50g: an interesting RAND anomaly

[DavidM]

(03-17-2018 09:17 PM)John Keith Wrote:  It is well known that the least significant bits of the output of a PRNG have a shorter period than the more significant bits. There was a lo-o-o-o-ng discussion of this issue on comp.sys.hp48 but I can't find the link at the moment.

Is there a way to sample the most significant bits or would that not fit your application?

It wouldn't be hard to do in this case, but knowing that zeros are more likely to creep into that side of the seed, I'd prefer to simply shift the seed to the right 2-3 nibbles. I tested this earlier with some success, but while that was successful at improving the results of this specific scenario, I have to wonder if it's simply trading the current problem for a different one elsewhere.

Any insight is appreciated!