updated 20 aug 2024
new May 2012
This hardware random number generator I built with what I believe is the same device used to produce this famous 1955 book, One Million Random Digits and 100,000 Normal Deviates.
Randomness isn't a property of anything. It isn't part of nature; it's a made-up human "problem". Any consensus of what constitutes a definition of "random" is local to some particular discipline that needs it. The whole business is a bit disturbing. For a taste of this, check out this Wikipedia entry on random sequence.
(True randomness is a slippery and difficult concept, entirely unnatural. Vast fields of rigorous study require random sequences; while some portions of the concept of random are easy enough to grasp, there are subtleties that make it deeply weird. (And deeply boring, to most people, many of whom might in fact stop buying lottery tickets if they bothered to know anything about statistics and probability. The people who run lotteries certainly do.)
The Gas Tube Number Generator contains a hardware true random number sequence generator, from which it "extracts" (sic) umm, nothingness whatsoever, and upon a literal hand-wave selects a visual graphical symbol from a random sequence of binary digits. It is therefore an oracle of nothing; it rigorously, provably, provides nor contains any information, bias, opinion, or predictability. Which, as it turns out, is a lot of work.
At the dawn of automatic electronic computing machinery, 1948, the Rand Corporation built an electronic machine with which to generate the contents of a book titled One Million Random Digits and 100,000 Normal Deviates. The book is a large mathematical table, rigorously proven, perversely or hilariously, to contain absolutely no information -- a sort of anti-table. Instructions for its use read like a card trick. It contains a random sequence of decimal digits from an anti-oracle; the heart of it an electronic component that reliably produced noise when operated outside of it's original design parameters. Rand's machine was a difficult challenge for the time, but the "high quality" sequence of random numbers so produced, at great expense, was sorely needed for theoretical work on nuclear bombs, amongst other things.
The Rand machine's internal generator of randomness was a noise source, a type 6D4 gas-filled thyratron electron tube (valve), in a perpendicular magnetic field. The 6D4 was designed during World War Two to be a simple electronic relay, but it was found to have a second career as an excellent source of electronic noise.
As far as I can tell, the underlying phenomenon, the chaotic motion of electrons within the tube, was first described and exploited by Cobine and Curry, 1946, with practical ramifications of electrons in a magnetic field documented back to the beginning of the 20th century. How or why Cobine and Curry came upon this phenomenon and applied it to noise generation remains unknown to me after months of research. I am fairly certain that the above paper is the first publication of this as a usable source of noise, as nearly all applications of the 6D4 as noise source date after it's publication.
The technique is mentioned in an obscure Air Force/RAND newsletter (1954) and the noise-generation aspect of the 6D4 thyratron made it into the Sylvania databook, so by then it was hardly obscure.
The Gas Tube Number Generator uses the same 6D4 tube as its source of randonmess (they are/were common as dirt on eBay, I paid a buck for one). Like Cobine and Curry, I have placed the 6D4 thyratron in a cross magnetic field of approximately 375 gauss, oriented exactly as described in the article. I obtained exactly the same results, as seen on my late-20th-century oscilloscope, photo below. My machinery deviates from theirs in that I made my magnet electrical, the current for which is regulated by a small computer, so that can be influenced by human behavior, as described further on. Which, of course, is a perversion of their intent.
In 1948 the electronic state of the art was such that it was a substantial engineering feat to eke reliable non-sense from the thing; RAND's George W. Brown jests in a 1949 RAND history (document P113):
My own personal hope for the future is that we won't have to build any more random digit generators.
Things have changed such that it is now feasible in a (demented) artist's workshop.
Where my Atomic Number Generator used the natural decay of uranium ore from Grants, New Mexico for it's own particular story, the Gas Tube Number Generator explores messier human ends.
Clearly, if a sequence of numbers is truly random, it does not follow any pattern. But what is "pattern"? There are patterns that humans perceive innately, and many more patterns discernable only through aids such as mathematics or machinery. There are other criteria for determining randomness, but that's a central one for human purposes. It is certain that a truly random number sequence, that meets whatever rigorous standards you'd like to apply, shall contain no pattern.
Hold that thought for a minute.
Now the pursuit of science has been for some time moving away from external absolutes ("the sun is the center of the universe") in both practical and theoretical ways. Even matter hardly stands still; it turns out that there is no "ground" from which to measure anything, quite literally everything is measured against some (often arbitrary) standard.
That said, many "absolutes" remain, but they are all essentially relationships (ratios), not concrete "things": the ratio of circumference to radius, speed of light (though that is being assailed, it hasn't budged yet), quanta, etc. Many things that seem like "things" are made up, like an ampere of electrons, or side-effects of biology, like "red".
Back to randomness: random sequences are one of the few remaining external absolute standards. There exists the concept of "absolute zero" temperature, the utter lack of thermal energy, but I do not think anyone has any illusion that it "exists" in any concrete way. But the complete lack of information -- true randomness -- while as abstract as absolute zero temperature, still requires a physical, existent, machine to produce it, though even that isn't that simple -- for a hardware random number sequence generator used as a source for mathematical ends can't really be trusted, as is, it must be constantly tested -- "yup, still can't find any pattern in there".
So what the hell is really going on here? We (humans) make a black box to exacting standards, a thing that dynamically gets rid of any pattern, information, whatever. But this can never truly be proven, the best that can be said is "none of the patterns has been found". I'm not implying that the machine will produce an arbitrarily long un-discerned "pattern" (like the pattern of digits of pi, that visually formed a circle, in the movie CONTACT) or other quasi-religious paraphenomena.
I consider a true random number sequence generator -- such as mine, though I will likely never bother to rigorously test it -- to be different from pre-quantum "standards" apparatus (standard cells, platinum meter bars, etc), and more analogous to modern "relative standards" -- time and length measurement based upon the resonance of a particular molecule, where the accuracy is determined by our ability to measure it. But even here randomness is an inverse of measurement -- it requires believing that we can't measure anything, eg. we can't find any pattern.
I just think this is a little bit weird. Not doubting it's usefulness or correctness, just saying it's a little bit weird. So I made a weird machine.
The machine is made of long-obsolete but lovely and reliable devices; a Leeds and Northrup taut wire mirror galvanometer, the aforementioned 6D4 gas thyratron, a 6F6 "Magic Eye" tube display, a tiny D'Arsonval type meter, a Pinlight 17-segment filamentary character display (the latter two more modern than you'd think). And finally, the phosphor screen oscillograph.
Suspended mirror galvanometers are amongst the oldest electrical instruments made, dating to around 1820. One step up after that guy's frog-leg experiment. One was the output device of the first trans-Atlantic undersea cables. They can be extremely sensitive; nanoampere region. They are extremely slow (10+ seconds side to side). They are very accurate and linear. Mine's a fairly common one, 1930s to 1950s, removed from an instrument.
Basically a coil is suspended in a magnetic field (the big U shaped
horseshoe magnet) on a thin wire, a tiny mirror attached to the coil rotates in
the magnetic field induced by current through the coil, here, produced via
analogWrite() in an Arduino. The deflection of the mirror is
small; a narrowly focused light beam (in GTNG the blue laser, projected upward
from inside the box, to a small mirror on a brass stanchion visible within the
"U" of the magnet) aims the blue light onto the tiny mirror. The relatively
long (6 inches) distance to the phosphor screen provides angular
"amplification" of the dot's deflection.
Blue light is necessary to cause the strontium aluminate to fluoresce. Most bench instruments used old filament lamps and lenses to project a thin line onto ground glass with a more conventional (sic) reticle (I have one here). I wanted the more nifty phophor, which caused me no end of grief.
These are alien and lovely "information" displays; two semi-conical fluorescent screens, semi-independently controllable, "wide" to "narrow". They were used as FM radio tuning indicators and the like in high-end radio and stereo gear, 1940s through 1960s, Macintosh, Klipsh, etc. They eventually grow dim but can flicker quite rapidly.
Here, a generated binary 0 deflects the left, a 1 deflects the right. They should average equal deflections if digit balance is good.
Magic Eye tubes are a pain to interface to 5-volt microprocessors since the grids operate at -250 volts or something. I used a circuit called a current mirror built with high voltage transistors; easy. They get hot as do the various voltage-dropping resistors. They remain plentiful on eBay.
Interestingly I can find no data of any kind on these display devices. They're not that rare nor very obsolete; they are extremely rugged, daylight-visible and were used in avionics into the modern area. Likely my search term sucks.
There exist LED equivalents. I emphasized their oddness here by messing with brightness, ramping current up and down slowly to emphasize the physicality of the glowing wires.
Left alone, the Gas Tube Number Generator internally produces random binary digits at a rate of approximately 100,000 to 250,000 per second. The computer samples this stream approximately 10,000 times per second, assembling a 16-digit binary number that in effect ranges from 0 to 65535, decimal. This process remains hidden inside the machine.
Every so often -- randomly, between 20 to 30 seconds -- one of the numbers from this random stream is sampled, and used to create a symbol on the visual character display on the front of the machine. This character persists for a few seconds, and the display reverts to shuffling through randomly lit segments.
There is also a light-sensitive device on the front panel, below the white lamp. When "sufficient" change in lighting occurs (eg. from a waved hand) the machine produces a randomly selected character. Smaller changes in light, insufficient to produce a character, serve to perturb the noise tube's magnet current. This certainly affects the distribution of chaotic electrons within the tube, which are the source of the noise. If hand-wagging is continuous, the perturbations become larger, hence the magnetic field grows very large or very small, which is clearly audible in the loudspeaker on the front panel of the machine.
The machine still works as intended, but the blue/violet laser pen that lights the phosphor screen stopped producing light, even though I paid twelve whole dollars for it. Imagine my outrage. I finally got around to buying what I hope is a better quality laser diode, a claimed 120 mW. I was forced to learn about collimation and managed to dig the collimator out of the pen, fabricate a simple mount, and focus it barely good enough to light the screen, which will have to do.
Here's some notes and photos from the August 2024 work.
The machine was intended to have a plexiglas cover and I may shortly pay to have one fabricated.
Unsurprisingly the cameras in phones surpasses everything affordable in 2012 so I took photos of the machine and its innards while I was replacing the laser.
There's a lot of stuff in the box, because electron tubes are quite power hungry (heat producing); four voltages are needed 5, 6, 12, 250 volts DC. A lot of attention went into airflow design so it stays cool with two small fans.
The fundamentals here speak to many things:
matter and energy are different states of the same thing. under extreme circumstances matter can be converted directly to energy; atom(s) of one element splits into one or more atoms of other elements, and release energy. however, state changes of this kind are rare (eg. "nuclear energy").
far more common are lower-energy state-change within atoms involving mainly electrons. this is called "chemical" or "electrical" energy. these interactions are unthinkably frequent and are the basis for overwhelmingly-most change in matter, living or otherwise.
information is a state (a property) of matter; information is pattern observed or imposed upon matter. there is no information without matter, and all information transfer involves energy transfer.
patterns "emerge" -- are observed (or created) by us, beings composed of matter -- defined as information, against a background of non-information -- chaos, or noise.
"randomness" is defined by states of matter in which humans or their agents cannot detect patterns. non-patterned states of matter humans declare to be informationless.
there is no ground state, the earth beneath our feet is in motion, everything is literally, measurably, relative to everything else. there are no absolutes, anywhere, and all patterns are temporary, even yours and mine.
these things we know.
I was recently asked, 'why do all this?', the implication being that i could have simply faked it. To me, this question drags in a complex of assumptions about what art is; and questions like this are why, when asked 'what i do', i say 'i make things'. i don't generally offer to call myself an artist. i am an artist; i make art; i'm not ashamed of it, or confused, or being arch or contrary. but so many people drag in unquestioned baggage with label-words like 'artist' that i try to describe myself with verbs, instead.
Why do this? (Meaning, why make an arguably real, rigorous hardware random number generator when simulation would seem to suffice; no one would notice.) Because i wanted to know. I think to many, maybe most, people art means 'artifice', illusion, or fake. Often it is, and that's great. But there's a lack of dialog in assumptions that belies indifference or ignorance.
I make things to ask questions, not answer them, to interrogate the world. Physics and science and the philosophy thereof is what's interesting to me.
The noise generator circuit, surrounding the 6D4, is very simple; the tube is diode-wired, grid to cathode, with about 6 mA current (20K ohm, 250 volts) derived experimentally. The noise signal is taken off the resistor in the cathode (plate tied to +250V) for ground-referenced output. A high-Z resistor network, diode bounder and cap feed an LH353 op amp wired as voltage follower (impedence transformer). The LF353's output drives an LM311 comparator trimpotted for symmetry, this drives half a CD4013 flipflop for digit output.
The random digit feeds an Arduino pin, sampled in the main loop. The noise signal is too fast for interrupts with this old processor. Randomness is derived from the time between transitions of the clipped broadband noise, which is centered on 700 KHz, which is so much faster than any sampling the Arduino could do that many of the concerns of a "real" RNG are moot. This machine generates random numbers slowly, still thousands of times faster than the characters displayed. I used less rigor than would be necessary for a "real" RNG but far more than needed for my purposes.
The magnet current control was interesting, the coils are high value but low Q inductors, the current essentially DC with a 1 to 10 Hz wiggle induced from the front panel sensor. It's driven by an emitter follower that likes to oscillate (damped so it won't). Magnet current is around 80 mA, I think (poor notes).
It's hard to appreciate just how sensitive a taut wire, suspended mirror galvo is. And how slow... the galvo is driven by an Arduino analogWrite() output, 0 to 5V. The galvo itself has a one (1) ohm resistor in parallel, and 15K ohms in series. The galvo coil is something likje 300 ohms; you can't measure it, the meter current would destroy the fantastically delicate coil. Maximum current through the 1 ohm/galvo is 300 microamps, I'm too lazy to calc the current through the coil. A microamp or so for full scale deflection.
The galvo is extremely slow; drive current changes are bucked by the huge back EMF of the coil in the magnetic field, so it's about 10 seconds left to right.
Other random circuitry: a pair of current mirrors drive the eye tube grids (5V analog to 250V swing); SN74LV8153's do serial to parallel for the Pinlight display (driven with SoftwareSerial).
Here are the 2012 construction details. Things like this start with the basic idea, breadboarded proof of concept (noise, etc), an aesthetic plan worked up, then designed as build proceeds...