Posted on July 01, 2013 at 23:39
This is a blind implementation, as there's no code framework. Assumes TIM3 is slaved off TIM1, and Period=65535; (N-1)
uint32_t captured[6];
void TIM3_IRQHandler(void)
{
uint32_t timebase;
uint16_t timehigh1, timehigh2, timelow;
do
{
timehigh1 = TIM3->CNT; // High order, 1/65536th rate on slave
timelow = TIM1->CNT; // Low order, high rate, will change by a dozen or so ticks
timehigh2 = TIM3->CNT; // Catch rollover, if it occurred
}
while(timehigh1 != timehigh2); // Overflowed, try over, perhaps 1 additional iteration
timebase = ((uint32_t)timehigh1 <<
16
) | (uint32_t)timelow;
// Then for each channel
if (TIM_GetITStatus(TIM3, TIM_IT_CC1) == SET)
{
uint16_t
ccr
=
TIM3
->CCR1; // note use of local/static with same names to limit typos
static uint16_t ccrlast;
static uint32_t timelast;
uint32_t timedelta, ccrdelta;
/* Clear TIM3 Capture compare interrupt pending bit */
TIM_ClearITPendingBit(TIM3, TIM_IT_CC1);
timedelta = (timebase - timelast) >> 16; // error perhaps a few 100 cycles, but we truncate by 65536, so just rounding noise
timelast = timebase;
ccrdelta = (uint32_t)(ccr - ccrlast); // implicitly & 0xFFFF
ccrlast = ccr;
captured[4] = ccrdelta + (timedelta << 16); // Use an array! inherit channel 4 designation
} // sourcer32@gmail.com
// ..
}
Now there is at least one flaw in this method, if the period of the input signal is close to the 16-bit overflow it might underestimate. One way of addressing this would be to compute a rough delta, and compare that to the fine computation. If the rough estimate was around 60000 cycles more, then you have identified a cycle slip in the high order and would correct the fine computation with a += 655
