Weird overflow behavior of character type variable

Question

I am trying to design an ON-OFF Keying decoder for an RFID reader project I've been working on. Below is the code that runs on a PIC16F887 @20MHZ. The compiler is Mikroelektronika mikroC PRO for PIC.

What I was trying to do is to toggle PORTC.1 every 200uS. The code runs as wished, but only for about 5.8 msec, as I measured with the oscilloscope. After that, PORTC.1 do not toggle.

I am guessing that somehow it is related to the overflow of the fiftymicrosand data_time variables. If we increase fiftymicrosvariable every 50 usec - actually this is what progam do, it will overflow after 255*50 microseconds which is equal to 12.75 msec.

Whenever I change the type of the variables fiftymicrosand data_time to integer, the problem diminishes.

Could you help me understand what is the reason behind this behavior?

volatile unsigned char fiftymicros=0;
volatile unsigned char fiftymicroscounter=0;
volatile unsigned char data_time=0;
volatile unsigned int milis=0;
volatile unsigned int milis_ctr=0;

void calculate_data_time()
{
     static volatile unsigned char ctr=0;
     static volatile unsigned char ind=0;
     if(fiftymicros-data_time>3)
     {
       data_time=fiftymicros;
       PORTC.F1=~PORTC.F1;
     }
}

void interrupt()
{
     if(INTCON.T0IF)
     {
      TMR0=198;
      INTCON.T0IF=0;
      fiftymicros++;
      if(++fiftymicroscounter==20) { fiftymicroscounter=0; milis++; }
     }
    if(INTCON.RBIF)
     {
      //data_time=fiftymicros;
      INTCON.RBIF=0;
     }
}

void main()
{
     ANSEL =0;
     ANSELH=0;
     TRISA=0; PORTA=0;
     TRISB=0x01; PORTB=0;
     TRISC=0; PORTC=0;
     TRISD=0; PORTD=0;
     TRISE=0; PORTE=0;
     OPTION_REG.INTEDG=1;
     OPTION_REG.T0CS=0;
     OPTION_REG.PSA=0;
     OPTION_REG.PS0=1;
     OPTION_REG.PS1=0;
     OPTION_REG.PS2=0;
     INTCON.TMR0IE=1;
     INTCON.RBIE=1;
     INTCON.GIE=1;
     UART1_Init(9600);
     Delay_ms(100);
     milis_ctr=milis;
     while(1)
     {
      calculate_data_time();
     }
}

Answer 1

If the compiler is ANSI C compliant then the two unsigned chars will be promoted to integers based on the Usual Arithmetic Conversions specified in the ANSI C standard. If data_time is greater than fiftymicros due to fiftymicros overflowing, then the result of fiftymicros - data_time will be a negative number.

With the original code, if fiftymicros = 0 and data_time = 251, the result of fiftymicros - data_time will be -251

You can force the compiler to do what you want by changing the code to this:

if ((unsigned char)(fiftymicros - data_time) > 3)
{
    ...
}

With this code, if fiftymicros = 0 and data_time = 251, the result of (unsigned char)(fiftymicros - data_time) will be 5.

Answer 2

I think it all boils down to:

 if(fiftymicros-data_time>3)
 {
   data_time=fiftymicros;
   PORTC.F1=~PORTC.F1;
 }

I don't know how that compiler handles the >3, but it may be performing a signed comparison. So, here's the scenario: Assume fiftymicros just incremented to 255, and data_time is 251 (these numbers could be slightly smaller and the result would be the same). The if would be true so data_time would be set to 255. Then interrupt() increments fiftymicros, which causes it to roll over to 0 (because fiftymicros is an unsigned char). Now, all future calculations fiftymicros-data_time will result in a negative (signed) number, which will always be less than 3.

A simple solution would be:

 if(fiftymicros>3)
 {
   fiftymicros=0;
   PORTC.F1=~PORTC.F1;
 }

If you still want data_time for other reasons, then add data_time+=fiftymicros; before fiftymicros=0;.