how to read signal from magnetic RPM sensor

[merk]

I would like to connect motorcycle magnetic RPM sensor to PIC16F1827 MCU. Output signal is AC voltage in a range from 2 to 10V (measuring with multimeter, in reality signal goes from negativ to positiv voltage each time gear tooth passes the coil. I was thinking to first use a diode to rectify AC then drive this through the MOSFET and to the PIC CCP input. I would use PIC's ST input on CCP module in capture mode to measure time it takes the signal to change its state from high to low voltage. Also I would be reading hall effect speed sensor, but this sensor already has digital output. Please note the wiring diagram marked with red square.

 

[duffy]

Looks ok to me, assuming the coil never goes over 10V. You could diode-clamp it to the +12V rail if there's a possibility otherwise. Maybe add a 10μF filter cap on the 5V side, I don't trust the two .1's to do all the filtering even with just a 7-segment LED load. You are going to need to activate the weak pullups on RA0 and RA1... but you probably already know that. I would suggest doing the measurement from rising-edge to rising-edge instead of measuring the width of a positive pulse, you will get more accurate reads that way.

 

[ClydeCrashKop]

I worked on a tachometer a while ago. For low frequencies, someone recommended reading time between pulses. I tried that and found that as the frequency increases, the time between pulses shortens exponentially. Think about it, the difference between 1 RPM and 2 RPMs is 30 seconds. The difference between 10,001 RPMs and 10,002 RPMs is a whole lot less than 30 seconds.

 

[duffy]

I tried that and found that as the frequency increases, the time between pulses shortens exponentially.

Good point!

So it would probably make more sense to set a timer for 0.46875 seconds, count tach pulses during this interval, then multiply by 128 (left-shift 7x), and this will give you RPM.

 

[Mr RB]

Or maybe better, to count the uS time of a full tach pulse period, then do one simple calc;
RPM = (60000000 / period_in_uS)

 

[duffy]

then do one simple calc; RPM = (60000000 / period_in_uS)

Just check to make sure there's any room left for the code after you load the floating-point math library into that chip with 4k of memory and 384 bytes of RAM.

 

[Mr RB]

I get your point.

But it doesn't require floating point code, only an unsigned long (32bit) division. Maybe 200 words of ROM for the 32bit_unsigned_div function.

If ROM is limited, it can be done with almost no ROM by using successive division like this (a system I have used in the past);

unsigned long period_uS;
unsigned long math1;
unsigned long result;

// 32bit unsigned long division using minimal code size.
math1 = 60000000;
result = 0;
while(math1 >= period_uS)
{
  math1 -= period_uS;
  result++;
}

The whole thing ends up at about 40 to 50 ROM words or so.