crazy boost converter for LED use is too noisy!

[BW]

Hi,

I designed a, in retrospective, quite sick boosting DC-DC converter
for driving a LED-string. I boost using the LTC3783 controller and an
external FET, from 12V to around 80-90V at 0.5A!

Coming in at a duty-cycle of around the maximum the LTC3783 wants,
85%, it is needless to say that the inductor and associated PCB traces
are, well, magnetically active... (avg. current through the inductor
is around 3A, switching at +/- 1A approx. at 1 MHz)

The result works, somewhat, but is horrible - the output capacitor
makes this large clunking noise when the circuit is PWM'ed and the
current regulation is horrible as well, perhaps because the LTC3783
does delicate sensing of voltages around 100mV in the feedback-loop,
at the same time as there is an induced 100 mV in most leads due to
the magnetics it seems..

Should I thrown the design in the bin and start over with a more
prudent approach ? I guess the high-field high duty-cycle version will
never be "quiet" even with better PCB layout and better shielded
inductors ?

The LTC3783 demo-board only does 12V->36V, maybe for a reason!

Best regards,
BJorn

 

[Robert Baer]

Hi,

I designed a, in retrospective, quite sick boosting DC-DC converter
for driving a LED-string. I boost using the LTC3783 controller and an
external FET, from 12V to around 80-90V at 0.5A!

Coming in at a duty-cycle of around the maximum the LTC3783 wants,
85%, it is needless to say that the inductor and associated PCB traces
are, well, magnetically active... (avg. current through the inductor
is around 3A, switching at +/- 1A approx. at 1 MHz)

The result works, somewhat, but is horrible - the output capacitor
makes this large clunking noise when the circuit is PWM'ed and the
current regulation is horrible as well, perhaps because the LTC3783
does delicate sensing of voltages around 100mV in the feedback-loop,
at the same time as there is an induced 100 mV in most leads due to
the magnetics it seems..

Should I thrown the design in the bin and start over with a more
prudent approach ? I guess the high-field high duty-cycle version will
never be "quiet" even with better PCB layout and better shielded
inductors ?

The LTC3783 demo-board only does 12V->36V, maybe for a reason!

Best regards,
BJorn

If you *hear* anything from a capacitor, the displacement current is
going to kill it.
Seems your design needs improvement.

 

[D from BC]

Hi,

I designed a, in retrospective, quite sick boosting DC-DC converter
for driving a LED-string. I boost using the LTC3783 controller and an
external FET, from 12V to around 80-90V at 0.5A!

Coming in at a duty-cycle of around the maximum the LTC3783 wants,
85%, it is needless to say that the inductor and associated PCB traces
are, well, magnetically active... (avg. current through the inductor
is around 3A, switching at +/- 1A approx. at 1 MHz)

The result works, somewhat, but is horrible - the output capacitor
makes this large clunking noise when the circuit is PWM'ed and the
current regulation is horrible as well, perhaps because the LTC3783
does delicate sensing of voltages around 100mV in the feedback-loop,
at the same time as there is an induced 100 mV in most leads due to
the magnetics it seems..

Should I thrown the design in the bin and start over with a more
prudent approach ? I guess the high-field high duty-cycle version will
never be "quiet" even with better PCB layout and better shielded
inductors ?

The LTC3783 demo-board only does 12V->36V, maybe for a reason!

Best regards,
BJorn

I'm just learning to make best guesses for pcb and component
parasitics in spice for smps design.
Throw in some Cs and Ls here and there and a sim'd smps gets really
ugly.
Ringing, spikes, oscillations, feedback, mistiming, heat, loss of
control, EMI,.... huhhhh

For even more fun, make a controller from scratch too


D from BC
British Columbia
Canada.

 

[[email protected]]

Hi,

I designed a, in retrospective, quite sick boosting DC-DC converter
for driving a LED-string. I boost using the LTC3783 controller and an
external FET, from 12V to around 80-90V at 0.5A!

Coming in at a duty-cycle of around the maximum the LTC3783 wants,
85%, it is needless to say that the inductor and associated PCB traces
are, well, magnetically active... (avg. current through the inductor
is around 3A, switching at +/- 1A approx. at 1 MHz)

The result works, somewhat, but is horrible - the output capacitor
makes this large clunking noise when the circuit is PWM'ed and the
current regulation is horrible as well, perhaps because the LTC3783
does delicate sensing of voltages around 100mV in the feedback-loop,
at the same time as there is an induced 100 mV in most leads due to
the magnetics it seems..

Should I thrown the design in the bin and start over with a more
prudent approach ? I guess the high-field high duty-cycle version will
never be "quiet" even with better PCB layout and better shielded
inductors ?

The LTC3783 demo-board only does 12V->36V, maybe for a reason!

Best regards,
BJorn

The low sense voltage is to achieve high efficiency, i.e. you don't
have too many LEDs in a string. If you are doing 90V output, dropping
a bandgap worth of voltage across a sense resistor isn't that big of a
deal, percentage wise. I'm trying to recall the hack. Here it is on
the maxim website:
http://www.maxim-ic.com/appnotes.cfm/an_pk/3243
This isn't specific to any part, but rather topology.

I haven't done this first hand, so feel free to say I'm blowing it out
my arse. There is always some sort of gotcha in real life.

 

[[email protected]]

I'm just learning to make best guesses for pcb and component
parasitics in spice for smps design.
Throw in some Cs and Ls here and there and a sim'd smps gets really
ugly.
Ringing, spikes, oscillations, feedback, mistiming, heat, loss of
control, EMI,.... huhhhh

For even more fun, make a controller from scratch too

D from BC
British Columbia
Canada.

The chips themselves often don't work the first pass, so I suggest not
rolling your own. A few companies have tried this with disastrous
results, i.e. recalls.

I did a self timed boost converter that needed a second pass due to
not modeling enough stray on the pin where the inductor and diode
intersect. Of course, if you are building a discrete design, your
design mistakes are not as costly.

 

[BW]

If you *hear* anything from a capacitor, the displacement current is
going to kill it.
Seems your design needs improvement.

Yes, I've come to this conclusion as well The output capacitor is
seeing voltages go quickly up to 80V, then while the circuit is not
used the cap voltage leaks down to say 40V, then another PWM cycle
comes and raises it to 80V again. What I'm guessing is that it is this
repetitive high dV/dt behaviour that is 1) sounding and 2) destroying
the cap (which in itself is a quite interesting phenomena which I'd
like to know more about, but maybe more on that later .

Apart from avoiding the dV/dt itself, are there any capacitor
parameters or brand/models that are more tolerant to this ? E.g.
putting multiple smaller capacitors in parallel. The current capacitor
is a 1 uF/100V X7R part.

Best regards,
Bjorn

 

[BW]

---
The low sense voltage is to achieve high efficiency, i.e. you don't
have too many LEDs in a string. If you are doing 90V output, dropping
a bandgap worth of voltage across a sense resistor isn't that big of a
deal, percentage wise. I'm trying to recall the hack. Here it is on
the maxim website:http://www.maxim-ic.com/appnotes.cfm/an_pk/3243
This isn't specific to any part, but rather topology.

You mean the part in the Maxim note where there is a current sense
resistor that says "7.5?" on it ?

Anyway, yes, thanks for the idea. The 0.5A passing through the sense
resistor is PWM'ed at a quite low duty cycle so I could increase the
resistor and voltage drop somewhat and take the resistor heat with no
problem probably. This would probably reduce the impact of the noise
on the current regulation feedback at least.

I'm still afraid of the magnetic coupling to other parts of the
circuit though, and EMI in general, so I guess I have to revise the
driving scheme anyway..

Thanks,

Bjorn

 

[Tim Williams]

Why 90V and not, say, 25V * 4?

LEDs seem to parallel just fine under peak current...the "joule thief" I
made runs at 5MHz and the two white LEDs light up just fine in parallel.
Peak voltage is something like 3-4V.

Tim

 

[BW]

Why 90V and not, say, 25V * 4?

LEDs seem to parallel just fine under peak current...the "joule thief" I
made runs at 5MHz and the two white LEDs light up just fine in parallel.
Peak voltage is something like 3-4V.

If the Vf-spread is truly random, this would probably work fine. A big
problem is that if one of the strings are opened, the other will get
the full current and will burn out as well (unless you also skip the
current regulation and go with the old resistor-in-series current
limiter).

Another problem might be if the different strings accidentally are
manufactured from different batches having non-random Vf - for
example, some manufacturers have 3 or more Vf bins in the production
process, it is random from which bin you get your LED's but might not
be random in the individual bags. Then one of your 4 parallell strings
will need 18 volts and the other 25 volts. This might or might not be
visible/desirable depending on the v-i characteristics of course..

/Bjorn

 

[Don Klipstein]

Yes, I've come to this conclusion as well The output capacitor is
seeing voltages go quickly up to 80V, then while the circuit is not
used the cap voltage leaks down to say 40V, then another PWM cycle
comes and raises it to 80V again. What I'm guessing is that it is this
repetitive high dV/dt behaviour that is 1) sounding and 2) destroying
the cap (which in itself is a quite interesting phenomena which I'd
like to know more about, but maybe more on that later .

Apart from avoiding the dV/dt itself, are there any capacitor
parameters or brand/models that are more tolerant to this ? E.g.
putting multiple smaller capacitors in parallel. The current capacitor
is a 1 uF/100V X7R part.

This sounds like a ceramic capacitor. I have known these to be
piezoelectric, even microphonic. If it is not getting warm, it's probably
OK - but I would be leery about putting anything in production that has
capacitors producing audible sounds.

I did see mention of a current sense resistor in another post in this
thread. I would try increasing it to get the current down (your duty
cycle will have to increase).

I might also suggest a non-ceramic capacitor, of a kind with an AC
voltage rating and noted as suitable for use in pulse forming networks,
snubbers, etc.

- Don Klipstein ([email protected])

 

[Joerg]

The chips themselves often don't work the first pass, so I suggest not
rolling your own. A few companies have tried this with disastrous
results, i.e. recalls.

I did a self timed boost converter that needed a second pass due to
not modeling enough stray on the pin where the inductor and diode
intersect. Of course, if you are building a discrete design, your
design mistakes are not as costly.

I've designed a few of those but semi-discrete, the only chip in there
being a Schmitt. All the way up to 12:1 step-up. Works fine, in
production for >10 years now. The trick is to do an RF-style layout and
_not_ split any grounds. And don't use boutique chips or the folks from
purchasing will be all over you some day.

 

[qrk]

Hi,

I designed a, in retrospective, quite sick boosting DC-DC converter
for driving a LED-string. I boost using the LTC3783 controller and an
external FET, from 12V to around 80-90V at 0.5A!

Coming in at a duty-cycle of around the maximum the LTC3783 wants,
85%, it is needless to say that the inductor and associated PCB traces
are, well, magnetically active... (avg. current through the inductor
is around 3A, switching at +/- 1A approx. at 1 MHz)

The result works, somewhat, but is horrible - the output capacitor
makes this large clunking noise when the circuit is PWM'ed and the
current regulation is horrible as well, perhaps because the LTC3783
does delicate sensing of voltages around 100mV in the feedback-loop,
at the same time as there is an induced 100 mV in most leads due to
the magnetics it seems..

Should I thrown the design in the bin and start over with a more
prudent approach ? I guess the high-field high duty-cycle version will
never be "quiet" even with better PCB layout and better shielded
inductors ?

The LTC3783 demo-board only does 12V->36V, maybe for a reason!

Best regards,
BJorn

If the circuit ran fine in LTspice, your pcb layout may be at fault.
Switchers are very sensitive to pcb layout. I've seen switcher
controllers go berzerk because the designer thought ground was ground.
If you don't understand that wierd comment, then you need to read the
section on layout in the data sheet.

When probing your circuit, you need to pay attention to oscilloscope
probe ground lead length if you're using a 10x passive probe. You
can't use the standard ground lead on a scope probe. Active
differential probes work great for the low voltage parts on a
switcher.

LEDs look like resistors in over-current conditions. I've sucessfully
ran 6 strings of IR emitters in parallel running 1 amp thru each LED.
The current sharing balances out fairly well. The emitters were rated
for 50mA continuous. Amazingly, they worked for over 1 million pulses
at 9A with a very low duty cycle.

You might want to post your pcb layout on ABSE for the switcher
section to see if that could be the problem.

 

[BW]

When probing your circuit, you need to pay attention to oscilloscope
probe ground lead length if you're using a 10x passive probe. You
can't use the standard ground lead on a scope probe. Active
differential probes work great for the low voltage parts on a
switcher.

Thanks, yes, I'm going to dig out the active probes for further
probing.

LEDs look like resistors in over-current conditions. I've sucessfully
ran 6 strings of IR emitters in parallel running 1 amp thru each LED.
The current sharing balances out fairly well. The emitters were rated
for 50mA continuous. Amazingly, they worked for over 1 million pulses
at 9A with a very low duty cycle.

Interesting! I should hook up the LED's on a breadboard and see if
they behave as nice..

Which of these two solutions for a low-dutycycle flash would you
consider chosing then:

1) same as my original booster, but more LED's in parallell thus lower
voltage and higher current

2) booster with the same lower voltage as in 1), but with a current as
low as the PWM duty cycle charging a cap and then "flashing" using a
standard linear current regulator for the LED's

The result would be the same, but the circuit in 2 would be more
complex but would put much lower demands on the switcher.. OTOH it
would perhaps put more demand on the cap, which would have to be
specified for high di/dt use.

/Bjorn

 

[neon]

WHY BIULT SUCH A BOOSTER WHAT FOR ? LEDs are diodes of 3-4v and I=30-100 ma each. if you go directly across the line add a diode and resistors for the same effect .

 

[[email protected]]

You mean the part in the Maxim note where there is a current sense
resistor that says "7.5?" on it ?

Anyway, yes, thanks for the idea. The 0.5A passing through the sense
resistor is PWM'ed at a quite low duty cycle so I could increase the
resistor and voltage drop somewhat and take the resistor heat with no
problem probably. This would probably reduce the impact of the noise
on the current regulation feedback at least.

I'm still afraid of the magnetic coupling to other parts of the
circuit though, and EMI in general, so I guess I have to revise the
driving scheme anyway..

Thanks,

Bjorn

Actually, I screwed up. Take a look at this chip:
http://www.maxim-ic.com/quick_view2.cfm/qv_pk/2451
This is about as simple as a boost gets. Now look at the voltage
divider. Replace the resistor from "output" to "FB" with your LED
string. The feedback loop will set the voltage across the lower
resistor to that of a bandgap. The value of the resistor sets the
current in the LEDs.

This is how some of the early LED drivers were done. Now a full 1.24V
across the lower resistor wastes power, but if you have a long LED
string, the waste is not significant.

You don't need to use a Maxim boost. There are a few vendors of this
minimal style boost chip. I'm pretty sure National makes these as well
as JRC. The advantage here is the voltage levels are higher, so there
is less of a noise margin issue.

 

[Robert Baer]

Yes, I've come to this conclusion as well The output capacitor is
seeing voltages go quickly up to 80V, then while the circuit is not
used the cap voltage leaks down to say 40V, then another PWM cycle
comes and raises it to 80V again. What I'm guessing is that it is this
repetitive high dV/dt behaviour that is 1) sounding and 2) destroying
the cap (which in itself is a quite interesting phenomena which I'd
like to know more about, but maybe more on that later .

Apart from avoiding the dV/dt itself, are there any capacitor
parameters or brand/models that are more tolerant to this ? E.g.
putting multiple smaller capacitors in parallel. The current capacitor
is a 1 uF/100V X7R part.

Best regards,
Bjorn

Maybe 10uF will give less change, but that will be at the expense of
a much larger charging current over a brief time, (over?) stressing
something else.

 

[qrk]

Thanks, yes, I'm going to dig out the active probes for further
probing.


Interesting! I should hook up the LED's on a breadboard and see if
they behave as nice..

Which of these two solutions for a low-dutycycle flash would you
consider chosing then:

1) same as my original booster, but more LED's in parallell thus lower
voltage and higher current

2) booster with the same lower voltage as in 1), but with a current as
low as the PWM duty cycle charging a cap and then "flashing" using a
standard linear current regulator for the LED's

The result would be the same, but the circuit in 2 would be more
complex but would put much lower demands on the switcher.. OTOH it
would perhaps put more demand on the cap, which would have to be
specified for high di/dt use.

/Bjorn

I would use a lower voltage, since it's easier to design a boost
converter. There are a few converters with an integrated driver
transistor if you keep the voltage below 36 volts. Since you don't
show a schematic, hard to say what your circuit is doing.

In the past, I have used a boost converter and set the current so it
will charge a capacitor bank in a timely fashion to a specified
voltage. If you connect a small capacitor from the voltage feedback
input to ground, you can slightly overcharge the capacitor bank which
will shut down the boost converter until you fire the emitters. Use
some beefy ceramic capacitors (around 10uF) near the IR emitter string
to handle the fast edges. Use long life, low esr aluminum
electrolytics for the bulk capacitance. Look thru Digi-Key for
suitable capacitors.

This sort of scheme will give you a output pulse with decreasing light
output over time. The amount of droop is controlled by bulk
capacitance and pulse width.

Since you are over driving the emitters, be sure that the light output
remains stable with a stiff supply for your particular scenario. If it
starts to droop, you are overheating the emitter die. To look at the
light output, you need to use a reverse biased PIN photo detector if
you're below tens of microsecond pulse widths. Hamamatsu S5973-01
works well and is about $20. Pearson current transformers work well
for measuring the pulse current.

The IR flash unit I built incorporated 14 parallel strings. Each
string had 10 series connected emitters. Current was 1.0 A thru each
emitter. I used a power FET to flash the emitters. Pulse duration was
in the single digit microsecond region.

 

[[email protected]]

Thanks, yes, I'm going to dig out the active probes for further
probing.


Interesting! I should hook up the LED's on a breadboard and see if
they behave as nice..

Which of these two solutions for a low-dutycycle flash would you
consider chosing then:

1) same as my original booster, but more LED's in parallell thus lower
voltage and higher current

2) booster with the same lower voltage as in 1), but with a current as
low as the PWM duty cycle charging a cap and then "flashing" using a
standard linear current regulator for the LED's

The result would be the same, but the circuit in 2 would be more
complex but would put much lower demands on the switcher.. OTOH it
would perhaps put more demand on the cap, which would have to be
specified for high di/dt use.

/Bjorn

While I'm of the school that pulsing LEDs is more efficient than
continuous illumination, there are some that make the opposite
argument. Why not try the design I suggested in my second post, which
would dump a regulated current into one string of LEDs.

If you want to pulse the LEDs with regulated current, i.e. your
solution #2, consider something along the lines of "current regulated
diodes." They are just jfets running at IDSS. I think it would be
difficult for any feedback circuit to regulate the current over a
short pulse.

 

[BW]

While I'm of the school that pulsing LEDs is more efficient than
continuous illumination, there are some that make the opposite
argument. Why not try the design I suggested in my second post, which
would dump a regulated current into one string of LEDs.

The pulsing in this case is because the camera does not take images
continously anyway.

The design I started with was the same as in the Maxim application you
suggested - and the problem was partly the low feedback voltage, which
I easily raised to 1V, but also partly (and mostly) the inherent noise-
generating capability of switching a 4-5 amp current at 1 mhz through
a large inductor on a PCB... that will never be silent, it seems.

If you want to pulse the LEDs with regulated current, i.e. your
solution #2, consider something along the lines of "current regulated
diodes." They are just jfets running at IDSS. I think it would be
difficult for any feedback circuit to regulate the current over a
short pulse.

This is interesting. I looked into the current regulating diodes but
most off-the-shelf parts seemt o be in the < 50 mA range while I need
500 mA. Should I "roll my own" with a JFET then perhaps ? And spice
simulate it..

(The jfet is a feedback circuit as well of course, albeit with a
shorter roundtrip maybe than an op-amp and a BJT....)

/Bjorn

 

[[email protected]]

The pulsing in this case is because the camera does not take images
continously anyway.

The design I started with was the same as in the Maxim application you
suggested - and the problem was partly the low feedback voltage, which
I easily raised to 1V, but also partly (and mostly) the inherent noise-
generating capability of switching a 4-5 amp current at 1 mhz through
a large inductor on a PCB... that will never be silent, it seems.


This is interesting. I looked into the current regulating diodes but
most off-the-shelf parts seemt o be in the < 50 mA range while I need
500 mA. Should I "roll my own" with a JFET then perhaps ? And spice
simulate it..

(The jfet is a feedback circuit as well of course, albeit with a
shorter roundtrip maybe than an op-amp and a BJT....)

/Bjorn

The JFET is open loop.

I didn't catch that this was for a flash. What kind of synchronization
time do you require? What I"m thinking is you could just "charge" and
inductor, then dump the energy into the string of LEDs. How does the
camera tell the "flash" to turn off? Or do you put out the same amount
of light each time? What I suggest is writing an objective first, then
the circuit topology will follow.