Library that allows for higher resolution and speed than standard mbed PWM library using same syntax (drop-in replacement).

Dependents:   PwmOscillator FastStepDriver TLC5940 CameraTest ... more

FastPWM is a library that unlocks alot more of the potential of the mbed's PWM units than the normal PWM library. It is currently available for the LPC1768, LPC11u24, KLxxZ, K20D50M and most STM32 targets (those lacking are fairly easy to add). (Since I generally forget to update this list, if you want to know if your target is supported see if it compiles).

The two main points it allows for compared to the regular PwmOut library is clock cycle precision PWM and (automated) changing prescalers. It isn't perfect yet, but for now it will have to do ;). For those familiar with version 1, version 2 is almost completely rewritten to add more functions.


FastPWM is largely a drop-in replacement for the normal PwmOut library. All the same functions are available, with some extras.



All prescaler options are disabled for 32-bit PWM units currently, the prescaler is fixed at 1


With this function you can set the value of the prescaler. Aditionally the second argument of the constructor is used for the same to directly set it from the constructor. It returns the actual prescaler which is set. If the requested one isn't available it is always rounded up (unless it is larger than the maximum prescaler).

There are three options for this function. Any value larger than zero will simply be set. (Yes it is signed, so yes you cannot use the full 32-bit prescaler if your device supports it, I cannot imagine why you possibly would want that). If the value is zero dynamic prescaling is disabled and the current prescaler is returned. If the value is -1 dynamic prescaling is enabled and the current prescaler is returned.

So what is dynamic prescaling? This is the default option for FastPWM, don't use any prescaler option and it is enabled. To start with the negative, it adds quite some processing cycles, so changing the period takes longer. Luckily generally the PWM period is constant. The good part is that it automatically adapts the prescaler unit to give as much accuracy as possible: It gives highest accuracy for the duty-cycle, and also allows you to generate a wide range of periods. For example you can now create a LED blinking at 1Hz with FastPWM on the LPC11u24/Nucleo 16-bit PWM units. (On the KL25Z this isn't possible due to limitted value of the prescaler).

As the nice warning message above says, this is currently only implemented for 16-bit PWM units, simply because normally you won't need it for 32-bit PWM units. For those it is automatically disabled, and you cannot enable it. However for example the majority of the PWM units of the LPC11u24 can't be used to make servo signals with PwmOut, with FastPWM they can.

TL;DR, by default it uses dynamic prescaling. Unless period is changed very often just keep it on default and enjoy your larger range of possible periods and higher accuracy.



These two functions allow you to directly write the pwm period and pulsewidth in clock ticks. This is useful if you need to have very little overhead. It is dependent on which device you use, since they have different clock rates. You can get the current clock speed of your device with SystemCoreClock.


PwmOut uses floats for setting the time in seconds, and ints for milliseconds and microseconds. All three of those don't give enough accuracy to fully use the PWM units. Which is why FastPWM uses besides int for milliseconds and microseconds, it uses doubles for seconds and also for microseconds. Generally it is adviced to use these doubles, sometimes you might need to explicitly cast your variables to doubles.

Currently setting pulsewidth in microseconds with an int is a risk with some prescaler values (not on the 32-bit timers). See known-issues.

Adding other microcontrollers

Look at the other device files on how to add other microcontrollers. Functions that need to be implemented:

  • initFastPWM(): Any setups required can be done here. Must set the number of bits of the PWM unit.
  • pulsewidth_ticks( uint32_t ticks ): Set the pulsewidth in ticks
  • period_ticks( uint32_t ticks ): Set the period in ticks
  • getPeriod(): Return the period in ticks
  • setPrescaler(uint32_t reqScale): Set the prescaler. If reqScale is zero, return the current prescaler value only. Otherwise set the requested prescaler, if needed round up to first available prescaler, return the actually set prescaler. If the PWM unit is 32-bit (or for another reason), you can set dynamicPrescaler as false, disabling the dynamic prescaler.

Known Issues

  • Changing the prescaler manually does not adapt periods/pulsewidth
    • Manually re-set the period of each FastPWM object on that PWM unit, this should also set the duty cycle.
  • Changing the period of one FastPWM object does not keep the duty cycle of other PWM objects on that PWM unit constant, but the pulsewidth.
    • Manually re-set the duty cycle of other PWM objects.
  • PwmOut objects run at wrong speed when you use FastPWM
    • Don't use PwmOut objects.
  • On certain prescaler values setting period/pulsewidth in especially microsecond integers, also to lesser extend also millisecond integers, can result in wrong values.
    • The problem is that the number of clock ticks per microsecond/millisecond as integers are pre-calculated for improved speed. However if it isn't an integer number that gives an error.
    • Solution is to preferably use doubles (or ticks). On the 32-bit pwm units this is not an issue, so for them it doesn't matter.
    • I am planning to have a further look into it, but I expect it to stay an issue.

Here the TL;DR is: Preferably set the period/prescaler once at the beginning before setting the duty-cycle/pulsewidth. If that isn't possible, take into account duty cyles need to be set again. And preferably use doubles.


Some of the ideas are 'loaned' from Jochen Krapf's fork of the original FastPWM:

Wed Oct 30 03:00:00 2019 +0000
Replace new/delete with malloc/free for void *. Deleting void pointer is undefined behavior in C++.

Who changed what in which revision?

UserRevisionLine numberNew contents of line
Sissors 11:e0a8f0fcb1c9 1 #if defined(TARGET_KLXX) || defined(TARGET_K20D50M)
Sissors 11:e0a8f0fcb1c9 2
Sissors 11:e0a8f0fcb1c9 3 #include "FastPWM.h"
Sissors 11:e0a8f0fcb1c9 4
Sissors 11:e0a8f0fcb1c9 5 void FastPWM::initFastPWM( void ) {
Sissors 11:e0a8f0fcb1c9 6 bits = 16;
Sissors 11:e0a8f0fcb1c9 7 }
Sissors 11:e0a8f0fcb1c9 8
Sissors 11:e0a8f0fcb1c9 9 void FastPWM::pulsewidth_ticks( uint32_t ticks ) {
Sissors 11:e0a8f0fcb1c9 10 *(_pwm.CnV) = ticks;
Sissors 11:e0a8f0fcb1c9 11 }
Sissors 11:e0a8f0fcb1c9 12
Sissors 11:e0a8f0fcb1c9 13 void FastPWM::period_ticks( uint32_t ticks ) {
Sissors 11:e0a8f0fcb1c9 14 *(_pwm.MOD) = ticks - 1;
Sissors 11:e0a8f0fcb1c9 15 }
Sissors 11:e0a8f0fcb1c9 16
Sissors 11:e0a8f0fcb1c9 17 uint32_t FastPWM::getPeriod( void ) {
Sissors 11:e0a8f0fcb1c9 18 return *(_pwm.MOD) + 1;
Sissors 11:e0a8f0fcb1c9 19 }
Sissors 11:e0a8f0fcb1c9 20
Sissors 11:e0a8f0fcb1c9 21 uint32_t FastPWM::setPrescaler(uint32_t reqScale) {
Sissors 11:e0a8f0fcb1c9 22
Sissors 11:e0a8f0fcb1c9 23 //Yes this is ugly, yes I should feel bad about it
Sissors 11:e0a8f0fcb1c9 24 volatile uint32_t *TPM_SC = _pwm.MOD - 2;
Sissors 11:e0a8f0fcb1c9 25
Sissors 11:e0a8f0fcb1c9 26 const char prescalers[] = {1, 2, 4, 8, 16, 32, 64, 128};
Sissors 11:e0a8f0fcb1c9 27
Sissors 11:e0a8f0fcb1c9 28 //If prescaler is 0, return current one
Sissors 11:e0a8f0fcb1c9 29 if (reqScale == 0)
Sissors 11:e0a8f0fcb1c9 30 return (prescalers[(*TPM_SC) & 0x07]);
Sissors 11:e0a8f0fcb1c9 31
Sissors 11:e0a8f0fcb1c9 32 uint32_t retval = 0;
Sissors 11:e0a8f0fcb1c9 33 char bin;
Sissors 11:e0a8f0fcb1c9 34
Sissors 11:e0a8f0fcb1c9 35 for (bin = 0; bin<8; bin++) {
Sissors 11:e0a8f0fcb1c9 36 retval = prescalers[bin];
Sissors 11:e0a8f0fcb1c9 37 if (retval >= reqScale)
Sissors 11:e0a8f0fcb1c9 38 break;
Sissors 11:e0a8f0fcb1c9 39 }
Sissors 14:b30038fbba51 40 if (bin == 8)
Sissors 14:b30038fbba51 41 bin = 7;
Sissors 11:e0a8f0fcb1c9 42
Sissors 11:e0a8f0fcb1c9 43 //Clear lower 5 bits, write new value:
Sissors 11:e0a8f0fcb1c9 44 char clockbits = *TPM_SC & (3<<3);
Sissors 11:e0a8f0fcb1c9 45
Sissors 11:e0a8f0fcb1c9 46 //For some reason clearing them takes some effort
Sissors 11:e0a8f0fcb1c9 47 while ((*TPM_SC & 0x1F) != 0)
Sissors 11:e0a8f0fcb1c9 48 *TPM_SC &= ~0x1F;
Sissors 11:e0a8f0fcb1c9 49
Sissors 11:e0a8f0fcb1c9 50
Sissors 14:b30038fbba51 51 *TPM_SC = bin + clockbits;
Sissors 11:e0a8f0fcb1c9 52
Sissors 11:e0a8f0fcb1c9 53 return retval;
Sissors 11:e0a8f0fcb1c9 54 }
Sissors 11:e0a8f0fcb1c9 55 #endif