1 00:00:00,090 --> 00:00:02,490 The following content is provided under a Creative 2 00:00:02,490 --> 00:00:04,030 Commons license. 3 00:00:04,030 --> 00:00:06,330 Your support will help MIT OpenCourseWare 4 00:00:06,330 --> 00:00:10,720 continue to offer high-quality educational resources for free. 5 00:00:10,720 --> 00:00:13,320 To make a donation or view additional materials 6 00:00:13,320 --> 00:00:17,280 from hundreds of MIT courses, visit MIT OpenCourseWare 7 00:00:17,280 --> 00:00:18,450 at ocw.mit.edu. 8 00:00:25,988 --> 00:00:28,470 DENNIS FREEMAN: Hello. 9 00:00:28,470 --> 00:00:31,590 So welcome to the last lecture in 003. 10 00:00:31,590 --> 00:00:34,350 So there are more lectures scheduled. 11 00:00:34,350 --> 00:00:35,940 But I don't believe in trying to cram 12 00:00:35,940 --> 00:00:39,780 new material in the last couple of hours before the exam. 13 00:00:39,780 --> 00:00:43,770 So no new material after today, although it will be helpful 14 00:00:43,770 --> 00:00:46,200 if you continue to try to work on problems, 15 00:00:46,200 --> 00:00:49,510 and try to internalize what we did more recently. 16 00:00:49,510 --> 00:00:50,830 That's the idea. 17 00:00:50,830 --> 00:00:52,890 So I'm not trying to pump new information in. 18 00:00:52,890 --> 00:00:55,260 That doesn't mean that there aren't things 19 00:00:55,260 --> 00:00:58,450 that you should be doing. 20 00:00:58,450 --> 00:01:00,900 But after today, no new information. 21 00:01:00,900 --> 00:01:03,900 In fact, today there is no new information. 22 00:01:03,900 --> 00:01:05,760 There will be a session on Tuesday. 23 00:01:05,760 --> 00:01:10,410 But the idea is primarily to get feedback from you people. 24 00:01:10,410 --> 00:01:12,960 As I mentioned last time, we've changed a lot of things. 25 00:01:12,960 --> 00:01:15,000 We change a lot of things every term. 26 00:01:15,000 --> 00:01:17,280 And our primary source of information 27 00:01:17,280 --> 00:01:19,110 is the end of term surveys. 28 00:01:19,110 --> 00:01:19,950 Tell us what works. 29 00:01:19,950 --> 00:01:20,910 Tell us what didn't work. 30 00:01:20,910 --> 00:01:21,750 Tell us what you liked. 31 00:01:21,750 --> 00:01:22,958 Tell us what you didn't like. 32 00:01:22,958 --> 00:01:24,900 If you didn't like the electronic submission 33 00:01:24,900 --> 00:01:26,917 of homework, tell us. 34 00:01:26,917 --> 00:01:28,500 If you didn't like the tutor, tell us. 35 00:01:28,500 --> 00:01:31,320 If you would rather have individual office hours instead 36 00:01:31,320 --> 00:01:33,900 of block office hours, it's a good chance for you 37 00:01:33,900 --> 00:01:37,560 to tell us what you think. 38 00:01:37,560 --> 00:01:39,960 And it's especially important that you fill out 39 00:01:39,960 --> 00:01:42,180 the online subject evaluation. 40 00:01:42,180 --> 00:01:45,390 So please, if you have a laptop with a certificate on it, 41 00:01:45,390 --> 00:01:47,880 feel free to bring it to lecture next time. 42 00:01:47,880 --> 00:01:49,960 That's one of the main ideas. 43 00:01:49,960 --> 00:01:56,360 But we'll also spend some time just discussing 44 00:01:56,360 --> 00:02:00,414 what you would suggest for improvements for next time. 45 00:02:00,414 --> 00:02:01,330 Questions or comments? 46 00:02:04,690 --> 00:02:07,710 OK, so today is mostly just for fun. 47 00:02:07,710 --> 00:02:10,419 Today is mostly just to tell you about the way 48 00:02:10,419 --> 00:02:14,410 6.003 plays out in technologies that you 49 00:02:14,410 --> 00:02:17,050 might have some interest in. 50 00:02:17,050 --> 00:02:20,470 And I chose a technology that I am particularly interested in. 51 00:02:20,470 --> 00:02:23,350 Because as I probably have mentioned in the past, 52 00:02:23,350 --> 00:02:26,440 like many of you, I was afflicted with an addiction 53 00:02:26,440 --> 00:02:28,300 with music at about your age. 54 00:02:28,300 --> 00:02:29,350 I recovered. 55 00:02:29,350 --> 00:02:30,160 You will too. 56 00:02:32,760 --> 00:02:34,720 But I was very interested in it. 57 00:02:34,720 --> 00:02:36,340 It was a very hot topic. 58 00:02:36,340 --> 00:02:40,270 And I want to talk about some of the things that 003 has done 59 00:02:40,270 --> 00:02:44,140 to make that industry very different today from what it 60 00:02:44,140 --> 00:02:48,430 was when I was an undergraduate. 61 00:02:48,430 --> 00:02:54,850 It may not surprise you, but it surprises me 62 00:02:54,850 --> 00:02:58,300 that the technology that we used was really not 63 00:02:58,300 --> 00:03:00,250 very different from the technology 64 00:03:00,250 --> 00:03:02,290 that the inventor invented. 65 00:03:02,290 --> 00:03:06,397 So who invented the phonograph? 66 00:03:06,397 --> 00:03:07,480 You can look at the notes. 67 00:03:07,480 --> 00:03:08,688 You're supposed to know that. 68 00:03:12,010 --> 00:03:13,270 Just shout. 69 00:03:13,270 --> 00:03:15,005 Makes me feel better. 70 00:03:15,005 --> 00:03:15,880 AUDIENCE: [INAUDIBLE] 71 00:03:15,880 --> 00:03:17,005 DENNIS FREEMAN: Thank you-- 72 00:03:17,005 --> 00:03:17,950 Edison. 73 00:03:17,950 --> 00:03:23,915 So Edison was, of course, a genius and a prolific inventor. 74 00:03:23,915 --> 00:03:25,540 At the time he invented the phonograph, 75 00:03:25,540 --> 00:03:28,600 he was interested in two things that he 76 00:03:28,600 --> 00:03:31,840 was trying to draw relations between, 77 00:03:31,840 --> 00:03:34,330 telegraphy and telephony. 78 00:03:34,330 --> 00:03:35,770 We talked about telegraphy before. 79 00:03:35,770 --> 00:03:39,910 Telegraphy, telegraph, was a way of getting messages 80 00:03:39,910 --> 00:03:41,890 point to point. 81 00:03:41,890 --> 00:03:45,100 So person A wanted to get a message to person B. 82 00:03:45,100 --> 00:03:48,460 And it would be transmitted via the telegraph system. 83 00:03:48,460 --> 00:03:50,080 Now I kind of oversimplified the way 84 00:03:50,080 --> 00:03:51,496 the telegraph system worked when I 85 00:03:51,496 --> 00:03:52,864 talked about it the last time. 86 00:03:52,864 --> 00:03:54,530 It really worked a lot more like message 87 00:03:54,530 --> 00:03:57,714 passing than you might have thought. 88 00:03:57,714 --> 00:03:59,380 What would happen is that there wouldn't 89 00:03:59,380 --> 00:04:02,440 be a direct connection between person A and person B. 90 00:04:02,440 --> 00:04:05,930 It might have to go through six relays. 91 00:04:05,930 --> 00:04:08,270 So in order to get to California, 92 00:04:08,270 --> 00:04:10,850 it might be important that a message that 93 00:04:10,850 --> 00:04:13,880 originated in Boston would go to New York, 94 00:04:13,880 --> 00:04:16,730 and from New York to Pennsylvania, 95 00:04:16,730 --> 00:04:20,000 from Pennsylvania to Chicago, et cetera. 96 00:04:20,000 --> 00:04:24,080 So there may not be a direct route over long distances. 97 00:04:24,080 --> 00:04:26,930 So the way that was done, you would take your message 98 00:04:26,930 --> 00:04:32,990 to the telegraph office A. The operator would key it in. 99 00:04:32,990 --> 00:04:35,790 It would get written down by the receiver, 100 00:04:35,790 --> 00:04:39,750 say, in New York, who would listen to the message, 101 00:04:39,750 --> 00:04:44,140 write it out, basically reproducing your message there, 102 00:04:44,140 --> 00:04:47,990 then key it in to the next relay. 103 00:04:47,990 --> 00:04:50,900 So it might actually be keyed in a dozen times 104 00:04:50,900 --> 00:04:52,730 before it hits the final destination. 105 00:04:52,730 --> 00:04:55,370 And that was tedious, laborious, hard. 106 00:04:55,370 --> 00:04:59,630 And Edison was interested in fixing that. 107 00:04:59,630 --> 00:05:02,990 And to do so, he invented a paper tape system, 108 00:05:02,990 --> 00:05:06,400 so that the receiver wouldn't need to listen it. 109 00:05:06,400 --> 00:05:10,539 They would just punch out paper tape as the clicks came in. 110 00:05:10,539 --> 00:05:12,080 So as the dots and dashes came in it, 111 00:05:12,080 --> 00:05:13,700 it would make holes in paper. 112 00:05:13,700 --> 00:05:15,350 The holes in paper, then, could be 113 00:05:15,350 --> 00:05:18,050 used to transmit to the next station. 114 00:05:18,050 --> 00:05:20,510 So Edison did that. 115 00:05:20,510 --> 00:05:22,040 He was interested in thinking about 116 00:05:22,040 --> 00:05:23,600 whether you could do the same sort of thing 117 00:05:23,600 --> 00:05:25,880 with other kinds of signals besides telegraph signals. 118 00:05:25,880 --> 00:05:28,070 And that's the origin of the phonograph. 119 00:05:28,070 --> 00:05:31,580 He was interested in, could you send a voice message 120 00:05:31,580 --> 00:05:35,570 to a relay, have it automatically recorded, 121 00:05:35,570 --> 00:05:39,704 so that it could then be rebroadcast over the next link? 122 00:05:39,704 --> 00:05:41,870 So he was kind of imagining a telephone network that 123 00:05:41,870 --> 00:05:46,610 was based on the same principle that the telegraph network had 124 00:05:46,610 --> 00:05:47,720 been. 125 00:05:47,720 --> 00:05:50,960 So the idea was to try to record an audio signal 126 00:05:50,960 --> 00:05:55,820 of the type that would come across a telephone wire. 127 00:05:55,820 --> 00:05:59,690 So he got together with his aide, Batchelor, 128 00:05:59,690 --> 00:06:01,100 and made this drawing. 129 00:06:01,100 --> 00:06:04,460 This was 1877. 130 00:06:04,460 --> 00:06:06,440 And 30 hours later-- 131 00:06:06,440 --> 00:06:09,080 kind of rapid prototyping for the time-- 132 00:06:09,080 --> 00:06:11,750 his machinist made this. 133 00:06:11,750 --> 00:06:13,610 So that's the first phonograph. 134 00:06:13,610 --> 00:06:21,230 The idea was you wrap that disk with a piece of wax paper. 135 00:06:21,230 --> 00:06:23,480 You turn the crank. 136 00:06:23,480 --> 00:06:29,320 As you turn the crank, this thing is connected to a screw, 137 00:06:29,320 --> 00:06:32,450 so that it's moving this way at the same time this big thing is 138 00:06:32,450 --> 00:06:34,820 going around that way. 139 00:06:34,820 --> 00:06:38,300 This is a diaphragm, so that you shout in here. 140 00:06:38,300 --> 00:06:40,710 And there's a needle connected on the other end, 141 00:06:40,710 --> 00:06:43,190 so that as you shout, the needle gets 142 00:06:43,190 --> 00:06:48,500 moved with audio frequencies contained in your voice. 143 00:06:48,500 --> 00:06:50,300 So how do you do this, then? 144 00:06:50,300 --> 00:06:52,970 So you grab the handle. 145 00:06:52,970 --> 00:06:55,700 You put a new sheet of wax paper on it. 146 00:06:55,700 --> 00:06:56,810 You grab the handle. 147 00:06:56,810 --> 00:07:01,200 You turn at a constant speed, and shout. 148 00:07:01,200 --> 00:07:05,300 OK, it took some coordination. 149 00:07:05,300 --> 00:07:09,140 The remarkable thing is it worked. 150 00:07:09,140 --> 00:07:12,986 So Edison said, Mary had a little lamb. 151 00:07:12,986 --> 00:07:17,390 And on replaying it, the device said, Mary had a little lamb. 152 00:07:17,390 --> 00:07:20,020 The very first try, it worked. 153 00:07:20,020 --> 00:07:22,220 Now it didn't work all that great. 154 00:07:22,220 --> 00:07:25,535 What do you suppose happened? 155 00:07:25,535 --> 00:07:26,970 AUDIENCE: [INAUDIBLE] 156 00:07:26,970 --> 00:07:27,886 DENNIS FREEMAN: Noise? 157 00:07:27,886 --> 00:07:29,538 Of course, there was noise, yes. 158 00:07:29,538 --> 00:07:30,900 AUDIENCE: [INAUDIBLE] 159 00:07:30,900 --> 00:07:32,525 DENNIS FREEMAN: Frequency was terrible. 160 00:07:34,669 --> 00:07:35,710 Anything more disastrous? 161 00:07:35,710 --> 00:07:36,340 Yeah? 162 00:07:36,340 --> 00:07:37,510 AUDIENCE: [INAUDIBLE] 163 00:07:37,510 --> 00:07:39,509 DENNIS FREEMAN: Cranking is not at all constant. 164 00:07:39,509 --> 00:07:42,520 So you can make it sound like Alvin the chipmunk if you like. 165 00:07:45,980 --> 00:07:46,670 Something worse. 166 00:07:50,120 --> 00:07:53,734 How long do you think the recording lasted? 167 00:07:53,734 --> 00:07:55,210 AUDIENCE: [INAUDIBLE]. 168 00:07:55,210 --> 00:07:57,586 DENNIS FREEMAN: Yeah, about once. 169 00:07:57,586 --> 00:08:01,262 AUDIENCE: [INAUDIBLE]. 170 00:08:01,262 --> 00:08:02,720 DENNIS FREEMAN: So the problem was, 171 00:08:02,720 --> 00:08:07,190 to get the energy back out, you have to-- 172 00:08:07,190 --> 00:08:12,500 so after recording-- so clean sheet of wax paper, 173 00:08:12,500 --> 00:08:15,760 crank, crank, crank, yell, scream, 174 00:08:15,760 --> 00:08:18,340 and now reset the needle to the beginning. 175 00:08:18,340 --> 00:08:21,010 And now crank without screaming. 176 00:08:21,010 --> 00:08:22,990 In order to get the message back out, 177 00:08:22,990 --> 00:08:25,960 you had to push on the needle, so that it would follow 178 00:08:25,960 --> 00:08:29,100 the indentations on the wax. 179 00:08:29,100 --> 00:08:31,860 But if you pushed hard enough for it to follow it, 180 00:08:31,860 --> 00:08:34,559 what would happen? 181 00:08:34,559 --> 00:08:37,130 It would erase the message. 182 00:08:37,130 --> 00:08:40,429 So it was kind of a destructive read out. 183 00:08:40,429 --> 00:08:42,799 It didn't last very long. 184 00:08:42,799 --> 00:08:47,850 Because the act of reading it back out tended to erase it. 185 00:08:47,850 --> 00:08:51,960 So he redid the thing with tinfoil. 186 00:08:51,960 --> 00:08:54,250 That worked much better. 187 00:08:54,250 --> 00:08:57,780 And in fact, it caught on like overnight. 188 00:08:57,780 --> 00:09:00,060 So within two years, he was invited 189 00:09:00,060 --> 00:09:01,770 to the National Academies of Sciences 190 00:09:01,770 --> 00:09:03,180 to give a talk on this. 191 00:09:03,180 --> 00:09:06,000 That's him posing at the National Academy of Sciences 192 00:09:06,000 --> 00:09:09,210 with his then phonograph. 193 00:09:09,210 --> 00:09:12,460 They loved it. 194 00:09:12,460 --> 00:09:15,260 Now despite the fact they loved it, 195 00:09:15,260 --> 00:09:17,750 nobody knew what to do with it. 196 00:09:17,750 --> 00:09:20,570 Everybody thought it was just the neatest thing in the world. 197 00:09:20,570 --> 00:09:22,550 But they didn't have a clue what to do with it. 198 00:09:22,550 --> 00:09:25,590 And of course, Edison, one, was a bit self-promoting. 199 00:09:25,590 --> 00:09:28,940 So he wrote articles about it, and tried to explain. 200 00:09:28,940 --> 00:09:31,160 In the patent, he listed about seven or eight things 201 00:09:31,160 --> 00:09:32,990 you could do with this thing. 202 00:09:32,990 --> 00:09:37,070 Today they sound a little far-fetched. 203 00:09:37,070 --> 00:09:42,080 Recording important historical addresses, 204 00:09:42,080 --> 00:09:44,870 medical transcription-- 205 00:09:44,870 --> 00:09:46,780 the one that seems most-- 206 00:09:46,780 --> 00:09:49,610 I don't know-- sort of gruesome to me, 207 00:09:49,610 --> 00:09:52,610 recording the last will and testament of a person 208 00:09:52,610 --> 00:09:55,070 as they're dying. 209 00:09:55,070 --> 00:09:59,230 Now that sounds gruesome to me. 210 00:09:59,230 --> 00:10:01,280 Then imagine what it was like back then. 211 00:10:01,280 --> 00:10:05,260 Back then, nobody had ever heard somebody's voice 212 00:10:05,260 --> 00:10:06,040 who was not alive. 213 00:10:10,800 --> 00:10:13,140 The technology didn't exist. 214 00:10:13,140 --> 00:10:15,090 So all of the sudden, he's proposing 215 00:10:15,090 --> 00:10:17,630 to record the dying breath of, say-- 216 00:10:17,630 --> 00:10:19,970 it kind of mortified everybody. 217 00:10:19,970 --> 00:10:21,960 What Edison didn't like was the idea 218 00:10:21,960 --> 00:10:24,420 that it would be used for music. 219 00:10:24,420 --> 00:10:25,830 That was trivial. 220 00:10:25,830 --> 00:10:28,410 He was doing important science. 221 00:10:28,410 --> 00:10:31,350 So he didn't like the idea that this would be used for music. 222 00:10:31,350 --> 00:10:33,900 And so he downplayed that his entire life. 223 00:10:33,900 --> 00:10:39,300 Nevertheless, the commercial success was music. 224 00:10:39,300 --> 00:10:45,510 He was trying to promote this as a system for recording speech, 225 00:10:45,510 --> 00:10:49,350 medical transcription, buying breaths, historical talks. 226 00:10:49,350 --> 00:10:50,520 That didn't really catch on. 227 00:10:50,520 --> 00:10:53,520 This was a playback device. 228 00:10:53,520 --> 00:10:56,160 This really did catch on. 229 00:10:56,160 --> 00:10:58,050 This was about 60 years later. 230 00:10:58,050 --> 00:11:00,886 It's virtually identical. 231 00:11:00,886 --> 00:11:02,760 It became a piece of furniture that everybody 232 00:11:02,760 --> 00:11:05,460 had in their house. 233 00:11:05,460 --> 00:11:08,100 The cylinders had gone back to wax. 234 00:11:08,100 --> 00:11:10,410 Alexander Graham Bell had worked on the problem. 235 00:11:10,410 --> 00:11:14,010 He figured out how to make long-lasting wax. 236 00:11:14,010 --> 00:11:17,920 This crank cranked a spring. 237 00:11:17,920 --> 00:11:20,205 So you could crank it up, but over time it would play. 238 00:11:20,205 --> 00:11:22,330 So you didn't need to go at exactly the right speed 239 00:11:22,330 --> 00:11:24,090 anymore-- 240 00:11:24,090 --> 00:11:25,920 a number of big improvements. 241 00:11:25,920 --> 00:11:30,660 The cylinders held two minutes of recording. 242 00:11:30,660 --> 00:11:33,780 So they weren't exactly long-playing. 243 00:11:33,780 --> 00:11:42,620 OK, this was the same model of phonograph I had in college. 244 00:11:42,620 --> 00:11:46,680 And it's actually remarkably similar. 245 00:11:46,680 --> 00:11:50,340 So there's some differences. 246 00:11:50,340 --> 00:11:52,560 There was no crank. 247 00:11:52,560 --> 00:11:55,140 We had electricity. 248 00:11:55,140 --> 00:11:56,265 It had a motor. 249 00:11:56,265 --> 00:11:57,660 So the motor propelled it. 250 00:11:57,660 --> 00:11:59,100 It was no longer a cylinder. 251 00:11:59,100 --> 00:12:01,020 Now it's a desk. 252 00:12:01,020 --> 00:12:01,815 There was no screw. 253 00:12:04,700 --> 00:12:08,690 So how does the needle track the grooves if there's no screw? 254 00:12:11,325 --> 00:12:12,710 AUDIENCE: [INAUDIBLE] 255 00:12:12,710 --> 00:12:14,085 DENNIS FREEMAN: There's a spiral. 256 00:12:14,085 --> 00:12:15,409 So the records had a spiral. 257 00:12:15,409 --> 00:12:15,950 That's right. 258 00:12:19,760 --> 00:12:22,400 This was the thing that read the grooves. 259 00:12:22,400 --> 00:12:27,020 What held that in the groove? 260 00:12:27,020 --> 00:12:27,850 It was gravity. 261 00:12:30,500 --> 00:12:33,080 So it was the same kind of idea. 262 00:12:33,080 --> 00:12:41,150 You had to push on the needle to hold the needle in the groove, 263 00:12:41,150 --> 00:12:45,020 so that the needle could report the vibrations that 264 00:12:45,020 --> 00:12:48,110 were recorded there. 265 00:12:48,110 --> 00:12:54,515 But this was much classier than the version in Edison's time. 266 00:12:54,515 --> 00:13:01,670 And in fact, you could tell how serious the audio file was 267 00:13:01,670 --> 00:13:04,604 by the different kinds of performance enhancements 268 00:13:04,604 --> 00:13:05,520 of this sort of thing. 269 00:13:05,520 --> 00:13:09,660 So for example, this little knob, 270 00:13:09,660 --> 00:13:12,995 this controls the weight, the force 271 00:13:12,995 --> 00:13:14,711 that's placed on the needle. 272 00:13:14,711 --> 00:13:16,460 If you had a good set up, you could set it 273 00:13:16,460 --> 00:13:18,890 for about 1.2 grams. 274 00:13:18,890 --> 00:13:21,050 If you had a bad set up, it'd go up to 2 grams. 275 00:13:21,050 --> 00:13:29,879 So you could tell how serious a person was by how much force 276 00:13:29,879 --> 00:13:30,795 was required to track. 277 00:13:34,010 --> 00:13:36,790 So for example, the primary thing 278 00:13:36,790 --> 00:13:38,540 that determined how well the system worked 279 00:13:38,540 --> 00:13:41,092 was the cartridge. 280 00:13:41,092 --> 00:13:42,550 So here's a blow up of a cartridge. 281 00:13:42,550 --> 00:13:46,320 The cartridge-- anybody who is serious 282 00:13:46,320 --> 00:13:47,970 had a diamond-tipped needle. 283 00:13:47,970 --> 00:13:49,980 If they didn't have a diamond-tipped needle, 284 00:13:49,980 --> 00:13:51,990 you just didn't even think about them. 285 00:13:51,990 --> 00:13:54,120 So everybody I knew had a diamond-tipped needle. 286 00:13:54,120 --> 00:13:56,970 That was just the way it was. 287 00:13:56,970 --> 00:13:59,160 The least expensive kind of cartridges 288 00:13:59,160 --> 00:14:03,020 was a little piece of piezoceramic. 289 00:14:03,020 --> 00:14:07,470 The piezoceramic was stiff, so that, 290 00:14:07,470 --> 00:14:09,900 as you played the records, they tended to wear out. 291 00:14:09,900 --> 00:14:12,450 Because you had to put a lot of force on it. 292 00:14:12,450 --> 00:14:17,400 Mine, I had the V15 type 4. 293 00:14:17,400 --> 00:14:20,280 So this was a magnetic device. 294 00:14:20,280 --> 00:14:22,890 So there were a tiny little electromagnets, 295 00:14:22,890 --> 00:14:24,450 so that the up and down motions would 296 00:14:24,450 --> 00:14:26,130 be converted into an electrical signal 297 00:14:26,130 --> 00:14:28,700 by magnetism, not by piezoelectric effect. 298 00:14:28,700 --> 00:14:32,550 And you could track that with a much smaller force. 299 00:14:32,550 --> 00:14:35,460 And I should just mention just so you know that we 300 00:14:35,460 --> 00:14:36,810 were serious back then-- 301 00:14:36,810 --> 00:14:40,890 so this cost 160 then dollars. 302 00:14:40,890 --> 00:14:45,780 There's been a factor of six or so inflation, this cost $1,000. 303 00:14:45,780 --> 00:14:49,160 And if you were serious, you had these. 304 00:14:49,160 --> 00:14:54,650 So that's kind of the technology as it existed 305 00:14:54,650 --> 00:14:56,690 when I was an undergraduate. 306 00:14:56,690 --> 00:14:59,150 And it was good. 307 00:14:59,150 --> 00:15:02,480 It had existed for over 100 years. 308 00:15:02,480 --> 00:15:07,100 But it was still basically the same thing Edison had done. 309 00:15:07,100 --> 00:15:09,590 The media-- oh, I should show it just in case 310 00:15:09,590 --> 00:15:12,020 you've never seen a media. 311 00:15:12,020 --> 00:15:13,730 So this is a record. 312 00:15:13,730 --> 00:15:14,670 This was my record. 313 00:15:20,080 --> 00:15:23,892 And the problem is, of course, they scratch. 314 00:15:23,892 --> 00:15:25,600 So over time, these things would scratch. 315 00:15:25,600 --> 00:15:27,599 So the technologies for things like, 316 00:15:27,599 --> 00:15:29,140 how do you keep them from scratching, 317 00:15:29,140 --> 00:15:30,020 how do you wear them out? 318 00:15:30,020 --> 00:15:32,680 And you can also appreciate sort of where the music industry is 319 00:15:32,680 --> 00:15:33,490 coming from. 320 00:15:33,490 --> 00:15:35,680 This costs about $15-- 321 00:15:35,680 --> 00:15:38,770 15 then dollars. 322 00:15:38,770 --> 00:15:41,660 So with factor six, it's about $100. 323 00:15:41,660 --> 00:15:44,110 So from the point of view of buying music, 324 00:15:44,110 --> 00:15:46,792 it was much more expensive back then. 325 00:15:46,792 --> 00:15:49,000 And from the point of view of the recording industry, 326 00:15:49,000 --> 00:15:49,720 think of it. 327 00:15:49,720 --> 00:15:51,400 You had all these addicted people 328 00:15:51,400 --> 00:15:54,010 wanting to buy this stuff. 329 00:15:54,010 --> 00:15:55,120 And it wears out. 330 00:15:57,760 --> 00:16:00,190 So you play it for two or three years, 331 00:16:00,190 --> 00:16:02,090 you have to buy a new one. 332 00:16:02,090 --> 00:16:04,560 OK, so you can sort of appreciate where the music 333 00:16:04,560 --> 00:16:05,730 industry is coming from. 334 00:16:05,730 --> 00:16:08,360 Because things have changed. 335 00:16:08,360 --> 00:16:14,070 OK, so distortions, expensive, fragile, blah, blah, blah. 336 00:16:14,070 --> 00:16:17,280 Then in the 1980s, everything changed. 337 00:16:17,280 --> 00:16:20,400 In the 1980s, there was the invention of the CD. 338 00:16:20,400 --> 00:16:23,250 The CD was invented by Philips Corp 339 00:16:23,250 --> 00:16:25,830 and Sony Corp working together. 340 00:16:25,830 --> 00:16:29,930 And just everything was revolutionized. 341 00:16:29,930 --> 00:16:35,960 So we'll talk a little bit about how the CD works. 342 00:16:35,960 --> 00:16:38,300 The important thing for the discussion 343 00:16:38,300 --> 00:16:43,410 is that, unlike records, CDs are virtually indestructible. 344 00:16:43,410 --> 00:16:44,660 They have very low distortion. 345 00:16:44,660 --> 00:16:47,060 They are next to free. 346 00:16:47,060 --> 00:16:50,390 And all of that is technology that's 347 00:16:50,390 --> 00:16:51,800 in large part enabled by 6.003. 348 00:16:51,800 --> 00:16:54,920 Of And that's what I want to actually talk about. 349 00:16:54,920 --> 00:16:57,585 So what is a CD? 350 00:16:57,585 --> 00:17:00,620 A CD is not very different from a record. 351 00:17:00,620 --> 00:17:02,979 So a CD has tracks. 352 00:17:02,979 --> 00:17:04,520 The information is written in tracks. 353 00:17:04,520 --> 00:17:06,369 The tracks are written as a spiral. 354 00:17:06,369 --> 00:17:10,359 The spiral starts at the inside and spirals out. 355 00:17:10,359 --> 00:17:14,319 And so the idea is that you read a spiral of messages. 356 00:17:14,319 --> 00:17:16,780 And just like a record-- 357 00:17:16,780 --> 00:17:19,450 the way you make a record is called embossing. 358 00:17:19,450 --> 00:17:21,319 You make a master. 359 00:17:21,319 --> 00:17:28,460 And then you take molten plastic, and press it, 360 00:17:28,460 --> 00:17:32,780 and copy the patterns to make that record. 361 00:17:32,780 --> 00:17:35,570 In fact, there is a big technology on the formula 362 00:17:35,570 --> 00:17:38,030 that you used for the vinyl. 363 00:17:38,030 --> 00:17:40,460 One of the secrets of RCA at the time when 364 00:17:40,460 --> 00:17:44,540 I was in graduate school, they had a secret recipe 365 00:17:44,540 --> 00:17:48,680 that included dropping four slices of American cheese 366 00:17:48,680 --> 00:17:50,300 in their vinyl producing. 367 00:17:50,300 --> 00:17:51,890 Because somebody had accidentally once 368 00:17:51,890 --> 00:17:53,450 dropped a sandwich. 369 00:17:53,450 --> 00:17:56,970 And those records came out good. 370 00:17:56,970 --> 00:17:59,030 So there were magical things like that that 371 00:17:59,030 --> 00:18:01,670 were trade secrets at the time. 372 00:18:01,670 --> 00:18:05,600 OK, so that technology was basically embossing. 373 00:18:05,600 --> 00:18:09,410 And that's the same thing you do when you make a CD. 374 00:18:09,410 --> 00:18:14,720 CD is poly carbonate, which is injection molded. 375 00:18:14,720 --> 00:18:18,500 So you make a master. 376 00:18:18,500 --> 00:18:21,150 The master is usually made out of aluminum. 377 00:18:21,150 --> 00:18:23,150 The master is made very similar to the way 378 00:18:23,150 --> 00:18:24,530 that Edison would make a master. 379 00:18:24,530 --> 00:18:26,870 There's a big device with an electromagnet 380 00:18:26,870 --> 00:18:30,710 with a small stylus and it punches them. 381 00:18:30,710 --> 00:18:35,520 But then you copy the CDs by injection molding. 382 00:18:35,520 --> 00:18:38,510 So you take that original. 383 00:18:38,510 --> 00:18:40,320 And there's actually a complicated process. 384 00:18:40,320 --> 00:18:44,930 You use the original to make a half a dozen fathers. 385 00:18:44,930 --> 00:18:47,030 The fathers are used to make mothers. 386 00:18:47,030 --> 00:18:48,530 Each father makes 20 mothers. 387 00:18:48,530 --> 00:18:51,050 Then each of the mothers make children. 388 00:18:51,050 --> 00:18:54,050 And so you end up being able to get a lot of copies 389 00:18:54,050 --> 00:18:55,175 out of the original master. 390 00:18:58,460 --> 00:19:01,520 There's a reflective layer of aluminum 391 00:19:01,520 --> 00:19:03,920 that's evaporated on top of the polycarbonate. 392 00:19:03,920 --> 00:19:09,320 And then there's a thin layer of plastic put over the top. 393 00:19:09,320 --> 00:19:14,450 That results in something that's extremely robust. 394 00:19:14,450 --> 00:19:18,050 And we'll see in a minute how they make it robust. 395 00:19:18,050 --> 00:19:22,640 But this dimension is like over a millimeter. 396 00:19:22,640 --> 00:19:29,450 So this is the CD version of the same album. 397 00:19:29,450 --> 00:19:34,460 So this surface is the information reading side. 398 00:19:34,460 --> 00:19:36,930 There's a millimeter of stuff here. 399 00:19:36,930 --> 00:19:39,830 There's a millimeter of polycarbonate protecting 400 00:19:39,830 --> 00:19:42,320 this surface that has the information 401 00:19:42,320 --> 00:19:44,030 from the environment. 402 00:19:44,030 --> 00:19:47,549 So in fact, if you want to destroy somebody's CD, 403 00:19:47,549 --> 00:19:49,340 you might think the way to do that would be 404 00:19:49,340 --> 00:19:51,080 to write in felt tip on here. 405 00:19:51,080 --> 00:19:52,310 No, don't do it that way. 406 00:19:52,310 --> 00:19:56,150 Use a ballpoint pen on this side. 407 00:19:56,150 --> 00:19:59,060 Because that layer of plastic is very thin. 408 00:19:59,060 --> 00:20:00,760 So use a ballpoint pen here. 409 00:20:00,760 --> 00:20:02,500 It will completely screw it up. 410 00:20:02,500 --> 00:20:08,335 So it's very robust when it gets imperfections on the read side. 411 00:20:10,930 --> 00:20:15,350 OK, so that's kind of the structure. 412 00:20:15,350 --> 00:20:19,510 It's virtually indestructible is the point. 413 00:20:19,510 --> 00:20:22,090 So then, how do you get the information on it? 414 00:20:22,090 --> 00:20:25,540 Well, the information is coded in these patterns. 415 00:20:25,540 --> 00:20:27,350 The patterns are small. 416 00:20:27,350 --> 00:20:29,785 The pattern width is half micron. 417 00:20:29,785 --> 00:20:34,940 A human hair is 50 to 100 microns thick. 418 00:20:34,940 --> 00:20:37,020 And I have white here. 419 00:20:37,020 --> 00:20:38,790 Well, I used to have blonde hair. 420 00:20:38,790 --> 00:20:40,160 I have gray here now. 421 00:20:40,160 --> 00:20:42,470 Ignoring that for the moment, I used have blond hair. 422 00:20:42,470 --> 00:20:45,860 Blonde hair is about 50 microns diameter. 423 00:20:45,860 --> 00:20:48,830 Black hair is about 100 microns diameter. 424 00:20:48,830 --> 00:20:50,630 So depending on the color of your hair, 425 00:20:50,630 --> 00:20:55,100 it would obscure 30 to 60 tracks if you 426 00:20:55,100 --> 00:20:56,720 were to lay a hair across it. 427 00:20:56,720 --> 00:20:59,960 So these are very, very small patterns. 428 00:20:59,960 --> 00:21:08,960 And so the trick for how you code them is a big problem. 429 00:21:08,960 --> 00:21:12,657 Before we get there, though, how do you code the audio? 430 00:21:12,657 --> 00:21:14,240 You won't be too surprised to find out 431 00:21:14,240 --> 00:21:16,190 that it's a sample data system. 432 00:21:16,190 --> 00:21:19,060 So we do sampling just the way we've talked about it before. 433 00:21:19,060 --> 00:21:22,550 Audio, you can hear sounds from about 20 hertz 434 00:21:22,550 --> 00:21:24,170 to about 20 kilohertz. 435 00:21:24,170 --> 00:21:26,330 This is an ideogram. 436 00:21:26,330 --> 00:21:28,880 Audiogram's plot is a function of frequency. 437 00:21:28,880 --> 00:21:30,950 What's the minimum amplitude required 438 00:21:30,950 --> 00:21:33,129 to be able to hear that frequency? 439 00:21:33,129 --> 00:21:35,045 You're most sensitive in the kilohertz region. 440 00:21:37,600 --> 00:21:40,072 The scale over here is DB in something 441 00:21:40,072 --> 00:21:41,280 we call sound pressure level. 442 00:21:41,280 --> 00:21:43,880 Sound pressure level is the centigrade equivalent 443 00:21:43,880 --> 00:21:45,590 for hearing. 444 00:21:45,590 --> 00:21:47,090 So 0 is just audible. 445 00:21:47,090 --> 00:21:49,580 100 starts to hurt. 446 00:21:49,580 --> 00:21:53,910 120 is permanent damage kind of thing. 447 00:21:53,910 --> 00:21:59,990 So you can hear about 100 decibels over a range 448 00:21:59,990 --> 00:22:02,890 from 20 to 20 kilohertz. 449 00:22:02,890 --> 00:22:07,260 And so the way they coded it for the CD 450 00:22:07,260 --> 00:22:10,200 was sampled at 44.1 kilohertz. 451 00:22:10,200 --> 00:22:13,380 They chose that funny number to be 452 00:22:13,380 --> 00:22:16,560 a little bit bigger than twice the highest 453 00:22:16,560 --> 00:22:19,590 frequency that you can hear. 454 00:22:19,590 --> 00:22:21,450 But they made it the funny number 455 00:22:21,450 --> 00:22:26,950 because they wanted it to be an integer multiple of 60 456 00:22:26,950 --> 00:22:29,580 so it was easy to synchronize with television signals. 457 00:22:29,580 --> 00:22:31,300 That's where the funny number came from. 458 00:22:31,300 --> 00:22:37,470 So 44.1 kilohertz divided by 60 is some integer. 459 00:22:37,470 --> 00:22:40,500 The problem is, of course, that they wanted to be greedy. 460 00:22:40,500 --> 00:22:43,590 They wanted all the bits to code audio. 461 00:22:43,590 --> 00:22:47,880 So they only sampled barely over the frequency 462 00:22:47,880 --> 00:22:49,860 that was required. 463 00:22:49,860 --> 00:22:52,715 So according to the sampling theorum, you need 40. 464 00:22:58,640 --> 00:23:02,700 So if you had a sample data system running at 44, 465 00:23:02,700 --> 00:23:09,430 any amount of signal above 22 would alias and sound terrible. 466 00:23:09,430 --> 00:23:11,180 So you need to put an anti-aliasing filter 467 00:23:11,180 --> 00:23:12,060 to prevent that. 468 00:23:14,820 --> 00:23:16,820 We've kind of cheated the whole way through 003. 469 00:23:16,820 --> 00:23:18,960 We always said, when you need a low-pass filter, 470 00:23:18,960 --> 00:23:19,890 use an ideal one. 471 00:23:19,890 --> 00:23:22,640 Well, they don't exist. 472 00:23:22,640 --> 00:23:26,030 we use ideal filters because they're conceptually simple. 473 00:23:26,030 --> 00:23:27,770 There is no such thing. 474 00:23:27,770 --> 00:23:31,820 If you actually try to make a filter using designs like Russ 475 00:23:31,820 --> 00:23:33,110 talked about in recitation. 476 00:23:33,110 --> 00:23:35,780 Butterworth filters, things like that, they 477 00:23:35,780 --> 00:23:38,000 have transitioned bandwidths. 478 00:23:38,000 --> 00:23:40,310 So there is a range of frequencies 479 00:23:40,310 --> 00:23:42,880 for which the filter is neither fully on nor fully off. 480 00:23:42,880 --> 00:23:45,490 And it's in transition. 481 00:23:45,490 --> 00:23:48,440 You would like that range of frequencies to be very small, 482 00:23:48,440 --> 00:23:49,880 so you don't waste bandwidth. 483 00:23:52,940 --> 00:23:56,180 So they would like to reconstruct up to 40. 484 00:23:56,180 --> 00:24:02,210 And they allowed 4 kilohertz in the guard band. 485 00:24:02,210 --> 00:24:04,580 That was intentional, because they 486 00:24:04,580 --> 00:24:06,770 didn't want to have a lot of bits that were not 487 00:24:06,770 --> 00:24:09,037 useful for recording audio. 488 00:24:09,037 --> 00:24:11,120 That's the reason they only went up to 44, instead 489 00:24:11,120 --> 00:24:14,690 of, say, 80, which would have made the anti-aliasing problem 490 00:24:14,690 --> 00:24:16,700 simple. 491 00:24:16,700 --> 00:24:20,900 But it complicated the making of the anti-alias filter. 492 00:24:20,900 --> 00:24:24,980 Because there is sounds that come into microphones 493 00:24:24,980 --> 00:24:26,150 about 20 kilohertz. 494 00:24:26,150 --> 00:24:29,390 It's just that you can't hear them. 495 00:24:29,390 --> 00:24:33,060 But if you were to record those sounds, and then sample them 496 00:24:33,060 --> 00:24:36,879 at 40 kilohertz, you would hear the alias. 497 00:24:36,879 --> 00:24:38,420 And it would be highly objectionable. 498 00:24:38,420 --> 00:24:40,790 So you have to anti-alias. 499 00:24:40,790 --> 00:24:42,650 And that's actually hard. 500 00:24:42,650 --> 00:24:47,270 Because if you'd like to hear up to 20, 501 00:24:47,270 --> 00:24:52,650 and if you're sampling at 44.1, 20 502 00:24:52,650 --> 00:24:56,340 reflected down to 24.1, which means 503 00:24:56,340 --> 00:25:00,930 that you want the transition band to be 34 kilohertz. 504 00:25:04,030 --> 00:25:06,540 And if you think about the design 505 00:25:06,540 --> 00:25:09,960 of a filter like a Butterworth filter, 506 00:25:09,960 --> 00:25:15,512 in order to get a transition that attenuates by ADDB-- 507 00:25:15,512 --> 00:25:17,220 think about the dynamic range of hearing. 508 00:25:17,220 --> 00:25:22,170 You have to make the loudest sounds small enough, 509 00:25:22,170 --> 00:25:23,730 so that you can no longer hear them. 510 00:25:23,730 --> 00:25:28,050 So in general, you have to attenuate by about ADDB. 511 00:25:28,050 --> 00:25:29,820 You need about 50 poles. 512 00:25:29,820 --> 00:25:37,030 To get an ADD attenuation using a Butterworth design 513 00:25:37,030 --> 00:25:39,610 in 4 kilohertz at 20 kilohertz. 514 00:25:39,610 --> 00:25:43,470 Everybody sort of know what I'm talking about? 515 00:25:43,470 --> 00:25:45,920 That's impossible. 516 00:25:45,920 --> 00:25:50,060 Lining up filters-- so the way the Butterworth filters work, 517 00:25:50,060 --> 00:25:50,690 you remember? 518 00:25:50,690 --> 00:25:59,240 So you have a bunch of poles like so. 519 00:25:59,240 --> 00:26:01,550 The way you think about that is that you 520 00:26:01,550 --> 00:26:07,160 start with a design that looks like a not-so-good low-pass 521 00:26:07,160 --> 00:26:09,530 filter. 522 00:26:09,530 --> 00:26:14,010 Then you add these poles, which give you 523 00:26:14,010 --> 00:26:19,180 a peaky response like that, so that when this is going down, 524 00:26:19,180 --> 00:26:23,800 the peakiness here helps to cancel that. 525 00:26:23,800 --> 00:26:26,740 Then you do that again. 526 00:26:26,740 --> 00:26:30,820 So you can think about these successive pole pairs 527 00:26:30,820 --> 00:26:35,060 as sharpening the transition by pushing up 528 00:26:35,060 --> 00:26:38,400 on the center frequency and rolling off 529 00:26:38,400 --> 00:26:39,790 on the edge frequency. 530 00:26:39,790 --> 00:26:42,850 But in order for that to work, they have to all be lined up. 531 00:26:42,850 --> 00:26:44,320 So if you've got 50 of them, you've 532 00:26:44,320 --> 00:26:47,380 got to be able to line these up to about a percent or so. 533 00:26:47,380 --> 00:26:50,830 Otherwise, they don't perform the way they're supposed to, 534 00:26:50,830 --> 00:26:53,840 and they can actually make things worse. 535 00:26:53,840 --> 00:27:00,130 So making an analog filter of that type 536 00:27:00,130 --> 00:27:02,980 was impossible then and is difficult now. 537 00:27:02,980 --> 00:27:06,190 Here is sort of the state of the art. 538 00:27:06,190 --> 00:27:08,400 So this actually does have ADDB of roll 539 00:27:08,400 --> 00:27:10,060 off, because it's a fancier design. 540 00:27:10,060 --> 00:27:12,070 As an elliptic design. 541 00:27:12,070 --> 00:27:14,620 The trick here, though, is that all the parts 542 00:27:14,620 --> 00:27:16,510 are laser trimmed. 543 00:27:16,510 --> 00:27:18,250 So they're all done on one substrate. 544 00:27:18,250 --> 00:27:20,110 All the inductors, all the capacitors 545 00:27:20,110 --> 00:27:22,652 are done on a single substrate. 546 00:27:22,652 --> 00:27:24,610 Then they're laser trimmed, so that they follow 547 00:27:24,610 --> 00:27:27,550 the exactly the right places. 548 00:27:27,550 --> 00:27:30,850 Then it's encased in aluminum to help 549 00:27:30,850 --> 00:27:34,030 to hold the temperature constant and to keep 550 00:27:34,030 --> 00:27:37,420 the electromagnetic interference out. 551 00:27:37,420 --> 00:27:42,290 And even so, this has just 11 pole pairs. 552 00:27:42,290 --> 00:27:47,600 So this is not nearly adequate for that kind of a design. 553 00:27:47,600 --> 00:27:56,000 So what we do instead is we use something called oversampling. 554 00:27:56,000 --> 00:27:58,820 So instead of sampling at 44.1, which 555 00:27:58,820 --> 00:28:01,190 is what you would think would happen in a CD, 556 00:28:01,190 --> 00:28:04,680 you sample it four times that. 557 00:28:04,680 --> 00:28:07,590 By sampling at a very high frequency, 558 00:28:07,590 --> 00:28:11,970 you make it easy to make an anti-aliasing filter. 559 00:28:11,970 --> 00:28:14,130 Because now, you only need to worry 560 00:28:14,130 --> 00:28:18,690 about saving the frequencies up to 20 kilohertz. 561 00:28:18,690 --> 00:28:26,100 But you've got 170 kilohertz to do the transition. 562 00:28:26,100 --> 00:28:27,990 So the filter design becomes very easy. 563 00:28:27,990 --> 00:28:32,820 In fact, a typical design only has five poles, and so just 564 00:28:32,820 --> 00:28:33,880 a handful of capacitors. 565 00:28:33,880 --> 00:28:38,020 Just two capacitors and two op amps will do that. 566 00:28:40,960 --> 00:28:47,212 So then you've got a system that is sampled at 176 kilohertz. 567 00:28:47,212 --> 00:28:48,420 What's the problem with that? 568 00:28:53,560 --> 00:28:56,623 What's the problem with sampling at 176 kilohertz? 569 00:29:00,700 --> 00:29:02,630 Too much information. 570 00:29:02,630 --> 00:29:05,277 So what you've done by sampling at the higher frequency 571 00:29:05,277 --> 00:29:07,610 is you've increased the amount of information that needs 572 00:29:07,610 --> 00:29:10,210 to be stored by a factor of 4. 573 00:29:10,210 --> 00:29:14,980 So instead of being able to record 74 minutes of music, 574 00:29:14,980 --> 00:29:18,310 you can record 74 divided by 4, 18 minutes of music. 575 00:29:18,310 --> 00:29:20,800 Nobody would like that, right? 576 00:29:20,800 --> 00:29:23,960 So how do you fix that? 577 00:29:23,960 --> 00:29:27,300 AUDIENCE: [INAUDIBLE] 578 00:29:27,300 --> 00:29:29,820 DENNIS FREEMAN: Downsample, exactly. 579 00:29:29,820 --> 00:29:36,690 So what you do is you run that discrete time signal that 580 00:29:36,690 --> 00:29:42,330 was sampled at 176 through a discrete low-pass filter, which 581 00:29:42,330 --> 00:29:46,050 you implemented digitally with arbitrary precision as much 582 00:29:46,050 --> 00:29:48,520 as you want. 583 00:29:48,520 --> 00:29:57,930 So here's a design based on a filter whose length is 200. 584 00:29:57,930 --> 00:30:01,350 By making a length 200 discrete time filter, 585 00:30:01,350 --> 00:30:04,222 you can capture the basic sine T over T shape 586 00:30:04,222 --> 00:30:06,180 that you're trying to get with an ideal filter. 587 00:30:09,300 --> 00:30:14,680 So here with 200 samples, it's easy to reproduce 588 00:30:14,680 --> 00:30:17,500 a large number of ripples similar to the large number 589 00:30:17,500 --> 00:30:19,850 of ripples that I wanted over here, 590 00:30:19,850 --> 00:30:22,540 but now with digital multiplies, not 591 00:30:22,540 --> 00:30:26,490 with resistors and capacitors. 592 00:30:26,490 --> 00:30:30,420 And this shows where the poles-- 593 00:30:30,420 --> 00:30:32,940 this is an all 0 filter. 594 00:30:32,940 --> 00:30:34,350 It's a finite length filter. 595 00:30:34,350 --> 00:30:36,090 We didn't talk about that. 596 00:30:36,090 --> 00:30:39,270 It's an easy way to implement filter design. 597 00:30:39,270 --> 00:30:41,620 It's an easy way to design filters. 598 00:30:41,620 --> 00:30:43,530 It's a very popular way of doing it. 599 00:30:43,530 --> 00:30:44,810 So it's an all 0 filter. 600 00:30:44,810 --> 00:30:46,560 And you can sort of figure out by thinking 601 00:30:46,560 --> 00:30:50,100 about the spacing of the zeros relative to the unit 602 00:30:50,100 --> 00:30:53,940 circle, how that implements a low-pass filter. 603 00:30:53,940 --> 00:30:56,850 That was done with an automated optimizing 604 00:30:56,850 --> 00:30:59,540 algorithm for finding the best low-pass filter. 605 00:30:59,540 --> 00:31:03,780 The method is called the Parks-McClellan algorithm. 606 00:31:03,780 --> 00:31:05,580 And it makes a very nice filter. 607 00:31:05,580 --> 00:31:10,680 So this shows the transition of ADDB 608 00:31:10,680 --> 00:31:17,280 passing 20 kilohertz signals and attenuating 24 and above. 609 00:31:17,280 --> 00:31:22,130 So you get a very nice filter by doing that in discrete time. 610 00:31:22,130 --> 00:31:26,174 So then having done the digital filter, 611 00:31:26,174 --> 00:31:27,590 you can throw away the frequencies 612 00:31:27,590 --> 00:31:30,700 that would alias and then downsample. 613 00:31:30,700 --> 00:31:31,314 Yeah? 614 00:31:31,314 --> 00:31:34,237 AUDIENCE: Is the filter in those zero locations 615 00:31:34,237 --> 00:31:35,470 standardized by anybody? 616 00:31:35,470 --> 00:31:37,060 Or does any manufacturer-- 617 00:31:37,060 --> 00:31:39,470 DENNIS FREEMAN: Any manufacturer-- yes. 618 00:31:39,470 --> 00:31:41,132 Every manufacturer is different. 619 00:31:41,132 --> 00:31:42,590 What's happening over time, though, 620 00:31:42,590 --> 00:31:45,290 is that the chips are becoming standardized. 621 00:31:45,290 --> 00:31:48,487 So if you buy a particular chip, it has a particular filter set. 622 00:31:48,487 --> 00:31:50,070 But it's not standardized in the code. 623 00:31:53,750 --> 00:32:01,340 OK, so that's how you generate a signal that is robust. 624 00:32:01,340 --> 00:32:03,530 And it's basically just 003. 625 00:32:03,530 --> 00:32:06,980 Now we have to think about, how would you make the player? 626 00:32:06,980 --> 00:32:09,200 Because what we'd like to do is enable 627 00:32:09,200 --> 00:32:16,140 you to use a CD in a more portable fashion 628 00:32:16,140 --> 00:32:19,140 than you could use this technology. 629 00:32:19,140 --> 00:32:21,060 So the question is now, how would you 630 00:32:21,060 --> 00:32:27,120 make a player that can pick up this information off the CD, 631 00:32:27,120 --> 00:32:30,540 so that you can reproduce it in a Walkman 632 00:32:30,540 --> 00:32:35,440 or in a more portable environment? 633 00:32:35,440 --> 00:32:38,620 So I had this brilliant idea about two years ago 634 00:32:38,620 --> 00:32:45,070 that I would take this CD and I would use my microscope. 635 00:32:45,070 --> 00:32:47,770 I may have mentioned I light microscopy. 636 00:32:47,770 --> 00:32:48,880 I went into my microscope. 637 00:32:48,880 --> 00:32:50,630 I used my research-quality microscope, 638 00:32:50,630 --> 00:32:52,810 which costs $70,000. 639 00:32:52,810 --> 00:32:57,730 And I'll take a beautiful picture of the bits on here. 640 00:32:57,730 --> 00:33:03,100 And I'll write a problem set for 003 where you have to decode 641 00:33:03,100 --> 00:33:04,780 the bitstream. 642 00:33:04,780 --> 00:33:06,940 I thought it was brilliant. 643 00:33:06,940 --> 00:33:09,520 And it was like a week before the homework 644 00:33:09,520 --> 00:33:11,520 had to be written, you know, the standard thing. 645 00:33:11,520 --> 00:33:14,020 And of course, I couldn't get it to work in time. 646 00:33:14,020 --> 00:33:18,940 The problem was that even my $70,000 research microscope 647 00:33:18,940 --> 00:33:22,620 has trouble seeing the bits. 648 00:33:22,620 --> 00:33:25,520 The bits are really small. 649 00:33:25,520 --> 00:33:31,020 So when I first tried it, I couldn't see anything. 650 00:33:31,020 --> 00:33:32,760 So I fiddled around with the microscope. 651 00:33:32,760 --> 00:33:35,280 I tried all different kinds of magnifications. 652 00:33:35,280 --> 00:33:37,260 I tried all kinds of different optical tricks. 653 00:33:37,260 --> 00:33:39,190 But I know I couldn't figure out anything. 654 00:33:39,190 --> 00:33:39,780 I couldn't get it to work. 655 00:33:39,780 --> 00:33:41,029 So I got my graduate students. 656 00:33:41,029 --> 00:33:43,200 They're always much better at this sort of thing. 657 00:33:43,200 --> 00:33:44,880 And we finally got it to work. 658 00:33:44,880 --> 00:33:47,520 We cheated like crazy. 659 00:33:47,520 --> 00:33:50,610 So we had to use something called water immersion. 660 00:33:50,610 --> 00:33:53,490 So you put a drop of water on the surface, 661 00:33:53,490 --> 00:33:55,170 so that you can use the-- 662 00:33:55,170 --> 00:34:00,620 you lose resolution when you go through a glass 663 00:34:00,620 --> 00:34:03,380 to air interface. 664 00:34:03,380 --> 00:34:05,960 That's a bunch of microscopy that you don't need to know. 665 00:34:05,960 --> 00:34:07,010 But you do. 666 00:34:07,010 --> 00:34:08,810 So there are tricks in microscopy 667 00:34:08,810 --> 00:34:13,250 where you can avoid having the air interface by using water. 668 00:34:13,250 --> 00:34:16,880 The water is more closely optically matched to glass. 669 00:34:16,880 --> 00:34:20,120 So what we did was we did immersion microscopy, 670 00:34:20,120 --> 00:34:22,969 put a drop of water on here, put the lens in the water. 671 00:34:22,969 --> 00:34:25,780 And that's the picture I got. 672 00:34:25,780 --> 00:34:32,580 So the point is that I had to work to get a $70,000 673 00:34:32,580 --> 00:34:37,290 microscope to read the bits off here. 674 00:34:37,290 --> 00:34:39,120 And I sort of gave up with the idea 675 00:34:39,120 --> 00:34:41,340 that I would record a track. 676 00:34:41,340 --> 00:34:45,210 Because it was just too painful to just get this one image. 677 00:34:45,210 --> 00:34:51,510 Now the trick is they make players that cost $10. 678 00:34:51,510 --> 00:34:55,650 What do they know that I don't know? 679 00:34:55,650 --> 00:34:58,230 So here I am with a $70,000 research microscope 680 00:34:58,230 --> 00:35:00,860 having a difficult time getting the bits off. 681 00:35:00,860 --> 00:35:07,070 How do they make a CD player for $10? 682 00:35:07,070 --> 00:35:09,499 Just for fun, I did the same thing with a DVD. 683 00:35:09,499 --> 00:35:11,790 After I figured out how to do the immersion microscopy, 684 00:35:11,790 --> 00:35:14,840 I figured I'd at least-- so this is a DVD done 685 00:35:14,840 --> 00:35:17,800 with the same sort of thing. 686 00:35:17,800 --> 00:35:19,810 So how do they do it? 687 00:35:19,810 --> 00:35:21,250 They do very clever things. 688 00:35:21,250 --> 00:35:23,265 The first trick they use is interference. 689 00:35:23,265 --> 00:35:25,560 So I talked about interference last time 690 00:35:25,560 --> 00:35:30,510 with our standing wave illumination microscope. 691 00:35:30,510 --> 00:35:32,400 They do interference. 692 00:35:32,400 --> 00:35:36,221 So I told you that this is polycarbonate with an aluminum 693 00:35:36,221 --> 00:35:36,720 coding. 694 00:35:36,720 --> 00:35:39,930 The aluminum is actually working like a mirror. 695 00:35:39,930 --> 00:35:41,380 And that's the intent. 696 00:35:41,380 --> 00:35:46,780 So they don't actually try to code block and white. 697 00:35:46,780 --> 00:35:49,090 What they try to code is distance. 698 00:35:49,090 --> 00:35:51,250 So the idea is that you take a laser beam, 699 00:35:51,250 --> 00:35:55,240 and you out reflect it off of the CD much 700 00:35:55,240 --> 00:35:58,580 like it reflects off a mirror. 701 00:35:58,580 --> 00:36:03,260 And then the features that were embossed 702 00:36:03,260 --> 00:36:05,420 are at different depths. 703 00:36:05,420 --> 00:36:08,000 And the depth's chosen very specifically 704 00:36:08,000 --> 00:36:11,250 to be lambda over 4, the wavelength of the light. 705 00:36:11,250 --> 00:36:15,290 They use a 700 nanometer laser. 706 00:36:15,290 --> 00:36:24,290 And the features are offset by lambda over 4. 707 00:36:24,290 --> 00:36:27,410 The idea is that-- 708 00:36:27,410 --> 00:36:32,730 so if you have a feature, and you have-- 709 00:36:32,730 --> 00:36:40,120 so you compare, what would be the timing for this laser beam 710 00:36:40,120 --> 00:36:42,106 if it struck the mirror? 711 00:36:42,106 --> 00:36:43,480 You compare that to what it would 712 00:36:43,480 --> 00:36:45,250 be if it struck a feature. 713 00:36:45,250 --> 00:36:48,790 And you rig it so that those two times 714 00:36:48,790 --> 00:36:52,610 are different by lambda over 2, lambda over 4 in, 715 00:36:52,610 --> 00:36:56,140 and lambda over four out. 716 00:36:56,140 --> 00:37:02,830 So then if you rig the beam to be half of the light 717 00:37:02,830 --> 00:37:08,190 falls on both, what's the sum of these waveforms 718 00:37:08,190 --> 00:37:13,240 going to be if you have a component that's 719 00:37:13,240 --> 00:37:16,630 at the mirror and a component at the indented? 720 00:37:16,630 --> 00:37:19,810 If these are different in timing by lambda over 2 721 00:37:19,810 --> 00:37:22,240 when you add them, the answer is? 722 00:37:25,138 --> 00:37:26,110 AUDIENCE: [INAUDIBLE] 723 00:37:26,110 --> 00:37:27,550 DENNIS FREEMAN: Yeah, you get complete destructive 724 00:37:27,550 --> 00:37:28,091 interference. 725 00:37:28,091 --> 00:37:29,950 You get an out answer of 0. 726 00:37:29,950 --> 00:37:31,780 So the idea in the reading of them 727 00:37:31,780 --> 00:37:35,530 is to use interference and rig it 728 00:37:35,530 --> 00:37:40,380 so that 0s are represented by no light 729 00:37:40,380 --> 00:37:43,780 and 1s are represented by lots of light. 730 00:37:43,780 --> 00:37:48,910 So that's the first clever trick they do. 731 00:37:48,910 --> 00:37:52,540 Then the question is, how do you focus it? 732 00:37:52,540 --> 00:37:56,470 When I was doing my research microscope, the research 733 00:37:56,470 --> 00:38:00,730 microscope-- and first off, it sits on a half ton of granite 734 00:38:00,730 --> 00:38:02,050 to make it all very stable. 735 00:38:02,050 --> 00:38:05,530 Because when you're doing the focusing, 736 00:38:05,530 --> 00:38:07,360 you don't want-- so the vibrations 737 00:38:07,360 --> 00:38:13,230 of the floor in my lab are about 100 micrometers. 738 00:38:13,230 --> 00:38:14,920 So it's about 100 micrometers. 739 00:38:14,920 --> 00:38:19,160 And I have to have the stability of the system sub-micron. 740 00:38:19,160 --> 00:38:23,060 So I use a table made out of a half ton of granite. 741 00:38:23,060 --> 00:38:24,890 And the microscope sits on top of that. 742 00:38:24,890 --> 00:38:27,370 And that makes everything nice and stable. 743 00:38:27,370 --> 00:38:29,150 They can't do that. 744 00:38:29,150 --> 00:38:32,470 So how do they control the focus without using a half-- 745 00:38:32,470 --> 00:38:35,470 I mean, it wouldn't work to make a Walkman that 746 00:38:35,470 --> 00:38:38,620 is a half ton granite, right? 747 00:38:38,620 --> 00:38:41,560 So what do they do to enable them 748 00:38:41,560 --> 00:38:45,430 to do the same kind of focusing that I was doing? 749 00:38:45,430 --> 00:38:47,050 They use feedback. 750 00:38:47,050 --> 00:38:52,480 So the idea is to substitute feedback for precision. 751 00:38:52,480 --> 00:38:55,210 So rather than trying to make the parts precise, 752 00:38:55,210 --> 00:38:59,900 they put the imprecise parts in a feedback loop-- 753 00:38:59,900 --> 00:39:03,210 very powerful method. 754 00:39:03,210 --> 00:39:04,610 So here's what they do. 755 00:39:04,610 --> 00:39:05,990 They take the laser. 756 00:39:05,990 --> 00:39:09,030 It's trying to focus on to the CD that has patterns in it. 757 00:39:12,040 --> 00:39:15,850 And so there's a beam splitter, so that part of the light 758 00:39:15,850 --> 00:39:18,150 goes over to the detector. 759 00:39:18,150 --> 00:39:18,650 Excuse me. 760 00:39:18,650 --> 00:39:19,660 I didn't do that right. 761 00:39:19,660 --> 00:39:21,520 The laser, it squirts up to here, 762 00:39:21,520 --> 00:39:23,380 goes straight through the beam splitter, 763 00:39:23,380 --> 00:39:25,660 comes back through a focusing lens, 764 00:39:25,660 --> 00:39:27,910 bounces off the beam splitter through another lens, 765 00:39:27,910 --> 00:39:29,930 and hits a detector. 766 00:39:29,930 --> 00:39:32,170 So the detector is this way. 767 00:39:32,170 --> 00:39:33,462 And I'm showing this view then. 768 00:39:33,462 --> 00:39:34,795 So the detector should be there. 769 00:39:34,795 --> 00:39:36,280 I'm sharing a view of it over here. 770 00:39:36,280 --> 00:39:39,760 And it's rigged, so that these two lenses-- 771 00:39:39,760 --> 00:39:43,720 one of the lenses is spherical and one is cylindrical. 772 00:39:43,720 --> 00:39:45,330 What that means is-- 773 00:39:45,330 --> 00:39:48,260 you remember I've talked about microscopy in the past. 774 00:39:48,260 --> 00:39:51,530 Out of focus means blurred. 775 00:39:51,530 --> 00:39:55,080 Blurred for a spot means bigger. 776 00:39:55,080 --> 00:40:01,100 As a microscope goes out of focus, the spot gets bigger. 777 00:40:01,100 --> 00:40:03,230 What you do with these two lenses-- 778 00:40:03,230 --> 00:40:05,870 because one is circularly symmetric, 779 00:40:05,870 --> 00:40:08,210 and the other is slenderly symmetrical, 780 00:40:08,210 --> 00:40:13,010 what you do is you make the beam in this direction 781 00:40:13,010 --> 00:40:17,538 go out of focus as it's going into focus in this direction. 782 00:40:22,380 --> 00:40:28,410 So because you get two lenses, one is symmetric in xy. 783 00:40:28,410 --> 00:40:33,690 And one has no magnification in y, but big magnification in x. 784 00:40:33,690 --> 00:40:37,620 You can rig it so that this dot will go out 785 00:40:37,620 --> 00:40:45,230 of focus horizontally while it's coming into focus vertically. 786 00:40:45,230 --> 00:40:53,070 The result is that, if you move the CD closer, it becomes fat. 787 00:40:53,070 --> 00:40:57,640 And if you move the CD further, it becomes skinny. 788 00:40:57,640 --> 00:41:03,070 So all they need to do, then, is focus up and down 789 00:41:03,070 --> 00:41:05,080 until the dot is as close to being 790 00:41:05,080 --> 00:41:07,120 circular as they can get it. 791 00:41:07,120 --> 00:41:10,370 That's why they use a quadrant detector. 792 00:41:10,370 --> 00:41:14,100 So the quadrant detector lets them calculate 793 00:41:14,100 --> 00:41:18,140 a plus c minus b plus d. 794 00:41:18,140 --> 00:41:21,140 And you would like that answer to be 0. 795 00:41:21,140 --> 00:41:23,280 So now you use that as your error. 796 00:41:23,280 --> 00:41:27,620 They use the light that hits this quadrant detector 797 00:41:27,620 --> 00:41:30,530 to compute this number, put that in a feedback loop 798 00:41:30,530 --> 00:41:34,160 that drives the stage up and down to focus. 799 00:41:34,160 --> 00:41:36,410 So here's an ancient CD. 800 00:41:36,410 --> 00:41:38,350 The CD would be here. 801 00:41:38,350 --> 00:41:42,140 Here is the reading lens. 802 00:41:42,140 --> 00:41:46,490 Here is a platform that has an electromagnet under it that 803 00:41:46,490 --> 00:41:48,335 will move it up and down 100 microns. 804 00:41:54,410 --> 00:42:00,240 So this electromagnet is under the stage. 805 00:42:00,240 --> 00:42:03,760 So this whole thing goes up and down 806 00:42:03,760 --> 00:42:06,570 in a feedback loop that tries to make this error 807 00:42:06,570 --> 00:42:12,090 signal be 0, tries to make the pattern circularly symmetric. 808 00:42:12,090 --> 00:42:15,210 So it's extremely clever. 809 00:42:15,210 --> 00:42:21,430 Then you have this pattern. 810 00:42:21,430 --> 00:42:23,470 So the light is twice as big. 811 00:42:23,470 --> 00:42:26,650 Half the photons are supposed to fit in the pit. 812 00:42:26,650 --> 00:42:28,210 And half of the photons are supposed 813 00:42:28,210 --> 00:42:32,290 to hit in the land that surrounds the pit, 814 00:42:32,290 --> 00:42:44,190 so that as the CD is spinning, if the spot illuminates 815 00:42:44,190 --> 00:42:46,260 a place that's entirely in the pit, 816 00:42:46,260 --> 00:42:50,640 the dot is small because of interference. 817 00:42:50,640 --> 00:42:53,190 And if it hits a place that doesn't have a pit, 818 00:42:53,190 --> 00:42:56,930 the dot is large, because there's no interference. 819 00:42:56,930 --> 00:42:58,340 But then you have the problem of, 820 00:42:58,340 --> 00:43:04,860 how do you keep the dot reading the center of the groove? 821 00:43:04,860 --> 00:43:07,080 So in this technology, we kept the dot 822 00:43:07,080 --> 00:43:11,040 in the center of the groove by using gravity. 823 00:43:11,040 --> 00:43:13,560 We obviously don't want to touch that surface. 824 00:43:13,560 --> 00:43:15,300 So that's not the technology we use. 825 00:43:15,300 --> 00:43:18,210 We don't use a screw. 826 00:43:18,210 --> 00:43:19,065 We use feedback. 827 00:43:22,410 --> 00:43:27,420 So here what we do, instead of projecting a single dot, 828 00:43:27,420 --> 00:43:29,700 we use a beam splitter that turns 829 00:43:29,700 --> 00:43:32,880 the one dot into three dots. 830 00:43:32,880 --> 00:43:35,100 And the three dots are positioned 831 00:43:35,100 --> 00:43:38,970 so that they hit two more photo sensors, one 832 00:43:38,970 --> 00:43:42,520 on each side of the main photo center. 833 00:43:42,520 --> 00:43:46,560 And the idea is that, if the light is perfectly centered 834 00:43:46,560 --> 00:43:50,810 on a track, these will have equal brightness. 835 00:43:50,810 --> 00:43:56,490 But now if the beam is off center, one of the dots 836 00:43:56,490 --> 00:44:00,600 will be bright, because it hit all land, compared 837 00:44:00,600 --> 00:44:05,070 to the other one, which some of the light is hitting the pit. 838 00:44:05,070 --> 00:44:09,340 So this is just like the head tracking thing in 601. 839 00:44:09,340 --> 00:44:13,720 So whichever of the two photo sensors gets more light, 840 00:44:13,720 --> 00:44:17,120 you move the track to compensate for that. 841 00:44:17,120 --> 00:44:19,910 So it all works in a feedback loop. 842 00:44:19,910 --> 00:44:27,380 So the idea here is the same sort of deal. 843 00:44:27,380 --> 00:44:28,910 You have two sensors here. 844 00:44:28,910 --> 00:44:33,940 And you use the error, which is the difference between the two, 845 00:44:33,940 --> 00:44:36,260 to move the head to the track. 846 00:44:36,260 --> 00:44:39,120 And the head motion is this thing over here. 847 00:44:39,120 --> 00:44:42,210 So here's a rack and pinion connected up to a motor. 848 00:44:42,210 --> 00:44:45,680 And so you have that motor servo-ed back 849 00:44:45,680 --> 00:44:50,630 to some photo sensors over here that keep the beam centered. 850 00:44:54,470 --> 00:44:59,210 So the idea, then, is that what you get for $10 851 00:44:59,210 --> 00:45:02,780 to read this thing is completely remarkable. 852 00:45:02,780 --> 00:45:07,010 The idea is that it's got very complicated servo mechanisms 853 00:45:07,010 --> 00:45:11,480 that can read very small things that even a $70,000 854 00:45:11,480 --> 00:45:14,960 microscope has trouble reading. 855 00:45:14,960 --> 00:45:16,130 And it's stable. 856 00:45:16,130 --> 00:45:18,980 And you can run them, or you can jostle them around, 857 00:45:18,980 --> 00:45:20,660 and jog with them. 858 00:45:20,660 --> 00:45:24,650 And that's all because they use feedback, rather than trying 859 00:45:24,650 --> 00:45:27,170 to make precision machining. 860 00:45:27,170 --> 00:45:30,770 And then they make the signals robust by using 861 00:45:30,770 --> 00:45:34,065 the kind of DT signal processing that we talked about. 862 00:45:34,065 --> 00:45:36,290 So anyway, the point of today was just 863 00:45:36,290 --> 00:45:40,010 to give it some motivation of the way 6.003 factors 864 00:45:40,010 --> 00:45:43,380 into technologies that you might be interested in. 865 00:45:43,380 --> 00:45:44,680 And this is just one example. 866 00:45:44,680 --> 00:45:47,150 I could have chosen any number of different devices 867 00:45:47,150 --> 00:45:49,020 and they would have been similar. 868 00:45:49,020 --> 00:45:50,960 So I hope you enjoyed the course. 869 00:45:50,960 --> 00:45:54,060 I hope that you have fun with the final. 870 00:45:54,060 --> 00:45:58,100 Please come on Tuesday, and tell us 871 00:45:58,100 --> 00:46:00,630 what we should do differently next time. 872 00:46:00,630 --> 00:46:02,180 Thanks.