WEBVTT 1 00:00:00.160 --> 00:00:03.600 Welcome to the deep dive. Today. We're getting into something 2 00:00:03.919 --> 00:00:08.759 pretty intricate the world of radio frequency integrated circuit design, 3 00:00:09.720 --> 00:00:11.080 our FIC design. 4 00:00:11.199 --> 00:00:13.400 Yeah, this is the magic behind you know, your phone 5 00:00:13.480 --> 00:00:17.000 connecting to the network, your Wi Fi, Bluetooth, all that 6 00:00:17.079 --> 00:00:18.519 wireless stuff we take for granted. 7 00:00:18.679 --> 00:00:22.480 It really is hidden complexity, isn't it? So? Our mission today. 8 00:00:23.120 --> 00:00:27.320 We've got some textbook excerpts, core material on RFICs and 9 00:00:27.359 --> 00:00:29.640 we want to pull out the key insights. What are 10 00:00:29.679 --> 00:00:33.280 the big challenges, what's surprising, our counterintuitive about making these 11 00:00:33.359 --> 00:00:35.520 tiny radio chips work exactly? 12 00:00:35.560 --> 00:00:39.920 We'll hit the fundamentals, things like noise distortion, matching impedances, 13 00:00:40.159 --> 00:00:43.079 how components behave on silicon, and then look at how 14 00:00:43.119 --> 00:00:47.280 those fundamentals play out in the actual circuit blocks, you know, mixers, oscillators, amplifiers, 15 00:00:47.719 --> 00:00:48.399 the whole chain. 16 00:00:48.479 --> 00:00:50.799 Okay, let's lift the lid on the engineering that makes 17 00:00:50.840 --> 00:00:53.880 wireless communication possible. Where do we start? What are the 18 00:00:53.880 --> 00:00:56.079 fundamental headaches for an RFIC designer? 19 00:00:56.159 --> 00:00:58.920 Right? Well, when you're working at these frequencies gigaherds usually 20 00:00:59.240 --> 00:01:02.479 and on a tiny you're immediately juggling a whole set 21 00:01:02.520 --> 00:01:05.840 of constraints. Okay, Like what things like making sure your 22 00:01:05.840 --> 00:01:08.599 circuit works over the right frequency range, getting enough gain, 23 00:01:09.280 --> 00:01:12.000 making sure it's stable and doesn't oscillate when you don't 24 00:01:12.000 --> 00:01:16.879 want it to. Then there's noise distortion that's nonlinearity, getting 25 00:01:16.920 --> 00:01:19.680 impedance is matched upright, and just dealing with the heat, 26 00:01:19.840 --> 00:01:21.359 the power dissipation. 27 00:01:21.599 --> 00:01:23.799 And I guess these all interact. You tweak one thing 28 00:01:23.920 --> 00:01:25.079 and something else gets worse. 29 00:01:25.239 --> 00:01:28.319 Oh. Absolutely, it's a constant balancing act. Improve the gain, 30 00:01:28.760 --> 00:01:31.640 your stability might suffer, or maybe distortion goes up. That's 31 00:01:31.680 --> 00:01:34.840 why it's challenging. And maybe the most fundamental limit you 32 00:01:34.920 --> 00:01:35.680 hit is noise. 33 00:01:35.959 --> 00:01:41.079 Ah, noise, the ultimate party pooper for sensitive electronics. What's 34 00:01:41.120 --> 00:01:43.480 the source say about noise in rfices. 35 00:01:43.920 --> 00:01:46.959 Well, it starts with the basics, thermal noise, just the 36 00:01:47.079 --> 00:01:50.799 random jiggling of electrons and resistors. The noise power is 37 00:01:50.799 --> 00:01:53.400 related to temperature and resistance. You know the classic four 38 00:01:53.480 --> 00:01:57.599 KTR thing for noise voltage squared perhtz. It's white noise, 39 00:01:57.920 --> 00:01:59.840 pretty flat with frequency at least up to very high 40 00:01:59.840 --> 00:02:01.079 freequo standard stuff. 41 00:02:01.319 --> 00:02:04.359 But there was something interesting about the available noise power. 42 00:02:04.680 --> 00:02:08.400 Yes, this is really key. The absolute maximum of noise 43 00:02:08.439 --> 00:02:11.560 power you can theoretically, pull out of any resistor doesn't matter. 44 00:02:11.599 --> 00:02:14.759 What its value is is k time's t, where K 45 00:02:14.919 --> 00:02:18.599 is Boltzmann's constant and t is temperature in kelvin. 46 00:02:18.560 --> 00:02:21.360 Just kt regardless of resistance, And. 47 00:02:21.319 --> 00:02:23.800 If you consider the noise over a certain bandwidth B, 48 00:02:24.520 --> 00:02:27.719 then the total available noise power is KTB. That's like 49 00:02:28.000 --> 00:02:31.639 the fundamental floor set by physics, and crucially, an antenna 50 00:02:31.680 --> 00:02:34.719 picking up signals from the environment, its available noise power 51 00:02:34.759 --> 00:02:36.800 is also modeled as KTB. 52 00:02:36.560 --> 00:02:40.080 So that KTB is the absolute minimum noise level you're 53 00:02:40.080 --> 00:02:42.879 starting with just from the source itself, before your circuit 54 00:02:42.919 --> 00:02:44.560 even touches the signal exactly. 55 00:02:44.879 --> 00:02:47.319 But then your circuit does add its own noise. 56 00:02:47.199 --> 00:02:49.560 Right, and that's where the noise figure comes in Precisely. 57 00:02:49.800 --> 00:02:53.080 Noise figure or NF, measures how much worse the signal 58 00:02:53.080 --> 00:02:55.800 to noise ratio gets as the signal goes through your circuit. 59 00:02:56.039 --> 00:02:58.520 It's basically the ratio of the actual output noise to 60 00:02:58.560 --> 00:03:01.680 the noise you get if the circuit itself were perfectly noiseless. 61 00:03:01.800 --> 00:03:03.879 An ideal circuit has an NF of one. 62 00:03:03.879 --> 00:03:07.199 Or zero dB, and in a receiver chain with multiple 63 00:03:07.240 --> 00:03:10.319 amplifiers and mixers and stuff. How does that noise add up? 64 00:03:10.520 --> 00:03:14.560 That's critical for system design. The source explains how noise 65 00:03:14.560 --> 00:03:18.280 figures cascade. The noise from the first stage gets added directly, 66 00:03:18.879 --> 00:03:21.479 but the noise from the second stage it gets divided 67 00:03:21.520 --> 00:03:23.520 by the gain of the first stage before it adds 68 00:03:23.520 --> 00:03:26.360 to the total. Noise from the third stage gets divided 69 00:03:26.360 --> 00:03:28.280 by the gain of the first and second stages, and 70 00:03:28.319 --> 00:03:28.639 so on. 71 00:03:28.919 --> 00:03:31.280 Ah, So the first stage is hugely important. 72 00:03:31.360 --> 00:03:34.919 Domit really? That first block Usually a low noise amplifier 73 00:03:35.039 --> 00:03:37.919 or LNA pretty much sets the noise performance for the 74 00:03:37.919 --> 00:03:41.560 whole receiver. You need to have good gain and crucially 75 00:03:41.919 --> 00:03:45.599 a very low noise figure itself. If the LNA does 76 00:03:45.599 --> 00:03:49.159 its job well, the noise added by later stages like mixers, 77 00:03:49.159 --> 00:03:51.719 which can be noisier, has much less impact on the 78 00:03:51.719 --> 00:03:52.680 overall system at. 79 00:03:52.639 --> 00:03:55.800 Ff Okay, so noise limits how faint a signal you 80 00:03:55.840 --> 00:03:57.639 can hear? What about the other end, how strong a 81 00:03:57.680 --> 00:03:59.879 signal can you handle before things go haywire? 82 00:04:00.080 --> 00:04:04.479 That's distortion right, exactly linearity or the lack there of nonlinearity. 83 00:04:04.840 --> 00:04:08.360 Real circuits aren't perfectly linear. Double the input doesn't always 84 00:04:08.400 --> 00:04:10.800 exactly double the output, especially when signals get. 85 00:04:10.719 --> 00:04:13.639 Large how do designers model that predict the distortion? 86 00:04:14.000 --> 00:04:16.279 A standard way is using a mathematical trick, a power 87 00:04:16.319 --> 00:04:18.839 series expansion. You approximate the output as a sum of 88 00:04:18.920 --> 00:04:22.720 terms a constant dc offset A term proportional to the 89 00:04:22.720 --> 00:04:25.839 input that's your ideal linear game. K one a term 90 00:04:25.839 --> 00:04:28.639 proportional to the input squared at K two. You put cbe, 91 00:04:28.680 --> 00:04:30.160 K three and so on, and. 92 00:04:30.120 --> 00:04:33.120 The K two K three terms those are the troublemakers. 93 00:04:33.279 --> 00:04:36.040 They are the odd order terms like K three cause 94 00:04:36.079 --> 00:04:39.759 things like gain compression or symmetrical clipping. Even order terms 95 00:04:39.800 --> 00:04:42.240 like K two show up strongly in things like diodes 96 00:04:42.360 --> 00:04:45.000 or when there's asymmetry. But the real issue in RF 97 00:04:45.040 --> 00:04:47.000 is often what happens when you have multiple. 98 00:04:46.600 --> 00:04:48.759 Signals, like in a real wireless environment. 99 00:04:48.839 --> 00:04:51.480 Right put two different frequencies, say F one and F two, 100 00:04:51.959 --> 00:04:54.399 into a circuit with that K three term, you get 101 00:04:54.399 --> 00:04:56.560 outputs of frequencies like two F one F two and 102 00:04:56.560 --> 00:04:59.639 two F two F one. These are the infamous third 103 00:04:59.759 --> 00:05:02.240 order intermodulation products IM three. 104 00:05:02.279 --> 00:05:03.240 And why are they so bad? 105 00:05:03.639 --> 00:05:06.199 Because if F one and F two are relatively close together, 106 00:05:06.560 --> 00:05:09.639 those IM three products can land right inside your desired 107 00:05:09.720 --> 00:05:13.399 channel or spill over into the channel next door, and 108 00:05:13.439 --> 00:05:16.519 once they're generated, they're almost impossible to filter out if 109 00:05:16.560 --> 00:05:20.000 they're close to your signal. The source also mentions related 110 00:05:20.040 --> 00:05:23.959 effects like composite second order CSO and composite triple bt 111 00:05:24.199 --> 00:05:27.639 CDB when you have many many tones like in cable 112 00:05:27.680 --> 00:05:28.439 TV systems. 113 00:05:28.680 --> 00:05:31.920 Okay, so we need ways to quantify how linear or 114 00:05:31.959 --> 00:05:34.639 nonlinear a circuit is. What are the key metrics? 115 00:05:34.879 --> 00:05:37.800 One big one is the one dB compression point. Us 116 00:05:37.839 --> 00:05:39.920 are called P one dB. It's the input power level 117 00:05:40.160 --> 00:05:42.399 where the actual gain of the circuit has dropped by 118 00:05:42.439 --> 00:05:45.920 one decibel compared to its ideal small signal. Linear gain 119 00:05:46.519 --> 00:05:49.199 gives you a practical idea of the maximum signal power 120 00:05:49.199 --> 00:05:52.199 the circuit can handle before it starts significantly compressing or 121 00:05:52.199 --> 00:05:53.199 distorting the signal. 122 00:05:53.360 --> 00:05:57.120 Okay, that's about gain saturation. What about those IM three proders? Specifically? 123 00:05:57.360 --> 00:05:59.600 For that, we use the intercet point, most commonly the 124 00:05:59.639 --> 00:06:02.639 third or interset point or IP three. It can be 125 00:06:02.639 --> 00:06:05.920 specified at the input IP three or output OIP three. 126 00:06:06.199 --> 00:06:08.839 It's a theoretical point. You usually find it by plotting 127 00:06:08.879 --> 00:06:10.639 the power of the desired signal and the power of 128 00:06:10.680 --> 00:06:13.399 the I three products on a log log scale versus 129 00:06:13.439 --> 00:06:17.000 input power. The lines are straight, but with different slopes 130 00:06:17.439 --> 00:06:20.439 where they would intersect if the circuit didn't compress first. 131 00:06:20.680 --> 00:06:21.759 That's the IP three. 132 00:06:21.680 --> 00:06:24.240 So it's extrapolated, not a real operating point. 133 00:06:24.439 --> 00:06:27.720 Usually yes, but it's incredibly useful. A higher IP three 134 00:06:27.800 --> 00:06:30.079 value means the circuit is more linear, it can handle 135 00:06:30.079 --> 00:06:33.399 stronger signals before generating problematic levels of IM three distortion. 136 00:06:34.040 --> 00:06:36.879 There's also a similar metric for second order effects IP two. 137 00:06:37.079 --> 00:06:39.720 Okay, So noise sets the floor the minimum signal you 138 00:06:39.759 --> 00:06:42.519 can detect, and linearity characterized by P one, d B 139 00:06:42.639 --> 00:06:44.879 and I three sets a sort of ceiling on the 140 00:06:44.920 --> 00:06:47.839 maximum signal you can handle without too much distortion. 141 00:06:47.800 --> 00:06:50.639 Exactly, And the difference between that floor and that ceiling 142 00:06:50.800 --> 00:06:53.000 that's your dynamic range, is the range of signal powers 143 00:06:53.000 --> 00:06:54.439 the circuit can process effectively. 144 00:06:54.519 --> 00:06:54.600 Ye. 145 00:06:54.800 --> 00:06:57.319 You want a wide dynamic range obviously, so you can 146 00:06:57.360 --> 00:07:01.240 pick up weak signals, but also tolerates strong interfering signals 147 00:07:01.240 --> 00:07:03.319 without getting swamped by noise or distortion. 148 00:07:03.720 --> 00:07:06.240 And the source mentioned bandwidth affects this. 149 00:07:06.480 --> 00:07:10.160 Right, because the total noise power is KTB. A wider 150 00:07:10.199 --> 00:07:14.160 bandwidth larger B means a higher noise floor. So for 151 00:07:14.199 --> 00:07:18.560 the same circuit linearity, increasing the bandwidth directly reduces your 152 00:07:18.600 --> 00:07:20.959 dynamic range. It's a fundamental trade off. 153 00:07:21.079 --> 00:07:24.439 Makes sense. Noise and linearity huge challenges, but none of 154 00:07:24.480 --> 00:07:27.240 that matters if you can't efficiently move the signal between 155 00:07:27.279 --> 00:07:31.040 stages or connect to the antenna. Impedance matching crucial. 156 00:07:31.399 --> 00:07:34.879 Impedance matching in RF is mostly about maximizing power transfer. 157 00:07:35.160 --> 00:07:37.360 You get maximum power from a source to a load 158 00:07:37.680 --> 00:07:40.519 when the load impedance is the complex conjugate of the 159 00:07:40.560 --> 00:07:44.959 source impedance. It's also important for preventing reflections on transmission lines, 160 00:07:45.079 --> 00:07:47.360 which become a big deal at high frequencies. And sometimes 161 00:07:47.399 --> 00:07:50.920 you might match for optimal noise performance instead of maximum power. 162 00:07:51.199 --> 00:07:54.800 And that standard fifty er impedance we always hear about, Yeah. 163 00:07:54.600 --> 00:07:58.759 Fifty oms is a very common characteristic impedance for RF systems, cables, 164 00:07:59.120 --> 00:08:01.879 test equipment makes it easier to connect things together. 165 00:08:02.040 --> 00:08:04.920 The source brings up the Smith chart as the go 166 00:08:04.959 --> 00:08:07.399 to tool here. Why is it so powerful? 167 00:08:07.680 --> 00:08:12.639 The Smith chart is just ingenious. It's a graphical way 168 00:08:12.839 --> 00:08:16.680 to plot all possible impedances with a positive real part 169 00:08:16.920 --> 00:08:21.680 onto a single circular chart. It directly relates impedance to 170 00:08:21.759 --> 00:08:26.319 the reflection coefficient how much power gets reflected back from 171 00:08:26.360 --> 00:08:28.959 an impedance mismatch. The center of the chart is a 172 00:08:29.000 --> 00:08:31.120 perfect match zero reflection, and. 173 00:08:31.079 --> 00:08:33.559 You can see how adding components moves you around. 174 00:08:33.320 --> 00:08:35.879 The chart exactly. Adding a component in series move you 175 00:08:35.879 --> 00:08:39.159 along circles of constant resistance. Adding a component in parallel 176 00:08:39.240 --> 00:08:42.440 moves you along circles of constant conductance. You can literally 177 00:08:42.480 --> 00:08:44.399 trace out a path on the chart from your starting 178 00:08:44.440 --> 00:08:47.279 impedance to your target impedance usually fifty olms or the 179 00:08:47.320 --> 00:08:49.960 conjugate match by adding inductors and capacitors. 180 00:08:50.000 --> 00:08:52.960 And you prefer inductors and capacitors because they don't add noise. 181 00:08:53.360 --> 00:08:57.679 Ideally, yes, purely reactive components LS and cs just store 182 00:08:57.720 --> 00:09:01.320 and release energy. They don't dissipate power like resistors, so 183 00:09:01.360 --> 00:09:04.279 they don't add thermal noise. That's why matching networks are 184 00:09:04.279 --> 00:09:07.600 typically built using them. The source also mentions things like 185 00:09:07.639 --> 00:09:11.519 converting between series and parallel RC or RL networks, which 186 00:09:11.559 --> 00:09:14.639 involves the quality factor or a queue, and using tapped 187 00:09:14.679 --> 00:09:16.639 capacitors or transformers for matching too. 188 00:09:16.759 --> 00:09:20.440 Okay, so we need resistors, capacitors, inductors, but we need 189 00:09:20.480 --> 00:09:22.440 to build them on the silicon chip. That sounds like 190 00:09:22.440 --> 00:09:23.879 where things get really tricky. 191 00:09:24.240 --> 00:09:27.279 Well, it's arguably one of the biggest differentiators and challenges 192 00:09:27.320 --> 00:09:30.960 in rfiic design compared to say, board level RF design. 193 00:09:31.519 --> 00:09:34.240 The materials and the physical structures you have available in 194 00:09:34.279 --> 00:09:37.720 a standard silicon CMOS or bike MOS process, the thin 195 00:09:37.799 --> 00:09:41.759 metal layers, the insulating dielectrics, the silicon substrate itself, they 196 00:09:41.759 --> 00:09:43.679 aren't ideal for RF passive components. 197 00:09:43.679 --> 00:09:45.000 What kind of problems pop up? 198 00:09:45.279 --> 00:09:49.159 Well, for resistors you have sheet resistance limitations, but a 199 00:09:49.159 --> 00:09:52.799 bigger issue at RF is the skin effect. At high frequencies, 200 00:09:52.840 --> 00:09:55.799 the current crowds towards the surface of a conductor instead 201 00:09:55.799 --> 00:09:59.559 of flowing uniformly through it. This effectively reduces the cross 202 00:09:59.559 --> 00:10:03.519 sectional area the current uses, increasing the resistance. The source 203 00:10:03.559 --> 00:10:07.039 even has an example showing this effect can significantly increase 204 00:10:07.080 --> 00:10:11.960 the resistance of typical on chip metal lines at gigahertz frequencies. 205 00:10:11.519 --> 00:10:15.200 So more loss just from the wires themselves, and parasitics 206 00:10:15.279 --> 00:10:16.559 everywhere unavoidable. 207 00:10:16.679 --> 00:10:20.080 Every piece of metal has some parasitic capacitance to neighboring 208 00:10:20.080 --> 00:10:22.799 metal and to the silicon substrate below it, and every 209 00:10:22.840 --> 00:10:26.000 current loop which includes every signal path, and this return 210 00:10:26.120 --> 00:10:30.000 path has some parasitic inductance. At low frequencies, you might 211 00:10:30.000 --> 00:10:33.080 ignore these, but at RF they can totally dominate the 212 00:10:33.080 --> 00:10:34.440 intended behavior of your circuit. 213 00:10:34.600 --> 00:10:37.279 Inductors seem like a particular pain point on chip. 214 00:10:37.240 --> 00:10:40.000 They really are. You typically make them as spiral patterns 215 00:10:40.080 --> 00:10:42.799 using the top thickest metal layers, but they suffer from 216 00:10:42.840 --> 00:10:45.960 a couple of major problems. First, their quality factor Q 217 00:10:46.559 --> 00:10:47.519 is usually quite low. 218 00:10:47.799 --> 00:10:49.120 Reminds what Q means here. 219 00:10:49.279 --> 00:10:51.720 Q is basically the ratio of energy stored in the 220 00:10:51.720 --> 00:10:55.480 inductor to the energy dissipated per cycle. A low Q 221 00:10:55.720 --> 00:10:57.919 means the inductor is lossy. It acts like it has 222 00:10:57.960 --> 00:11:01.559 a significant series resistance. The source mentions typical on chip 223 00:11:01.639 --> 00:11:04.799 Q values around five or so two gigaherds. That's not 224 00:11:04.919 --> 00:11:06.759 great compared to off ship components. 225 00:11:06.799 --> 00:11:08.159 And why is low Q bad? 226 00:11:08.559 --> 00:11:12.159 It adds loss, which reduces gain in tune circuits, broadens 227 00:11:12.200 --> 00:11:15.919 filter responses, and adds noise. The other big issue is 228 00:11:15.960 --> 00:11:19.679 self resonance. Because of the parasitic capacitance between the windings 229 00:11:19.679 --> 00:11:22.759 of the spiral, the inductor only actually looks inductive up 230 00:11:22.759 --> 00:11:26.679 to a certain frequency. Above that frequency the capacitance dominates 231 00:11:26.679 --> 00:11:29.200 and the whole thing starts acting like a capacitor. The 232 00:11:29.240 --> 00:11:32.399 self resonant frequency SRF sets an upper limit on the 233 00:11:32.399 --> 00:11:35.759 inductor's useful operating range and limbs how large an inductance 234 00:11:35.840 --> 00:11:37.679 value you can practically achieve on chip. 235 00:11:37.799 --> 00:11:39.960 I saw a note about differential que being better. 236 00:11:40.120 --> 00:11:42.799 Yeah, if you build a symmetric inductor for a differential 237 00:11:42.840 --> 00:11:46.759 signal path, some of the loss mechanisms, particularly coupling to 238 00:11:46.799 --> 00:11:49.679 the substrate, can look like common mode effects and have 239 00:11:49.799 --> 00:11:53.000 less impact on the differential signal, so the que measured 240 00:11:53.000 --> 00:11:56.559 differentially can be higher than for a single ended inductor 241 00:11:56.600 --> 00:12:00.720 connected to ground. Designers also use tricks like patterned ground 242 00:12:00.759 --> 00:12:04.000 shields underneath the spiral to try and reduce substraate. 243 00:12:03.639 --> 00:12:07.080 Losses, and even the tiny wires connecting the chip to 244 00:12:07.120 --> 00:12:08.559 the outside world matter. 245 00:12:09.200 --> 00:12:12.799 The bond wires absolutely critical. This is a huge practical issue. 246 00:12:12.960 --> 00:12:15.639 A single millimeter of bond wire can easily have an 247 00:12:15.679 --> 00:12:20.080 inductance of around a nano henry at gigahertz frequencies. That's 248 00:12:20.080 --> 00:12:23.279 the significant impedance. The source highlights that you have to 249 00:12:23.320 --> 00:12:26.960 account for bondwire inductance and also the mutual inductance between 250 00:12:27.000 --> 00:12:30.840 adjacent bondwires. Designers might use multiple bond wires in parallel 251 00:12:30.840 --> 00:12:34.320 to reduce inductance, carefully place ground wires between signal wires, 252 00:12:34.440 --> 00:12:37.679 or use differential signaling, which helps cancel out some inductive effects. 253 00:12:37.840 --> 00:12:39.759 Wow, just getting a signal on and off the ship 254 00:12:39.799 --> 00:12:41.879 is an RF design problem in itself. What about the 255 00:12:41.919 --> 00:12:44.320 active devices, the transistors the work courses. 256 00:12:44.440 --> 00:12:49.279 Yeah, usually mosvats, nms and pmos and CMS processes or 257 00:12:49.320 --> 00:12:54.320 sometimes bipolar junction transistors pjts. For RF, you care about 258 00:12:54.320 --> 00:12:57.879 their small signal behavior, things like transconductance UK, which is 259 00:12:58.039 --> 00:13:00.000 how much uppercurrent change you get for an input volti 260 00:13:00.080 --> 00:13:02.440 to change and output resistance. 261 00:13:02.000 --> 00:13:03.639 ROAD and how fast they are right. 262 00:13:04.240 --> 00:13:07.039 The key figures of merit for speed are FT the 263 00:13:07.159 --> 00:13:11.159 unity current gain frequency, and fimax the maximum frequency of oscillation. 264 00:13:11.840 --> 00:13:14.159 These tell you roughly the maximum frequency at which the 265 00:13:14.159 --> 00:13:17.279 transistor can provide useful gain. They depend heavily on how 266 00:13:17.320 --> 00:13:21.480 you bias the transistor, and again on internal parasitic capacitances 267 00:13:21.639 --> 00:13:24.840 within the transistor structure. And of course, transistors add noise 268 00:13:24.879 --> 00:13:27.840 to mainly thermal noise in the channel. For mosfetz and 269 00:13:27.919 --> 00:13:30.399 shot noise associated with current flow and bgts. 270 00:13:30.519 --> 00:13:34.440 Okay, So we have these imperfect passive components noisy, fast 271 00:13:34.480 --> 00:13:37.879 but not infinitely fast transistors. How do they come together 272 00:13:37.879 --> 00:13:40.879 in the main RF circuit blocks. Let's start with mixers. 273 00:13:41.200 --> 00:13:45.159 Mixers are frequency translators. Their main job, usually in a receiver, 274 00:13:45.480 --> 00:13:47.799 is to shift the high frequency signal coming from the 275 00:13:47.840 --> 00:13:50.720 antenna of the RF signal down to a lower, more 276 00:13:50.759 --> 00:13:53.639 manageable frequency called the intermediate frequency or IF. 277 00:13:53.759 --> 00:13:54.559 And how do they do that? 278 00:13:54.600 --> 00:13:58.399 Shifting it relianes on nonlinearity. You feed the RS signal 279 00:13:58.480 --> 00:14:01.000 and another signal generated locally on the chip called a 280 00:14:01.039 --> 00:14:05.039 local oscillator or L signal into a nonlinear circuit element. 281 00:14:05.519 --> 00:14:09.840 The nonlinearity creates new frequencies, including the sum RF plus 282 00:14:10.000 --> 00:14:13.600 LO and the difference rfl O frequencies. You then filter 283 00:14:13.679 --> 00:14:15.840 the output to keep just the one you want, usually 284 00:14:15.879 --> 00:14:19.240 the difference frequency for down conversion. The source mentions the 285 00:14:19.279 --> 00:14:21.720 Gilbert cell, which is a very common and clever mixer 286 00:14:21.720 --> 00:14:23.440 circuit topology used in ICs. 287 00:14:23.639 --> 00:14:25.519 What's a major headache for mixer design? 288 00:14:25.879 --> 00:14:28.559 One of the biggest is the image frequency. See for 289 00:14:28.600 --> 00:14:31.799 a given low frequency and a desired IF frequency is 290 00:14:31.840 --> 00:14:35.159 a IF R fl O. There's another RF frequency that 291 00:14:35.159 --> 00:14:37.519 will also mix down at the same if. That's the 292 00:14:37.519 --> 00:14:41.000 image frequency located at LO plus IF or loif if 293 00:14:41.120 --> 00:14:45.000 rfzl IF. If there's a signal at that image frequency, 294 00:14:45.200 --> 00:14:47.320 it will overlap with your desired signal at the mixer 295 00:14:47.360 --> 00:14:48.759 output and cause interference. 296 00:14:49.039 --> 00:14:50.679 Ah. So you need to get rid of the image 297 00:14:50.720 --> 00:14:52.080 signal before it hits the mixer. 298 00:14:52.240 --> 00:14:55.480 Ideally yes, with filtering yeah, or you can use specific 299 00:14:55.559 --> 00:14:59.679 mixer architectures called image reject mixers that inherently cancel out 300 00:14:59.679 --> 00:15:03.159 the image signal. Good image rejection IR is a really 301 00:15:03.200 --> 00:15:07.200 important spec for a receiver. Mixers also add noise, of course, 302 00:15:07.279 --> 00:15:10.799 both thermal noise and noise from the LO signal itself. 303 00:15:11.120 --> 00:15:14.240 Mixing down the source points out things like using a strong, 304 00:15:14.480 --> 00:15:17.879 sharp edged LLO signal can actually help improve both noise 305 00:15:17.960 --> 00:15:19.960 and linearity in some mixer types. 306 00:15:20.039 --> 00:15:22.679 Okay, moving on to that LO signal generator itself. The 307 00:15:22.720 --> 00:15:25.840 oscillator needs to be super stable and clean, right. 308 00:15:25.840 --> 00:15:28.799 Absolutely vital. An oscillator's job is to generate a precise, 309 00:15:29.159 --> 00:15:33.440 stable frequency reference. The basic principle is positive feedback. You 310 00:15:33.480 --> 00:15:35.840 take an amplifier loop its output back to its input 311 00:15:35.840 --> 00:15:38.919 through a frequency selective network like a resonant tank, and 312 00:15:39.039 --> 00:15:41.559 if at one specific frequency the gain around that loop 313 00:15:41.600 --> 00:15:44.360 is exactly one and the total phase shift is zero 314 00:15:44.519 --> 00:15:47.360 or three hundred and sixty degrees, the circuit will oscillate 315 00:15:47.360 --> 00:15:50.799 spontaneously at that frequency. That's the Barkhausen criterion and. 316 00:15:50.759 --> 00:15:52.919 This idea of negative resistance right. 317 00:15:53.279 --> 00:15:55.840 The frequency is usually set by a resonant tank circuit, 318 00:15:55.919 --> 00:15:59.679 typically an inductor and capacitor. But as we discussed on hip, 319 00:15:59.759 --> 00:16:03.519 tanks are lossy. They have resistance representing by a low queue. 320 00:16:04.799 --> 00:16:08.120 To sustain oscillation, the active part of the oscillator circuit 321 00:16:08.200 --> 00:16:11.240 needs to effectively cancel out that loss. It does this 322 00:16:11.360 --> 00:16:14.799 by providing what looks like a negative resistance, which injects 323 00:16:14.960 --> 00:16:17.679 energy into the tank to perfectly compensate for the energy 324 00:16:17.720 --> 00:16:19.879 being dissipated by the tank's positive resistance. 325 00:16:20.159 --> 00:16:22.799 So you need enough gain or negative resistance to get 326 00:16:22.799 --> 00:16:26.159 it started. What stops the oscillation amplitude running. 327 00:16:25.879 --> 00:16:29.519 Away nonlinearity again, but this time it's useful. As the 328 00:16:29.559 --> 00:16:33.120 oscillation starts and the signal amplitude builds up, the active 329 00:16:33.159 --> 00:16:36.240 devices in the oscillator naturally start to saturate or compress. 330 00:16:36.600 --> 00:16:39.279 This reduces the effective gain or the magnitude of the 331 00:16:39.279 --> 00:16:42.919 negative resistance. The amplitude stabilizes exactly at the point where 332 00:16:42.919 --> 00:16:45.919 the loop gain becomes equal to one. The source also 333 00:16:45.960 --> 00:16:49.039 mentions some subtle effects like bias point shifts during startup 334 00:16:49.120 --> 00:16:51.879 due to rectification of harmonics from this nonlinearity. 335 00:16:51.960 --> 00:16:56.120 Okay, stable frequency, stable amplitude. But the bane of oscillator 336 00:16:56.159 --> 00:16:58.360 designers is phase noise, isn't it? Oh? 337 00:16:58.440 --> 00:17:02.519 Absolutely? Phase noise is probably the most critical performance metric 338 00:17:02.600 --> 00:17:07.079 for many oscillators. It represents tiny, random fluctuations in the 339 00:17:07.119 --> 00:17:10.279 phase of the oscillator's signal over time, which is equivalent 340 00:17:10.319 --> 00:17:14.160 to fluctuations in its instantaneous frequency. It's noise, but it 341 00:17:14.200 --> 00:17:16.720 shows up as sidebands, sort of like a skirt of 342 00:17:16.759 --> 00:17:19.240 noise power around the perfect single tone you wanted. 343 00:17:19.359 --> 00:17:21.079 And why is phase noise so bad? 344 00:17:21.519 --> 00:17:24.359 In a receiver, the lo phase noise mixes with strong 345 00:17:24.400 --> 00:17:28.160 interfering signals and spreads their energy into your desired channel, 346 00:17:28.359 --> 00:17:31.720 potentially drowning out your weak signal. In a transmitter, it 347 00:17:31.759 --> 00:17:35.720 spreads your transmitted power into adjacent channels, causing interference to others. 348 00:17:36.039 --> 00:17:39.519 It fundamentally limits channel spacing and data rates In wireless systems. 349 00:17:39.599 --> 00:17:41.559 What determines how bad the phase noise is. 350 00:17:41.839 --> 00:17:45.880 Several things, often summarized by Lesen's model. A higher Q 351 00:17:46.119 --> 00:17:49.160 resonator tank is much better for phase noise. That's a 352 00:17:49.160 --> 00:17:52.160 big reason why off chip crystal oscillators are so much 353 00:17:52.240 --> 00:17:55.759 cleaner than on chip elc oscillators. More power in the 354 00:17:55.759 --> 00:17:59.160 oscillator generally helps up to a point, the noise figure 355 00:17:59.160 --> 00:18:03.200 of the active device matters, and crucially, low frequency noise 356 00:18:03.200 --> 00:18:05.640 sources like one half noise or flicker noise from the 357 00:18:05.680 --> 00:18:08.480 transistors or a noise on the control voltage used to 358 00:18:08.519 --> 00:18:12.279 tune the oscillator frequency for a character can get up 359 00:18:12.319 --> 00:18:17.200 converted by the oscillator's nonlinearity and significantly degrade the phase noise, 360 00:18:17.599 --> 00:18:20.799 especially close into the carrier frequency. The source had a 361 00:18:20.799 --> 00:18:23.720 good example highlighting the impact of noise on the tuning line. 362 00:18:23.799 --> 00:18:26.480 Any tricks to improve it, besides getting a higher que. 363 00:18:26.480 --> 00:18:29.559 Careful biasing and device sizing to minimize flicker noise helps. 364 00:18:30.039 --> 00:18:33.079 Some architectures are inherently better than others, and techniques like 365 00:18:33.119 --> 00:18:36.759 automatic amplitude control aec loops can help stabilize the operating 366 00:18:36.759 --> 00:18:39.400 point and make the phase noise more robust. To variations. 367 00:18:39.599 --> 00:18:42.160 All right, let's touch on filters and RFICs. What are 368 00:18:42.200 --> 00:18:43.039 their main jobs? 369 00:18:43.240 --> 00:18:46.759 Filters are the gatekeepers of frequency. They select the frequencies 370 00:18:46.799 --> 00:18:49.799 you want and reject the ones you don't, so defining 371 00:18:49.799 --> 00:18:52.359 the channel bandwidth, getting rid of strong out of van 372 00:18:52.440 --> 00:18:56.319 blockers or interferers, and as we mentioned, providing that critical 373 00:18:56.359 --> 00:18:57.880 image rejection before the mixer. 374 00:18:58.200 --> 00:19:00.720 And how are they built on chip? Often using those 375 00:19:00.720 --> 00:19:02.160 same LC resonators. 376 00:19:02.359 --> 00:19:05.119 Yeah, simple filters often use resonators. For example, using a 377 00:19:05.160 --> 00:19:08.359 parallel LC tank as load for an amplifier creates a 378 00:19:08.359 --> 00:19:11.599 band pass filter response around the resonant frequency. You can 379 00:19:11.640 --> 00:19:15.359 also use series or parallel resonators to create notches or 380 00:19:15.400 --> 00:19:19.440 band stop filters. The source showed a neat trick putting 381 00:19:19.440 --> 00:19:22.240 a series LC resonator in the emitter or source path 382 00:19:22.319 --> 00:19:26.160 of an LNA at its resonant frequency tuned to the 383 00:19:26.200 --> 00:19:29.480 image frequency. It presents very high impedance which kills the 384 00:19:29.480 --> 00:19:33.359 amplifier's gain, specifically at the image frequency, creating a notch 385 00:19:33.359 --> 00:19:34.640 filter right at the input. 386 00:19:35.160 --> 00:19:37.759 What are the challenges with on ship filters? Besides the 387 00:19:37.799 --> 00:19:38.880 low queue. 388 00:19:38.559 --> 00:19:41.240 Stability can be an issue, especially if you try to 389 00:19:41.240 --> 00:19:44.640 make active filters with game but maybe the biggest practical 390 00:19:44.680 --> 00:19:49.000 problem is sensitivity to process variations. Those on chip, inductor 391 00:19:49.000 --> 00:19:51.480 and capacitor values can vary quite a bit from chip 392 00:19:51.480 --> 00:19:54.559 to chip or krong a wafer. This means the center 393 00:19:54.599 --> 00:19:57.720 of frequency, bandwidth and rejection depth of your filter can 394 00:19:57.759 --> 00:20:02.000 also vary significantly, making it hard to meet tight specifications reliably. 395 00:20:02.720 --> 00:20:05.519 The source had an example showing how tolerance affects image 396 00:20:05.519 --> 00:20:06.319 rejection depth. 397 00:20:06.480 --> 00:20:10.799 Okay, last, big block power amplifiers PAS getting the signal 398 00:20:10.839 --> 00:20:11.880 boosted up to transmit. 399 00:20:12.319 --> 00:20:14.640 A PA's job is to take the relatively weak signal 400 00:20:14.640 --> 00:20:17.720 from the preceding stages and deliver a hefty amount of power, 401 00:20:17.839 --> 00:20:20.799