Would you like to inspect the original subtitles? These are the user uploaded subtitles that are being translated:
1
00:00:09,519 --> 00:00:13,679
hi welcome to signalpad i have a really
2
00:00:11,919 --> 00:00:16,000
special episode for you guys today
3
00:00:13,679 --> 00:00:18,320
i have the privilege of introducing you
4
00:00:16,000 --> 00:00:20,640
to the keysight uxr series
5
00:00:18,320 --> 00:00:22,160
real-time oscilloscope this oscilloscope
6
00:00:20,640 --> 00:00:24,880
isn't just a little bit better than the
7
00:00:22,160 --> 00:00:28,560
study art it's a giant leap forward
8
00:00:24,879 --> 00:00:31,518
it can do 256 giga sample per second
9
00:00:28,559 --> 00:00:33,759
with 110 gigahertz of analog bandwidth
10
00:00:31,518 --> 00:00:35,519
on four channels simultaneously
11
00:00:33,759 --> 00:00:37,439
and it does that with a 10 bit
12
00:00:35,520 --> 00:00:38,960
resolution a to d converter
13
00:00:37,439 --> 00:00:40,919
and it doesn't matter how many charge
14
00:00:38,960 --> 00:00:43,679
channels you have active because it does
15
00:00:40,920 --> 00:00:44,559
256 gig sampling all the channels at the
16
00:00:43,679 --> 00:00:47,039
same time
17
00:00:44,558 --> 00:00:48,799
this means that it captures processes
18
00:00:47,039 --> 00:00:51,679
displays and stores over
19
00:00:48,799 --> 00:00:53,280
10 terabit per second of information in
20
00:00:51,679 --> 00:00:55,679
just this one box it's
21
00:00:53,280 --> 00:00:57,039
just unbelievable in order to get to
22
00:00:55,679 --> 00:00:59,119
this kind of specification
23
00:00:57,039 --> 00:01:00,879
they've had to innovate at every layer
24
00:00:59,119 --> 00:01:02,558
of the design and everything had to be
25
00:01:00,878 --> 00:01:04,478
custom made from scratch
26
00:01:02,558 --> 00:01:06,560
from the front-end connector to the
27
00:01:04,478 --> 00:01:08,879
pre-amplifier to the samplers
28
00:01:06,560 --> 00:01:10,079
to the a2d converters and even to the ic
29
00:01:08,879 --> 00:01:11,839
that interfaces
30
00:01:10,079 --> 00:01:13,359
the memory to the hdd converter had to
31
00:01:11,840 --> 00:01:15,600
be completely custom made
32
00:01:13,359 --> 00:01:17,840
in order to accommodate this massive
33
00:01:15,599 --> 00:01:19,519
amount of data that's being captured
34
00:01:17,840 --> 00:01:21,600
they also had to create their own
35
00:01:19,519 --> 00:01:23,359
calibrator module with nist calibration
36
00:01:21,599 --> 00:01:24,719
certificate which they can ship to you
37
00:01:23,359 --> 00:01:26,719
and you can calibrate your scope
38
00:01:24,719 --> 00:01:29,200
directly in the lab there's really
39
00:01:26,719 --> 00:01:30,560
nothing else like this and the fact that
40
00:01:29,200 --> 00:01:32,320
they have gone to this point is really
41
00:01:30,560 --> 00:01:32,799
amazing if i told you a couple of years
42
00:01:32,319 --> 00:01:34,000
ago
43
00:01:32,799 --> 00:01:36,079
these numbers you would have never
44
00:01:34,000 --> 00:01:36,719
believed it now this 10-bit atd
45
00:01:36,078 --> 00:01:38,000
converter
46
00:01:36,719 --> 00:01:39,359
which is sitting at the heart of this
47
00:01:38,000 --> 00:01:40,400
scope is the same one that's in the
48
00:01:39,359 --> 00:01:42,399
asterisk
49
00:01:40,400 --> 00:01:43,759
which i did a full teardown and review
50
00:01:42,399 --> 00:01:44,399
and that that's been refined and
51
00:01:43,759 --> 00:01:46,399
upgraded
52
00:01:44,399 --> 00:01:47,680
to be put in here now not only are we
53
00:01:46,399 --> 00:01:49,840
going to take a look at it
54
00:01:47,680 --> 00:01:51,759
and see how it behaves i'm going to take
55
00:01:49,840 --> 00:01:53,200
it apart well not quite take it apart
56
00:01:51,759 --> 00:01:54,719
but they've really sent me
57
00:01:53,200 --> 00:01:56,799
the acquisition board and they've sent
58
00:01:54,719 --> 00:01:58,798
me the front-end sampling modules
59
00:01:56,799 --> 00:02:00,079
i'm going to take off the part look at
60
00:01:58,799 --> 00:02:01,920
it under the microscope
61
00:02:00,078 --> 00:02:03,679
and really give you a detailed view of
62
00:02:01,920 --> 00:02:04,640
how it works what kind of engineers gone
63
00:02:03,680 --> 00:02:05,920
into it
64
00:02:04,640 --> 00:02:07,840
i'm going to take a look at how they've
65
00:02:05,920 --> 00:02:10,000
accomplished this massive amount of
66
00:02:07,840 --> 00:02:12,159
bandwidth it's actually 113 gigahertz
67
00:02:10,000 --> 00:02:13,280
which is extraordinary there's so much
68
00:02:12,159 --> 00:02:15,039
you can do with this there's
69
00:02:13,280 --> 00:02:16,719
it opens up test and measurement
70
00:02:15,039 --> 00:02:18,079
capability for optical coherent
71
00:02:16,719 --> 00:02:19,598
communication and even wireless
72
00:02:18,080 --> 00:02:21,520
communication if you want to capture
73
00:02:19,598 --> 00:02:23,039
many wireless channels at the same time
74
00:02:21,520 --> 00:02:24,640
like nothing else before
75
00:02:23,039 --> 00:02:26,318
so i'm really eager to take a look at it
76
00:02:24,639 --> 00:02:27,679
it's going to be a long video broken
77
00:02:26,318 --> 00:02:29,199
into a few sections you can look at the
78
00:02:27,680 --> 00:02:29,599
description to jump to the section you
79
00:02:29,199 --> 00:02:31,919
want
80
00:02:29,598 --> 00:02:33,199
this is still a prototype so for really
81
00:02:31,919 --> 00:02:34,559
detailed testing we're going to have to
82
00:02:33,199 --> 00:02:36,079
wait a little bit longer but i still do
83
00:02:34,560 --> 00:02:37,120
a couple of experiments with it which i
84
00:02:36,080 --> 00:02:38,719
think shows you how
85
00:02:37,120 --> 00:02:40,560
amazing it is and what it can be done
86
00:02:38,719 --> 00:02:41,840
with it so without any waiting let's go
87
00:02:40,560 --> 00:02:43,519
check it out
88
00:02:41,840 --> 00:02:44,878
so before we look at the architecture of
89
00:02:43,519 --> 00:02:46,879
the new oscilloscope i think it's
90
00:02:44,878 --> 00:02:48,479
worthwhile to appreciate how people have
91
00:02:46,878 --> 00:02:50,878
been doing this kind of
92
00:02:48,479 --> 00:02:52,479
high frequency sampling in the past or
93
00:02:50,878 --> 00:02:55,518
at least up to this point
94
00:02:52,479 --> 00:02:55,840
now tektronix likroy and keysight all
95
00:02:55,519 --> 00:02:58,480
have
96
00:02:55,840 --> 00:03:00,719
employed various techniques to overcome
97
00:02:58,479 --> 00:03:01,359
the limitations of time interleave a td
98
00:03:00,719 --> 00:03:03,680
converters
99
00:03:01,360 --> 00:03:04,879
for example lacroix does the digital
100
00:03:03,680 --> 00:03:06,400
bandwidth interleaving which is
101
00:03:04,878 --> 00:03:08,560
something that they have invented
102
00:03:06,400 --> 00:03:10,000
when i did the full teardown analysis of
103
00:03:08,560 --> 00:03:12,158
their scope we talked
104
00:03:10,000 --> 00:03:14,080
extensively about how that architecture
105
00:03:12,158 --> 00:03:15,598
works by splitting the frequency
106
00:03:14,080 --> 00:03:17,840
into two bands and then looking at the
107
00:03:15,598 --> 00:03:19,598
esp and combining them now keysight has
108
00:03:17,840 --> 00:03:21,439
their own real edge technology
109
00:03:19,598 --> 00:03:22,719
which works in a somewhat similar way it
110
00:03:21,439 --> 00:03:24,239
has some differences
111
00:03:22,719 --> 00:03:26,080
between their implementation and the
112
00:03:24,239 --> 00:03:27,680
electro implementation and of course the
113
00:03:26,080 --> 00:03:29,200
tektronix asynchronous
114
00:03:27,680 --> 00:03:30,879
timing to leave architecture which came
115
00:03:29,199 --> 00:03:33,119
a little bit later also
116
00:03:30,878 --> 00:03:35,199
uses some tricks on how to with dsp to
117
00:03:33,120 --> 00:03:36,959
correct how they sample the front end by
118
00:03:35,199 --> 00:03:39,199
using completely asynchronous
119
00:03:36,959 --> 00:03:40,878
sampling now all of these techniques
120
00:03:39,199 --> 00:03:41,598
they work to some extent but they all
121
00:03:40,878 --> 00:03:43,759
suffer from
122
00:03:41,598 --> 00:03:45,679
some basic limitations these these
123
00:03:43,759 --> 00:03:47,120
techniques of combining signals
124
00:03:45,680 --> 00:03:49,280
afterwards in dsp
125
00:03:47,120 --> 00:03:51,039
always has the limitations of noise
126
00:03:49,280 --> 00:03:52,959
being overlapped
127
00:03:51,039 --> 00:03:54,719
the tones being generated at the
128
00:03:52,959 --> 00:03:57,199
boundaries where frequencies overlapping
129
00:03:54,719 --> 00:03:58,239
it's a quite difficult a problem and dsp
130
00:03:57,199 --> 00:03:59,759
intensive problem
131
00:03:58,239 --> 00:04:01,599
to solve this and they work and they
132
00:03:59,759 --> 00:04:02,560
have been demonstrated up to about 70
133
00:04:01,598 --> 00:04:04,719
gigahertz or so
134
00:04:02,560 --> 00:04:06,080
from all of these companies but what
135
00:04:04,719 --> 00:04:08,318
keysight wanted to do here
136
00:04:06,080 --> 00:04:09,599
is to go back to really do time
137
00:04:08,318 --> 00:04:12,878
interleaving basically
138
00:04:09,598 --> 00:04:14,639
truly do 264 giga sample per second
139
00:04:12,878 --> 00:04:16,399
really and and get all the bandwidth
140
00:04:14,639 --> 00:04:18,399
from the front-end process by the
141
00:04:16,399 --> 00:04:20,000
single asic all the way up to 110
142
00:04:18,399 --> 00:04:20,478
gigahertz and that's exactly what they
143
00:04:20,000 --> 00:04:22,879
have
144
00:04:20,478 --> 00:04:25,120
done here and this is a very basic
145
00:04:22,879 --> 00:04:27,439
representation of their instrument it's
146
00:04:25,120 --> 00:04:28,399
very simple it's just pre-amplifier
147
00:04:27,439 --> 00:04:30,240
samplers
148
00:04:28,399 --> 00:04:32,879
and other samplers and then buffers and
149
00:04:30,240 --> 00:04:34,720
adcs exactly how a time interleaved
150
00:04:32,879 --> 00:04:36,000
architecture would be like and this time
151
00:04:34,720 --> 00:04:38,400
into leave architecture is
152
00:04:36,000 --> 00:04:40,000
time interleaved in several layers uh
153
00:04:38,399 --> 00:04:40,719
first is timing to leave between
154
00:04:40,000 --> 00:04:42,240
chipsets
155
00:04:40,720 --> 00:04:44,320
and then within the chip says the time
156
00:04:42,240 --> 00:04:45,040
interleaved between adc's and we'll talk
157
00:04:44,319 --> 00:04:46,959
about that
158
00:04:45,040 --> 00:04:48,879
in detail but really the architecture is
159
00:04:46,959 --> 00:04:50,399
pretty straightforward and this is what
160
00:04:48,879 --> 00:04:52,719
the architecture of pretty much
161
00:04:50,399 --> 00:04:54,319
every oscilloscope and that relies on
162
00:04:52,720 --> 00:04:55,040
time interleaving architecture without
163
00:04:54,319 --> 00:04:56,879
doing any
164
00:04:55,040 --> 00:04:58,319
fancy frequency interleaving or anything
165
00:04:56,879 --> 00:04:59,360
like that this is what it's going to
166
00:04:58,319 --> 00:05:01,120
look like
167
00:04:59,360 --> 00:05:02,479
so now that we know this let's go ahead
168
00:05:01,120 --> 00:05:04,000
and take a look at these blocks a couple
169
00:05:02,478 --> 00:05:04,719
of these pictures here for example the
170
00:05:04,000 --> 00:05:06,160
front end is
171
00:05:04,720 --> 00:05:07,520
pretty interesting the acquisition board
172
00:05:06,160 --> 00:05:08,880
is interesting and i have those and
173
00:05:07,519 --> 00:05:11,918
let's go take a look and see
174
00:05:08,879 --> 00:05:14,639
how they're made and here we have all
175
00:05:11,918 --> 00:05:15,680
the hardware required for 256 giga
176
00:05:14,639 --> 00:05:17,439
sample per second
177
00:05:15,680 --> 00:05:19,600
of data acquisition storage and
178
00:05:17,439 --> 00:05:20,079
processing there's a lot here to talk
179
00:05:19,600 --> 00:05:21,600
about
180
00:05:20,079 --> 00:05:23,519
and we're lucky enough to have all the
181
00:05:21,600 --> 00:05:24,080
hardware to take a look at it now here's
182
00:05:23,519 --> 00:05:26,879
the main
183
00:05:24,079 --> 00:05:28,478
110 gigahertz front-end sampler module
184
00:05:26,879 --> 00:05:30,240
here's the two-channel version for the
185
00:05:28,478 --> 00:05:31,918
lower models we'll talk about it as well
186
00:05:30,240 --> 00:05:33,280
and of course the entire acquisition
187
00:05:31,918 --> 00:05:34,799
board adc's
188
00:05:33,279 --> 00:05:37,519
the chipset that interfaces with the
189
00:05:34,800 --> 00:05:38,720
adcs and the hypercube memory module
190
00:05:37,519 --> 00:05:40,560
which is all the way at the back and
191
00:05:38,720 --> 00:05:42,800
we'll take a look at these in detail
192
00:05:40,560 --> 00:05:45,038
now what's amazing is that we're also
193
00:05:42,800 --> 00:05:46,639
given the ability to take a close look
194
00:05:45,038 --> 00:05:49,120
at the front-end module where all the
195
00:05:46,639 --> 00:05:50,400
magic with 110 gigahertz of bandwidth
196
00:05:49,120 --> 00:05:51,680
and sampling happens
197
00:05:50,399 --> 00:05:54,399
you can see here that the back of the
198
00:05:51,680 --> 00:05:56,079
lid we have rf absorbers strategically
199
00:05:54,399 --> 00:05:57,359
placed on top of the rfi season some of
200
00:05:56,079 --> 00:05:58,800
the critical traces
201
00:05:57,360 --> 00:06:01,600
this is to be expected because
202
00:05:58,800 --> 00:06:02,800
everything here is in a faraday cage
203
00:06:01,600 --> 00:06:04,319
you may not be able to see through the
204
00:06:02,800 --> 00:06:05,120
camera but we will be able to see when
205
00:06:04,319 --> 00:06:07,759
we look at it
206
00:06:05,120 --> 00:06:09,360
closely that these walls are all coming
207
00:06:07,759 --> 00:06:10,720
off the surface of the pcb and you're
208
00:06:09,360 --> 00:06:11,600
creating all these cavities and if
209
00:06:10,720 --> 00:06:13,120
you're not careful
210
00:06:11,600 --> 00:06:14,800
you're going to create second modes in
211
00:06:13,120 --> 00:06:15,360
there and it will be such a high quality
212
00:06:14,800 --> 00:06:17,120
factor
213
00:06:15,360 --> 00:06:18,879
you will get resonances and potentially
214
00:06:17,120 --> 00:06:20,240
even oscillations so they've
215
00:06:18,879 --> 00:06:21,918
obviously thought of that this is what
216
00:06:20,240 --> 00:06:23,199
they do all the time
217
00:06:21,918 --> 00:06:25,198
now we're going to take a close look in
218
00:06:23,199 --> 00:06:27,038
the front end i'm interested to see how
219
00:06:25,199 --> 00:06:28,240
the analog data is handled how is the
220
00:06:27,038 --> 00:06:29,759
clocking handle
221
00:06:28,240 --> 00:06:31,759
clocking for a to d converters is
222
00:06:29,759 --> 00:06:33,439
critical if you're not if you don't have
223
00:06:31,759 --> 00:06:35,439
a line clocks but for these samplers
224
00:06:33,439 --> 00:06:36,240
you're going to have tones and spurs all
225
00:06:35,439 --> 00:06:37,600
over the place
226
00:06:36,240 --> 00:06:39,439
in the spectrum after the data
227
00:06:37,600 --> 00:06:41,039
converters so there's a ton of
228
00:06:39,439 --> 00:06:43,120
engineering that's gone into this
229
00:06:41,038 --> 00:06:44,159
as well as multiple technologies used to
230
00:06:43,120 --> 00:06:45,680
build this front-end
231
00:06:44,160 --> 00:06:47,840
so i'm going to take a really close look
232
00:06:45,680 --> 00:06:49,598
at this and analyze it in a way reverse
233
00:06:47,839 --> 00:06:51,359
engineer it to some extent
234
00:06:49,598 --> 00:06:52,719
and then we'll jump into looking at a
235
00:06:51,360 --> 00:06:54,080
data converter and
236
00:06:52,720 --> 00:06:56,400
the chips that they've created for the
237
00:06:54,079 --> 00:06:57,519
interface the dsp there's a ton of stuff
238
00:06:56,399 --> 00:06:59,120
in here and of course
239
00:06:57,519 --> 00:07:00,560
the two-channel version so without
240
00:06:59,120 --> 00:07:02,240
keeping you waiting let's go and take a
241
00:07:00,560 --> 00:07:03,280
zoomed in view of this and i'll go over
242
00:07:02,240 --> 00:07:06,160
it step by step
243
00:07:03,279 --> 00:07:07,598
and see how it works so let's take a
244
00:07:06,160 --> 00:07:09,199
look at this front end and really
245
00:07:07,598 --> 00:07:11,038
analyze exactly what's going on and
246
00:07:09,199 --> 00:07:12,639
there's a lot of things happening here
247
00:07:11,038 --> 00:07:14,318
but we should kind of break it down step
248
00:07:12,639 --> 00:07:14,800
by step i think the easiest place to
249
00:07:14,319 --> 00:07:16,720
start
250
00:07:14,800 --> 00:07:18,879
is just the front-end so the input
251
00:07:16,720 --> 00:07:20,639
signal enters the device from here
252
00:07:18,879 --> 00:07:22,159
this is going to be a one millimeter
253
00:07:20,639 --> 00:07:24,160
connector in the hundred and
254
00:07:22,160 --> 00:07:26,000
ten gigahertz version of this instrument
255
00:07:24,160 --> 00:07:27,680
now before this connector there is an
256
00:07:26,000 --> 00:07:29,439
electromechanical attenuator
257
00:07:27,680 --> 00:07:31,280
and that electromechanical antenna sits
258
00:07:29,439 --> 00:07:34,000
between the front panel connector
259
00:07:31,279 --> 00:07:34,879
and the sampler module here that you see
260
00:07:34,000 --> 00:07:36,720
and the reason they put the
261
00:07:34,879 --> 00:07:38,079
electromechanical attenuator is because
262
00:07:36,720 --> 00:07:40,800
there is no way to get
263
00:07:38,079 --> 00:07:42,399
the massive dynamic range required from
264
00:07:40,800 --> 00:07:44,160
you know a volt per division all the way
265
00:07:42,399 --> 00:07:46,318
down to a millivolt per division going
266
00:07:44,160 --> 00:07:48,879
directly into the front-end amplifier
267
00:07:46,319 --> 00:07:50,800
so they divide the task of attenuation
268
00:07:48,879 --> 00:07:53,120
between the mechanical
269
00:07:50,800 --> 00:07:54,720
and the solid state attenuator or the
270
00:07:53,120 --> 00:07:56,560
amplifier that's the front and this is
271
00:07:54,720 --> 00:07:57,919
common this is done all the time
272
00:07:56,560 --> 00:07:59,680
however you have to appreciate the
273
00:07:57,918 --> 00:08:02,159
difficulty in doing this because
274
00:07:59,680 --> 00:08:03,519
the loss and the frequency flatness of
275
00:08:02,160 --> 00:08:05,280
the mechanical attenuator
276
00:08:03,519 --> 00:08:07,120
the interfaces between the connectors
277
00:08:05,279 --> 00:08:08,638
and the module all of those are going to
278
00:08:07,120 --> 00:08:11,199
make it difficult to get
279
00:08:08,639 --> 00:08:12,400
110 gigahertz bandwidth signal into this
280
00:08:11,199 --> 00:08:14,400
module so they have
281
00:08:12,399 --> 00:08:15,679
custom design all of these these modules
282
00:08:14,399 --> 00:08:17,519
are made to each other
283
00:08:15,680 --> 00:08:19,519
there's no cable in between everything's
284
00:08:17,519 --> 00:08:21,359
assembled mechanically together
285
00:08:19,519 --> 00:08:24,000
so having said that let's assume that
286
00:08:21,360 --> 00:08:25,840
the 110 gigahertz signal simply enters
287
00:08:24,000 --> 00:08:27,199
this front-end module after the
288
00:08:25,839 --> 00:08:29,198
mechanical attenuator
289
00:08:27,199 --> 00:08:30,800
we have a very small trace then we have
290
00:08:29,199 --> 00:08:31,759
two amplifiers in a row there's two
291
00:08:30,800 --> 00:08:34,080
chipsets here
292
00:08:31,759 --> 00:08:35,759
these are indium phosphite front-end
293
00:08:34,080 --> 00:08:38,959
amplifiers and vgas
294
00:08:35,759 --> 00:08:40,080
so your solid state valuable gain
295
00:08:38,958 --> 00:08:42,239
amplifier control
296
00:08:40,080 --> 00:08:44,720
and the front-end amplifier that handles
297
00:08:42,240 --> 00:08:46,799
110 gigahertz of signal is all done here
298
00:08:44,720 --> 00:08:48,320
the system linearity the system noise
299
00:08:46,799 --> 00:08:50,399
and the performance is determined
300
00:08:48,320 --> 00:08:52,320
by these front-end amplifiers you can
301
00:08:50,399 --> 00:08:54,000
imagine how difficult it is to design
302
00:08:52,320 --> 00:08:55,920
this to have such a noise to actually
303
00:08:54,000 --> 00:08:58,559
get anything meaningful from the a2d
304
00:08:55,919 --> 00:09:00,240
converters that follow after many stages
305
00:08:58,559 --> 00:09:02,000
the amount of components that are
306
00:09:00,240 --> 00:09:02,799
actually cascaded before you hit the
307
00:09:02,000 --> 00:09:04,639
a2ds is
308
00:09:02,799 --> 00:09:06,159
staggering and to be able to meet the
309
00:09:04,639 --> 00:09:08,720
linearity and the noise for that
310
00:09:06,159 --> 00:09:10,319
is really quite amazing so having said
311
00:09:08,720 --> 00:09:12,160
that this indium phosphide front-end
312
00:09:10,320 --> 00:09:14,000
amplifiers are designed by keysight
313
00:09:12,159 --> 00:09:15,439
in their own foundry they have their own
314
00:09:14,000 --> 00:09:17,839
foundry that make indium phosphite
315
00:09:15,440 --> 00:09:20,160
devices up to 600 or 700 gigahertz
316
00:09:17,839 --> 00:09:20,880
ftf max i can't quite remember the exact
317
00:09:20,159 --> 00:09:22,319
numbers but
318
00:09:20,879 --> 00:09:24,559
you know needless to say this is a very
319
00:09:22,320 --> 00:09:26,080
fast process so now signal coming out of
320
00:09:24,559 --> 00:09:28,559
here at this point
321
00:09:26,080 --> 00:09:30,639
is still a 110 gigahertz signal now
322
00:09:28,559 --> 00:09:32,239
there's also a path coming out here
323
00:09:30,639 --> 00:09:33,759
all the way going into this device here
324
00:09:32,240 --> 00:09:35,519
this is an edge trigger
325
00:09:33,759 --> 00:09:37,679
this is a 24 gigahertz edge trigger
326
00:09:35,519 --> 00:09:38,560
device it's actually not really used in
327
00:09:37,679 --> 00:09:40,719
this because
328
00:09:38,559 --> 00:09:42,000
24 gigahertz of edge trigger is kind of
329
00:09:40,720 --> 00:09:43,680
a legacy
330
00:09:42,000 --> 00:09:45,278
architecture for being able to detect
331
00:09:43,679 --> 00:09:47,199
edges up to 24 gigahertz but when you
332
00:09:45,278 --> 00:09:49,600
have 110 gigahertz of bandwidth
333
00:09:47,200 --> 00:09:50,720
you don't really want to use the 24 gig
334
00:09:49,600 --> 00:09:54,159
edge trigger anymore
335
00:09:50,720 --> 00:09:56,240
this entire oscilloscope uses a digital
336
00:09:54,159 --> 00:09:58,719
trigger so all of that is handled by the
337
00:09:56,240 --> 00:10:00,159
dsp custom digital asic that they have
338
00:09:58,720 --> 00:10:01,680
designed that follows the
339
00:10:00,159 --> 00:10:03,519
a to d converts we'll talk about that
340
00:10:01,679 --> 00:10:06,000
when we get there but the hardware for
341
00:10:03,519 --> 00:10:07,278
an analog 24 gigahertz edge trigger is
342
00:10:06,000 --> 00:10:09,278
still present here
343
00:10:07,278 --> 00:10:11,278
the signal after that goes into our
344
00:10:09,278 --> 00:10:14,559
first sampler now here's the first
345
00:10:11,278 --> 00:10:15,919
really complex uh device that is also an
346
00:10:14,559 --> 00:10:18,078
indium phosphide
347
00:10:15,919 --> 00:10:20,399
is presented now this device is going to
348
00:10:18,078 --> 00:10:22,719
take her 110 gigahertz signal
349
00:10:20,399 --> 00:10:23,759
is going to sample it with a 64
350
00:10:22,720 --> 00:10:26,959
gigahertz clock
351
00:10:23,759 --> 00:10:28,559
in four phases in order to generate four
352
00:10:26,958 --> 00:10:31,439
signals coming out each of them now
353
00:10:28,559 --> 00:10:32,719
having bandwidth of up to 32 gigahertz
354
00:10:31,440 --> 00:10:34,959
this is how you get
355
00:10:32,720 --> 00:10:36,959
the reduction in required bandwidth as
356
00:10:34,958 --> 00:10:38,958
the stages follow on this is again
357
00:10:36,958 --> 00:10:40,879
a normal architecture for time
358
00:10:38,958 --> 00:10:42,639
interleaved data converters
359
00:10:40,879 --> 00:10:44,639
so therefore we will have four paths one
360
00:10:42,639 --> 00:10:46,480
path over here one path over here
361
00:10:44,639 --> 00:10:48,720
one path over here and one path over
362
00:10:46,480 --> 00:10:50,959
here so we would expect these four paths
363
00:10:48,720 --> 00:10:52,480
to go into the next stages of samplers
364
00:10:50,958 --> 00:10:53,679
in order to be able to further break
365
00:10:52,480 --> 00:10:54,320
down the bandwidth so that you can
366
00:10:53,679 --> 00:10:57,199
actually
367
00:10:54,320 --> 00:10:58,160
use the a2d converters inputs so how do
368
00:10:57,200 --> 00:11:01,278
we do this
369
00:10:58,159 --> 00:11:03,679
64 gigahertz for phase sampling
370
00:11:01,278 --> 00:11:05,278
that itself is really difficult so for
371
00:11:03,679 --> 00:11:06,799
to understand that we have to go from
372
00:11:05,278 --> 00:11:07,838
the side where the clock is now
373
00:11:06,799 --> 00:11:09,599
generated
374
00:11:07,839 --> 00:11:11,360
now there's an ic over here which you
375
00:11:09,600 --> 00:11:13,040
can barely see the edge of this is a
376
00:11:11,360 --> 00:11:16,320
hittite part this is a
377
00:11:13,039 --> 00:11:17,519
12 to 16 gigahertz to up to 17 gigahertz
378
00:11:16,320 --> 00:11:20,800
amplifier it has
379
00:11:17,519 --> 00:11:22,240
a p1 db of 26 dbm so you can put out
380
00:11:20,799 --> 00:11:23,679
tons of power
381
00:11:22,240 --> 00:11:26,320
which is required because they're going
382
00:11:23,679 --> 00:11:28,879
to divide that clock signal further down
383
00:11:26,320 --> 00:11:30,800
now it appears that the 16 gigahertz
384
00:11:28,879 --> 00:11:32,559
system clock comes from here but if i
385
00:11:30,799 --> 00:11:33,439
understand correctly that itself is
386
00:11:32,559 --> 00:11:35,679
generated from an
387
00:11:33,440 --> 00:11:38,000
eight gigahertz fundamental clock which
388
00:11:35,679 --> 00:11:39,838
is spread throughout the entire chassis
389
00:11:38,000 --> 00:11:41,120
connecting all the different samplers
390
00:11:39,839 --> 00:11:41,760
and connecting all the different atd
391
00:11:41,120 --> 00:11:44,399
converters
392
00:11:41,759 --> 00:11:46,399
to a common synthesized clock which
393
00:11:44,399 --> 00:11:48,639
again is all designed by keysight
394
00:11:46,399 --> 00:11:50,440
generating an eight gigahertz clock
395
00:11:48,639 --> 00:11:52,958
clean enough to be able to feed
396
00:11:50,440 --> 00:11:54,720
256 giga sample per second equivalent
397
00:11:52,958 --> 00:11:55,759
sampling is really challenging
398
00:11:54,720 --> 00:11:57,759
now i don't have that board
399
00:11:55,759 --> 00:11:59,120
unfortunately but needless to say
400
00:11:57,759 --> 00:12:01,360
it's you know in the in the 10
401
00:11:59,120 --> 00:12:03,679
femtosecond range clock which is
402
00:12:01,360 --> 00:12:05,200
really amazing so now that it doubled
403
00:12:03,679 --> 00:12:06,958
that apparently at some point and then
404
00:12:05,200 --> 00:12:07,680
we get the 16 gigahertz clock and there
405
00:12:06,958 --> 00:12:10,399
is a first
406
00:12:07,679 --> 00:12:12,239
power splitter here that power splitter
407
00:12:10,399 --> 00:12:14,559
splits the signal into two parts
408
00:12:12,240 --> 00:12:16,159
one side of it goes in here now this
409
00:12:14,559 --> 00:12:18,638
disappears into the board
410
00:12:16,159 --> 00:12:20,159
but you can kind of roughly uh see where
411
00:12:18,639 --> 00:12:21,440
it would go it would have to come all
412
00:12:20,159 --> 00:12:22,879
the way over here and this goes
413
00:12:21,440 --> 00:12:24,639
underneath this module
414
00:12:22,879 --> 00:12:27,200
and then comes out of here and pops into
415
00:12:24,639 --> 00:12:29,039
this so now you have a 16 gigahertz
416
00:12:27,200 --> 00:12:31,360
signal going into this chain
417
00:12:29,039 --> 00:12:32,719
over here first they do some some
418
00:12:31,360 --> 00:12:33,919
filtering to make sure you know there's
419
00:12:32,720 --> 00:12:35,680
only a single tone
420
00:12:33,919 --> 00:12:37,599
and then they double it once here so
421
00:12:35,679 --> 00:12:39,439
that times two
422
00:12:37,600 --> 00:12:40,639
over here that's doubled so now it's at
423
00:12:39,440 --> 00:12:43,200
32 gigahertz
424
00:12:40,639 --> 00:12:45,278
once again they filter it very carefully
425
00:12:43,200 --> 00:12:48,160
and then they double it again
426
00:12:45,278 --> 00:12:48,879
one more time now we are sitting at 64
427
00:12:48,159 --> 00:12:51,039
gigahertz
428
00:12:48,879 --> 00:12:52,159
now single is still single-ended which
429
00:12:51,039 --> 00:12:53,599
means that it needs to become
430
00:12:52,159 --> 00:12:54,319
differentials there's after some
431
00:12:53,600 --> 00:12:56,320
filtering
432
00:12:54,320 --> 00:12:58,560
there's a hybrid here you can see very
433
00:12:56,320 --> 00:13:00,639
closely the hybrid splits the signal
434
00:12:58,559 --> 00:13:01,759
into two pieces now that you have two
435
00:13:00,639 --> 00:13:03,839
signals
436
00:13:01,759 --> 00:13:05,519
that are in 180 degree out of phase and
437
00:13:03,839 --> 00:13:07,200
it appears that they put them through
438
00:13:05,519 --> 00:13:08,799
two separate amplifiers here
439
00:13:07,200 --> 00:13:10,240
so they have two single nano distributed
440
00:13:08,799 --> 00:13:12,559
amplifiers again this is all in
441
00:13:10,240 --> 00:13:13,759
indium phosphide now you have a 64
442
00:13:12,559 --> 00:13:15,838
gigahertz signal
443
00:13:13,759 --> 00:13:17,838
on one side and a 64 gigahertz signal on
444
00:13:15,839 --> 00:13:19,120
the other side so you get yourself a
445
00:13:17,839 --> 00:13:21,920
differential
446
00:13:19,120 --> 00:13:24,320
a 64 gigahertz signal going into the
447
00:13:21,919 --> 00:13:26,479
front end indium phosphite sampler
448
00:13:24,320 --> 00:13:28,879
but that's of course not in four phases
449
00:13:26,480 --> 00:13:29,759
you need to generate four phases from 64
450
00:13:28,879 --> 00:13:31,120
gigahertz
451
00:13:29,759 --> 00:13:33,120
normally if you're in a silicon
452
00:13:31,120 --> 00:13:33,600
germanium process or in a cmos process
453
00:13:33,120 --> 00:13:35,200
that's
454
00:13:33,600 --> 00:13:37,120
not too difficult but in indium
455
00:13:35,200 --> 00:13:39,440
phosphide you gain
456
00:13:37,120 --> 00:13:40,320
linearity and noise and dynamic range
457
00:13:39,440 --> 00:13:42,720
but you lose
458
00:13:40,320 --> 00:13:44,480
complexity of integration so making
459
00:13:42,720 --> 00:13:46,240
phase shifters that are adjustable in
460
00:13:44,480 --> 00:13:48,079
indium phosphide is quite difficult
461
00:13:46,240 --> 00:13:49,759
now i happen to know that the way they
462
00:13:48,078 --> 00:13:50,958
do this is that they double the 64
463
00:13:49,759 --> 00:13:53,600
gigahertz yet again
464
00:13:50,958 --> 00:13:54,638
to 128 gigahertz and then they process
465
00:13:53,600 --> 00:13:56,560
that back down
466
00:13:54,639 --> 00:13:58,159
in order to get the four phases which is
467
00:13:56,559 --> 00:14:00,319
a common technique you can use
468
00:13:58,159 --> 00:14:02,480
if you have tons of bandwidth and tons
469
00:14:00,320 --> 00:14:03,600
of frequency content you can do this in
470
00:14:02,480 --> 00:14:05,120
a very fast process
471
00:14:03,600 --> 00:14:06,879
going the opposite direction just so
472
00:14:05,120 --> 00:14:08,959
that you can create your phases
473
00:14:06,879 --> 00:14:10,159
so now you have four phases of 64
474
00:14:08,958 --> 00:14:12,479
gigahertz in here
475
00:14:10,159 --> 00:14:15,039
which you can use to sample the
476
00:14:12,480 --> 00:14:18,000
front-end 110 gigahertz signal into four
477
00:14:15,039 --> 00:14:19,838
sections so all of that is again fully
478
00:14:18,000 --> 00:14:21,440
key side design all of these filters
479
00:14:19,839 --> 00:14:23,199
every single component you see here
480
00:14:21,440 --> 00:14:25,199
has been custom designed to make this
481
00:14:23,198 --> 00:14:26,719
happen you still need phase shifters
482
00:14:25,198 --> 00:14:27,599
inside the front end because you need to
483
00:14:26,720 --> 00:14:30,480
make sure that the
484
00:14:27,600 --> 00:14:31,920
alignment between the 64 gigahertz clock
485
00:14:30,480 --> 00:14:33,440
in four phases is good
486
00:14:31,919 --> 00:14:35,519
and they do take care of that again
487
00:14:33,440 --> 00:14:37,680
inside the three five process
488
00:14:35,519 --> 00:14:39,039
amazing stuff so now you have all these
489
00:14:37,679 --> 00:14:41,919
four signals now you can
490
00:14:39,039 --> 00:14:43,120
get away from indium phosphide and and
491
00:14:41,919 --> 00:14:46,240
make the complexity
492
00:14:43,120 --> 00:14:48,078
more and create signals that can be
493
00:14:46,240 --> 00:14:49,839
processed in a silicon germanium process
494
00:14:48,078 --> 00:14:51,439
you don't need as much bandwidth anymore
495
00:14:49,839 --> 00:14:53,680
the bandwidth of these signals coming
496
00:14:51,440 --> 00:14:56,800
out of here is no longer uh
497
00:14:53,679 --> 00:14:58,799
110 gigahertz is only up to 32 gigahertz
498
00:14:56,799 --> 00:15:00,719
now all these traces that you see these
499
00:14:58,799 --> 00:15:02,799
are all thin film deposits these are all
500
00:15:00,720 --> 00:15:04,320
again done by keysight this is all
501
00:15:02,799 --> 00:15:06,399
completely custom made
502
00:15:04,320 --> 00:15:07,680
and you can see individual cavities here
503
00:15:06,399 --> 00:15:09,120
this is this has a depth of
504
00:15:07,679 --> 00:15:10,879
a couple of millimeters so when i put
505
00:15:09,120 --> 00:15:13,120
the lid on top of this
506
00:15:10,879 --> 00:15:14,000
all of these are in fahrenheit cages and
507
00:15:13,120 --> 00:15:16,078
you can see that they
508
00:15:14,000 --> 00:15:18,000
isolate the traces from each other and
509
00:15:16,078 --> 00:15:21,519
these are high bandwidth
510
00:15:18,000 --> 00:15:24,240
very low skew good group delay paths
511
00:15:21,519 --> 00:15:26,000
created to make the signal as nicely
512
00:15:24,240 --> 00:15:29,198
preserved as much as possible before
513
00:15:26,000 --> 00:15:30,720
it hits the next samplers there
514
00:15:29,198 --> 00:15:32,240
now going forward let me change the
515
00:15:30,720 --> 00:15:34,000
color here so that we can
516
00:15:32,240 --> 00:15:36,480
do the second stage with a different
517
00:15:34,000 --> 00:15:38,240
color all right let's take a look at the
518
00:15:36,480 --> 00:15:40,000
what happens after that let's go back to
519
00:15:38,240 --> 00:15:41,839
the clock so now the other path of the
520
00:15:40,000 --> 00:15:42,720
clock the other 16 gigahertz clock goes
521
00:15:41,839 --> 00:15:44,959
into this
522
00:15:42,720 --> 00:15:47,278
hybrid here which then further splits
523
00:15:44,958 --> 00:15:50,159
the signal into four so now you got one
524
00:15:47,278 --> 00:15:51,278
two three four pads each of them 16
525
00:15:50,159 --> 00:15:54,480
gigahertz again
526
00:15:51,278 --> 00:15:56,078
you can use this to feed the samplers
527
00:15:54,480 --> 00:15:57,920
of the stages that follow so now
528
00:15:56,078 --> 00:15:59,919
following over here we go into this
529
00:15:57,919 --> 00:16:01,439
here's the another sampler here is
530
00:15:59,919 --> 00:16:03,599
another sampler
531
00:16:01,440 --> 00:16:04,560
going over here is the third one and
532
00:16:03,600 --> 00:16:06,879
going over here
533
00:16:04,559 --> 00:16:07,599
here's our fourth one now these clocks
534
00:16:06,879 --> 00:16:09,919
that are
535
00:16:07,600 --> 00:16:11,759
disappearing into the pcb you can see
536
00:16:09,919 --> 00:16:13,120
the trace is over there one of them pops
537
00:16:11,759 --> 00:16:16,000
over here
538
00:16:13,120 --> 00:16:17,120
the other one pops over here this one
539
00:16:16,000 --> 00:16:19,360
where does it go
540
00:16:17,120 --> 00:16:21,440
right there and the last one goes all
541
00:16:19,360 --> 00:16:23,039
the way here and pops out of here
542
00:16:21,440 --> 00:16:25,279
and if you look closely you can see that
543
00:16:23,039 --> 00:16:27,120
they take the 16 gigahertz signal and
544
00:16:25,278 --> 00:16:29,600
not only do they filter it but they also
545
00:16:27,120 --> 00:16:30,959
passively create a 180 degree
546
00:16:29,600 --> 00:16:33,199
hybrid here so there's a balance in
547
00:16:30,958 --> 00:16:34,159
there plus the filter feeding the 16
548
00:16:33,198 --> 00:16:35,758
gigahertz clock
549
00:16:34,159 --> 00:16:37,600
into the sampler here now this is a
550
00:16:35,759 --> 00:16:39,759
silicon germanium sampler
551
00:16:37,600 --> 00:16:41,680
you can generate four phases from the 16
552
00:16:39,759 --> 00:16:42,399
gigahertz in here fairly straightforward
553
00:16:41,679 --> 00:16:45,278
so there's
554
00:16:42,399 --> 00:16:46,879
four phases of 16 gigahertz sampling the
555
00:16:45,278 --> 00:16:49,120
front-end
556
00:16:46,879 --> 00:16:51,759
signal coming over here at a bandwidth
557
00:16:49,120 --> 00:16:55,198
of up to 32 gigahertz which then creates
558
00:16:51,759 --> 00:16:56,879
four lanes one two three four
559
00:16:55,198 --> 00:16:59,278
out of each of the samplers so you can
560
00:16:56,879 --> 00:17:01,838
see that this splits into four yet again
561
00:16:59,278 --> 00:17:03,439
and four over here one more time and
562
00:17:01,839 --> 00:17:05,759
then four over here one two
563
00:17:03,440 --> 00:17:07,120
three four and then four over here as
564
00:17:05,759 --> 00:17:10,240
the last one so now
565
00:17:07,119 --> 00:17:13,678
your 110 gigahertz front-end signal
566
00:17:10,240 --> 00:17:15,919
has now been broken into 16 differential
567
00:17:13,679 --> 00:17:17,280
pairs and these 16 differential lines
568
00:17:15,919 --> 00:17:18,959
can now be fed
569
00:17:17,279 --> 00:17:20,558
into the a to b converters because now
570
00:17:18,959 --> 00:17:21,120
they're low enough in bandwidth each of
571
00:17:20,558 --> 00:17:23,119
them has
572
00:17:21,119 --> 00:17:26,239
a bandwidth of just under eight
573
00:17:23,119 --> 00:17:27,838
gigahertz now 16 times eight is 128
574
00:17:26,240 --> 00:17:28,480
gigahertz the bandwidth of the scope is
575
00:17:27,838 --> 00:17:31,200
limited to
576
00:17:28,480 --> 00:17:32,000
about 113. so each of these process a
577
00:17:31,200 --> 00:17:34,080
little bit less than
578
00:17:32,000 --> 00:17:35,038
eight gigahertz of signal but now that
579
00:17:34,079 --> 00:17:38,079
you have this
580
00:17:35,038 --> 00:17:39,759
you can feed that directly into an a2d
581
00:17:38,079 --> 00:17:43,279
converter all you need is that your a
582
00:17:39,759 --> 00:17:45,679
d converter must now accept four lanes
583
00:17:43,279 --> 00:17:47,359
uh each of them up to eight gigahertz of
584
00:17:45,679 --> 00:17:48,960
bandwidth now if you do that which means
585
00:17:47,359 --> 00:17:49,279
that your a2d converters would have to
586
00:17:48,960 --> 00:17:52,640
be
587
00:17:49,279 --> 00:17:54,000
each 64 giga sample per second which is
588
00:17:52,640 --> 00:17:57,440
exactly what it is now
589
00:17:54,000 --> 00:17:58,160
the a2d converters one here second a to
590
00:17:57,440 --> 00:18:00,640
b here
591
00:17:58,160 --> 00:18:01,600
third a to d here and fourth a to d here
592
00:18:00,640 --> 00:18:04,720
all have to work
593
00:18:01,599 --> 00:18:08,240
in parallel so four times 64
594
00:18:04,720 --> 00:18:10,720
you get 256 giga sample per second
595
00:18:08,240 --> 00:18:12,160
but don't don't be fooled by the fact
596
00:18:10,720 --> 00:18:13,600
that you keep splitting these signals
597
00:18:12,160 --> 00:18:15,038
and therefore you make the job of the a
598
00:18:13,599 --> 00:18:18,000
to the converter easier
599
00:18:15,038 --> 00:18:19,599
the isolation and the alignment and the
600
00:18:18,000 --> 00:18:21,279
matching between these paths
601
00:18:19,599 --> 00:18:22,959
is critical because if you don't have
602
00:18:21,279 --> 00:18:24,798
matching between them then you're going
603
00:18:22,960 --> 00:18:26,400
to get spurs in the spectrum of the a2d
604
00:18:24,798 --> 00:18:28,319
converters you can't just split the
605
00:18:26,400 --> 00:18:29,919
signal without worrying about
606
00:18:28,319 --> 00:18:31,038
how much phase noise you have how much
607
00:18:29,919 --> 00:18:32,400
phase alignment you have how much
608
00:18:31,038 --> 00:18:34,240
amplitude balance you have
609
00:18:32,400 --> 00:18:37,200
that's what the magic of this front end
610
00:18:34,240 --> 00:18:38,960
is is to be able to take 110 gigahertz
611
00:18:37,200 --> 00:18:41,600
produce a signal that can actually be
612
00:18:38,960 --> 00:18:42,558
processed by 64 gigahertz per second a
613
00:18:41,599 --> 00:18:45,439
2d converter
614
00:18:42,558 --> 00:18:47,200
now just the adc itself at 10 bits with
615
00:18:45,440 --> 00:18:48,960
64 giga sum per second
616
00:18:47,200 --> 00:18:50,798
is itself an engineering marvel because
617
00:18:48,960 --> 00:18:52,240
it's not easy to make that
618
00:18:50,798 --> 00:18:54,558
but you can see that they've broken the
619
00:18:52,240 --> 00:18:56,640
front end a little bit differently
620
00:18:54,558 --> 00:18:58,399
than it is in the s series oscilloscope
621
00:18:56,640 --> 00:18:59,120
which i did a full teardown and review
622
00:18:58,400 --> 00:19:00,880
of before
623
00:18:59,119 --> 00:19:02,719
this is the same adc that has been
624
00:19:00,880 --> 00:19:04,320
upgraded and refined
625
00:19:02,720 --> 00:19:05,839
for this application so you can see how
626
00:19:04,319 --> 00:19:08,079
they use that and
627
00:19:05,839 --> 00:19:10,319
they just tweak the front end with new
628
00:19:08,079 --> 00:19:11,240
devices and new configurations in order
629
00:19:10,319 --> 00:19:14,399
to enable
630
00:19:11,240 --> 00:19:15,679
256 gigs per second sampling now if you
631
00:19:14,400 --> 00:19:18,400
look closely again
632
00:19:15,679 --> 00:19:19,280
the adc inputs from each other are fully
633
00:19:18,400 --> 00:19:21,280
isolated
634
00:19:19,279 --> 00:19:22,798
and this reason is because it is a
635
00:19:21,279 --> 00:19:25,599
little bit easier to handle
636
00:19:22,798 --> 00:19:26,480
crosstalk within each adc then between
637
00:19:25,599 --> 00:19:28,798
the adcs
638
00:19:26,480 --> 00:19:31,360
so the isolation between the lanes of
639
00:19:28,798 --> 00:19:33,440
these ones coming out is very important
640
00:19:31,359 --> 00:19:35,359
i also want you to appreciate how
641
00:19:33,440 --> 00:19:36,000
difficult it must be to isolate the
642
00:19:35,359 --> 00:19:38,319
clock
643
00:19:36,000 --> 00:19:39,359
from leaking into the a2d converters
644
00:19:38,319 --> 00:19:41,439
because you have
645
00:19:39,359 --> 00:19:43,199
16 gigahertz clocks everywhere in
646
00:19:41,440 --> 00:19:44,880
multiple phases you have multiple
647
00:19:43,200 --> 00:19:47,038
multipliers and doublers which create
648
00:19:44,880 --> 00:19:48,799
harmonics and intermodulation products
649
00:19:47,038 --> 00:19:50,079
all of these things coexist in this
650
00:19:48,798 --> 00:19:51,759
front-end module
651
00:19:50,079 --> 00:19:53,359
all of which have to be filtered
652
00:19:51,759 --> 00:19:54,879
properly in order to
653
00:19:53,359 --> 00:19:57,119
make sure that they don't feed into the
654
00:19:54,880 --> 00:19:59,280
a2d converters so really is
655
00:19:57,119 --> 00:20:01,439
extraordinary a ton of engineering has
656
00:19:59,279 --> 00:20:02,158
gone into making this happen you can see
657
00:20:01,440 --> 00:20:04,159
custom
658
00:20:02,159 --> 00:20:05,520
packaging there's devices from different
659
00:20:04,159 --> 00:20:07,120
technologies is
660
00:20:05,519 --> 00:20:08,639
is amazing and what is also interesting
661
00:20:07,119 --> 00:20:10,479
is everything is wire bonded
662
00:20:08,640 --> 00:20:12,000
now they don't use flip chip technology
663
00:20:10,480 --> 00:20:13,599
anywhere here because
664
00:20:12,000 --> 00:20:15,519
you can calibrate some of those effects
665
00:20:13,599 --> 00:20:16,558
out and with clever design you can get
666
00:20:15,519 --> 00:20:19,359
away with using
667
00:20:16,558 --> 00:20:20,720
wire bonds even up to 110 gigahertz is
668
00:20:19,359 --> 00:20:22,719
which itself is quite
669
00:20:20,720 --> 00:20:24,720
interesting to see so there it is that's
670
00:20:22,720 --> 00:20:26,798
the front end so now you have an idea
671
00:20:24,720 --> 00:20:28,079
of what happens to the 110 gigahertz
672
00:20:26,798 --> 00:20:31,359
signal and how does it
673
00:20:28,079 --> 00:20:32,639
end up being fed into four adc's at 64
674
00:20:31,359 --> 00:20:34,479
giga sample per second
675
00:20:32,640 --> 00:20:36,080
so now we can go ahead and take a look
676
00:20:34,480 --> 00:20:37,839
at the acquisition module
677
00:20:36,079 --> 00:20:39,678
and keeping in mind exactly what is
678
00:20:37,839 --> 00:20:40,399
happening here and see how that signal
679
00:20:39,679 --> 00:20:43,798
is processed
680
00:20:40,400 --> 00:20:46,640
by the a2d converters and here is the
681
00:20:43,798 --> 00:20:49,200
256 giga sample per second
682
00:20:46,640 --> 00:20:51,038
acquisition board it is massive and it
683
00:20:49,200 --> 00:20:53,360
has so many components on it
684
00:20:51,038 --> 00:20:55,359
so many devices designed by keysight now
685
00:20:53,359 --> 00:20:57,519
keep in mind there are four of these
686
00:20:55,359 --> 00:20:58,719
inside a four channel oscilloscope
687
00:20:57,519 --> 00:20:59,240
that's because that's how you're going
688
00:20:58,720 --> 00:21:02,640
to get
689
00:20:59,240 --> 00:21:04,960
256 gigahertz per second 110 gigahertz
690
00:21:02,640 --> 00:21:06,240
per channel regardless of how the
691
00:21:04,960 --> 00:21:07,679
channels are configured so it doesn't
692
00:21:06,240 --> 00:21:09,599
matter if you have one channel on
693
00:21:07,679 --> 00:21:11,200
or if you have four channels on you have
694
00:21:09,599 --> 00:21:12,879
one of these dedicated per channel
695
00:21:11,200 --> 00:21:14,000
that's one of the reasons why it's 1.3
696
00:21:12,880 --> 00:21:16,080
million dollars
697
00:21:14,000 --> 00:21:17,200
because you have so much independent
698
00:21:16,079 --> 00:21:19,038
redundant hardware
699
00:21:17,200 --> 00:21:20,400
to get the performance per channel
700
00:21:19,038 --> 00:21:22,879
totally unaffected
701
00:21:20,400 --> 00:21:23,600
by how much data you're collecting so as
702
00:21:22,880 --> 00:21:26,640
we said
703
00:21:23,599 --> 00:21:28,399
this needs to have four 64 giga sample
704
00:21:26,640 --> 00:21:30,000
per second a 2d converters in order to
705
00:21:28,400 --> 00:21:32,400
be able to to get to 256
706
00:21:30,000 --> 00:21:33,759
gig sampling and with the 16 inputs
707
00:21:32,400 --> 00:21:36,400
coming from the front end
708
00:21:33,759 --> 00:21:36,879
sampler i just showed you so here's one
709
00:21:36,400 --> 00:21:39,200
two
710
00:21:36,880 --> 00:21:41,440
three four a to d converters and these
711
00:21:39,200 --> 00:21:43,919
are the each of these are 64 gig
712
00:21:41,440 --> 00:21:45,038
10 bits with four inputs you can see
713
00:21:43,919 --> 00:21:47,038
some of the traces
714
00:21:45,038 --> 00:21:49,038
with the microstrip or strip lines that
715
00:21:47,038 --> 00:21:51,038
are going into it there's a front-end
716
00:21:49,038 --> 00:21:52,720
high-speed connector and an alignment
717
00:21:51,038 --> 00:21:54,640
pin that allows you to interface with
718
00:21:52,720 --> 00:21:55,440
the front-end sampler so all the analog
719
00:21:54,640 --> 00:21:57,600
signals
720
00:21:55,440 --> 00:21:59,038
have to travel as little as possible and
721
00:21:57,599 --> 00:22:00,319
they've come up with this solution with
722
00:21:59,038 --> 00:22:01,679
this connector
723
00:22:00,319 --> 00:22:03,839
there is a front-end sample there
724
00:22:01,679 --> 00:22:05,440
modulator this thing is quite heavy
725
00:22:03,839 --> 00:22:07,918
to lift oh i have it the other way
726
00:22:05,440 --> 00:22:08,480
around like this and this simply plugs
727
00:22:07,919 --> 00:22:11,440
in
728
00:22:08,480 --> 00:22:12,319
to here like so there it is that's one
729
00:22:11,440 --> 00:22:13,840
channel
730
00:22:12,319 --> 00:22:15,439
and this then has to connect to the
731
00:22:13,839 --> 00:22:17,119
front panel of the instrument so the
732
00:22:15,440 --> 00:22:18,400
electromechanical antenna sits right
733
00:22:17,119 --> 00:22:19,199
over here and right after that is the
734
00:22:18,400 --> 00:22:20,798
front end
735
00:22:19,200 --> 00:22:22,720
of the instrument itself so then you
736
00:22:20,798 --> 00:22:24,000
stack four of these so you can imagine
737
00:22:22,720 --> 00:22:25,519
the thermal management and the
738
00:22:24,000 --> 00:22:27,839
mechanical management of this
739
00:22:25,519 --> 00:22:28,639
itself is quite a nightmare now even
740
00:22:27,839 --> 00:22:31,839
though you have
741
00:22:28,640 --> 00:22:33,360
four 64 gig a2d converters here how do
742
00:22:31,839 --> 00:22:35,759
you deal with the data coming out of
743
00:22:33,359 --> 00:22:37,359
these because each of these is at 64gb
744
00:22:35,759 --> 00:22:39,119
10 with 640
745
00:22:37,359 --> 00:22:40,798
gigabit per second of data needs to be
746
00:22:39,119 --> 00:22:42,959
handled not only that
747
00:22:40,798 --> 00:22:43,839
these are all working together so you
748
00:22:42,960 --> 00:22:46,640
need to have
749
00:22:43,839 --> 00:22:47,759
the alignment the calibration all the
750
00:22:46,640 --> 00:22:49,759
coefficients
751
00:22:47,759 --> 00:22:51,440
to handle with the channel channels
752
00:22:49,759 --> 00:22:54,000
mismatches and channel
753
00:22:51,440 --> 00:22:55,440
frequency group delay variations all of
754
00:22:54,000 --> 00:22:56,159
those parameters have to be somehow
755
00:22:55,440 --> 00:22:58,159
handled
756
00:22:56,159 --> 00:22:59,840
directly from the after the a to d
757
00:22:58,159 --> 00:23:00,960
converters and more importantly what
758
00:22:59,839 --> 00:23:02,959
about the trigger
759
00:23:00,960 --> 00:23:05,440
this thing has a full digital trigger
760
00:23:02,960 --> 00:23:06,798
per channel up to 110 gigahertz which
761
00:23:05,440 --> 00:23:07,759
means that you need to also be able to
762
00:23:06,798 --> 00:23:09,918
do that
763
00:23:07,759 --> 00:23:11,919
live on the data as it comes in of
764
00:23:09,919 --> 00:23:12,880
course you can't dump that data to a cpu
765
00:23:11,919 --> 00:23:15,840
there's no way
766
00:23:12,880 --> 00:23:17,840
which means you have to create a custom
767
00:23:15,839 --> 00:23:19,599
digital signal processor chipset to
768
00:23:17,839 --> 00:23:20,480
handle all of that and of course that's
769
00:23:19,599 --> 00:23:23,759
what they've done
770
00:23:20,480 --> 00:23:25,919
one two three four each of these is
771
00:23:23,759 --> 00:23:27,440
independently connected and working
772
00:23:25,919 --> 00:23:29,120
together at the same time to each of the
773
00:23:27,440 --> 00:23:30,080
a to the converter so everything you see
774
00:23:29,119 --> 00:23:31,678
under these
775
00:23:30,079 --> 00:23:33,359
eight heat sinks that's keysight
776
00:23:31,679 --> 00:23:35,440
technology designed and
777
00:23:33,359 --> 00:23:36,959
manufactured by them just so that you
778
00:23:35,440 --> 00:23:38,159
can package by them just so you can put
779
00:23:36,960 --> 00:23:39,919
it on this board
780
00:23:38,159 --> 00:23:41,200
now after that you need to put the data
781
00:23:39,919 --> 00:23:43,759
into some kind of memory
782
00:23:41,200 --> 00:23:45,360
the data is coming in at huge rates here
783
00:23:43,759 --> 00:23:47,599
and the only way to handle that is
784
00:23:45,359 --> 00:23:48,479
you have to use a very fast memory like
785
00:23:47,599 --> 00:23:50,558
a hypercube
786
00:23:48,480 --> 00:23:52,159
it's a 3d integrated memory these are
787
00:23:50,558 --> 00:23:52,798
made by micron there's four of them
788
00:23:52,159 --> 00:23:54,559
again
789
00:23:52,798 --> 00:23:56,960
and these actually have more than a two
790
00:23:54,558 --> 00:23:59,119
gigabyte of of storage or two giga
791
00:23:56,960 --> 00:24:00,798
sample per of storage there's much more
792
00:23:59,119 --> 00:24:02,399
but the scope doesn't use all of the
793
00:24:00,798 --> 00:24:03,759
memory only uses a fraction of it
794
00:24:02,400 --> 00:24:06,240
because of the 32-bit
795
00:24:03,759 --> 00:24:06,798
architecture you can only address 4 gig
796
00:24:06,240 --> 00:24:08,400
in total
797
00:24:06,798 --> 00:24:09,599
so anyway that's a minor detail and
798
00:24:08,400 --> 00:24:10,720
perhaps something that they will change
799
00:24:09,599 --> 00:24:12,639
in the future but there is
800
00:24:10,720 --> 00:24:14,640
memory available more than what is
801
00:24:12,640 --> 00:24:16,240
expected for the scope
802
00:24:14,640 --> 00:24:18,080
so now we have this chain i actually
803
00:24:16,240 --> 00:24:19,919
have the ics
804
00:24:18,079 --> 00:24:21,278
for each of them and we should be able
805
00:24:19,919 --> 00:24:21,600
to take a look at it and i'll show you
806
00:24:21,278 --> 00:24:23,278
that
807
00:24:21,599 --> 00:24:24,719
in just a second to show you what goes
808
00:24:23,278 --> 00:24:26,558
into those and they're decapped
809
00:24:24,720 --> 00:24:27,519
so we can even look at what's underneath
810
00:24:26,558 --> 00:24:28,399
them although you're really not going to
811
00:24:27,519 --> 00:24:29,679
be able to see
812
00:24:28,400 --> 00:24:31,519
very much because it's all flip chip
813
00:24:29,679 --> 00:24:32,400
technology so take a moment and
814
00:24:31,519 --> 00:24:33,679
appreciate
815
00:24:32,400 --> 00:24:35,278
what's happening here how much
816
00:24:33,679 --> 00:24:36,798
technology is involved because the front
817
00:24:35,278 --> 00:24:38,079
has this indium phosphide
818
00:24:36,798 --> 00:24:40,319
then we're going to get to silicon
819
00:24:38,079 --> 00:24:41,759
germanium and then there is a two chips
820
00:24:40,319 --> 00:24:43,918
inside the hd converter
821
00:24:41,759 --> 00:24:46,558
silicon germanium and then 65 nanometers
822
00:24:43,919 --> 00:24:48,559
cmos and after that you go into a dsp
823
00:24:46,558 --> 00:24:49,918
which itself is either 24 or 28
824
00:24:48,558 --> 00:24:51,759
nanometers cmos so
825
00:24:49,919 --> 00:24:53,679
they've designed across so many
826
00:24:51,759 --> 00:24:56,158
technologies to make this happen
827
00:24:53,679 --> 00:24:58,159
this is a lot of work and it is really
828
00:24:56,159 --> 00:24:59,679
amazing to see how many engineers
829
00:24:58,159 --> 00:25:01,919
have to work together and how complex
830
00:24:59,679 --> 00:25:03,679
this system is so let's look at it
831
00:25:01,919 --> 00:25:05,600
a little bit from the edge here this is
832
00:25:03,679 --> 00:25:07,440
a fairly thick board as you can imagine
833
00:25:05,599 --> 00:25:10,480
because these ics have a
834
00:25:07,440 --> 00:25:11,840
lot of bga balls on the back of them
835
00:25:10,480 --> 00:25:12,720
when you look at the density of this
836
00:25:11,839 --> 00:25:15,519
package here
837
00:25:12,720 --> 00:25:16,880
and that's the digital processor part so
838
00:25:15,519 --> 00:25:18,558
then all of that of course has to go
839
00:25:16,880 --> 00:25:21,120
into the board so then this might be a
840
00:25:18,558 --> 00:25:22,319
maybe a 32 layer board not quite sure
841
00:25:21,119 --> 00:25:24,079
it's a power connector
842
00:25:22,319 --> 00:25:26,079
massive massive power connector here on
843
00:25:24,079 --> 00:25:26,960
the left some digital connectors on the
844
00:25:26,079 --> 00:25:30,399
other side and
845
00:25:26,960 --> 00:25:32,640
check this out this is just
846
00:25:30,400 --> 00:25:34,320
beautiful a lot of linear technology
847
00:25:32,640 --> 00:25:36,080
dc-dc converters these are the best
848
00:25:34,319 --> 00:25:37,759
pretty much you can buy in the world you
849
00:25:36,079 --> 00:25:38,399
can see some of the other traces going
850
00:25:37,759 --> 00:25:39,759
in
851
00:25:38,400 --> 00:25:41,200
there's some cable cut here it's most
852
00:25:39,759 --> 00:25:42,720
likely they were injecting clock because
853
00:25:41,200 --> 00:25:43,759
the clock also needs to go into these
854
00:25:42,720 --> 00:25:45,839
atd converters
855
00:25:43,759 --> 00:25:48,000
the clock needs to be synchronized
856
00:25:45,839 --> 00:25:50,158
aligned to the clock of the front end oh
857
00:25:48,000 --> 00:25:51,759
it's just an absolute nightmare
858
00:25:50,159 --> 00:25:53,840
i'm going to zoom in a little bit so you
859
00:25:51,759 --> 00:25:55,519
can see some of the beautiful layout
860
00:25:53,839 --> 00:25:57,759
that is done here i mean take a look at
861
00:25:55,519 --> 00:25:59,918
these decoupling capacitors is just
862
00:25:57,759 --> 00:26:01,200
insane and they're all at 45 degree
863
00:25:59,919 --> 00:26:03,120
angle because you can't
864
00:26:01,200 --> 00:26:04,960
you can't fit them otherwise this thing
865
00:26:03,119 --> 00:26:06,479
is is absolutely beautiful
866
00:26:04,960 --> 00:26:07,679
i think i'm gonna have to take a couple
867
00:26:06,480 --> 00:26:09,038
of pictures of these and make a
868
00:26:07,679 --> 00:26:11,679
wallpaper
869
00:26:09,038 --> 00:26:12,720
out of this it's just amazing to see how
870
00:26:11,679 --> 00:26:15,919
much is going on
871
00:26:12,720 --> 00:26:17,120
at the back of this board so all the
872
00:26:15,919 --> 00:26:18,640
power handling all the
873
00:26:17,119 --> 00:26:20,639
calculations for how to make sure
874
00:26:18,640 --> 00:26:21,679
there's no noise coupling there is
875
00:26:20,640 --> 00:26:22,960
what do you do with the digital noise
876
00:26:21,679 --> 00:26:24,240
what do you do with analog noise it's
877
00:26:22,960 --> 00:26:26,079
just crazy
878
00:26:24,240 --> 00:26:27,519
that the board this board design is is
879
00:26:26,079 --> 00:26:30,879
wonderful so now we can
880
00:26:27,519 --> 00:26:33,519
flip it back over here and let's take a
881
00:26:30,880 --> 00:26:34,799
look at inside the package of the a2d
882
00:26:33,519 --> 00:26:37,038
converter
883
00:26:34,798 --> 00:26:38,639
now this is the same a2d converter as
884
00:26:37,038 --> 00:26:41,038
the s-series scope
885
00:26:38,640 --> 00:26:42,240
let me find a good angle here so we can
886
00:26:41,038 --> 00:26:44,480
also zoom into it
887
00:26:42,240 --> 00:26:46,240
a little bit more there it is so this
888
00:26:44,480 --> 00:26:46,798
a2d converter itself is a two-chip
889
00:26:46,240 --> 00:26:48,558
module
890
00:26:46,798 --> 00:26:50,480
which is a front-end silicon germanium
891
00:26:48,558 --> 00:26:52,720
module that front-end handles the
892
00:26:50,480 --> 00:26:55,278
amplifying and aligning and whatever it
893
00:26:52,720 --> 00:26:57,600
is required to feed the signal into the
894
00:26:55,278 --> 00:27:00,079
64 giga sample per second four channel
895
00:26:57,599 --> 00:27:01,759
input a 2d converter which is in 65
896
00:27:00,079 --> 00:27:04,000
nanometer cmos it's a very
897
00:27:01,759 --> 00:27:06,158
straightforward uh packaging nothing
898
00:27:04,000 --> 00:27:07,679
unusual there you can see the front end
899
00:27:06,159 --> 00:27:08,720
they have done a little bit differently
900
00:27:07,679 --> 00:27:10,480
in terms of where to
901
00:27:08,720 --> 00:27:12,319
place the balls to create good signal
902
00:27:10,480 --> 00:27:13,278
integrity and so on so two chip
903
00:27:12,319 --> 00:27:15,759
solutions sigi
904
00:27:13,278 --> 00:27:16,720
and cmos a hybrid adc module same thing
905
00:27:15,759 --> 00:27:19,278
you will find
906
00:27:16,720 --> 00:27:20,240
directly in the s series scope which
907
00:27:19,278 --> 00:27:22,558
looks quite nice
908
00:27:20,240 --> 00:27:24,319
so then now if you if you're keeping
909
00:27:22,558 --> 00:27:25,519
track we have a front-end indium
910
00:27:24,319 --> 00:27:28,079
phosphate amplifier
911
00:27:25,519 --> 00:27:28,720
front-end indium phosphide sampler sigi
912
00:27:28,079 --> 00:27:31,359
sampler
913
00:27:28,720 --> 00:27:32,558
cd pre-amplifier 65 nanometer cmos
914
00:27:31,359 --> 00:27:34,079
that's how the data
915
00:27:32,558 --> 00:27:35,839
ultimately gets to the a2d converter
916
00:27:34,079 --> 00:27:37,519
that's just crazy crazy
917
00:27:35,839 --> 00:27:38,879
and then if you can also take a look
918
00:27:37,519 --> 00:27:40,558
similarly
919
00:27:38,880 --> 00:27:42,000
to the chipset that the airspeed which
920
00:27:40,558 --> 00:27:42,720
is actually right on top of underneath
921
00:27:42,000 --> 00:27:44,880
this heatsink
922
00:27:42,720 --> 00:27:46,079
you can take that cup cap off that's a
923
00:27:44,880 --> 00:27:48,559
fairly large
924
00:27:46,079 --> 00:27:49,759
ic there this is a fully digital ic
925
00:27:48,558 --> 00:27:52,558
which interfaces with the
926
00:27:49,759 --> 00:27:54,240
a2d converters as i said tons and tons
927
00:27:52,558 --> 00:27:56,319
of ios in the back there
928
00:27:54,240 --> 00:27:58,079
very nice very nice design as to be
929
00:27:56,319 --> 00:28:00,960
expected it looks like a
930
00:27:58,079 --> 00:28:02,398
regular flip chip bga package there all
931
00:28:00,960 --> 00:28:04,159
of this heatsink of course
932
00:28:02,398 --> 00:28:05,678
has to come from the top flip chip gives
933
00:28:04,159 --> 00:28:07,039
you that advantage to get the heat from
934
00:28:05,679 --> 00:28:08,000
the top of the ic which is most
935
00:28:07,038 --> 00:28:10,000
efficient
936
00:28:08,000 --> 00:28:11,599
and uh these are most likely some test
937
00:28:10,000 --> 00:28:13,119
uh test ports here during
938
00:28:11,599 --> 00:28:14,639
during factory testing maybe it's
939
00:28:13,119 --> 00:28:16,158
interesting to see what they're doing
940
00:28:14,640 --> 00:28:18,399
here around the border there
941
00:28:16,159 --> 00:28:19,200
and yeah it's it's 28 diameter 24.99 we
942
00:28:18,398 --> 00:28:20,798
are not sure but
943
00:28:19,200 --> 00:28:22,240
it has all the interfaces required to
944
00:28:20,798 --> 00:28:23,679
grab data from day to day and to
945
00:28:22,240 --> 00:28:26,319
interface with the hypercube
946
00:28:23,679 --> 00:28:28,159
memory from micron so there it is that's
947
00:28:26,319 --> 00:28:31,278
uh that's kind of what it looks like
948
00:28:28,159 --> 00:28:34,080
to get from a 110 gigahertz analog input
949
00:28:31,278 --> 00:28:35,038
all the way to 256 gigahertz per second
950
00:28:34,079 --> 00:28:38,000
digital
951
00:28:35,038 --> 00:28:39,839
sampling there now because this
952
00:28:38,000 --> 00:28:41,519
front-end module is kind of
953
00:28:39,839 --> 00:28:42,959
designed in such a way to accept
954
00:28:41,519 --> 00:28:44,879
broken-down signals
955
00:28:42,960 --> 00:28:46,240
down to you know eight up to eight
956
00:28:44,880 --> 00:28:47,919
gigahertz per channel
957
00:28:46,240 --> 00:28:49,679
it means that you can take advantage of
958
00:28:47,919 --> 00:28:53,159
the architecture of this sampler
959
00:28:49,679 --> 00:28:55,840
and break it into two and make two
960
00:28:53,159 --> 00:28:56,399
128 giga sample per second front end so
961
00:28:55,839 --> 00:28:59,678
you can
962
00:28:56,398 --> 00:29:02,639
operate this half at 128 gig
963
00:28:59,679 --> 00:29:04,240
and this app and 128 gig and created two
964
00:29:02,640 --> 00:29:06,320
channel other scope
965
00:29:04,240 --> 00:29:08,159
with half the sample rate if you don't
966
00:29:06,319 --> 00:29:09,519
need to go to 110 gigahertz and that's
967
00:29:08,159 --> 00:29:11,440
exactly what they have done
968
00:29:09,519 --> 00:29:13,278
and that's what the purpose of this
969
00:29:11,440 --> 00:29:15,200
other sampler module is
970
00:29:13,278 --> 00:29:17,359
if you look at this other sampler module
971
00:29:15,200 --> 00:29:19,278
it has the same interface connector
972
00:29:17,359 --> 00:29:20,798
so you can connect directly to the same
973
00:29:19,278 --> 00:29:22,000
front end here
974
00:29:20,798 --> 00:29:23,759
and the only thing that's different is
975
00:29:22,000 --> 00:29:24,480
that instead of having the indium
976
00:29:23,759 --> 00:29:25,839
phosphite
977
00:29:24,480 --> 00:29:27,519
you don't need any of phosphate anymore
978
00:29:25,839 --> 00:29:29,359
because the front-end bandwidth can be
979
00:29:27,519 --> 00:29:31,440
limited to 32 gigahertz if you don't
980
00:29:29,359 --> 00:29:33,278
need to go above 32 gigahertz well
981
00:29:31,440 --> 00:29:35,759
we can give you two channels like this
982
00:29:33,278 --> 00:29:37,359
and you can have samplers that are
983
00:29:35,759 --> 00:29:39,200
the same sticky samplers which are much
984
00:29:37,359 --> 00:29:40,879
simpler now to work with
985
00:29:39,200 --> 00:29:42,960
and the front end can be entirely done
986
00:29:40,880 --> 00:29:44,880
in silicon germanium reduces the cost
987
00:29:42,960 --> 00:29:46,880
significantly the components are
988
00:29:44,880 --> 00:29:48,159
much much fewer everything is flip
989
00:29:46,880 --> 00:29:48,640
shipped over here there's no wire
990
00:29:48,159 --> 00:29:51,200
bonding
991
00:29:48,640 --> 00:29:52,799
this is all reflow process makes things
992
00:29:51,200 --> 00:29:54,640
significantly lower cost
993
00:29:52,798 --> 00:29:56,558
and now you can have two channels up to
994
00:29:54,640 --> 00:29:58,880
about 32 gigahertz a bandwidth or so
995
00:29:56,558 --> 00:29:59,839
but then each of them can be 128 giga
996
00:29:58,880 --> 00:30:02,880
sample per second
997
00:29:59,839 --> 00:30:05,359
now this is the first time in history
998
00:30:02,880 --> 00:30:07,440
ever that you have a front-end module
999
00:30:05,359 --> 00:30:08,639
that can be configured in such a
1000
00:30:07,440 --> 00:30:10,798
flexible way
1001
00:30:08,640 --> 00:30:12,320
all the way up to 256 giga sample per
1002
00:30:10,798 --> 00:30:14,480
second so if you buy a scope
1003
00:30:12,319 --> 00:30:16,398
that has this in it you can keep
1004
00:30:14,480 --> 00:30:17,919
upgrading it by just simply changing the
1005
00:30:16,398 --> 00:30:19,119
front-end module by sending it to
1006
00:30:17,919 --> 00:30:20,559
factory they'll just change this but
1007
00:30:19,119 --> 00:30:22,719
they'll keep this piece
1008
00:30:20,558 --> 00:30:24,000
in your scope so the same chassis with
1009
00:30:22,720 --> 00:30:26,640
the same serial number
1010
00:30:24,000 --> 00:30:27,839
can be upgraded across a huge range of
1011
00:30:26,640 --> 00:30:29,840
bandwidths
1012
00:30:27,839 --> 00:30:31,278
as your needs change and that lowers the
1013
00:30:29,839 --> 00:30:33,519
ownership cost
1014
00:30:31,278 --> 00:30:34,398
of the ecosystem that's one of the key
1015
00:30:33,519 --> 00:30:35,918
factors here
1016
00:30:34,398 --> 00:30:37,359
is that you don't want to have to keep
1017
00:30:35,919 --> 00:30:38,480
throwing everything out and you see
1018
00:30:37,359 --> 00:30:40,398
here's the same
1019
00:30:38,480 --> 00:30:42,079
clocking scheme used in the back so it's
1020
00:30:40,398 --> 00:30:43,759
essentially identical
1021
00:30:42,079 --> 00:30:45,278
to the other one all you need to do is
1022
00:30:43,759 --> 00:30:47,038
just change the front end
1023
00:30:45,278 --> 00:30:48,960
uh back to this one if you want to go to
1024
00:30:47,038 --> 00:30:50,319
higher frequencies everything stays the
1025
00:30:48,960 --> 00:30:52,558
same and then the software
1026
00:30:50,319 --> 00:30:53,439
handles it really quite beautiful
1027
00:30:52,558 --> 00:30:55,119
amazing
1028
00:30:53,440 --> 00:30:56,798
amazing stuff going on here the amount
1029
00:30:55,119 --> 00:30:58,079
of engineering that keysight engineers
1030
00:30:56,798 --> 00:31:00,000
should really be proud
1031
00:30:58,079 --> 00:31:01,278
to put something out here on the market
1032
00:31:00,000 --> 00:31:02,960
so there it is that's a
1033
00:31:01,278 --> 00:31:04,798
detailed look at how it works how it
1034
00:31:02,960 --> 00:31:05,200
compares and how it's manufactured but
1035
00:31:04,798 --> 00:31:07,440
now
1036
00:31:05,200 --> 00:31:09,120
the fun part let's go turn the thing on
1037
00:31:07,440 --> 00:31:10,880
put some signal into it and see how it
1038
00:31:09,119 --> 00:31:13,359
behaves
1039
00:31:10,880 --> 00:31:14,799
so right now we have the scope at 500
1040
00:31:13,359 --> 00:31:16,398
millivolt per division which is the
1041
00:31:14,798 --> 00:31:18,480
maximum input signal
1042
00:31:16,398 --> 00:31:20,079
and this is absorbed by the front end
1043
00:31:18,480 --> 00:31:21,919
and mechanical attenuator
1044
00:31:20,079 --> 00:31:23,839
but we can go all the way down to two
1045
00:31:21,919 --> 00:31:25,519
millivolt per division in hardware
1046
00:31:23,839 --> 00:31:27,119
and there's a one millivolt per division
1047
00:31:25,519 --> 00:31:28,399
which is the further amplification
1048
00:31:27,119 --> 00:31:30,719
that's done internally
1049
00:31:28,398 --> 00:31:32,398
so i can go all the way down to a two
1050
00:31:30,720 --> 00:31:33,919
millivolt per division
1051
00:31:32,398 --> 00:31:35,599
and this is the the best that the
1052
00:31:33,919 --> 00:31:37,759
hardware can do so now the mechanical
1053
00:31:35,599 --> 00:31:39,678
attenuator is at zero db attenuation
1054
00:31:37,759 --> 00:31:41,839
and the front-end indium phosphide
1055
00:31:39,679 --> 00:31:44,080
pre-amplifier is at the maximum gain
1056
00:31:41,839 --> 00:31:46,000
this is going to be the the lowest noise
1057
00:31:44,079 --> 00:31:48,879
the scope is going to be able to support
1058
00:31:46,000 --> 00:31:50,798
now traditionally this noise goes up for
1059
00:31:48,880 --> 00:31:52,480
some of the other competing products but
1060
00:31:50,798 --> 00:31:54,000
they've kept the noise really low and
1061
00:31:52,480 --> 00:31:54,798
we'll take a look at that it's really
1062
00:31:54,000 --> 00:31:56,640
amazing
1063
00:31:54,798 --> 00:31:58,158
and right now we're sitting at one micro
1064
00:31:56,640 --> 00:32:00,480
second per division over here
1065
00:31:58,159 --> 00:32:01,919
but check out how far you can go it's
1066
00:32:00,480 --> 00:32:03,839
crazy so
1067
00:32:01,919 --> 00:32:05,360
we can continue to go forward right now
1068
00:32:03,839 --> 00:32:07,038
we are at uh
1069
00:32:05,359 --> 00:32:09,278
one nanosecond per division and i can
1070
00:32:07,038 --> 00:32:12,240
keep going all the way to one
1071
00:32:09,278 --> 00:32:14,960
picosecond per division one of these
1072
00:32:12,240 --> 00:32:15,359
horizontal divisions is one picosecond
1073
00:32:14,960 --> 00:32:17,440
that's
1074
00:32:15,359 --> 00:32:19,918
an extraordinary thing to see and this
1075
00:32:17,440 --> 00:32:20,399
is what you get from 256 giga sample per
1076
00:32:19,919 --> 00:32:21,919
second
1077
00:32:20,398 --> 00:32:24,239
so let's go back all the way to one
1078
00:32:21,919 --> 00:32:25,759
microsecond and the reason the reason i
1079
00:32:24,240 --> 00:32:26,960
want to go to one microsecond is because
1080
00:32:25,759 --> 00:32:29,759
i want to capture
1081
00:32:26,960 --> 00:32:30,319
a lot of noise and measure that noise by
1082
00:32:29,759 --> 00:32:33,359
doing an
1083
00:32:30,319 --> 00:32:34,558
rms calculation directly on it so right
1084
00:32:33,359 --> 00:32:38,639
now the scope is
1085
00:32:34,558 --> 00:32:41,440
set to 113 gigahertz full bandwidth and
1086
00:32:38,640 --> 00:32:43,120
113 gigahertz of full trigger bandwidth
1087
00:32:41,440 --> 00:32:45,278
so this is as much noise you can
1088
00:32:43,119 --> 00:32:47,518
ever collect from the front end because
1089
00:32:45,278 --> 00:32:48,798
this is full bandwidth at play here
1090
00:32:47,519 --> 00:32:51,599
so let's go ahead and see how much noise
1091
00:32:48,798 --> 00:32:54,240
that is i've already set that up here
1092
00:32:51,599 --> 00:32:56,319
there it is so the rms noise for a one
1093
00:32:54,240 --> 00:33:00,000
microsecond period
1094
00:32:56,319 --> 00:33:01,439
at 256 giga sample per second over 113
1095
00:33:00,000 --> 00:33:05,119
gigahertz of bandwidth
1096
00:33:01,440 --> 00:33:08,720
is on average about
1097
00:33:05,119 --> 00:33:10,959
just under 700 micro volts rms that's
1098
00:33:08,720 --> 00:33:12,558
wonderful that's amazing to see and it's
1099
00:33:10,960 --> 00:33:14,798
at minus 50 dbm
1100
00:33:12,558 --> 00:33:16,398
that's the in dbm equivalent of it this
1101
00:33:14,798 --> 00:33:17,359
is with nothing connected to the input
1102
00:33:16,398 --> 00:33:20,319
of the oscilloscope
1103
00:33:17,359 --> 00:33:21,599
as you can clearly see now this noise is
1104
00:33:20,319 --> 00:33:23,918
for the full bandwidth
1105
00:33:21,599 --> 00:33:24,639
however if you don't need 113 gigahertz
1106
00:33:23,919 --> 00:33:27,120
of bandwidth
1107
00:33:24,640 --> 00:33:28,320
you can reduce this equivalent noise by
1108
00:33:27,119 --> 00:33:30,319
filtering everything
1109
00:33:28,319 --> 00:33:32,480
that you don't want it can be all done
1110
00:33:30,319 --> 00:33:33,759
in dsp because the adcs are capturing
1111
00:33:32,480 --> 00:33:36,399
the entire bandwidth
1112
00:33:33,759 --> 00:33:37,200
all together at the same time anyway and
1113
00:33:36,398 --> 00:33:38,719
there is no
1114
00:33:37,200 --> 00:33:40,798
hardware filter in front of this it's
1115
00:33:38,720 --> 00:33:41,759
just not really possible to put anything
1116
00:33:40,798 --> 00:33:42,798
in the front of this
1117
00:33:41,759 --> 00:33:44,640
and that wouldn't degrade the
1118
00:33:42,798 --> 00:33:45,440
performance so they just capture
1119
00:33:44,640 --> 00:33:47,278
everything
1120
00:33:45,440 --> 00:33:48,480
and process and deal with it in the
1121
00:33:47,278 --> 00:33:50,558
digital dsp
1122
00:33:48,480 --> 00:33:52,319
chipset that i have designed directly
1123
00:33:50,558 --> 00:33:54,558
after the a2d converters so we can go
1124
00:33:52,319 --> 00:33:56,158
ahead and reduce the bandwidth
1125
00:33:54,558 --> 00:33:59,038
let's set the bandwidth from a let's say
1126
00:33:56,159 --> 00:34:01,278
113 gigahertz set it to 70 gigahertz
1127
00:33:59,038 --> 00:34:02,480
and keep an eye out on this number now
1128
00:34:01,278 --> 00:34:06,240
we are sitting
1129
00:34:02,480 --> 00:34:08,159
at 50 at 500 micro volt rms so you can
1130
00:34:06,240 --> 00:34:09,599
see that the noise is gradually going to
1131
00:34:08,159 --> 00:34:11,119
come down
1132
00:34:09,599 --> 00:34:12,800
as you reduce the bandwidth so if you
1133
00:34:11,119 --> 00:34:15,280
don't need 70 gigahertz no problem
1134
00:34:12,800 --> 00:34:16,159
we can try 30 gigahertz here you go at
1135
00:34:15,280 --> 00:34:17,919
30 gigahertz
1136
00:34:16,159 --> 00:34:20,320
we're going to be down to about 300
1137
00:34:17,918 --> 00:34:22,480
micro volt rms which is tiny
1138
00:34:20,320 --> 00:34:23,760
it's it's by far the best there is in
1139
00:34:22,480 --> 00:34:27,039
this kind of performance
1140
00:34:23,760 --> 00:34:29,440
now if i go all the way to one gigahertz
1141
00:34:27,039 --> 00:34:31,119
of course you can see at one gigahertz
1142
00:34:29,440 --> 00:34:33,760
the noise almost completely disappears
1143
00:34:31,119 --> 00:34:35,919
we're sitting at 73 micro volt
1144
00:34:33,760 --> 00:34:37,679
on average or about 0 the mean is about
1145
00:34:35,918 --> 00:34:41,199
80 micro volt on average
1146
00:34:37,679 --> 00:34:44,000
now keep in mind that this 80 micro volt
1147
00:34:41,199 --> 00:34:46,878
average noise is the noise from a front
1148
00:34:44,000 --> 00:34:48,559
end that supports 113 gigahertz of
1149
00:34:46,878 --> 00:34:51,118
bandwidth if you were to build a front
1150
00:34:48,559 --> 00:34:52,878
end only for one gigahertz of bandwidth
1151
00:34:51,119 --> 00:34:54,639
sure you can build that perhaps even
1152
00:34:52,878 --> 00:34:55,519
less noise and you can see that in the s
1153
00:34:54,639 --> 00:34:58,000
series scope
1154
00:34:55,519 --> 00:35:00,719
but this is the same front end that is
1155
00:34:58,000 --> 00:35:02,400
capturing 113 gigahertz of bandwidth and
1156
00:35:00,719 --> 00:35:04,879
for that to go this low
1157
00:35:02,400 --> 00:35:06,639
is just crazy and it's amazing to see
1158
00:35:04,880 --> 00:35:08,240
that the process and the design
1159
00:35:06,639 --> 00:35:09,598
of the front-end amplifier and of course
1160
00:35:08,239 --> 00:35:10,399
the noise from attenuators and
1161
00:35:09,599 --> 00:35:13,200
everything is
1162
00:35:10,400 --> 00:35:14,639
is quite good and this is how they get
1163
00:35:13,199 --> 00:35:16,239
some of the performances that they are
1164
00:35:14,639 --> 00:35:17,519
able to do and i'll show you that at the
1165
00:35:16,239 --> 00:35:19,838
very end of the video
1166
00:35:17,519 --> 00:35:21,920
by doing ffts and so on and you can see
1167
00:35:19,838 --> 00:35:23,519
exactly how much noise can be captured
1168
00:35:21,920 --> 00:35:25,280
so from the noise point of view this is
1169
00:35:23,519 --> 00:35:28,079
amazing and i will show you the
1170
00:35:25,280 --> 00:35:29,359
the enob as a function of noise as well
1171
00:35:28,079 --> 00:35:31,200
one of the other very interesting thing
1172
00:35:29,358 --> 00:35:33,679
about this design is that
1173
00:35:31,199 --> 00:35:34,239
the effective number of bits of the
1174
00:35:33,679 --> 00:35:36,319
scope
1175
00:35:34,239 --> 00:35:37,358
scales with the bandwidth you're using
1176
00:35:36,320 --> 00:35:39,760
and that's because it's
1177
00:35:37,358 --> 00:35:40,799
purely limited by the noise and it's not
1178
00:35:39,760 --> 00:35:43,040
limited by
1179
00:35:40,800 --> 00:35:44,640
harmonics and by distortion and because
1180
00:35:43,039 --> 00:35:46,159
if it was limited by distortion
1181
00:35:44,639 --> 00:35:47,679
even if you reduce the bandwidth you
1182
00:35:46,159 --> 00:35:49,519
wouldn't be able to get rid of it that's
1183
00:35:47,679 --> 00:35:51,358
a huge plus for this scope
1184
00:35:49,519 --> 00:35:53,199
that you can change the enop as a
1185
00:35:51,358 --> 00:35:54,639
function of bandwidth and we'll show you
1186
00:35:53,199 --> 00:35:56,000
that plots at the end
1187
00:35:54,639 --> 00:35:57,759
now having said that well how do you
1188
00:35:56,000 --> 00:35:59,519
calibrate the scope like this
1189
00:35:57,760 --> 00:36:00,880
you know you spend a 1.3 million dollars
1190
00:35:59,519 --> 00:36:01,599
you get a four channel scope on your
1191
00:36:00,880 --> 00:36:03,920
bench
1192
00:36:01,599 --> 00:36:05,680
and now you want to calibrate it and
1193
00:36:03,920 --> 00:36:07,440
align all the channels to each other
1194
00:36:05,679 --> 00:36:09,440
and make sure that the frequency
1195
00:36:07,440 --> 00:36:11,440
response and the group delayed response
1196
00:36:09,440 --> 00:36:13,838
of all the channels are matched you
1197
00:36:11,440 --> 00:36:15,599
would need a 110 gigahertz source and
1198
00:36:13,838 --> 00:36:16,480
you would need a nist calibration in
1199
00:36:15,599 --> 00:36:17,760
order to do this
1200
00:36:16,480 --> 00:36:19,760
and of course you don't want to send the
1201
00:36:17,760 --> 00:36:20,560
scope to keysight every time you need to
1202
00:36:19,760 --> 00:36:22,640
calibrate it
1203
00:36:20,559 --> 00:36:24,400
that wouldn't be practical so what they
1204
00:36:22,639 --> 00:36:25,039
have done is that they have created a
1205
00:36:24,400 --> 00:36:27,280
probe
1206
00:36:25,039 --> 00:36:28,960
that plugs in directly to the front and
1207
00:36:27,280 --> 00:36:30,880
that probe is their own design that
1208
00:36:28,960 --> 00:36:33,358
creates a very sharp edge
1209
00:36:30,880 --> 00:36:34,800
that allows you to calibrate the scope
1210
00:36:33,358 --> 00:36:36,078
to the output of the probe
1211
00:36:34,800 --> 00:36:38,000
and i'll show you that probe in just a
1212
00:36:36,079 --> 00:36:39,200
second and tell you how it works it's a
1213
00:36:38,000 --> 00:36:41,358
good clever way
1214
00:36:39,199 --> 00:36:43,118
of being able to do calibration directly
1215
00:36:41,358 --> 00:36:43,920
in your lab so you never have to send
1216
00:36:43,119 --> 00:36:46,720
the unit back
1217
00:36:43,920 --> 00:36:48,720
to get it miscalibrated and here's a
1218
00:36:46,719 --> 00:36:50,159
close look at the calibrator
1219
00:36:48,719 --> 00:36:52,559
now the front end of the calibrator
1220
00:36:50,159 --> 00:36:54,000
directly mates with the front connector
1221
00:36:52,559 --> 00:36:56,078
of the unit it's a one millimeter
1222
00:36:54,000 --> 00:36:57,679
ruggedized connector and everything in
1223
00:36:56,079 --> 00:36:58,320
here and this module is designed by
1224
00:36:57,679 --> 00:37:00,159
keysight
1225
00:36:58,320 --> 00:37:01,680
it has their own three five process
1226
00:37:00,159 --> 00:37:03,199
components and i will show you from a
1227
00:37:01,679 --> 00:37:03,838
photograph and we'll talk about how it
1228
00:37:03,199 --> 00:37:06,480
works
1229
00:37:03,838 --> 00:37:07,679
now as i said this is a nist calibration
1230
00:37:06,480 --> 00:37:09,599
unit meaning that
1231
00:37:07,679 --> 00:37:11,598
the exact response of it up to the edge
1232
00:37:09,599 --> 00:37:13,119
of this connector is stored in here
1233
00:37:11,599 --> 00:37:14,800
and once you connect this to the unit it
1234
00:37:13,119 --> 00:37:15,519
will download this response into the
1235
00:37:14,800 --> 00:37:17,039
instrument
1236
00:37:15,519 --> 00:37:18,559
and the instrument will try and match
1237
00:37:17,039 --> 00:37:18,960
this response from the measurements it's
1238
00:37:18,559 --> 00:37:20,960
getting
1239
00:37:18,960 --> 00:37:22,559
so it doesn't even matter but that
1240
00:37:20,960 --> 00:37:23,679
response is as long as it has enough
1241
00:37:22,559 --> 00:37:25,519
frequency content
1242
00:37:23,679 --> 00:37:26,719
to meet the 110 gigahertz requirement
1243
00:37:25,519 --> 00:37:28,639
but that's pretty clever
1244
00:37:26,719 --> 00:37:30,319
and a good simple way to do this as long
1245
00:37:28,639 --> 00:37:31,519
as you can create this device and i'll
1246
00:37:30,320 --> 00:37:33,838
show you what that is
1247
00:37:31,519 --> 00:37:35,599
in just a second and now here are some
1248
00:37:33,838 --> 00:37:36,880
of the connectors that it comes
1249
00:37:35,599 --> 00:37:39,039
with the unit as well this is for
1250
00:37:36,880 --> 00:37:39,838
example to convert the ruggedized one
1251
00:37:39,039 --> 00:37:41,920
millimeter
1252
00:37:39,838 --> 00:37:44,078
connector to a v connector there's also
1253
00:37:41,920 --> 00:37:45,519
k connectors and here's a one millimeter
1254
00:37:44,079 --> 00:37:47,599
to one millimeter connector
1255
00:37:45,519 --> 00:37:48,559
and the advantage of these is that you
1256
00:37:47,599 --> 00:37:50,320
will just basically
1257
00:37:48,559 --> 00:37:52,078
age these connectors and in case
1258
00:37:50,320 --> 00:37:53,599
something breaks it would be these ones
1259
00:37:52,079 --> 00:37:55,519
as opposed to the front of the unit
1260
00:37:53,599 --> 00:37:57,039
itself these are much cheaper to replace
1261
00:37:55,519 --> 00:37:58,480
than of course the front end connector
1262
00:37:57,039 --> 00:38:00,000
of the instrument it will be
1263
00:37:58,480 --> 00:38:01,679
significantly
1264
00:38:00,000 --> 00:38:03,280
more difficult you have to send it into
1265
00:38:01,679 --> 00:38:04,078
keysight but these also come with the
1266
00:38:03,280 --> 00:38:05,839
uxa
1267
00:38:04,079 --> 00:38:07,359
the spectrum analyzer as well so let's
1268
00:38:05,838 --> 00:38:08,639
go ahead and connect this to the scope
1269
00:38:07,358 --> 00:38:12,239
but before that let me show you what's
1270
00:38:08,639 --> 00:38:13,838
in it and exactly how it works
1271
00:38:12,239 --> 00:38:16,078
and here is what is inside this
1272
00:38:13,838 --> 00:38:16,400
calibrator probe here so this module
1273
00:38:16,079 --> 00:38:18,000
that
1274
00:38:16,400 --> 00:38:19,680
we were just looking at this is what it
1275
00:38:18,000 --> 00:38:21,358
looks like on the inside
1276
00:38:19,679 --> 00:38:22,879
so there is an ic in the front which
1277
00:38:21,358 --> 00:38:24,319
take we'll take a look in just a second
1278
00:38:22,880 --> 00:38:25,280
and there's a signal deck which can be
1279
00:38:24,320 --> 00:38:27,599
fed directly
1280
00:38:25,280 --> 00:38:30,079
into this ic now this signal is going to
1281
00:38:27,599 --> 00:38:32,000
come from the auxiliary port of the
1282
00:38:30,079 --> 00:38:34,000
oscilloscope itself which generates a
1283
00:38:32,000 --> 00:38:35,920
square wave and the rise time of that
1284
00:38:34,000 --> 00:38:38,559
square of it is nowhere close
1285
00:38:35,920 --> 00:38:39,680
to 110 gigahertz equivalent bandwidth so
1286
00:38:38,559 --> 00:38:42,400
you're going to have to
1287
00:38:39,679 --> 00:38:43,838
increase the edge sharpness and reduce
1288
00:38:42,400 --> 00:38:45,039
the rise and fall times before it
1289
00:38:43,838 --> 00:38:47,199
becomes useful
1290
00:38:45,039 --> 00:38:49,039
for doing this calibration up to 110
1291
00:38:47,199 --> 00:38:50,639
gigahertz which is exactly what this
1292
00:38:49,039 --> 00:38:52,880
is so after a little bit of filtering
1293
00:38:50,639 --> 00:38:55,118
over here the square wave that's coming
1294
00:38:52,880 --> 00:38:57,119
from the auxiliary port is going to go
1295
00:38:55,119 --> 00:38:58,240
through this indium phosphide limiting
1296
00:38:57,119 --> 00:39:00,000
amplifier
1297
00:38:58,239 --> 00:39:01,838
there are multiple stages of limiting
1298
00:39:00,000 --> 00:39:04,000
amplifier in a row here and every time
1299
00:39:01,838 --> 00:39:05,199
you limit this with very high gain and
1300
00:39:04,000 --> 00:39:06,880
very high accuracy
1301
00:39:05,199 --> 00:39:08,480
you reduce the rise and fall times and
1302
00:39:06,880 --> 00:39:10,720
they do this a couple of times until
1303
00:39:08,480 --> 00:39:12,559
until it's super sharp uh with the rise
1304
00:39:10,719 --> 00:39:14,000
time only a couple of picoseconds
1305
00:39:12,559 --> 00:39:16,320
and then they feed that out from the
1306
00:39:14,000 --> 00:39:17,920
connector to the scope now this is
1307
00:39:16,320 --> 00:39:19,838
really impressive this actually means
1308
00:39:17,920 --> 00:39:21,200
that the rise and fall times right here
1309
00:39:19,838 --> 00:39:23,599
at the edge of the ic
1310
00:39:21,199 --> 00:39:24,879
is even more because without this
1311
00:39:23,599 --> 00:39:25,680
interface and the connector and
1312
00:39:24,880 --> 00:39:26,960
everything because
1313
00:39:25,679 --> 00:39:28,799
all of that is going to of course
1314
00:39:26,960 --> 00:39:30,480
further limit the bandwidth it's quite
1315
00:39:28,800 --> 00:39:32,400
difficult to get all of that
1316
00:39:30,480 --> 00:39:34,320
working quite nicely so having said that
1317
00:39:32,400 --> 00:39:35,680
this is what's inside this all designed
1318
00:39:34,320 --> 00:39:37,359
by keysight again
1319
00:39:35,679 --> 00:39:39,598
from scratch just for the purposes of
1320
00:39:37,358 --> 00:39:42,400
being able to do this nist calibration
1321
00:39:39,599 --> 00:39:43,838
directly on this oscilloscope itself
1322
00:39:42,400 --> 00:39:45,440
which is really quite impressive
1323
00:39:43,838 --> 00:39:47,119
so now that we know how it works
1324
00:39:45,440 --> 00:39:47,920
internally we can go ahead and connect
1325
00:39:47,119 --> 00:39:50,800
it to the scope
1326
00:39:47,920 --> 00:39:52,960
and see how it behaves and here i have
1327
00:39:50,800 --> 00:39:54,720
the calibrator directly connected to
1328
00:39:52,960 --> 00:39:56,639
channel one of the instrument
1329
00:39:54,719 --> 00:39:58,319
and the interface to the front of the
1330
00:39:56,639 --> 00:39:59,039
scope the probe interface is also
1331
00:39:58,320 --> 00:40:00,640
connected
1332
00:39:59,039 --> 00:40:02,480
and the auxiliary output of the scope
1333
00:40:00,639 --> 00:40:04,239
which generates the square wave directly
1334
00:40:02,480 --> 00:40:05,838
goes to the calibrator
1335
00:40:04,239 --> 00:40:08,399
now as you know as i showed you what's
1336
00:40:05,838 --> 00:40:10,078
inside that the calibrator here those
1337
00:40:08,400 --> 00:40:12,079
edges coming from the auxiliary are
1338
00:40:10,079 --> 00:40:14,480
going to be continuously sharpened
1339
00:40:12,079 --> 00:40:15,839
up until to a few picoseconds so that we
1340
00:40:14,480 --> 00:40:17,358
can calibrate this scope with it and
1341
00:40:15,838 --> 00:40:19,519
this also shows you
1342
00:40:17,358 --> 00:40:21,519
how what kind of fast edges can actually
1343
00:40:19,519 --> 00:40:24,079
be captured here by the scope
1344
00:40:21,519 --> 00:40:26,159
right now we're looking at 512 million
1345
00:40:24,079 --> 00:40:27,359
points 200 microsecond per division we
1346
00:40:26,159 --> 00:40:28,879
can see the square root but we're
1347
00:40:27,358 --> 00:40:29,440
looking at it from really really far
1348
00:40:28,880 --> 00:40:31,200
away
1349
00:40:29,440 --> 00:40:33,679
so we're going to zoom in and see how
1350
00:40:31,199 --> 00:40:35,039
fast those edges actually are
1351
00:40:33,679 --> 00:40:37,118
so let's go ahead and see what happens
1352
00:40:35,039 --> 00:40:38,880
so i'm going to go all the way down so
1353
00:40:37,119 --> 00:40:40,000
right now we're at one microsecond keep
1354
00:40:38,880 --> 00:40:42,559
zooming in
1355
00:40:40,000 --> 00:40:44,639
five nanosecond one nanosecond and we're
1356
00:40:42,559 --> 00:40:46,318
going to go here there's 10 picosecond
1357
00:40:44,639 --> 00:40:49,279
per division and check it out we're
1358
00:40:46,318 --> 00:40:52,000
looking at a 3.3 picosecond
1359
00:40:49,280 --> 00:40:53,119
edge that's how fast and how sharp those
1360
00:40:52,000 --> 00:40:55,358
edges actually are
1361
00:40:53,119 --> 00:40:57,200
and this is why you're able to use this
1362
00:40:55,358 --> 00:41:00,159
as a missed calibrator
1363
00:40:57,199 --> 00:41:01,279
the exact group delay frequency content
1364
00:41:00,159 --> 00:41:03,440
phase information
1365
00:41:01,280 --> 00:41:05,119
that's coming out of the calibrator has
1366
00:41:03,440 --> 00:41:07,200
been transferred to the scope
1367
00:41:05,119 --> 00:41:08,960
and the scope is going to calibrate
1368
00:41:07,199 --> 00:41:09,679
itself and change the coefficients
1369
00:41:08,960 --> 00:41:13,440
required
1370
00:41:09,679 --> 00:41:15,440
to match that exactly so the calibration
1371
00:41:13,440 --> 00:41:16,480
plane of the calibrator is at the edge
1372
00:41:15,440 --> 00:41:18,079
of the connector
1373
00:41:16,480 --> 00:41:19,760
and the plane of the measurement of the
1374
00:41:18,079 --> 00:41:21,440
scope is at the connector of
1375
00:41:19,760 --> 00:41:22,960
the scope itself and those two are going
1376
00:41:21,440 --> 00:41:23,519
to match perfectly and that's how you
1377
00:41:22,960 --> 00:41:26,000
get
1378
00:41:23,519 --> 00:41:27,199
this calibration on the channel and it
1379
00:41:26,000 --> 00:41:30,559
looks amazing i mean
1380
00:41:27,199 --> 00:41:32,799
to get a a 3.2 picosecond edge
1381
00:41:30,559 --> 00:41:34,400
measured live on an oscilloscope it's
1382
00:41:32,800 --> 00:41:35,280
never been done before for a real-time
1383
00:41:34,400 --> 00:41:36,800
oscilloscope
1384
00:41:35,280 --> 00:41:38,560
and we can also go ahead and do a
1385
00:41:36,800 --> 00:41:42,480
calculation and do
1386
00:41:38,559 --> 00:41:42,480
a quick function to show
1387
00:41:42,559 --> 00:41:46,000
what the pulse response looks like and
1388
00:41:44,639 --> 00:41:48,719
here's the pulse response
1389
00:41:46,000 --> 00:41:50,800
and you can also see even the ripple the
1390
00:41:48,719 --> 00:41:52,719
ringing that comes from the sharp edge
1391
00:41:50,800 --> 00:41:53,519
just simply because the interface is not
1392
00:41:52,719 --> 00:41:54,879
perfect
1393
00:41:53,519 --> 00:41:56,880
it doesn't matter by the way that it's
1394
00:41:54,880 --> 00:41:58,240
ringing because that ring is included in
1395
00:41:56,880 --> 00:42:00,318
this calibration
1396
00:41:58,239 --> 00:42:02,239
performance so it doesn't matter what it
1397
00:42:00,318 --> 00:42:04,000
looks like it just needs to match
1398
00:42:02,239 --> 00:42:06,000
that response exactly and that's the
1399
00:42:04,000 --> 00:42:07,280
beauty of it each calibrator can be a
1400
00:42:06,000 --> 00:42:09,119
little bit different
1401
00:42:07,280 --> 00:42:10,800
but you can still perfectly calibrate
1402
00:42:09,119 --> 00:42:12,720
the scope as long as there is a
1403
00:42:10,800 --> 00:42:14,318
certain performance that each calibrator
1404
00:42:12,719 --> 00:42:17,039
meets and you can see the
1405
00:42:14,318 --> 00:42:18,639
ripples here and look at this impulse
1406
00:42:17,039 --> 00:42:21,199
response here
1407
00:42:18,639 --> 00:42:22,719
this is in giga volts per second that's
1408
00:42:21,199 --> 00:42:24,639
how sharp this edge is this
1409
00:42:22,719 --> 00:42:25,838
is amazing to see this and this is a
1410
00:42:24,639 --> 00:42:27,279
really good way
1411
00:42:25,838 --> 00:42:29,279
to quickly do this and not only can you
1412
00:42:27,280 --> 00:42:31,359
do the calibration for the phase
1413
00:42:29,280 --> 00:42:32,480
and group delay and frequency response
1414
00:42:31,358 --> 00:42:34,318
of the channel itself
1415
00:42:32,480 --> 00:42:36,000
this also aligns the channels to each
1416
00:42:34,318 --> 00:42:37,679
other which is another important thing
1417
00:42:36,000 --> 00:42:38,318
because you have a single source coming
1418
00:42:37,679 --> 00:42:40,639
in
1419
00:42:38,318 --> 00:42:42,239
and that source is self-aligned to all
1420
00:42:40,639 --> 00:42:43,759
the channels therefore and that's
1421
00:42:42,239 --> 00:42:45,439
that's how you can get everything
1422
00:42:43,760 --> 00:42:47,119
calibrated to each other which looks
1423
00:42:45,440 --> 00:42:49,119
great
1424
00:42:47,119 --> 00:42:50,480
so here's the setup to produce a very
1425
00:42:49,119 --> 00:42:51,440
high frequency signal to the
1426
00:42:50,480 --> 00:42:56,159
oscilloscope
1427
00:42:51,440 --> 00:42:58,880
now i'm using an hp 83752b to generate
1428
00:42:56,159 --> 00:43:00,719
a signal up to 20 gigahertz and then
1429
00:42:58,880 --> 00:43:01,119
pass it to this military multiplier
1430
00:43:00,719 --> 00:43:03,039
times
1431
00:43:01,119 --> 00:43:04,559
six that's going to generate signals up
1432
00:43:03,039 --> 00:43:06,318
to 120 gigahertz
1433
00:43:04,559 --> 00:43:08,159
and there is a waveguide to coax
1434
00:43:06,318 --> 00:43:09,759
converter and a one millimeter cable
1435
00:43:08,159 --> 00:43:10,879
which goes to the front of the unit
1436
00:43:09,760 --> 00:43:12,800
i think there is something a little
1437
00:43:10,880 --> 00:43:13,519
poetic about using such an old
1438
00:43:12,800 --> 00:43:15,599
instrument
1439
00:43:13,519 --> 00:43:17,679
to measure such a new instrument there's
1440
00:43:15,599 --> 00:43:18,880
a enormous gap in time and technology
1441
00:43:17,679 --> 00:43:20,239
between these two units
1442
00:43:18,880 --> 00:43:21,920
and i wonder if the engineers who
1443
00:43:20,239 --> 00:43:23,519
designed this ever envision
1444
00:43:21,920 --> 00:43:25,280
reaching this point which is really
1445
00:43:23,519 --> 00:43:26,800
amazing the other advantage of using
1446
00:43:25,280 --> 00:43:28,560
this is that because this is a such a
1447
00:43:26,800 --> 00:43:30,560
high phase noise
1448
00:43:28,559 --> 00:43:32,400
unit and it has its own imperfections
1449
00:43:30,559 --> 00:43:33,759
and it has poor harmonic performance
1450
00:43:32,400 --> 00:43:35,200
you will be able to capture some of
1451
00:43:33,760 --> 00:43:36,000
those imperfections directly with the
1452
00:43:35,199 --> 00:43:38,078
oscilloscope
1453
00:43:36,000 --> 00:43:40,000
and examine how the multiplier actually
1454
00:43:38,079 --> 00:43:42,240
behaves as the input signal to
1455
00:43:40,000 --> 00:43:43,599
it increases and how it's harmonic
1456
00:43:42,239 --> 00:43:45,039
changes so it's going to be an
1457
00:43:43,599 --> 00:43:45,519
interesting setup and an interesting
1458
00:43:45,039 --> 00:43:47,039
result
1459
00:43:45,519 --> 00:43:48,639
so let's go ahead and zoom to the screen
1460
00:43:47,039 --> 00:43:50,159
and see what we get
1461
00:43:48,639 --> 00:43:52,480
so let's take a look at the output
1462
00:43:50,159 --> 00:43:54,719
spectrum that this instrument produces
1463
00:43:52,480 --> 00:43:56,960
right now i have channel one set to full
1464
00:43:54,719 --> 00:44:00,399
band with 113 gigahertz
1465
00:43:56,960 --> 00:44:03,358
at the full sample rate 384 000
1466
00:44:00,400 --> 00:44:05,599
points and 150 nanoseconds per division
1467
00:44:03,358 --> 00:44:07,920
this allows me to compute the fft
1468
00:44:05,599 --> 00:44:11,680
and the fft will span anywhere from zero
1469
00:44:07,920 --> 00:44:13,519
hertz all the way to up to 128 gigahertz
1470
00:44:11,679 --> 00:44:14,879
at the resolution bandwidth of one
1471
00:44:13,519 --> 00:44:15,920
megahertz we can in fact
1472
00:44:14,880 --> 00:44:18,000
look at that just to make sure the
1473
00:44:15,920 --> 00:44:19,599
resolution movement is one megahertz
1474
00:44:18,000 --> 00:44:21,119
there it is one megahertz resolution
1475
00:44:19,599 --> 00:44:23,599
bandwidth and check it
1476
00:44:21,119 --> 00:44:25,119
check the noise floor here the displayed
1477
00:44:23,599 --> 00:44:28,960
average noise floor
1478
00:44:25,119 --> 00:44:31,680
for this setting is around minus 80 dbm
1479
00:44:28,960 --> 00:44:33,838
this is extraordinary for such a
1480
00:44:31,679 --> 00:44:34,078
broadband front end for an oscilloscope
1481
00:44:33,838 --> 00:44:35,519
and
1482
00:44:34,079 --> 00:44:37,280
there's nothing connected to the input
1483
00:44:35,519 --> 00:44:39,920
right now and take a look
1484
00:44:37,280 --> 00:44:41,599
there are almost no tones at the output
1485
00:44:39,920 --> 00:44:43,599
of this adc
1486
00:44:41,599 --> 00:44:45,680
meaning every single harmonic of every
1487
00:44:43,599 --> 00:44:46,400
single clock that's generated is so well
1488
00:44:45,679 --> 00:44:49,039
filtered
1489
00:44:46,400 --> 00:44:51,039
and so well accounted for here there is
1490
00:44:49,039 --> 00:44:52,480
only a little bit of a signal at 64
1491
00:44:51,039 --> 00:44:54,079
gigahertz in the middle of the spectrum
1492
00:44:52,480 --> 00:44:56,159
and that's from the front-end sampler
1493
00:44:54,079 --> 00:44:58,318
very difficult to get rid of that signal
1494
00:44:56,159 --> 00:45:01,519
but even that is sitting at around minus
1495
00:44:58,318 --> 00:45:03,039
76 or 75 dbm on average
1496
00:45:01,519 --> 00:45:04,480
and perhaps in the future version the
1497
00:45:03,039 --> 00:45:05,039
final version it might be even better
1498
00:45:04,480 --> 00:45:07,280
than this
1499
00:45:05,039 --> 00:45:09,358
so having said that this is not even at
1500
00:45:07,280 --> 00:45:11,200
the highest sensitivity because we are
1501
00:45:09,358 --> 00:45:13,279
at 50 millivolts per division
1502
00:45:11,199 --> 00:45:15,118
if i were to reduce this all the way
1503
00:45:13,280 --> 00:45:17,040
down to 2 millivolts per division
1504
00:45:15,119 --> 00:45:19,200
the displayed average noise floor at the
1505
00:45:17,039 --> 00:45:20,880
same settings is going to continue to
1506
00:45:19,199 --> 00:45:23,759
get better and better for instance
1507
00:45:20,880 --> 00:45:24,400
i can go down to 20 millivolt per
1508
00:45:23,760 --> 00:45:25,599
division
1509
00:45:24,400 --> 00:45:26,880
and you can see that the noise is
1510
00:45:25,599 --> 00:45:28,720
already much lower even though the
1511
00:45:26,880 --> 00:45:29,519
resolution manual is exactly the same as
1512
00:45:28,719 --> 00:45:31,519
it was before
1513
00:45:29,519 --> 00:45:33,759
so if you need to make a very sensitive
1514
00:45:31,519 --> 00:45:34,639
measurement you can get noise floor much
1515
00:45:33,760 --> 00:45:37,119
better than
1516
00:45:34,639 --> 00:45:38,480
86 dbm simply because you can go all the
1517
00:45:37,119 --> 00:45:40,720
way down to 2 millivolt per division
1518
00:45:38,480 --> 00:45:42,880
which is the limit of the hardware
1519
00:45:40,719 --> 00:45:45,199
front end sensitivity now of course it
1520
00:45:42,880 --> 00:45:46,880
means that this the logic signal you can
1521
00:45:45,199 --> 00:45:48,000
capture is going to shrink as well but
1522
00:45:46,880 --> 00:45:48,720
if you need to make sensitive
1523
00:45:48,000 --> 00:45:50,480
measurements
1524
00:45:48,719 --> 00:45:52,239
that's the way to do it so let's go back
1525
00:45:50,480 --> 00:45:53,199
to where we were
1526
00:45:52,239 --> 00:45:54,479
because we're going to do some
1527
00:45:53,199 --> 00:45:56,799
measurements here that's going to
1528
00:45:54,480 --> 00:45:58,719
require a larger vertical spacing there
1529
00:45:56,800 --> 00:46:00,560
so there it is back to where we were
1530
00:45:58,719 --> 00:46:02,318
now i want to investigate and see what
1531
00:46:00,559 --> 00:46:05,039
happens to this multiplier
1532
00:46:02,318 --> 00:46:06,400
as i increase input signal power to it
1533
00:46:05,039 --> 00:46:09,039
and normally multipliers
1534
00:46:06,400 --> 00:46:10,720
have a nominal input signal be under
1535
00:46:09,039 --> 00:46:12,639
which they don't operate very well but
1536
00:46:10,719 --> 00:46:14,480
i'm curious to see how it behaves as i
1537
00:46:12,639 --> 00:46:16,000
change the input signal so right now i
1538
00:46:14,480 --> 00:46:19,760
have it at 18 gigahertz
1539
00:46:16,000 --> 00:46:21,358
so 18 gigahertz times 6 is 108 gigahertz
1540
00:46:19,760 --> 00:46:22,560
that's what i expect the multiplier to
1541
00:46:21,358 --> 00:46:24,159
produce
1542
00:46:22,559 --> 00:46:25,838
first we're going to give it minus 10
1543
00:46:24,159 --> 00:46:27,279
dbm and see what happens
1544
00:46:25,838 --> 00:46:29,358
so let's go ahead and turn the power on
1545
00:46:27,280 --> 00:46:31,119
to the multiplier and the signal on
1546
00:46:29,358 --> 00:46:32,480
there it is now you can see as soon as i
1547
00:46:31,119 --> 00:46:34,800
turn the signal on there are some
1548
00:46:32,480 --> 00:46:36,159
additional tones appearing in the fft
1549
00:46:34,800 --> 00:46:38,400
now you don't see it at all in the
1550
00:46:36,159 --> 00:46:40,159
vertical here because it's so small
1551
00:46:38,400 --> 00:46:41,519
but we can see that the instrument is
1552
00:46:40,159 --> 00:46:45,039
capturing a signal
1553
00:46:41,519 --> 00:46:48,318
at 72 gigahertz at minus 58 dbm
1554
00:46:45,039 --> 00:46:49,679
but 72 gigahertz is 18 times 4 it's not
1555
00:46:48,318 --> 00:46:51,279
18 times 6
1556
00:46:49,679 --> 00:46:53,519
which means that this multiplier right
1557
00:46:51,280 --> 00:46:55,839
now has a larger times
1558
00:46:53,519 --> 00:46:56,559
4 output than it does at time six but
1559
00:46:55,838 --> 00:46:58,559
even the time
1560
00:46:56,559 --> 00:46:59,759
six can be seen right over here is a
1561
00:46:58,559 --> 00:47:02,318
tiny little peak
1562
00:46:59,760 --> 00:47:04,240
coming out at 108 gigahertz we're not
1563
00:47:02,318 --> 00:47:05,920
recording it because it's below
1564
00:47:04,239 --> 00:47:07,519
the peak level that i have defined so
1565
00:47:05,920 --> 00:47:09,119
it's not capturing that peak there
1566
00:47:07,519 --> 00:47:11,199
but let's go ahead and increase the
1567
00:47:09,119 --> 00:47:11,838
signal and see how these two harmonics
1568
00:47:11,199 --> 00:47:13,439
interact
1569
00:47:11,838 --> 00:47:14,880
and what happens with the input signal
1570
00:47:13,440 --> 00:47:15,920
up to the output signal as i increase
1571
00:47:14,880 --> 00:47:17,680
that so here we go
1572
00:47:15,920 --> 00:47:19,039
going higher and higher you can see that
1573
00:47:17,679 --> 00:47:21,838
they're both growing
1574
00:47:19,039 --> 00:47:23,440
and the times four is still growing
1575
00:47:21,838 --> 00:47:24,639
we're going to continue over here let's
1576
00:47:23,440 --> 00:47:26,480
stop here for a second
1577
00:47:24,639 --> 00:47:28,960
and right now the two of them are almost
1578
00:47:26,480 --> 00:47:32,880
the same amplitude the 72 gigahertz is
1579
00:47:28,960 --> 00:47:35,039
minus 38 the 108 gigahertz signal is -41
1580
00:47:32,880 --> 00:47:37,039
so this multiplier is still not working
1581
00:47:35,039 --> 00:47:38,719
very well but i expect that as i
1582
00:47:37,039 --> 00:47:40,239
increase the input signal eventually
1583
00:47:38,719 --> 00:47:42,078
we're going to reach a point
1584
00:47:40,239 --> 00:47:44,000
where the multipliers input devices and
1585
00:47:42,079 --> 00:47:44,559
the amplifiers which are tuned to
1586
00:47:44,000 --> 00:47:47,199
produce
1587
00:47:44,559 --> 00:47:48,640
time six are going to operate in large
1588
00:47:47,199 --> 00:47:49,279
signal and they're going to get rid of
1589
00:47:48,639 --> 00:47:51,039
the other
1590
00:47:49,280 --> 00:47:52,480
harmonic that's undesired so let's go
1591
00:47:51,039 --> 00:47:53,920
ahead and try that
1592
00:47:52,480 --> 00:47:56,000
we're going to keep increasing it and i
1593
00:47:53,920 --> 00:47:57,119
expect the times 4 to begin to shrink
1594
00:47:56,000 --> 00:47:59,119
there it is you can see
1595
00:47:57,119 --> 00:48:02,400
it is shrinking now and i'm going to go
1596
00:47:59,119 --> 00:48:04,318
all the way to an input signal of 0 dbm
1597
00:48:02,400 --> 00:48:06,079
let's go ahead and stop at 0 dbm there
1598
00:48:04,318 --> 00:48:08,880
it is and check it out
1599
00:48:06,079 --> 00:48:10,640
here's our fundamental output this is
1600
00:48:08,880 --> 00:48:13,838
sitting at 108 gigahertz
1601
00:48:10,639 --> 00:48:15,838
minus 12 dbm and the 72 gigahertz signal
1602
00:48:13,838 --> 00:48:18,400
is at about -30 dbm
1603
00:48:15,838 --> 00:48:19,519
so that gives us 18 db of rejection
1604
00:48:18,400 --> 00:48:21,760
between the fourth
1605
00:48:19,519 --> 00:48:24,000
and the sixth harmonic of the input
1606
00:48:21,760 --> 00:48:26,000
signal which isn't really good
1607
00:48:24,000 --> 00:48:28,318
but perhaps for this kind of multiplier
1608
00:48:26,000 --> 00:48:30,559
screening off this also tells us that
1609
00:48:28,318 --> 00:48:31,679
if your system is sensitive to a 72
1610
00:48:30,559 --> 00:48:34,160
gigahertz signal
1611
00:48:31,679 --> 00:48:35,519
even though this is a wr10 waveguide if
1612
00:48:34,159 --> 00:48:37,118
you're sensitive to this signal you're
1613
00:48:35,519 --> 00:48:37,759
going to have to put a waveguide filter
1614
00:48:37,119 --> 00:48:39,358
after this
1615
00:48:37,760 --> 00:48:41,200
otherwise this is going to continue to
1616
00:48:39,358 --> 00:48:42,719
be present in your system
1617
00:48:41,199 --> 00:48:44,318
and the fundamental signal sitting all
1618
00:48:42,719 --> 00:48:45,519
the way out there is of course now
1619
00:48:44,318 --> 00:48:47,039
significantly larger
1620
00:48:45,519 --> 00:48:48,318
and you can see a whole bunch of other
1621
00:48:47,039 --> 00:48:50,239
harmonics here these are all the
1622
00:48:48,318 --> 00:48:50,960
multiples of 18 as well as all the other
1623
00:48:50,239 --> 00:48:52,879
harmonics
1624
00:48:50,960 --> 00:48:54,079
that this old synthesizer is producing
1625
00:48:52,880 --> 00:48:55,838
they're all mixing
1626
00:48:54,079 --> 00:48:57,440
intermodulation products are showing up
1627
00:48:55,838 --> 00:49:00,239
they're just all over the place
1628
00:48:57,440 --> 00:49:02,079
now to do a meticulous test to figure
1629
00:49:00,239 --> 00:49:03,919
out which of these is coming from what
1630
00:49:02,079 --> 00:49:05,519
and if any of them are coming from the
1631
00:49:03,920 --> 00:49:07,039
a2d converters of the oscilloscope
1632
00:49:05,519 --> 00:49:09,199
you're going to have to do any
1633
00:49:07,039 --> 00:49:10,400
a much better source is required and i'm
1634
00:49:09,199 --> 00:49:12,000
going to save that for
1635
00:49:10,400 --> 00:49:13,920
when we get the final version of the
1636
00:49:12,000 --> 00:49:14,880
instrument but nonetheless we can
1637
00:49:13,920 --> 00:49:16,880
clearly see
1638
00:49:14,880 --> 00:49:18,000
that this instrument is going to have a
1639
00:49:16,880 --> 00:49:20,079
very strong
1640
00:49:18,000 --> 00:49:22,079
fourth harmonic coming out this also
1641
00:49:20,079 --> 00:49:23,680
means that we're going to be able to
1642
00:49:22,079 --> 00:49:25,280
capture this in the time domain
1643
00:49:23,679 --> 00:49:26,799
and if i capture it in the time domain
1644
00:49:25,280 --> 00:49:27,680
we are going to be able to see some
1645
00:49:26,800 --> 00:49:29,599
double edging
1646
00:49:27,679 --> 00:49:31,279
because of all this smart strong
1647
00:49:29,599 --> 00:49:33,119
harmonics present in the spectrum
1648
00:49:31,280 --> 00:49:34,720
we can actually test that let's go ahead
1649
00:49:33,119 --> 00:49:36,160
and turn off the ffd since we're done
1650
00:49:34,719 --> 00:49:38,078
with it
1651
00:49:36,159 --> 00:49:40,719
there it is and let's go ahead and try
1652
00:49:38,079 --> 00:49:42,880
and zoom in
1653
00:49:40,719 --> 00:49:44,959
a little bit more let's keep going until
1654
00:49:42,880 --> 00:49:47,599
we see the sinusoid
1655
00:49:44,960 --> 00:49:49,280
and here we go there is our sinusoid you
1656
00:49:47,599 --> 00:49:52,000
can see it very clearly this is a
1657
00:49:49,280 --> 00:49:53,760
108 gigahertz sinusoid and you can see
1658
00:49:52,000 --> 00:49:54,318
double edging and triple aging and some
1659
00:49:53,760 --> 00:49:56,480
of this
1660
00:49:54,318 --> 00:49:58,000
because it has strong harmonics and look
1661
00:49:56,480 --> 00:50:00,240
at how sharp the trigger
1662
00:49:58,000 --> 00:50:02,800
is and this is because the trigger has
1663
00:50:00,239 --> 00:50:03,919
full 113 gigahertz of bandwidth done in
1664
00:50:02,800 --> 00:50:06,000
the digital domain
1665
00:50:03,920 --> 00:50:07,760
in the digital signal processor and it's
1666
00:50:06,000 --> 00:50:09,119
doing a fantastic job and you can see
1667
00:50:07,760 --> 00:50:10,800
how the signals spread
1668
00:50:09,119 --> 00:50:12,240
as they move away from the trigger point
1669
00:50:10,800 --> 00:50:13,359
because of all the other harmonics that
1670
00:50:12,239 --> 00:50:14,799
are present
1671
00:50:13,358 --> 00:50:16,799
but the trigger is working perfectly
1672
00:50:14,800 --> 00:50:17,519
fine now the spectrum signals that we
1673
00:50:16,800 --> 00:50:19,519
measured
1674
00:50:17,519 --> 00:50:20,800
traditionally you can really only do
1675
00:50:19,519 --> 00:50:22,480
that with
1676
00:50:20,800 --> 00:50:23,760
a spectrum analyzer and the only
1677
00:50:22,480 --> 00:50:25,440
spectrum analyzer in the world that can
1678
00:50:23,760 --> 00:50:26,079
give you the spectrum of this from dc to
1679
00:50:25,440 --> 00:50:28,480
110
1680
00:50:26,079 --> 00:50:30,318
is the uxa which is also from keysight
1681
00:50:28,480 --> 00:50:32,240
so now you can do this on their
1682
00:50:30,318 --> 00:50:33,838
oscilloscopes as well which is just
1683
00:50:32,239 --> 00:50:34,558
madness to be able to do this kind of
1684
00:50:33,838 --> 00:50:36,400
measurements
1685
00:50:34,559 --> 00:50:37,680
up to these frequencies it's really
1686
00:50:36,400 --> 00:50:39,838
impressive
1687
00:50:37,679 --> 00:50:41,598
i also wanted to show you some primary
1688
00:50:39,838 --> 00:50:42,558
measurements that keysight has done on
1689
00:50:41,599 --> 00:50:44,079
this instrument
1690
00:50:42,559 --> 00:50:46,079
and keep in mind that these are done on
1691
00:50:44,079 --> 00:50:47,280
a prototype this instrument's not
1692
00:50:46,079 --> 00:50:48,960
completely finished yet so these
1693
00:50:47,280 --> 00:50:50,079
measurements are likely going to change
1694
00:50:48,960 --> 00:50:51,679
and probably improve
1695
00:50:50,079 --> 00:50:53,599
by the time the final instrument is
1696
00:50:51,679 --> 00:50:56,318
released now here the instrument is set
1697
00:50:53,599 --> 00:50:57,838
to a bandwidth of 70 gigahertz
1698
00:50:56,318 --> 00:51:00,000
and at this bandwidth they're trying to
1699
00:50:57,838 --> 00:51:01,679
measure the flatness of the front-end
1700
00:51:00,000 --> 00:51:03,199
response as well as the effective number
1701
00:51:01,679 --> 00:51:03,919
of bits and noises some of the other
1702
00:51:03,199 --> 00:51:05,598
measurements
1703
00:51:03,920 --> 00:51:07,519
now if you look here you can see how
1704
00:51:05,599 --> 00:51:08,160
flat the frequency response is i mean
1705
00:51:07,519 --> 00:51:11,599
this is
1706
00:51:08,159 --> 00:51:13,440
within plus some minus 0.75 db or so all
1707
00:51:11,599 --> 00:51:15,039
the way to 70 gigahertz
1708
00:51:13,440 --> 00:51:16,639
which is a 3db bandwidth of the
1709
00:51:15,039 --> 00:51:17,679
instrument and this is not even the
1710
00:51:16,639 --> 00:51:19,440
final version
1711
00:51:17,679 --> 00:51:21,118
it's very difficult to accomplish this
1712
00:51:19,440 --> 00:51:22,240
they also measure the enum but if you
1713
00:51:21,119 --> 00:51:24,720
look at the enop
1714
00:51:22,239 --> 00:51:26,239
from dc all the way up to 70 gigahertz
1715
00:51:24,719 --> 00:51:27,838
this dip you see here you have to ignore
1716
00:51:26,239 --> 00:51:30,078
because that's cut off fighter
1717
00:51:27,838 --> 00:51:31,599
by the dsp anyway so really the
1718
00:51:30,079 --> 00:51:32,559
measurement that's only valid all the
1719
00:51:31,599 --> 00:51:35,440
way from dc
1720
00:51:32,559 --> 00:51:37,599
to 70 gigahertz the enop is just below 6
1721
00:51:35,440 --> 00:51:38,960
db and it's completely flat regardless
1722
00:51:37,599 --> 00:51:40,880
of the input frequency
1723
00:51:38,960 --> 00:51:43,199
this means that it's entirely noise
1724
00:51:40,880 --> 00:51:44,960
limited and it's not distortion limited
1725
00:51:43,199 --> 00:51:46,558
so that if you were to continue to
1726
00:51:44,960 --> 00:51:48,159
increase to
1727
00:51:46,559 --> 00:51:49,839
reduce the bandwidth of the front end
1728
00:51:48,159 --> 00:51:50,799
the enub is going to continue to get
1729
00:51:49,838 --> 00:51:52,960
better and better
1730
00:51:50,800 --> 00:51:54,318
up until a certain point where
1731
00:51:52,960 --> 00:51:55,599
distortion might kick in
1732
00:51:54,318 --> 00:51:57,759
this is the kind of measurements that i
1733
00:51:55,599 --> 00:51:58,559
would want to do on the final version of
1734
00:51:57,760 --> 00:52:00,000
the instrument
1735
00:51:58,559 --> 00:52:01,519
but to be able to get this is really
1736
00:52:00,000 --> 00:52:02,880
quite impressive the noise we've already
1737
00:52:01,519 --> 00:52:05,199
measured which matches these
1738
00:52:02,880 --> 00:52:07,920
numbers now we can go all the way to 110
1739
00:52:05,199 --> 00:52:09,358
gigahertz so in this case 113 gigahertz
1740
00:52:07,920 --> 00:52:11,599
and again you can see the front-end
1741
00:52:09,358 --> 00:52:13,679
bandwidth here fairly flat and this
1742
00:52:11,599 --> 00:52:14,880
is going to further improve once they
1743
00:52:13,679 --> 00:52:17,279
complete the instrument
1744
00:52:14,880 --> 00:52:18,079
is really quite impressive same with the
1745
00:52:17,280 --> 00:52:21,280
enoch is
1746
00:52:18,079 --> 00:52:22,559
again flat all the way up to 113
1747
00:52:21,280 --> 00:52:24,640
gigahertz
1748
00:52:22,559 --> 00:52:26,400
similarly just around five and a half db
1749
00:52:24,639 --> 00:52:27,440
which i'm sure it's going to be improved
1750
00:52:26,400 --> 00:52:28,960
a little bit more too
1751
00:52:27,440 --> 00:52:30,720
i'm very curious what would be the e
1752
00:52:28,960 --> 00:52:32,559
knob let's see at one gigahertz
1753
00:52:30,719 --> 00:52:34,239
with a one gigahertz filtered bandwidth
1754
00:52:32,559 --> 00:52:34,559
and up to one gigahertz input frequency
1755
00:52:34,239 --> 00:52:36,399
but
1756
00:52:34,559 --> 00:52:38,960
i mean take a look at this is flat all
1757
00:52:36,400 --> 00:52:39,920
the way up to 112 gigahertz completely
1758
00:52:38,960 --> 00:52:42,400
noise limited
1759
00:52:39,920 --> 00:52:44,159
this is really good and it promises to
1760
00:52:42,400 --> 00:52:47,280
be able to get better and better enough
1761
00:52:44,159 --> 00:52:48,399
as the frequency changes and because of
1762
00:52:47,280 --> 00:52:50,240
this low noise
1763
00:52:48,400 --> 00:52:51,760
performance of the oscilloscope and
1764
00:52:50,239 --> 00:52:53,439
because of the good effective number of
1765
00:52:51,760 --> 00:52:55,440
bits we get we can do measurements
1766
00:52:53,440 --> 00:52:56,639
at a performance level never
1767
00:52:55,440 --> 00:53:00,240
accomplished before
1768
00:52:56,639 --> 00:53:02,719
so here's an example of a 64 gigabyte 64
1769
00:53:00,239 --> 00:53:04,239
coherent modulation that's being used in
1770
00:53:02,719 --> 00:53:07,199
the scope as a receiver here
1771
00:53:04,239 --> 00:53:10,078
and you're hitting an evm of 2.8 percent
1772
00:53:07,199 --> 00:53:12,159
for 64 gigabytes 64 qm that's more than
1773
00:53:10,079 --> 00:53:14,480
600 gigabit per second equivalent
1774
00:53:12,159 --> 00:53:17,358
rate for coherent communication now the
1775
00:53:14,480 --> 00:53:19,039
previous best measurement was at 5.4
1776
00:53:17,358 --> 00:53:20,558
percent evm and this improvement is
1777
00:53:19,039 --> 00:53:23,199
primarily due to the fact
1778
00:53:20,559 --> 00:53:25,119
that channel to channel a jitter is less
1779
00:53:23,199 --> 00:53:26,719
than 35 femtoseconds
1780
00:53:25,119 --> 00:53:28,400
and that means that the constellation is
1781
00:53:26,719 --> 00:53:29,679
preserved going through the oscilloscope
1782
00:53:28,400 --> 00:53:31,760
and ultimately digitized
1783
00:53:29,679 --> 00:53:33,199
and processed by the dsp is really quite
1784
00:53:31,760 --> 00:53:35,040
amazing furthermore
1785
00:53:33,199 --> 00:53:39,358
you can do things that could never been
1786
00:53:35,039 --> 00:53:41,679
done before this is a 64 gigabyte 256 qm
1787
00:53:39,358 --> 00:53:43,039
that's a terabit per second equivalent
1788
00:53:41,679 --> 00:53:46,239
coherent modulation
1789
00:53:43,039 --> 00:53:47,599
the vm is again 2.6 percent nobody has
1790
00:53:46,239 --> 00:53:50,078
ever done this measurement
1791
00:53:47,599 --> 00:53:51,519
on an oscilloscope before all right i
1792
00:53:50,079 --> 00:53:53,039
hope you enjoyed this video
1793
00:53:51,519 --> 00:53:54,480
there was a lot of information and i
1794
00:53:53,039 --> 00:53:56,239
definitely really enjoyed making it i
1795
00:53:54,480 --> 00:53:57,599
only had this scope for a couple of days
1796
00:53:56,239 --> 00:53:59,039
so i didn't have a lot of time to do
1797
00:53:57,599 --> 00:53:59,920
detailed experiments but then i can't
1798
00:53:59,039 --> 00:54:02,159
wait for the
1799
00:53:59,920 --> 00:54:04,559
final full version to be out with four
1800
00:54:02,159 --> 00:54:06,318
channels so we can do some crazy setup
1801
00:54:04,559 --> 00:54:07,839
and see what we can capture with it now
1802
00:54:06,318 --> 00:54:09,358
having said that i'd love to hear what
1803
00:54:07,838 --> 00:54:10,318
you guys think in the comments section
1804
00:54:09,358 --> 00:54:12,239
let me know what you thought of the
1805
00:54:10,318 --> 00:54:12,480
oscilloscope and we can discuss some of
1806
00:54:12,239 --> 00:54:13,838
the
1807
00:54:12,480 --> 00:54:15,358
details of its design and if you have
1808
00:54:13,838 --> 00:54:15,920
any questions i'll try to answer them
1809
00:54:15,358 --> 00:54:17,519
i'm sure
1810
00:54:15,920 --> 00:54:19,200
the keysight staff would also love to
1811
00:54:17,519 --> 00:54:20,880
answer your questions this is a
1812
00:54:19,199 --> 00:54:21,598
wonderful accomplishment really for
1813
00:54:20,880 --> 00:54:23,358
engineering
1814
00:54:21,599 --> 00:54:36,000
and i think we should all celebrate that
1815
00:54:23,358 --> 00:54:36,000
i'll see you in the comments section
129738
Can't find what you're looking for?
Get subtitles in any language from opensubtitles.com, and translate them here.