Clock with a radioactive timebase

Similar projects worth following
I have a "thing" for clocks with unusual timebases. Interesting and inventive clock displays and wondrous clock cases are fine, I guess, but it's what makes a clock "tick" that fascinates me. Having had great fun with my bendulum clock project (which still runs great), I went looking for an entirely different approach to mark the passage of time, and decided to try to build a clock driven by radioactive decay. The decay of 241-Am, in particular.

This project is by no means finished. I'm not even sure it can be finished, but I'm sneaking up on it and thought I'd share what I've got so far. Suggestions are more than welcome.

Why another clock?

The clocks people use to tell the time of day consist of a display to show what time it is, a timebase to drive the display, and a case to hold them. The number of ways people have found to build clocks -- to vary the display, the timebase and the case -- is truly astonishing. Sadly, since the development of the super-cheap, accurate quartz clock movement almost all of the interesting new clocks use that as their timebase and only vary the case and the display.

There are exceptions, of course. Some interesting new clocks use traditional mechanical movements such as pendulums or balance wheels. Others use the 50 or 60 Hz of the AC power as their timebase. A few use standard time signals such as those from WWV, or the GPS constellation to keep their primary timebase synchronized closely with Coordinated Universal Time. But for the most part, hobbyist clock builds and commercial clocks for consumer use all employ a quartz movement and differ from one another just in their the case and display.

For commercially made clocks the reason is obvious, but why do our clock builds go down this same path so often? Personally, I find the timebase part of a time-of-day clock just as fascinating as the case and the display. Maybe more so. I guess that’s what motivated my bendulum clock build.

The fun and success I had with that build led me to think about other ways I might make a timebase for a clock. In the course of that build I struggled for quite a while with the fact that its bendulum, like all physically oscillating timebases, from pendulums to quartz crystals, changes its rate slightly as the temperature changes. Whenever something physical oscillates back and forth, thermal expansion tends to change the rate at which it does so. To counteract this, you can either compensate for it by observing how it changes and counteracting it in the design (the approach I took with the bendulum) or try to hold the temperature constant (as is done with an “oven controlled crystal oscillator” or OCXO). But is there another approach all together?

How is this one different?

After having the question roll around in my head for quite a while, one day a possible answer just sort of arrived out of nowhere: radioactive decay.

The average number of decay events that occur in a sample of a long-lived radioisotope changes very slowly over time and in a way that is completely predictable. Importantly, the rate of decay doesn’t change at all with changes in temperature, pressure or other environmental conditions. What I started to think about was whether I could derive, say, a 1Hz timebase from the radioactive decay of a small, safe sample of some long-lived radioactive substance.

The problem is that although the rate of decay is completely predictable over the long haul, every decay event is completely random in time. A sample small enough to be both obtainable and safe will necessarily have a number of decay events per second that’s pretty small. That means the number of decay events that occur in any given second will be unpredictable.

Suppose you had a sample you knew underwent 10 decay events per second, on average. If you drove a clock by counting decay events, moving the second hand ahead by one each time you got to a multiple of ten, the clock would keep time. On average. But in the short term it would always be wrong. Sometimes the clock’s “seconds” would occur quickly and sometimes there would be long pauses between them. If you’ve ever heard a Geiger counter you know how irregularly the clicks come. I don’t see any way to directly derive a regular timebase out of that.

But the fact that the average decay occurs so predictably means that we should be able to use such a radioactive sample to adjust the run rate of a not-so-good 1Hz timebase so that it stays very close to correct and over...

Read more »

  • Disappointing, but Not Unexpected

    Dave Ehnebuske06/02/2018 at 18:18 0 comments

    Well, I've now logged how many clicks my Geiger counter saw each minute since some time on May 20. It's more than clear that the click rate is falling gradually. When I first became suspicious a while back, the rate was about 50 clicks per second and now it's about 45. The radioactive source I'm using is the little dot of 226Ra stuck right up next to the GM tube that I've used for some time now. Here's the data:

    As with earlier charts, the blue line is the average of the whole run up to that point and the red line is an exponentially decaying moving average with a decay period of an hour -- a measure of the behavior during the previous hour. For my purpose -- building a timebase -- this is not good at all.

    I'm relatively certain the problem is with the long-term stability of my Geiger counter, not the physical setup or the radioactive source. It's still possible, I suppose, that there's an environmental problem (e.g. variability due to Rn), though there's no obvious diurnal pattern to the data or correlation that I can see with the weather.

    Time, I think, to look into a different (type of) detector to see how it behaves over time. I need to do some research on the alternatives.

  • What Counts as "Flying"?

    Dave Ehnebuske05/12/2018 at 17:25 0 comments

    Observation shows that the change in the click rate for my Geiger counter has, for now at least, stopped changing (so much) over time. Here's the data from a recent run of well over a day:

    As you can see, the average is no longer falling as it was. In fact it's behaving within statistical limits. If it had behaved this way from the beginning, I would not have become suspicious it was changing.

    Quite why this should be, I'm not at all sure. Maybe the brand new Geiger counter just needed to run for a while, and now all's well. Maybe it's just not very stable? Maybe its sensitive to "the phase of the moon"? I haven't found anything yet that would indicate it's fundamentally unstable or subject to malign influences, but then I wouldn't: its been stable for a while now.

    Such as it is, that's the good news: results so far don't rule out getting this to work. But there's bad news, too. A little research shows that the sensitivity of Geiger counters of the sort I'm using tends change over time. None of the sources I've found say why the sensitivity changes, but people talk about needing to calibrate "survey Geiger counters" against a standard every few months. That doesn't bode well for what's supposed to be a source of truth. I'm beginning to worry I need a different sort of detector.

    Research, both into stability and into alternative detectors continues.

  • Pig Airport

    Dave Ehnebuske05/05/2018 at 16:17 0 comments

    Well, I built the first version of a complete clock using the Geiger counter, an interface shield, an Arduino Uno and a Lavet motor clock display. The Arduino sketch uses detected clicks from the Geiger counter to figure out updated guesses estimates of how many Arduno micros() there are in a standard second and uses that to drive the clock display. The idea is to calibrate the speed of the Arduino's not-too-accurate ceramic resonator clock using the count of the decays detected by the Geiger counter.

    While I'm pretty happy with the way it works, continued observation of the click rate from the Geiger counter worries me. The click rate continues to decline slowly with time. At the beginning of the week it was just under 51 clicks per second, now it's just under 48.  Since I don't want to build the equivalent of  "pig airport" -- a bunch of fancy, complicated infrastructure based on the fundamentally flawed idea that pigs can fly -- I need to figure out why the click rate is not constant.

    As [Thomas] observed, there are several critical aspects of the way the Geiger counter detector is built that could be the cause of what I see. I've tried to make sure the radioactive source itself is stable and in a stable relationship with the GM tube. So, I'll look first at things I haven't tried to control. For example, perhaps the voltage on the GM tube is wandering around. Or maybe changes like this are just what one sees if a Geiger counter is run continuously for a week. For that I'll need to do some research.

  • A Longer Run

    Dave Ehnebuske05/02/2018 at 14:46 0 comments

    To test whether the slow decline in the click rate I saw in the several-hour run I talked about a few days ago, I did a much longer run using the same setup as last time. Here is the result:

    As with the chart from the previous run, the blue line is the average number of clicks per second to that point in the run. The red line is the moving average with a decay of an hour. This run is ambiguous enough that I've decided to carry on with the experiment using a "hotter" 137Cs source Getting one could take a while, so in the meantime, I'll assume things will work out and plan to carry on with developing the code for the 1Hz timebase.

  • New Radioactive Source

    Dave Ehnebuske04/30/2018 at 16:37 5 comments

    After an extremely productive discussion here on, I decided to change the source of decay events I'm using. I swapped out the 241Am source for a 226Ra source I had obtained a while back on fleaBay. It claims to be a 6k CPM source intended for calibrating old Civil Defense Geiger counters from the cold war days. 226Ra breaks down to 222Rn by emitting an alpha particle and a gamma photon. (See the radium decay series chart here.) My Geiger counter doesn't detect alpha particles, but it does detect gamma photons.

    The idea here is to raise the rate at which decays are detected to the point where I get enough of them that the statistics will produce an acceptably accurate clock. The mathematics are covered on a very useful page (here) that [Ted Yapo] brought to my attention. What it nets out to is the higher the rate of detected decay events ("clicks"), the more accurate the clock. For example, I need at 168 clicks per second for the clock to achieve (within 1 standard deviation) a 1-minute-a-week accuracy. To achieve a 1-minute-year accuracy, I'd need to detect 8760 clicks per second.

    So, I swapped in my 226Ra source and measured the click rate against a realtime clock with an accuracy of 2 sec/year. So far, the result is a little disturbing.

    The good news is that the rate is up quite a bit. The average over the 15.5-hour run is a little over 50 clicks per second. That's not enough. I need more than three times that many even to get to an accuracy of a minute a week. It looks like I'll need a much "hotter" source. More disturbing is that, so far, the average shows a trend:

    During the run, the number of clicks detected was measured once a minute. In the chart, the blue line is the average of the click rate over time. The red line is the exponential decay moving average with a decay period of an hour -- basically a measure of the rate during the course of the preceding hour. As you can see, both measures decrease with time.

    Ack! Where did that come from? Is my Geiger counter getting tired of counting this quickly? An artifact of running for too short a time? An experimental error? At 50.9 clicks per second and 15.5 hours of run time, the SD is 5.9e-4, so the decrease looks like it could be real.

    Time will tell. I'd welcome your thoughts.

View all 5 project logs

Enjoy this project?



Thomas wrote 04/29/2018 at 19:53 point

As I see it, the approach has something in common with a mechanical clock:

* the exposure of the detector to the source must be constant (angle, offset)

* in the case of alpha particles properties of the air must be constant (better: no air at all)

* the sensitivity of the detector threshold must not depend on temperature, voltage, aging, etc.

In other words: any variation in parameters that change the sensitivity of the setup needs to  managed.

  Are you sure? yes | no

Dave Ehnebuske wrote 05/01/2018 at 00:34 point

So, like other physical devices, its behavior is determined by certain aspects of its construction that must be controlled to get the desired behavior. I agree completely.

  Are you sure? yes | no

Thomas wrote 05/01/2018 at 07:27 point

Please note that the application of a statistical theory was proposed to the technical problem (the observer assumes a Poisson process, which is probably true for the 241-Am source but not for the device that has a propensity for observing a decay event). I think that your device is a particularly interesting for applying the "propensity theory of probability" (Objectivity), not the "Bayesian theory of probability" (Subjectivity).

20th century scientific theory dealt a lot with the truth contents of theories. Here is a good introduction the central part of the concept of Verisimilitude: . The underlying problem is one of decision (how to assess Verisimilitude). Personally I believe that Herbert A. Simon's concept of Bounded Rationality applies here (i.e. be pragmatic).

  Are you sure? yes | no

A. M. Aitken wrote 04/27/2018 at 23:15 point


This is awesome.  And so much more feasible than I'd have guessed.  The one issue I have with Ted's maths is that it assumes you know the effective activity to high precision at the start. It will have to be calibrated and if this is done for a month it will be the bigger source of error of the running clock after a year.

Regarding the description, 2cps is odd.  Either the source is of very very low activity (like a cloud chamber alpha source) or maybe the Geiger tube is not a mica window version?

  Are you sure? yes | no

Dave Ehnebuske wrote 04/28/2018 at 00:15 point

Glad you like it.

The calibration "plan," such as it is, will be to adjust the running speed over time in the same way that you'd adjust a "normal" clock. Even if I knew the specific activity of the source, I can't see how I could capture all of the decays, so I'd be stuck doing that anyway, I think.

As to the 2cps problem, yes, it's a pretty small source -- one harvested from a defunct smoke detector -- and the Geiger tube isn't one with a mica window, so no alpha particles which is what 241-Am mostly decays into. After Ted's comment, I moved the source to right next to the Geiger tube. Now it counts about 6cps. Maybe I should switch to 226-Ra and see how quickly a little Arduino can count.

  Are you sure? yes | no

A. M. Aitken wrote 04/28/2018 at 14:40 point

The Am source as Ted said should be producing over 30k alphas a second, of which about half should be exiting the surface of the foil. How close that is to the real world I'm not sure. Without a thin window you are probably detecting low energy gammas and the odd x-ray at very low efficiency. 6cps works out at around three quarters of an hour in a year. I wouldn't switch to a different source, but if you want better performance I would change the detector.

  Are you sure? yes | no

Ted Yapo wrote 04/28/2018 at 17:45 point

You can buy 10uCi 137Cs sources license-free in the US - I have one - and I've seen it saturate a small tube before.  Lots of gammas.  But, they're not cheap at all - United Nuclear sells them, along with a number of other isotopes.  Electronic Goldmine sells radium watch hands for around $5 when they go on sale (Listed at $10 now).  But 226Ra is also alpha-decay - you only see gammas from the daughters. Lastly, Walmart sells a $5 ionization smoke detector - I've used the button from one in a spinthariscope before.

I've occasionally seen Russian mica-windowed tubes on ebay relatively cheap.

  Are you sure? yes | no

Dave Ehnebuske wrote 04/28/2018 at 19:16 point

Thanks, everyone, for the feedback. It's very helpful.

I didn't use a counter with a mica-window GM tube because I'd either have to pay quite a bit for a complete unit or I'd need to design and build one myself. After repeatedly failing in my attempts to build a PIN-diode alpha detector, contemplating the likely difficulty of things like building the required a high-voltage power supply led me to a low-end, pre-built Geiger counter with a glass GM tube. I'm thinking I'll stick with that for now and try a "hotter" source of gammas.

A 137-Cs source isn't cheap but it's not crazy expensive, either. 137-Cs has a half-life of 30.5 years, so I'll need to compensate for it's decline in activity over time, but that shouldn't be too hard.

  Are you sure? yes | no

Ted Yapo wrote 04/28/2018 at 19:58 point

@Dave Ehnebuske so, the thing about 137Cs and the 30-year half life.  After 30 years, the activity from the 137Cs has dropped to half.  But, the "missing half" is now a bunch of different radio-daughters, each decaying at their own rate and adding to the activity. So, the activity from the resulting soup of species doesn't follow an exponential decay exactly.  I don't know how much of an effect this will have - maybe someone has done the calculations somewhere.

EDIT: I looked it up, 137Cs decays via beta to Ba-137m, which actually yields the gamma decay.  The resulting 137Ba is stable, so ignore the above!

  Are you sure? yes | no

A. M. Aitken wrote 04/28/2018 at 20:24 point

That mica tube is quite slow and is rated for a lifespan of 50'000'000 counts, which would be 3 months at 6cps, the rate of the current tube. It may be organic quenched, Russian tubes are a bit of a lottery.

  Are you sure? yes | no

Thomas wrote 04/27/2018 at 19:51 point

It's not uncommon to use radio-controlled clocks as a server timebase. Your radio-activity controlled clock could do that, too, but it could also be applied as an entropy source for cryptography ;-)

  Are you sure? yes | no

Dave Ehnebuske wrote 04/27/2018 at 22:47 point

Yes, entropy is something my little button of 241-Am has lots of. But making a cryptographic RNG that would pass scrutiny with the crypto-folk is probably a lot harder than making a clock!

  Are you sure? yes | no

Thomas wrote 04/28/2018 at 01:16 point

  Are you sure? yes | no

Thomas wrote 04/29/2018 at 08:57 point

I ordered a smoke detector "ion chamber) with "<0.8 µCi 241-Am". Now I'm looking for a reasonably priced PIN diode that can be fully exposed. The author of the link I posted used a BPX-61, which is rather expensive, and I don't know how difficult it is to remove the glass window. I also have a small (rather insensitive) Geiger-Müller tube as a fall-back.

  Are you sure? yes | no

Dave Ehnebuske wrote 04/29/2018 at 14:35 point

@Thomas I tried walking down the PIN diode detector road. I found it pretty frustrating -- my design skills with extremely small signals and op-amps just isn't good enough. I found this reference quite helpful, though:

Maybe it'll help you, too.

  Are you sure? yes | no

Thomas wrote 04/29/2018 at 19:11 point

Thanks, that looks like valuable information even if the BPW34 (unexposed PIN diode chip) isn't likely to be sensitive to alpha particles. The paper above uses an exposed chip.

In the meantime I found a mention that almost any diode will work for detecting alpha particles if it's large enough, e.g. the BE junction of a 2N3055 transistor.

Inserting a 241-Am source into a TO-3 case might be a good option :-)

  Are you sure? yes | no

Dave Ehnebuske wrote 04/27/2018 at 14:58 point

Thanks very much for this comment, Ted. It and the referenced page help enormously!

You've understood precisely what I'm trying to do.

I have been wondering for some time how to calculate how accurate a clock like the one I'm playing with could be, and this tells me exactly how to calculate that. It also tells me how important it is that I design the thing to capture as many of the decay events as I can.

Thanks again for the comment!

  Are you sure? yes | no

Ted Yapo wrote 04/27/2018 at 16:47 point

Glad you found it useful.  Very cool project!

The unstated assumption (although discussed on the linked page) is that for large N, the Poisson distribution is well approximated by a Gaussian, so you can use the 68-95-99 rule with the SD.  The page says that this approximation is good even for N=20, so it seems OK even for a few seconds of this kind of "atomic clock" ;-)

  Are you sure? yes | no

Ted Yapo wrote 04/27/2018 at 02:25 point

Are you counting detected decays as a timebase? I think you can calculate the expected accuracy using a Poisson distribution.  This page has some pertinent info:

The key point seems to be that the variance of the Poisson distribution is equal to the mean, or the standard deviation to the square root of the mean.  For a 1 uCi sample (say a button harvested from a smoke detector), the mean count would be 37,000/sec.  For the moment, assume you can somehow detect them all.  One standard deviation after 1s is 192 counts (= sqrt(37000)), so you have a 68% chance of being closer than 5ms (= 192/37000).  Very good.

Now, wait a year.  The mean number of counts is 365*24*60*60*37,000 =  1.17e12.  The SD is 1.08e6, equal to 29 seconds.  So, you have a 68% chance of being within 30 seconds accurate, or a 95% chance of being accurate within a minute (2-sigma).  This is not that bad really.

The numbers get a little worse if you only detect a fraction of the decays (which, of course, you will), but at first glance, it looks feasible.

Check my math and method, because, you know, I could be wrong.  This is a very interesting project!

  Are you sure? yes | no

Similar Projects

Does this project spark your interest?

Become a member to follow this project and never miss any updates