RANG timing spec?

RE: RANG timing spec?

MartinBL — Fri, 15 Sep 2017 08:55:56 GMT

As mentioned in the specs "The time [t_RNG,UNI] to generate a byte cannot be guaranteed". The RNG uses Von Neumann’s debiasing algorithm which compares pairs of bits generated by the RNG and if both values in the pair are the same ('0' or '1') they are discarded. Hence, there is a probability that you will get long sequences of 0's or 1's and that the process will take a (relatively) long time.

RE: RANG timing spec?

bluebeam — Thu, 14 Sep 2017 16:36:18 GMT

Oh thanks. I didn't look the electrical spec. I'm using 52832.

The question is still about the worse case scenario. Can I say if the typical time to generate one byte is 120us, then it'll take 1.2ms to generate 10 bytes? If occasionally it takes, say 3ms to generate 1 byte then the total time will be so much different. It seems very undeterministic. How should I think about this from deisgn perspective. Particularly I need to do this in between connection interval. A few ms variation will be pretty big impact on design.

RE: RANG timing spec?

MartinBL — Thu, 14 Sep 2017 11:38:41 GMT

Hi,

What chip are you looking at? We have numbers for the nRF51 in the Product spec v3.3:

Values for the nRF52 can be found here.

Both chips also have an event register called VALRDY that you can use to get notified for every new random number written to the VALUE register.

A simple way to use it is like this:

while(NRF_RNG->EVENTS_VALRDY != 1)
{
    // Do nothing, but wait
}