Questions about pi filters

[forza764]

Hi, in psu and cpu vrm i see that there are a coil and capacitor filter in the outputs, i read somewhere that these filters are tuned to certain frecuency.

Questions:

1- If i change the cap value the filter will work ok?
2- Does the coil need to be changed to another value too?
3- What performs better, badly tuned Pi Filter or just a cap without coil?

[Th3_uN1Qu3]

You can put in larger caps with no worries. In general, going to a smaller cap is what causes issues. Look at all the blank capacitor spots on low-end motherboards. Do you think they change anything in the control circuit on the higher performance models? Nope... they don't. Just more fets, more caps and that's it. Any issues that could appear will be at very low load - and since the CPU always uses some power, this won't be a problem.

Regulated switchmode supplies cannot operate without an inductor (coil). The reference voltage in the controller is a triangle wave, also known as a ramp. The switching devices work in pulses. Pulsed current flowing thru an inductor generates a triangle wave, so now the loop can be closed. The pulses are always the same height, their width is what changes. It follows that if you just put a capacitor on the output, it would always try to charge to the peak voltage of the pulse and keep that value, so the voltage would stay the same regardless of the width of the pulse, hence the supply will have no regulation. Of course, in the real world, putting a load on the aforementioned cap will produce a voltage drop, and ripple will appear on the output, and the cap would be put under large stress and eventually fail. This is why unregulated supplies, no matter if they're fed by a switcher or a conventional mains transformer, always have huge capacitor banks.

By varying the width of a pulse train sent thru an inductor, a current ramp of variable height is obtained thru that inductor, with almost no power loss since the inductor is a reactive device and has (near) zero DC resistance. Then the capacitor averages this ramp, and a voltage regulator is obtained. This is how a buck regulator, like your CPU VRM, works: http://schmidt-walter.eit.h-da.de/sm...w_hilfe_e.html

[forza764]

If the smaller caps cause issues, why in poly-mods they change from say...3300uF to 820uF polys?

[shovenose]

Due to lower ESR, it's typically accepted that you only need half the capacitance when you're switching from electrolytic to polymner. However, this is best for motherboards; it's not a good idea in PSUs...

[PCBONEZ]

VRM's have a minimum uF needed. It's called 'bulk capacitance'.
[It's in Intel's VRM Tech Product Spec [TPS] for whichever VRM design.]
After that it's all about ESR.
A bigger can means lower ESR.
When they use lytics they over-kill bulk capacitance [uF] in order to get lower ESR.
.
VRM-high typically needs at least 1200uF [total]
Vcore typically needs at least 5500uF [total]
[Those aren't exact numbers but are typical in VRMs on poly-as-original boards.]
In both cases [VRM-high and Vcore] more uF is better because bulk capacitance helps with transients.
Cheap boards and boards for low power CPUs sometimes use less bulk capacitance than those numbers.
Servers usually use more.
.

[Longbow]

Originally Posted by forza764

Hi, in psu and cpu vrm i see that there are a coil and capacitor filter in the outputs, i read somewhere that these filters are tuned to certain frecuency.

You can look at it this way, but it is more involved than just filtering secondary supply ripple. The values of these parts are chosen specifically for a particular power supply design and are not generic in any way. The LC values affect the way the supply handles changing current demands on the output.

Questions:

1- If i change the cap value the filter will work ok?

Hard to say. The values are chosen for a reason, not just because the manufacturer wants to save a few bucks on larger capacitors.

2- Does the coil need to be changed to another value too?

see above. The design of the output filters cannot be isolated from the design of the whole supply.

3- What performs better, badly tuned Pi Filter or just a cap without coil?

see above. The term filter used in this case may not be the best choice. These components adjust time constants within the supply that are important to maintain. Each supply will have different values based on many variables: output voltage, output current requirements, slew rate requirements, slew rate capability, and so on. Simply stated, you can't just change one thing and not affect the whole. That statement holds for all physical and electrical systems.

The term "bulk capacitance" is used to differentiate the large value, high microfarad electrolytics from the <1 uF ceramic bypass capacitors which are also needed to stabilize operation at the higher frequencies. In the case of the LC values at the p.s. output, the C does refer to the large value caps.

[PCBONEZ]

Longbow is covering PSUs primarily.
I want to add some about VRMs.
.
VRMs tuned to a frequency, not really..
[That might be contrary to previous thought.]
The MOSFETs are fired by PWM which changes Pulse Width with the changing needs [load] of the CPU.
That means the Ripple Frequency will be variable with the load on the CPU.
They compensate for not being able to tune [optimize] it for a given frequency by making ESR so low that all frequencies pass to ground.
The huge total uF values seen in Vcore with lytic caps is NOT because they need that much uF in Vcore.
More caps => lower ESR
Bigger caps => lower ESR
The Total uF is large because they are using many large caps -> to lower the ESR.
... And not because that much uF is needed.
.
Polymer have much lower ESR than Lytics so they can achieve the low Total ESR without over-killing the uF in the process.
That does not mean some value of uF isn't needed in Vcore, it just means they don't have to over-kill it just to get low ESR.
.
I gave typical values for all Polymer VRM's in the earlier post.
- That should be used in place of the 'half rule' that won't die.
- The 'half rule' was invented when poly mods were young and experimental [which isn't the case anymore] and, while it generally works, there are times it will excessively limit your options or even get you into trouble.
[Example-1: If you have 9x 820uF in a Vcore (which is fairly common) and you halved it, you would have way too little total uF. - That's trouble.]
[Example-2: If you have 9x 3300uF in a Vcore (also common) and you halved it, you would limiting yourself to larger uF poly caps than you need.]
In both examples using 9x 680uF Poly [with low ESR] would be ideal even for a high end CPU. [6120uF total. - See earlier post.]
.