Random Periodicity Test Code for arduino (compiled in Arduino IDE)

rand-period-test.c

/*
 * Test sketch based on rand-period-test.c by @unprovable 

 * His sketch shown an anomaly with Arduino's PRNG, in which the LCG period seems to be unreliable.
 * Also the function seems to be quite slow; iterating over the stream until we get back to the beginning
 * takes 199340888ms (55.3724689 hours). It takes 1.5 minutes (approx) to generate 1M numbers.

 * So I decided to repeat the test with a textbook implementation of a standard LCG to see if we can get a
 * reliable generator with a stable period (and hopefully faster runtime) under the same conditions.

 * In my test, this script takes ~47 seconds each 1M numbers, and 103345725 ms (28.707146 hours) in total, 
 * with a period that matches the theoretical expected value of 7FFFFFFF, so if you are going to do random
 * with an arduino, even a regular standard LCG implementation is way better than what you get out of the
 * box.
 
 * NOTE: Tests performed on an Arduino Nano (Atmega328p processor) @ 16Mhz.
 */
long rand_i, rand_j, rand_k;
unsigned long i = 0, j = 0, k = 0;
long startTime;
long lastTime;
long now;


unsigned long lcg_seed = 0x50000L; // Could be any value

unsigned long rand2(){
  lcg_seed = (lcg_seed * 1103515245L + 12345L) % 2147483648L;
  return lcg_seed;
}

void setup() {
  // put your setup code here, to run once:
  rand_i = rand2();
  rand_j = rand2();
  rand_k = rand2();
  Serial.begin(57600);
  while (!Serial) {}
  Serial.print("starting num rand_i: "); Serial.println(rand_i, HEX);
  Serial.print("starting num rand_j: "); Serial.println(rand_j, HEX);
  Serial.print("starting num rand_k: "); Serial.println(rand_k, HEX);
  pinMode(LED_BUILTIN, OUTPUT);
  Serial.print("using rand_k = ");Serial.println(rand_k);
  startTime = millis();
  lastTime = startTime;
}

void loop() {
  
  // put your main code here, to run repeatedly:
  if ((i % 1000000) == 0) {
    now = millis();
    Serial.print("Up to ");
    Serial.print(i);
    Serial.print(" iterations in ");
    Serial.print((long)(now - lastTime), DEC);
    Serial.println(" ms");
    lastTime = now;
  }
  if (rand2() == rand_k) {
    now = millis();
    Serial.print("MATCH!! We got a match on "); Serial.print(rand_k); Serial.print(" in "); Serial.print(i, HEX); Serial.println(" operations.");
    //i = 0;
    while(true){
      Serial.print("MATCH!! We got a match on "); Serial.print(rand_k); Serial.print(" in "); Serial.print(i, HEX); Serial.println(" operations.");
      Serial.print("-- It took ");
      Serial.print((long)(now - startTime), DEC);
      Serial.println(" ms");
      digitalWrite(LED_BUILTIN, HIGH);
      delay(50000);
      digitalWrite(LED_BUILTIN, LOW);
      delay(500);
    }; // stop the duino so we can see what the magic number is :-P Keep printing in case the serial port goes to sleep...
  }
  i++;
}

If you are wondering how this code compares to unprovable's sketch but you don't want to run a diff: This sketch is exactly the same, but it uses a standard LCG (implemented in rand2()) instead of Arduino's random().

It also has a global time counter to know how much time it takes to run this whole thing, which is something I also added to my local copy of his sketch when I ran the test on my Arduino. That's how I know how much time it took, and was able to compare.

If you run mine and unprovable's sketch on an Arduino other than a Nano @ 16Mhz, let me know! It would be quite interesting to know what the performance is like, and what result you get, on other Arduinos, with different processors and/or clock speeds.