TPS75003 Buck feedback network and feedforward capacitor

(my take on a quick read through the data sheet) The minimum on/off time aspects of this control scheme mean that the feedback amplifier is not integrating (or acting as a PI controller) on the output voltage but is acting essentially as a comparator to detect the output voltage passing through the setpoint. The Cff cap is altering the shape of the ripple voltage to tweak the point in the ripple that switches the comparator. The normal lead lag compensation for PI controller tuning is not really applicable to this design. There is a big interaction between the ESR of the output cap (which determines the form of the ripple voltage) and the optimum value of the Cff capacitor. The entire effect of Cff takes place in essentially a single operation cycle as it phase shifts the switching points of the comparator with respect to the ripple voltage.

Am I stating the painfully obvious, (or am I completely off the beam) or does this overflight give you any new insight?

Right. I knew this, but I let myself get confused by looking at other buck controllers.

In this case (see TPS64002 datasheet figure 27) we are using a very low ESR ceramic capacitor that leaves very little ripple. If the output capacitor ESR is low, then Cff will have questionable benefit regarding ripple sense timing, hence the need for R1b. This is why I started to question the function of Cff around R1a in this application. It seems like Cff will act more like an integrator, slowing the rise time of the feedback pin as current flows through R1b during Ton. Does this sound right?

================================

Greg Neff VP Engineering

Greg Neff wrote: (snip)

Agreed. The filter inductor has a phase lead effect somewhat analogous to that of Cff.

Yes. If the output is well filtered at the switching frequency, connecting Cff to it is just about the same as connecting Cff to ground, but ground is more predictable.

Once you have the full 90 degrees lead produced by R1a, I would think you could tune it by moderating the effect with Cff between the feedback node and ground, rather than to the output. This gives you essentially two knobs to turn, R1a for magnitude and Cff for phase adjustment. The only advantage I can see to connecting Cff to the output is to adapt the tuning to cover unit to unit variations in output cap ESR. I would want to test any selected values for R1a and Cff with a pair of output capacitors selected as examples of high and low ESR extremes (in the final layout), to make sure the supply doesn't misbehave in some strange way over this range.