Junk box 2N3904 Avalanche Pulse Generator

One problem everyone comes across with test and measurement equipment, home made equipment, and ebay buys is how do you test it?  Well the most common way is to take your equipment in to a NIST certified calibration lab for testing.  Well this is neither cheap nor practical for many people.

But there are ways of creating cheap and practical references at home to test against your equipment.

Today we will be focusing on the avalanche pulse generator, this is a piece of equipment designed to give the fastest possible rising pulse so you can check the rise time of a frontend and calculate its bandwidth, it also has some other uses but we will get into that later.

The basic concept is rather simple, you operate a transistor in its avalanche region and once it breaks down or is triggered the current will flow out far more rapidly than it would in normal operation like a bursting dam or a spark gap.

Avalanche can be accomplished in many semiconductors like diodes but in my research the simple NPN transistor works best with the ability to reliably trigger and recover from the avalanche.  Not every transistor can reliably avalanche, not only different transistors but clones of those transistors and even the same transistor from different batches as transistors are not designed to operate in this region.

In this circuit i used a very common 2N3904 after reading this  wonderful post by Kerry Wong

I did do a number of modifications as shown by my schematic below.

Pulse Generator Schematic

First let’s start off with the voltage generator, in this i used an AP34063 DC/DC converter taken from an old router, this generates ~35 volts, at first i put it threw a simple 4 stage voltage multiplier forgetting about how large of a loss is created with a voltage multiplier being used in an switching supply like this where true AC is not created, in the end i needed 7 stages to create 125v
After trial and error with avalanche breakdown with the 2N3904 i found that 120v DC is required to actually avalanche my particular transistor, 125v DC gives me some nice leway to be sure it avalanches properly. Below is what my finished circuit looks like made entirely from junkbox parts

Finished 125v supply on scarp perf

The next thing we need to look at is the actual avalanche circuit itself, the 125v is fed in through an 220k resistor to limit current and provide a high impedance for the coax stub, This is than fed into the emitter of the transistor than into a coax stub, this stub is an about 1.25M long piece of stranded core 50 ohm coax terminated (with a dab of hot glue on the end to prevent a shock as it does sit at about 125v), this stub both gives some capacitance and increase the ontime of the transistor to give a decent square wave, I’m not entirely sure how this stub works but i assume it dumps the charge at very low impedance in to the transistor keeping it avalanching for a short period longer. The transistor avalanches in to 2 100ohm resistors presenting it with a 50ohm load and allowing voltage on a tap at the collector to dump into the SMA connector i used (mainly because i dont have any BNCs on hand)

Normally enough charge will charge up on the base to allow the transistor to trigger and avalanche, After an hour of figuring out why this would not work i noticed placing my multimeter probe caused it to avalanche, i placed a lower resistance across the base thinking that was it, it was not, i placed a capacitance across the base thinking that might have been it and that was also a failure
I figured i must have to trigger it as i have seen in Jay_Diddy_B’s post on the EEVBlog forums so i used a junkbox 555cn timer IC in a dead bug configuration on top of the avalanche board with a few random bits and bobs to create an about 1khz 10% duty pulse, this then goes through a resistor and an AC coupling capacitor to generate a proper trigger.
but somthing odd happens, even tho only 1khz is triggering the avalanche it appears to be avalanching at about 8-9khz, not too worried about this i chose not to look too hard into it.
The avalanche board is shown below on some thin scrap FR4 i had laying around in massive sheets

Hotglued Avalanche and trigger circuit


Now time to run some tests on my fastest scope i had (100mhz 1gsps gratten scope)

Looks like a rather large pulse …

As you can see the pulse is father high in amplitude, far higher in amplitude than one would want, so the only option was to insert a 10db 3ghz attenuator to be able to measure it properly, after the attenuator was installed everything appeared to ready to run some measurements!

More reasonable signal being measured

Now we can see a rise time of about 1.56ns and an 1/ΔT of 641mhz, i have heard of 3 different ways of measuring proper from this … 0.3*ΔT, 0.35*ΔT and 0.4*ΔT … so splitting the difference i went to 0.35*ΔT giving the frontend of this scope a total maximum bandwidth of 224.35MHz, this makes sense as they do have a 200mhz model of this scope and even tho i have confirmed that one uses a different ADC it does not appear to use a different frontend

Now how many quantization levels does it have? Well switching it over to a dots view should show us

As you can see it appears to quantize every 1ns giving us our 1 billion samples per second lining up to the 1gsps of our ADCs

Now im also curious at when the pulse gets triggered, so i attached chanel 2 to the trigger output and set my scope to trigger off that

You can see a few things in this, one it appears to start the avalanche when the trigger is at about 187mV … this is close enough to the 200mV i was expecting, you can also see the difference in rise times between the trigger pulse and the avalanche pulse, also you can see what appears to be reflections from the pulse finding its way back to the trigger, but nothing very concerning to insert any ferrites.

I have to say this was a fun $0 project, but if i were to spend some cash what would i have done differently knowing what i do now?

Well for one i would not use an 2N3904, the BFR505 appears to be a better solution at a simple 30v avalanche of ~200-300pS
i would also integrate an distributed attenuator in the circuit it itself to keep the pulse under 5V and use a much faster rising trigger

but for now this is a fine tool!

Now what else can this be used for? You have seen testing scopes but testing any kind of digitizer or frontend is possible, testing speed of logic gates is also an option, testing coax for its velocity factor and impedance is tempting also, with modification this can be used as a very fast trigger for a number of things, testing pulse detection can also be done

And another thing that slipped my mind during testing is using it with a splitter to test timing skew between channels

Well i did it, attached the pulse generator to a homemade T, some coax and SMA to BNC as shown below in order to test the channel differential

And we run a test with some cursors

As you can see a 160ps/6.25GHz differential is barely anything to shake a stick at in a 100MHz scope and can probably mostly be chalked up to error in measurement and in my setup

Noise Source Reverse Engineering and Improvements

I ordered a noise source from ebay for about $20USD, It came in good shape and working order just 2 weeks later. I say “[in] working order” rather loosely as the thing got hot, very very hot, easily over 70*c after just 10 minutes around the LNAs. Also it leaked quite a bit, properly terminating it and placing my cellphone on top of it caused me to lose reception almost immediately.

So, curious how it worked I got my flashlight and some reading glasses and started looking up part numbers. Soon after I was able to reverse engineer the unit (thanks in no small part by the mostly one layer design)

Not wanting to do a destructive tear-down I was unable to pull values of the caps but I think one can work them out.

The LNAs were interesting, I thought they were 2 different types but after looking up numbers it appears as if they are all the same SBB5089Z as U2 and U4 being manufactured by RFMD and U3 being manufactured by Sirenza Microdevices, Not sure why this is but both data sheets appear to be copy and pastes of each other

Another interesting aspect is they are driven directly from the 12v unregulated input and they appear to be relying solely on a 100R resistor to drop the voltage down to the 5v limit, not something I would have done with the cost of 5v regulators or even a zener diode but it appears to have worked.

Now how does this work? Well it appears to be a fairly common reverse biased diode noise source using a 24v zener diode supplied in what looks like a standard voltage regulator arrangement supplied with 38.5v, This is than AC coupled, put threw a 3dB attenuator then amplified by 60dB or so with 3 LNAs, put threw another attenuator than AC coupled to the output.

The first attenuator appear to be used simply to help maintain 50 ohms from the zener, and the last attenuator appears to simply not let the LNAs experience a high impedance if disconnected, This is a fairly common practice.

The 38.5v is created with a fairly standard and cheap MC34063AC DC-DC controller setup in a boost configuration. Its output is not as well filtered as I would like appearing to just have a few uF cap on the output.
Now let’s have a look at how we can improve this, What would make this unit more usable would be to pull away all this excess heat, we do this with an old hefty heatsink I have used on many projects I cut up. Next we must deal with the leakage, to do this I soldered some leads on to the ground leads of the LNAs (this will also help with some cooling) along with the ground lead of the DC-DC output cap, I than poked some holes in a piece of copper sheeting and soldered the leads to that copper piece as shown below

Step 2 was to place the module inside of an Altoids tin, this won’t give the best shielding but it will be enough to prevent any issues. After drilling out 2 holes for the new input jack and cutting a slit for the heatsink I than glued the heatsink to the bottom of the Altoids tin with a few springs to maintain a low impedance to ground, then glued the module to the top of that with some epoxy heatsink compound. Next was soldering the tin to the ground of the module wherever I could to prevent the creation of any sort of cavity filter, resonator, or oscillator, along with tacking down the copper shield.

One last little detail was poking a hole in the case to let the LED shine through, I than made a rudimentary light-pipe using a glob of hot glue and some copper tape to use as a reflector. It’s not great but the already bright and over driven green LED shined threw wonderfully!

Lets now have a look at how well it works from 25-1800Mhz


Noise Source Turned OFF


Noise source turned ON

NOTE: The dips at about 225Mhz and 450Mhz are due to a glitch in the software and are not actually present

As you can see it generates a good -50-60dB of noise, This eliminates any fear of radiating too much noise from your antenna under test while keeping it well above the -75dB noise floor of my reciever


Whats next? Well we will use this to characterize some antennas!