iceNINE Tech

- Homebrew Really Fast Pulse Generator -

This web page describes my experiences with a really fast pulse generator that I built. I used a schematic from Linear Technology App Note 47 (pages 93 & 94) [1]; I assume that they may hold some sort of rights to this design. I feel, then, that I should explicitly state that I am not attempting to infringe on anyone's rights here; I am not selling anything or claiming that this design is in any way mine.

So. There are many reasons that you might want a really really fast-rise-time pulse. Jim Williams, author of LT App Note 47, can write much more eloquently [2] on this matter than I, so I will only give my reason - signal chain speed verification. I use a lot of Tektronix 7000-series scopes, which have mainframes (rated to a given speed), and plug-in amplifiers (rated to (often) another speed). Then you plug in a probe, with another frequency response. And, all of this stuff is both fairly old, and originally built to a higher quality (ie, faster speed) than guaranteed by the spec sheets.

Basically, I wanted a pulse that will drive a given signal chain (probe -> amp -> mainframe) to the limit of its rise time, so that I can measure that. There is a direct relationship between rise time and bandwidth, at least for a gaussian system like an analog 'scope. This relationship is somewhat different for some digital scopes - check your manuals.

Freq = 0.35 / Trise

This means that the one can evaluate the -3dB point of a system (like a scope) thusly:

3.5nS -> 100MHz, 1.75nS -> 200MHz, 350pS -> 1GHz (etc.)

(Click the image for a more readable version.)

Mr Williams gives a good bibliography (in the aforementioned app note) on transistor avalance-mode operation, which is how this circuit works. I will not attempt to descibe it, other than to say that it is elegantly simple. I initially used two bench supplies in series to get the 90V needed, which meant that the whole deal was only three resistors, a capacitor, and a transistor! The simple 1.5V -> 90V switcher added to the part count, but made it easier/safer to use (and much more portable!). While using the power supplies, I found that my assembly would begin to pulse at under 70V, but only slowly (less than 100KHz). Increasing the voltage sped up the rep rate, until at 90V I am seeing about 200KHz.

You can get the transistors at Digikey; they are not true 2N2369's (which come in a TO-18 metal can, or maybe TO-39), but will work fine (in my experience). I used the PN2369 (a Fairchild Semi part), in a TO-92. There are two others you may want to try, the MPS2369 (by ON Semi), or the PN2369A (again by Fairchild). I tried the latter (briefly, just curious if it would avalanche), and it appeared to work. These are all under 50 cents US in onesies (at Digikey). Also, I used a random 150uH inductor, I think it was rated for a few hundred mA. As the switcher only draws <5mA, I think many small inductors should work (but Your Milage May Vary).

If you are building one of these, and want a nice professional finish, I'd recommend one of the small, blue metal boxes that Pomona makes, with their high-quality BNC hardware attached. Put the whole deal in there, replace one of the bulkhead connectors with a toggle switch, and you're good to go. Try P/N 3605 (PDF datasheet). They're a little pricey, but very nice. Sometimes Digikey (there's a theme here) has them in stock.

And here are some results:

This is an early version of the assembly. I was aiming for fast rise only, and the abberation at the trailing edge is a result. The rise time appears to be approximately 350pS, which is the rated rise time for the mainframe. This implies that the entire signal chain is rated for at least 1GHz.

This is on a 7104 (1GHz analog) scope, with a 7A29 (nominally 1GHz, 50 ohm) vertical amp. The pulser is directly connected [3] to the vertical amp (no cable). The dual timebase shown is 10nS/div and 500pS/div. I used a 7M13 to do the custom text.

The camera used for all these pictures is a C5-B, loaded with whatever Polaroid film I could find that they still make, that fits. These are all ~1.5 sec exposures, using the scope's graticle illumination (no flash), developed for 45 seconds. You can see that the CRT of this scope is somewhat smaller than the camera's max image size (especially compared to the next shot).

7A29, 7B92A, in a 7104 (click for larger)

This shows the pulse shape (only), with the circuits' final configuration. I ended up using a 5pF capacitor to get the smoothest trailing edge.

Here I am using a system with _way_ less bandwidth. This is a 7D20 digital sampling unit (a whole 20 Megasamples), mounted in a 7403N (a little-known mainframe, possessing the same (large) size CRT as the common 7603). I have the timebase turned all the way up, and the 10x HMAG engaged, _and_ the maximum amount of averaging (256 sweeps), as well as the 'vector' mode (which fills in the distance between dots with a guess). If you look closely, you can see the individual dots that make up the pulse. The two bright points are the cursors, positioned as best as can be done - they need to be on actual samples, not interpolated lines, and there aren't many of them in this pulse!

Using the formula I gave earlier, a rise time of 2.5nS works out to 140MHz. Interestingly, there is a small sticker on the front of this plugin that says '-3dB @ 145MHz'. So I guess we're all in agreement.

Note that I had to use an external 50 ohm terminator, as this plugins' inputs are 1 Meg only.

7D20 in a 7403N (click for larger)

This scope is clearly not up to the task at all, though it's trying hard. The TPS2024 is a 200MHz-rated scope, running at 2G/S per channel. This waveform should be the same as that shown by the 7D20, but shows some interesting trailing-edge abberations.

The built-in rise-time measurement was used to get a figure of 1.85ns. It is not shown here because the scope cannot tell, when you have more than one 'rise' on a screen, which one you intend for it to measure.

Calculating the bandwidth from rise time (using the traditional, gaussian (analog) formula) gives approximately 190MHz. OK.


The exact circuit used to generate this pulse is the same as that used for the first, 7104-derived picture. It is interesting that the trailing-edge of the pulse is here much cleaner. I wonder what's really going on here. This sort of thing makes me think of Hiesenberg's Uncertainty Principle - it's not often that it really comes into play, but fast stuff like this is one of those times!

I borrowed a little time on a super-duper HP 1500MHz (only 4G/S, though) scope, and put it into single-channel max-averaging mode, to get the cleanest measurements I could. You can click on the image to the right, and get a larger, readable version.

HP 1500MHz Scope (click for larger)


[1] Variations of this circuit appear in other LT app notes; they are all recommended reading if you are interested in analog (real world) electronics.

[2] See the app note in question.

[3] I had to double-terminate this one (using an external 50 ohm terminator as well as the plugins' internal 50 ohm termination), as I had not yet put the internal attenuating resistor in place, and the pulse amplitude too high to get it all on screen at once.

Last updated February 2, 2006
Original content copyright © Christian Weagle unless otherwise indicated.