How Do i Increase Resolution with CCP Capture?

santana · Joined: 16 Oct 2003 Posts: 3

Hello Guys
I need some help with me project as it has being driving me mad all
week.
I am using Pic16F877 capture mode to measure a frequency with varies from 1Khz to 5Khz.
This works fine up to +/- 5Hz accuracy using a 20mhz Xtal.
For example my pic display say 2410 Hz and if the frequency increase to
say 2413Hz the pic displays 2415Hz the same thing happens if the frequency decreaces. ( it displays 2405Hz )
i need to get at least +/- 1Hz accuracy
i am thinking of upgrading to 18F452 at 40Mhz but this would only give me (if i am right -/+ 2.5Hz resolution)

Is there a way i can do this ?

Santana

Ttelmah · Guest

santana · Joined: 16 Oct 2003 Posts: 3

i am using 0.2 of a second, and multiplying the value by five, to get 5Hz resolution.
i thought of doing it the way you discribed , by applying a 1Hz pulse from
my RTC (DS1307) to RB0 and then use RC0 for the measurement but this
don't work with my application as the frequency i an measure would have disappeared by the time 1 second is up.
i have to make the reading within 0.2 of a second.
this is because i am measuring the frequency of a vibrating wire which is sweept and it stops vibrating after 0.3 of a second

Santana

PCM programmer · Joined: 06 Sep 2003 Posts: 21708

You could use a CCP pre-scaler of 16, and a Timer1 pre-scaler of 1.
This gives you a timer clock of 5 MHz.

Let's do an example for 5000 Hz and 5001 Hz input frequencies to
the CCP pin.

5000 Hz / 16 = 312.5 Hz
5 MHz / 312.5 Hz = 16000 clocks counted by the CCP between edges.

Now do it for 5001 Hz input:

5001 Hz / 16 = 312.5625 Hz
5 MHz / 312.5625 Hz = 15996.8 clocks (actual CCP count = 15996)

OK, now let's do the frequency calculations, based on those counts.
This is the math that you have to do in your C source code.
For the 5000 Hz example:

(5,000,000 x 16) / 16000 clocks = 5000 Hz as the calculated frequency.

For 5001 Hz:

(5,000,000 x 16) / 15996 clocks = 5001 Hz

So this works OK. But what about the 1000 Hz example ?
In that case, the number of counts will be 5 times higher, or about
80,000 counts. That exceeds the size of the Timer1, which is only
16 bits. So you have 3 choices. You can reduce the CCP pre-scaler
to 4 (that's the next available choice). Or, you could change the
Timer1 pre-scaler to 2. Or, you could extend the Timer1 resolution
to be 24 bits. I have some sample code on how to do that, here:
http://www.ccsinfo.com/forum/viewtopic.php?t=906

The problem with changing pre-scalers, is that you have to detect
when to do it, and design code for that. After looking at the problem,
I decided the most simple way was to keep the Timer1 prescaler of 1,
and just extend it to 24 bits.

santana · Joined: 16 Oct 2003 Posts: 3

Thanks for that example i would be experimenting
with this tomorrow and does this mean that i wouldn't need to upgrade to
18F452 ?

Santana

Ttelmah · Guest

SherpaDoug · Joined: 07 Sep 2003 Posts: 1640 Location: Cape Cod Mass USA

If you are measuring frequency by counting pulses, by counting an INTEGER number of pulses, you can not measure a difference less than one pulse. At 0.2 sec you can not measure a change of less than 5Hz because 5Hz is one pulse.

The only solution is to measure the PERIOD of your signal and calculate frequency. At the start of the measurement you look for the first rising edge. Then using a very fast clock you measure the time till the next rising edge. If you know how many cycles of tone there will be you may be able to count several consecutive cycles for more accuracy.
_________________
The search for better is endless. Instead simply find very good and get the job done.