How to Program the TMS320F28069 ePWM Module for Space Vector Modulation?

78sys,

We offer a space vector software module as part of both controlSUITE Digital Motor Control Library (register based macro approach) and MotorWare (Object Oriented API for our InstaSPIN solutions)

install controlSUITE and look here for the files and library docs

C:\ti\controlSUITE\libs\app_libs\motor_control\math_blocks\v4.2

and there are multiple examples (anything that is doing FOC) in any of the DRV83 or HVMotor kits

C:\ti\controlSUITE\development_kits

Install MotorWare and look at

C:\ti\motorware\motorware_1_01_00_11\sw\modules\svgen

The documenation for all MotorWare components are done in Doxygen, accessed through HTML once you run MotorWare.exe

Every project in MotorWare uses the svgen (or svgen_current reconstruction) implementation.

While Control Suite had some Space Vector code, trying to extract it for just open-loop sine wave was more difficult than going back to first principles and deriving open-loop sinusoidal space vector from scratch, so that's what I did. Using  ePWM1, 2, and 3 for the 3-phase, and executing the Space Vector pwm in &epwm2_isr interrupt routine, which is ok since epwm2 syncs with ePWM1 and ePWM3. Outputs full-amplitude sine wave, as expected with Space Vector, and alot simpler code than all the Control Suite code.

Only problem is that I'm using 6 if/else's to identify the sector, and then using trig (sin()) to compute the pwm. This uses about 3200 instruction cycles in the isr, which is alot when running at 12kHz.

Any suggestions for structure to use fewer instruction cycles than the if/else progression?

svgen in MotorWare.

SVGEN_run()

static inline void SVGEN_run(SVGEN_Handle handle,const MATH_vec2 *pVab,MATH_vec3 *pT)
{

  _iq Vmax,Vmin,Vcom;
  _iq Va,Vb,Vc;
  _iq Va_tmp = -(pVab->value[0]>>1);
  _iq Vb_tmp = _IQmpy(SVGEN_SQRT3_OVER_2,pVab->value[1]);
 
  Va = pVab->value[0];  //alpha
  Vb = Va_tmp + Vb_tmp; //_IQmpy(SVGEN_SQRT3_OVER_2, Vb_tmp);  //-0.5*alpha + sqrt(3)/2 * beta;
  Vc = Va_tmp - Vb_tmp; //_IQmpy(SVGEN_SQRT3_OVER_2, Vb_tmp);  //-0.5*alpha - sqrt(3)/2 * beta;
 
  Vmax=0;
  Vmin=0;

  // find order Vmin,Vmid,Vmax
  if (Va > Vb)   
  {
    Vmax = Va;
    Vmin = Vb;
  }
  else
  {
    Vmax = Vb;
    Vmin = Va;
  }

  if (Vc > Vmax)
  {
    Vmax = Vc;
  }
  else if (Vc < Vmin)
  {
    Vmin = Vc;
  }
  
  Vcom = (Vmax+Vmin)>>1;  //0.5*(Vmax+Vmin)

  // Subtract common-mode term to achieve SV modulation
  pT->value[0] = (Vcom - Va);
  pT->value[1] = (Vcom - Vb);
  pT->value[2] = (Vcom - Vc);
 
  return;
} // end of SVGEN_run() function

Isn't that just Insta-Spin and feedback (FOC)?

The MotorWare modules and drivers can be used with non-InstaSPIN solutions (though I doubt many are).

What I posted is just general SV transform code, has no knowledge of FAST or FOC.

According to the web site, Motor Ware is (1) for Insta-Spin and (2) integrated for feedback (current = torque and speed/position):

http://www.ti.com/tool/motorware

It would be a big help if TI had generic, user-friendly SV code  with easy, step-by-step instrucitions on how to integrate and use the code in a generic application, with or without feedback.

One more question -- what's the fastest way to come up to speed on the CLA in the F28069?

I think the CLA could help out with removing control burden from the main processor.

Thanks.

Hi Chris,

Thanks for the info. The Piccolo with the CLA provides enough parallel processing power to eliminate need for FPGAs in many high-performance (high control bandwidth) embedded control applications because of the combination of its low-cost parallel processing and high-speed 12-bit ADCs (which would need interfacing to FPGAs externally, in most cases).

Went to a CLA workshop in Tampa area a couple years ago, built a simple CLA app, but need a refresher.

Thank you.