EEVblog #306 - Jim Williams Pulse Generator

[EEVblog]

Dave.

[PeteInTexas]

I'm curious about hooking it up to your "upgraded" Rigol.

[envisionelec]

Interesting. I have never seen that App Note. I once generated pulses using the propagation delay of some CMOS gates. I was getting nanosecond pulses with picosecond rise times, shorter than what this generates...and driving an LED circuit for ultra high speed photography where a laser could not be used. The PCB was otherworldly. It was handmade from copper tape and Kapton film and was packaged into a tiny Pomona EMI sealed case.

Applied for the patent at the urging of my co-workers, but someone beat me to the punch just three months prior. 

[EEVblog]

Interesting. I have never seen that App Note. I once generated pulses using the propagation delay of some CMOS gates. I was getting nanosecond pulses with picosecond rise times, shorter than what this generates...

Circuit?

Dave.

[free_electron]

hmm. if i have a spare moment i'll breadboard this thingie... i have access to a 16GHz bandwidth scope... let's see what it really does... should be interesting to see.
-edit- i looked at the pcb layout a bit. he's got way too much stray capacitance around the emitter of the transistor. the copper should be peeld back. this thing can go faster ..

[jahonen]

I'm not sure if the pulse gives correct rise time figures to calculate the bandwidth. I (and Leo Bodnar) did some experiments in the past, with my 300 MHz and 6 GHz scopes. I got 820 ps rise time for my Agilent MSO6034A:

https://www.eevblog.com/forum/projects-designs-and-technical-stuff/scope-rising-time/msg10635/#msg10635 (see next posts also)

That would translate to 488 MHz bandwidth (using factor of 0.4) which sounds a bit unrealistic, but using a coax for energy storage to get a true step, gives 1.16 ns which leads to much more realistic value of 345 MHz for a 300MHz scope banner specification.

For a reliable step and rise time measurements, one must replace the capacitor with a piece of coax (preferably with a low-loss rigid type).

Regards,
Janne

[EEVblog]

-edit- i looked at the pcb layout a bit. he's got way too much stray capacitance around the emitter of the transistor. the copper should be peeld back. this thing can go faster

I agree. Although I doubt you'll get much faster than what Jim Williams did. This circuit has been much built over the years and I don't recall anyone ever getting anything substantially better than Jim did.
I will be able to measure my actual board soon with a better scope. I don't expect a huge increase in performance by stripping back the ground plane in this instance. But yes, further improvement can be eeked out I'm sure.

Dave.

[mikeselectricstuff]

Don't forget to ask Agilent if you can try it with one of their higher-end scopes while you're there , assuming they bother keeping any in Oz...!
 

[peter.mitchell]

When you "upgrade" a scope, not just via firmware, do they put in some daughter boards or do they do a mac style upgrade?

video related:

[EEVblog]

Don't forget to ask Agilent if you can try it with one of their higher-end scopes while you're there , assuming they bother keeping any in Oz...!

They do, and one is coming to my lab shortly...

Dave.

[jahonen]

I'm not sure if the pulse gives correct rise time figures to calculate the bandwidth. I (and Leo Bodnar) did some experiments in the past, with my 300 MHz and 6 GHz scopes. I got 820 ps rise time for my Agilent MSO6034A:

https://www.eevblog.com/forum/projects-designs-and-technical-stuff/scope-rising-time/msg10635/#msg10635 (see next posts also)

That would translate to 488 MHz bandwidth (using factor of 0.4) which sounds a bit unrealistic, but using a coax for energy storage to get a true step, gives 1.16 ns which leads to much more realistic value of 345 MHz for a 300MHz scope banner specification.

For a reliable step and rise time measurements, one must replace the capacitor with a piece of coax (preferably with a low-loss rigid type).

Regards,
Janne

Just checked the bandwidth using a spectrum analyzer tracking generator, and -3 dB seems to be about at 380 MHz. So 820 ps is definitely too optimistic figure.

Regards,
Janne

[tnt]

Don't forget to ask Agilent if you can try it with one of their higher-end scopes while you're there , assuming they bother keeping any in Oz...!

They do, and one is coming to my lab shortly...

Dave.

Great, I must confess I was kind of hoping for this :p

As mentioned by others the pulse is so short that the scope doesn't display it fully (falls before it reached peak). But if you know the "real" pulse shape, and the "measured" pulse shape on the DUT, I'm pretty sure it would be possible to correctly find the bandwidth. The PCB has holes to solder a coax BTW. I tried with 30 m of LMR240 and it worked fine even though it does impair the practicality to have 30 m of coax attached :p


To comment on the other posts:

The output trace is meant to have a 50 ohm impedance, and when using it with a bnc cable I would expect the capacitance of the cable dominates anyway. The layout is certainly not perfect, but it's the first revision of the PCB and it turned out "good enough" for my use so I didn't bother making a second revision.

Btw, the 90V supply is made easily available on a test pad on the PCB so that if you want to do a "jim williams" type of construction on the connector itself, you can reuse the switcher part and only have to worry about the 4-5 components to solder on the BNC.

Cheers,

     Sylvain

[jahonen]

Given potential of this circuit, I think it is not probably worth the time to start optimizing the layout much further, although it would be certainly possible to use a 3D EM simulator to optimize the cutouts beneath of each component pad and via to get 50 ohm impedance everywhere in the signal path Also, too wide microstrips start to behave in a non-TEM way (which causes a dispersion) when frequency rises enough but that doesn't probably happen until above 10 GHz or so.

Regards,
Janne

[BravoV]

Just fyi, saved & bookmarked this info while a go about other transistor that is faster than 2N2369, from this interesting post -> http://www.electronicspoint.com/avalanche-transistors-t212987.html

Edit : Datasheet of BFG541 NPN 9 GHz Wideband transistor here -> http://www.nxp.com/documents/data_sheet/BFG541_CNV.pdf

The subject of avalanche-mode pulse generators comes up here now and again. A colleague pointed out to me that a Philips BFG541 transistor will avalanche nicely at around 50V or so.

Confirmed. It's quite fast too: I measured a 150ps risetime, using a Tek S-6 sampler plugin in a 7000 series mainframe. It's a little over twice as fast as the 2n2369 I usually select for this purpose.

 I thought some people around here might like to know...

 Anyone here know any other transistors that will also avalanche fast? (I know of Zetex. Not so great.)

 Jeroen Belleman

[Omicron]

This circuit pops up now and then. But the thing is: you can't use it for bandwidth measurements! The formula used is only valid if the pulse is wide enough for the oscilloscope to reach the actual signal amplitude. This has been discussed before on the forum I think. This particular pulse is too fast for the scope to show the correct amplitude. The "rise time" you measure this way is meaningless, the scope never reaches the 90% point. Rise time is not 10% to 90% of what you see on the screen, it's 10% to 90% of what actual signal amplitude is there at the input!

I believe that in later application notes Jim provides circuits that generate much longer pulses.

[tnt]

As Janne and I mentionned above, in a latter app note you have the same circuit but using a coax hardline as energy storage and this creates more of a 'step' than a pulse. Just need to solder a coax at the right place, but it makes it a bit impractical.

[Omicron]

Indeed, I recall the use of the hard line.

Also, you have to remember that Agilent and Tek always specify bandwidth of the entire system, i.e. including the probes they provide with the scope. So if you measure the bandwidth of the scope itself on the 50 Ohms input you should measure a bandwidth that is slightly larger. 345MHz seems perfectly on spec for a 300MHz scope.

[EEVblog]

As mentioned by others the pulse is so short that the scope doesn't display it fully (falls before it reached peak). But if you know the "real" pulse shape, and the "measured" pulse shape on the DUT, I'm pretty sure it would be possible to correctly find the bandwidth.

What needs to be remembered also, is that you can't overestimate your bandwidth with this. So at worst, whatever the circumstances, your scope will have a better bandwidth than what you measure 

Dave.

[nitro2k01]

Couldn't you do a quick bodge and put a small piece of wire between the female connectors? (Bare wire, so short that you can make the shields touch.)

[dda]

As mentioned by others the pulse is so short that the scope doesn't display it fully (falls before it reached peak). But if you know the "real" pulse shape, and the "measured" pulse shape on the DUT, I'm pretty sure it would be possible to correctly find the bandwidth.

What needs to be remembered also, is that you can't overestimate your bandwidth with this. So at worst, whatever the circumstances, your scope will have a better bandwidth than what you measure 

Dave.

If the peak begins to fall before the scope can detect it, isnt the 'true' rise time underestimated, and thus dividing 0.35 by a smaller number means BW is larger and so is overestimated?

[Omicron]

What needs to be remembered also, is that you can't overestimate your bandwidth with this. So at worst, whatever the circumstances, your scope will have a better bandwidth than what you measure 

Dave.

Actually that is not true. If you measure your bandwidth using the rise time obtained from the narrow pulse your estimate is always going to be way too high. For example see the number Jahonen obtained for the 300MHz scope. The reason is that what you read as the "rise time" is too low because in reality the curve should have been allowed to continue to the actual amplitude of the input signal. You never reach this real 90% point and what you think is the 90% point from what you see on the screen is happening way too early. Or in other terms: the higher the amplitude of the input pulse, the higher your bandwidth reading is going to be.

[jahonen]

What needs to be remembered also, is that you can't overestimate your bandwidth with this. So at worst, whatever the circumstances, your scope will have a better bandwidth than what you measure 

Dave.

Actually that is not true. If you measure your bandwidth using the rise time obtained from the narrow pulse your estimate is always going to be way too high. For example see the number Jahonen obtained for the 300MHz scope. The reason is that what you read as the "rise time" is too low because in reality the curve should have been allowed to continue to the actual amplitude of the input signal. You never reach this real 90% point and what you think is the 90% point from what you see on the screen is happening way too early. Or in other terms: the higher the amplitude of the input pulse, the higher your bandwidth reading is going to be.

I have also noticed that if one performs an integration of the pulse with a scope integration math function and then measures the rise time of this integrated pulse (a step), the resulting figure is much closer to the real value (this might be indeed pessimistic). But I guess that this requires that the pulse is much faster than the scope.

Out of interest, I measured the spectrum of the pulser output signal. It seems that it could be flatter, but that is probably ok'ish up to something like 500 MHz or so, depending on the criteria. Certainly adequate for 100 MHz or so. First roll-off seems to begin quite early.



Regards,
Janne

[free_electron]

The output trace is meant to have a 50 ohm impedance, and when using it with a bnc cable I would expect the capacitance of the cable .

I am going to build it dead bug style.
The problems with the are the following :
 - you have ground on the top layer around the emitter. That should not be there. The 50 ohms is made between top and bottom layer, not on the layer itself. Any ground on top layer froms stray capacitance
- the pins of the transistor sit in pads that go through the board. This gives again stray capacitance , inductance and it forms also an unterminated antenna.

I will flip the transistor upside down , solder the case to a chunk of copper , solder an end-launch sma close to it , put two 100 ohm smd resistors in 'tombstone' and bend the emitter wire so it laus on top of the resistors and touches the center pin of the sma. That way the whole emitter node is floating. Nothing is sticking out.

For a second revision of the board you can do that easily.
Drill a round hole in the pcb so you can drop the body of the transistor in th hole with the legs sticking up. And provide only smd pads to solder the transistor wires.

I'll whip up a drawing of what i mean later today. Off to the local junkshop to get some of those transistors. They have 1800 in stock...

[tnt]

Hi,

I did a few tests and I think the results are pretty interesting.

First a description of the equipement involved:

 - Rigol 1052E modified for 100 MHz
 - Agilent 3000-X 350 MHz
 - Agilent 3000-X 500 MHz

 - E4433B 4GHz RF signal generator

 - Pulse generator
 - Pulse generator + 1m RG402 semirigid coax as energy storage (see picture)


The first test I did was to get the 'true' -3dB bandwidth of the scopes by first feeding them a known amplitude 10 MHz signal and then rising the frequency until I reached a 3 dB attenuation. Results are as follow:

 - Rigol 1052E mod 100 MHz => 115 MHz
 - Agilent 3000-X 350 MHz  => 405 MHz
 - Agilent 3000-X 500 MHz  => 650 MHz


I then measured the rise time with the pulse generator alone:

 - Rigol 1052E mod 100 MHz => 1.28 ns ( ~ 273 MHz using 0.35 / t_r)
 - Agilent 3000-X 350 MHz  => 0.71 ns ( ~ 560 MHz using 0.4 / t_r)
 - Agilent 3000-X 500 MHz  => 0.54 ns ( ~ 740 MHz using 0.4 / t_r)

As you see the bandwidth is too high (I used gaussian model for the rigol and the maximally flat model for the Agilent), as several people pointed out, the pulse is too short and so can't be measure accurately by low bandwidth scope. The more bandwidth you have the closer you get to reality but for low-end scope it's _way_ over estimated !

I then tested to measure the rise time of the integral of the signal using the math function as Janne suggested. The rigol doesn't have this so it's only for the agilent:

 - Agilent 3000-X 350 MHz  => 1.15 ns ( ~ 350 MHz using 0.4 / t_r)
 - Agilent 3000-X 500 MHz  => 0.76 ns ( ~ 530 MHz using 0.4 / t_r)

It's more realistic but this time it's a bit too pessimistic (altough not that much, see later) but it could just be dumb luck ... more samples needed.

On the PCB, I left pads to solder a coax hardline as Jim Williams describes in a later app note, so here I soldered 1m of RG402 semirigid coax to it.

 - Rigol 1052E mod 100 MHz => 2.85 ns ( ~ 120 MHz using 0.35 / t_r)
 - Agilent 3000-X 500 MHz  => 0.73 ns ( ~ 550 MHz using 0.4 / t_r)

As you can see the Rigol is pretty much spot on, but the Agilent result is too pessimistic ...

However we must remember that the pulse generator itself has a ~ 350 ps rise time, and so the total system rise time is defined as :

t_sys ^ 2 = t_scope ^ 2 + t_pulsegen ^ 2

and so if t_sys = 0.73 ns  and  t_pulsegen = 0.35 ns  we get that the scope rise time must be 0.64 ns which correspond to an approximate bandwidth of 625 MHz which is what was expected.

So all in all I would say:

 - Add a coax hardline (15 cm should be enough) and that allows to really measure the scope bandwidth
 - Take the pulse gen into account for anything expected to be > 350 MHz.
 - Ideally known the caracteristics of your particular pulse gen to compensate for them as accurately as possible.

Cheers,

    Sylvain

[Lukas]

Why are you all fiddling around with avalanche pulse generators? We're not in the '60s anymore!
PECL gates such as the MC100EPT20 provide 120ps rise/fall time in a convenient SO8 package without arcane dead bug technology. Just put one on a PCB with two SMA connectors, hook it up to 3.3V and feed it with the input of a signal generator or include an oscillator on the PCB. When 120ps aren't enough, there are gates with 30ps rise/fall time, such as the NBSG11 (very expensive, free samples). Very fast comparators also have output drivers that produce fast edges.