1
00:00:00,040 --> 00:00:02,460
The following content is
provided under a Creative

2
00:00:02,460 --> 00:00:03,870
Commons license.

3
00:00:03,870 --> 00:00:06,910
Your support will help MIT
OpenCourseWare continue to

4
00:00:06,910 --> 00:00:10,560
offer high quality educational
resources for free.

5
00:00:10,560 --> 00:00:13,460
To make a donation or view
additional materials from

6
00:00:13,460 --> 00:00:17,740
hundreds of MIT courses, visit
MIT OpenCourseWare at

7
00:00:17,740 --> 00:00:18,990
ocw.mit.edu.

8
00:00:25,930 --> 00:00:29,510
PROFESSOR: Today I want to
talk a little bit about

9
00:00:29,510 --> 00:00:30,890
designing control systems.

10
00:00:30,890 --> 00:00:34,120
This will finish up
our discussions

11
00:00:34,120 --> 00:00:36,750
on signals and systems.

12
00:00:36,750 --> 00:00:39,480
Let me then just briefly
review where we are.

13
00:00:39,480 --> 00:00:42,790
Hopefully this might help you
also for perspective with

14
00:00:42,790 --> 00:00:45,320
regard to thinking about
the exam tonight.

15
00:00:45,320 --> 00:00:47,080
We've looked at a bunch
of different kinds of

16
00:00:47,080 --> 00:00:50,890
representations for discrete
time systems.

17
00:00:50,890 --> 00:00:55,320
The easiest, most concise method
we looked at was the

18
00:00:55,320 --> 00:00:58,700
representation using a
difference equation.

19
00:00:58,700 --> 00:01:03,690
That's mathematically as
concise as you can get.

20
00:01:03,690 --> 00:01:07,500
But it doesn't tell you
important things like who's

21
00:01:07,500 --> 00:01:10,370
the input and who's the output
and what are all the different

22
00:01:10,370 --> 00:01:12,620
ways that you can get
through the system

23
00:01:12,620 --> 00:01:14,290
from input to output?

24
00:01:14,290 --> 00:01:18,670
So for that question block
diagrams are nice.

25
00:01:18,670 --> 00:01:20,680
Block diagrams are graphical.

26
00:01:20,680 --> 00:01:24,500
It makes it very easy to see
when there is, for example, a

27
00:01:24,500 --> 00:01:29,830
cyclic path through
the network.

28
00:01:29,830 --> 00:01:30,780
But they're graphic.

29
00:01:30,780 --> 00:01:34,100
They're not nearly so concise
as difference equations.

30
00:01:34,100 --> 00:01:36,220
So then we went on
to operators.

31
00:01:36,220 --> 00:01:42,780
Operators are just as concise
as difference equations but

32
00:01:42,780 --> 00:01:46,400
they contain additional
information because the

33
00:01:46,400 --> 00:01:49,890
operators have an implicit
argument.

34
00:01:49,890 --> 00:01:52,500
So there's an input, which is
the argument to the operator,

35
00:01:52,500 --> 00:01:55,220
and there's an output, which is
the output of the operator.

36
00:01:55,220 --> 00:01:59,210
So you can tell who is the input
and who is the output.

37
00:01:59,210 --> 00:02:00,340
So that's good.

38
00:02:00,340 --> 00:02:03,040
That sort of combines the
strengths of difference

39
00:02:03,040 --> 00:02:05,340
equations and block diagrams.

40
00:02:05,340 --> 00:02:07,530
You end up with a concise
representation that has

41
00:02:07,530 --> 00:02:12,660
complete information about
the signal flow paths.

42
00:02:12,660 --> 00:02:17,700
Furthermore, you can analyze
the operators by using

43
00:02:17,700 --> 00:02:23,210
polynomial mathematics and that
gives rise to the notion

44
00:02:23,210 --> 00:02:25,870
of a system functional.

45
00:02:25,870 --> 00:02:29,470
And that's a very nice closure
because that represents an

46
00:02:29,470 --> 00:02:34,590
abstraction that lets us think
about a whole system as though

47
00:02:34,590 --> 00:02:40,740
it were just one part, one
thing, one operator.

48
00:02:40,740 --> 00:02:44,390
So we use that structure then,
all of those representations,

49
00:02:44,390 --> 00:02:47,730
to try to learn about
feedback.

50
00:02:47,730 --> 00:02:50,820
And first off, in the block
diagram it's very easy to see

51
00:02:50,820 --> 00:02:53,000
that any time you have feedback
-- feedback so

52
00:02:53,000 --> 00:02:56,930
enormously powerful that we want
to use it in design --

53
00:02:56,930 --> 00:02:59,070
but you can see immediately
from the structure of the

54
00:02:59,070 --> 00:03:04,000
block diagram that if you have
feedback then you have cycles.

55
00:03:04,000 --> 00:03:04,880
Why is that interesting?

56
00:03:04,880 --> 00:03:08,060
Well, that's interesting because
if you have cycles

57
00:03:08,060 --> 00:03:13,560
then even transient inputs can
generate persistent outputs.

58
00:03:13,560 --> 00:03:15,940
So that's a kind of behavior
that we would like to

59
00:03:15,940 --> 00:03:18,210
understand.

60
00:03:18,210 --> 00:03:20,790
From the nature of feedback
it generates cycles.

61
00:03:20,790 --> 00:03:24,070
From the nature of cycles it
generates persistent responses

62
00:03:24,070 --> 00:03:26,800
even if there's no input.

63
00:03:26,800 --> 00:03:29,690
And we saw that we could
characterize those by thinking

64
00:03:29,690 --> 00:03:34,450
about those responses for
one part at a time.

65
00:03:34,450 --> 00:03:39,280
And those parts we thought of as
poles and the responses to

66
00:03:39,280 --> 00:03:41,840
a single pole we called modes.

67
00:03:41,840 --> 00:03:45,730
So we thought through a way of
decomposing the response of a

68
00:03:45,730 --> 00:03:51,200
complicated system in terms
of a number of additive

69
00:03:51,200 --> 00:03:53,900
components that are
based on poles.

70
00:03:53,900 --> 00:03:59,800
Poles are just the base of
a geometric sequence.

71
00:03:59,800 --> 00:04:02,520
So today then what I want to
do is use that framework to

72
00:04:02,520 --> 00:04:03,930
think about design.

73
00:04:03,930 --> 00:04:08,650
How would you optimize the
design of a controller?

74
00:04:11,210 --> 00:04:15,550
Looking back to where we've
been, way back in Lab 4,

75
00:04:15,550 --> 00:04:19,220
ancient history, we looked at
how you could program the

76
00:04:19,220 --> 00:04:20,959
robot to approach a wall.

77
00:04:20,959 --> 00:04:24,330
And we saw that depending on how
you set up that system you

78
00:04:24,330 --> 00:04:27,920
could get very different
performances.

79
00:04:27,920 --> 00:04:32,510
And what we'd like to do is have
a way that we can design

80
00:04:32,510 --> 00:04:35,580
for performance without
actually building it.

81
00:04:35,580 --> 00:04:38,160
The kind of thing that we built
in the lab, Lab 4, was

82
00:04:38,160 --> 00:04:41,180
so easy that building it to
determine its behaviors was

83
00:04:41,180 --> 00:04:42,290
not a bad problem.

84
00:04:42,290 --> 00:04:47,960
But In general if you were
building a 777, there's more

85
00:04:47,960 --> 00:04:49,800
than one pole.

86
00:04:49,800 --> 00:04:54,280
And you wouldn't necessarily
want to test drive all of the

87
00:04:54,280 --> 00:04:56,820
bad configurations.

88
00:04:56,820 --> 00:04:59,070
So we'd like to be able to
understand that kind of a

89
00:04:59,070 --> 00:05:01,200
problem analytically.

90
00:05:01,200 --> 00:05:04,170
We'd like to be able
to analyze it.

91
00:05:04,170 --> 00:05:08,790
So using the different
representations you can

92
00:05:08,790 --> 00:05:11,200
generate a very concise
representation just thinking

93
00:05:11,200 --> 00:05:12,580
in terms of difference
equations.

94
00:05:12,580 --> 00:05:15,310
You all did this in Lab 4.

95
00:05:15,310 --> 00:05:18,950
And you get a single difference
equation that tells

96
00:05:18,950 --> 00:05:21,540
you in principle everything
there is that you could know,

97
00:05:21,540 --> 00:05:23,650
but not in a form that's
very easy to analyze.

98
00:05:27,520 --> 00:05:29,400
It's a bit better if you
translate the difference

99
00:05:29,400 --> 00:05:33,510
equation into a block diagram
because now you can see that

100
00:05:33,510 --> 00:05:36,220
this system of equations
has in fact two

101
00:05:36,220 --> 00:05:37,200
feedback loops in it.

102
00:05:37,200 --> 00:05:38,200
Two cycles.

103
00:05:38,200 --> 00:05:41,570
Two things that might
potentially generate

104
00:05:41,570 --> 00:05:44,820
persistent responses to
transient signals, which could

105
00:05:44,820 --> 00:05:46,770
then degrade performance.

106
00:05:46,770 --> 00:05:50,480
If the transient signal lasts
ten years it might be a bad

107
00:05:50,480 --> 00:05:51,020
controller.

108
00:05:51,020 --> 00:05:57,870
If the 777 hits turbulence
and never stabilizes

109
00:05:57,870 --> 00:06:00,060
that would be bad.

110
00:06:00,060 --> 00:06:04,950
If small disturbances got bigger
with time that might be

111
00:06:04,950 --> 00:06:06,770
bad, right?

112
00:06:06,770 --> 00:06:10,740
So we can see that this simple
controller described by these

113
00:06:10,740 --> 00:06:13,110
difference equations has
the potential to do

114
00:06:13,110 --> 00:06:14,180
that sort of stuff.

115
00:06:14,180 --> 00:06:18,780
And we'd like to understand,
when does it?

116
00:06:18,780 --> 00:06:22,560
The easiest way to think about
analyzing this is to focus

117
00:06:22,560 --> 00:06:27,810
first on the inner loop and ask
the question, what's the

118
00:06:27,810 --> 00:06:31,520
functional representation for
that box which we would call

119
00:06:31,520 --> 00:06:32,660
an accumulator?

120
00:06:32,660 --> 00:06:38,040
This box, this thing,
accumulates at its output, the

121
00:06:38,040 --> 00:06:41,110
sum of all the things that ever
came in so we call it an

122
00:06:41,110 --> 00:06:43,160
accumulator.

123
00:06:43,160 --> 00:06:45,420
So what's the functional
representation of an

124
00:06:45,420 --> 00:06:46,440
accumulator?

125
00:06:46,440 --> 00:06:48,160
Well, we just do polynomial
math.

126
00:06:48,160 --> 00:06:54,120
Easy so we can recognize from
the block diagram that the

127
00:06:54,120 --> 00:07:00,540
signal Y could be constructed
by applying R to W.

128
00:07:00,540 --> 00:07:03,590
But we can also see in the block
diagram that W is the

129
00:07:03,590 --> 00:07:09,680
sum of X and Y. And then if we
take the left hand side and

130
00:07:09,680 --> 00:07:13,010
the right hand side of this
double equation we get

131
00:07:13,010 --> 00:07:16,100
something that involves just X
and Y, which we can solve for

132
00:07:16,100 --> 00:07:20,530
the ratio Y over X. Which then
says that the ratio is R over

133
00:07:20,530 --> 00:07:22,550
(1 minus R).

134
00:07:22,550 --> 00:07:25,025
That's a functional
representation for the effect

135
00:07:25,025 --> 00:07:27,910
of the accumulation.

136
00:07:27,910 --> 00:07:31,460
That's also something that comes
up so frequently in the

137
00:07:31,460 --> 00:07:35,450
design of control systems
that we give it a name.

138
00:07:35,450 --> 00:07:38,550
We call this Black's Equation.

139
00:07:38,550 --> 00:07:43,120
And it's especially useful to
avoid these little trivial

140
00:07:43,120 --> 00:07:46,930
steps in algebra, to just
jump to the answer.

141
00:07:46,930 --> 00:07:48,630
So let's see that everybody's
following me.

142
00:07:51,740 --> 00:07:54,070
The equation for this box is
the thing that we will call

143
00:07:54,070 --> 00:07:55,710
Black's Equation.

144
00:07:55,710 --> 00:07:56,980
It's not mysterious.

145
00:07:56,980 --> 00:08:01,270
It's something that you could
derive, so derive it.

146
00:08:01,270 --> 00:08:04,350
Figure out the functional form
for the system that goes from

147
00:08:04,350 --> 00:08:10,330
X to Y and figure out which
of these forms is correct.

148
00:08:10,330 --> 00:08:13,730
(1) through (4), or (5) if none
of the above applies.

149
00:08:13,730 --> 00:08:16,120
So take 30 seconds, figure
out the answer.

150
00:08:16,120 --> 00:08:17,390
I'm going to ask you
to raise your hand.

151
00:08:17,390 --> 00:08:18,640
You're free to talk
to your neighbors.

152
00:09:11,990 --> 00:09:12,060
OK.

153
00:09:12,060 --> 00:09:13,560
So everybody who raised your
hands tell me what the

154
00:09:13,560 --> 00:09:15,610
right answer is.

155
00:09:15,610 --> 00:09:16,130
OK, wonderful.

156
00:09:16,130 --> 00:09:17,360
It's about 95%.

157
00:09:17,360 --> 00:09:17,770
No.

158
00:09:17,770 --> 00:09:20,463
It's about 100% correct out of
about 95% participation.

159
00:09:22,990 --> 00:09:26,800
So the answer you can form just
like we did before, no

160
00:09:26,800 --> 00:09:29,250
particular tricks, using
simple algebra.

161
00:09:29,250 --> 00:09:32,530
Simple algebra you get
F over (1 minus FG).

162
00:09:32,530 --> 00:09:34,820
The thing I want you to
recognize is kind of a

163
00:09:34,820 --> 00:09:37,610
graphical way of thinking
about that.

164
00:09:37,610 --> 00:09:39,630
And you can just memorize
F over (1 minus

165
00:09:39,630 --> 00:09:41,720
FG) and that's fine.

166
00:09:41,720 --> 00:09:44,540
But there's some interesting
things that the designer

167
00:09:44,540 --> 00:09:47,020
thinks about.

168
00:09:47,020 --> 00:09:51,180
This functional form is F,
that's the forward gain.

169
00:09:51,180 --> 00:09:54,055
That's the gain through the
system starting at the input

170
00:09:54,055 --> 00:09:57,230
and going directly
to the output.

171
00:09:57,230 --> 00:10:04,930
So this form says that the
closed loop gain, the system

172
00:10:04,930 --> 00:10:11,350
function that results when the
loop is closed, is just the

173
00:10:11,350 --> 00:10:15,820
forward gain, F, divided by
(1 minus the loop gain).

174
00:10:15,820 --> 00:10:21,340
The loop gain is the product
around the loop once.

175
00:10:21,340 --> 00:10:23,050
So that's just a convenient
way of thinking about it.

176
00:10:23,050 --> 00:10:26,660
Any time you have a feedback
system of this type you can

177
00:10:26,660 --> 00:10:28,830
think about the closed loop
response as being the forward

178
00:10:28,830 --> 00:10:34,660
path F divided by (1 minus
loop gain FG).

179
00:10:34,660 --> 00:10:37,810
So the answer was 1.

180
00:10:37,810 --> 00:10:40,930
And generally we'll see two
different ways of representing

181
00:10:40,930 --> 00:10:42,370
the system.

182
00:10:42,370 --> 00:10:44,620
Sometimes we'll represent it
with a plus as we did with the

183
00:10:44,620 --> 00:10:46,120
accumulator.

184
00:10:46,120 --> 00:10:47,960
Many times we'll represent
it with a minus.

185
00:10:47,960 --> 00:10:50,080
That's the way we think about
control problems.

186
00:10:50,080 --> 00:10:51,890
We put in the minus because we'd
like to think about the

187
00:10:51,890 --> 00:10:55,140
controller as trying to make
something go to 0.

188
00:10:55,140 --> 00:10:57,140
So when you take the difference,
that gives us an

189
00:10:57,140 --> 00:10:58,730
error signal.

190
00:10:58,730 --> 00:11:00,720
And then we can think about the
controller being the thing

191
00:11:00,720 --> 00:11:03,120
that drives that error to 0.

192
00:11:03,120 --> 00:11:06,450
But however you think about it,
there's two forms that are

193
00:11:06,450 --> 00:11:07,980
very closely related.

194
00:11:07,980 --> 00:11:10,420
They really differ by just a
minus sign which you could

195
00:11:10,420 --> 00:11:13,500
think of as just multiplying
G. So it's sort of like the

196
00:11:13,500 --> 00:11:17,020
right hand side is just
a minus G plugged into

197
00:11:17,020 --> 00:11:19,360
the left hand side.

198
00:11:19,360 --> 00:11:21,970
So then the way you use this
idea, you think about the

199
00:11:21,970 --> 00:11:24,690
block diagram and you say, OK I
can replace this thing with

200
00:11:24,690 --> 00:11:26,360
an equivalent system
which is R over (1

201
00:11:26,360 --> 00:11:32,480
minus R) and then repeat.

202
00:11:32,480 --> 00:11:36,500
So the R over (1 minus R) means
that, if you think about

203
00:11:36,500 --> 00:11:39,890
Black's Equation now for this
loop, you should think about

204
00:11:39,890 --> 00:11:46,840
the forward gain, K minus T, R
over (1 minus R), that's this.

205
00:11:46,840 --> 00:11:49,240
Divided by (1 plus
the loop gain).

206
00:11:49,240 --> 00:11:53,820
So 1 plus, and the loop gain,
well, the wire down here just

207
00:11:53,820 --> 00:11:55,480
has a gain of 1.

208
00:11:55,480 --> 00:11:56,625
So the loop gain and
the forward gain

209
00:11:56,625 --> 00:11:58,500
are the same thing.

210
00:11:58,500 --> 00:12:01,170
So you get this kind
of expression which

211
00:12:01,170 --> 00:12:02,200
simplifies to this.

212
00:12:02,200 --> 00:12:04,890
There's two things I want
you to see about this.

213
00:12:04,890 --> 00:12:10,270
First off, I want you to see
that even though the

214
00:12:10,270 --> 00:12:13,460
simple-minded way of plugging
in said that we should have

215
00:12:13,460 --> 00:12:20,420
got a quotient of quotients N
over D over N over D, it's

216
00:12:20,420 --> 00:12:22,690
simplified to a single ratio.

217
00:12:25,430 --> 00:12:29,700
If you design a system out of
just adders, gains and delays,

218
00:12:29,700 --> 00:12:31,550
that will always be true.

219
00:12:31,550 --> 00:12:33,740
There's a closure.

220
00:12:33,740 --> 00:12:37,750
It will always be the case
that the functional that

221
00:12:37,750 --> 00:12:40,560
represents a system of that
form will always have the

222
00:12:40,560 --> 00:12:43,920
property that it's a polynomial
in R divided by

223
00:12:43,920 --> 00:12:45,820
another polynomial in R,
that's just the way

224
00:12:45,820 --> 00:12:48,100
polynomials work.

225
00:12:48,100 --> 00:12:51,200
The other thing is that you
can now start to interpret

226
00:12:51,200 --> 00:12:56,140
what the behavior of this could
represent in a simpler

227
00:12:56,140 --> 00:12:59,740
form than thinking about this.

228
00:12:59,740 --> 00:13:03,820
So this kind of a representation
that leads to

229
00:13:03,820 --> 00:13:06,590
intuition about what the
behavior should be is very

230
00:13:06,590 --> 00:13:08,510
helpful when you're thinking
about design.

231
00:13:08,510 --> 00:13:13,850
And in particular this
particular thing says that if

232
00:13:13,850 --> 00:13:18,800
we think about a simpler system
that could generate

233
00:13:18,800 --> 00:13:23,690
that same response we can
generate some intuition for

234
00:13:23,690 --> 00:13:27,410
how we would like to set
the parameter k.

235
00:13:27,410 --> 00:13:33,030
So in particular this system
functional which we generated

236
00:13:33,030 --> 00:13:38,060
for this system could equally
apply to that system.

237
00:13:41,200 --> 00:13:42,450
Is that clear?

238
00:13:44,900 --> 00:13:50,610
So there's a numerator, which
I've represented here.

239
00:13:50,610 --> 00:13:52,810
There's a numerator, which
has an R in it and has

240
00:13:52,810 --> 00:13:54,010
a minus kT in it.

241
00:13:54,010 --> 00:13:56,890
I'm going to, for reasons that
you'll see in a minute, I'm

242
00:13:56,890 --> 00:13:59,360
going to call something P0
because it's the pole.

243
00:14:02,710 --> 00:14:05,620
So the numerator I've
represented here and this

244
00:14:05,620 --> 00:14:09,160
denominator has this form.

245
00:14:09,160 --> 00:14:11,830
And I wrote it that way because
this is the canonical

246
00:14:11,830 --> 00:14:14,000
form for the way of thinking
about a pole.

247
00:14:16,530 --> 00:14:19,120
So what I showed is that even
though this was a more

248
00:14:19,120 --> 00:14:24,910
complicated system you can think
about it as the cascade

249
00:14:24,910 --> 00:14:29,395
of a delay, a gain,
and a pole.

250
00:14:33,010 --> 00:14:36,270
The pole can be calculated from
the gain, the pole is the

251
00:14:36,270 --> 00:14:41,890
multiplier for R, so the
pole is (1 plus kT).

252
00:14:41,890 --> 00:14:46,350
So if I were to choose kT to
be minus 0.2 then the pole

253
00:14:46,350 --> 00:14:49,460
would be at 0.8.

254
00:14:49,460 --> 00:14:55,170
If the pole is at 0.8 then the
mode, the natural response to

255
00:14:55,170 --> 00:14:59,070
the pole, would have the
form P to the n.

256
00:14:59,070 --> 00:15:02,590
They always have the form
geometric P to the n, so it

257
00:15:02,590 --> 00:15:05,490
would look like 0.8 to the n.

258
00:15:05,490 --> 00:15:09,030
Because of the pre-multiplier of
1 minus P0 the whole thing

259
00:15:09,030 --> 00:15:12,460
gets multiplied by 0.2 and
because of the delay the whole

260
00:15:12,460 --> 00:15:15,240
thing gets shifted
to the right.

261
00:15:15,240 --> 00:15:19,440
The important thing is that by
thinking about manipulating

262
00:15:19,440 --> 00:15:24,070
this as an operator we can
recognize and simplify the

263
00:15:24,070 --> 00:15:25,580
form of the behavior.

264
00:15:25,580 --> 00:15:27,650
That gives us an intuitive
grasp over how

265
00:15:27,650 --> 00:15:29,170
to best choose kT.

266
00:15:31,710 --> 00:15:32,960
It's all clear?

267
00:15:35,270 --> 00:15:38,940
Now, the behaviors that we're
interested in are not always

268
00:15:38,940 --> 00:15:40,160
unit-sample responses.

269
00:15:40,160 --> 00:15:42,500
We do unit-sample responses
because they're the easiest

270
00:15:42,500 --> 00:15:44,800
possible thing we could
think of, right?

271
00:15:44,800 --> 00:15:51,230
A unit-sample is the simplest
non-zero signal.

272
00:15:51,230 --> 00:15:54,980
A unit-sample is the signal that
is different from 0 in

273
00:15:54,980 --> 00:15:56,980
exactly one place --

274
00:15:56,980 --> 00:15:59,950
the easiest possible place, 0.

275
00:15:59,950 --> 00:16:01,600
And it has its easy-as-possible

276
00:16:01,600 --> 00:16:02,830
non-zero value --

277
00:16:02,830 --> 00:16:04,060
1.

278
00:16:04,060 --> 00:16:06,590
So we focus on the unit-sample
signal because it's the

279
00:16:06,590 --> 00:16:09,090
easiest possible signal
we could think about.

280
00:16:09,090 --> 00:16:10,960
But when we're thinking about
behaviors we're often thinking

281
00:16:10,960 --> 00:16:13,810
about other things.

282
00:16:13,810 --> 00:16:16,900
Often we'll think about
the step response.

283
00:16:16,900 --> 00:16:18,300
Here I have illustrated
the way you would

284
00:16:18,300 --> 00:16:20,790
measure a step response.

285
00:16:20,790 --> 00:16:28,240
A step response is what would
happen if the output were

286
00:16:28,240 --> 00:16:35,080
initially 0, if we were at rest,
and suddenly we turned

287
00:16:35,080 --> 00:16:39,620
on a signal that was
constant at 1.

288
00:16:39,620 --> 00:16:42,740
That would happen in the robot
case if we started the robot

289
00:16:42,740 --> 00:16:46,730
close to the wall, at
rest, near 0 --

290
00:16:46,730 --> 00:16:53,220
so that the output signal is
close to 0, but the desired

291
00:16:53,220 --> 00:16:54,680
input was a meter behind.

292
00:16:57,550 --> 00:17:00,350
Then that would be an input
signal that started at time

293
00:17:00,350 --> 00:17:04,230
equals 0 equal to 1 and
persisted forever at 1.

294
00:17:04,230 --> 00:17:05,910
And the result then
would be what we

295
00:17:05,910 --> 00:17:08,230
call the step response.

296
00:17:08,230 --> 00:17:12,490
The step response is typically
easier to measure in the lab

297
00:17:12,490 --> 00:17:13,910
than is the unit-sample
response.

298
00:17:13,910 --> 00:17:15,880
So we use the unit-sample
response when we're thinking

299
00:17:15,880 --> 00:17:18,270
analytically, when we're doing
calculations, and we use the

300
00:17:18,270 --> 00:17:19,440
step response when
we're in the lab

301
00:17:19,440 --> 00:17:20,690
trying to measure something.

302
00:17:23,760 --> 00:17:27,089
And the whole theory wouldn't
be very useful if there

303
00:17:27,089 --> 00:17:30,780
weren't a close relationship
between those two things.

304
00:17:30,780 --> 00:17:33,780
This diagram illustrates
the relationship.

305
00:17:33,780 --> 00:17:38,020
If we think about a system H
for which we would like to

306
00:17:38,020 --> 00:17:41,930
find the step response, the step
response to that system

307
00:17:41,930 --> 00:17:43,930
is what would the system
do if you put the

308
00:17:43,930 --> 00:17:47,080
unit-step into the system.

309
00:17:47,080 --> 00:17:49,290
I've represented the unit-step
here as u[n].

310
00:17:52,220 --> 00:17:57,480
u[n], the signal that is 0 for
n less than 0, and 1 for n

311
00:17:57,480 --> 00:18:00,220
bigger than or equal to 0 --

312
00:18:00,220 --> 00:18:03,570
is just the accumulation of
the delta function, the

313
00:18:03,570 --> 00:18:06,290
unit-sample.

314
00:18:06,290 --> 00:18:13,420
So this system, the cascade of
an accumulator with H, would

315
00:18:13,420 --> 00:18:17,910
measure the step response of H
if it were driven with the

316
00:18:17,910 --> 00:18:19,160
sample signal.

317
00:18:21,490 --> 00:18:26,480
Because of the properties of
polynomials and because block

318
00:18:26,480 --> 00:18:31,920
diagrams follow the rules for
polynomials, we can flip these

319
00:18:31,920 --> 00:18:37,750
whenever the systems both start
at rest, and if we flip

320
00:18:37,750 --> 00:18:41,630
those we get a different
interpretation.

321
00:18:41,630 --> 00:18:46,260
What this says is that if you
were to take H and stimulate

322
00:18:46,260 --> 00:18:50,750
it with the unit-sample you
would get h, little h, which

323
00:18:50,750 --> 00:18:53,000
we would call the unit-sample
response because it's the

324
00:18:53,000 --> 00:18:57,010
response to the system when the
input is the unit-sample.

325
00:18:57,010 --> 00:19:00,480
So if you measured h with the
unit-sample rather then with

326
00:19:00,480 --> 00:19:05,520
the unit-step you would get the
unit-sample response from

327
00:19:05,520 --> 00:19:08,590
which you could generate the
step response by running it

328
00:19:08,590 --> 00:19:09,840
through an accumulator.

329
00:19:11,980 --> 00:19:15,040
So what that says is there's a
close association, there's a

330
00:19:15,040 --> 00:19:17,430
close relationship, between the
unit-sample response and

331
00:19:17,430 --> 00:19:18,780
the unit-step response.

332
00:19:18,780 --> 00:19:22,060
One is the accumulation
of the other.

333
00:19:22,060 --> 00:19:24,410
The unit-step response is the
accumulated response to the

334
00:19:24,410 --> 00:19:25,660
unit-sample response.

335
00:19:28,020 --> 00:19:32,140
So that means that in that
previous example where we saw

336
00:19:32,140 --> 00:19:36,280
that setting kT equal to minus
0.2 resulted in this

337
00:19:36,280 --> 00:19:39,970
unit-sample response, that
would correspond to this

338
00:19:39,970 --> 00:19:41,400
unit-step response.

339
00:19:41,400 --> 00:19:46,630
All you do is for every sample
you calculate, for this

340
00:19:46,630 --> 00:19:52,140
response you calculate the sum
of say, n equals 0, you would

341
00:19:52,140 --> 00:19:57,360
take the sum of all of the
previous answers in H[n].

342
00:19:57,360 --> 00:19:59,520
It's the accumulation.

343
00:19:59,520 --> 00:20:04,830
So it starts at 0 since the sum
of all those numbers is 0.

344
00:20:04,830 --> 00:20:09,930
Then at time 1 it becomes the
sum from here back so it

345
00:20:09,930 --> 00:20:11,860
becomes 0.2.

346
00:20:11,860 --> 00:20:14,160
Then here it's the sum
from here back.

347
00:20:14,160 --> 00:20:16,750
And if you add these all up
it becomes a number that

348
00:20:16,750 --> 00:20:17,860
approaches 1.

349
00:20:17,860 --> 00:20:19,250
Not surprisingly, right?

350
00:20:19,250 --> 00:20:22,460
If you've got a feedback system
and if you started the

351
00:20:22,460 --> 00:20:25,710
robot up against the wall and
the desired position was one

352
00:20:25,710 --> 00:20:31,020
meter behind you would
monotonically approach 1, OK?

353
00:20:31,020 --> 00:20:33,350
And what you can see is that if
you change the value of the

354
00:20:33,350 --> 00:20:40,590
pole, here I've changed the kT
from minus 0.2 which is what

355
00:20:40,590 --> 00:20:44,360
the previous answer was, to
minus 0.8, I've changed the

356
00:20:44,360 --> 00:20:47,880
value of the pole,
the unit-sample

357
00:20:47,880 --> 00:20:49,300
response got faster.

358
00:20:52,860 --> 00:20:55,290
And the unit-step response
also got faster.

359
00:20:55,290 --> 00:20:59,970
The point is there are different
kinds of performance

360
00:20:59,970 --> 00:21:03,300
metrics that we might want to
use, unit-sample response,

361
00:21:03,300 --> 00:21:07,260
unit-step response, but the
responses of all of them, you

362
00:21:07,260 --> 00:21:11,580
can tell something about the
response to all of them from

363
00:21:11,580 --> 00:21:13,350
the response of the unit-sample
signal.

364
00:21:13,350 --> 00:21:16,000
That's why we focus so much
on the unit-sample signal.

365
00:21:16,000 --> 00:21:17,740
It's not because it's the
most popular thing

366
00:21:17,740 --> 00:21:18,840
to use in the lab.

367
00:21:18,840 --> 00:21:22,830
It's because it's the easiest
thing to calculate with and it

368
00:21:22,830 --> 00:21:25,000
gives us insight into things
that we would like to measure

369
00:21:25,000 --> 00:21:27,690
in the lab.

370
00:21:27,690 --> 00:21:33,360
So for this very simple system
what you can show is that

371
00:21:33,360 --> 00:21:36,360
there's only a few possible
behaviors, a few

372
00:21:36,360 --> 00:21:40,160
categories of behaviors.

373
00:21:40,160 --> 00:21:46,370
If you were to choose kT to be
between 0 and 1, then the

374
00:21:46,370 --> 00:21:51,285
pole, which is (1 plus kT) would
also be between 0 and 1.

375
00:21:54,150 --> 00:21:58,840
Since the system has a single
pole you can say a lot about

376
00:21:58,840 --> 00:22:02,930
the response from the numerical
value of the pole.

377
00:22:02,930 --> 00:22:05,540
If the pole is between 0 and 1
then the response is going to

378
00:22:05,540 --> 00:22:08,700
be monotonic and converging.

379
00:22:11,510 --> 00:22:15,530
That results because the
unit-sample response was

380
00:22:15,530 --> 00:22:21,650
positive only and decayed
towards 0.

381
00:22:21,650 --> 00:22:32,380
Because it's positive only
it means monotonic --

382
00:22:32,380 --> 00:22:35,080
goes to 0 and makes
it converge.

383
00:22:35,080 --> 00:22:39,550
So you can infer properties
about the unit-step's response

384
00:22:39,550 --> 00:22:42,620
from properties of the pole
just like we could infer

385
00:22:42,620 --> 00:22:45,720
properties of the unit-sample
response.

386
00:22:45,720 --> 00:22:49,520
If you changed kT to be between
minus 2 and 1 you get

387
00:22:49,520 --> 00:22:52,010
a P0 that's between
minus 1 and 0.

388
00:22:52,010 --> 00:22:53,260
Again that's just
that equation.

389
00:22:55,750 --> 00:23:00,260
That says that the response
will be alternating.

390
00:23:00,260 --> 00:23:03,160
So the sign of the unit-sample
response goes

391
00:23:03,160 --> 00:23:04,410
positive then negative.

392
00:23:07,560 --> 00:23:10,650
It still converges in the sense
that the unit sample

393
00:23:10,650 --> 00:23:13,840
response approaches 0.

394
00:23:13,840 --> 00:23:17,200
And what that means for the unit
step-response is that the

395
00:23:17,200 --> 00:23:20,190
unit-step response will
converge toward 1.

396
00:23:28,730 --> 00:23:30,820
The sign doesn't alternate
around 0, the sign

397
00:23:30,820 --> 00:23:32,070
alternates around 1.

398
00:23:35,170 --> 00:23:37,370
So again, you can infer the
properties of the unit-step

399
00:23:37,370 --> 00:23:39,710
response from the properties of
the unit-sample response.

400
00:23:39,710 --> 00:23:44,480
And if you have kT that is less
than minus 2 then you get

401
00:23:44,480 --> 00:23:46,300
a P0 that's less than
minus 1 and

402
00:23:46,300 --> 00:23:47,980
that's a divergent response.

403
00:23:47,980 --> 00:23:51,200
So the point is you can infer
properties about the control

404
00:23:51,200 --> 00:23:56,120
system by thinking about the
poles of the system where here

405
00:23:56,120 --> 00:23:57,950
I've illustrated it for
a simple system that

406
00:23:57,950 --> 00:23:59,200
only has one pole.

407
00:24:01,780 --> 00:24:02,720
OK.

408
00:24:02,720 --> 00:24:03,760
I told you a bunch of facts.

409
00:24:03,760 --> 00:24:05,260
Now you figure out something.

410
00:24:05,260 --> 00:24:10,340
How would I choose k for this
system to get the quote, "best

411
00:24:10,340 --> 00:27:00,640
performance?"

412
00:27:00,640 --> 00:27:04,340
So which value of kT would give
the fastest convergence

413
00:27:04,340 --> 00:27:06,575
for the unit-sample signal?

414
00:27:11,670 --> 00:27:18,980
OK, participation is down but
the hit rate is still good.

415
00:27:18,980 --> 00:27:20,980
Virtually everybody who
volunteered to answer got the

416
00:27:20,980 --> 00:27:23,580
right answer.

417
00:27:23,580 --> 00:27:25,055
The most popular
answer was (2).

418
00:27:25,055 --> 00:27:27,950
Why is the answer (2)?

419
00:27:27,950 --> 00:27:29,870
What's the range of
possibilities

420
00:27:29,870 --> 00:27:31,120
that we could get?

421
00:27:35,560 --> 00:27:40,710
If we choose k, or kT, we
could choose kT to be--

422
00:27:40,710 --> 00:27:44,190
what's the range of kT
that we could use?

423
00:27:44,190 --> 00:27:46,260
Minus 2 to 0?

424
00:27:46,260 --> 00:27:51,060
[INAUDIBLE] we could use kT,
any real number, right?

425
00:27:51,060 --> 00:27:53,350
We couldn't use imaginary
numbers because that doesn't

426
00:27:53,350 --> 00:27:55,830
sort of make sense for
a real system.

427
00:27:55,830 --> 00:27:59,330
But we could choose
any real number.

428
00:27:59,330 --> 00:28:04,400
The real numbers map, according
to this chart, the

429
00:28:04,400 --> 00:28:06,710
real numbers map to a different
real number.

430
00:28:06,710 --> 00:28:09,670
If you choose kT you can figure
out where is the pole

431
00:28:09,670 --> 00:28:11,770
by that mapping.

432
00:28:11,770 --> 00:28:14,735
Where would you put the pole to
get the fastest response?

433
00:28:19,660 --> 00:28:22,230
If you have your choice of
putting the pole anywhere on

434
00:28:22,230 --> 00:28:24,400
this red line, that red line or
that red line, where would

435
00:28:24,400 --> 00:28:28,970
you put it and why?

436
00:28:33,870 --> 00:28:37,140
AUDIENCE: Just inside
the [INAUDIBLE]

437
00:28:37,140 --> 00:28:39,020
PROFESSOR: Putting it inside the
unit circle would probably

438
00:28:39,020 --> 00:28:40,862
be a good idea because?

439
00:28:40,862 --> 00:28:42,130
AUDIENCE: [UNINTELLIGIBLE]

440
00:28:42,130 --> 00:28:42,920
PROFESSOR: Yeah.

441
00:28:42,920 --> 00:28:44,170
If you didn't put it
inside the unit

442
00:28:44,170 --> 00:28:45,854
circle it wouldn't converge.

443
00:28:45,854 --> 00:28:48,060
That's right.

444
00:28:48,060 --> 00:28:49,820
So you like [UNINTELLIGIBLE]
the inside.

445
00:28:49,820 --> 00:28:52,450
Given the choice of anywhere
here and anywhere here which

446
00:28:52,450 --> 00:28:53,350
would you choose?

447
00:28:53,350 --> 00:28:54,050
How would you choose it?

448
00:28:54,050 --> 00:28:54,400
Yeah.

449
00:28:54,400 --> 00:28:55,280
AUDIENCE: Derive.

450
00:28:55,280 --> 00:28:57,414
PROFESSOR: Derive.

451
00:28:57,414 --> 00:28:58,908
And how do you get that?

452
00:28:58,908 --> 00:29:00,158
AUDIENCE: [UNINTELLIGIBLE]

453
00:29:03,390 --> 00:29:04,884
PROFESSOR: What would happen
if it was close to

454
00:29:04,884 --> 00:29:06,134
[UNINTELLIGIBLE]?

455
00:29:08,106 --> 00:29:10,630
It converges quite
quickly, right?

456
00:29:10,630 --> 00:29:13,470
Poles always converge
geometrically.

457
00:29:13,470 --> 00:29:16,600
The base of the geometric
is the pole value.

458
00:29:16,600 --> 00:29:19,150
So you'd like the pole to be
as small as possible to get

459
00:29:19,150 --> 00:29:22,140
the convergence as
fast as possible.

460
00:29:22,140 --> 00:29:24,620
That make sense to everybody?

461
00:29:24,620 --> 00:29:29,790
So in particular for this
example if you chose kT to be

462
00:29:29,790 --> 00:29:37,010
minus 1 in that limit then this
entire factor goes away.

463
00:29:37,010 --> 00:29:42,820
So the entire response
degenerates to R and R is not

464
00:29:42,820 --> 00:29:45,970
instantaneous but it's
pretty fast.

465
00:29:45,970 --> 00:29:49,730
What that says is that
you get to the final

466
00:29:49,730 --> 00:29:54,710
value in one step.

467
00:29:54,710 --> 00:29:58,860
So if the input consisted of
a unit sample, which has

468
00:29:58,860 --> 00:30:02,010
non-zero value only at 0, the
output would have non-zero

469
00:30:02,010 --> 00:30:09,150
value only at 1, right?

470
00:30:09,150 --> 00:30:13,440
Thinking about the way that
works in practice, think about

471
00:30:13,440 --> 00:30:15,310
the robot and think about
we're trying to

472
00:30:15,310 --> 00:30:17,000
drive toward the wall.

473
00:30:17,000 --> 00:30:23,570
If we made kT be minus one, and
just for the sake of being

474
00:30:23,570 --> 00:30:27,420
concrete let me say that
T is about 1/10.

475
00:30:27,420 --> 00:30:30,960
That's what the sampling period
is for the robots we

476
00:30:30,960 --> 00:30:32,880
use in the lab.

477
00:30:32,880 --> 00:30:38,720
If T were 1/10 then the best
k would be minus 10.

478
00:30:38,720 --> 00:30:42,550
And what that says is that if
we were 1 meter away from

479
00:30:42,550 --> 00:30:47,350
where we want to be we would set
the velocity to 10 meters

480
00:30:47,350 --> 00:30:48,620
per second.

481
00:30:48,620 --> 00:30:53,400
What that says is that if we
started 1 meter away from

482
00:30:53,400 --> 00:30:56,880
where we want to be, so this
is intended to represent

483
00:30:56,880 --> 00:31:01,390
position on the same axis that
this has showed, so if we

484
00:31:01,390 --> 00:31:07,630
started here and we wanted to be
here, time is plotted down.

485
00:31:07,630 --> 00:31:11,930
If we use the rule that we just
specified then we would

486
00:31:11,930 --> 00:31:15,400
set the velocity given this
condition which is 1 meter

487
00:31:15,400 --> 00:31:16,780
away from where we want to be.

488
00:31:16,780 --> 00:31:20,070
We would set the velocity
to be 10.

489
00:31:20,070 --> 00:31:25,380
If we set the velocity to be 10
then after 1 unit of time,

490
00:31:25,380 --> 00:31:28,820
after 1/10 of a second, we are
1 meter to the right, which

491
00:31:28,820 --> 00:31:31,130
just happens to be exactly
where we want to be.

492
00:31:33,910 --> 00:31:36,920
Had we chosen k to be bigger
we would have overshot.

493
00:31:36,920 --> 00:31:39,950
Had we chosen k to be smaller
we would have undershot.

494
00:31:39,950 --> 00:31:45,170
k equals 10 gave us precisely
the right answer so that we

495
00:31:45,170 --> 00:31:47,020
get there in one fell swoop.

496
00:31:47,020 --> 00:31:50,240
Then on the very next step we
would compute a velocity of 0

497
00:31:50,240 --> 00:31:53,670
because we are at where
we want to be so

498
00:31:53,670 --> 00:31:56,080
we would stay there.

499
00:31:56,080 --> 00:31:59,240
And that condition would
persist forever.

500
00:31:59,240 --> 00:32:03,350
The idea would be this simple
system provides a way that we

501
00:32:03,350 --> 00:32:05,700
could set the gain so we could
get to where we want

502
00:32:05,700 --> 00:32:07,580
to be in one step.

503
00:32:07,580 --> 00:32:08,830
It's hard to beat that.

504
00:32:12,010 --> 00:32:14,330
The problem that results and the
reason you didn't see that

505
00:32:14,330 --> 00:32:18,060
good behavior in the lab was
that the sensors in the robot

506
00:32:18,060 --> 00:32:21,260
don't work instantaneously.

507
00:32:21,260 --> 00:32:23,780
They introduce delay.

508
00:32:23,780 --> 00:32:26,920
And as an idealization of that
delay I want to think through

509
00:32:26,920 --> 00:32:28,250
the same problem.

510
00:32:28,250 --> 00:32:32,760
But now let's say that the
sensor delays the input to the

511
00:32:32,760 --> 00:32:36,100
sensor which is the output
of the system.

512
00:32:36,100 --> 00:32:40,980
Let's say that the sensor
introduces a delay of 1, so

513
00:32:40,980 --> 00:32:45,610
now instead of reporting d
sensed, which was d0[n], it

514
00:32:45,610 --> 00:32:46,860
reports d0[n minus 1].

515
00:32:50,960 --> 00:32:54,090
So now what would happen?

516
00:32:54,090 --> 00:33:00,250
Now with the delay, if I started
here and if by some

517
00:33:00,250 --> 00:33:06,750
mysterious process I was here
at time n equals 1, then I

518
00:33:06,750 --> 00:33:10,080
would calculate my
new velocity.

519
00:33:10,080 --> 00:33:12,130
What would be my new
velocity here?

520
00:33:12,130 --> 00:33:14,590
I'm right where I want to be.

521
00:33:14,590 --> 00:33:17,564
What would be my new velocity
if I assume that the sensor

522
00:33:17,564 --> 00:33:21,460
has a delay of [UNINTELLIGIBLE]?

523
00:33:21,460 --> 00:33:21,947
AUDIENCE: 10.

524
00:33:21,947 --> 00:33:23,410
PROFESSOR: 10.

525
00:33:23,410 --> 00:33:27,920
Because of the delay the sensor
is reporting that I'm a

526
00:33:27,920 --> 00:33:31,850
meter away from where
I want to be.

527
00:33:31,850 --> 00:33:34,870
So the controller calculates,
oh, I need to

528
00:33:34,870 --> 00:33:35,990
go forward a meter.

529
00:33:35,990 --> 00:33:38,710
I'll set the velocity to 10.

530
00:33:38,710 --> 00:33:42,320
So having set the velocity to 10
and then one step goes by,

531
00:33:42,320 --> 00:33:46,330
now we're completely
on the wrong side.

532
00:33:46,330 --> 00:33:49,210
That's what happens when you
put delay into the system.

533
00:33:49,210 --> 00:33:52,970
So because we're basing this
decision on where we were last

534
00:33:52,970 --> 00:33:57,220
time we go to the wrong place.

535
00:33:57,220 --> 00:33:59,880
So now we're here.

536
00:33:59,880 --> 00:34:02,264
What will the controller
say next?

537
00:34:02,264 --> 00:34:03,150
AUDIENCE: [UNINTELLIGIBLE]

538
00:34:03,150 --> 00:34:04,480
PROFESSOR: Stay here.

539
00:34:04,480 --> 00:34:07,650
You're in a great place.

540
00:34:07,650 --> 00:34:09,810
I'm really 1 meter too close.

541
00:34:09,810 --> 00:34:12,429
In fact I banged
into the wall.

542
00:34:12,429 --> 00:34:15,469
But the sensor is telling me I'm
exactly where I want to be

543
00:34:15,469 --> 00:34:16,719
so stay here.

544
00:34:20,080 --> 00:34:26,090
Say I didn't kill myself, what
will the velocity next be?

545
00:34:26,090 --> 00:34:27,199
Minus 10.

546
00:34:27,199 --> 00:34:29,610
So now I tell myself
to go back.

547
00:34:29,610 --> 00:34:30,860
That's probably a good move.

548
00:34:33,620 --> 00:34:38,130
But now I still think I'm too
close to the wall so I tell

549
00:34:38,130 --> 00:34:40,889
myself to continue to back up.

550
00:34:40,889 --> 00:34:44,639
The idea is that I get
poor performance.

551
00:34:44,639 --> 00:34:48,920
The delay had a devastating
effect on the way that the

552
00:34:48,920 --> 00:34:50,790
controller worked.

553
00:34:50,790 --> 00:34:54,190
Even though it's a tiny change
to the way the system works it

554
00:34:54,190 --> 00:34:57,060
has a devastating effect
on behavior.

555
00:34:57,060 --> 00:34:59,400
We'd like to be able to predict
that without having to

556
00:34:59,400 --> 00:35:01,240
measure it.

557
00:35:01,240 --> 00:35:03,690
Here's the same equations except
that I put a delay in

558
00:35:03,690 --> 00:35:05,440
the sensor.

559
00:35:05,440 --> 00:35:08,300
Here is the same block diagram
but I've represented a delay

560
00:35:08,300 --> 00:35:09,550
in the sensor path.

561
00:35:13,660 --> 00:35:17,060
So now the question is, what's
the new functional

562
00:35:17,060 --> 00:35:20,910
representation for that
control system?

563
00:37:53,740 --> 00:37:55,530
So what's the answer?

564
00:37:55,530 --> 00:37:59,365
Can you rate the functional form
for this system as one of

565
00:37:59,365 --> 00:38:02,195
(1), (2), (3), or (4), or
is it none of the above?

566
00:38:07,200 --> 00:38:12,610
About 1/3 participation and
about 100% correct.

567
00:38:12,610 --> 00:38:13,860
The answer's four.

568
00:38:16,360 --> 00:38:18,872
You get to use Black's Equation
or however you'd like

569
00:38:18,872 --> 00:38:20,530
to think about that.

570
00:38:20,530 --> 00:38:23,280
You can think about reducing the
inner loop the same as we

571
00:38:23,280 --> 00:38:26,930
did before and then think about
this as forward over (1

572
00:38:26,930 --> 00:38:30,760
plus loop gain) but now the loop
gain has R squared in it

573
00:38:30,760 --> 00:38:35,810
instead of R. We
get this form.

574
00:38:35,810 --> 00:38:39,410
How does this form differ from
when the R wasn't here?

575
00:38:39,410 --> 00:38:42,140
What's the difference between
R not there and R is there?

576
00:38:49,020 --> 00:38:49,275
What's the answer?

577
00:38:49,275 --> 00:38:50,525
Anyone?

578
00:38:54,415 --> 00:38:55,665
AUDIENCE: [UNINTELLIGIBLE]

579
00:38:59,265 --> 00:39:00,235
PROFESSOR: The squared term.

580
00:39:00,235 --> 00:39:04,940
So this term in the previous
form was just an R and in this

581
00:39:04,940 --> 00:39:08,720
form is a square and
so what's that do?

582
00:39:08,720 --> 00:39:10,330
What's the importance
of the fact that

583
00:39:10,330 --> 00:39:13,660
there's a square there?

584
00:39:13,660 --> 00:39:15,790
Two poles, right?

585
00:39:15,790 --> 00:39:18,850
We now have a polynomial in
the denominator that is

586
00:39:18,850 --> 00:39:21,750
quadratic in R.

587
00:39:21,750 --> 00:39:23,560
And what that's going to do is
it's going to give us two

588
00:39:23,560 --> 00:39:26,650
poles instead of one.

589
00:39:26,650 --> 00:39:28,250
The importance of that is that
now we're going to have to

590
00:39:28,250 --> 00:39:31,140
think through-- we previously
categorized what were all the

591
00:39:31,140 --> 00:39:34,600
behaviors you could
get from one pole.

592
00:39:34,600 --> 00:39:39,050
The behaviors you can get from
one pole were monotonic

593
00:39:39,050 --> 00:39:45,590
divergence, non-monotonic
alternating divergence,

594
00:39:45,590 --> 00:39:51,840
monotonic convergence,
alternating convergence.

595
00:39:51,840 --> 00:39:53,220
So there were four behaviors
that were

596
00:39:53,220 --> 00:39:54,590
possible with one pole.

597
00:39:54,590 --> 00:39:56,390
Now we have to think through
what are all the possible

598
00:39:56,390 --> 00:39:58,780
behaviors that we could
get with two poles.

599
00:39:58,780 --> 00:40:00,750
Different problem.

600
00:40:00,750 --> 00:40:03,530
Hopefully they're related.

601
00:40:03,530 --> 00:40:08,480
So here, the way we would find
out what the poles are is take

602
00:40:08,480 --> 00:40:15,730
this expression, substitute for
every R, 1 over z, turn

603
00:40:15,730 --> 00:40:18,410
the ratio of polynomials
in R into a ratio of

604
00:40:18,410 --> 00:40:19,660
polynomials in z.

605
00:40:22,700 --> 00:40:25,580
To do that in this case I had
to multiply numerator and

606
00:40:25,580 --> 00:40:26,880
denominator by z squared.

607
00:40:29,830 --> 00:40:33,190
Having done that I get a second
order polynomial in z

608
00:40:33,190 --> 00:40:35,960
in the bottom so there's two
poles which are the roots of

609
00:40:35,960 --> 00:40:38,960
that polynomial.

610
00:40:38,960 --> 00:40:40,420
And that's just a quadratic
equation.

611
00:40:43,100 --> 00:40:46,050
The interesting thing now is
to map out what are all the

612
00:40:46,050 --> 00:40:48,410
possible behaviors that that
system can give us.

613
00:40:51,350 --> 00:40:54,470
It's important to realize that's
a simple generalization

614
00:40:54,470 --> 00:40:57,600
of what we saw before.

615
00:40:57,600 --> 00:41:01,090
It will be the case that any
system that we construct out

616
00:41:01,090 --> 00:41:06,330
of adders, gains and delays will
have the property that we

617
00:41:06,330 --> 00:41:09,010
can write the system functional
as a ratio of

618
00:41:09,010 --> 00:41:16,930
polynomials in R. By the factor
theorem we will always

619
00:41:16,930 --> 00:41:20,340
be able to factor
the denominator.

620
00:41:20,340 --> 00:41:23,270
And by the notion of partial
fractions we'll always be able

621
00:41:23,270 --> 00:41:26,660
to write some complicated
expression like that in terms

622
00:41:26,660 --> 00:41:28,960
of a sum of parts.

623
00:41:28,960 --> 00:41:31,700
Each part being first order.

624
00:41:31,700 --> 00:41:34,960
The intuition we get from this
is that what we ought to do is

625
00:41:34,960 --> 00:41:40,630
factor the denominator, find
the poles, and associate a

626
00:41:40,630 --> 00:41:44,090
behavior with each
of those poles.

627
00:41:44,090 --> 00:41:48,080
Here's what the problem looks
like for the two pole problem.

628
00:41:48,080 --> 00:41:52,450
If we have the general form
given here and if we start by

629
00:41:52,450 --> 00:41:59,760
thinking about kT having a small
magnitude, if kT has a

630
00:41:59,760 --> 00:42:02,300
small magnitude then we have 1/2
plus or minus the square

631
00:42:02,300 --> 00:42:05,150
root of 1/2 squared.

632
00:42:05,150 --> 00:42:06,680
So that's 1/2 plus
or minus 1/2.

633
00:42:06,680 --> 00:42:08,990
That's 0 or 1.

634
00:42:08,990 --> 00:42:13,470
So the poles for this system,
if you make k be very small,

635
00:42:13,470 --> 00:42:17,040
the poles are at 0,
near 0 and near 1.

636
00:42:17,040 --> 00:42:21,140
Is that a good system response
or a bad system response?

637
00:42:21,140 --> 00:42:21,550
Bad.

638
00:42:21,550 --> 00:42:22,800
Why?

639
00:42:25,221 --> 00:42:27,797
Well, we're trying to think
through the behavior of the

640
00:42:27,797 --> 00:42:30,520
second order system by thinking
about the separate

641
00:42:30,520 --> 00:42:34,280
behaviors of each
of the poles.

642
00:42:34,280 --> 00:42:37,360
Is this a good pole
or a bad pole?

643
00:42:37,360 --> 00:42:39,860
Why?

644
00:42:39,860 --> 00:42:43,760
The response is always pole
[UNINTELLIGIBLE].

645
00:42:43,760 --> 00:42:48,800
The mode associated with the
response at a pole near one is

646
00:42:48,800 --> 00:42:51,380
something near one to the end.

647
00:42:51,380 --> 00:42:54,130
That never converges.

648
00:42:54,130 --> 00:42:57,520
If you start with some error
the error persists forever.

649
00:42:57,520 --> 00:43:00,030
Well, that's not good.

650
00:43:00,030 --> 00:43:06,410
If wind turbulence knocks you
into a decline in your

651
00:43:06,410 --> 00:43:12,230
airplane and it persists
forever, that's not good.

652
00:43:12,230 --> 00:43:14,440
You would like those
things to damp out.

653
00:43:14,440 --> 00:43:15,700
So this pole is bad.

654
00:43:15,700 --> 00:43:16,950
How about that pole?

655
00:43:20,150 --> 00:43:22,560
That one has a response
that decays quickly.

656
00:43:22,560 --> 00:43:24,870
But the problem is that when
you add the two pieces

657
00:43:24,870 --> 00:43:26,660
together, that was the reason
I showed you this

658
00:43:26,660 --> 00:43:30,450
decomposition, you can think
about the polynomial being

659
00:43:30,450 --> 00:43:34,280
factored and being broken
into a number of parts.

660
00:43:34,280 --> 00:43:37,180
The part that's associated with
the pole near one has a

661
00:43:37,180 --> 00:43:39,590
response that goes
for a long time.

662
00:43:39,590 --> 00:43:44,390
So that will asymptotically
dominate your response.

663
00:43:44,390 --> 00:43:47,950
So we refer to this as
a dominant pole.

664
00:43:47,950 --> 00:43:51,070
This pole dominates
the response.

665
00:43:51,070 --> 00:43:55,040
That's a way of inferring the
behavior of two poles from the

666
00:43:55,040 --> 00:43:59,820
sum of single poles.

667
00:43:59,820 --> 00:44:01,970
In this particular case there's
one pole that matters

668
00:44:01,970 --> 00:44:03,170
more than the other one.

669
00:44:03,170 --> 00:44:06,280
So we call that pole
the dominant pole.

670
00:44:06,280 --> 00:44:11,990
If you were to make kT more
negative, So here's the

671
00:44:11,990 --> 00:44:12,720
general form.

672
00:44:12,720 --> 00:44:16,510
If you make kT negative, you can
make the thing under the

673
00:44:16,510 --> 00:44:20,240
radical sign go towards 0.

674
00:44:20,240 --> 00:44:24,400
If you made the thing under the
radical sign go to 0 then

675
00:44:24,400 --> 00:44:26,790
you would get two
poles at 1/2.

676
00:44:30,780 --> 00:44:34,370
So here we would see that if kT
were minus 1/4, if kT were

677
00:44:34,370 --> 00:44:37,550
minus 1/4 we would have 1/2
squared, which is plus 1/4.

678
00:44:37,550 --> 00:44:41,150
Minus 1/4 would give us
0 under the radical.

679
00:44:41,150 --> 00:44:43,470
So we would get two
poles at 1/2.

680
00:44:43,470 --> 00:44:44,720
Is that good or bad?

681
00:44:47,870 --> 00:44:51,510
Well, it's better than the
previous example, right?

682
00:44:51,510 --> 00:44:55,040
Because each of those poles is
associated with the response

683
00:44:55,040 --> 00:44:59,410
where the error gets half what
it used to be on every step.

684
00:44:59,410 --> 00:45:00,660
So it converges.

685
00:45:06,560 --> 00:45:10,720
If going from 0 to minus 1/4 is
good, then going to minus

686
00:45:10,720 --> 00:45:12,900
1/2 might be better, right?

687
00:45:12,900 --> 00:45:19,680
If you continue that trend, say
you make kT be minus 1, if

688
00:45:19,680 --> 00:45:23,660
kT is minus 1 then you get
1/2 squared minus one.

689
00:45:23,660 --> 00:45:28,360
So 1/2 squared is 1/4, minus
1 would be minus 3/4.

690
00:45:28,360 --> 00:45:30,560
That gives us a complex
pole here.

691
00:45:33,140 --> 00:45:38,280
So we get two poles that are
right on the unit circle.

692
00:45:38,280 --> 00:45:39,110
What's that mean?

693
00:45:39,110 --> 00:45:41,900
That means oscillations.

694
00:45:41,900 --> 00:45:46,070
Oscillations is something you
can't get with one pole with a

695
00:45:46,070 --> 00:45:48,240
real system.

696
00:45:48,240 --> 00:45:51,750
Oscillations result from
a poll that has

697
00:45:51,750 --> 00:45:53,360
an imaginary component.

698
00:45:53,360 --> 00:45:55,160
If the system is real
you could only get

699
00:45:55,160 --> 00:46:00,330
such poles in pairs.

700
00:46:00,330 --> 00:46:02,540
So it's this pair that
makes sense for

701
00:46:02,540 --> 00:46:05,750
a real valued system.

702
00:46:05,750 --> 00:46:07,520
And that gives rise to
oscillations and that's

703
00:46:07,520 --> 00:46:11,140
exactly what we saw here.

704
00:46:11,140 --> 00:46:15,030
So we can associate the
oscillations that we saw in

705
00:46:15,030 --> 00:46:21,090
the simulated lab experiment
with poles that have imaginary

706
00:46:21,090 --> 00:46:22,340
components.

707
00:46:25,230 --> 00:46:28,550
So what would be the period of
the oscillation in the system

708
00:46:28,550 --> 00:46:33,620
given by 1/2 plus
j root 3 over 2?

709
00:46:50,735 --> 00:46:52,202
AUDIENCE: [INAUDIBLE]

710
00:46:52,202 --> 00:46:54,464
PROFESSOR: Excuse me?

711
00:46:54,464 --> 00:46:55,398
Excuse me?

712
00:46:55,398 --> 00:46:57,740
AUDIENCE: Where did the
root three come from?

713
00:46:57,740 --> 00:47:01,010
PROFESSOR: The previous page.

714
00:47:01,010 --> 00:47:04,190
If you substitute minus 1.

715
00:47:48,370 --> 00:47:49,625
So what's the period
of the oscillation?

716
00:47:53,286 --> 00:47:57,171
So the period's represented
by 5 converged into 6.

717
00:47:57,171 --> 00:48:00,440
So how do you get 6?

718
00:48:00,440 --> 00:48:03,130
The easiest way to think about
that is to think about the

719
00:48:03,130 --> 00:48:07,020
poles being expressed
in polar notation.

720
00:48:07,020 --> 00:48:09,770
The poles we previously said
were 1/2 plus or minus

721
00:48:09,770 --> 00:48:11,330
j root 3 over 2.

722
00:48:11,330 --> 00:48:15,390
That's the same as e plus
or minus j pi over 3.

723
00:48:15,390 --> 00:48:20,300
It's easier to use that form
because if you take that form,

724
00:48:20,300 --> 00:48:25,020
so if you think about e to the
j, what was it, 2 pi over 3?

725
00:48:27,760 --> 00:48:29,010
Pi over 3.

726
00:48:33,240 --> 00:48:39,150
So if you think about that form,
that's the pole, we can

727
00:48:39,150 --> 00:48:42,550
write that that way.

728
00:48:46,140 --> 00:48:50,190
Then the inside has
a magnitude of 1.

729
00:48:50,190 --> 00:48:53,360
So we can think about that just
being a magnitude of 1

730
00:48:53,360 --> 00:48:55,850
and an angle of pi over 3.

731
00:48:55,850 --> 00:49:00,010
So when you raise that to the n,
the magnitude to the n, one

732
00:49:00,010 --> 00:49:04,820
to the n is always 1, and the
angle raised to the n, it just

733
00:49:04,820 --> 00:49:07,010
increases linearly with n.

734
00:49:07,010 --> 00:49:12,230
So the angle goes from pi over
3 to 2pi over 3 to pi to 4pi

735
00:49:12,230 --> 00:49:15,030
over 3, et cetera.

736
00:49:15,030 --> 00:49:18,510
So you can think about this
going from pi over 3, 2pi over

737
00:49:18,510 --> 00:49:20,350
3, pi, 4, 5, 6.

738
00:49:22,930 --> 00:49:25,990
It takes n equals 6 to get
around to where it started so

739
00:49:25,990 --> 00:49:27,240
the period is 6.

740
00:49:32,750 --> 00:49:36,000
If you were to further change
the game, if you were to make

741
00:49:36,000 --> 00:49:40,960
it even more negative, the poles
would go outside the

742
00:49:40,960 --> 00:49:41,560
unit circle.

743
00:49:41,560 --> 00:49:44,100
And then what would happen?

744
00:49:44,100 --> 00:49:46,060
AUDIENCE: [UNINTELLIGIBLE]

745
00:49:46,060 --> 00:49:46,550
PROFESSOR: Right.

746
00:49:46,550 --> 00:49:49,500
So that's completely
unacceptable.

747
00:49:49,500 --> 00:49:52,470
The point is that by changing
the gain you can get any

748
00:49:52,470 --> 00:49:57,050
behavior on this figure which
is called the root locus.

749
00:49:57,050 --> 00:50:00,670
So root meaning the root
of a polynomial.

750
00:50:00,670 --> 00:50:04,480
Locus meaning the acceptable
values of points.

751
00:50:04,480 --> 00:50:08,280
So the root locus shows you
all the possible behaviors

752
00:50:08,280 --> 00:50:10,470
they could result from
this system.

753
00:50:10,470 --> 00:50:15,460
So given that root locus, how
would you choose k to make

754
00:50:15,460 --> 00:50:17,545
your system response as
fast as you could?

755
00:51:19,830 --> 00:51:22,250
So what value of kT
would you want?

756
00:51:22,250 --> 00:51:23,500
Everyone raise your hands.

757
00:51:26,122 --> 00:51:27,670
That's very good.

758
00:51:27,670 --> 00:51:31,780
So the most popular answer
is number (2).

759
00:51:31,780 --> 00:51:34,350
So why would the answer
be number (2)?

760
00:51:34,350 --> 00:51:35,600
What do you look at?

761
00:51:39,500 --> 00:51:40,496
Yeah?

762
00:51:40,496 --> 00:51:43,484
AUDIENCE: [INAUDIBLE]

763
00:51:43,484 --> 00:51:44,980
PROFESSOR: Remember what
we're trying to do.

764
00:51:44,980 --> 00:51:48,500
We're trying to infer properties
of the behavior of

765
00:51:48,500 --> 00:51:53,530
this second order system from
the pole locations.

766
00:51:53,530 --> 00:51:57,120
We know that there's an
expansion that lets us expand

767
00:51:57,120 --> 00:52:01,930
the system in terms of the sum
of two first order responses.

768
00:52:01,930 --> 00:52:06,130
The slowest of the first order
responses will dominate

769
00:52:06,130 --> 00:52:07,860
eventually.

770
00:52:07,860 --> 00:52:09,720
So what we need to
look at is the

771
00:52:09,720 --> 00:52:13,570
slowest of the two responses.

772
00:52:13,570 --> 00:52:15,940
We would like to know
of the two poles

773
00:52:15,940 --> 00:52:18,400
which one is the slowest?

774
00:52:18,400 --> 00:52:23,220
The slowest is the one that's
closest to the unit circle.

775
00:52:23,220 --> 00:52:27,690
So we would get the fastest
response when the slowest one

776
00:52:27,690 --> 00:52:29,110
is as fast as possible.

777
00:52:32,040 --> 00:52:39,480
As the poles initially go toward
each other from 0 to 1

778
00:52:39,480 --> 00:52:42,470
this one is getting faster, this
one is getting slower.

779
00:52:42,470 --> 00:52:45,460
So the slowest one
is this one.

780
00:52:45,460 --> 00:52:49,590
So the slowest one is fastest
when they meet.

781
00:52:49,590 --> 00:52:52,320
And then when they diverge
does the slowest one get

782
00:52:52,320 --> 00:52:54,636
faster or slower?

783
00:52:54,636 --> 00:52:56,500
It's already slower
because it gets

784
00:52:56,500 --> 00:52:59,980
closer to the unit circle.

785
00:52:59,980 --> 00:53:02,610
So you get the fastest response
whenever you get the

786
00:53:02,610 --> 00:53:08,050
two poles both colliding at 1/2
and that was the case that

787
00:53:08,050 --> 00:53:13,610
happened when kT was minus 1/4
from two or three slides ago.

788
00:53:13,610 --> 00:53:18,910
So the idea then is to try to
infer what would be the

789
00:53:18,910 --> 00:53:22,430
behavior of this higher order
system by thinking about the

790
00:53:22,430 --> 00:53:27,190
behaviors of the individual
components, here the poles.

791
00:53:27,190 --> 00:53:29,960
And what we saw was something
that's in fact a very

792
00:53:29,960 --> 00:53:31,500
important general trend.

793
00:53:31,500 --> 00:53:35,600
What we saw was that we first
analyzed the wall finder

794
00:53:35,600 --> 00:53:39,110
system assuming there was
no delay in the sensor.

795
00:53:39,110 --> 00:53:42,130
And we found that that system
was characterized by a single

796
00:53:42,130 --> 00:53:45,870
pole and we had the design
freedom of putting that pole

797
00:53:45,870 --> 00:53:49,650
anywhere we wanted to
on the real axis.

798
00:53:49,650 --> 00:53:54,390
And that allowed us to choose
the pole to be at 0 which gave

799
00:53:54,390 --> 00:53:57,200
terrific performance.

800
00:53:57,200 --> 00:53:59,580
The interesting thing that
happened when you add just one

801
00:53:59,580 --> 00:54:02,910
more pole by putting a delay
in the sensor, you make the

802
00:54:02,910 --> 00:54:05,560
system more complicated and
now you can't possibly get

803
00:54:05,560 --> 00:54:08,030
nearly so good behavior.

804
00:54:08,030 --> 00:54:11,460
The behavior is a lot worse
than it was before.

805
00:54:11,460 --> 00:54:15,390
And in fact, if you were to do
the same kind of analysis by

806
00:54:15,390 --> 00:54:22,550
putting yet another delay in the
sensor you would find even

807
00:54:22,550 --> 00:54:24,550
worse behavior.

808
00:54:24,550 --> 00:54:27,810
The idea then, the
generalization of the way the

809
00:54:27,810 --> 00:54:31,070
behaviors is working, generally
speaking adding

810
00:54:31,070 --> 00:54:35,650
delays inside a feedback loop
is a destabilizing thing.

811
00:54:35,650 --> 00:54:40,440
Generally as the number of
delays increases you end up

812
00:54:40,440 --> 00:54:44,730
having to back off on the
maximum gain that you can use

813
00:54:44,730 --> 00:54:49,020
because the system becomes
less stable.

814
00:54:49,020 --> 00:54:54,440
So the overall moral is that
delays are bad generally.

815
00:54:54,440 --> 00:54:56,560
I mean, you could concoct some
kind of a weird scheme where

816
00:54:56,560 --> 00:54:57,460
that wouldn't be true.

817
00:54:57,460 --> 00:54:59,820
But it's actually hard to
concoct such a weird scheme.

818
00:54:59,820 --> 00:55:02,270
In general, and in virtually
every physical system that

819
00:55:02,270 --> 00:55:05,090
you'll run into, adding
delays makes the

820
00:55:05,090 --> 00:55:07,030
system harder to stabilize.

821
00:55:07,030 --> 00:55:09,010
And that's the big message.

822
00:55:09,010 --> 00:55:13,390
And the system that we looked at
in the lab, the wall finder

823
00:55:13,390 --> 00:55:17,190
was actually quite hard
because the number

824
00:55:17,190 --> 00:55:18,820
of delays was large.

825
00:55:18,820 --> 00:55:21,810
If you try to track where delays
can enter the robot

826
00:55:21,810 --> 00:55:25,460
system they get in at very
many different places.

827
00:55:25,460 --> 00:55:28,270
In the physical sensor, in the
microprocessor, in the

828
00:55:28,270 --> 00:55:31,230
conversion from analog to
digital, there's a number of

829
00:55:31,230 --> 00:55:32,850
delays in that system.

830
00:55:32,850 --> 00:55:36,200
And that's why it becomes
hard to stabilize.

831
00:55:36,200 --> 00:55:36,780
OK.

832
00:55:36,780 --> 00:55:39,450
So that's the main content
for today.

833
00:55:39,450 --> 00:55:42,150
What I want to do is give you
one more practice question.

834
00:55:45,040 --> 00:55:47,710
The big problem that I want you
to think about from today

835
00:55:47,710 --> 00:55:51,870
is how do you characterize
performance?

836
00:55:51,870 --> 00:55:55,940
When we had a single pole
performance was easy to talk

837
00:55:55,940 --> 00:56:02,170
about because performance was
diverging monotonically,

838
00:56:02,170 --> 00:56:07,730
diverging alternating,
converging monotonically,

839
00:56:07,730 --> 00:56:09,220
converging alternating.

840
00:56:09,220 --> 00:56:10,970
There were four kinds
of behaviors.

841
00:56:10,970 --> 00:56:13,750
When we went to second order
we saw some new behaviors.

842
00:56:13,750 --> 00:56:17,230
It could become oscillatory.

843
00:56:17,230 --> 00:56:20,040
What I'd like you to do now is
think not just about those

844
00:56:20,040 --> 00:56:21,600
properties but many
other properties.

845
00:56:21,600 --> 00:56:23,190
So here's some questions.

846
00:56:23,190 --> 00:56:26,150
Think about the system on the
top and I'd like you to infer

847
00:56:26,150 --> 00:56:28,950
properties about that system.

848
00:56:28,950 --> 00:56:34,350
In particular, does this system
have three poles?

849
00:56:34,350 --> 00:56:37,360
Is the unit sample response, is
there a way to write that

850
00:56:37,360 --> 00:56:41,630
as the sum of three geometric
sequences?

851
00:56:41,630 --> 00:56:44,430
What's the unit sample
response?

852
00:56:44,430 --> 00:56:47,220
And is one of the poles
that z equals 1?

853
00:56:47,220 --> 00:56:50,080
So think about the system,
think about five ways of

854
00:56:50,080 --> 00:56:52,890
characterizing it and tell
me how many of those five

855
00:56:52,890 --> 00:56:54,230
characterizations is correct.

856
01:00:58,700 --> 01:01:00,045
So how many of the properties
are true?

857
01:01:03,733 --> 01:01:04,983
AUDIENCE: [UNINTELLIGIBLE]

858
01:01:12,535 --> 01:01:15,480
Probably 2/3 correct?

859
01:01:15,480 --> 01:01:16,730
PROFESSOR: How many poles?

860
01:01:20,190 --> 01:01:23,355
How do you get three?

861
01:01:23,355 --> 01:01:24,834
Where are the poles?

862
01:01:24,834 --> 01:01:27,792
AUDIENCE: [UNINTELLIGIBLE]

863
01:01:27,792 --> 01:01:28,285
PROFESSOR: How do
I find poles?

864
01:01:28,285 --> 01:01:30,257
What do I do?

865
01:01:30,257 --> 01:01:31,736
Yes?

866
01:01:31,736 --> 01:01:35,187
AUDIENCE: You use Black's
Equation [UNINTELLIGIBLE]

867
01:01:35,187 --> 01:01:38,110
the top you can express
[UNINTELLIGIBLE]

868
01:01:38,110 --> 01:01:41,540
so use Black's Equation to
express the system function as

869
01:01:41,540 --> 01:01:44,480
R cubed over (1 minus
R-cubed) then--

870
01:01:44,480 --> 01:01:44,920
PROFESSOR: That's right.

871
01:01:44,920 --> 01:01:49,680
AUDIENCE: --the denominator
as an order of 3 and

872
01:01:49,680 --> 01:01:52,080
you combine the 3s.

873
01:01:52,080 --> 01:01:55,060
PROFESSOR: So a little more
formally, we would take this

874
01:01:55,060 --> 01:02:01,160
thing and we would rewrite that
with R goes to 1 over z.

875
01:02:01,160 --> 01:02:09,070
So we get 1 over z cubed, 1
minus (1 over z cubed), which

876
01:02:09,070 --> 01:02:12,810
is then, clearing the z cubes
we would get 1 over

877
01:02:12,810 --> 01:02:14,820
(z cubed minus 1).

878
01:02:17,340 --> 01:02:19,290
How many poles?

879
01:02:19,290 --> 01:02:19,890
Three.

880
01:02:19,890 --> 01:02:22,774
What are the poles of
z cubed minus 1?

881
01:02:25,750 --> 01:02:28,230
Three poles in what?

882
01:02:28,230 --> 01:02:29,718
AUDIENCE: 0 [UNINTELLIGIBLE]

883
01:02:33,190 --> 01:02:33,686
PROFESSOR: And so let's vote.

884
01:02:33,686 --> 01:02:35,174
Let's take a vote.

885
01:02:35,174 --> 01:02:39,142
There's two poles
at z equals 1.

886
01:02:39,142 --> 01:02:41,622
Yes?

887
01:02:41,622 --> 01:02:42,614
No?

888
01:02:42,614 --> 01:02:43,606
AUDIENCE: [UNINTELLIGIBLE]

889
01:02:43,606 --> 01:02:45,094
PROFESSOR: Why not?

890
01:02:45,094 --> 01:02:46,344
AUDIENCE: [UNINTELLIGIBLE]

891
01:02:48,100 --> 01:02:51,050
PROFESSOR: So there's
two poles.

892
01:02:51,050 --> 01:02:53,550
I made a [UNINTELLIGIBLE]
plane.

893
01:02:53,550 --> 01:02:54,800
Where's the poles?

894
01:02:57,340 --> 01:03:00,140
Well, you could factor
it, right?

895
01:03:00,140 --> 01:03:02,690
If you factored it you'd
find that there is

896
01:03:02,690 --> 01:03:06,680
a pole at 1, right?

897
01:03:06,680 --> 01:03:10,170
But then there's two more
poles like that.

898
01:03:14,480 --> 01:03:20,290
So the poles are the three
roots of 1, which can be

899
01:03:20,290 --> 01:03:30,210
written like 1 e to the j, 2pi
over 3, and e to the j minus

900
01:03:30,210 --> 01:03:31,460
2pi over 3.

901
01:03:35,520 --> 01:03:37,070
Which pole was the
dominant pole?

902
01:03:37,070 --> 01:03:38,320
AUDIENCE: [UNINTELLIGIBLE]

903
01:03:43,022 --> 01:03:46,130
PROFESSOR: OK, bad question.

904
01:03:46,130 --> 01:03:49,016
What's a better question?

905
01:03:49,016 --> 01:03:50,350
AUDIENCE: How many dominant
poles are there?

906
01:03:50,350 --> 01:03:51,970
PROFESSOR: How many dominant
poles are there?

907
01:03:51,970 --> 01:03:53,100
That's a much better
question, yes.

908
01:03:53,100 --> 01:03:55,130
There's sort of three
poles that are

909
01:03:55,130 --> 01:03:56,630
equally dominant, right?

910
01:03:56,630 --> 01:04:00,420
They all have the
same magnitude.

911
01:04:00,420 --> 01:04:02,000
Why do we talk about
dominant poles?

912
01:04:02,000 --> 01:04:03,250
What are dominant
poles good for?

913
01:04:07,610 --> 01:04:12,510
If I told you that I had a pole
at 3 and a pole at minus

914
01:04:12,510 --> 01:04:14,110
1, which one's the
dominant pole?

915
01:04:14,110 --> 01:04:14,590
AUDIENCE: 3.

916
01:04:14,590 --> 01:04:16,510
PROFESSOR: Why?

917
01:04:16,510 --> 01:04:17,000
AUDIENCE: Greater magnitude.

918
01:04:17,000 --> 01:04:18,620
PROFESSOR: Greater magnitude.

919
01:04:18,620 --> 01:04:21,440
Why do we care?

920
01:04:21,440 --> 01:04:24,300
We don't care, right?

921
01:04:24,300 --> 01:04:25,930
What's good about the
dominant pole?

922
01:04:31,690 --> 01:04:35,320
Well, we can write this response
as something that

923
01:04:35,320 --> 01:04:39,400
looks like three to the n
plus minus 1 to the n.

924
01:04:39,400 --> 01:04:42,260
If you let n get big enough
the only one that

925
01:04:42,260 --> 01:04:45,240
matters is 3 to the n.

926
01:04:45,240 --> 01:04:49,210
So if all you care about is
exactly how the plane was

927
01:04:49,210 --> 01:04:53,710
flying the instant before it hit
the ground then you would

928
01:04:53,710 --> 01:04:55,990
only need to worry
about long time.

929
01:04:55,990 --> 01:04:58,150
And if you only worry about
long time you only need to

930
01:04:58,150 --> 01:05:02,030
worry about the pole that's
worst behaved.

931
01:05:02,030 --> 01:05:03,770
That's where the concept
comes from.

932
01:05:03,770 --> 01:05:06,740
So none of these poles are
particularly worse behaved

933
01:05:06,740 --> 01:05:08,810
than the others.

934
01:05:08,810 --> 01:05:10,510
What's the unit-sample response

935
01:05:10,510 --> 01:05:11,760
associated with that pole?

936
01:05:17,420 --> 01:05:20,998
We have a name for that
[INAUDIBLE] right?

937
01:05:20,998 --> 01:05:21,992
AUDIENCE: It's huge.

938
01:05:21,992 --> 01:05:25,040
PROFESSOR: It's
[UNINTELLIGIBLE].

939
01:05:25,040 --> 01:05:26,520
What's the unit-sample response

940
01:05:26,520 --> 01:05:27,780
associated with this pole?

941
01:05:33,280 --> 01:05:37,466
Well, it's got a complex
value right?

942
01:05:37,466 --> 01:05:40,870
So the unit-sample response
associated with that pole is e

943
01:05:40,870 --> 01:05:47,050
to the j 2 pi over 3 n.

944
01:05:47,050 --> 01:05:49,430
That's a complex number.

945
01:05:49,430 --> 01:05:56,040
That's 1 at time 0 and e to the
j 2 pi over 3n at time 1

946
01:05:56,040 --> 01:06:01,420
and e to the j 4 pi over
3 at time two.

947
01:06:01,420 --> 01:06:04,950
So it goes from here at 0 to
here at 1 to here at 2,3, 4,

948
01:06:04,950 --> 01:06:06,200
5, 6, 7, 8.

949
01:06:08,950 --> 01:06:15,100
What's the period
of this pole?

950
01:06:15,100 --> 01:06:18,230
What's the period of the
unit-sample responses

951
01:06:18,230 --> 01:06:20,550
associated with that pole?

952
01:06:20,550 --> 01:06:21,440
3.

953
01:06:21,440 --> 01:06:23,580
Because it takes 3
to get around to

954
01:06:23,580 --> 01:06:24,650
where you started again.

955
01:06:24,650 --> 01:06:28,392
What's the period
of this pole?

956
01:06:28,392 --> 01:06:28,853
3.

957
01:06:28,853 --> 01:06:30,697
You just spin around backward.

958
01:06:30,697 --> 01:06:31,485
What's the period
of the response

959
01:06:31,485 --> 01:06:34,746
associated with that pole?

960
01:06:34,746 --> 01:06:36,162
Bad question.

961
01:06:36,162 --> 01:06:38,530
All right, what's a
better question?

962
01:06:38,530 --> 01:06:40,254
Is there a period associated
with it?

963
01:06:40,254 --> 01:06:42,614
You could say that period
1 [UNINTELLIGIBLE]

964
01:06:42,614 --> 01:06:43,864
definition of period.

965
01:06:46,390 --> 01:06:48,350
What's the period
of this pole?

966
01:06:51,822 --> 01:06:55,130
Dumb question, right?

967
01:06:55,130 --> 01:06:58,150
Period implies repeat.

968
01:06:58,150 --> 01:07:01,840
If the response repeats itself
after some time then we would

969
01:07:01,840 --> 01:07:03,090
say the response is periodic.

970
01:07:06,450 --> 01:07:09,270
Neither of those poles,
well, the minus 1 is.

971
01:07:09,270 --> 01:07:13,020
Is the minus 1 pole periodic?

972
01:07:13,020 --> 01:07:15,285
Yes.

973
01:07:15,285 --> 01:07:18,907
What's the difference between
periodic and alternation?

974
01:07:18,907 --> 01:07:21,160
Does a [UNINTELLIGIBLE]

975
01:07:21,160 --> 01:07:22,090
alternate?

976
01:07:22,090 --> 01:07:23,070
AUDIENCE: Yes.

977
01:07:23,070 --> 01:07:24,215
PROFESSOR: Does it oscillate?

978
01:07:24,215 --> 01:07:26,019
AUDIENCE: No.

979
01:07:26,019 --> 01:07:28,420
PROFESSOR: Bad question.

980
01:07:28,420 --> 01:07:33,410
Alternate is a word that we
invented for one pole.

981
01:07:33,410 --> 01:07:35,730
Because the response
alternated in sine.

982
01:07:35,730 --> 01:07:39,350
The unit-sample response in one
pole, where the pole is a

983
01:07:39,350 --> 01:07:41,670
negative number, alternated
in sine.

984
01:07:41,670 --> 01:07:44,790
So we gave that a name.

985
01:07:44,790 --> 01:07:47,310
Alternation is not necessarily
something that we would like

986
01:07:47,310 --> 01:07:50,210
to associate with a higher
order system.

987
01:07:50,210 --> 01:07:53,220
Periodic is perfectly reasonable
to talk about for a

988
01:07:53,220 --> 01:07:54,580
higher order system.

989
01:07:54,580 --> 01:07:58,920
Periodic merely means that if y
of n were a periodic signal

990
01:07:58,920 --> 01:08:04,360
that I could express y of
n plus n as y of n.

991
01:08:04,360 --> 01:08:05,610
That would be periodic.

992
01:08:07,610 --> 01:08:10,320
If the thing repeats itself we
would say it's periodic.

993
01:08:10,320 --> 01:08:12,650
So the point of going over this
stuff is just to give you

994
01:08:12,650 --> 01:08:15,670
some exercise in thinking
about how to think about

995
01:08:15,670 --> 01:08:18,109
properties of systems.

996
01:08:18,109 --> 01:08:19,830
We develop properties initially

997
01:08:19,830 --> 01:08:23,290
thinking about one pole.

998
01:08:23,290 --> 01:08:26,010
Those properties were easy.

999
01:08:26,010 --> 01:08:30,910
Converging, diverging,
monotonic, alternating.

1000
01:08:30,910 --> 01:08:34,729
When we try to think about
corresponding properties of

1001
01:08:34,729 --> 01:08:38,040
higher order systems we can't
simply map the simple

1002
01:08:38,040 --> 01:08:40,090
properties of first order
systems into the other.

1003
01:08:40,090 --> 01:08:42,189
We have to think about more
complicated things.

1004
01:08:42,189 --> 01:08:44,520
Then we think about things
like dominant.

1005
01:08:44,520 --> 01:08:46,760
If one of the poles has a bigger
magnitude than the

1006
01:08:46,760 --> 01:08:49,870
other then for large times we
can ignore the smaller one.

1007
01:08:55,420 --> 01:08:56,795
What happens for short time?

1008
01:09:00,149 --> 01:09:03,850
Does this response monotonically
increase with

1009
01:09:03,850 --> 01:09:05,210
time for all time?

1010
01:09:11,792 --> 01:09:13,250
No.

1011
01:09:13,250 --> 01:09:16,892
Since the response associated
with minus 1 alternates in

1012
01:09:16,892 --> 01:09:23,399
sine, for short times, for times
with n close to 0, that

1013
01:09:23,399 --> 01:09:26,750
can be just as important
as this one.

1014
01:09:26,750 --> 01:09:32,439
So the dominant pole idea tells
you how things work when

1015
01:09:32,439 --> 01:09:34,590
you have large times.

1016
01:09:34,590 --> 01:09:36,819
It doesn't necessarily tell you
how things work when you

1017
01:09:36,819 --> 01:09:38,020
have small times.

1018
01:09:38,020 --> 01:09:42,060
How about, the unit sample
response is the sum of three

1019
01:09:42,060 --> 01:09:42,790
geometrics.

1020
01:09:42,790 --> 01:09:44,040
Yes or no?

1021
01:09:46,865 --> 01:09:48,320
What are the three geometrics?

1022
01:09:53,170 --> 01:09:54,625
And the answer to that's yes.

1023
01:09:54,625 --> 01:09:55,110
That's very important.

1024
01:09:55,110 --> 01:10:01,720
The three geometrics over here
are this pole to the n plus

1025
01:10:01,720 --> 01:10:08,830
this pole to the n plus
something that goes with this

1026
01:10:08,830 --> 01:10:12,770
other pole to the n.

1027
01:10:12,770 --> 01:10:14,840
Now it's a weighted sum but
the weights are not

1028
01:10:14,840 --> 01:10:16,090
necessarily 0.

1029
01:10:21,120 --> 01:10:23,440
The slide that I showed you
for the partial fraction

1030
01:10:23,440 --> 01:10:26,540
decomposition, you can always
write a higher order system as

1031
01:10:26,540 --> 01:10:29,460
a sum of first order factors.

1032
01:10:29,460 --> 01:10:31,940
That's the partial fraction
expansion.

1033
01:10:31,940 --> 01:10:34,070
That doesn't mean the weights
are all unity.

1034
01:10:36,830 --> 01:10:39,720
Number (2), can you write the
unit-sample response as the

1035
01:10:39,720 --> 01:10:41,470
sum of three geometric
signals?

1036
01:10:41,470 --> 01:10:41,860
Yes.

1037
01:10:41,860 --> 01:10:44,290
There it is.

1038
01:10:44,290 --> 01:10:46,390
And if you're really good at
complex math you could find

1039
01:10:46,390 --> 01:10:48,530
out a, b and c.

1040
01:10:48,530 --> 01:10:50,960
And that would tell you the
unit-sample response and that

1041
01:10:50,960 --> 01:10:52,570
would tell you the answer
to (3) and (4).

1042
01:10:52,570 --> 01:10:54,407
Is the unit-sample response 0,
0, 0, 1, 0, 0, 0, 1, 0, 0, 0,

1043
01:10:54,407 --> 01:10:56,738
1, 0, 0, 0, 1?

1044
01:10:56,738 --> 01:10:59,410
Or 1, 0, 0, 0, 1, 1 --

1045
01:10:59,410 --> 01:11:01,110
whatever.

1046
01:11:01,110 --> 01:11:05,140
Is it one of those two or
something different?

1047
01:11:05,140 --> 01:11:09,076
And how do you figure it out?

1048
01:11:09,076 --> 01:11:10,060
How do you figure
out the unit?

1049
01:11:10,060 --> 01:11:11,310
Is the unit-sample response--

1050
01:11:13,560 --> 01:11:14,524
Is number (2) correct?

1051
01:11:14,524 --> 01:11:17,174
Is the unit-sample response
0, 0, 0, 1, 0, 0,

1052
01:11:17,174 --> 01:11:18,424
0, 1, 0, 0, 0, 1--

1053
01:11:21,680 --> 01:11:23,390
Yes?

1054
01:11:23,390 --> 01:11:26,100
How do you know that?

1055
01:11:26,100 --> 01:11:27,841
You could solve that equation.

1056
01:11:32,611 --> 01:11:34,719
Is there an easier way?

1057
01:11:34,719 --> 01:11:36,156
Yeah?

1058
01:11:36,156 --> 01:11:38,551
AUDIENCE: I wrote it as
a difference equation.

1059
01:11:38,551 --> 01:11:39,988
PROFESSOR: Just write the
difference equation.

1060
01:11:39,988 --> 01:11:40,946
Exactly.

1061
01:11:40,946 --> 01:11:43,820
So even though I bad-mouth
difference equations a lot,

1062
01:11:43,820 --> 01:11:45,900
here it's easy.

1063
01:11:45,900 --> 01:11:49,480
In fact, you can see
it in the network.

1064
01:11:49,480 --> 01:11:51,500
Thinking about the difference
equation would be easy.

1065
01:11:51,500 --> 01:11:53,500
Thinking about the network
would be easy.

1066
01:11:53,500 --> 01:11:55,270
If we think about the
unit-sample response of this

1067
01:11:55,270 --> 01:12:00,330
thing started at rest, rest
means this is 0, this is 0,

1068
01:12:00,330 --> 01:12:03,270
and this is 0 initially.

1069
01:12:03,270 --> 01:12:08,060
Unit-sample response means
this becomes 1 at time 0.

1070
01:12:08,060 --> 01:12:13,000
At time 0 this is 1, this is
0, this is 0, this is 0.

1071
01:12:13,000 --> 01:12:27,070
So if I think about what's the
time response look like and I

1072
01:12:27,070 --> 01:12:31,820
did plus, this is
0, 1, 0, 0, 0.

1073
01:12:31,820 --> 01:12:33,610
The first answer is 0.

1074
01:12:33,610 --> 01:12:34,170
Clock ticks.

1075
01:12:34,170 --> 01:12:35,420
What happens?

1076
01:12:38,450 --> 01:12:40,394
This is 1.

1077
01:12:40,394 --> 01:12:43,630
This becomes 1.

1078
01:12:43,630 --> 01:12:44,570
This doesn't change.

1079
01:12:44,570 --> 01:12:45,510
That doesn't change.

1080
01:12:45,510 --> 01:12:47,620
This goes to 0.

1081
01:12:47,620 --> 01:12:48,640
0 comes around here.

1082
01:12:48,640 --> 01:12:50,270
That goes to 0.

1083
01:12:50,270 --> 01:12:52,640
So that's the answer
at time 1.

1084
01:12:52,640 --> 01:12:55,542
What happens at time 2?

1085
01:12:55,542 --> 01:12:57,070
Just keep working it.

1086
01:12:57,070 --> 01:13:01,850
The clock ticks, this goes to 1,
this goes to 0, these stay

1087
01:13:01,850 --> 01:13:05,330
0, this stays 0.

1088
01:13:05,330 --> 01:13:07,870
That's the answer at
time equals 2.

1089
01:13:07,870 --> 01:13:09,790
Now the clock ticks.

1090
01:13:09,790 --> 01:13:13,480
Now this comes over to here,
that means it comes back here.

1091
01:13:13,480 --> 01:13:14,450
This is still 0.

1092
01:13:14,450 --> 01:13:17,750
That comes to 1.

1093
01:13:17,750 --> 01:13:19,520
That's the answer for time 3.

1094
01:13:19,520 --> 01:13:20,940
Now the clock ticks.

1095
01:13:20,940 --> 01:13:22,010
And you can see the whole
thing would just

1096
01:13:22,010 --> 01:13:24,000
repeat itself now.

1097
01:13:24,000 --> 01:13:27,170
Is the response periodic?

1098
01:13:27,170 --> 01:13:28,070
Yes.

1099
01:13:28,070 --> 01:13:29,200
The response is periodic.

1100
01:13:29,200 --> 01:13:31,118
What's the period?

1101
01:13:31,118 --> 01:13:32,086
3.

1102
01:13:32,086 --> 01:13:34,506
And I can see [INAUDIBLE] if
the [INAUDIBLE] there it's

1103
01:13:34,506 --> 01:13:37,530
going to have be related to
the period over here.

1104
01:13:37,530 --> 01:13:39,290
These periods are not same.

1105
01:13:39,290 --> 01:13:42,300
This period is 3, this period is
3, this period is 1, if you

1106
01:13:42,300 --> 01:13:44,720
want to call that a period.

1107
01:13:44,720 --> 01:13:47,390
But they are related.

1108
01:13:47,390 --> 01:13:47,780
OK.

1109
01:13:47,780 --> 01:13:52,900
The point of this exercise is
to illustrate two things.

1110
01:13:52,900 --> 01:13:57,680
We inferred properties of first
order systems by looking

1111
01:13:57,680 --> 01:14:01,700
at a single pole which for a
real system could only behave

1112
01:14:01,700 --> 01:14:04,570
in one of four different ways.

1113
01:14:04,570 --> 01:14:06,810
Second order system introduced
new behaviors.

1114
01:14:06,810 --> 01:14:08,885
Now we can oscillate, which
we couldn't do before.

1115
01:14:11,540 --> 01:14:14,170
Having got to oscillation,
oscillation came about because

1116
01:14:14,170 --> 01:14:16,060
of complex numbers.

1117
01:14:16,060 --> 01:14:18,130
If you go to higher order
systems nothing

1118
01:14:18,130 --> 01:14:19,380
new happens in algebra.

1119
01:14:22,360 --> 01:14:25,490
There's no such thing as
meta-complex numbers, right?

1120
01:14:25,490 --> 01:14:28,200
Complex is as bad as it gets.

1121
01:14:28,200 --> 01:14:30,720
So you can have complex
numbers.

1122
01:14:30,720 --> 01:14:34,870
The higher order behaviors can
still have complex numbers but

1123
01:14:34,870 --> 01:14:38,360
you have to think when we ask
you, what's the property of a

1124
01:14:38,360 --> 01:14:39,700
higher order system.

1125
01:14:39,700 --> 01:14:42,620
You can think about it in terms
of the individual parts

1126
01:14:42,620 --> 01:14:45,320
but it requires some thinking.

1127
01:14:45,320 --> 01:14:45,740
OK.

1128
01:14:45,740 --> 01:14:46,710
Good luck tonight.

1129
01:14:46,710 --> 01:14:47,960
See you then.