Optimizing SPWM code to run in one clock cycle

Projects Programming
1

I found this great article on SPWM inverters where they used a (albeit faster) FPGA system to generate a SPWM signal to convert DC into AC. I wanted to do the same thing with an Arduino, and my code (which is based off of their approach) can be found here:

// Credit for procedures in code goes to the 2008 paper
// "Digital SPWM synthesis for the design of single phase inverters" 
// By Lucien Ngalamou and Leary Myers

#define A1_PIN 5
#define A2_PIN 4
#define B1_PIN 3
#define B2_PIN 2

#define CLK_SPEED 16000000.0 //MAY NOT BE 16MHZ EXACT
#define FREQUENCY 500.0
#define PULSE_COUNT 512
#define OUTPUT_VOLTAGE_PERCENTAGE 100.0

long N = round(1/(2*FREQUENCY*PULSE_COUNT*(1/CLK_SPEED)));
long m = 0;
long D_m = 0;
long cnt = 0;

long sin_lookup[PULSE_COUNT];

void setup() {
  pin_init();
  generate_lookups();
}

void loop() {
  noInterrupts();

  if(cnt==N){
    cnt=0;
    m%=PULSE_COUNT;
    m++;
    
    D_m=sin_lookup[m];
  }
  
  if(D_m > 0){
    digitalWrite(B1_PIN, LOW);
    digitalWrite(B2_PIN, LOW);
    digitalWrite(A1_PIN, HIGH);
    digitalWrite(A2_PIN, HIGH);
    D_m--;
  }
  else{
    digitalWrite(A1_PIN, LOW);
    digitalWrite(A2_PIN, LOW);
    digitalWrite(B1_PIN, HIGH);
    digitalWrite(B2_PIN, HIGH);
  }
  
  cnt++;
}

void generate_lookups(){
  for(long i=0; i<PULSE_COUNT; i++){
    sin_lookup[i]=round((OUTPUT_VOLTAGE_PERCENTAGE/100)*N*sin((i*PI)/PULSE_COUNT));
  }
}

void pin_init(){
  pinMode(A1_PIN, OUTPUT);
  pinMode(A2_PIN, OUTPUT);
  pinMode(B1_PIN, OUTPUT);
  pinMode(B2_PIN, OUTPUT);
}

The A1, A2, B1, and B2 pins correspond to the sets of diagonals on an h bridge where I have connected a resistor and capacitor to an oscilloscope to measure the readings.

Its a nice sine wave, however the frequency is not 1:1 with what is in my code, and I suspect that this is because the code in my loop() function is not being executed in one clock cycle exactly.

How could I better optimize the section contained in loop() to run faster, other than getting a faster Arduino? (I'm currently using an Arduino mega with a 16MHz clock)

2

??

Typically, one machine instruction is executed in one clock cycle. The loop() function might be translated into dozens or hundreds of machine instructions.

digitalWrite() alone takes around 4 microseconds, if I recall correctly, or about 64 machine instructions.

Any of the timers can be used to generate PWM signals without intervention of the processor.

3

oh hey, you helped me with my mpu6050, what's going on jreminton?

anyways, the paper I was referencing used a 'reference clock' in their calculations, which every cycle was based off of said reference clock. If my code takes multiple cycles to execute what would be expected to complete in a single cycle (or close to it), it throws off my calculations. whereas the expected clock frequency is 16MHz, in actuality it might be operating closer to 8Mhz.

And while I could start testing out different clock frequencies to find which best suites my code, I think I should try optimizing it first. I think its unlikely ill be able to get it to one clock cycle exactly, but I know at the very least more could be done to get it closer to said margin.

4

Select all of your diagonals pins on the same port of controller (for example pins 6, 7, 8, 9 - al on the PORT_H of the atmega2560).

It give you ability to write to the four pins in one-two MC instruction:

#define A1_PIN 6  //PH3
#define A2_PIN 7  //PH4
#define B1_PIN 8  //PH5
#define B2_PIN 9  //PH6

DDRH |= 0x00011110;  // setup pins PH3-6 as OUTPUT

PORTH |= 0x00011000; // set PH3 PH4 as HIGH, others unchanged
PORTH &= 0x11111001; // set PH5 PH6 as LOW, others unchanged
5

As mentioned above, instead of digitalWrite(), use direct port access (look that up for the details).

Example, instead of digitalWrite(4,1) on an Arduino Uno, use PORTD |= 1<<4;

The latter executes in one machine cycle, about 64x faster.

6

This topic was automatically closed 180 days after the last reply. New replies are no longer allowed.