Buck Converter Output Capacitor

[QQ]

Hi all,

I would be grateful if you can help me clear this basic doubt
regarding buck converters:

During startup of a buck converter I notice there is an "inrush"
current that flows into the output capacitor. I think that this inrush
current can be reduced by increasing the soft start capacitance. But
increasing the soft start cap reduces the rate at which the output
voltage comes up, which is something I don't want to do. So I select
FET's which can handle the inrush spike and and inductor which doesnt
saturate at the level of current. Also I inrease the R_over_current
sense resistance value. But what parameter of the output capacitor
should I be looking at to determine whether the output cap can handle
the inrush current?

The datasheet for the Cap seems to specify Rated Voltage, Rated Cap,
EST, Rated ripple current, Tangent of loss angle, Leakage current.

Thanks,
QQ

 

[R.Legg]

q_q_404@hotmail.com (QQ) wrote in message news:<9ab724f3.0405072123.30308e3c@posting.google.com>...

Hi all,

I would be grateful if you can help me clear this basic doubt
regarding buck converters:

During startup of a buck converter I notice there is an "inrush"
current that flows into the output capacitor. I think that this inrush
current can be reduced by increasing the soft start capacitance. But
increasing the soft start cap reduces the rate at which the output
voltage comes up, which is something I don't want to do. So I select
FET's which can handle the inrush spike and and inductor which doesnt
saturate at the level of current. Also I inrease the R_over_current
sense resistance value. But what parameter of the output capacitor
should I be looking at to determine whether the output cap can handle
the inrush current?

Current ratings in a capacitor are thermally derived. The only parts
I'm aware of that are sensitive to surge currents are some solid
tantalums.

With fet's and inductors as series limiters, there is generally no
concern for the capacitor, as fets are more likely to be damaged
first. Even when the charge transfer is from capacitor to capacitor,
or source impedance is extremely low, damage is more likely to occur
first in the switch - even if (or perhaps especially if) it's a
mechanical one.

In a buck regulator, if the current is controlled in any way, it's the
controlling switch that will suffer first, if only from agravated
ground bounce affecting drive integrity. You should use the soft start
to indirectly limit currents, with the aim to keep components within
their intended operating ranges and to avoid disturbing the input
supply, if the current is not controlled directly. This control need
not limit the rise severely, but disabling it entirely is not a wise
step.

How large a dV/dT are you attempting on the output and for what
reason? How large a filter capacitor are you using and how was it
selected?

RL

 

[R.Legg]

q_q_404@hotmail.com (QQ) wrote in message news:<9ab724f3.0405072123.30308e3c@posting.google.com>...

saturate at the level of current. Also I inrease the R_over_current
sense resistance value. But what parameter of the output capacitor
should I be looking at to determine whether the output cap can handle
the inrush current?

I forgot to mention that the capacitor parameter controlling peak
current is most commonly found as a dV/dT limit. Ipk = C x dV/dt. This
is usually applied to film capacitors and reflects the ruggedness of
internal metalization and it's contact integrity to the schoopage.

RL

 

[James Meyer]

On 8 May 2004 07:51:49 -0700, legg@magma.ca (R.Legg) posted this:

How large a dV/dT are you attempting on the output and for what
reason? How large a filter capacitor are you using and how was it
selected?

RL

Good question. One way to reduce the inrush current is to reduce the
value of the output capacitor to a value as small as the converter can tolerate
consistent with the ripple requirements.

Jim

 

[Joerg]

Hi James,

Not necessarily. A switcher inductor attempts to instantly dump whatever its peak
current is into that output cap. Smaller caps might not live long if they were
rated way under that current level.

Regards, Joerg
http://www.analogconsultants.com

 

[QQ]

Thank you all for your informative answers. I have another question
which I would like to know the answer to:

On one of the (voltage mode) buck converters I have designed the
switching frequency is 700KHz. The output voltage which has a ripple
of about 5mV is modulated by a low frequency sinusoid (25KHz
frequency) of ripple about 15mV. I have used appropriate probing
techniques. I am quite new at this and I dont know why this is
happening? Instability is my guess but I had designed for a phase
margin of about 45deg and gain margin of about -70db. Any ideas as to
what the problem could be?

Thanks
QQ

 

[R.Legg]

q_q_404@hotmail.com (QQ) wrote in message news:<9ab724f3.0405090044.297da670@posting.google.com>...

Thank you all for your informative answers. I have another question
which I would like to know the answer to:

On one of the (voltage mode) buck converters I have designed the
switching frequency is 700KHz. The output voltage which has a ripple
of about 5mV is modulated by a low frequency sinusoid (25KHz
frequency) of ripple about 15mV. I have used appropriate probing
techniques. I am quite new at this and I dont know why this is
happening? Instability is my guess but I had designed for a phase
margin of about 45deg and gain margin of about -70db. Any ideas as to
what the problem could be?

Check your measurement set-up.

If a computer is involved, disconnect it to see if there's any
difference. Common culprit is ground noise through a serial port or
similar hookup.

I note that you haven't answered any of the questions your previous
post raised, nor have you enlightened us on any cure that may have
been found to the originally posted problem, in the meanwhile. That's
pretty shabby use of a two-way communication channel.

RL

 

[Roger Gt]

"QQ" <q_q_404@hotmail.com> wrote in message
news:9ab724f3.0405090044.297da670@posting.google.com...
: Thank you all for your informative answers. I have another
question
: which I would like to know the answer to:
:
: On one of the (voltage mode) buck converters I have designed the
: switching frequency is 700KHz. The output voltage which has a
ripple
: of about 5mV is modulated by a low frequency sinusoid (25KHz
: frequency) of ripple about 15mV. I have used appropriate probing
: techniques. I am quite new at this and I dont know why this is
: happening? Instability is my guess but I had designed for a
phase
: margin of about 45deg and gain margin of about -70db. Any ideas
as to
: what the problem could be? Thanks QQ

There may be some random resonance in the circuit, not at all
uncommon in power circuits. Also hard to find if your not used to
this level of trouble shooting.

 

[QQ]

Sorry Legg, I didnt answer the previous question. I want dv/dt to be
fairly large becasue this DC/DC is on a plug in card. Some of the
power supplies that the chip uses already are present when this card
is plugged in. So this particular rail cannot come up too slowly or
the chip begins to draw current from this rail. Anyway I think I have
found a compromise value of Rocset but thanks for your answers. It was
quite informative.

Measurement setup seems to be fine since I cannot see a similar effect
on other DC/DC rails. No computer is connected to the device. One
other points I should probably mention. The ouput cap is 100uF
ceramic. Also I have "heard" of a condition called subharmonic
oscillation due to instability. Do you think this is an instability
issue? I tested the power supplies with load tranisents and it does
seem to recover from load transients fast enough which makes me
suspect that that is not the problem. Also the low frequency ripple is
in the range of 15mV peak to peak while the "actual" ripple voltage is
5mV peak to peak only. I think this low frequency ripple is of a very
low level which again makes me think, not an instability issue.

What are your comments on this?

Thanks,
QQ













legg@magma.ca (R.Legg) wrote in message news:<e715b5cc.0405090817.2cad6872@posting.google.com>...

q_q_404@hotmail.com (QQ) wrote in message news:<9ab724f3.0405090044.297da670@posting.google.com>...
Thank you all for your informative answers. I have another question
which I would like to know the answer to:

On one of the (voltage mode) buck converters I have designed the
switching frequency is 700KHz. The output voltage which has a ripple
of about 5mV is modulated by a low frequency sinusoid (25KHz
frequency) of ripple about 15mV. I have used appropriate probing
techniques. I am quite new at this and I dont know why this is
happening? Instability is my guess but I had designed for a phase
margin of about 45deg and gain margin of about -70db. Any ideas as to
what the problem could be?

Check your measurement set-up.

If a computer is involved, disconnect it to see if there's any
difference. Common culprit is ground noise through a serial port or
similar hookup.

I note that you haven't answered any of the questions your previous
post raised, nor have you enlightened us on any cure that may have
been found to the originally posted problem, in the meanwhile. That's
pretty shabby use of a two-way communication channel.

RL

 

[legg]

On 9 May 2004 18:06:02 -0700, q_q_404@hotmail.com (QQ) wrote:

Measurement setup seems to be fine since I cannot see a similar effect
on other DC/DC rails. No computer is connected to the device. One
other points I should probably mention. The ouput cap is 100uF
ceramic. Also I have "heard" of a condition called subharmonic
oscillation due to instability. Do you think this is an instability
issue? I tested the power supplies with load tranisents and it does
seem to recover from load transients fast enough which makes me
suspect that that is not the problem. Also the low frequency ripple is
in the range of 15mV peak to peak while the "actual" ripple voltage is
5mV peak to peak only. I think this low frequency ripple is of a very
low level which again makes me think, not an instability issue.

Is this condition occuring when plugged into the system, or when it is
stand-alone?

It is not subharmonic instability, as 25KHz is too far off to be a
half, third or quarter tone of 700KHz.

If the supply is running in a system with other converters operating
at a near-frequency, this could be a beat-frequency. This occurs when
another unit is operating 25KHz away from the DUT. It is not
guaranteed to be inaudible as the frequency can be anywhere within
your circuit's tolerances. Multiple units in the same system only
avoid this through synchronization.

You may find that it is an effect already common in the system, using
previous supply types in parallel, unless they were somehow configured
to avoid it.

If there is any sharing circuitry, you might check them out for
contributing funny business as well.

By the way, at what frequency is your '45degree margin' obtained?

Your inrush problem sounds like a shortcoming in the control
circuitry, or mating configuration. Paralleled and hot-swappable
circuits have to be designed for the function - they will not occur
accidentally.

RL

 

[QQ]

Hi Legg,

Thanks for your time. I am getting more and more convinced after
showing the waveform to others that it is probe noise. They say that
the scope probes can pick up random small magnitude noise components
which could show up when the voltage scale is reduced to the 5mV per
div range. Also load transient testing gives good recovery which is
comforting. Its a digital tektronix scope by the way. And the probe is
'fashioned' from a 50 ohm BNC cable to avoid noise pickup
(ironically!).

The gain crossover frequency is designed to be about 20 KHz and that
is the point where the phase margn is lowest ie. 45 deg. I havent
verified this using a frequency response analyzer though.

Ive heard that stability can be tested using load transients. How
exactly does one set that up? Of course if the supply recovers on load
transients then it is stable. But is it possible to say, just looking
at load transient response, what the gain margin and phase margin is
etc.?

Thanks,
QQ













legg <legg@nospam.magma.ca> wrote in message news:<3qlt90dcish1mn5701mbtoptkl7lc3a0of@4ax.com>...

On 9 May 2004 18:06:02 -0700, q_q_404@hotmail.com (QQ) wrote:

Measurement setup seems to be fine since I cannot see a similar effect
on other DC/DC rails. No computer is connected to the device. One
other points I should probably mention. The ouput cap is 100uF
ceramic. Also I have "heard" of a condition called subharmonic
oscillation due to instability. Do you think this is an instability
issue? I tested the power supplies with load tranisents and it does
seem to recover from load transients fast enough which makes me
suspect that that is not the problem. Also the low frequency ripple is
in the range of 15mV peak to peak while the "actual" ripple voltage is
5mV peak to peak only. I think this low frequency ripple is of a very
low level which again makes me think, not an instability issue.

Is this condition occuring when plugged into the system, or when it is
stand-alone?

It is not subharmonic instability, as 25KHz is too far off to be a
half, third or quarter tone of 700KHz.

If the supply is running in a system with other converters operating
at a near-frequency, this could be a beat-frequency. This occurs when
another unit is operating 25KHz away from the DUT. It is not
guaranteed to be inaudible as the frequency can be anywhere within
your circuit's tolerances. Multiple units in the same system only
avoid this through synchronization.

You may find that it is an effect already common in the system, using
previous supply types in parallel, unless they were somehow configured
to avoid it.

If there is any sharing circuitry, you might check them out for
contributing funny business as well.

By the way, at what frequency is your '45degree margin' obtained?

Your inrush problem sounds like a shortcoming in the control
circuitry, or mating configuration. Paralleled and hot-swappable
circuits have to be designed for the function - they will not occur
accidentally.

RL

 

[legg]

On 11 May 2004 00:56:47 -0700, q_q_404@hotmail.com (QQ) wrote:

div range. Also load transient testing gives good recovery which is
comforting. Its a digital tektronix scope by the way. And the probe is
'fashioned' from a 50 ohm BNC cable to avoid noise pickup
(ironically!).

And no other plotters, printers or other devices are connected to the

Tek?

Using 'fashioned' test harnessing can mean you end up having to ship
one of them with each unit. A calibrated scope probe, plugged into a
hard-soldered socket, is probably a better solution, even if you have
to add an amplifier to get measurable amplitudes.

The gain crossover frequency is designed to be about 20 KHz and that
is the point where the phase margn is lowest ie. 45 deg. I havent
verified this using a frequency response analyzer though.

Ive heard that stability can be tested using load transients. How
exactly does one set that up? Of course if the supply recovers on load
transients then it is stable. But is it possible to say, just looking
at load transient response, what the gain margin and phase margin is
etc.?

I thought you said you'd done load transient testing on the DUT.
Fiddle with your compensation and note the differences. Don't forget
to vary the supply voltage over it's range. Apply the heat gun and
freeze spray, if you don't have a thermal chamber. Look at the input
current, too, while hitting the output with the load change.

The main kinks in load transient testing are in obtaining meaningful
response around zero load or near the device's limit, and also in
getting good definition and control of the load current wavefront
di/dt. You will find that most shunts give erroneously high reports,
as their bandwidth is abysmal.

RL

 

[John Devereux]

q_q_404@hotmail.com (QQ) writes:

Hi Legg,

Thanks for your time. I am getting more and more convinced after
showing the waveform to others that it is probe noise. They say that
the scope probes can pick up random small magnitude noise components
which could show up when the voltage scale is reduced to the 5mV per
div range. Also load transient testing gives good recovery which is
comforting. Its a digital tektronix scope by the way. And the probe is
'fashioned' from a 50 ohm BNC cable to avoid noise pickup

Well, is the signal still there with the circuit unpowered?

Try unplugging everything else in the vicinity (computers, power
adaptors, everything you can except your circuit and the scope!).


--

John Devereux

 

[Dario]

NTC resistors are widely used to ward off inrush current attack !



QQ wrote:

Hi all,

I would be grateful if you can help me clear this basic doubt
regarding buck converters:

During startup of a buck converter I notice there is an "inrush"
current that flows into the output capacitor. I think that this inrush
current can be reduced by increasing the soft start capacitance. But
increasing the soft start cap reduces the rate at which the output
voltage comes up, which is something I don't want to do. So I select
FET's which can handle the inrush spike and and inductor which doesnt
saturate at the level of current. Also I inrease the R_over_current
sense resistance value. But what parameter of the output capacitor
should I be looking at to determine whether the output cap can handle
the inrush current?

The datasheet for the Cap seems to specify Rated Voltage, Rated Cap,
EST, Rated ripple current, Tangent of loss angle, Leakage current.

Thanks,
QQ

 

[Terry Given]

"Dario" <dsafari@yahoo.com> wrote in message
news:40A4FD01.B67953BC@yahoo.com...

NTC resistors are widely used to ward off inrush current attack !

yes they are. And if you look carefully, you will find (as I did) that
virtually NO NTC can handle the energy required to charge a typical bus cap
when the NTC is hot - they tend to explode, instead. which they do.
Unfortunately, NTC's are designed to start cold then run hot - so power
cycling is a great way to make them explode. Legg, whats your take on NTC's?

Terry

 

[legg]

On Sat, 15 May 2004 17:00:18 -0700, "Terry Given"
<the_domes@xtra.co.nz> wrote:

"Dario" <dsafari@yahoo.com> wrote in message
news:40A4FD01.B67953BC@yahoo.com...

NTC resistors are widely used to ward off inrush current attack !


yes they are. And if you look carefully, you will find (as I did) that
virtually NO NTC can handle the energy required to charge a typical bus cap
when the NTC is hot - they tend to explode, instead. which they do.
Unfortunately, NTC's are designed to start cold then run hot - so power
cycling is a great way to make them explode. Legg, whats your take on NTC's?

Your problem was control of rising output voltage in a hotswap
circuit. Methods of charge equalization, prior to shorting the output
capacitors to the functioning bus, are what I thought you should be
looking at, if the controlling circuitry was not inherently
bamboozled.

The problem is normally considered to be one of firstly avoiding dips
and spikes in the functioning bus due to the back-surge or turn-on
overshoot; secondly avoiding connector pin damage - an issue in larger
or higher current circuits; then finally, of course, you want the
newly plugged-in circuit to run normally.

NTC's have an energy rating. The problem of operation, hot, is not
that the energy increases, as this energy is usually determined by the
capacitors and application voltage. The problem is that they no longer
provide the intended inrush limit.

If they are present when someone tries to energize a circuit that has
already failed, then obviously the energy that they experience is much
higher than normal. An exploding NTC is a sign of an over-sized fuse
or overstressed NTC.

RL

 

[Terry Given]

"legg" <legg@nospam.magma.ca> wrote in message
news:cneba09uh4gvphujgtke4gdqmp2sm7t3jg@4ax.com...

Your problem was control of rising output voltage in a hotswap
circuit. Methods of charge equalization, prior to shorting the output
capacitors to the functioning bus, are what I thought you should be
looking at, if the controlling circuitry was not inherently
bamboozled.

The problem is normally considered to be one of firstly avoiding dips
and spikes in the functioning bus due to the back-surge or turn-on
overshoot; secondly avoiding connector pin damage - an issue in larger
or higher current circuits; then finally, of course, you want the
newly plugged-in circuit to run normally.

NTC's have an energy rating. The problem of operation, hot, is not
that the energy increases, as this energy is usually determined by the
capacitors and application voltage. The problem is that they no longer
provide the intended inrush limit.

I did some work on a number of smps that had NTC inrush limiting. As usual
Legg is spot-on wrt energy ratings, and that .5CV^2 governs E, not the NTC
resistance. What the NTC does govern, however, is the peak power
dissipation, V^2/Rntc (ie the rate of change of energy). If the NTC is hot,
E doesnt change but Ppeak does - for 100:1 cold:hot resistance, Ppeak is
100x higher when hot. It is this I found most NTC's couldnt handle. The eqpt
in question was 90-265Vac rated, Rhot = 1R, so (265*1.41)^2/1 = 115kW peak
pulse power; (115*1.41)^2/1 = 26.3kW. I tested a range of NTCs, and couldnt
find one that would survive this (and fit in the space allowed). Some (the
smaller ones) exploded, most failed short-circuit.

When I discovered this, I went and looked at some real units in the field
(115Vac), and found about half had shorted NTCs. Note that this argument
alos applies to resistors, too - most people consider average power
dissipation, but ignore peak power (I have re-worked several smps designs
where large FET gates were driven from +12V supplies thru 4R7 0805
resistors, which see about 30W peak, and eventually fail)

If they are present when someone tries to energize a circuit that has
already failed, then obviously the energy that they experience is much
higher than normal. An exploding NTC is a sign of an over-sized fuse
or overstressed NTC.

RL

I once had to do a forensic analysis on an ac motor controller that caught
fire. soft-charge was resistive, bypassed by a relay. The actual problem was
that the customer, in a fit of inspired stupidity, had shorted the DC bus
terminals together. These of course were after the soft-charge circuit, the
relay of which was controlled by the main smps - no DC bus, no relay, so the
customer had dropped the national grid across the soft-start resistors and
left it there. It turns out that RTV burns nicely at 900 degrees C, and of
course the input current was nowhere neare enough to blow the fuses. We
ended up selecting Vitrohm for the soft-charge resistors, as they failed
open-circuit under these conditions. The customer also had to foot the bill.

Terry