Integrating feedforward, average output voltage and transient response into the voltage control loop. ARG...
This has always been a "last 10%" tweaking function of the SMPS engineer. Maximizing input regulation while providing good output regulation, especially in a multiple output voltage condition requires a lot of attention. Transient response is not only a trait of the analog compensation loop, but could be treated as a hardware override subcircuit. Getting all these circuits to play nicely together can be a great challenge.

Perhaps, I should elaborate. Traditionally, feedforword, within the analog control domain, is when a small current through a high-value resistor from the DC input voltage is summed (or subtracted) to the input to the voltage error amp. Add to that, a possible multiple output cross-sensing circuit can be the other part of the voltage feedback circuit, it can become a bit of a problem to have tightly regulated and accurate output voltage(s) with a wide input voltage range. It is more a bit of time allocation. Modeling would be difficult, Playing with the physical circuit might be more expeditious.

There are several new philosophies, within the digital domain, is to handle transient increases in output load currents may be handled by an association of new hardware design and new software triggers. Within the analog domain it may be a high-speed comparator and an override of the output PWM with extra on-time to be better than only the analog feedback loop, this is important for the lower output voltages, (MCUs and ASICs). This is a problem for the analog Si designers.
In short, as analog SMPS designers, selecting an analog controller is as involved as selecting an MCU for a best product performance.

Feedforward schemes became just a fringe operation once current-mode came along. The current-mode inherently has feedforward built into it, and probably 90% of power supplies use this for control.

Recent talk about digital control, observers, other integrated high-frequency approaches have brought back voltage mode control again, and the need to supplement it with feedforward.

Some of the very first PWM control circuits did this long ago, in the 1960's. Ramps were built with input voltage values, achieving feedforward, current-mode, and state observers before they even had a name.

Marty,

I admit to being almost entirely ignorant about "feedforward" issues.

But that may be because I've developed a SPICE-based averaged switchmode circuit model that does not, in any way, require the consideration of feedforward concepts, yet (I believe) still is exceptionally accurate. In essence, my circuit model incorporates mostly all relevant behaviors of the switchmode circuit WITHOUT the use of (what seems to me to be unnecessary and artificial) multiple frequency-domain feedback/feedforward blocks.

I believe that the switchmode circuit, including voltage feedback and/or current mode control, has its own overall properties and should not need to be described by an artificially-defined transfer function block structure. Then again, I may be wrong (except that I've demonstrated good results)!

So my question is, why are we even talking about feedforward effects?

Enlighten me! I know I'm ignorant, so you can't really insult me. Go for it! But be prepared for an honest response and possibly further questions.

I look forward to (somewhat) polite responses.

Robert, As Ray has mentioned, feedforward has mostly become antiquated in the current mode controlled world. Current-mode responds instantaneously to changes to the input voltage. In voltage-mode control, the output would decrease or increase as to the level of the input voltage and the output-only sensing loop would take time to respond to the change in operating parameters . Feed forward would sense the input change at its earliest point and include it within the compensation loop. Some of the digital control loops are voltage mode, which make more sense in a high power forward-mode environment where comparator and A/D jitter (due to slow current slopes) can cause problems in system stability.
It would not be a bad idea to include a feedforward link into your model, even though you do not use it. It would render some more flexibility into your model for variations.

Hello all, feedforward is still used, for instance in dc-dc telecom bricks where voltage mode is still widely implemented. You mainly find it in active-clamp forward topologies where the PWM control ramp is made with an RC network but the R is connected to Vin rather than a reference voltage (or in place of a fixed current source if you prefer). That way, you have a charging current that varies with Vin. Since the ramp slope is changed, you modify the PWM block gain. The idea is to compensate the Vin/Vpeak dc gain you have in the voltage-mode buck transfer function. If Vpeak in this quotient is replaced by kFF.Vin, the Vin term disappears and you are left with 1/kFF, shielding your buck topology insensitive to input variations, small-signal wise. In an average model, you replace the PWM block of gain1/Vpeak by a block 1/V(vin)*kFF where kFF is the feedforward term that depends on R and C and vin is the input voltage node.

Marty, this is a little far out but I have wondered why the users (the high power digital computer chip designers) did not add a pin or output word that would allow the power supply controller to know that it was now increasing its power demand. This would act like a feedforward load signal, which current mode control does not offer and the output voltage monitoring cannot see until the output filter has discharged at least a tiny bit. The small amount of time lost just adds to the slow response in flyback and boost converters.

Thank you all for your generous and instructive comments.

I just may harbour a misconception about feedforward in current mode control (CMC) feedback. Years ago, I believe I saw a number of papers describing averaged methods for modeling CMC feedback that seemed (at least a few of them) to imply that a model for averaged CMC control MUST include, as an inherent part of the feedback model, a "feedforward block", sometimes described explicitly in these papers, and external to the core "PWM modulator". This did not appear likely to me, as the "PWM modulator" and the basic, canonical CMC feedback seemed to be physically separate (the basic CMC circuit used inputs such as the output voltage and inductor current from the PWM modulator but did not NECESSARILY include any explicit feedforward from the input voltage).

I now get the impression that you folks are indicating that a "feedforward" block is not an INHERENT part of a simple CMC, but is, instead, an "optional" feedback block that can be used to improve circuit performance. This concept makes a lot more sense to me, and is sending me on an entirely new learning path!

I'm presently reviewing some of the older papers that may have got me thinking "in the wrong direction". Will take me a while to fully digest!

I am very pleased that each of you responders has been very polite and generous with your instructive replies! I hope this linkedin forum remains a place for such transfer of knowledge to those (like myself) who still need to learn more about this topic.

I may respond further if I figure this out better myself, BUT, I would be greatly appreciative and interested in any further informative comments re my confusion about the necessity/optionality of CMC feedforward blocks!

Robert,

Feedforward can be thought of as positive inputs (as opposed to negative feedback) into a control summing node such that various changes being fed forward (of the system inputs and outputs) cause the controlled quantity (usually the output voltage) to remain invariant in the face of changes in the system inputs and outputs (for example input voltage or output current). If feedforward is perfect, then the feedback will have almost no error to correct in spite of large changes. In practice the perturbation being fed forward may reduce the work the feedback loop has to do by a factor of 10x to 30x. Feed forward can turn easily into unstable positive feedback if excessively applied.

True output current (measured at the load side of the output capacitor) is not often available, so feedforward of that parameter is not commonly done. On the other hand, input voltage is very easy to measure and can be fed forward to control the slope of the compensation ramp to good effect, even with peak current mode control.

Robert, I believe Marty is discussing an actual power supply controller approach not a model.
True current mode controllers have an inherent instability that require compensation usually done with voltage from the ramp generator not a feedforward in the sense of input voltage variations or load current variation.

David, I also feel that measuring output current after the filter would be questionable which I why I would like a signal from the load source that is about to or just started using more current. As with most feedforward approaches they are approximate and require a feedback loop to correct for DC errors and maybe reduce dynamic errors. If the system was really fast and stable all we would need is feedback and of course a similar claim could be made for feedforward that if we precisely knew the transfer properties and they were stable over life we could compensate for them with a feedforward system and not need negative feedback but parts have initial variation and drift over life not to mention loads being at least a little different than expected and or specified.
Thus negative feedback is used to minimize error over life, but I believe Marty is questioning whether we may not improve the system by adding some limited feedforward. Allowing us to potentially kick start the controller in the correct direction before the feedback senses or could otherwise start correcting for the error.

All, We have gotten a little of track. Feed-forward is not placing positive feedback around the negative feedback error amplifier, but it is to sum in the influence of the input voltage into the positive or negative input to this amplifier.

Chris' comment is an example of placing this effect into the positive input by changing the ramp rate of the error amp. In the olden days, the TL949 brought out both amplifier inputs and a reference pin. The input offset errors were larger in those days and the effect could not be predicted consistently. So production line testing placed a resistor of varying values into either the + or - input of the error amp. Its intended effect was to not have dips or peaks in the output (after the delays of the feedback loop of voltage mode control) onto the output.
Todays digital systems can perform both current-mode, voltage mode or a combination of both. The scaling factor of the input voltage into the control algorithms is equivalent to this resistor. More parameter sensing is better for any control system, if it is cost effective or if it is digitally timely. Of course, more non-linear control is possible.
The more things change, the more they stay the same.

Marty,

I respectfully suggest that you are unnecessarily restricting your classification of feedforward to just its most common implementation. I stand by my previous comment as being accurate both regarding the broad scope of feedforward in general and in my description of its purpose and operation.

For example, take a hypothetical case of output current feedforward in a simple current mode buck regulator. Given that output current (at the load, not the inductor) is somehow magically measured (the difficulty of doing this is why it is not often done), then the value of this current may be summed into the current mode reference coming from the voltage error amplifier. If the load suddenly demands more current (say, a one amp step change), then the inductor current reference will just as suddenly go up by one amp without the voltage loop ever having to notice a dip in the output.

Feedforward cuts down errors ahead of feedback so that feedback does not have to work so hard (or at all in a perfect world).

Feedforward of load current has definitely already been explored in quite some detail. But as you say, the difficulties of sensing it rapidly and accurately make it a technique that is just not used very often.

For the RHP zero converters, you have to adjust the scaling of the sensing with duty cycle
to define the new reference of current. Not so easy to do with analog control over a wide range.

Your point is well taken on feedforward - its big advantage is in relieving the demands on the feedback. I have used it in many situations, including power factor correction where the distortion can be reduced with feedforward.