Low voltage drop capacitance multiplier

[John Devereux]

Hi,

The "capacitance multiplier" circuit can be used for protecting a
sensitive circuit from noise on the supply rail:

Vin _______ _________ Vout
| \_^
R |
'-------|
C
|
--------------------

This circuit will drop 1V or so. Is there something with lower voltage
drop?

A normal LDO voltage regulator requires a specific Vin value. This is
not ideal since if Vin gets too low, it will go out of regulation and
(presumably) start to let noise through.

What I want is a circuit that will track Vin, with only a couple of
hundred mV drop, suppressing anything above a kHz or so.

Perhaps there is a way to hack a normal LDO to do this?

 

[Fred Bloggs]

Hi,

The "capacitance multiplier" circuit can be used for protecting a
sensitive circuit from noise on the supply rail:

Vin _______ _________ Vout
| \_^
R |
'-------|
C
|
--------------------

This circuit will drop 1V or so. Is there something with lower voltage
drop?

A normal LDO voltage regulator requires a specific Vin value. This is
not ideal since if Vin gets too low, it will go out of regulation and
(presumably) start to let noise through.

What I want is a circuit that will track Vin, with only a couple of
hundred mV drop, suppressing anything above a kHz or so.

Perhaps there is a way to hack a normal LDO to do this?

The newer LDOs from National are unconditionally stable with arbitrarily
large output capacitors and any ESR and 10's mV I/O differential, check
them out.

 

[John Devereux]

The newer LDOs from National are unconditionally stable with
arbitrarily large output capacitors and any ESR and 10's mV I/O
differential, check them out.

Hi Fred,

I'll take a look.

But, do they maintain line regulation during dropout (for kHz+
frequencies), in the way that the "capacitance multiplier" does?

 

[Fred Bloggs]

Hi Fred,

I'll take a look.

But, do they maintain line regulation during dropout (for kHz+
frequencies), in the way that the "capacitance multiplier" does?

Dropout means loop is saturated so there is no control and all you have
left for filtering is RDS,ON x C. But if ESR is milli-ohm and RDS,on is
ohms, then something like 40dB attenuation of KHz ripple should be
obtainable.

 

[Fred Bloggs]

Dropout means loop is saturated so there is no control and all you have
left for filtering is RDS,ON x C. But if ESR is milli-ohm and RDS,on is
ohms, then something like 40dB attenuation of KHz ripple should be
obtainable.

Eh- that will only work for the PFET types, the NFET ones charge-pump
the gate drive to achieve ultra-low RDS,ON.

 

[Phil Hobbs]

The newer LDOs from National are unconditionally stable with arbitrarily
large output capacitors and any ESR and 10's mV I/O differential, check
them out.

Yeah, but they still rely on loop gain to suppress ripple, which is
slow, whereas the cap multiplier relies on low C_ce and high Early
voltage, both of which are fast. You can easily get > 100 dB ripple
suppression in the tens of kilohertz with a two-pole cap multiplier made
from an MMBT3904, which is really tough to do with feedback. For lower
dropouts, it's possible to do transformer tricks, or to use a power op
amp--google for the Kanner Kap. Still not as good as a cap multiplier.

If by chance you have a negative supply, you can also
capacitance-multiply ground instead of Vin, and adjust the offset
voltage anywhere you like. The disadvantage is that this spoils the
commonality of ground, but it's sometimes worthwhile.

Alternatively, if you can get at a higher (unregulated) supply voltage,
you can filter that separately and add a current source from there to
the base of the pass transistor to pull it up a bit--maybe you can get
Vin-Vout down to 300 mV this way without losing all your beta, at least
if the load current is smallish.

Cheers,

Phil Hobbs

 

[Winfield Hill]

John Devereux wrote...

The "capacitance multiplier" circuit can be used for protecting
a sensitive circuit from noise on the supply rail:

Vin _______ _________ Vout
| \_^
R |
'-------|
C
|
--------------------

This circuit will drop 1V or so. Is there something with lower
voltage drop?

A normal LDO voltage regulator requires a specific Vin value.

Right, an LDO is not suitable for you.

What I want is a circuit that will track Vin, with only a couple
of hundred mV drop, suppressing anything above a kHz or so.

You have two choices, neither one very attractive. You can use
a PNP or PMOS pass element with an active-regulator circuit that
tracks the average value of Vin, but that's a painful circuit to
make and leaves you with a brute-force compensated output. Or you
can modify your circuit above to drive the NPN base from a voltage
that's 400mV or so above Vin. To get this voltage you'll need to
add a simple dc-dc converter with its output stacked on Vin, and
appropriately regulate its output. For myself, often in this very
situation, I settle for the modest drop across the pass element,
and design the follow-on circuitry appropriately.

One way to improve the attenuation-vs-voltage-drop tradeoff is to
make the filter active by splitting the base resistor and bypassing
it from the output. The resulting 12dB/octave cutoff slope allows
you to use lower-value resistors (less base-current voltage drop)
and still get improved 120Hz and high-frequency noise attenuation.

.. ,---||----------,
.. | |
.. Vin ---+------ | ---- C E ---+------ out
.. | | B
.. | | |
.. '-/\/\--+--/\/\--+---||--- gnd

An issue not always considered in these circuits is, what happens
to the transistor's dissipation in the event of a short circuit?
Unless Vin current limits at a fairly low current, the transistor
may be exposed to a damaging power-dissipation level. I deal with
this issue by adding a collector resistor, like a small 3W WW type
power resistor. The resistor is chosen for less than 400mV drop
at the maximum operating current. The tradeoffs in selecting this
protection resistor reveal one more problem to solve in any attempt
to design such a circuit with a voltage drop under about 700mV.

One last comment. If this type of filter is used directly after
the 60Hz rectifier storage filter capacitor, where the ripple may
be 500mV or more, the 0.5 Vbe average drop won't be high enough to
allow proper filtering. In such a case, using a logic-level power
MOSFET pass element not only gets you the extra voltage you need,
but allows using much higher filter-resistor values, and eliminates
the need for an awkward current-limit resistor, because the MOSFET
delivers high output currents without excessive base-resistor drop.

Following a MOSFET filter stage with a three-terminal regulator is
one way to get a very quiet high-current regulated power supply,
with micro-volt ripple levels.

(One awkward issue you'll encounter is understanding the MOSFET's
subthreshold region of operation to predict the Vgs value. Sadly,
this issue won't be dealt with in the FET's datasheet. Moreover,
available Spice models won't show the correct Vgs value either.
But we do discuss the theory and give you guidance in AoE.)

 

[John Larkin]

Hi,

The "capacitance multiplier" circuit can be used for protecting a
sensitive circuit from noise on the supply rail:

Vin _______ _________ Vout
| \_^
R |
'-------|
C
|
--------------------

This circuit will drop 1V or so. Is there something with lower voltage
drop?

A normal LDO voltage regulator requires a specific Vin value. This is
not ideal since if Vin gets too low, it will go out of regulation and
(presumably) start to let noise through.

What I want is a circuit that will track Vin, with only a couple of
hundred mV drop, suppressing anything above a kHz or so.

Perhaps there is a way to hack a normal LDO to do this?

How about using an inductor?

John

 

[John Devereux]

Yeah, but they still rely on loop gain to suppress ripple, which is
slow, whereas the cap multiplier relies on low C_ce and high Early
voltage, both of which are fast. You can easily get > 100 dB ripple
suppression in the tens of kilohertz with a two-pole cap multiplier
made from an MMBT3904, which is really tough to do with feedback. For
lower dropouts, it's possible to do transformer tricks, or to use a
power op amp--google for the Kanner Kap. Still not as good as a cap
multiplier.

That is one weird circuit. It's certainly low dropout!
..
..
.. In --------------------------- Out
.. | | |
.. | |\ |
.. '-CAP---| --CAP-'
.. |/
.. | Gain -A
.. 0V ----------------------------
..

I cannot quite get my head around that, I will have to fire up the
simulator.

If by chance you have a negative supply, you can also
capacitance-multiply ground instead of Vin, and adjust the offset
voltage anywhere you like. The disadvantage is that this spoils the
commonality of ground, but it's sometimes worthwhile.

No negative supply I'm afraid.

Alternatively, if you can get at a higher (unregulated) supply
voltage, you can filter that separately and add a current source from
there to the base of the pass transistor to pull it up a bit--maybe
you can get Vin-Vout down to 300 mV this way without losing all your
beta, at least if the load current is smallish.

I can't really do this either.

Cheers,

Phil Hobbs

The man himself... Of course it was your book I got the basic circuit
from in the first place.

 

[John Devereux]

How about using an inductor?

It would need to be too big, I think. Similarly with just using a BFC.

 

[John Larkin]

That is one weird circuit. It's certainly low dropout!
.
.
. In --------------------------- Out
. | | |
. | |\ |
. '-CAP---| --CAP-'
. |/
. | Gain -A
. 0V ----------------------------
.

I cannot quite get my head around that, I will have to fire up the
simulator.

That's sometimes used in high-voltage power supplies to get low
ripple, where active components wouldn't be practical. But the amp
can't be powered directly from the rail it's filtering; it can be
powered by a secondary rail that's RC filtered from the main rail!

Since this is a virtual capacitor, it must be able to store enough
energy to accomplist the filtering job. That requirement limits what
such tricks can accomplish.

John

 

[John Devereux]

John Devereux wrote...

Right, an LDO is not suitable for you.


You have two choices, neither one very attractive. You can use
a PNP or PMOS pass element with an active-regulator circuit that
tracks the average value of Vin, but that's a painful circuit to
make and leaves you with a brute-force compensated output. Or you
can modify your circuit above to drive the NPN base from a voltage
that's 400mV or so above Vin. To get this voltage you'll need to
add a simple dc-dc converter with its output stacked on Vin, and
appropriately regulate its output. For myself, often in this very
situation, I settle for the modest drop across the pass element,
and design the follow-on circuitry appropriately.

OK, Thanks. For me also, it does look like it will be simpler to
design the follow on circuitry to accept the lower voltage.

One way to improve the attenuation-vs-voltage-drop tradeoff is to
make the filter active by splitting the base resistor and bypassing
it from the output. The resulting 12dB/octave cutoff slope allows
you to use lower-value resistors (less base-current voltage drop)
and still get improved 120Hz and high-frequency noise attenuation.

. ,---||----------,
. | |
. Vin ---+------ | ---- C E ---+------ out
. | | B
. | | |
. '-/\/\--+--/\/\--+---||--- gnd

I saw that version in Phils book, too! It still leaves you with about
a 0.8V drop though.

<SNIP good stuff>

What do you think about that "Kanner Kap" circuit!?

 

[Winfield Hill]

John Devereux wrote...

What do you think about that "Kanner Kap" circuit!?

I've used that scheme before, it's useful. But having thought
of it myself, I had no idea it had been patented, US 4,710,861.
Now I'm trying to remember if I used this before 1986. Ahem,
when did the US patent life change from 17 to 20 years?

 

[John Devereux]

John Devereux wrote...

I've used that scheme before, it's useful. But having thought
of it myself, I had no idea it had been patented, US 4,710,861.
Now I'm trying to remember if I used this before 1986. Ahem,
when did the US patent life change from 17 to 20 years?

I do remember seeing a similar things for high voltage power supply
stabilisation, as John Larkin reminded me. (In AoE, IIRC). Also, I
have used an active filter chip that worked similarly.

But it is the idea of having the input, output and supply all being
the same wire that is making my brain hurt!

<http://www.edn.com/archives/1996/060696/graphs/12df4fg2.htm>

Taken from

<http://www.edn.com/archives/1996/060696/12df4.htm>

 

[Phil Hobbs]

One way to improve the attenuation-vs-voltage-drop tradeoff is to
make the filter active by splitting the base resistor and bypassing
it from the output. The resulting 12dB/octave cutoff slope allows
you to use lower-value resistors (less base-current voltage drop)
and still get improved 120Hz and high-frequency noise attenuation.

. ,---||----------,
. | |
. Vin ---+------ | ---- C E ---+------ out
. | | B
. | | |
. '-/\/\--+--/\/\--+---||--- gnd

The Sallen-Key trick gives you a much sharper knee, but because of the
sneak path through the cap to the output, it limits your ultimate
attenuation. It's usually better just to split R, as you did, but
return both caps to ground.

Cheers,

Phil Hobbs

 

[Winfield Hill]

Phil Hobbs wrote...

The Sallen-Key trick gives you a much sharper knee, but because of
the sneak path through the cap to the output, it limits your ultimate
attenuation. It's usually better just to split R, as you did, but
return both caps to ground.

.. Vin ---+------------- C E ------- out
.. | B
.. | |
.. '-/\/\--+--/\/\--+---||--- gnd
.. |
.. '----||--- gnd

The optimum configuration to use would appear to depend on the goal.
A spice trial with two 430-ohm resistors, two 22uF caps, a 2n4401,
and a 500-ohm load (32mA at 16V) shows the two in a dead heat 35dB
down at 120Hz, the Sallen-Key form winning at 360Hz (66dB compared
to 54dB), and the both-caps-to-ground form winning above 500Hz
(52dB compared to 70dB at high frequencies - the latter assumes the
cap's esr is less than 1 ohm above say 3kHz). Using a 5k, 3mA load
erased the Sallen-Key advantage over two-caps-to-ground above 170Hz
(the Sallen-Key was still better by 13dB at 120Hz, where it enjoyed
a resonant dip). Above 10kHz the two-caps-to-ground form won by an
impressive 35 dB.

In summary, the Sallen-Key appears superior for hum reduction tasks
if properly designed, and two-caps-to-ground appears superior for
high-frequency noise reduction. And either form dramatically beats
using the same total capacitance with one twice-value resistor,
typically by 20dB in the critical line-frequency-harmonics region.

 

[Phil Hobbs]

In summary, the Sallen-Key appears superior for hum reduction tasks
if properly designed, and two-caps-to-ground appears superior for
high-frequency noise reduction. And either form dramatically beats
using the same total capacitance with one twice-value resistor,
typically by 20dB in the critical line-frequency-harmonics region.

Yes, that's more or less what I've found too. I'm usually much more
worried about tens-of-kilohertz crap from SMPS ripple.

Cheers,

Phil Hobbs

 

[Winfield Hill]

Winfield Hill wrote...

Phil Hobbs wrote...

The Sallen-Key trick gives you a much sharper knee, but because of
the sneak path through the cap to the output, it limits your ultimate
attenuation. It's usually better just to split R, as you did, but
return both caps to ground.

. Vin ---+------------- C E ------- out
. | B
. | |
. '-/\/\--+--/\/\--+---||--- gnd
. |
. '----||--- gnd

[ snip ]
Above 10kHz the two-caps-to-ground form won by an impressive 35 dB.

In summary, the Sallen-Key appears superior for hum reduction tasks
if properly designed, and two-caps-to-ground appears superior for
high-frequency noise reduction. And either form dramatically beats
using the same total capacitance with one twice-value resistor,
typically by 20dB in the critical line-frequency-harmonics region.

Substituting a MOSFET for the BJT (admittedly with a poor model),
using the same values (430 ohms and 22uF), the region of Sallen-Key
superiority shrinks, and doesn't occur at a useful frequency unless
the values are carefully chosen. The two-caps-to-ground form wins
by 35 to 45dB above 1kHz -- a nice result. But of course, when using
a MOSFET one wouldn't use small resistors and big electrolytics. If
the values are scaled by 10x the Sallen-Key form doesn't look so bad.

If the values are scaled by ~200x to 220k and 0.1uF (makes sense to
me), the filter's performance is significantly improved, e.g. -50dB
at 120Hz. Surprisingly, the Sallen-Key and two-caps-to-ground forms
are nearly identical with a 500-ohm load (both have nice 86dB dip at
360Hz - can we believe that?), and are better than -67dB above 1kHz.
With a light 5k load the two-caps-to-ground form pulls ahead by a
modest 6dB, above 2kHz. But with the 67 to 86 attenuation results
shown by the Spice analysis, I'd want to spend some time fixing the
2n7000 FET's subthreshold model before taking them too seriously.

.. 2n7000
.. Vin ---+------------- D S ----- out
.. | G * use a 10V gate-source
.. | 220k 220k | 0.1 zener if Vin > 20 volts
.. '-/\/\--+--/\/\--+---||--- gnd
.. | 0.1
.. '----||--- gnd

Before leaving this investigation, in which I used resistive loads,
the issue of capacitive loading should be taken into account. For
example, if a three-terminal regulator follows the noise filter, a
separate input capacitor is good for the regulator's stability.
Or if other "ordinary" linear circuitry follows, that'll certainly
need bypass caps.

Note, if a BJT is used, as shown below, a small base resistor or
ferrite bead might be wise to dampen RF oscillation, viz,

.. ___
.. Vin ---+-------------- C E ---+-------|___|--- out
.. | B | |
.. | | '----||---+-- gnd
.. '-/\/-+-/\/-+-/\/-'
.. | '----||--- gnd
.. '--||-- gnd

A quick spice look with the MOSFET: adding load capacitors didn't
show much difference, more HF attenuation (78dB), and both plots
coming together above 20kHz. However, in these high-attenuation
regions, poor capacitor esr modeling can create significant errors.

 

[Tim Williams]

It would need to be too big, I think. Similarly with just using a BFC.

Tube guys don't care...
http://www.hammondmfg.com/193.htm

Tim

 

[Winfield Hill]

Tim Williams wrote...

John Devereux wrote ...

Tube guys don't care... http://www.hammondmfg.com/193.htm

Right, use brute force. E.g, check out this heavy beast:
cat no 193Q, 10H, 500mA, 53ohms, 1kV, 21 pounds.