Agilent has oscilloscopes to 63 GHz /160GS/s. How is that possible?

What type of technology is used to make such a ($439,824) product? How do you sample at that rate? 160GS/s implies some sort of clock at 160Ghz. How is that done? I don't understand how this can be done at any price the speeds are so otherworldly.

Based on EEVblog tear-downs I've watched, I think the answer is "dark sorcery and witchcraft" :smiley:

Well if it can be done, Agilent is a company that could do it. As a spin-off/de-vesture of the old HP instrumentation company they were always among or leading in the state of art of measurement teck.

Lefty

I had an 8 GHz scope a while back, and it is done with 8 bit flash convertors and you use multiple ones and interleave the data. I am sure they have a white paper. For example, if you have 16 convertors, they each need to sample at 10 GS/sec.

KeithRB:
I had an 8 GHz scope a while back, and it is done with 8 bit flash convertors and you use multiple ones and interleave the data. I am sure they have a white paper. For example, if you have 16 convertors, they each need to sample at 10 GS/sec.

Out of curiosity I have looked at TI/National, Analog, and LT's sites and I think the fastest converter that is available is TI at about 5 Gsps and that is at 7-bit. The datasheet shows it's actually 4 ADCs embedded in one chip with some clock management. So maybe you use 32 of those? But how the heck do you interleave them so that they are sampling at the correct point in time?

Note that you only get 63 GHz if you are using two channels, you are limited to "only" 33 GHz with all 4, so they are clearly interleaving multiple samplers.

Need to use something faster than an Arduino 8)

CrossRoads:
Need to use something faster than an Arduino 8)

Nothing a little liquid nitrogen and a faster crystal can't fix. 8)

I realize I should have posted these links:

First thing I buy if I win the Powerball:

http://www.home.agilent.com/en/pd-2108888-pn-DSAX96204Q/infiniium-high-performance-oscilloscope-63-ghz?nid=-33202.1010838&cc=US&lc=eng

TI's obscenely fast converter:

http://www.ti.com/product/lm97600

Even if you can sample that fast, how do you put test leads on a circuit that is moving at a rate that necessitates that kind of speed? Seems like the capacitance of any test lead would wreck it. Crazy stuff.

SirNickity:
Even if you can sample that fast, how do you put test leads on a circuit that is moving at a rate that necessitates that kind of speed? Seems like the capacitance of any test lead would wreck it. Crazy stuff.

I think the answer is "active probes" (MSRP about $8,000 each for this speed, but you get a set with the scope), whatever that is. Because I can repeat what Agilent says doesn't mean I understand it.

Or you put 50 ohm sample ports.

JoeN:

KeithRB:
I had an 8 GHz scope a while back, and it is done with 8 bit flash convertors and you use multiple ones and interleave the data. I am sure they have a white paper. For example, if you have 16 convertors, they each need to sample at 10 GS/sec.

Out of curiosity I have looked at TI/National, Analog, and LT's sites and I think the fastest converter that is available is TI at about 5 Gsps and that is at 7-bit. The datasheet shows it's actually 4 ADCs embedded in one chip with some clock management. So maybe you use 32 of those? But how the heck do you interleave them so that they are sampling at the correct point in time?

HP/Agilent were never afraid to design their own hybrid chips so there is no telling how they are doing it. They used to publish a nice Journal that had articles on their technology but I can't find that Agilent do that.

There is a white paper on this scope which describes a 10 ns clock with really clean edges - sub-sub 1 ns risetime.

Need to use something faster than an Arduino

I'm not sure if this is what HP uses, but besides using multiple interleaved ADCs, there
is a technique called "equivalent time sampling" that can be used if you have a stable
trigger capability and a repetitive signal.

Basically, what you do is trigger at the same point in the repetitive waveform multiple
times, up to 20 or so, and with each successive trigger you introduce an increasing
time-delay on when the ADC sampling is started. Then, you reconstruct the original
repetitive waveform by re-ordering the samples from memory to the screen. It's not
difficult.

Some years ago, I did this using a 20-Mhz PIC, which has a max sampling rate of about
50-Ksps, which is normally good for sampling about a 5 Khz sinewave, if you want to
display it [for good display, the signal must be much slower than the Nyquist frequency,
eg about 10 samples per period].

With ETS on the PIC, I was able to get an equivalent sampling rate of 1-Msps. With this
I was able to resolve a 50-Khz sinewave with very good fidelity. Even better, when using
an FFT spectral display on the samples, I was able to follow an aliased sinewave all the way
to about 250-Khz, as I recall. It really is magic, :-).

You can make a 100GSample/sec oscilloscope with arduino as well:

  1. sync the signal at point A
  2. take a S/H sample at t=A+10ps, ADC and store data (d1)
  3. sync the signal at point A
  4. take a S/H sample at t=A+11ps, ADC and store data (d2)
  5. sync the signal at point A
  6. take a S/H sample at t=A+12ps, ADC and store data (d3)
    etc..
    after ie. 256 such loops you'll get d1..d256 from t=10ps to t=266ps with 1ps resolution..
    :slight_smile:
    PS: as described above by oric-dan in detail :slight_smile:

Agilent separates the "real-time" specs from the repetitive sampling specs.

According to the manual they use custom indium phosphide ICs on the input - seems to be the
fastest semiconductor substrate material available (although it seems the actual devices are
built out of InGaAs built onto an InP surface).

InP technology is touted as being able to reach THz speeds BTW. Its also the most brittle
semiconductor - don't drop that 'scope!

SirNickity:
Even if you can sample that fast, how do you put test leads on a circuit that is moving at a rate that necessitates that kind of speed? Seems like the capacitance of any test lead would wreck it. Crazy stuff.

At microwave frequencies all signals are carried by stripline or microstrip or coax or waveguide or similar,
no issue with capacitance when a correctly terminated transmission line is used. Yes it is tricky to put test
leads on a circuit, active probes have to have very small dimensions I think, and you'd normally design in test
ports I suspect! The manual mentions the probes can be upgraded for higher bandwidth in the future... Shudder
to think how expensive the probes are actually.

I would love to know how many of these $500K scopes they sell a year and to what companies/individuals. Note normal 10 week delivery estimate, I suspect they build to order rather then ship from inventory.

Lefty

LeCroy is better (?):

SiGe technology
:slight_smile: