Software PWM frequency

Datman · Guest

Hi all. I am trying to design a 4 channel PWM controller, variable duty cycle, and also variable frequency from 30Hz to 2kHz. The reason for doing so is that the product will be used to drive valves from many different manufacturers, all with different PWM frequency requirements.

I have tried several approaches to get the full range i need. I'm using PIC16F690, currently with 20Mhz crystal. Firstly i have a method where an 8 bit timer just runs without using interrupt. The value of the desired PWM duty is compared with the current value in the timer buffer to produce the square output. Thus the frequency is determined by how fast the timer increments its buffer. So I am limited here to a set few frequencies as determined by the prescaler, and not the frequencies in between.

RLScott · Joined: 10 Jul 2007 Posts: 465

You could sacrifice duty-cycle resolution for an increase in PWM frequency. For example, instead of counting 0,1,2,3,...99,100 count 0,2,4,6,...,98,100. This will give you double the PWM speed.

Another solution is to using instruction times instead of a timer. Your loop would look something like this:

John P · Joined: 17 Sep 2003 Posts: 331

I'd find it easier to work in terms of elapsed time rather than frequency. And what resolution do you need for the PWM duty cycles? What I see as the major problem is that you'd need to subdivide a 2KHz rate (= 500usec/cycle) into as many segments as possible to get any reasonable choice of duty cycles at the maximum frequency.

Suppose you had a basic clock running at 20KHz, or 50usec per cycle. Could you do everything you need to do with increments of 50usec? If so, then I think the project can work. If these valves just dumb old solenoids, they shouldn't be too fussy about niceties like exact frequency, because they're so highly inductive that they even out the current all by themselves.

Datman · Guest

RLScott · Joined: 10 Jul 2007 Posts: 465

If you insist on a software PWM, then it is going to be pretty compute-intensive. Any interrupt from the SPI port is going to corrupt one cycle of the duty cycle being generated. If it is allowed to stop all PWM output whenever a new duty-cycle is commanded, then the SPI interrupt would not be a problem. But if you have to change duty-cycle by SPI command smoothly on-the-fly, then I think you are screwed.

I made a duty-cycle generator with a 12F675 running at 4 MHz. It took in an analog command instead of a SPI command, and it only needed to generate one channel of PWM. Also, I used Timer 0 interrupts and the frequency was fixed at power-up. What you are trying to do is much harder.

Are there any regularities in the 4 duty-cycles that you could take advantage of to simplify the design? Or do you really need 4 duty cycles that vary independently over the 0-100% range?

Robert Scott
Real-Time Specialties

Datman · Guest

unfortunately, 4 that vary over the full 0 - 100% range is what is required. Confused

Its a tough one, I think I may have to try and talk the client into a 30Hz - 1.2kHz spec lol. This i can achieve using interrupts. Its hard to know what most solenoids require, they are trying to cover their asses i guess, but generally the ones i have come across are lower frequency Question

John P · Joined: 17 Sep 2003 Posts: 331

It seems to me that 1.2KHz with 2% resolution leads to a basic clock rate of 60KHz. That's 16.7usec per cycle, or just over 80 instructions with a 20MHz crystal. And in that time, the interrupt has to be properly set up and ended, I mean registers saved etc. Can it work?

In another thread someone proposed not actually responding to interrupts at all, but just checking the flag in the main routine. That might be worth a look. You could poll the flag for incoming communication, too. I do that all the time when I need an incorruptible timer interrupt, and it works OK.