Significant ringing with the Jim Williams 2N2369A pulse generator from AN47

[6SN7WGTB]

I've knocked up a pulse gen using a 2N2369A as per the AN47 design. Fed with 90V from my stabilised PSU, via a screened cable.

The pulse seems as 'fast' as my 'scope can measure (around 600ps, with ca. 15ps SD).

However there is a significant ring. I have tried to make the layout as best as possible. Any thoughts please of what I'm doing wrong? And/or might this be a case of trying various 2N2369As?

The decoupling cap on the PSU input makes no difference to the ring, and I've tried with and without the ferrite: the SD of the pulse rise time is lower with both, but neither affect the ring.

The yellow braided sleeve contains the 1MΩ resistor from the PSU input.

Thanks.

[TimFox]

What is your output termination (50 ohms?)?

[6SN7WGTB]

Very good question. The 'scope is 1MΩ, and I immediately thought that I should add a 50Ω inline termination.

So I did and it made no difference.

Hmm. Then I realised the circuit has a 50Ω resistor from output to ground anyway and given that this is right on the centre pin to ground I can't see what another 50Ω termination will achieve given my little box is plugged onto the 'scope BNC. Anyway, it didn't...!

EDIT: have snipped off collector lead to body and soldered the 1MΩ and 2p2 direct to the case. No improvement.

[Benta]

Could you be a bit more specific about AN-47?
Eg, page # reference. I find no pulse generator in there.

[macboy]

Will all due respect to the brilliant Jim Williams, you can't measure scope rise time with the AN-47 pulser. This circuit (this specific version of this circuit) provides a high voltage impulse of very short duration. In order to measure rise time, one needs a fast edge step waveform, not an impulse.

Your scope response to the impulse seems reasonable. An old analog scope with a gaussian rolloff characteristic may show something closer to the app note, a nice symmetric rise and fall without ringing. Most modern digital scopes have a sharper rolloff, and this always results in ringing of the impulse response (and correspondingly, some overshoot in the step response). Try taking the FFT of the impulse waveform. You will be pleasantly surprised. As long as the input impulse is a close enough approximation to a theoretical impulse (i.e. very short duration), then the FFT of that will reveal the frequency response of the scope (actually of the system: the scope+interconnects+pulser itself). You can get a similar result using a fast edge step , using a math function to take the derivative with time dV/dt, which converts the step into an impulse, then take the FFT of this computed impulse waveform. Sadly, not all scopes allow FFT of a math result (the dV/dt impulse). My LeCroy allows it (obviously), but my Rigol MSO5xxx does not. One example of an obtainable fast edge step generator is the Leo Bodnar pulser, search this forum.

You might be able to use the scope math functions to work backwards to a step response from the impulse response that you have. Just take the integral over time. This should result in a step, almost certainly with some ringing. Then measure the rise time of the resulting integral / step waveform in the traditional way. Make sure there is no zero offset when measuring the impulse (preceding and following the impulse should be exactly zero), or the integrated step will have a strange slope/drift to it.

In a later application note (I'll try to look up the number later at home), Jim presented an updated version of the 2N2369 pulser which used a length of coaxial cable in place of the capacitor, plus other tweaks. This resulted in a step waveform corresponding to the length of the cable, very approximately 1 ns per foot of length. This fundamentally changed the circuit from an impulse generator to a step generator, and allowed direct rise time measurements on the resulting waveform.

[porter]

Could you be a bit more specific about AN-47?
Eg, page # reference. I find no pulse generator in there.

check out
APPENDIX D
Measuring Probe-Oscilloscope Response

[RoGeorge]

I would remove the base inductance (the bead), in the AN47 there seem to be no ferrite bead.  Also I would make a single ground point, and solder the decoupling capacitor there, right near the transistor.

If you probe the signal without any additional cable (using only the oscilloscope probe), it should work as it is, without any additional 50 ohms.  Probe must be on dividing by ten, not on 1x, and must use the small GND spring at the tip of the probe, no long GND clip wire, and no tip hook, just the naked probe with the short spring GND.

[T3sl4co1l]

From the horse-mouth: https://www.analog.com/media/en/technical-documentation/application-notes/an47fa.pdf p.93

The impulse can easily be extended: simply replace C1 with a transmission line (open circuit) stub.

I made this years ago,
https://www.seventransistorlabs.com/Images/Avalanche%20Pulse%20Generator.gif
https://www.seventransistorlabs.com/Images/AvalancheGen.jpg
Layout isn't particularly good, but I don't need it for pristine impulses/steps.  (Notice the BNC connectors are bonded to ground, there's a solder joint directly to the thread.)

A 2N3904 is used, which breaks down around 100V; YMMV.  It can also be triggered (via the transformer), though mostly I use it free-running.

I later made an improvement to the trigger,
https://www.seventransistorlabs.com/Images/Self-Powered_Schmitt.png
https://www.seventransistorlabs.com/Images/AvalancheGen2.jpg
which was here using a ~300V transistor in avalanche to switch a grid PFN for other testing.

Avalanche breakdown is simply going from a high to low resistance state; physically, a minute conductive filament pierces the N-P-N semiconductor stack, presumably triggered by an avalanche cascade, and apparently the mechanism is local heating so intense that the current filament becomes intrinsic (and highly conductive; T > 300°C?). thus punching through the base layer, momentarily shorting C and E together.  (A good 10µs or so after the impulse, things cool down, carriers recombine, and it becomes nonconductive again.)   This is an incredibly tiny amount of material to heat up; even so, it's able to sink tremendous current: the 2N3904 above is doing about 1 or 2A peak, and by 5-10A I think, destruction quickly ensues.  Characterized devices are available from Diodes Inc: https://www.diodes.com/assets/Datasheets/ZTX415.pdf able to carry 30A or more, albeit at somewhat slower risetime.

Large devices can be made to avalanche as well, but it's usually fatal: a power transistor's entire junction capacitance discharged through a single filament quickly vaporizes the area, punching a hole through the transistor: the failed device has a modest resistance C-E, but still "transists" otherwise, evidencing the highly localized failure mode.

(Incidentally, the same mode of operation affects MOSFETs as well, which is why avalanche peak current, and drain rising dV/dt (particularly in body diode reverse recovery), have limits.  It's generally hard to get a MOSFET's parasitic BJT to activate, but that is precisely what these ratings concern, and exceeding them causes runaway or point failure, very quickly (~ns) indeed.)

Tim

[RoGeorge]

Another thing I would try before anything else would be to disable sin(x)/x interpolation in the oscilloscope, and/or to put the display in dots mode.  The interpolation algorithm should help only for slower signals, but it may draw artifacts when trying to represent signals faster than the oscilloscope's bandwidth.

I don't know if this is the case for your oscilloscope model, but it's easy to try if disabling sinx/x helps.  My Rigol DS1054Z does some ugly things with very fast pulses, and in my model the sin(x)/x can be disable when only one channel is active, for more than one channel is always on (with disable greyed out).

[T3sl4co1l]

Another thing I would try before anything else would be to disable sin(x)/x interpolation in the oscilloscope, and/or to put the display in dots mode.  The interpolation algorithm should help only for slower signals, but it may draw artifacts when trying to represent signals faster than the oscilloscope's bandwidth.

I don't know if this is the case for your oscilloscope model, but it's easy to try if disabling sinx/x helps.  My Rigol DS1054Z does some ugly things with very fast pulses, and in my model the sin(x)/x can be disable when only one channel is active, for more than one channel is always on (with disable greyed out).

This.  Note it's showing 1ns/div but only 8GSps.  My old TDS460 (350MHz, 100MSps real time) does a whopping 50GSps at that scale -- not that it's very useful, the ET sampling fills in extraordinarily slowly (but, that's a design issue I'm pretty sure, and good ET acquisition would be ~instantaneous on repetitive signals).  At 8GSps, there are far more pixels than samples on screen (what, 50-100 px/div?), and thus significant interpolation to be done.

How points between samples are to be rendered, is an open question, but a common choice is the brick-wall reconstruction filter: in other words, sinc interpolation.  Given that the device bandwidth is only so-and-so, the sinc kernel spans this-and-that many samples (out to some modest cutoff window).  We can thus construct a filter (probably FIR is used, but whatever works) and filter samples through it, and obtain an up-sampled output to send to the display.  The biggest downside to this is the significant overshoot on very fast signals, overshoot that doesn't match the actual analog response (which tends to be more of a single or double pole transfer function, limited by input amp BW or ADC sampling aperture).

Tim

[6SN7WGTB]

Could you be a bit more specific about AN-47?
Eg, page # reference. I find no pulse generator in there.

Pages 93,94,95, August 1991 AN-47.

[6SN7WGTB]

Lots of feedback overnight - thank you all. Will digest today. (I am in the UK).

BTW, am not seeking to use this for measuring 'scope/probe response - it is for some TDR tests. It's all hobby stuff, but I know I should be able to get a better pulse.

I do note with interest that the 'scope may be rendering the signal incorrectly. Will look at that too.

I did previously build a 2N3904 version - in fcat two: one SMD and one full fat. The SMD one was minute, bt oddly the pulse was actually sharper on the full fat one. And on the SMD I did try with and without additional inductance in Tr leads to mimic full fat one.

Will report back

[iMo]

Mind the type of the resistors there in your gadget is not optimal, imho, as the TH types have large parasitic L and C. Every mm/pF/nH counts at those frequencies (aka edges).
Also the TH orange capacitor is not optimal, moreover it shall be wired at the transistor, afaik (doublecheck, if it is the aprox 2pF one).

[jonpaul]

Homage to Our old friend Jim Williams ...brilliant analog  circuit man to thevery end.

His avalance pulser is fine but not the best solution.

The TEK TD pulsers can be better and the best is the Leo Bodnar 30 and 40 pS pulsers.

For the OP, the scope is showing typical ringing due to poor probing technique.

ANY probes, cables, terms are  suspect especially any leads.

Our test setup:  NO cables,  reduce VSWR reflections, attn 20 dB 

Leo Bodnar BNC pulser>>>Mini Circuits HAT-20 50 Ohm 20 dB atten>>TEK 2467B on 50 Ohms input term.

See the old  "show us your square wave"  thread for images...

Enjoy,

Jon

[6SN7WGTB]

Mind the type of the resistors there in your gadget is not optimal, imho, as the TH types have large parasitic L and C. Every mm/pF/nH counts at those frequencies (aka edges).
Also the capacitor in not optimal, moreover it shall be wired at the transistor, afaik (doublecheck).

I did debate resistors: carbon comp, carbon film, or metal film. Which would you suggest and why?

The 2p2 is soldered to the Tr leg and straight to ground - so what am I doing wrong?

[RoGeorge]

some TDR tests

Have you seen this video from w2aew:   #323: Measure length of coax, etc. with your scope, a battery and a resistor - simple TDR? 

[iMo]

Mind the type of the resistors there in your gadget is not optimal, imho, as the TH types have large parasitic L and C. Every mm/pF/nH counts at those frequencies (aka edges).
Also the capacitor in not optimal, moreover it shall be wired at the transistor, afaik (doublecheck).

I did debate resistors: carbon comp, carbon film, or metal film. Which would you suggest and why?

The 2p2 is soldered to the Tr leg and straight to ground - so what am I doing wrong?

Ok, I have not seen the 2pF on the top picture there.
You have to use the smallest SMD parts, with almost zero leads, afaik people make the 50ohm one as two -three smdes in parallel soldered directly to the BNC/SMA connector (from the ground metal part to the mid pin). Google for it there is a plethora of designs also here in eevblog afaik.

[T3sl4co1l]

I mean, "smallest" is relative.  The electrical length of 300ps is still some 60mm, so using components and connections much shorter than this is acceptable -- you do still have to be mindful of their strays, though.

Keep in mind, Tektronix built their golden age solid-state scopes (say, 485, 2465) in THT -- 1/4W axial resistors and etc. throughout.  The secret sauce between 465, 475 and 485 (and beyond) was moving more amps from discrete transistors to ceramic hybrids and monolithic ICs.  Not that 350MHz is exactly in the 300ps domain, and not that these designs were exactly simple to begin with (not to mention the decades of iteration that led to them), but that's merely a simple proportion, and not even a large one -- these had 1ns risetime, so a factor of 3, say.  If a 10mm-long resistor is good enough for that, a 3mm-long chip (3216 / 1206) is plenty here!*

*For the same resistances, and ratios to Zo and what have you, of course.  There is some further advantage, as chip resistors are usually zig-zag or pinch design, whereas axials are usually spiral (of a few turns), so THT resistors end up a few mm (maybe even cm) longer than body length alone.

Also, "smallest" is almost certainly not desirable here: the peak power of 100V into 50 ohms is probably enough to blow a, say, 0201 chip?  Or, enough voltage to just arc over completely.  Even smaller are available, but one begins to strain the "possible" requirement, given the capability of human hands, and availability and capability of applicable (likely: budget proto) fabs.

Tim

[tggzzz]

BTW, am not seeking to use this for measuring 'scope/probe response - it is for some TDR tests. It's all hobby stuff, but I know I should be able to get a better pulse.

Several excellent reasons

The issue with using a pulse for TDR and scope response is that what you see is the convolution of the scope and the input pulse. In order to be able to separate out the scope/TDR response, you have to have characterised the pulse. Chicken vs egg

A better way is to create something where the design creates a known output, one which is only compromised by the construction imperfections. Possibilities include:

• add stub line on avalanche pulser. Not sure how construction imperfections will affect the waveform purity
• current mode output from a fast comparator. I believe this is in Leo Bodnar's device. Limited voltage swing, which is unimportant for many purposes
• fast "TTL" logic. This can drive a 2.5V pulse into 50ohms with a ~250ps risetime. See the "show us your square wave" thread

Have fun

[6SN7WGTB]

Ok, I have not seen the 2pF on the top picture there.
You have to use the smallest SMD parts, with almost zero leads, afaik people make the 50ohm one as two -three smdes in parallel soldered directly to the BNC/SMA connector (from the ground metal part to the mid pin). Google for it there is a plethora of designs also here in eevblog afaik.
[/quote]

This was an SMD version, admittedly with a 2N3904, but it would not better 1250ps.

It was very small - mounted inside a phono plug that plugged into a phone-BNC converter directly onto the 'scope BNC.

[6SN7WGTB]

For the OP, the scope is showing typical ringing due to poor probing technique.

It is plugged onto the 'scope BNC, so there is not probe or probing.

[tggzzz]

For the OP, the scope is showing typical ringing due to poor probing technique.

It is plugged onto the 'scope BNC, so there is not probe or probing.

Is the scope input a "real 50ohms", or is it merely 50ohms//15pF - i.e. just a 50ohm resistor banged across the input?

[RoGeorge]

It is plugged onto the 'scope BNC, so there is not probe or probing.

Usually the scope input has a considerable capacitance when compared to a 10x probe tip.  Try measuring with a probe set on 10x, and the pulse should be sharper.

[tggzzz]

It is plugged onto the 'scope BNC, so there is not probe or probing.

Usually the scope input has a considerable capacitance when compared to a 10x probe tip.  Try measuring with a probe set on 10x, and the pulse should be sharper.

That's a case of RTFM, of course.

I wouldn't regard the difference between, say, 10pF and 15pF as being "considerable". I would regard inserting a complex RLCZ network (i.e. the probe) as being more significant.

[6SN7WGTB]

Just to be really clear this device is plugged into the 'scope BNC. No probe, No cable. It is a BNC plug with the components mounted on it.

It has a 50Ω resistor right across the centre pin to shell, so a 50Ω termination will make, and did make, no difference. The 'scope is however a 1MΩ input. Not sure what capacitance, but vaguely recall 10-15p.

I will experiment with settings as I need to bottom out what impact the 'scope may be having before I consume time on the pulse generator.