Difference between revisions of "Arduino: realtime clock test"

Revision as of 18:35, 14 February 2016

Description of test

There is only one oscillator for primary mCPU on Arduino UNO board. It is running on 16 MHz frequency and have decent stability. The questions if it is enough stable to measuer wall clock using Arduino UNO board. The best way how to get answer is to test it.

Code described in this article is measuring time starting at pre-set value. Idea is to keep this running for some time and then check difference between precise clock and clock measured by Arduino UNO.

Technical description

• Arduino UNO board contain 16MHz crystal. Timer1 is 16bit. Once configured to auto-reload at 62500 and take clock source via 1/256 pre-scaler I get interrupt exactly at one second time (16MHz / 256 = 62500 ticks per second, Timer1 is calculating from zero). Clock update routine is invoked via interrupt handler.
• To see "current" time I configured UART. Data are send on each update.
• Currently I don't know how long it will take to setup hardware and start "counting" time so i created small routine to set time. There is interrupt handler for received characters. Once this routine receive character "s" it reset time to predefined time in code.
• Reset of device is indicated by rapid blinking "L" diode. On each update diode change state (e.g. from On to Off and vice versa).
• I cut "reset-en" to ensure that microprocessor is not accidentally reset by connected computer. Ability to reset device from computer is not necessary for me as I use avr dragon in ISP mode. In future I am planing to use my own boot-loader with code update capabilities included in my code.
• I changed prescaler after first test. In beginning i used /1024 but this value is so high for fine tuning. Changing top value by "1" (one) cause approximately 4 seconds difference per day.

Source code

Here is source code of my test. Please note that this code designed only for this test. There are several parts which can result in race condition or are creating additional load to CPU. For example there is no need to update whole date/time string each second.

#define F_CPU 16000000
#define BAUD 9600

#include <avr/io.h>
#include <avr/interrupt.h>
#include <stdint.h>
#include <avr/sleep.h>
#include <util/setbaud.h>
#include <util/atomic.h>
#include <util/delay.h>

volatile struct {
        uint8_t month;
        uint8_t day;
        uint8_t hour;
        uint8_t minute;
        uint8_t second;
} time;

//                          012345678901234567890123456789
volatile uint8_t buf[30] = "2012-00-00 00:00:00          \r";
volatile uint8_t i;

ISR(USART_RX_vect) {
        uint8_t tmp = UDR0;

        if (tmp == 's') {
                // set initial time
                time.month = 8;
                time.day = 14;
                time.hour = 20;
                time.minute = 30;
                time.second = 0;
        };

}

ISR(USART_TX_vect) {
        if(i < 30) {
                UDR0 = buf[i];
                i++;
        } else {
                i = 0;
        }
}

ISR(TIMER1_COMPA_vect) {

        time.second++;
        if (time.second > 59) {
                time.second = 0;
                time.minute++;
        if (time.minute > 59) {
                time.minute = 0;
                time.hour++;
        if (time.hour > 23) {
                time.hour = 0;
                time.day++;
        if (time.day > 31) {
                time.day = 1;
                time.month++;
        if (time.month > 12) {
                time.month = 1;
        }}}}};

        buf[17] = time.second / 10 + '0';
        buf[18] = time.second % 10 + '0';

        buf[14] = time.minute / 10 + '0';
        buf[15] = time.minute % 10 + '0';

        buf[11] = time.hour / 10 + '0';
        buf[12] = time.hour % 10 + '0';

        buf[8] = time.day / 10 + '0';
        buf[9] = time.day % 10 + '0';

        buf[5] = time.month / 10 + '0';
        buf[6] = time.month % 10 + '0';

        UDR0 = ' ';
        PORTB = ( time.second & 0x01 ) ? _BV(PORTB5) : 0 ;

}

void main (void) {
        // setup serial port
        // 9600, 8 bit, 1 stop, no parity, interupt handler on RX/TX
        UBRR0H = UBRRH_VALUE;
        UBRR0L = UBRRL_VALUE;
        UCSR0B = _BV(RXCIE0) | _BV(TXCIE0) | _BV(RXEN0) | _BV(TXEN0);
        UCSR0C = _BV(UCSZ01) | _BV(UCSZ00);

        // Timer1 top value (for autoreload)
        // Note: this is two byte operation
        //       it should be secured by atomic block
        //       when accessing while interupts are enabled
        OCR1A = 62500;

        // Time1, 1/256, autoreload, interupt on compare
        TCCR1B = _BV(CS12) | _BV(WGM12);
        TIMSK1 = _BV(OCIE1A);

        //setup "L" led and blink to indicate startup
        DDRB = _BV(DDB5);
        {
          uint8_t i;
          for(i = 0; i < 10; i++){
                PORTB = _BV(PORTB5);
                _delay_ms ( 50 );
                PORTB = 0;
                _delay_ms ( 50 );
          }
        }

        // enable interupts
        sei();

        // send "start" character;
        UDR0= 's';

        // all done, wait for interupt
        for(;;) sleep_mode();
}

I'll keep ruining this code for several days to see what will be difference between "atomic" time and "my" time.

Result

After 24 hours of unattended run i get +5 seconds difference compared to stratum 1 NTP server. Apparently crystal is of by ~926 Hz (approximately 0.006 % ). For timing of microprocessor it is very good result, but not for clock. If stable frequency of crystal on Arduino UNO board is 16,000,925.93 Hz (as calculated using time offset after 24 hours of run) then it is possible to correct it by changing TOP value of Timer1 (OCR1A).

I'll use automatic time correction in my next project so for me this test end successfully. I'll create also code to correct "speed" of local clock. I think it will be interesting to compare auto corrected TOP value of Timer1 along with other Arduino UNO boards for next project.