These are some initial thoughts for the prototype second entropy condenser, called a Zenergraph:-
The DO-41 package that the diode comes in is annoying. It is a clear envelope, and the Zener effect is very susceptible to extraneous lighting. More light, the better the diode regulates. Unfortunately for us, the noise level drops dramatically. The diode will either have to be replaced with a another type that is an opaque encapsulation, or steps taken to screen this type from all light. An idea might be to epoxy pot both the diode and the 2N3904 transistor together. No light and short leads between diode and transistor to reduce external noise pick up. Hmm.
Notice the rating of the Zener diode. Typical circuits use 5 - 10 volt Zeners. But at ReallyReallyRandom we believe that more is better, and more voltage is better still. This was the prototype. The actual Zenergraph will raise the voltage to +/-15 volts which means that we can use a 24V Zener diode with headroom to spare.
This is what was tested on a breadboard:-
A whopping 180mV peak to peak output from the collector of the 2N3904 transistor. Resistor current was measured at 5uA but the 14nA current through the Zener diode is only a SPICE estimate. 14nA is a little low to measure directly. The 2N3904's hFE was measured at 170, but this was using a digital multi meter using an unknown base current as it is not detailed in the meter manual. So the actual hFE at 5uA may not be as actually measured as hFE tend to be somewhat dependant on the collector current. Whilst SPICE reports a 14nA Zener diode current, if you do the maths, that would equate to a hFE of 357, which is odd and seems a little high for a 2N3904. Stranger things have happened though at sea and in the dark bowels of the REALLYREALLYRANDOM command centre.
To quickly confirm the quality of the random noise, a 2 minute sample was recorded via Audacity. It looks like this, which is characteristically flat as for random noise:-
The sample was normalised to a volume level of 0.8 (to account for further amplification stages that would be included) and exported as an 8 bit .wav file (to mimic readings from an 8 bit analogue to digital converter). This file was measured for randomness using the compression technique. We estimate an entropy generation rate of 31%, equating to 110 kbits/s. Dom sez "Not bad."