It was a simple 6 LED project.

Push a button, the LED's flash, then one led lights up representing the random number generated.

set a register with 00 0001
rotate left
check for overflow
reset value if overflow

Running on a AT89C4051 at 12 MHz using assembly language.

Button press interrupts generator and stores number for display after a "random" LED flashing routine (you know, for show).

The project was submitted to a lab in South Africa who tested the unit for randomness and failed their test. 30,000 manual button presses produced a propability that lamp #1 would light up 4% less then the other 5.

I talked to their math expert and he theorizes that the generator was running to fast. I should put a delay (1000 x) into the assembly code slowing the random number generator down.

Any thoughts on this scheme fixing the problem?

I used a ring counter so I could plug the value right into the serial LED driver and display the "win".

The other alternative was to set a seed of #06H and DJNZ the number. Catch it on a button press and convert it to a single bit location for display.

A little more software but I don't see the difference in performance in randomness between RL and DJNZ at effectively 12 MHz.

No time for a hardware change as the testing facility already has the board and expects a software change to fix the problem.

Does anyone have one of those fancy software simulators, and the time, to try both methods and see which one would solve this problem?