SMPS input filter

[cupoftea]

Hi,
Equation 23 of the LM5149 datasheet (page 32) is incorrect for the input filter shown in
Fig 9_2 of page 31. Would you agree with this?
It totally ignores CD and CIN for a start. (bearing in mind that CF is likely to be a large 'lytic.

LM5149 datasheet

https://www.ti.com/lit/ds/symlink/lm5149.pdf?ts=1713696712373

Also, on page 32 it says...

Adding an input filter to a switching regulator modifies the control-to-output transfer function.

Would you agree, this is not accurate?, if the filters output impedance is ten times less than the SMPS Zin at all frequencies up to the x'over freq,
then there is no adjustment to the con_to_output tran func.
___---___---________________________________________________________________________________________________
Also, do you agree that the conducted EMC scan shown at bottom right of page 48 of the LM5143 datasheet (below)
(conducted EMC from 30MHz to 108MHz) will be all Common Mode emissions?

LM5143 datasheet:

https://www.ti.com/lit/ds/symlink/lm5143.pdf?ts=1713697779917&ref_url=https%253A%252F%252Fwww.ti.com%252Fproduct%252FLM5143

 

[BradtheRad]

The diagram shows a kind of Pi (π-shape) filter for the incoming DC supply. From seeing discussions I've had the impression CF is there for 'general purposes' in filtering rogue noise or momentary voltage droops. It doesn't interact in a specific manner with neighboring components. Its Farad value isn't governed by an electronic formula since its value can be increased or decreased as suits the designer.

However I do see the choke input LIN interacting with the capacitor CD which follows the choke. They create an LC 2nd-order filter which rings severely on startup. Their resonant frequency looks exactly like eq. 23 if you change CF to CD. In this sense I think you're post is correct.

My simulations show this LC resonating is so severe and continues for many cycles so as to send reverse current waveforms backward up the supply. It needs to be damped by a resistor RD as used in the diagram.

This same tendency is warned against by amplifier builders, that this LC crossover filter should not be operated without a load. The reason is because brief occurrences of the resonant frequency continue to ring, easily get out of hand and destroy components.

However when that LC arrangement is proper values, and works as intended, it has the effect of smoothing supply input current to a smooth flat waveform in my simulations. Years ago I saw a member refer to a professor wanting it implemented in their project, and I saw its purpose and its effectiveness. The CF type of filter does its own filtering job, however the choke has its effect on Ampere storage and Ampere flow, doing a further filtering job.

 

[cupoftea]

Hi...on a related subject regarding SMPS input filters........

Page 66-77 of Basso Book "switch mode power supplies" covers Input filters for SMPS.
snva538 by ti.com also covers Input filters of SMPS.

snva538

https://www.ti.com/lit/an/snva538/snva538.pdf?ts=1713952246374&ref_url=https%253A%252F%252Fwww.google.com%252F

The "DCM design guide" (A "DCM" is a type of DCDC SMPS module) by VicorPower also covers Input filters for SMPS on pages 11 to 33

DCM design guide

https://www.vicorpower.com/documents/design_guides/DG-DCM-Design-Guide-VICOR.pdf

....Near bottom of page 23 of "DCM design guide" it states, (regarding an input LC filter for an SMPS)....
QUOTE>>>> Select the cut-off frequency below the crossover frequency (20kHz) of the DCM. <<<<UNQUOTE

That is, its saying the the cut-off frequency of the SMPS LC input filter should be below the SMPS's crossover frequency.

Neither Basso, nor ti.com, nor any other reference that i can find make this stipulation. Why do
VicorPower stipulate it?

 

[BradtheRad]

There was a previous thread about this topic. One or more daily experts here stated the capacitor input filter is sufficient in most cases (for real life). Certain power sources (batteries) seem to have slight inductance built in. So usually the resistance in power supply hookups is sufficient to fulfill the R value. My simulations shows the RC integrator does a partial job.

Furthermore I find undamped LC 2nd order (exampled in your links) sets up wandering resonant up-down Ampere levels which are so severe and long-lasting that current flows backwards to a positive supply. It can seem more trouble than it's worth to include the undamped L & C. Even in simulations I've had to adjust and wait. However a damping resistance must be added even though it reduces efficiency. Our sense of efficiency discourages putting in additional resistance. Usually we wish to reduce resistance in the input wiring.

However certain power sources only deliver a certain Ampere level. Real life examples:
* weak batteries
* solar panels
* Power over ethernet.
These benefit from the ability of an LC 2nd order input filter to smooth the Ampere waveform to a flat level so it never goes above power supply specs. It does require effort to select proper L C R values.

 

[cupoftea]

Thanks..
On the same subject..The attached (LTspice and PNG's) shows a Buck converter with a feedback loop bandwidth of 10kHz
being fed by an input filter with cut
-off frequency of 15.9kHz.
The Buck is perfectly stable, with no problems even after full load to noLoad transient
and vice versa.

This demo's that the VicorPower App Note stipulation that the Input Filter fco should
be below the SMPS's Crossover frequency is inaccurate.

As such, why have they stated this?

After all, if the Buck's bandwidth is 10kHz, then it wont even be able to "see" any resonance at 15.9kHz...
...so why would this be important? The Buck's input impedance will no longer be negative above 10kHz.

Surely as long as the Buck's input impedance is greater than the input filter's output impedance (as
"seen" from the Buck"), -and this over all frequencies equal to or less than the Buck's
crossover frequency....then all will be well?
[not yet attached due to server error]

 

[cupoftea]

Hi, again on SMPS inut filters...
We are doing a Buck with 24Vin, 13Vout, 450kHz, 15A out.

The 24Vin is provided by whatever the customer connects there…we don’t know. As such our input filter is five 1210, 10uF, 50V MLCC’s, with three 470uF electrolytic capacitors. The electro’s only have 0.5A ripple rating, so we don’t want much of the ripple getting pulled from them. (There is a total of 6A ACrms ripple current)

There is a 1uH filter inductor upstream of this.

(The electro’s are for damping, and each has a 330mR resistor placed in series with it).

Obviously we always need to keep the input filter’s output impedance approx. ten times less than the buck’s Zin at frequencies less than the loop crossover frequency……and of course, have no idea what input supply the customer may connect. (it may have an LC output filter with big L and small C for all we know…and thence we would need the highly damped input filter.)

Anyway, we know that the ripple current will divide between the MLCC’s and electro’s downstream of the 1uH inductor (very little ripple gets passed the 1uH). It will divide according to the capacitor impedances. The thing is, if we measure impedance of the electro’s using a frequency analyzer…how do we know that the CLR circuit that we find, will be similar in all batchs of this electro?

EDH477M050S9SAA

https://www.mouser.co.uk/datasheet/2/447/KEM_A4049_EDH-3316662.pdf

 

[cupoftea]

Hi Please may i add the following, on the subject of big filters being needed even for high frequency SMPS...

The pointlessness of high switching frequency DCDC SMPS:

We are doing a DCDC Buck SMPS, 24Vin to 12Vout at 300W.

We could do it at say 500kHz switching frequency and so we only need a few 10’s of uF at the input filter, (with a 4x electro as a damper) and even less at the output. -So a very small solution…or is it?....

…Well, yes, if, that is, you are supplying it from a typical lab power supply with a huge output capacitor, and the load is just a resistor.

However, if you are selling it to the general PSU market, and your customers could supply it from a multitude of different sources, and may connect loads to it which may have up to several mF of input capacitance, or more…then you certainly cannot get away with such small filter capacitor banks, and , in fact, you would have been better off just choosing a lower switching frequency like 100kHz or so. -Your solution would likely be no bigger despite the lower switching frequency.

This is because in order to be able to handle the connection of large capacitance loads to the output, you need to have the output capacitor bank on the actual SMPS being relatively large. Also, at the input, maybe the source will have a big output inductor, and small capacitor downstream of that, and so your PSU will need an input filter comprising a relatively large input capacitor bank, so as to avoid input filter instability when the source gets connected.

As such, why do we always see encouragement of high switching frequency solutions.? They have more switching losses, and often , cannot be made much smaller than a lower frequency solution.

 

[RB2024]

One of you mentioned TI document SNVA538. I'm trying to go through it and have reached page 10 where it shows the optimized ratios for the multiple section filter. L1 is given in terms of L. L2 relates to L1. Ld relates to L2. C2 relates to C1. Rd relates to L1 and C1.

What does C1 relate to? How do I relate it all to a desired Fc?

I think they made a leap somewhere and I just missed it.

 

[BradtheRad]

What does C1 relate to? How do I relate it all to a desired Fc?

Select C1 value so it: 1) supplements the incoming supply during greatest current draw, then 2) receives charge during minimum current draw. Thus the cap stays at about the same voltage as the incoming supply.
Select inductor value in order to admit sufficient current.

Slower frequency needs larger value C1.

 

[RB2024]

Ah! Perseverence. I have figured out what I need to do by working all of the examples. Shame on me for not doing that sooner.