Why should I calibrate the oscillator in AVR programming

Question

I'm new to AVR programming. I found a sample code on web; for a simple USART communication with PC. I have a little doubt there.

The main loop starts like this;

void main(){
  OSCCAL_calibration(); 
  USARTinit();
  //start communicating with PC
}

I can't understand the reason for calibrating the oscillator, using OSCCAL_calibration(); function.

---

FUNCTIONS

OSCCAL_calibration() function

void OSCCAL_calibration(void){
    unsigned char calibrate = 0;
    int temp;
    unsigned char tempL;
    CLKPR = (1<<CLKPCE);
    CLKPR = (1<<CLKPS1) | (1<<CLKPS0);
    TIMSK2 = 0;
    ASSR = (1<<AS2);
    OCR2A = 200;
    TIMSK0 = 0;
    TCCR1B = (1<<CS10);
    TCCR2A = (1<<CS20);
    while((ASSR & 0x01) | (ASSR & 0x04));
    for(int i = 0; i < 10; i++)
    _delay_loop_2(30000);
    while(!calibrate){
        cli();
        TIFR1 = 0xFF;
        TIFR2 = 0xFF;
        TCNT1H = 0;
        TCNT1L = 0;
        TCNT2 = 0;
        while ( ! (TIFR2 && (1<<OCF2A)) );
        TCCR1B = 0; // stop timer1
        sei();
        if ( (TIFR1 && (1<<TOV1)) ){
            temp = 0xFFFF;
        }else{
            tempL = TCNT1L;
            temp = TCNT1H;
            temp = (temp << 8);
            temp += tempL;
        }
        if (temp > 6250){
            OSCCAL--;
        } else if (temp < 6120){
            OSCCAL++;
        }else
        calibrate = 1;
        TCCR1B = (1<<CS10);
    }
}

USARTinit() function

void USARTinit(){
    CLKPR = (1<<CLKPCE);
    CLKPR = (1<<CLKPS1);
    UBRR0H = 0;
    UBRR0L = 12;
    UCSR0A = (1<<U2X0);
    UCSR0B = (1<<RXEN0)|(1<<TXEN0)|(0<<RXCIE0)|(0<<UDRIE0);
    UCSR0C = (0<<UMSEL00)|(0<<UPM00)|(0<<USBS0)|(3<<UCSZ00)|(0<<UCPOL0);
}

---

I'm using Atmel Studio 6 and tested this with atmega2560 (actually, with my Arduino Mega). After a bit of changes, I could make it work. But it still works without the calibration function..

I'll itemize my questions as below.

1. What do you really do as calibrating the oscillator?
2. Is it a must?
3. Is there a similar function in PIC micro-controllers? (I'm a bit experienced in PIC programming. But never knew about something like that)

Also got a little doubt;

Why do you set a clock pre-scalar in USARTinit() function before setting the baud rate? can't you set the baud rate without a pre-scalar (which means, pre-scalar = 1)

Is it to save power or something? But i tried with pre-scalar=1, it seems not working (still trying). Yeah i've calculated the baudrate properly (using the given equation in datasheet).

Answer 1

If you are doing any timing related communications outside the microcontroller (serial, pushing spi to limits, etc) or keeping time or whatever then you need a more accurate clock.

It is not really about power, marginally perhaps, if the clock is a little slow then you use more power if a little fast then you save a little power.

Many but not all microcontrollers offer an internal R/C oscillator so that you dont need to have an external oscillator (extra components, extra cost). This is not one family vs another (avr, msp430, pic, etc) some chips within a family have internal oscillators some dont. The PIC's I used back in the day required an external, dont know the family in that detail today. How the calibration happens also varies from family to family.

Answer 2

I know nothing about this particular hardware but a quick Google showed up this datasheet

To quote:

The majority of the present AVR microcontrollers offer the possibility to run from an internal RC oscillator. The internal RC oscillator frequency can in most AVRs be calibrated to within +/-1% of the frequency specified in the datasheet for the device. This feature offers great flexibility and significant cost savings compared to using an external oscillator. The calibration performed in the Atmel factory is made at a fixed operating voltage and temperature (25°C, typically 5V). As the frequency of the internal RC oscillator is affected by both operating voltage and temperature, it may be desired to perform a secondary calibration, which matches the specific application environment. This secondary calibration can be performed to gain higher accuracy than the standard calibration offers, to match a specific operating voltage or temperature, or even to tune the oscillator to a different frequency.

Answer 3

The need for calibration depends on your actual hardware:

1. If your microcontroler uses a Xtal to generate its clock you don't have to calibrate anything, only choose the good divider, depending on your Xtal oscillation frequency.

2. If you only use an RC oscillator, then since it can have more important value tolerance you might want to calibrate the frequency, especially when using higher baud rates (like 19200 or more).

There is pre-scaller that you must set depending on your oscillator frequency, see the datasheet for more details.