High Vin LDO with truely low dropout in small package (long post)

rickman wrote:
> John Popelish wrote:
> 
>>John Popelish wrote:
>>
>>>rickman wrote:
>>>
>>>
>>>>We often need to power low current circuits (<100 mA) from a wide
>>>>battery voltage of 7 to 16.5 volts.  It is hard to find switching
>>>>regulators that will do this efficiently.
>>
>>This one (LT3470) has even better efficiency at low current
>>and a low parts count:
>>http://www.linear.com/pc/downloadDocument.do?navId=H0,C1,C1003,C1042,P8810,D6297
>>
>>Output at 3.3 volts:
>>
>>With 7 volts in and 1 mA out, 74%.
>>With 7 volts in and 100 Ma out. 80%.
>>
>>With 24 volts in and 1 mA out, 64%.
>>With 24 volts in and 100 mA out, 72%.
> 
> 
> Thanks for the tips, but I have looked at all of these devices as well
> as many more.  This is the third time I have done this same power
> supply search for these parameters.  If there was an inductive switcher
> out there that did a good job at 10 mA of current, I would have found
> it.  But the combination of low current and high Vin seems to be
> deadly.

Are you saying these efficiency specs are deadly, or wrong?

> In addition, all but a few of the high Vin regulators are
> non-synchrnous and many of those also require a boost diode.  By the
> time you add the sizes of all these components you are using a fair
> hunk of real estate and not getting much in the way of efficiency.

Yes, efficiency and low noise require some real estate.

> Designing an inductive switcher may well be complex.  But a multimode
> switched capacitor circuit is mostly a digital design along with
> careful control of the detailed timing.

And also has a maximum theoretical efficiency well below the 
100% theoretical efficiency of a buck regulator.  Switched 
capacitor voltage changes (with no inductors) are 
essentially RC processes.

> The only timing issue I am
> concerned about is that the P channel FETs must be driven through N
> channel FETs.  These devices add much more delay to the path than the
> logic circuits will.  To make sure there is no shoot through I will
> have to compensate these delays compared to the N channel FETs used on
> the low side of the caps.  I may have to add N channel FETs to drive
> the N channel FETs just to equalize the delays.
> 
> But what I asked for help on was selecting an LDO that meets my
> requirements.

I understand your request, I just don't yet understand why 
you are making it.  If you are severely limited in the EMI 
department, then you will have trouble with a switched 
capacitor step down circuit, as much as you will with an 
inductive buck regulator.  They both make noise.

I am suggesting that starting with something like the above 
buck regulator and spending your effort on noise 
containment, you will reach your goals and get higher 
efficiency and lower noise than is possible for the same 
real estate using your switched capacitor step down (without 
additional inductive noise filter components) and an LDO 
linear post regulator.

John Popelish wrote:
> Are you saying these efficiency specs are deadly, or wrong?

They are not correct for the mode I will be using the part.  On
inductive switchers a lot of power is used to keep the circuit
operating.  So at low power levels the efficiency is poor and can even
be beat by an LDO.  They get around this by essentially turning off the
switcher until the voltage drops enough to need the switcher again.  So
it runs in a burst mode with a higher ripple and a variable frequency.
I can't work with the variable frequency so I am stuck using the parts
in the PWM mode which has too low an efficiency.


> And also has a maximum theoretical efficiency well below the
> 100% theoretical efficiency of a buck regulator.  Switched
> capacitor voltage changes (with no inductors) are
> essentially RC processes.

Why is theoretical efficiency even an issue?  I have a design that over
a range of current will provide efficiencies between 70% and 95%
including the required drop out of the LDO.  Of course this is not
built or tested so it may end up having some higher losses than I
expect due to quiescent current.


> I understand your request, I just don't yet understand why
> you are making it.  If you are severely limited in the EMI
> department, then you will have trouble with a switched
> capacitor step down circuit, as much as you will with an
> inductive buck regulator.  They both make noise.

This circuit is not to deal with EMI.  Besides, this should have a lot
better EMI performance just because there is no inductor.  The reason
for this design is efficiency.


> I am suggesting that starting with something like the above
> buck regulator and spending your effort on noise
> containment, you will reach your goals and get higher
> efficiency and lower noise than is possible for the same
> real estate using your switched capacitor step down (without
> additional inductive noise filter components) and an LDO
> linear post regulator.

Please explain how I can improve the efficiency of the inductive
switcher at 10 to 30 mA of output current.  The simple inductive
switcher is not large.  But it is not efficient at low currents either.
 I don't understand how adding circuitry can improve that.

rickman wrote:
> John Popelish wrote:
(snip)
> I can't work with the variable frequency so I am stuck using the parts
> in the PWM mode which has too low an efficiency.

I don't understand why you can't work with variable 
frequency, if you keep the noise under control.

>>Switched
>>capacitor voltage changes (with no inductors) are
>>essentially RC processes.
> 
> Why is theoretical efficiency even an issue?  I have a design that over
> a range of current will provide efficiencies between 70% and 95%
> including the required drop out of the LDO.  Of course this is not
> built or tested so it may end up having some higher losses than I
> expect due to quiescent current.

Any time you connect two capacitors together that do not 
match in voltage, as much energy is lost as is transferred. 
  The switch is essentially a resistor in series with the 
charge transfer.

> This circuit is not to deal with EMI.  Besides, this should have a lot
> better EMI performance just because there is no inductor.

Inductors are not inherently noisy.  A shielded inductor can 
be part of a very useful noise filter.

> The reason for this design is efficiency.

You may teach me something new, if you can achieve high 
efficiency with a switched capacitor voltage changer.

And when charge plows between two of those capacitors 
through the low impedance of your switched, large current 
pulses (many times the average load current) will occur, and 
those can radiate a lot of noise, if you aren't careful.
> 
>>I am suggesting that starting with something like the above
>>buck regulator and spending your effort on noise
>>containment, you will reach your goals and get higher
>>efficiency and lower noise than is possible for the same
>>real estate using your switched capacitor step down (without
>>additional inductive noise filter components) and an LDO
>>linear post regulator.

> Please explain how I can improve the efficiency of the inductive
> switcher at 10 to 30 mA of output current.  The simple inductive
> switcher is not large.  But it is not efficient at low currents either.
>  I don't understand how adding circuitry can improve that.

The LT3470 claims an efficiency of 64% with 24 volts in, 3.3 
volts out and a 1 mA load (better with a 16 volt to 7 volt 
input).  If you add a small inductor to the input side and 
maybe a ferrite bead to the output, the noise level can be 
quite low.  The frequency is, however variable, since this 
part is a hysteretic controller.  But that means that it 
handles step load changes with guaranteed stability.  The 
main filter inductor could be less than 7mm square, like the 
100 uHy Sumida CDRH6D28NP-101ND @ $1 each:
http://www.sumida.com/en/products/pdf/CDRH6D28.pdf

John Popelish wrote:

> 
> Any time you connect two capacitors together that do not match in 
> voltage, as much energy is lost as is transferred.  The switch is 
> essentially a resistor in series with the charge transfer.
> 

Not quite true.

If the voltage difference is very small, the charging efficiency can be 
acceptable. Otherwise switched capacitor power sources would be totally 
useless. Rather than typically offering 95 percent efficiency.

On resistively charging a capacitor from zero, most of the energy loss 
happens early in the first time constant. By doing most of your charging 
four or five time constants out, the losses can be considerably lower.

This requires that the charge consumed per cycle be much less than the 
charge stored.

A fancier switchmode circuit that does an intermediate transfer to an 
inductance can also eliminate this problem.


-- 
Many thanks,

Don Lancaster                          voice phone: (928)428-4073
Synergetics   3860 West First Street   Box 809 Thatcher, AZ 85552
rss: http://www.tinaja.com/whtnu.xml   email: don@tinaja.com

Please visit my GURU's LAIR web site at http://www.tinaja.com

Don Lancaster wrote:
> John Popelish wrote:
> 
>>
>> Any time you connect two capacitors together that do not match in 
>> voltage, as much energy is lost as is transferred.  The switch is 
>> essentially a resistor in series with the charge transfer.
>>
> 
> Not quite true.
> 
> If the voltage difference is very small, the charging efficiency can be 
> acceptable. Otherwise switched capacitor power sources would be totally 
> useless. Rather than typically offering 95 percent efficiency.
> 
> On resistively charging a capacitor from zero, most of the energy loss 
> happens early in the first time constant. By doing most of your charging 
> four or five time constants out, the losses can be considerably lower.
 >
> This requires that the charge consumed per cycle be much less than the 
> charge stored.

I stand corrected on the efficiency possible.  I just 
simulated a 2 to 1 voltage switched capacitive voltage 
reducer, and if the switch on resistance was low enough, 
good efficiency was possible.  But the large current spikes 
I mentioned were also present.  If I use a two phase, high 
frequency 2 to 1 step down, everything quiets down pretty 
well.

> A fancier switchmode circuit that does an intermediate transfer to an 
> inductance can also eliminate this problem.

In article <1160326669.493440.41750@c28g2000cwb.googlegroups.com>, 
gnuarm@gmail.com says...
> John Popelish wrote:
> > Are you saying these efficiency specs are deadly, or wrong?
> 
> They are not correct for the mode I will be using the part.  On
> inductive switchers a lot of power is used to keep the circuit
> operating.  So at low power levels the efficiency is poor and can even
> be beat by an LDO.  They get around this by essentially turning off the
> switcher until the voltage drops enough to need the switcher again.  So
> it runs in a burst mode with a higher ripple and a variable frequency.
> I can't work with the variable frequency so I am stuck using the parts
> in the PWM mode which has too low an efficiency.
> 
> 

With a LOT of output capacitance, could you not end up with a
SMPS that  runs at your 600KHz for 1msec  and turns off for
100 msec.  With that kind of duty cycle,  you won't see much
EMI at anything other than 600KHz.

This would work if the power requirements are discontinuous---
part of the time at 100mA and part of the time at 10mA.  But
it would be a problem if the power required could be anywhere
in the range between 10 and 100mA.


Mark Borgerson

Mark Borgerson wrote:
> With a LOT of output capacitance, could you not end up with a
> SMPS that  runs at your 600KHz for 1msec  and turns off for
> 100 msec.  With that kind of duty cycle,  you won't see much
> EMI at anything other than 600KHz.
> 
> This would work if the power requirements are discontinuous---
> part of the time at 100mA and part of the time at 10mA.  But
> it would be a problem if the power required could be anywhere
> in the range between 10 and 100mA.

Or, you could make a sigma-delta regulator, that samples
the error signal at the 600Khz rate, and that would ensure all
edges are on 600Khz pickets, but some would be skipped at low powers.
-jg

In article <4529cc74$1@clear.net.nz>, no.spam@designtools.maps.co.nz 
says...
> Mark Borgerson wrote:
> > With a LOT of output capacitance, could you not end up with a
> > SMPS that  runs at your 600KHz for 1msec  and turns off for
> > 100 msec.  With that kind of duty cycle,  you won't see much
> > EMI at anything other than 600KHz.
> > 
> > This would work if the power requirements are discontinuous---
> > part of the time at 100mA and part of the time at 10mA.  But
> > it would be a problem if the power required could be anywhere
> > in the range between 10 and 100mA.
> 
> Or, you could make a sigma-delta regulator, that samples
> the error signal at the 600Khz rate, and that would ensure all
> edges are on 600Khz pickets, but some would be skipped at low powers.
> -jg

The key is that you need to skip enough 600KHz cycles so that the
power spectrum ends up with 600KHz peaks and 100Hz peaks (or some
other low frequency whose sum and difference with 600KHz stay
within a reasonable bandpass for 600KHz rejection.

Mark Borgerson

If I have understood you correctly, your system has two operating modes
a) low power and b) high power. Could you thus use two power supplies
that could be enabled and disabled using FET switches according to your
system requirements by this CPLD? In low power mode you would use LDO
and in high current mode you would use switched mode power supply+LDO.

I have not designed switched mode power supplies, but I guess if you
reduce the switching frequency, you will improve efficienfy with the
expence of ripple which could be "filtered" by this LDO. Could you get
away with for example 1 kHz switching frequency? Or could an adjustable
switching frequency be more feasable?

Please note, this was all pure speculation and I haven't tried this at
home :)

- Tim

This one came into my mind: How about using some small rechargeable
batteries or a very high capacitance condensator as an intermediate
power source which is charged by a 100mA step-down-switcher as needed.
When the charger (switching regulator) is operating in high current
mode, its efficiency remains quite good, maybe > 90%. You may want to
add an LDO to filter out the ripple created by charger and
battery/condensator. 

- Tim