LM3478 feedback adjustments

Hi Ronald,

We've received your query and will get back to you soon.

Thanks,
Anston

Question: Why is it moved to the Simple Switcher forum. Is LM3478 considered to be a Simple Switcher? I though all Simple Switchers have internal switches.

Hi Anston,

Thanks for the help.

I am affraid that there is no room on the board for a SEPIC. The LM3841 however, should fit. Can you tell me which of the mentioned improvements will help this application? Is this more or less a drop in replacement for the LM3478, or are there pitfalls I should be aware of.

Another question, probably more general. The product I am designing can be supplied by 12V and 24V. The buffer voltage can be set in a range from 12V to 40V. I tried putting some scenarios into Webench to see which values would be recommended for the feedback network. Unfortunately Webench will always optimize for that specific value. So each scenario will give a different value for the inductor and the feedback network. Unfortunately we can only use one inductor and one feedback network.
That leads me to the question: which values would best suit all scenarios?

I am not sure which values influence the feedback network. I guess it is both the switching frequency and the difference between input and output voltage. I want to fix the switching frequency and inductor. The only way to do this with WeBench seems to be to correct it for eacht scenario. However, It does seem that if I change the inductor value and switching frequency not all operating values are recalculated accordingly.

Is there a handle to which values will work best?

For the inductor a larger input voltage will require a larger inductance. On the other hand, the buffer capacitor is relatively large. Can I use the inductor value required by the lowest input voltage for the whole range, considering that the buffer will probably compensate for the extra ripple?

Which are the factors for the feedback network? I do not need the best impulse response. Actually I guess a slow impulse response is better. On the other hand the overshoot must be acceptable. I see that different operating values require vastly different values for the feedback network. What would be a good compromise? I would say the proportional component can be rather slow. The problem with the integrator is, that after an overshoot the feedback loop is broken for a while until the buffer drained to the set value. Maybe this can be corrected, as you suggested, for a bit with a capacitor in the feedback divider. But how do you find the best value? I did some experimenting a found that the switching noise is clearly visible at the feedback pin.

Hi Ronald,

We've received your request and will get back to you shortly.

Thanks,

Anston

Hi Anston,

Any news on this subject?

Ronald

Hi Ronald,

I missed replying to this. I'll take a look at this first thing tomorrow morning.

Thanks,

Anston

Hi Anston,

Thanks for your input. It is quite a lot of useful information.

There is one thing I do not yet quite understand. What will be the influence of 200K in series with 100pF if there is a 4-7nF in parallel to it? To me it seems that the feedback network will be 99% integrative like this. Or does it compensate with the current through the network rather than the voltage over the network?

Ronald

Hi Ronald,

The 4-7nF cap in parallel is optional. Use it if you see any instability or if a bode plot response shows up as a flat line indicating an absence of a pole in the frequency response.

Let me elaborate what those components do exactly. Rcomp and Ccomp create a zero to provide phase boost in the error amplifier. You need this to ensure there is enough phase margin at the crossover frequency. But if this ends up adding too much phase boost, Placing Ccomp2, now interacts with Rcomp and places a pole a little further down from the zero to improve the frequency response if it wasn't stable before. 

For your design, I see little need for Ccomp2. 

Thanks,

Anston

To follow up on this.

I made a new design. The LM3478 was replaced by an Lm3481. The inductor value was increased from 4.7uH to 22uH and the suggested compensation network was used. The output voltage is perfectly stable even when converting from 12V to 40V!

I am not sure what did the job, but I am sure it was not the compensation network. I did some experimenting with this and all variants give a stable output voltage. Maybe it was the inductor. 4.7uH is indeed a rather small value even though the buffer is relatively large. It is my guess that the LM3481 indeed gives much better performance than the LM3478.

I tested the pulse current source with pulses of 10A. When the pulse starts, the buffer starts to be drained. The LM3841 reacts immediately by recharging the buffer. There was a delay of no more than 50uS between idle and full on. This also resulted in a clicking noise in the inductor.

For this application this is undesirable because very tight regulation is not required. So I increased the comp. capacitor to 10nF and decreased the resistor to 2.7K. This results in a response after appr 1 ms and a smooth increase in power draw. Kind of a soft start. Quite useful for this application. I am quite happy with this behaviour. Of course this also causes overshoot, but I found that charging from 12V to 40V (cold start) only gave a 4V overshoot, which was corrected within the same second.  For the application this is acceptable.

There is only one slight problem. I can now see that on some occasions there is a sort of sawtooth superimposed on the compensation signal. This is not visible in the output voltage, but I think it needs correcting. So I probably need to install Ccomp2.

Update: I was partly wrong about my previous conclusion.

Increasing the inductor value definitely did a lot, but in the end the 100pf/200k compensation network gave a much more stable result.

Can you tell me some more about how you got to these values for gain and integration?
 Is there a handle to which gain values generally give the best results? The theory says that there is a certain range in which regulation is stable. But I practice I have no idea how large this range is.

I ask both because I am trying to get a better understanding of how to tune a feedback/compensation network in general, and because I am working on a change in my design that will add extra complexity.

I am working on adding regulation of the output current. The current is measured with a 100mr resistor. The current will be settable in a range of about 0.1-1A. In order to do this I set up an opamp for PI regulation. The reference is taken from an DA converter and the output is overriding the LM3481 feedback input through a diode.

My aim was to let the current feedback be more like an error signal than an actual regulation, so mainly proportional. however, as the opamp has limited bandwidth there wil be a limited bandwidth on the feedback anyway. I set the gain to 100, which should give a theoretical bandwidth of 47Khz.

When testing I found out that the LM3481 error amplifier oscillates over about the full regulation scale. The current feedback signal is clearly too slow to follow. It varies a few mv while the actual output current has a large ripple.

My first thought to solving this problem would be to make the LM3481 respond either slower or less powerful to the feedback signal. So here I am back at the beginning.
Can you tell me how to approach this? Lower gain will result in a less tight regulation, resulting in overshoot. Slower response may result in, again, overshoot. Is

there any way out of this?

Hi Ronald,

Happy to help! You are now venturing into control systems theory, explaining which is seldom easy. Hence, let me first ask you for your schematic based on which I can make further recommendations.

I read your post several times (very well described btw) to figure out if this approach will work, but I'm inclined to think it may not based on whether the following factors were taken into account:

a) You will need a reference level for the Op-Amp to compare the source signal to.

b) Make sure diode is the right polarity

c) Make sure there is sufficient Op-Amp common mode input voltage range.

d) How is the op-amp powered?

e) What happens when Output is shorted? Since Common mode range now extends to 0V.

f) Stability is incredibly difficult (since there maybe one or two op-amps in your external design that need to talk to internal compensation op-amps)

It might be easier to stabilize using the compensation pin of the IC. The values for which I generated using an in-house circuit calculator the last time. However using this approach renders the calculator ineffective. A portable version of this calculator is available through simulaton on Webench.

Hope this helps,

Thanks,

Anston