Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
WhatisTrueRandomness?
Mustbeunpredictable. Foragivensetofbinarydata'i',thei+1thbitcanonlybe predictedwith50%accuracy. Mustbeunbiasedandnonalgorithmic.Computerscan'tdo this(andneithercanyou!). Usefulforcryptography,science,andgames(gamblingand drinking).
TypesofRandomNumberGenerators
Pseudorandom(PRNG):Usesanalgorithmanda'secret' initialsequence.Thisiswhatyourcomputerdoes.
TrueRandom(TRNG):Samplesaphysicalsystemofhigh entropy.
Bothareeasytodesignwrongandtheyfailsilently!
TypesofTRNG
NonQuantum:Samplesacomplexsystemofhigh entropy(lavalamps,timebetweenkeypresses). Higherbandwidth,easiertoconstruct,numbersare notproducedbyanyobviousalgorithm. Iscomplexityasgoodasrandomness?
TypesofTRNG
QuantumTRNG(QTRNG):Samplesasimplesystemof highentropy(behaviorofsinglephotonsorparticles). Lowbandwidth,difficulttosamplequantumlevel phenomenae. However,theoutput'should'betrulyrandom!
MistakestoAvoid
Donotusemorethanone entropysourceordetector. Theauthorofthispaper1 suggestsagoodmethodtoo.
MistakestoAvoid
Donotuseacounter+CPU interrupts.Thenumberof eventswithinagiventimeis notarandomdistribution,it isaPoissondistribution2. Donottrytocalculate 'expected'timebetween events,either.
OurDesign
OurDesign
APINphotodiodeandopampsareusedasasolidstate particledetectorthatoperatesatlowvoltages.Itwas enclosedinaFaradaycage.
FlexPCB
BoardswereprintedonflexPCBforitsgoodrapid prototypingcharacteristics.
OurDesign
PulseshapingisdonebyaSchmitttriggerhexinverter.
OurDesign
Samplingandparallel outputaredonebyan ATtiny261MCU@ 8Mhz.
OurDesign
Demonstration
Hopefullynomagicbluesmoke
BasicAnalysisofOutput
Checkforbiasbycreatingsimplefrequencycharts. XmeansofYrandomvaluesshouldapproachthebinomial distributionforlargeXandY3. Keepinmindthat'proving'randomnessisimpossible throughhypothesistesting.
AdvancedAnalysis
NISTRandomNumberGenerationTechnicalWorking GroupStatisticalTestSuite2.04
AdvancedAnalysis
GivenablockofrandomdatadividedintoXbitstreamsand subjectedtoYtestsatsomethresholdalphavaluefor faliureZ,youexpecttruerandomdatatofailX*Y*Ztests. Minimum1/alphabitstreamsrequired(ouralpha=0.01) Weexpect187faliures,weobserve205. Faliuresdonotclusteronanyspecifictest.
FutureTechnology!
SinglephotonQTRNGsusingsinglephotonemittersand photomultipliertubes(Ebay!) Highbandwidth,zerobias!Noisotopes!Requireshigh voltageandvacuumtubes(seriously).
Acknowledgements
Foulabforaspacetowork,andfriendstowork with. TRNGdriveranddatamanagementscripts writtenbyfx
Wanttomakeoneyourself?
Ifallgoeswell,boardswillbeavailableinone monthatwww.legionheavyindustries.com
References
1:http://isi.cbs.nl/iamamember/CD2/pdf/545.PDF 2:http://en.wikipedia.org/wiki/Poisson_process 3:http://en.wikipedia.org/wiki/Binomial_distribution 4:http://csrc.nist.gov/groups/ST/toolkit/rng/index.html
Formoreinformation:
http://www.national.com/onlineseminar/2004/photodiode/PhotodiodeAmplifers.pdf http://jp.hamamatsu.com/resources/products/ssd/pdf/s1223_series_kpin1050e01.pdf http://www.fourmilab.ch/hotbits/how3.html