TMS320F280049: DCL_DF22_C1(DCL lib) VS CNTL_2P2Z(DP lib)

Jack,

The CNTL_2P2Z compensator is a fixed point implementation of what is called a "direct form 1" structure. The design incorporates a max/min clamp inside the loop to prevent windup. We sometimes want the compensator output to be clamped to a minimum of zero, for example if the output represents a PWM duty cycle, however if we do this in fixed point the compensator cannot recover quickly from positive saturation because the return path via a1 & a2 needs to see negative values. Therefore we set the lower limit of the two-sided clamp slightly negative (this is i_min), and re-clamp to the desired lower limit to zero after the compensator. This is why you see two clamps in the diagram. It's also explained in the text which follows the diagram in the DPlib user's guide.

The DCL_DF22_C1 is a "direct form 2 transposed" structure which has better numerical robustness than direct form 1 and is faaster to run since there is only one delay line. The C1 is a full implementation without clamping. If you want to implement clamping, you'll need to use the C2 and C3 functions together with a clamp between them. There is an example in the DCL (F28069_DF22) which demonstrates how this is done. The code is:

// run pre-computed controller
u2k = DCL_runDF22_C2(&controller2, ek);
v = DCL_runClamp_C1(&u2k, upperLim, lowerLim);
if (0 == v)
{
DCL_runDF22_C3(&controller2, ek, u2k);
}

Hoping this helps.

Regards,

Richard

Hi Jack,

The idea is to prevent the compensator states from "winding up", or growing too much, when the loop is saturated.  In saturation the loop is out of control, so nothing the controller does is changing anything at the output.  However the loop error will be non-zero, so the compensator will still try to correct for it.

By splitting the DF22 in this way we have the 'immediate' part which has no states; and the 'pre-computed' part which has two states.  These are respectively the C2 and C3 parts of the controller. The states are labelled x1 & x2 in the DF22 diagrams.

If the control is saturated we first test whether action by the immediate part brings it back within range.  If so, we pre-compute C3 for use next time.  If not, we don't pre-compute C3 so the states remain at their current values.  It's similar to integrator anti-windup in a PI(D) controller, except that we don't have an obvious integrator here.

Hope this makes it clearer.  Please let me know if I can help further.

Regards,

Richard

Hi Jack,

Yes the coefficients are the same between C1 and C2/C3.

You can use the SFRA to determine these coefficients, but be aware there is a notational difference between the coefficients of the DF22 and those of the older 2P2Z in DPlib.  In the latest SFRA documentation it says you can set the compensator type to DF22 so it shouldn't be an issue.

Regards,

Richard