:12:02.360 align:middle line:90% Suppose here's my load. 00:12:02.360 --> 00:12:04.100 align:middle line:84% And I'm talking about an input filter 00:12:04.100 --> 00:12:06.290 align:middle line:84% because I'm going to plug this thing into the wall. 00:12:06.290 --> 00:12:09.680 align:middle line:90% So here's my converter. 00:12:09.680 --> 00:12:14.480 align:middle line:84% And it's generating some ripple current ix. 00:12:14.480 --> 00:12:18.990 align:middle line:84% And here's whatever my filter is. 00:12:18.990 --> 00:12:22.790 align:middle line:84% And here I'm just going to draw my filter as a capacitor. 00:12:22.790 --> 00:12:24.650 align:middle line:84% What I'm going to do is I'm not going 00:12:24.650 --> 00:12:27.170 align:middle line:84% to plug this into my power source, which 00:12:27.170 --> 00:12:28.700 align:middle line:90% could be my vehicle battery. 00:12:28.700 --> 00:12:32.330 align:middle line:84% But instead, I'm going to go get another box 00:12:32.330 --> 00:12:36.050 align:middle line:84% and connect this up to a box that looks like this. 00:12:36.050 --> 00:12:41.970 align:middle line:84% It has a series inductance, L LISN, and a shunt capacitance, 00:12:41.970 --> 00:12:43.400 align:middle line:90% C LISN. 00:12:43.400 --> 00:12:46.820 align:middle line:84% Sometimes it has another capacitor here to ground. 00:12:46.820 --> 00:12:49.885 align:middle line:84% And then this is going to have some output. 00:12:49.885 --> 00:12:53.955 align:middle line:90% 00:12:53.955 --> 00:12:58.780 align:middle line:84% And then I'm going to connect this into the power source. 00:12:58.780 --> 00:12:59.620 align:middle line:90% So here we go. 00:12:59.620 --> 00:13:02.130 align:middle line:90% Here's zs. 00:13:02.130 --> 00:13:07.938 align:middle line:84% And here is vs. So this is whatever my usual source is. 00:13:07.938 --> 00:13:09.480 align:middle line:84% If this is my vehicle battery, that's 00:13:09.480 --> 00:13:11.160 align:middle line:90% what I'm going to plug it into. 00:13:11.160 --> 00:13:15.550 align:middle line:84% And this element here comes in a box. 00:13:15.550 --> 00:13:17.290 align:middle line:90% It's a standardized element. 00:13:17.290 --> 00:13:19.950 align:middle line:84% And then I'm going to go plug this output of this thing 00:13:19.950 --> 00:13:23.766 align:middle line:90% into a spectrum analyzer. 00:13:23.766 --> 00:13:29.790 align:middle line:90% 00:13:29.790 --> 00:13:31.290 align:middle line:90% And this is going to be-- 00:13:31.290 --> 00:13:35.820 align:middle line:84% the input of the spectrum analyzer has a 50-ohm input. 00:13:35.820 --> 00:13:38.730 align:middle line:84% So basically, my spectrum analyzer looks like a 50-ohm 00:13:38.730 --> 00:13:39.990 align:middle line:90% input. 00:13:39.990 --> 00:13:45.270 align:middle line:84% And the idea of this box is called a Line Impedance 00:13:45.270 --> 00:14:03.480 align:middle line:90% Stabilization Network, or LISN-- 00:14:03.480 --> 00:14:05.690 align:middle line:84% Line Impedance Stabilization Network. 00:14:05.690 --> 00:14:11.780 align:middle line:84% And its job is to present a consistent measurement 00:14:11.780 --> 00:14:13.640 align:middle line:90% point for EMI. 00:14:13.640 --> 00:14:15.240 align:middle line:90% So what does this thing do? 00:14:15.240 --> 00:14:17.450 align:middle line:84% It's its own kind of filter, but the idea 00:14:17.450 --> 00:14:22.430 align:middle line:84% is all the DC current I want to provide to my converter 00:14:22.430 --> 00:14:26.210 align:middle line:84% from my source will just pass through this inductor. 00:14:26.210 --> 00:14:29.720 align:middle line:84% This will have some cutoff frequency, 00:14:29.720 --> 00:14:32.600 align:middle line:84% and L LISN and C LISN in this capacitor, 00:14:32.600 --> 00:14:36.470 align:middle line:84% depending upon the LISN, will be specified. 00:14:36.470 --> 00:14:39.830 align:middle line:84% And below some frequency, all the current 00:14:39.830 --> 00:14:41.420 align:middle line:84% is just going to come from the input, 00:14:41.420 --> 00:14:44.030 align:middle line:84% and it will draw power just as normal. 00:14:44.030 --> 00:14:47.330 align:middle line:84% When I get to really, really high frequencies, tens 00:14:47.330 --> 00:14:49.580 align:middle line:84% of megahertz or whatever, the idea 00:14:49.580 --> 00:14:53.390 align:middle line:84% is that this path will look like an open circuit. 00:14:53.390 --> 00:14:55.560 align:middle line:84% C LISN will look like a short circuit. 00:14:55.560 --> 00:14:56.450 align:middle line:90% And I will be-- 00:14:56.450 --> 00:14:59.870 align:middle line:84% when I look back into here, if I look back 00:14:59.870 --> 00:15:02.210 align:middle line:84% this way, at radio frequencies, what I'm going to see 00:15:02.210 --> 00:15:04.730 align:middle line:90% is 50 ohms. 00:15:04.730 --> 00:15:09.560 align:middle line:84% And so whatever noise I conduct from my switching 00:15:09.560 --> 00:15:12.260 align:middle line:84% is going to be looking back into 50 ohms. 00:15:12.260 --> 00:15:15.440 align:middle line:84% And if we plug this test setup into here, 00:15:15.440 --> 00:15:18.430 align:middle line:84% where the zs is one thing, and somebody else plugs into it 00:15:18.430 --> 00:15:20.060 align:middle line:84% and zs is another thing, you're still 00:15:20.060 --> 00:15:23.420 align:middle line:84% going to get pretty close to the same results. 00:15:23.420 --> 00:15:25.888 align:middle line:84% So this box, this LISN, is only for testing. 00:15:25.888 --> 00:15:27.930 align:middle line:84% It's not something that goes with your converter. 00:15:27.930 --> 00:15:31.170 align:middle line:84% It's just part of the test stand to make measurements. 00:15:31.170 --> 00:15:33.600 align:middle line:84% And I give you examples of values. 00:15:33.600 --> 00:15:35.100 align:middle line:84% What are the values of those things? 00:15:35.100 --> 00:15:35.700 align:middle line:90% It depends. 00:15:35.700 --> 00:15:36.980 align:middle line:84% Are you building something for a car where 00:15:36.980 --> 00:15:38.272 align:middle line:90% they have one set of standards? 00:15:38.272 --> 00:15:41.772 align:middle line:84% Are you building it for FCC, where you plug into a wall, 00:15:41.772 --> 00:15:43.230 align:middle line:84% they have another set of standards? 00:15:43.230 --> 00:15:45.500 align:middle line:84% But all that is documented, and you test 00:15:45.500 --> 00:15:48.740 align:middle line:90% according to that standard. 00:15:48.740 --> 00:15:52.830 align:middle line:90% Any questions about this so far? 00:15:52.830 --> 00:15:55.560 align:middle line:84% AUDIENCE: So that capacitor over there, 00:15:55.560 --> 00:15:59.240 align:middle line:84% on the left side of the list of [INAUDIBLE]. 00:15:59.240 --> 00:16:03.563 align:middle line:90% 00:16:03.563 --> 00:16:05.230 align:middle line:84% DAVID PERREAULT: It depends on the LISN. 00:16:05.230 --> 00:16:07.110 align:middle line:90% So it depends on the standard. 00:16:07.110 --> 00:16:10.410 align:middle line:84% Some will just have an inductor and a capacitor. 00:16:10.410 --> 00:16:12.760 align:middle line:84% Others will have another capacitor on the other side. 00:16:12.760 --> 00:16:15.480 align:middle line:84% That's a function of-- what the LISN looks like exactly 00:16:15.480 --> 00:16:17.390 align:middle line:90% inside depends on the standard. 00:16:17.390 --> 00:16:19.910 align:middle line:90% 00:16:19.910 --> 00:16:21.000 align:middle line:90% Other questions? 00:16:21.000 --> 00:16:24.415 align:middle line:90% 00:16:24.415 --> 00:16:26.380 align:middle line:84% So presenting a non-LISN impedance 00:16:26.380 --> 00:16:30.130 align:middle line:84% lets me make repeatable measurements on my converter. 00:16:30.130 --> 00:16:34.240 align:middle line:84% And if I measure it this week, I get the same result 00:16:34.240 --> 00:16:35.590 align:middle line:90% as next week. 00:16:35.590 --> 00:16:37.720 align:middle line:84% Usually the way you would do this-- 00:16:37.720 --> 00:16:40.300 align:middle line:84% I should say, by the way, there's a lot of-- 00:16:40.300 --> 00:16:43.510 align:middle line:84% if you look up to get very, very repeatable measurements, 00:16:43.510 --> 00:16:44.930 align:middle line:84% they have all kinds of standards. 00:16:44.930 --> 00:16:49.600 align:middle line:84% Your converter has to be one meter or some specified 00:16:49.600 --> 00:16:51.260 align:middle line:90% height over a ground plane. 00:16:51.260 --> 00:16:55.222 align:middle line:84% It's like they make all the conditions exactly identical, 00:16:55.222 --> 00:16:56.680 align:middle line:84% which aren't real-world conditions, 00:16:56.680 --> 00:16:59.440 align:middle line:84% but they're just designed to make the test repeatable. 00:16:59.440 --> 00:17:00.983 align:middle line:90% And you follow that. 00:17:00.983 --> 00:17:03.400 align:middle line:84% So very often, if you're trying to qualify a power supply, 00:17:03.400 --> 00:17:08.619 align:middle line:84% you'll ultimately go to a test house that that's what they do. 00:17:08.619 --> 00:17:11.589 align:middle line:84% What you do in the lab is you can set up an approximation 00:17:11.589 --> 00:17:12.713 align:middle line:90% to that usually. 00:17:12.713 --> 00:17:14.380 align:middle line:84% I mean, you can set up a real test stand 00:17:14.380 --> 00:17:16.380 align:middle line:84% if you're a company that's making lots of parts. 00:17:16.380 --> 00:17:18.069 align:middle line:84% But at home, you can go buy a LISN 00:17:18.069 --> 00:17:20.398 align:middle line:90% and do an approximation to that. 00:17:20.398 --> 00:17:21.440 align:middle line:90% And it'll be pretty good. 00:17:21.440 --> 00:17:23.023 align:middle line:84% It's good enough for designing a power 00:17:23.023 --> 00:17:25.770 align:middle line:84% supply if you put a bunch of margin in your answers. 00:17:25.770 --> 00:17:27.380 align:middle line:90% And the way this thing-- 00:17:27.380 --> 00:17:30.870 align:middle line:84% the way the specifications are read is, for conducting EMI, 00:17:30.870 --> 00:17:34.400 align:middle line:84% you'll look at this response, spectrum analyzer 00:17:34.400 --> 00:17:38.210 align:middle line:84% configured a certain way with certain kinds quasi peak or peak 00:17:38.210 --> 00:17:38.880 align:middle line:90% detectors. 00:17:38.880 --> 00:17:40.970 align:middle line:84% There's a whole bunch of rules around it. 00:17:40.970 --> 00:17:44.150 align:middle line:84% And you're generally measuring the energy 00:17:44.150 --> 00:17:46.910 align:middle line:84% in different frequency bands and seeing, 00:17:46.910 --> 00:17:50.320 align:middle line:84% do they go above an allowed limit? 00:17:50.320 --> 00:17:53.580 align:middle line:84% And those allowed limits are really small. 00:17:53.580 --> 00:17:55.560 align:middle line:90% So if I thought I had-- 00:17:55.560 --> 00:17:58.680 align:middle line:90% 00:17:58.680 --> 00:18:03.900 align:middle line:84% this was tens of amps, here I might be required 00:18:03.900 --> 00:18:06.240 align:middle line:90% to get tens of microamps. 00:18:06.240 --> 00:18:08.580 align:middle line:84% So it's really small currents that you've 00:18:08.580 --> 00:18:12.280 align:middle line:90% got to attenuate down to. 00:18:12.280 --> 00:18:19.140 align:middle line:84% So when I show you, oh, we've got a capacitor here, 00:18:19.140 --> 00:18:22.350 align:middle line:84% the model for this would then be basically a capacitor 00:18:22.350 --> 00:18:23.970 align:middle line:84% in parallel with 50 ohms, and I'm 00:18:23.970 --> 00:18:26.570 align:middle line:84% trying to measure that voltage ripple on the 50 ohms, 00:18:26.570 --> 00:18:30.600 align:middle line:84% you're going to need more than a capacitor to make this work. 00:18:30.600 --> 00:18:33.730 align:middle line:84% So very often, you're going to need a multi-stage filter. 00:18:33.730 --> 00:18:36.870 align:middle line:84% I will say, one other thing, by the way, 00:18:36.870 --> 00:18:40.770 align:middle line:84% I've drawn this as if you have a LISN that you stick 00:18:40.770 --> 00:18:43.320 align:middle line:84% in between your power source and your converter, 00:18:43.320 --> 00:18:46.800 align:middle line:84% and you do, but typically, most specifications 00:18:46.800 --> 00:18:50.550 align:middle line:84% require that you put a LISN on this lead and a separate LISN 00:18:50.550 --> 00:18:52.710 align:middle line:90% on this lead. 00:18:52.710 --> 00:18:58.510 align:middle line:84% So there's actually two LISNs involved in a DC system test. 00:18:58.510 --> 00:19:02.530 align:middle line:84% We'll come back to why you would do that later. 00:19:02.530 --> 00:19:05.080 align:middle line:84% But the basic concept is not changed. 00:19:05.080 --> 00:19:08.160 align:middle line:90% It's the same idea. 00:19:08.160 --> 00:19:13.180 align:middle line:84% Now, let's just think of what we would need. 00:19:13.180 --> 00:19:17.250 align:middle line:84% I said, OK, a capacitor is not really 00:19:17.250 --> 00:19:20.100 align:middle line:84% going to cut it for this measurement. 00:19:20.100 --> 00:19:24.090 align:middle line:84% So let me think about my converter 00:19:24.090 --> 00:19:30.540 align:middle line:90% as some current ix or ix tilde. 00:19:30.540 --> 00:19:32.110 align:middle line:84% This is the ripple from my converter. 00:19:32.110 --> 00:19:35.420 align:middle line:90% 00:19:35.420 --> 00:19:42.300 align:middle line:84% If a converter-- if a capacitor is not going to be enough, 00:19:42.300 --> 00:19:46.080 align:middle line:84% what would be my basic building block? 00:19:46.080 --> 00:19:49.350 align:middle line:84% My basic building block might look something like this. 00:19:49.350 --> 00:19:52.540 align:middle line:84% It might be some filter capacitor Cf. 00:19:52.540 --> 00:19:54.180 align:middle line:84% And the first stage could be just 00:19:54.180 --> 00:19:56.430 align:middle line:84% my input capacitor, my converter, which I would always 00:19:56.430 --> 00:20:00.120 align:middle line:90% have and some inductor Lf. 00:20:00.120 --> 00:20:02.710 align:middle line:84% Now, this is not the LISN inductor. 00:20:02.710 --> 00:20:05.370 align:middle line:90% This is my filter inductor. 00:20:05.370 --> 00:20:07.060 align:middle line:90% So this is my filter. 00:20:07.060 --> 00:20:11.460 align:middle line:84% And then eventually I look back into the LISN impedance. 00:20:11.460 --> 00:20:15.750 align:middle line:84% And this would be z LISN looking here. 00:20:15.750 --> 00:20:19.076 align:middle line:84% And at ripple frequencies, maybe that looks like 50 ohms. 00:20:19.076 --> 00:20:21.750 align:middle line:90% 00:20:21.750 --> 00:20:28.200 align:middle line:84% So a basic building block is an Lc filter. 00:20:28.200 --> 00:20:32.460 align:middle line:84% What is the transfer function-- if I said this was ix tilde 00:20:32.460 --> 00:20:36.630 align:middle line:84% and this was iy tilde, what does the transfer function 00:20:36.630 --> 00:20:40.034 align:middle line:90% of this thing look like? 00:20:40.034 --> 00:20:46.750 align:middle line:84% The magnitude of iy tilde over ix tilde versus omega. 00:20:46.750 --> 00:20:48.780 align:middle line:84% And this is going to be a log scale, 00:20:48.780 --> 00:20:52.050 align:middle line:84% and this is going to be a log scale. 00:20:52.050 --> 00:20:55.890 align:middle line:84% At DC, all the current goes back. 00:20:55.890 --> 00:21:01.680 align:middle line:84% So at DC, I'm going to get this at really low frequencies. 00:21:01.680 --> 00:21:05.160 align:middle line:84% Then I'm going to have some region that's 00:21:05.160 --> 00:21:09.480 align:middle line:84% 1 over the square root of LfCf, which is going to be 00:21:09.480 --> 00:21:11.490 align:middle line:90% the cutoff of this filter. 00:21:11.490 --> 00:21:18.690 align:middle line:84% And at really high frequencies, this is going to fall off as 1 00:21:18.690 --> 00:21:26.410 align:middle line:84% over omega squared, or minus 40 db per decade. 00:21:26.410 --> 00:21:29.738 align:middle line:84% So the idea here is I can go get a bunch of attenuation. 00:21:29.738 --> 00:21:31.530 align:middle line:84% I'm going to put this cutoff of this filter 00:21:31.530 --> 00:21:36.630 align:middle line:84% well below my 150 kilohertz that I care about attenuating 00:21:36.630 --> 00:21:40.840 align:middle line:84% and live off of this attenuation out here. 00:21:40.840 --> 00:21:44.380 align:middle line:84% Now, why is that going as 1 over omega squared? 00:21:44.380 --> 00:21:48.400 align:middle line:84% Just as a reminder, the impedance of the capacitor, 00:21:48.400 --> 00:21:54.910 align:middle line:84% the magnitude of zc, goes as 1 over omega C. 00:21:54.910 --> 00:21:58.540 align:middle line:84% So this is getting to be a smaller impedance as frequency 00:21:58.540 --> 00:21:59.350 align:middle line:90% goes up. 00:21:59.350 --> 00:22:04.630 align:middle line:84% And z sub L is going as omega L. So basically, 00:22:04.630 --> 00:22:07.030 align:middle line:84% you get an omega squared factor because he's getting-- 00:22:07.030 --> 00:22:10.150 align:middle line:84% it's a current divider, and he's getting smaller 00:22:10.150 --> 00:22:12.580 align:middle line:84% inversely proportional to omega, and he's getting 00:22:12.580 --> 00:22:13.940 align:middle line:90% bigger proportional to omega. 00:22:13.940 --> 00:22:16.840 align:middle line:84% So you get a kind of an omega squared attenuation. 00:22:16.840 --> 00:22:19.810 align:middle line:84% So the way you might want to think about this 00:22:19.810 --> 00:22:23.170 align:middle line:84% is your filter's working because of the mismatch 00:22:23.170 --> 00:22:26.710 align:middle line:84% between the shunt path and the series path. 00:22:26.710 --> 00:22:30.010 align:middle line:84% And if I don't like this, I can build higher-order filters 00:22:30.010 --> 00:22:30.640 align:middle line:90% and so forth. 00:22:30.640 --> 00:22:33.220 align:middle line:90% And we'll talk about that. 00:22:33.220 --> 00:22:36.850 align:middle line:84% I haven't said what happens in the vicinity of cutoff. 00:22:36.850 --> 00:22:38.140 align:middle line:90% That's an important thing. 00:22:38.140 --> 00:22:40.010 align:middle line:84% And we're going to have to think about that. 00:22:40.010 --> 00:22:47.620 align:middle line:84% But let's for the moment not worry about-- 00:22:47.620 --> 00:22:49.580 align:middle line:84% not worry about what's going on here. 00:22:49.580 --> 00:22:51.400 align:middle line:90% We'll come back to that point. 00:22:51.400 --> 00:22:54.315 align:middle line:90% So any questions up to now? 00:22:54.315 --> 00:23:01.300 align:middle line:90% 00:23:01.300 --> 00:23:04.410 align:middle line:90% So I want to build this filter. 00:23:04.410 --> 00:23:08.796 align:middle line:84% And as I said, I need a lot of attenuation. 00:23:08.796 --> 00:23:11.700 align:middle line:90% 00:23:11.700 --> 00:23:14.230 align:middle line:84% If I'm going from 100 amps to 100 microamps, 00:23:14.230 --> 00:23:18.400 align:middle line:84% that's six orders of attenuation. 00:23:18.400 --> 00:23:25.350 align:middle line:84% So I'm going to probably need a pretty low cutoff filter here. 00:23:25.350 --> 00:23:27.400 align:middle line:84% I'm going to need a bunch of attenuation. 00:23:27.400 --> 00:23:30.510 align:middle line:84% And one of the things that people 00:23:30.510 --> 00:23:34.020 align:middle line:84% find about designing EMI filters, your average power 00:23:34.020 --> 00:23:36.780 align:middle line:90% electronics guy hates it. 00:23:36.780 --> 00:23:38.380 align:middle line:90% Why do they hate it? 00:23:38.380 --> 00:23:41.220 align:middle line:84% They hate it because you can very get-- 00:23:41.220 --> 00:23:45.810 align:middle line:84% you can very often get very kind of seemingly anomalous results. 00:23:45.810 --> 00:23:49.680 align:middle line:84% Like, you're doing stuff, and you're just getting noise, 00:23:49.680 --> 00:23:51.400 align:middle line:90% and you don't want this noise. 00:23:51.400 --> 00:23:56.170 align:middle line:90% And it becomes a problem. 00:23:56.170 --> 00:24:00.940 align:middle line:84% It's hard to get the good result that a simple picture like this 00:24:00.940 --> 00:24:03.100 align:middle line:90% would suggest that you can get. 00:24:03.100 --> 00:24:06.310 align:middle line:84% So let me tell you-- just give you 00:24:06.310 --> 00:24:08.170 align:middle line:84% a heads up about some of the things 00:24:08.170 --> 00:24:11.330 align:middle line:84% you got to pay attention to when you're doing this. 00:24:11.330 --> 00:24:17.740 align:middle line:84% And the first of these things is component parasitics. 00:24:17.740 --> 00:24:23.020 align:middle line:84% So if I were to look at this filter, and I'd say, jeez, 00:24:23.020 --> 00:24:25.370 align:middle line:90% what does this look like? 00:24:25.370 --> 00:24:26.770 align:middle line:90% Well, I said my capacitor-- 00:24:26.770 --> 00:24:29.980 align:middle line:90% 00:24:29.980 --> 00:24:32.740 align:middle line:84% if I looked at my capacitor, and I said, 00:24:32.740 --> 00:24:36.130 align:middle line:84% what is the magnitude of z sub C? 00:24:36.130 --> 00:24:41.800 align:middle line:84% I said that the magnitude of z sub C should fall off as 1 00:24:41.800 --> 00:24:43.780 align:middle line:90% over omega C, right? 00:24:43.780 --> 00:24:47.260 align:middle line:84% So this is the magnitude of z sub C. 00:24:47.260 --> 00:24:53.140 align:middle line:84% And if I had an ideal capacitor, it would do that forever. 00:24:53.140 --> 00:24:55.770 align:middle line:90% But what about a real capacitor? 00:24:55.770 --> 00:25:05.790 align:middle line:84% Well, a real capacitor, even though I think of it 00:25:05.790 --> 00:25:07.830 align:middle line:84% as an energy storage element, the plates 00:25:07.830 --> 00:25:11.670 align:middle line:90% do have some resistance. 00:25:11.670 --> 00:25:14.520 align:middle line:84% There's actually some resistance of the wiring 00:25:14.520 --> 00:25:16.090 align:middle line:90% going into that thing. 00:25:16.090 --> 00:25:21.000 align:middle line:84% So as a practical matter, a real capacitor 00:25:21.000 --> 00:25:26.700 align:middle line:84% is going to have some ESR, series resistance. 00:25:26.700 --> 00:25:28.270 align:middle line:90% So what's that going to do? 00:25:28.270 --> 00:25:35.350 align:middle line:84% Well, there's going to be some frequency above which this ESR 00:25:35.350 --> 00:25:39.720 align:middle line:84% has a bigger impedance than the capacitor, the ideal portion 00:25:39.720 --> 00:25:41.010 align:middle line:90% of the capacitor. 00:25:41.010 --> 00:25:45.330 align:middle line:84% So what I would expect is beyond some frequency, 00:25:45.330 --> 00:25:52.990 align:middle line:84% there's going to be some place where our R ESR is dominant. 00:25:52.990 --> 00:25:55.260 align:middle line:84% So instead of falling off forever as 1 over omega, 00:25:55.260 --> 00:25:57.840 align:middle line:90% it flattens out eventually. 00:25:57.840 --> 00:26:02.580 align:middle line:84% And then if I keep going up in frequency, 00:26:02.580 --> 00:26:07.170 align:middle line:84% I'm going to get to some point where 00:26:07.170 --> 00:26:12.180 align:middle line:84% I have some equivalent series inductance, L ESL. 00:26:12.180 --> 00:26:13.540 align:middle line:90% Why does that exist? 00:26:13.540 --> 00:26:15.785 align:middle line:84% Because there is space between the capacitors, 00:26:15.785 --> 00:26:17.160 align:middle line:84% and if I have current through it, 00:26:17.160 --> 00:26:19.020 align:middle line:84% there's some magnetic field in there. 00:26:19.020 --> 00:26:21.160 align:middle line:84% And there's just a little bit of inductance. 00:26:21.160 --> 00:26:23.460 align:middle line:84% It doesn't have to be very much, but we're 00:26:23.460 --> 00:26:26.370 align:middle line:84% talking about going to very high frequencies. 00:26:26.370 --> 00:26:30.420 align:middle line:84% So at some point, this ESL will even 00:26:30.420 --> 00:26:33.570 align:middle line:84% get to have a high-- because this is growing with omega, 00:26:33.570 --> 00:26:34.775 align:middle line:90% it's going to go up. 00:26:34.775 --> 00:26:37.710 align:middle line:90% 00:26:37.710 --> 00:26:42.060 align:middle line:84% So my overall contour of my impedance doesn't fall forever, 00:26:42.060 --> 00:26:45.240 align:middle line:84% but rather, it sort of does this kind of falls, 00:26:45.240 --> 00:26:46.560 align:middle line:90% flattens out, and goes up. 00:26:46.560 --> 00:26:48.810 align:middle line:90% And maybe this-- 00:26:48.810 --> 00:26:51.570 align:middle line:84% I've shown you a high ESR capacitor. 00:26:51.570 --> 00:26:54.330 align:middle line:84% Depending upon whether the ESR is greater than or less 00:26:54.330 --> 00:26:57.040 align:middle line:84% than the square root of L over C ESL-- 00:26:57.040 --> 00:27:03.060 align:middle line:84% L ESL over C, this will be very sharply tuned or very broad. 00:27:03.060 --> 00:27:07.540 align:middle line:84% But either way, you're going to get this kind of behavior. 00:27:07.540 --> 00:27:09.630 align:middle line:90% Why am I harping on this? 00:27:09.630 --> 00:27:14.460 align:middle line:84% Because if I thought of this as my real element, now 00:27:14.460 --> 00:27:18.930 align:middle line:84% I plug this thing in place of my capacitor. 00:27:18.930 --> 00:27:20.970 align:middle line:84% What would I expect my attenuation 00:27:20.970 --> 00:27:22.560 align:middle line:90% to look like in that case? 00:27:22.560 --> 00:27:25.955 align:middle line:84% Well, I said it was based on the mismatch 00:27:25.955 --> 00:27:27.330 align:middle line:84% between this impedance going down 00:27:27.330 --> 00:27:28.690 align:middle line:90% and this impedance going up. 00:27:28.690 --> 00:27:34.020 align:middle line:84% So eventually, instead of turning into an LC filter, 00:27:34.020 --> 00:27:37.170 align:middle line:90% it turns into an LR filter. 00:27:37.170 --> 00:27:39.480 align:middle line:84% So instead of falling off as 1 over omega squared, 00:27:39.480 --> 00:27:42.510 align:middle line:84% it's going to fall off as 1 over omega. 00:27:42.510 --> 00:27:46.260 align:middle line:84% And when I get to really high frequencies, 00:27:46.260 --> 00:27:49.140 align:middle line:84% ESL is going to dominate, and then I'm just going to have a-- 00:27:49.140 --> 00:27:53.280 align:middle line:84% I'm just going to have a current divider between L ESL and L 00:27:53.280 --> 00:27:55.020 align:middle line:90% filter. 00:27:55.020 --> 00:27:58.290 align:middle line:84% And in fact, what that's going to mean is my attenuation 00:27:58.290 --> 00:28:00.930 align:middle line:90% is just going to flatten out. 00:28:00.930 --> 00:28:03.450 align:middle line:84% So if I design my filter thinking, 00:28:03.450 --> 00:28:08.610 align:middle line:84% oh, jeez, that's how I get my six orders of magnitude, 00:28:08.610 --> 00:28:12.210 align:middle line:84% and I come back and I'm really dealing with something 00:28:12.210 --> 00:28:14.190 align:middle line:84% where at that frequency, this capacitor doesn't 00:28:14.190 --> 00:28:17.370 align:middle line:84% look like a capacitor, it looks like a little inductor, 00:28:17.370 --> 00:28:22.565 align:middle line:84% my EMI might be way higher than I thought it was going to be. 00:28:22.565 --> 00:28:23.940 align:middle line:84% My filtering is just going to get 00:28:23.940 --> 00:28:31.050 align:middle line:84% clobbered by the parasitic inductance of my capacitor. 00:28:31.050 --> 00:28:34.380 align:middle line:84% Now, I will say that, by the way, you know this-- 00:28:34.380 --> 00:28:36.790 align:middle line:84% and by the way, where does this kind of phenomena happen? 00:28:36.790 --> 00:28:40.150 align:middle line:84% Well, it depends on the size of your capacitor, 00:28:40.150 --> 00:28:41.650 align:middle line:84% the physical size of your capacitor. 00:28:41.650 --> 00:28:43.442 align:middle line:84% It depends on whether it's a good capacitor 00:28:43.442 --> 00:28:44.230 align:middle line:90% or not, et cetera. 00:28:44.230 --> 00:28:47.160 align:middle line:84% But very often, like 10 megahertz kind of frequencies 00:28:47.160 --> 00:28:50.160 align:middle line:84% is where these things are happening, as an example, 00:28:50.160 --> 00:28:52.692 align:middle line:84% or even 1 megahertz-- depends on the size of the component. 00:28:52.692 --> 00:28:54.900 align:middle line:84% And that's right in the band that you're caring about 00:28:54.900 --> 00:28:57.150 align:middle line:90% and trying to attenuate. 00:28:57.150 --> 00:29:01.860 align:middle line:84% So these bad parasitics of the components-- 00:29:01.860 --> 00:29:04.530 align:middle line:84% and usually the capacitor is the first thing to give out on you 00:29:04.530 --> 00:29:06.360 align:middle line:84% if you design the inductors right-- 00:29:06.360 --> 00:29:09.960 align:middle line:84% will cause you to get a lot less attenuation out of this LC 00:29:09.960 --> 00:29:12.390 align:middle line:90% filter than you thought. 00:29:12.390 --> 00:29:13.900 align:middle line:90% Any questions about that? 00:29:13.900 --> 00:29:17.450 align:middle line:90% 00:29:17.450 --> 00:29:20.485 align:middle line:90% So that's kind of a problem. 00:29:20.485 --> 00:29:26.310 align:middle line:90% 00:29:26.310 --> 00:29:30.390 align:middle line:90% What would I do about it? 00:29:30.390 --> 00:29:35.640 align:middle line:84% What I might do about it is not just have one capacitor 00:29:35.640 --> 00:29:38.910 align:middle line:84% but have a bunch of capacitors, some of which 00:29:38.910 --> 00:29:43.170 align:middle line:84% might be good at-- they're not a very high C value, 00:29:43.170 --> 00:29:46.350 align:middle line:84% but they don't have as much inductance. 00:29:46.350 --> 00:29:48.450 align:middle line:84% And so they look better at high frequencies 00:29:48.450 --> 00:29:51.120 align:middle line:84% than bigger capacitors that have better performance 00:29:51.120 --> 00:29:53.530 align:middle line:84% at low frequencies but not as good as high frequencies. 00:29:53.530 --> 00:29:56.520 align:middle line:84% So I can certainly do it by being careful 00:29:56.520 --> 00:29:59.550 align:middle line:84% about how I pick this component or picking multiple components 00:29:59.550 --> 00:30:01.540 align:middle line:84% to cover different frequency ranges. 00:30:01.540 --> 00:30:03.480 align:middle line:84% And if I have multiple components, by the way, 00:30:03.480 --> 00:30:04.930 align:middle line:84% I can get resonances between them. 00:30:04.930 --> 00:30:07.230 align:middle line:90% But that's another story. 00:30:07.230 --> 00:30:09.890 align:middle line:90% What else can I do? 00:30:09.890 --> 00:30:15.920 align:middle line:84% I might say, well, OK, if I'm limited by ESL, 00:30:15.920 --> 00:30:19.280 align:middle line:84% I will use a bigger Lf, but you're still only getting 00:30:19.280 --> 00:30:22.130 align:middle line:84% a divider that's a proportion between these two, 00:30:22.130 --> 00:30:23.550 align:middle line:90% the ratio of those two. 00:30:23.550 --> 00:30:28.340 align:middle line:84% If I want really more attenuation, what I can do 00:30:28.340 --> 00:30:30.540 align:middle line:90% is do a higher-stage filter. 00:30:30.540 --> 00:30:37.340 align:middle line:90% So here's ix tilde. 00:30:37.340 --> 00:30:41.420 align:middle line:90% Maybe I build a first stage-- 00:30:41.420 --> 00:30:46.670 align:middle line:90% I'll call this Cf1 and Lf1. 00:30:46.670 --> 00:30:48.697 align:middle line:84% And I'm leaving out damping and other things 00:30:48.697 --> 00:30:50.280 align:middle line:84% I'm going to talk about in the future. 00:30:50.280 --> 00:30:59.790 align:middle line:84% And then maybe I have Cf2 and Lf2 and so forth. 00:30:59.790 --> 00:31:01.650 align:middle line:90% And then maybe I have my-- 00:31:01.650 --> 00:31:06.020 align:middle line:84% I'm looking back into R LISN back here. 00:31:06.020 --> 00:31:09.110 align:middle line:84% So maybe I can get 40 db per decade out of this 00:31:09.110 --> 00:31:11.100 align:middle line:84% and 40 db per decade out of this. 00:31:11.100 --> 00:31:12.480 align:middle line:90% And maybe I do better. 00:31:12.480 --> 00:31:15.460 align:middle line:84% And it may be better to do that even in a lot of cases 00:31:15.460 --> 00:31:18.660 align:middle line:84% because even putting aside the parasitics, 00:31:18.660 --> 00:31:21.210 align:middle line:84% I can move the cutoff frequency down 1 00:31:21.210 --> 00:31:23.467 align:middle line:90% over the square root of LC. 00:31:23.467 --> 00:31:25.050 align:middle line:84% By using bigger and bigger components, 00:31:25.050 --> 00:31:27.240 align:middle line:84% I can move that cutoff filter down. 00:31:27.240 --> 00:31:30.750 align:middle line:84% Here, maybe I use smaller components, 00:31:30.750 --> 00:31:37.470 align:middle line:84% but instead of getting a 40 db per decade roll-off, 00:31:37.470 --> 00:31:39.240 align:middle line:84% I could theoretically get 80 out of this. 00:31:39.240 --> 00:31:41.060 align:middle line:84% So even if the net cutoff's higher, 00:31:41.060 --> 00:31:43.140 align:middle line:84% that fast attenuation will help me. 00:31:43.140 --> 00:31:47.810 align:middle line:84% So it's often better to go for smaller multi-stage filters 00:31:47.810 --> 00:31:53.660 align:middle line:84% than one big fat single filter stage trying 00:31:53.660 --> 00:31:56.780 align:middle line:90% to get the attenuation I want. 00:31:56.780 --> 00:32:00.020 align:middle line:84% And by the way, when you do this, for reasons I'm not really 00:32:00.020 --> 00:32:01.520 align:middle line:84% going to go into the details of this 00:32:01.520 --> 00:32:02.790 align:middle line:90% because we don't have time-- 00:32:02.790 --> 00:32:04.700 align:middle line:84% when you design these kind of things, 00:32:04.700 --> 00:32:08.660 align:middle line:84% it might be your inclination to say, oh, 00:32:08.660 --> 00:32:11.930 align:middle line:84% let me use the same LfCf cutoff here and here 00:32:11.930 --> 00:32:14.070 align:middle line:84% and put two filters with the same cutoff. 00:32:14.070 --> 00:32:16.440 align:middle line:84% It turns out to be better to stagger these things. 00:32:16.440 --> 00:32:19.070 align:middle line:84% So usually, you'd make this one a low cutoff 00:32:19.070 --> 00:32:21.012 align:middle line:84% and then this one a higher cutoff 00:32:21.012 --> 00:32:22.970 align:middle line:84% to deal with the higher frequency and so forth. 00:32:22.970 --> 00:32:25.460 align:middle line:84% It turns out to be a better kind of design strategy 00:32:25.460 --> 00:32:26.970 align:middle line:90% for these kind of filters. 00:32:26.970 --> 00:32:30.260 align:middle line:84% So you can cascade filter sections 00:32:30.260 --> 00:32:32.250 align:middle line:84% until you get the attenuation you need. 00:32:32.250 --> 00:32:35.468 align:middle line:84% And if you look in a typical offline power supply 00:32:35.468 --> 00:32:37.260 align:middle line:84% that you'd plug into the wall, they usually 00:32:37.260 --> 00:32:40.110 align:middle line:84% have a couple filter stages like this. 00:32:40.110 --> 00:32:42.710 align:middle line:90% Any questions about that? 00:32:42.710 --> 00:32:46.610 align:middle line:90% 00:32:46.610 --> 00:32:48.090 align:middle line:90% I should warn you too-- 00:32:48.090 --> 00:32:50.125 align:middle line:90% I said, oh, people hate these-- 00:32:50.125 --> 00:32:53.000 align:middle line:90% 00:32:53.000 --> 00:32:56.720 align:middle line:84% they hate dealing with the EMI because you go to measure stuff 00:32:56.720 --> 00:32:59.670 align:middle line:84% and then what you measure doesn't seem to make sense. 00:32:59.670 --> 00:33:06.100 align:middle line:84% And it all does make sense, but very small phenomena 00:33:06.100 --> 00:33:07.320 align:middle line:90% can mess you up. 00:33:07.320 --> 00:33:08.800 align:middle line:90% So think about this. 00:33:08.800 --> 00:33:10.590 align:middle line:84% This filter and it's cutoff regime, 00:33:10.590 --> 00:33:15.250 align:middle line:84% putting aside these parasitics I was talking about, 00:33:15.250 --> 00:33:17.350 align:middle line:90% could give you 80 db per decade. 00:33:17.350 --> 00:33:21.050 align:middle line:84% You can get really like, wham, lots of attenuation. 00:33:21.050 --> 00:33:23.510 align:middle line:84% And that's good because you need a lot of attenuation. 00:33:23.510 --> 00:33:24.770 align:middle line:90% But what's the downside? 00:33:24.770 --> 00:33:28.060 align:middle line:84% Well, one thing is the component parasitics themselves. 00:33:28.060 --> 00:33:29.800 align:middle line:90% And I focused on the capacitor. 00:33:29.800 --> 00:33:31.210 align:middle line:90% The inductors do have-- 00:33:31.210 --> 00:33:35.110 align:middle line:84% if you put a capacitance, inductors 00:33:35.110 --> 00:33:37.120 align:middle line:84% do have a parasitic parallel capacitance. 00:33:37.120 --> 00:33:38.980 align:middle line:84% You also have to worry about that one. 00:33:38.980 --> 00:33:42.010 align:middle line:84% But putting aside that parasitic, you also 00:33:42.010 --> 00:33:47.110 align:middle line:84% have layout parasitics, meaning if I do a crummy job 00:33:47.110 --> 00:33:51.220 align:middle line:84% and I don't stick the components into the board all the way 00:33:51.220 --> 00:33:54.970 align:middle line:84% or I put two components next to each other, 00:33:54.970 --> 00:33:59.160 align:middle line:84% they can have capacitance between them. 00:33:59.160 --> 00:34:00.942 align:middle line:84% So I can accidentally insert inductance, 00:34:00.942 --> 00:34:02.650 align:middle line:84% or I can accidentally insert capacitance. 00:34:02.650 --> 00:34:04.192 align:middle line:84% And one of the things that can happen 00:34:04.192 --> 00:34:09.570 align:middle line:84% is imagine this is my power supply input leads, 00:34:09.570 --> 00:34:13.392 align:middle line:84% and over here is where it's going to go off to-- 00:34:13.392 --> 00:34:14.850 align:middle line:84% I'm showing the LISN here, but this 00:34:14.850 --> 00:34:17.550 align:middle line:84% is where it goes off to the power supply input. 00:34:17.550 --> 00:34:20.670 align:middle line:84% What if I put those leads near each other? 00:34:20.670 --> 00:34:23.219 align:middle line:84% Well, then maybe what I would end up with 00:34:23.219 --> 00:34:26.730 align:middle line:84% is some small parasitic capacitance 00:34:26.730 --> 00:34:33.780 align:middle line:84% from this node past my filter to this node. 00:34:33.780 --> 00:34:35.440 align:middle line:90% And so what happens? 00:34:35.440 --> 00:34:37.530 align:middle line:84% This filter is doing some great job, 00:34:37.530 --> 00:34:39.730 align:middle line:90% massively killing the noise. 00:34:39.730 --> 00:34:44.219 align:middle line:84% But this, because this node has some switching ripple at it, 00:34:44.219 --> 00:34:45.989 align:middle line:84% it's being injected through this. 00:34:45.989 --> 00:34:48.469 align:middle line:84% Maybe it's a picofarad of capacitance. 00:34:48.469 --> 00:34:49.969 align:middle line:84% It doesn't even have to be very much 00:34:49.969 --> 00:34:52.860 align:middle line:84% if you have a high attenuation filter getting over to here 00:34:52.860 --> 00:34:55.367 align:middle line:84% and completely bypassing your filter. 00:34:55.367 --> 00:34:56.909 align:middle line:84% So you're getting bad results, you're 00:34:56.909 --> 00:34:59.730 align:middle line:84% measuring stuff over here, and you make this filter better 00:34:59.730 --> 00:35:00.670 align:middle line:90% and nothing happens. 00:35:00.670 --> 00:35:02.230 align:middle line:84% You make this filter better and nothing happens. 00:35:02.230 --> 00:35:02.850 align:middle line:90% Why? 00:35:02.850 --> 00:35:04.642 align:middle line:84% Because it's not going through your filter. 00:35:04.642 --> 00:35:06.390 align:middle line:90% It's going around your filter. 00:35:06.390 --> 00:35:10.590 align:middle line:84% And that's the kind of thing that drives people nuts. 00:35:10.590 --> 00:35:13.290 align:middle line:84% So you have to think very carefully about your layout. 00:35:13.290 --> 00:35:18.790 align:middle line:84% People really try to have a quiet side and a noisy side 00:35:18.790 --> 00:35:21.160 align:middle line:84% so that they stay physically far apart so you 00:35:21.160 --> 00:35:23.800 align:middle line:90% don't get capacitive bridging. 00:35:23.800 --> 00:35:25.550 align:middle line:84% You can also have magnetic bridging. 00:35:25.550 --> 00:35:26.650 align:middle line:90% If I have this-- 00:35:26.650 --> 00:35:29.530 align:middle line:84% if this loop with a lot of switching ripple 00:35:29.530 --> 00:35:32.440 align:middle line:84% has mutual inductance to this loop, 00:35:32.440 --> 00:35:34.360 align:middle line:84% even if there's no capacitance, I 00:35:34.360 --> 00:35:38.730 align:middle line:84% can get magnetically coupled right across through my layout 00:35:38.730 --> 00:35:43.410 align:middle line:84% and completely ruin my filter performance. 00:35:43.410 --> 00:35:48.500 align:middle line:84% We had some-- there's a power supply I worked on 00:35:48.500 --> 00:35:51.950 align:middle line:84% with a friend for commercial application. 00:35:51.950 --> 00:36:00.150 align:middle line:84% And we designed it, and it didn't quite make the-- 00:36:00.150 --> 00:36:01.200 align:middle line:90% couldn't pass the spec. 00:36:01.200 --> 00:36:03.060 align:middle line:90% We weren't passing the spec. 00:36:03.060 --> 00:36:05.820 align:middle line:84% We went and we moved on the board 00:36:05.820 --> 00:36:10.830 align:middle line:84% a trace about 3 or 4 millimeters. 00:36:10.830 --> 00:36:13.320 align:middle line:84% Suddenly we were passing it fine. 00:36:13.320 --> 00:36:15.600 align:middle line:90% It was exactly that phenomenon. 00:36:15.600 --> 00:36:18.150 align:middle line:84% There was a noisy port that was too close to the input. 00:36:18.150 --> 00:36:20.010 align:middle line:84% And we were capacitively coupling the energy 00:36:20.010 --> 00:36:20.753 align:middle line:90% into the input. 00:36:20.753 --> 00:36:22.420 align:middle line:84% And all we had to do was move the trace. 00:36:22.420 --> 00:36:24.870 align:middle line:84% And that trace, that parasitic isn't showing up 00:36:24.870 --> 00:36:26.377 align:middle line:90% in your schematic. 00:36:26.377 --> 00:36:28.210 align:middle line:84% It's not showing up anywhere, but it's there 00:36:28.210 --> 00:36:28.960 align:middle line:90% in the real world. 00:36:28.960 --> 00:36:30.110 align:middle line:90% And you have to go find it. 00:36:30.110 --> 00:36:32.890 align:middle line:84% We moved that one plane, and we were fine. 00:36:32.890 --> 00:36:36.490 align:middle line:84% So you've got to-- you've got to spend a lot of time thinking 00:36:36.490 --> 00:36:37.200 align:middle line:90% about this. 00:36:37.200 --> 00:36:39.520 align:middle line:84% What they do in commercial EMI filters 00:36:39.520 --> 00:36:42.580 align:middle line:84% is they often will have a filter stage like this 00:36:42.580 --> 00:36:44.767 align:middle line:84% and put this inside a shielded box 00:36:44.767 --> 00:36:47.350 align:middle line:84% and then have some feedthroughs going to the next filter stage 00:36:47.350 --> 00:36:50.110 align:middle line:84% and so forth, and they can build commercial EMI filters that way. 00:36:50.110 --> 00:36:51.410 align:middle line:90% And that's great. 00:36:51.410 --> 00:36:54.250 align:middle line:84% The sad news about being a power electronics designer 00:36:54.250 --> 00:36:55.990 align:middle line:84% is usually they won't let you do that. 00:36:55.990 --> 00:36:58.450 align:middle line:84% Nobody's going to let you have shielded boxes and so 00:36:58.450 --> 00:37:00.850 align:middle line:90% forth in your converter. 00:37:00.850 --> 00:37:03.430 align:middle line:84% So you've just got to be very clever about how you lay things 00:37:03.430 --> 00:37:05.950 align:middle line:84% out and keep things from coupling magnetically and keep 00:37:05.950 --> 00:37:07.750 align:middle line:84% things from coupling capacitively. 00:37:07.750 --> 00:37:11.170 align:middle line:84% And that's often an iterative process to get it to the point 00:37:11.170 --> 00:37:13.810 align:middle line:84% where you've realized what's going to happen 00:37:13.810 --> 00:37:14.960 align:middle line:90% with all the parasitics. 00:37:14.960 --> 00:37:18.400 align:middle line:84% So that's a way to, first of all, 00:37:18.400 --> 00:37:20.770 align:middle line:84% warn you what's to come if you go do this 00:37:20.770 --> 00:37:22.300 align:middle line:90% and to not get frustrated. 00:37:22.300 --> 00:37:27.870 align:middle line:84% Just look for where nonidealities, either layout 00:37:27.870 --> 00:37:30.810 align:middle line:84% parasitics or component parasitics, 00:37:30.810 --> 00:37:38.922 align:middle line:84% are going to come bite you, and then fix those. 00:37:38.922 --> 00:37:41.130 align:middle line:84% You've also got to worry about the measurements being 00:37:41.130 --> 00:37:44.210 align:middle line:90% right too. 00:37:44.210 --> 00:37:46.400 align:middle line:84% On one project I did with a colleague, 00:37:46.400 --> 00:37:50.510 align:middle line:84% we took our prototype to a test house, 00:37:50.510 --> 00:37:52.980 align:middle line:84% one of these commercial facilities, tested it, 00:37:52.980 --> 00:37:54.360 align:middle line:90% qualified the power supply. 00:37:54.360 --> 00:37:56.550 align:middle line:84% Yes, you can plug this thing into the wall. 00:37:56.550 --> 00:37:58.380 align:middle line:90% Went on our merry way. 00:37:58.380 --> 00:38:00.410 align:middle line:90% Company calls us up. 00:38:00.410 --> 00:38:03.590 align:middle line:84% And this was a power supply that went into a water purity 00:38:03.590 --> 00:38:06.380 align:middle line:84% meter that then went on to other semiconductor equipment. 00:38:06.380 --> 00:38:10.190 align:middle line:84% And get a call from the people we designed the power 00:38:10.190 --> 00:38:12.440 align:middle line:90% supply for. 00:38:12.440 --> 00:38:14.930 align:middle line:84% Yeah, our customers came back to us 00:38:14.930 --> 00:38:21.520 align:middle line:84% and said that their equipment was failing EMI tests, 00:38:21.520 --> 00:38:24.760 align:middle line:84% and they tracked their failure of the EMI test 00:38:24.760 --> 00:38:27.950 align:middle line:84% back to this one box that we make. 00:38:27.950 --> 00:38:30.930 align:middle line:84% And of course, we had designed the EMI filter for it. 00:38:30.930 --> 00:38:31.597 align:middle line:90% What's going on? 00:38:31.597 --> 00:38:34.180 align:middle line:84% We've got to fix this problem because their customer's unhappy 00:38:34.180 --> 00:38:35.525 align:middle line:90% because they're not passing FCC. 00:38:35.525 --> 00:38:37.150 align:middle line:84% But of course, we're puzzled because we 00:38:37.150 --> 00:38:39.880 align:middle line:90% think we've passed the test. 00:38:39.880 --> 00:38:41.917 align:middle line:84% Come to track back, what happened 00:38:41.917 --> 00:38:43.750 align:middle line:84% is we took it to a commercial test facility, 00:38:43.750 --> 00:38:46.090 align:middle line:84% and they had been doing work on their test setup. 00:38:46.090 --> 00:38:48.820 align:middle line:84% And their test setup had gotten messed up and wasn't 00:38:48.820 --> 00:38:50.180 align:middle line:90% grounded properly. 00:38:50.180 --> 00:38:54.280 align:middle line:84% So basically, the results they were showing from their test 00:38:54.280 --> 00:38:56.330 align:middle line:84% setup said it was passing, and it wasn't. 00:38:56.330 --> 00:38:57.260 align:middle line:84% We took it to a different facility. 00:38:57.260 --> 00:38:57.962 align:middle line:90% It was failing. 00:38:57.962 --> 00:39:00.670 align:middle line:84% And that's when we went back and changed the design, improved it, 00:39:00.670 --> 00:39:02.330 align:middle line:84% and suddenly it passed everything. 00:39:02.330 --> 00:39:05.557 align:middle line:84% So there's a lot of layout challenges. 00:39:05.557 --> 00:39:07.390 align:middle line:84% And there's a lot of measurement challenges. 00:39:07.390 --> 00:39:09.515 align:middle line:84% And you just got to be willing to dig through them. 00:39:09.515 --> 00:39:11.290 align:middle line:84% And eventually, you'll get something. 00:39:11.290 --> 00:39:14.740 align:middle line:84% But when you're dealing with power supplies, 00:39:14.740 --> 00:39:16.690 align:middle line:84% you don't try to design something that's 00:39:16.690 --> 00:39:20.630 align:middle line:84% going to pass by a few db the specifications, 00:39:20.630 --> 00:39:22.940 align:middle line:90% like by a small margin. 00:39:22.940 --> 00:39:24.440 align:middle line:84% You want to give yourself-- you want 00:39:24.440 --> 00:39:26.310 align:middle line:90% to you want to really pass it. 00:39:26.310 --> 00:39:29.000 align:middle line:84% Because the reason is, even with all 00:39:29.000 --> 00:39:32.750 align:middle line:84% that craziness to try to keep everything identical, 00:39:32.750 --> 00:39:36.560 align:middle line:84% you're going to have variations in your design in one test 00:39:36.560 --> 00:39:40.400 align:middle line:84% setup or another or component variations in the manufacturing 00:39:40.400 --> 00:39:42.840 align:middle line:90% that aren't quite the same. 00:39:42.840 --> 00:39:45.050 align:middle line:84% So you need to give yourself a bunch of margin 00:39:45.050 --> 00:39:46.590 align:middle line:90% around the specifications. 00:39:46.590 --> 00:39:48.710 align:middle line:84% And if we had had more margin in the first design, 00:39:48.710 --> 00:39:50.502 align:middle line:84% they probably wouldn't have even come back, 00:39:50.502 --> 00:39:52.670 align:middle line:84% even though the test results were-- 00:39:52.670 --> 00:39:54.990 align:middle line:84% it wasn't our fault that the test system was wrong. 00:39:54.990 --> 00:39:58.310 align:middle line:84% Actually, we got nicely paid to debug the EMI facility's 00:39:58.310 --> 00:40:01.650 align:middle line:84% problems for them, but that was another story. 00:40:01.650 --> 00:40:10.155 align:middle line:84% So just to illustrate these phenomena-- and so what 00:40:10.155 --> 00:40:11.280 align:middle line:90% are we going to talk about? 00:40:11.280 --> 00:40:13.613 align:middle line:84% We're going to talk about a bit more about filter design 00:40:13.613 --> 00:40:16.200 align:middle line:84% and a bunch of other phenomena that come up. 00:40:16.200 --> 00:40:20.660 align:middle line:84% But I just wanted to illustrate this whole question 00:40:20.660 --> 00:40:24.500 align:middle line:84% about, first of all, component parasitics, which 00:40:24.500 --> 00:40:27.170 align:middle line:84% are very real even if you do your best layout. 00:40:27.170 --> 00:40:32.750 align:middle line:84% But also, if you do a bad job, if you accidentally 00:40:32.750 --> 00:40:35.300 align:middle line:84% put a capacitance here or you accidentally-- 00:40:35.300 --> 00:40:39.200 align:middle line:84% because you don't place this capacitor very well, 00:40:39.200 --> 00:40:41.120 align:middle line:84% you have extra lead length going to it 00:40:41.120 --> 00:40:43.832 align:middle line:84% and you insert some inductance, you'll make everything worse. 00:40:43.832 --> 00:40:46.290 align:middle line:84% So you want to be really careful about your circuit layout. 00:40:46.290 --> 00:40:48.180 align:middle line:90% And we have a demo of this. 00:40:48.180 --> 00:40:52.860 align:middle line:84% And what this is, essentially, is a set of four filters here. 00:40:52.860 --> 00:40:54.667 align:middle line:90% These are just LC filters. 00:40:54.667 --> 00:40:56.250 align:middle line:84% And I'm going to let Rafa demonstrate. 00:40:56.250 --> 00:40:59.580 align:middle line:84% And what we have is a network analyzer. 00:40:59.580 --> 00:41:04.220 align:middle line:84% What a network analyzer does is it drives at one port, 00:41:04.220 --> 00:41:07.070 align:middle line:84% like our signal ix, and then it measures 00:41:07.070 --> 00:41:11.120 align:middle line:84% into another port, which would be like our signal at the LISN. 00:41:11.120 --> 00:41:16.320 align:middle line:84% And let's see the behavior of the same filter-- 00:41:16.320 --> 00:41:18.140 align:middle line:84% it has the same physical component, just 00:41:18.140 --> 00:41:21.240 align:middle line:84% placed slightly differently-- on the performance of the filter. 00:41:21.240 --> 00:41:25.372 align:middle line:84% And with that, I'll hand it over to Rafa and let her explain. 00:41:25.372 --> 00:41:27.330 align:middle line:84% GUEST SPEAKER: Do I have to use the microphone? 00:41:27.330 --> 00:41:28.885 align:middle line:90% DAVID PERREAULT: I think so. 00:41:28.885 --> 00:41:30.410 align:middle line:90% GUEST SPEAKER: OK. 00:41:30.410 --> 00:41:33.290 align:middle line:84% All right, so, like Dave mentioned, 00:41:33.290 --> 00:41:36.770 align:middle line:84% we have four different filters with four different layout 00:41:36.770 --> 00:41:37.770 align:middle line:90% strategies. 00:41:37.770 --> 00:41:39.830 align:middle line:84% In some of them, we are being careful 00:41:39.830 --> 00:41:42.840 align:middle line:84% about both the parasitic inductance and capacitance. 00:41:42.840 --> 00:41:46.010 align:middle line:90% In some of them, only one. 00:41:46.010 --> 00:41:48.440 align:middle line:84% DAVID PERREAULT: So yeah, I'll let Rafa-- 00:41:48.440 --> 00:41:52.440 align:middle line:84% I'll let Rafa do the testing while I do the explaining. 00:41:52.440 --> 00:41:55.190 align:middle line:90% So we have four filters. 00:41:55.190 --> 00:41:58.070 align:middle line:84% The one she's going to start with, 00:41:58.070 --> 00:42:00.270 align:middle line:84% it's an LC filter just like this. 00:42:00.270 --> 00:42:03.350 align:middle line:84% And all that happened is we were sloppy about, both, 00:42:03.350 --> 00:42:05.130 align:middle line:90% how we inserted the capacitor. 00:42:05.130 --> 00:42:09.540 align:middle line:84% So there's a little extra inductance in here. 00:42:09.540 --> 00:42:14.190 align:middle line:84% And the inductor, we place traces nearby. 00:42:14.190 --> 00:42:16.500 align:middle line:84% So there's a bit of extra capacitance across it. 00:42:16.500 --> 00:42:19.520 align:middle line:84% So that's the first trace we're going to see. 00:42:19.520 --> 00:42:21.130 align:middle line:90% And let's see what this does. 00:42:21.130 --> 00:42:24.400 align:middle line:90% 00:42:24.400 --> 00:42:27.010 align:middle line:84% GUEST SPEAKER: So this is the filter where we don't-- 00:42:27.010 --> 00:42:29.830 align:middle line:84% where we are not careful about the capacitance 00:42:29.830 --> 00:42:30.960 align:middle line:90% and the inductance. 00:42:30.960 --> 00:42:32.720 align:middle line:84% DAVID PERREAULT: So this is bad-- 00:42:32.720 --> 00:42:36.200 align:middle line:84% this is parasitic capacitance and parasitic inductance. 00:42:36.200 --> 00:42:38.240 align:middle line:84% And so this is the attenuation of the filter. 00:42:38.240 --> 00:42:40.240 align:middle line:84% And we can see that the filter attenuates down-- 00:42:40.240 --> 00:42:43.050 align:middle line:84% and by the way, we're going from 1 kilohertz to 20 megahertz 00:42:43.050 --> 00:42:43.550 align:middle line:90% here. 00:42:43.550 --> 00:42:46.180 align:middle line:84% You can see even before we get very far out, 00:42:46.180 --> 00:42:50.530 align:middle line:84% the filter ceases to attenuating and then it gets bad. 00:42:50.530 --> 00:42:52.360 align:middle line:84% And these were the good filter components, 00:42:52.360 --> 00:42:55.090 align:middle line:84% but we're just getting bad results. 00:42:55.090 --> 00:42:58.090 align:middle line:84% The next one we're going to try-- 00:42:58.090 --> 00:43:01.090 align:middle line:84% what we did was we said, OK, we're 00:43:01.090 --> 00:43:07.750 align:middle line:84% going to have OK layout of the capacitor, so 00:43:07.750 --> 00:43:10.150 align:middle line:84% no parasitic inductance or as little 00:43:10.150 --> 00:43:12.040 align:middle line:84% as we can get from the component itself, 00:43:12.040 --> 00:43:17.270 align:middle line:84% but we didn't fix the bad layout across the inductor. 00:43:17.270 --> 00:43:18.880 align:middle line:90% And so here we go. 00:43:18.880 --> 00:43:23.510 align:middle line:84% And what you can see is, yeah, this 00:43:23.510 --> 00:43:27.590 align:middle line:84% is a little bit better because of what we got, 00:43:27.590 --> 00:43:28.630 align:middle line:90% but it's still terrible. 00:43:28.630 --> 00:43:31.430 align:middle line:90% 00:43:31.430 --> 00:43:35.450 align:middle line:84% The next one, we have a bad layout 00:43:35.450 --> 00:43:38.630 align:middle line:84% with parasitic inductance in series with the capacitor, 00:43:38.630 --> 00:43:41.450 align:middle line:84% but we fixed the problem with the inductor. 00:43:41.450 --> 00:43:42.880 align:middle line:90% So we're fixing one at a time. 00:43:42.880 --> 00:43:46.460 align:middle line:90% 00:43:46.460 --> 00:43:49.580 align:middle line:84% And I should say, by the way, that this test, notice we don't 00:43:49.580 --> 00:43:50.820 align:middle line:90% have a ground plane set up. 00:43:50.820 --> 00:43:52.278 align:middle line:84% We don't have all this other stuff. 00:43:52.278 --> 00:43:54.590 align:middle line:84% So this test sub wouldn't qualify 00:43:54.590 --> 00:43:57.150 align:middle line:84% for getting repeatable results in all kinds of systems. 00:43:57.150 --> 00:43:58.230 align:middle line:90% But here we go. 00:43:58.230 --> 00:44:00.440 align:middle line:90% We have again-- 00:44:00.440 --> 00:44:03.380 align:middle line:84% I didn't fix both problems, and hence 00:44:03.380 --> 00:44:05.370 align:middle line:90% I still have bad performance. 00:44:05.370 --> 00:44:08.300 align:middle line:84% So this is the frustrating thing about these things, 00:44:08.300 --> 00:44:11.120 align:middle line:84% that, yeah, you fix one thing but not the other, 00:44:11.120 --> 00:44:14.100 align:middle line:84% and you get the same kind of bad result. 00:44:14.100 --> 00:44:17.660 align:middle line:90% Now, let's do the final case. 00:44:17.660 --> 00:44:20.330 align:middle line:84% And this final case is we've taken the same components, 00:44:20.330 --> 00:44:23.810 align:middle line:84% but we've tried to do a good job of laying out the filter. 00:44:23.810 --> 00:44:26.090 align:middle line:90% So this is our attempt to-- 00:44:26.090 --> 00:44:29.750 align:middle line:84% we're still going to have component parasitics, 00:44:29.750 --> 00:44:35.490 align:middle line:84% the internal resistance and inductance of the components. 00:44:35.490 --> 00:44:38.720 align:middle line:84% But we're going to try to do a slightly better job about how 00:44:38.720 --> 00:44:40.730 align:middle line:90% we configure them on the board. 00:44:40.730 --> 00:44:42.180 align:middle line:90% So here we go. 00:44:42.180 --> 00:44:43.000 align:middle line:90% What do we get now? 00:44:43.000 --> 00:44:45.960 align:middle line:90% 00:44:45.960 --> 00:44:48.570 align:middle line:90% Well, OK, there we go. 00:44:48.570 --> 00:44:49.410 align:middle line:90% Boom. 00:44:49.410 --> 00:44:53.040 align:middle line:84% Now, you say, well, is that very good? 00:44:53.040 --> 00:44:55.170 align:middle line:90% Did we change it very much? 00:44:55.170 --> 00:44:59.640 align:middle line:84% But notice that this log magnitude 00:44:59.640 --> 00:45:04.270 align:middle line:90% scale is 20 db per division. 00:45:04.270 --> 00:45:09.480 align:middle line:84% That means the noise of the blue trace is about a factor of 100 00:45:09.480 --> 00:45:12.750 align:middle line:84% below the noise at the white trace. 00:45:12.750 --> 00:45:15.160 align:middle line:84% So that's a really big difference. 00:45:15.160 --> 00:45:17.910 align:middle line:84% And for the others, it's maybe a factor of 50 00:45:17.910 --> 00:45:18.970 align:middle line:90% or something like that. 00:45:18.970 --> 00:45:23.000 align:middle line:84% So it's a substantial amount below. 00:45:23.000 --> 00:45:26.220 align:middle line:84% And this is just one example of just one filter stage, 00:45:26.220 --> 00:45:28.530 align:middle line:84% whether you've done a good job or a bad job 00:45:28.530 --> 00:45:30.100 align:middle line:90% with the parasitic elements. 00:45:30.100 --> 00:45:32.370 align:middle line:84% So when you go out to do this stuff-- 00:45:32.370 --> 00:45:35.430 align:middle line:84% this is the whole thing I want you to take away from this-- 00:45:35.430 --> 00:45:37.830 align:middle line:84% really think about all the second order 00:45:37.830 --> 00:45:40.500 align:middle line:84% effects, the parasitics of the components, 00:45:40.500 --> 00:45:41.890 align:middle line:90% and how you use them. 00:45:41.890 --> 00:45:45.240 align:middle line:84% And we try really hard to do things 00:45:45.240 --> 00:45:47.950 align:middle line:90% like keep the noisy input-- 00:45:47.950 --> 00:45:51.030 align:middle line:84% or the noisy converter away from the quiet input 00:45:51.030 --> 00:45:52.600 align:middle line:84% we're going to plug into the wall. 00:45:52.600 --> 00:45:54.570 align:middle line:84% Some days you just don't get away from it 00:45:54.570 --> 00:45:58.620 align:middle line:84% because they tell you, no, you can't 00:45:58.620 --> 00:45:59.890 align:middle line:90% change where our connector is. 00:45:59.890 --> 00:46:03.090 align:middle line:84% And just because the connector is right next to the noisy power 00:46:03.090 --> 00:46:05.460 align:middle line:90% supply, go figure it out. 00:46:05.460 --> 00:46:06.730 align:middle line:90% And you have to live with it. 00:46:06.730 --> 00:46:10.140 align:middle line:84% But if you can, try to keep your noisy stuff far away 00:46:10.140 --> 00:46:11.580 align:middle line:90% from your quiet stuff. 00:46:11.580 --> 00:46:16.770 align:middle line:84% Try to reduce the parasitics by being very careful about how you 00:46:16.770 --> 00:46:18.630 align:middle line:84% lay out your components-- low inductance 00:46:18.630 --> 00:46:21.810 align:middle line:84% insertion, low capacitance across the components, 00:46:21.810 --> 00:46:24.240 align:middle line:84% and that will go a long way towards getting 00:46:24.240 --> 00:46:26.520 align:middle line:84% around weird, anomalous behavior and getting 00:46:26.520 --> 00:46:28.142 align:middle line:90% to be as good as you can. 00:46:28.142 --> 00:46:29.850 align:middle line:84% And then even when you do that and you're 00:46:29.850 --> 00:46:32.700 align:middle line:84% getting some worse performance, getting worse 00:46:32.700 --> 00:46:34.180 align:middle line:90% around 20 megahertz, fine. 00:46:34.180 --> 00:46:36.180 align:middle line:84% Now we'll cascade another filter stage 00:46:36.180 --> 00:46:39.040 align:middle line:84% to handle the higher frequencies. 00:46:39.040 --> 00:46:41.210 align:middle line:90% So that's about it for today. 00:46:41.210 --> 00:46:43.120 align:middle line:84% Are there any questions before we wrap up 00:46:43.120 --> 00:46:45.770 align:middle line:84% this first introduction to EMI measurements? 00:46:45.770 --> 00:46:48.600 align:middle line:90% 00:46:48.600 --> 00:46:51.950 align:middle line:90% OK, we will see you next week. 00:46:51.950 --> 00:46:56.000 align:middle line:90%