Animated REG Graph
It is a little difficult to envision the measurement or instrumentation of the GCP/EGG project. A Random Event Generator is an uncommon device, although all of us are directly familiar with random processes in the world, including such things as flipping coins or watching cloud formations. The following animation shows what a bit of white noise looks like; the points range unpredictably above and below an average level. Though the random source is different, this may give some insight into the type of data we gather in the GCP, in which we look for slight changes in the fundamental randomicity, correlated with global events.
This graphic animation is not drawn from actual GCP data but is generated from a very cleverly designed random source based on the unpredictable movements of globules in a Lava Lamp. On an archived page of the Lavarand website, the device is explained, and a discussion of the statistics includes several levels of analysis.
Another clever demonstration of random processes, albeit using computer algorithms, can be found at an archive of the Uncertainty Foundation website, where they show with a Java applet a
Galton’s Desk, a Pachinko-like pinball machine that gradually builds a Gaussian distribution from random events.
As for the REGs used in the EGG project, the original source of random noise is different in the three cases. The PEAR REG uses Johnson noise from a resistor, and the Mindsong uses an FET — which might be essentially equivalent to a diode type. The Orion uses a pair of simple back-biased diodes which are XORed with each other. All three are based on quantum-level-unpredictable electronic noise.
Because real-world random sources will have some tiny but real biases that would be important in the huge datasets of the GCP, we ensure that the mean will be logically forced to be unbiased. The simplest method to do this is to impose a logical exclusive-or operation which inverts some of the bits, to make the probable contributions of 1 and 0 to the bitsum exactly equal. The XOR in the PEAR device is with a simple 1010101010 alternating bitstream, while that in the Mindsong device is with a randomly ordered (but optimized) table of all bytes with equal number of 1 and 0. This also removes all first order bias, but in addition, it reduces some possible time-series structure. For the Orion generator, there is no internal compensation, so the EGG software imposes a simple alternating bitstream XOR on the Orion data.