10 GHz USB Oscilloscope by Darwin Sabanovic

[fcb]

I was one of the few backers of Darwin's project on Kickstarter.

It would have been great to give it bench space here, and it would have been useful for something I've been working on.

[Lukas]

[...] ideas about the triggering.

What about random sampling? Continoulsy run the sampling clock and determine the time of the trigger event using a TDC.

Heh, you beat me to it. Either use a TDC from Acam or roll your own with an fpga.

@Lukas: what would you use for the dithered clock? A regular clock that you "dirtied up" with added noise to get your dithering? Or a noise generator + gated clock? I have used the latter for testing my fpga based TDC, but never played with the former.

Then again, if the resolution of those delay lines is good AND they can be adjusted fast enough then that could be better. It would really depend on the linearity of the delay line and the settling time (and resolution ).

Concerning the TDC: Acam and FPGAs aren't your only options. This guy has a really nice overview of available TDC technologies: http://www.ko4bb.com/~bruce/TDC.html Also see HP Journals from the 80s ... 90s. They described the TDCs they used in their random sampling digital oscilloscopes. They used multislope TDCs and achieved 1ps resolution IIRC. In some article, they also described how they dealt with clock dithering. My first try would be to alter the phase between acquisitions by using a slightly tunable VCXO.

In my sampling oscilloscope i'm using the 100EP196 LVPECL delay line. It allows digital coarse delay setting in steps of 10ps using a digital input and fine tuning by an analog control voltage. It drifts a lot over time/temperature, but the nonlinearity seems to stay somehow stable. The delay line can be configured as a ring oscillator to calibrate it. Once you've measured the nonlinearity it suffices to obtain the delays at zero and full delay. According to the datasheet, jitter is ~3ps.

[mrflibble]

Concerning the TDC: Acam and FPGAs aren't your only options. This guy has a really nice overview of available TDC technologies: http://www.ko4bb.com/~bruce/TDC.html Also see HP Journals from the 80s ... 90s. They described the TDCs they used in their random sampling digital oscilloscopes. They used multislope TDCs and achieved 1ps resolution IIRC. In some article, they also described how they dealt with clock dithering. My first try would be to alter the phase between acquisitions by using a slightly tunable VCXO.

Yeah, I know that list from Bruce. It's been a while since I went over the various options, but I seem to recall that the dual slope approach would mean a relatively low repetition rate. Maybe I misunderstood it at the time, but I didn't see how to reasonably get a 200 MHz repetition rate. And multislope was one bridge too far for me at that time.

Do you happen to know the HP Journal issue that deals with clock dithering?

In my sampling oscilloscope i'm using the 100EP196 LVPECL delay line. It allows digital coarse delay setting in steps of 10ps using a digital input and fine tuning by an analog control voltage. It drifts a lot over time/temperature, but the nonlinearity seems to stay somehow stable. The delay line can be configured as a ring oscillator to calibrate it. Once you've measured the nonlinearity it suffices to obtain the delays at zero and full delay. According to the datasheet, jitter is ~3ps.

Delay line configured as ring osc? As in have the CARRY4 chain (or whatever you use as delay element)  inline with the ring oscillator? So CARRY4 chain + bunch of inverters followed by a counter. And then run that for a fixed amount of time, as determined by at counter clocked with your regular sampling clock? I suppose that would work too. Another approach is a seperate ring osc. And yet another is to have an external noise source and then run a full calibration (count the bins). The last approach with external noise source is what I currently do. There might be some merit in having the delay line as part of a ring osc too though, mmmh.

[free_electron]

They are useless for day to day work , debugging work , signal analysis (analog or digital).

Absolute nonsense.  Anyone working on picosecond electronics will need a sampling scope to verify their designs.  Just because it's a design area you have no knowledge of doesn't mean it doesn't exist.

Like i said, except for some extreme things these machines are useless.
How many on the forum here do that kind of picocesond stuff ?

People in the 'industry' that actually do that stuff have access to proper machines. they wouldn;t be caught dead using this contraption.

I'm not downplaying the art involved in creating this machine. i couldn't do it so kudos to the project starter.
The danger is a bunch of hobbyist will flock to this thing "10Ghz for a nickel and a dime" and be VERY disappointed.

[branadic]

Concerning the TDC: Acam and FPGAs aren't your only options.

Sure, but they are for sure the cheapest on, for example GP22 <10,-€/pc.
TI also offers a TDC:

http://www.ti.com/product/ths788

but for a barefaced price of 121,78 € (Farnell).

[mrflibble]

Do they also sell the GP22 for < EUR 10 at single quantities to non-commercial entities aka private persons? Or lets say 3. Might be interesting for a little experiment.

[branadic]

The distributor for the integrated circuits made by ACAM is IS-Line GmbH. I bet they also sell to privat persons. I have an offer for 5pc. at a price of 13,02 € plus (Angelo Merte) taxes, which is still affordable for such a beast.

[pinkysbrein]

Is it really so difficult to make an integration + ADC TDC? You can just calibrate in between every measurement to get rid of temperature dependence.

[nctnico]

Another option is to use a programmable delay line. If you have 10ps steps then you could get an equivalent sampling rate of 100GHz. The TDS5xx/6xx/7xx can reach equivalent sampling times of 50GHz. There is a 'TDS520B component service manual' with all the diagrams. Maybe it tells or shows a bit of how Tek does that.

[mrflibble]

Another option is to use a programmable delay line.

Of course. I think the approach of a random trigger & use TDC to measure the delay of that trigger was just mentioned as another alternative.

There is a 'TDS520B component service manual' with all the diagrams. Maybe it tells or shows a bit of how Tek does that.

See for example: http://bee.mif.pg.gda.pl/ciasteczkowypotwor/Tek/ds520b-cm.pdf

[branadic]

Does anyone now how Darwin Sabanovic will proceed? I tryed to get in contact with him, but as before without success.
That reminds me of what someone else stated somewhere before, he's a poor sales and business man.

[David Hess]

And what is the bloody point of a sampling oscilloscope ? They are only useful on repetitive signals and you can look at eye diagrams ...

They do not suffer from overload recovery times and are uniquely useful for calibrating the transient response on reference flat pulse generators and therefor other instruments like real time oscilloscopes.  Their predictable frequency response allows them to be used as an alternative to thermal based instruments for accurate RF power measurement.

They are useless for day to day work , debugging work , signal analysis (analog or digital).

They certainly have their limitations and are not general purpose instruments but the same could be said for oscilloscopes faster than about 300 MHz where 50 ohm probe inputs are used.

None of the scope makers build em anymore.

They do still make them but only at the highest end where the highest bandwidth is needed although there are several USB based lower end commercial instruments available with Picoscope being the primary example.  Compare how much this project would have cost (since it failed to meet its funding requirements) for an 8 GHz sampling oscilloscope versus an 8 GHz digital storage oscilloscope.  Assuming you can live with its limitations, it will be much more economical for a given bandwidth.

[fcb]

Does anyone now how Darwin Sabanovic will proceed? I tryed to get in contact with him, but as before without success.
That reminds me of what someone else stated somewhere before, he's a poor sales and business man.

I'm not sure you can draw the conclusion that Darwin is a poor sales/business man, in-fact you could argue that walking away from this project makes business sense - it clearly failed to come close to the MVP threshold on KS.

Still, it's a pity. As a backer of the KS, I would have liked to own such a unit...

[eurofox]

I just bought the 11GHz version , delivery in 2-3 weeks 

I just build a ultra fast pulse generator, should be 50ps rising time (10-90%) but cannot measure it with my 1Ghz Agilent MSO-X scope (Rising time 450ps)

I will post the results of test the BS800E.

eurofox

[David Hess]

Interesting indeed. Wondering pretty much the same thing as Lukas. Which diodes did you use, and how did you hook things up?

So far mostly only simulation with BAS70 dual diodes. The trick is to feed a pulse into a shorted piece of stripline which makes the diodes conduct for a short period charging or discharging 2 capacitors (the voltage across is the sampled valued minus the previous sample).  The rise time of the pulse is driving the sampling window. I did some tests with a 2N2904 transistor in an avelanche pulse generator. The only time I needed 10V/div in 50 Ohm mode  on my scope

Trapped charge sampling bridges use the transition time of the trailing edge of the strobe to generate the sampling gate width.  Other sampling bridges use the width of the strobe which makes it more difficult to achieve very narrow sampling gate widths.

Trapped charge sampling bridges sometimes have problems because they generate a kick-out pulse much earlier than the sampling point so if your avalanche or tunnel diode pulse generator is too close to the sampling head, the kick-out may trigger it before the normal trigger occurs.  Anecdotally they also seem to have more blow-by problems but that may just be an implementation issue.

I am happy that suitable diodes for building sampling bridges are available from Avago (old HP) at inexpensive prices.  I wish there was a similar source for step recovery diodes.

[Lukas]

I just build a ultra fast pulse generator, should be 50ps rising time (10-90%)

Interesting, can you tell us more about it? Did you use a fast logic gate (OnSemi NBSG series) oder something more advanced?

[eurofox]

I just build a ultra fast pulse generator, should be 50ps rising time (10-90%)

Interesting, can you tell us more about it? Did you use a fast logic gate (OnSemi NBSG series) oder something more advanced?

It is based on a ultra fast comparator (SiGe) with basic rise time of 35ps (20-80%), output 200mV (50 ohm).
Very small board and housing including power supply from TRACO POWER (very small).
Triggered from any generator able to deliver square or pulse from ?? tested to 50Mhz with 0,4-2,5VPP (50 ohm)
Testing show it is very stable, no noise problems.
The output connections are very short to the SMA connector.

I need a "fast" oscilloscope for further testing and fine tuning. It is a first version prototype.

eurofox

[David Hess]

I just build a ultra fast pulse generator, should be 50ps rising time (10-90%)

Interesting, can you tell us more about it? Did you use a fast logic gate (OnSemi NBSG series) oder something more advanced?

It is based on a ultra fast comparator (SiGe) with basic rise time of 35ps (20-80%), output 200mV (50 ohm).
Very small board and housing including power supply from TRACO POWER (very small).
Triggered from any generator able to deliver square or pulse from ?? tested to 50Mhz with 0,4-2,5VPP (50 ohm)
Testing show it is very stable, no noise problems.
The output connections are very short to the SMA connector.

I need a "fast" oscilloscope for further testing and fine tuning. It is a first version prototype.

eurofox

Would that comparator be the Analog Devices ADCMP580 or one of its cousins?

The other integrated solution I have seen recommended for this sort of thing is the Micrel SY88922 laser driver which is slower but generates a larger output voltage swing.

[TerraHertz]

And what is the bloody point of a sampling oscilloscope ? They are only useful on repetitive signals and you can look at eye diagrams ...

Because most physical processes, physics, even electronic circuit behavior, are easily made to run repetitively. So for that old, cheap sampling scopes are fine. And thanks very much to everyone who avoids them, thus making them cheaper for the rest of us.

Chasing obscure rare data-stream related errors in communications protocols gives me the shudders. I'm very happy to leave that shit to others. And watch them pay King's Ransom amounts for real-time GHz scopes.

[madshaman]

For a sampling scope, I'm happy with my "ancient" HP 54120B sampling scope with 20Ghz test set, it's got nice 3.5mm inputs (which would add a lot to the cost of ANY scope input, and one also needs to take into account the need for sacrificial connectors, which also won't be cheap) and if I ever needed it, I could throw money away to get a 50Ghz test set.

That being said, old equipment is big, bulky, and not everyone appreciates its aesthetic beauty.  There's also only so much of it.

I can see many reasons why someone would want to own a new sampling scope, as long as it's useful to them.

[eurofox]

I just build a ultra fast pulse generator, should be 50ps rising time (10-90%)

Interesting, can you tell us more about it? Did you use a fast logic gate (OnSemi NBSG series) oder something more advanced?

It is based on a ultra fast comparator (SiGe) with basic rise time of 35ps (20-80%), output 200mV (50 ohm).
Very small board and housing including power supply from TRACO POWER (very small).
Triggered from any generator able to deliver square or pulse from ?? tested to 50Mhz with 0,4-2,5VPP (50 ohm)
Testing show it is very stable, no noise problems.
The output connections are very short to the SMA connector.

I need a "fast" oscilloscope for further testing and fine tuning. It is a first version prototype.

eurofox

Would that comparator be the Analog Devices ADCMP580 or one of its cousins?

The other integrated solution I have seen recommended for this sort of thing is the Micrel SY88922 laser driver which is slower but generates a larger output voltage swing.

Yep 

1/ In this ultra fast rising edge not many components are available
2/ It is cheap and easy aivalable
3/ Soldering this 16-LFCSP with a solder iron is  but I manage to do it, I try with hot air  , I have an oven but for one prototype with a few components 

eurofox

[eurofox]

And what is the bloody point of a sampling oscilloscope ? They are only useful on repetitive signals and you can look at eye diagrams ...

Because most physical processes, physics, even electronic circuit behavior, are easily made to run repetitively. So for that old, cheap sampling scopes are fine. And thanks very much to everyone who avoids them, thus making them cheaper for the rest of us.

Chasing obscure rare data-stream related errors in communications protocols gives me the shudders. I'm very happy to leave that shit to others. And watch them pay King's Ransom amounts for real-time GHz scopes.

Fully agree, apart from some specific datacom related problems in this very high speed most signals are repetitive and can be measured with a sampling scope.

With respect to remarks about probes, I think that on picosecond level probes are not really useful.
I solder an SMA connector on the place where I want to measure with very short wires, the enemy of picosecond pulses is capacity from PCB, wiring, cables.

eurofox

[David Hess]

Would that comparator be the Analog Devices ADCMP580 or one of its cousins?

The other integrated solution I have seen recommended for this sort of thing is the Micrel SY88922 laser driver which is slower but generates a larger output voltage swing.

Yep 

1/ In this ultra fast rising edge not many components are available
2/ It is cheap and easy available
3/ Soldering this 16-LFCSP with a solder iron is  but I manage to do it, I try with hot air  , I have an oven but for one prototype with a few components 

What I would like to try is using a differential output comparator or logic gate to emitter switch an RF transistor so the output pulse can be at a much higher level.  ECL does not have quite the output drive level to do this easily but a slower TTL/CMOS output comparator like an ADCMP603 could do it and the gain of the output transistor would restore some of the transition time performance.  That essentially duplicates how the Micrel SY88922 output stage works.

[Marco]

Since there is no current amplification in a common base amplifier I don't see how it could sharpen the edges much.

[eurofox]

Would that comparator be the Analog Devices ADCMP580 or one of its cousins?

The other integrated solution I have seen recommended for this sort of thing is the Micrel SY88922 laser driver which is slower but generates a larger output voltage swing.

Yep 

1/ In this ultra fast rising edge not many components are available
2/ It is cheap and easy available
3/ Soldering this 16-LFCSP with a solder iron is  but I manage to do it, I try with hot air  , I have an oven but for one prototype with a few components 

What I would like to try is using a differential output comparator or logic gate to emitter switch an RF transistor so the output pulse can be at a much higher level.  ECL does not have quite the output drive level to do this easily but a slower TTL/CMOS output comparator like an ADCMP603 could do it and the gain of the output transistor would restore some of the transition time performance.  That essentially duplicates how the Micrel SY88922 output stage works.

No problem with RF transistor to reach 5 or 10V but I think you will be in the nanosecond range and not on the picosecond range.