12 years ago.

Hi Erik, thanks a lot for your (fast) reply!

Your library FastPWM worked very well to create a custom square-wave frequency output. I got up to 48 MHz! (Though, the signal accuracy depends strongly on the frequency: for 12 / 24 / 48 MHz the frequency deviation is less than 1%. For others, e.g. 20 MHz, the deviation can be 20%.)

So far, I have a good configurable test signal. ( https://mbed.org/users/geotec/code/PwmOscillator/ )

Now, to the task of counting the signal pulses - i made two approaches:

First approach:

I connected the frequency output (pin p22) directly to the Timer2 capture pin p29 (CAP2.1) in order to count the pulses. Every rising edge on CAP2.1 created an interrupt on which a counter variable was incremented. This worked very well up to 1MHz. Above, it became increasingly inaccurate. At 2 MHz it stopped counting - seemed as if it couldn't separate the edges or interrupts anymore. I couldn't resolve this and had another idea...

Second approach:

I connected the frequency output (pin p22) directly to the Timer2 capture pin p29 (CAP2.1) and configured Timer2 to use this signal as external clock. This way, the timer count (TC) is no more incremented by the peripheral clock (PCLK, derived from main clock CCLK), but by the external square-wave frequency on pin22. This means, Timer2's "timestamp" (TC) is now the counter of the pulses on pin 29. This works quite good and up to 48 MHz! ( https://mbed.org/users/geotec/code/PulseCounter/ )

I tested this with an external frequency signal (9kHz) as reference, connected to Timer2's other capture pin p30 (CAP2.0). On rising edge, it reads the TC counts out of capture register (CR0). As far as i can see, the accuracy is quite good: the mbed counted the 48MHz-pulses between the rising edges of the 9kHz reference signal. The count was 1082120, resulting in a period of 22.54 ns. With an oscilloscope i measured a period of 22.50ns. So, in this case the deviation is less than 1%.

Hooray and thanks Erik!