Home/ Questions/Q 9291735
In Process

The Archive Base Latest Questions

Editorial Team
  • 0
Asked: 2026-06-18T20:43:08+00:00

I have a little code below. I use this code to output some 1s

  • 0

I have a little code below. I use this code to output some 1s and 0s (unsigned output[38]) from a GPIO of an embedded board.

My Question: the time between two output values (1, 0 or 0, 1) should be 416 microseconds as I define on clock_nanosleep below code, I also used sched_priority() for a better time resolution. However, an oscilloscope (pic below) measurement shows that the time between the two output values are 770 usec . I wonder why do I have that much inaccuracy between the signals?

PS. the board(beagleboard) has Linux 3.2.0-23-omap #36-Ubuntu Tue Apr 10 20:24:21 UTC 2012 armv7l armv7l armv7l GNU/Linux kernel, and it has 750 MHz CPU, top shows almost no CPU(~1%) and memory(~0.5%) is consumed before I run my code. I use an electronic oscilloscope which has no calibration problem. 

#include <stdio.h>
#include <stdlib.h> //exit();
#include <sched.h>
#include <time.h>

void msg_send();
struct sched_param sp;

int main(void){
      sp.sched_priority = sched_get_priority_max(SCHED_FIFO);
      sched_setscheduler(0, SCHED_FIFO, &sp);
      msg_send();
    return 0;
}

void msg_send(){
    unsigned output[38] = {0,1,0,1,0,1,0,1,0,1,0,1,0,1,0,1,0,1,0,1,0,0,1,1,0,0,1,1,0,0,1,1,0,0,1,1,0,1}; 

    FILE *fp8;
    if ((fp8 = fopen("/sys/class/gpio/export", "w")) == NULL){  //echo 139 > export
           fprintf(stderr,"Cannot open export file: line844.\n"); fclose(fp8);exit(1);
    }
    fprintf(fp8, "%d", 139);  //pin 3
    fclose(fp8);

    if ((fp8 = fopen("/sys/class/gpio/gpio139/direction", "rb+")) == NULL){
       fprintf(stderr,"Cannot open direction file - GPIO139 - line851.\n");fclose(fp8); exit(1);
    }
    fprintf(fp8, "out");
    fclose(fp8);

   if((fp8 = fopen("/sys/class/gpio/gpio139/value", "w")) == NULL) {
        fprintf(stderr,"error in openning value\n");  fclose(fp8); exit(1);
}

struct timespec req = { .tv_sec=0, .tv_nsec = 416000 }; //416 usec

/* here is the part that my question focus*/
    while(1){
        for(i=0;i<38;i++){
        rewind(fp8);
        fprintf(fp8, "%d", output[i]);
        clock_nanosleep(CLOCK_MONOTONIC ,0, &req, NULL);

        }
    }
}

enter image description here

EDIT: I have been reading for days that clock_nanosleep() or other nanosleep, usleep etc. does not guarantee the waking up on time. they usually provide to sleep the code for the defined time, but waking up the process depends on the CPU. what I found is that absolute time provides a better resolution (TIMER_ABSTIME flag). I found the same solution as Maxime suggests. however, I have a glitch on my signal when for loop is finalized. In my opinion, it is not good to any sleep functions to create a PWM or data output on an embedded platform. It is good to spend some time to learn CPU timers that platforms provide to generate the PWM or data out that has good accuracy.

Share

Leave an answer

You must login to add an answer.

Need An Account,

1 Answer

  1. Editorial Team

    I can’t figure out how a call to clock_getres() can solve your problem. In the man page, it’s said that only read the resolution of the clock.

    As Geoff said, using absolute sleeping clock should be a better solution. This can avoid the unespected timing delay from other code.

    struct timespec Time;
clock_gettime(CLOCK_REALTIME, &(Time));

while(1){
    Time.tv_nsec += 416000;
    if(Time.tv_nsec > 999999999){
        (Time.tv_sec)++;
        Time.tv_nsec -= 1000000000;
    }
    clock_nanosleep(CLOCK_REALTIME, TIMER_ABSTIME, &(Time), NULL);
    //Do something
}

    I am using this on fews programs I have for generating some regular message on ethernet network. And it’s working fine.

    • 0