A bit of fun I made up when my lab was filled with hundreds of MMI generators sampling tens of trillions of bits of entropy per second.
Generator prototype, 1KHz, about 1994
First large array – 64 generators producing a total of 1.024GHz, 2003
Individual MED array, 1GHz each, 2013
256 generators producing 400GHz each, 2010
That’s some nice generator porn.
You’ve certainly been evolving the hardware over the years. Thanks for the share.
Recently ANU have set a price per request, so we dont have a source of limitless raw unamplified entropy for experiments anymore. Is there a possibility to buy from you a psi-responsive REG without amplification and with high output rate?
I believe I still have some PQ128MS generators. They are 128Mbps generators with no deterministic post processing or bias amplification. I think that matches your specified requirement exactly, but you will of course have to set up your own server(s) to provide data to your users unless this is intended strictly for in-house testing. You can email me directly to discuss any business arrangements.
@ScottWilber
I have a question about bias amplification in MED. As far as I understood from the text “Bias Amplification Algorithms for MMI”, Random Walker is used there with a fixed number of input bits. However, it’s not clear to me, if the number of bits is fixed, what makes RW different from Majority Vote? Is the deviation size somehow taken into account in such an algorithm?
There are two variations of random walk bias amplification. One uses a fixed number of bits. That method gives the same result as majority voting. If a simple 1 or 0 output is desired, MV is simpler to compute, though probabilities and z-scores are more complex to compute if the results are treated as MV. Probabilities or z-scores are often used, for example to produce a linearized result, provide a weighting factor or to combine multiple results, so the algorithm chosen depends on the specific application.
The bias amplifier in MEDs uses a variable number of steps, that is, it keeps stepping until the walker reaches a preset bound – either the positive or the negative bound. The amplification is higher for this algorithm, but the variation in the number of input bits results in a variation in the time for the walker to produce an output bit. That is not an issue when the generation rate of amplified bits is high (100Kbps for the MED100K), because a relatively large number of those bits is typically used in an application, which averages out the time to produce a result.
Thanks for the clarification. It turns out that my assumption that it uses a fixed number of input bits to achieve a uniform output speed was wrong.
I have a question about PQ128MS. Is quantum shot noise making 99% influence on its output, or does chaotic entropy has bigger impact. I am reading your paper and it says that chaotic jitter is 400 times bigger than shot noise one. But i think i havent understood this correctly.
This is a somewhat complicated question to answer. It has to do with how entropy combines when there are different types in the same source. It’s been a while, but my recollection is that the theoretical quantum entropy is about 0.95-0.97, while the chaotic entropy is extremely close to 1.0, both referred to the output bits.
To begin, know that quantum entropy and chaotic entropy are sampled simultaneously since they are both always present in the entropy source. It is not possible to separate the results of one entropy type from the other by any mathematical or other means. What is important, and perhaps a little harder to understand, is each type of entropy contributes to the lack of predictability or true randomness of the output bits. The quantum component – if it were possible to measure separately – would allow the bits to be predicted about 60% of the time, while the nominal amount is just 50% (a 50/50 chance). The chaotic entropy, by itself, would make the predictability 50% plus 7.9 x 10^-31 (that is, 50.00000000000000000000000000000079%) – immeasurably close to the nominal 50/50 chance of a perfect random number.
While the two types of entropy cannot be measured separately (without making a special design of the entropy source), if an influence, for example, a mental influence or MMI, were to affect the outcome of the measurement only through a quantum process in the entropy source, that influence would still be able to “flip” a bit to the intended state. This would be still be possible, even in the presence of the chaotic entropy in the same entropy source.
Considering more deeply, the presence of the much larger chaotic entropy, if it were not influencable by mental intention, could indeed reduce the fraction of bits (reduce the effect size) that could be changeable by mental influence by purely quantum effects.
I don’t believe that chaotic entropy, as I define it in my papers, is outside the realm of mental influence. Rather, the amount of influence or effect size is inversely related to the amount of energy it takes to flip a bit to its intended state.
For this reason, the design of MMI generators (or REGs as others may call them) can have a profound influence on the effect size in an MMI system. It just turns out that everyone (except for my MMI generators) has used designs that don’t take advantage of this relationship. Even my current generators do not take this energy relationship to its potential limit. While I have considered several designs, they are substantially more difficult to build or mass produce – at the moment. It’s an area of active effort on my part.
