Fast PWM: Resolution in hertz

Question

I am trying to get a nice sounding varying PWM to DAC, and intend to use a Sallen Key as part of an active low pass filter.

I set the Fast PWM at a max of 255 (WGM1[0-3] = 5) and set/clear OC1A upon comparing with OCR1A (1<< COM1A1)

I also update the value of OCR1A in 64 distinct steps, changing those steps around 15'000 times a second to create a specific sinusoidal shape.

As the chip works at 16 Mega Hertz, does this mean that one cycle of PWM is 16'000'000/256= 62.5 Kilo Hertz? If not, what is its resolution

Thank you!

• Edgar Bonet: As you are directly poking into the hardware I/O registers, you have to read the datasheet. Look for the prescaler, and the bits CS1[0-2].
• Gerben: Assuming you have no prescaler, you'd indeed get around 62.5kHz. The 15kHz update rate for the sinusoidal is a bit odd, as it's 1/4.167 of the PWM update frequency. I'd probably use the timers interrupt to call a function that updates OC1A every 4th call.
• B7th: Thank you both for your answers! I read the datasheet but I guess without prescalers it's indeed 62.5kHz. The update will vary considerably as I'm trying to do the 12 tones of the scale. If I were to update OC1A every 4th PWM I'm not sure I'd be able to do a perfect period.

Answer 1

With the help of the comments, I decided to use variable PWM max values to match absolutely with the frequency. This resulted in a crispier sound that actually doesn't need the Sallen Key anymore.

For example, a low C natural, or 261.6Hz, updated 32 times for 8 full cycles of PWM give us, for a 16MHz chip, an period and amplitude of 238.

This means that having a prescaler of 8 (CS01), a period (OCR0A) AND a max potential duty cycle (ICR1) of 238, and going through 32 iterations of the sound sought for, I got the sound that I wanted and none of the wave groove that comes with a pulse duty cycle cut short.

Sure, it means less granularity when it comes to a Bb with a max duty cycle of 133, however it worked for me.