[MartinMember]

I’ve got my TrueRNGPro now and making my first test runs. The whitened modes are running fine with a very high degree of precision if I take a large enough sample.

But I don’t understand the results in raw binary mode. I implemented all the necessary data processing and bit operations to get 10 bit values from the RNGs. I get a mean value of ~477.5 which is not very close to the expected value of 511.5.

Next try was to right shift the 10 bit values by 2 bits to get 8 bit values. By that measure I tried to get rid of ADC bias which should be in the LSBs. Now my mean values are ~119 which is also not very close the the expected value of 127.5 and no quality improvement to the unshifted values.

Also I never get real low or real high values. The bound for the 10 bit values is approx. [140,820]. For the 8 bit values approx. [35,210]. The observed mean values are approximately in the middle of these ranges – so the distributions are likely Gaussian as expected, but with an noticable offset.

Furthermore most of the time the average values from RNG2 are slightly lower than the values from RNG1.

Is this what I should expect to receive in this mode?? I thoroughly tested my data processing and I don’t think there are numerical errors introduced by my code. What would be the recommended way to finally have a Gaussian distribution over 8 bits with a mean of 127.5. I can certainly do a rescaling to of the observed 10 bit range of [140,820] into the range [0,255] but maybe there is a more “correct” way?

[MartinMember]

Thanks for your answer.

2 cont.) You said the distribution is Gaussian in raw mode. This means the 8 received 10 Bit values for each RNG within one URB (16 values in total) have a Gaussian distribution considering their decimal values?

5) What exactly is meant by “the data contains a normal amount of ADC bias mostly in the LSB (least significant bit)”? Is this related to the precision of the used ADC in the TrueRNGPro, or is it related to a voltage bias? How does this affect my data in raw mode? Is the LSB not really random?

PS: I ordered my TrueRNGPro. Looking forward to it.

• This reply was modified 8 years, 2 months ago by  Martin.

[MartinMember]

And 4.: Does the raw mode increase data throughput since no more whitening is done in hardware?

[Author]

Posts