EEVBlog #1116 - The Capacitance Multiplier

[Wolfgang]

Agreed. The 1K is probably 1Ohm in reality.

[JS]

  With 1k the circuit is capable of removing about 150mVpp of garbage, as it sais, from a 12V rail, and it will change pretty linearly with the output voltage.

  In this circuit it's not easy to apply superposition, and analyze bias and signal independently. Other way you can think it would be the 15Ω resistor as part of the collector resistor and picking frome where the noise is smaller, or summing cancelling out the noise using an inverted signal.

  The point is the noise rejection capability is given from the DC and the ratio of those resistors, not from the signal itself. Then it's a matter of tweaking the gain for best cancellation and bias point for best efficency, depending on the shape of the ripple.

JS

[Cerebus]

  With 1k the circuit is capable of removing about 150mVpp of garbage, as it sais, from a 12V rail, and it will change pretty linearly with the output voltage.

Into an open circuit, yes. But with any kind of load? Remember, the LM317 is a 1 amp regulator. If it was drawn with a bijou regulator and a specified load, all well and good, but into an unspecified load with an implied maximum load of 1 amp and the shunt regulator capable of diverting at most 1mA/volt? The effectiveness is going to be very dependent on load.

[glarsson]

The text under the figure says it us for low current loads. The resistor on the LM317 output also prevents high current.

[coppercone2]

how many of these problems does using an op-amp in the capacitance circuit eliminate?

how about using a full fledged gyrator with some fast op-amps? (though I had problems with this)

and he says its for instrumentation power supplies. Nothing in instrumentation is going to use 1 amps on the 15V rail unless its something really fast or there are a TON of chips.

[coppercone2]

  With 1k the circuit is capable of removing about 150mVpp of garbage, as it sais, from a 12V rail, and it will change pretty linearly with the output voltage.

  In this circuit it's not easy to apply superposition, and analyze bias and signal independently. Other way you can think it would be the 15Ω resistor as part of the collector resistor and picking frome where the noise is smaller, or summing cancelling out the noise using an inverted signal.

  The point is the noise rejection capability is given from the DC and the ratio of those resistors, not from the signal itself. Then it's a matter of tweaking the gain for best cancellation and bias point for best efficency, depending on the shape of the ripple.

JS

Am I right in thinking that its biased by like 20uA in steady state w/10V rail?

I think its just a biased common emitter amplifier that uses the ac coupled noise as a input and uses the inverting output to cancel the signal? And its inverting output is current limited by 1k? and its output is isolated by 100k to the input so its not fighting itself

[Leo Bodnar]

Wenzel circuit uses LM317 just as an example of noisy (and cheap) LDO. 
It's intended use is powering very low noise OCXOs.  The heater is powered from a separate regulator.
Leo

Here is higher current version from the same appnote

[b_force]

What I normally do with the standard circuit is using a extra RC filter.
That gives a steeper roll of (24dB/oct) with a Q of 0.5

Another way is actually forming a sallen-key filter.

With these methods you can even get away with smaller capacitors.

[jasonhanjk]

MPC1700 has a ripple rejection of less than -10dB at the frequency of 10kHz. Should use LM317 for your video instead.

https://youtu.be/wopmEyZKnYo?t=352

[coppercone2]

the secret being unless you want to design for battery life don't use those. the mcp1700 has pretty much portable on 100% of its application suggestions

[Feynman]

Someone ever tried to tap the switching regulator's feedback signal behind a capacitance multiplier to regulate away its voltage drop? Is there any change the regulator is still stable (careful routing of the feedback signal assumed)?

[Wolfgang]

I would not be so much afraid of a routing induced problem, but a problem with loop stability.
Personally, I have never seen something like this in the wild.
Maybe some SPICE could help.

[T3sl4co1l]

You don't want to do that within the same loop, the purpose of the C-mult is to have a very low frequency pole.  The controller won't be able to react to its own changes.

There are very few situations where you need perfectly stable output voltage AND low noise, anyway.  Better to address the failings of your circuit, than to nurse it with an ever-more-complicated power supply.

Tim

[Wolfgang]

A capacitance multiplier in the loop makes a lot less sense than a postregulator.

With this, you could
- kill as much ripple as a capacitance multiplier
- have a fast load response
- increase accuracy

There are some Jim Williams Appnotes describing such designs.

[MT]

You don't want to do that within the same loop, the purpose of the C-mult is to have a very low frequency pole.
The controller won't be able to react to its own changes.

There are very few situations where you need perfectly stable output voltage AND low noise, anyway.

There are many situation where you need perfectly, very god output stability and low noise.

Better to address the failings of your circuit, than to nurse it with an ever-more-complicated power supply.

yes. Better to have a pre and post regulator and local circuitry regulators.

[VanitarNordic]

How the circuit should be in case of a negative input to the capacitance multiplier?

for example from the 7660 negative voltage generator

[Wolfgang]

Just turn the Zener around and use a PNP transistor.

[VanitarNordic]

Just turn the Zener around and use a PNP transistor.

Actually, I used NPN and the results were identical with when I used an PNP in this circuit!

Besides, if I just connect something at the output (output of the multiplier), a resistor (as a load) or even a 100nF capacitor or similar, then the noise appears again!

[AccountRemovedPerUsersRequest]

What kind of transistor characteristics I am looking in practice (for optimal/good practice)? Any suggestions in dpak?

A

[marth]

Thanks for the interesting video! If you model these circuits, a single bjt transistor circuit is best suited for this application. Better filtering and less voltage drop.

[marth]

All the same, everything is ambiguous and strongly depends on the output current. For a loaded circuit, the Darlington circuit is better.

[johofz]

I tried this circuit on a breadboard with a Mosfet and noticed that it gets significantly hot when drawing larger power. Does it come from the voltage drop across the Mosfet? It's on-resistance is 40mOhms and I drew about 0.8A at 24V input voltage. The voltage dropped to about 20V. If I calculate the power loss according to my assumption It comes to about 5W. That is huge!
Is this circuit not meant for bigger loads? Is there anything I could do to decrease power loss? I drive lots of optos from this power supply and use it to feed 3 linear regulators for +/-12V rails aswell as 5V and 3.3V rails.

[T3sl4co1l]

MOSFET drops Vgs(th) plus a bit, and the output impedance is 1/g_fs.  Both are higher than for a BJT at the same operating point, which is a big reason why this circuit is usually done with a BJT.

Tim