Feedback for low frequency PWM regulator

[JMini]

I'm building a PWM regulator for an incandescent bulb. Some of thos was
described in a thread called "RMS Approximation of PWM/Square wave". In any
case. Since there is no inductor/diode/cepacitor in the output stage I'll be
using an RMS converter (LTC1968). For the PWM section I'm using the MIC1557
(SOT-23 size 555 equiv) for a R-C sawtooth to a comparator (TLV7211a)
inverting input. I can choose the frequency (probably in the 200-800Hz
range). The feedback is sent through the LTC1968 RMS converter to the FB pin
(0.8V) of a tiny (SC-70) 5mA voltage regulator (OnSemi NCP102). It's really
just a powerful error amplifier. The Output of that is sent to the
non-inverting input of the comparator. So if the feedback voltage drops, the
NCP102 increases voltage ot the non-inverting input of the TLV7211, thus
increasing duty cycle. I've tested this type of layout on breadboard using
different components. I got to thinking though.
Question:
Would it be possible to use a resistor divider between the MIC1557 and
comparator to reduce the voltage of the sawtooth and feed it to the
NON-inverting comparator input and send the RMS converter output directly to
the INVERTING input of the comparator? That way, a falling output voltage
would cause a reduction in voltage to the INVERTING input and increase duty
cycle? I could reduce the component count.
I realize there is no true reference voltage in the system, but since the
1557 is fed from a fixed 5V source, the sawtooth would be a constant 5*1/3 to
5*2/3 V. Thoughts guys?

 

[Martin Griffith]

I'm building a PWM regulator for an incandescent bulb. Some of thos was
described in a thread called "RMS Approximation of PWM/Square wave". In any
case. Since there is no inductor/diode/cepacitor in the output stage I'll be
using an RMS converter (LTC1968). For the PWM section I'm using the MIC1557
(SOT-23 size 555 equiv) for a R-C sawtooth to a comparator (TLV7211a)
inverting input. I can choose the frequency (probably in the 200-800Hz
range). The feedback is sent through the LTC1968 RMS converter to the FB pin
(0.8V) of a tiny (SC-70) 5mA voltage regulator (OnSemi NCP102). It's really
just a powerful error amplifier. The Output of that is sent to the
non-inverting input of the comparator. So if the feedback voltage drops, the
NCP102 increases voltage ot the non-inverting input of the TLV7211, thus
increasing duty cycle. I've tested this type of layout on breadboard using
different components. I got to thinking though.
Question:
Would it be possible to use a resistor divider between the MIC1557 and
comparator to reduce the voltage of the sawtooth and feed it to the
NON-inverting comparator input and send the RMS converter output directly to
the INVERTING input of the comparator? That way, a falling output voltage
would cause a reduction in voltage to the INVERTING input and increase duty
cycle? I could reduce the component count.
I realize there is no true reference voltage in the system, but since the
1557 is fed from a fixed 5V source, the sawtooth would be a constant 5*1/3 to
5*2/3 V. Thoughts guys?

Phew,
Any chance of posting a circuit somewhere?


martin

 

[JMini]

Phew,
Any chance of posting a circuit somewhere?


martin

OK. I've uploaded a quick drawing to imageshack. PLEASE forgive the hand
drawing. I'm awful at getting this stuff into schematic software.
http://img386.imageshack.us/img386/9920/pwmschemajx8.jpg

 

[Phil Allison]

"Mook Johnson"

2) Why use Comverter. Average should be fine as you are only going to
create a DC average.

** Nonsense.

Lamps are rated for DC or rms AC voltage.

The average value of a PWM wave can easily be way under the rms equivalent
value - hence you will wind up with a blown lamp.



....... Phil

 

[Martin Griffith]

I guess the real question is, why? The light output will still be very
nonlinear on any control input. Why not feedback on the light?

John

Anyone played with these?
http://www.intersil.com/data/fn/fn7377.pdf


martin

 

[Spehro Pefhany]

For less that $1 you can get a PIC in SO8 (and smaller) which has built in
5v regulator, built in voltage reference, built in oscillator, 10 bit ADC,
10 bit PWM generator, and enough processing power to measure the supply
voltage, calculate and apply the required PWM duty at several hundred Hz
(and still have 4 pins left over).

Which SO-8 one(s) have an internal voltage reference (ie. not Vdd) for
the ADC?

If you can't do software the lamp filament *is* a thermistor providing
direct feedback of what you are actually trying to control (the filament
temperature). You can connect the lamp as one leg of a wheatstone bridge
and make a bistable circuit controlling the MOSFET from a comparator which
flips off when the filament resistance exceeds a value set by the rest of
the bridge resistors. A second comparator can make a timer to flip the
bistable on again after a fixed dead time or an oscillator which flips the
bistable on at constant frequency.

Both are one chip (+ maybe something to drive the MOSFET hard) solutions.

Best regards,
Spehro Pefhany

 

[Spehro Pefhany]

I know the uC path is the lowest component count path. I just have ZERO
knowledge with regard to uCs. This would be a simple thing if I had the first
idea what I was doing with those things. I've been pointed in the direction
of the ATTiny85 AVR as well as the AVRFreaks forum, but I need a lot more
learning. I don't even know what to buy to connect my PC to the board to
program the chip.

Also, the circuit really needs to control RMS current or RMS voltage. The
bulb mary vary by application, so it's resistance would be different from
bulb to bulb.

If you measure the voltage v(t), then the required PWM output %
is proportional to 1/v^2(t).

You'd need to calculate a voltage divider to give you the reference
voltage at maximum input, and specify the output % at that input.

Eg. 10V maximum input with 25% output at 10V in (say it's a 25 ohm 1W
bulb). Then at 6V in you'd have about 69.5% pwm %, for that same 1W
output. It would hit the end stop at 50% of the maximum input voltage
in this example, below which you could extinguish the light, flash it
or whatever, or simply allow it to drop naturally below that level.
Best regards,
Spehro Pefhany

 

[legg]

I'm building a PWM regulator for an incandescent bulb. Some of thos was
described in a thread called "RMS Approximation of PWM/Square wave". In any
case. Since there is no inductor/diode/cepacitor in the output stage I'll be
using an RMS converter (LTC1968). For the PWM section I'm using the MIC1557
(SOT-23 size 555 equiv) for a R-C sawtooth to a comparator (TLV7211a)
inverting input. I can choose the frequency (probably in the 200-800Hz
range). The feedback is sent through the LTC1968 RMS converter to the FB pin
(0.8V) of a tiny (SC-70) 5mA voltage regulator (OnSemi NCP102). It's really
just a powerful error amplifier. The Output of that is sent to the
non-inverting input of the comparator. So if the feedback voltage drops, the
NCP102 increases voltage ot the non-inverting input of the TLV7211, thus
increasing duty cycle. I've tested this type of layout on breadboard using
different components.

Why ?

This sort of arrangement will hit the lamp with the mother of all
turn-on surges.

Once stable (tee hee), you've got an rms voltage comparison to a
buried reference that bears no constant relationship to anything else
in the circuit, save the NCP102's reference voltage.

What are you trying to do?

RL

 

[JMini]

Couple of questions.

1) Why are you driving the high side of the bulb instead of the ground
side. Ground side is much easier becaust eh source is at ground so the
gate driver can be referenced to ground. Unless it is a P-channel and VIN
is less than 15V you're probabbly better off with a low side N-channel
(they are more robust).

2) Why use Comverter. Average should be fine as you are only going to
create a DC average. for average its as simple as a R-C if the pwm
frequency is high enough or a multi pole active filter. if it is lower.


If you comare average voltage in with a variable reference, you can have a
adjustable brightness curcuit that is linar with your pot adjustment
andfixed from external changes in VIN or temperature.

In reality, I will be using low side switching, but that would have required
that I illustrate the voltage dividers feeding the differential inputs of the
RMS converter. That's all.

 

[JMini]

I'm not so sure about the error amplifier part, but I see
why you are using an RMS converter, so you are controlling
the "effective" voltage to the lamp. I take it you are
after constant brightness. Of course, you could also
measure the lamp light output so you compensate for it aging
(and for the initial tolerance).

Are you very cramped for space, or is there some other good
reason you don't want to average the voltage of the pulses
with an inductor, so you can eliminate the RMS converter?
Having the inductor would lower the current ripple from the
power source and also the losses in the switch.

I am extremely cramped for space. the RMS converter is an 8-MSOP package. The
inductor would be HUGE. This regulator will carry about 10-11 amps RMS. I'm
also looking at keeping the frequency rather low to avoid a ton of switching
noise an reduce the phantoms that pop up when using high frequency PWM in
close proximity to other sentitive bits. I don't imagine 40+ kHz 10 Amps RMS
would play bery nicely just millimeters away from my RMS converter.

 

[JMini]

Why ?

This sort of arrangement will hit the lamp with the mother of all
turn-on surges.

Once stable (tee hee), you've got an rms voltage comparison to a
buried reference that bears no constant relationship to anything else
in the circuit, save the NCP102's reference voltage.

What are you trying to do?

RL

The input voltage will only be a few volts higher than the regulated output.
However, since the Vin is from batteries, the input voltage will be falling
the whole time, but I want constant RMS voltage to the bulb. This is for a
regulator for obscenely powerful flash lights. I recently built a
non-regulated version (PWM soft-start) that was 220W in a 3D Maglite size. It
has to be small (30mm round x 6mm high). The NCP102 has a built in
programmable softstart. I can stretch out the start-up over a full second or
more if I need to. I thought that the known min/max of the sawtooth would
provide a sort of reference. But the NCP102 looks like it might just be a
requirement That inrush current is monsterous. You're right. But
soft-starting will save the bulb. I plan on using the International rectifier
IRLR7843 for lower power applications and the IRF2804S for higher power ones.

 

[legg]

The input voltage will only be a few volts higher than the regulated output.
However, since the Vin is from batteries, the input voltage will be falling
the whole time, but I want constant RMS voltage to the bulb. This is for a
regulator for obscenely powerful flash lights. I recently built a
non-regulated version (PWM soft-start) that was 220W in a 3D Maglite size. It
has to be small (30mm round x 6mm high). The NCP102 has a built in
programmable softstart. I can stretch out the start-up over a full second or
more if I need to. I thought that the known min/max of the sawtooth would
provide a sort of reference. But the NCP102 looks like it might just be a
requirement That inrush current is monsterous. You're right. But
soft-starting will save the bulb. I plan on using the International rectifier
IRLR7843 for lower power applications and the IRF2804S for higher power ones.

The output of the rms-dc converter has a very slow response time -
measured in the 100s of milliseconds. In order to get the NCP102 to
work with this inside the feedback loop, you're going to have to slow
it's regulator down considerably. There's a model available if you
want to see what a pspice-type simulator shows.

Although the converter has differential inputs, which will simplify
interface to the low-side driver actually being employed, it's linear
output is in the 0-400mV range. How are you matching this to the
regulator's 800mV internal reference?

Fast-rising and falling current transients will radiate, even in an
800hZ pwm cicuit. The resonant frequency is determined by your
battery, lamp and switch wiring loops reacting with the fet's output
capacitance. Check it out with a scope. At this low frequency, you can
probably be generous with snubbers.

RL

 

[JMini]

The output of the rms-dc converter has a very slow response time -
measured in the 100s of milliseconds. In order to get the NCP102 to
work with this inside the feedback loop, you're going to have to slow
it's regulator down considerably. There's a model available if you
want to see what a pspice-type simulator shows.

Although the converter has differential inputs, which will simplify
interface to the low-side driver actually being employed, it's linear
output is in the 0-400mV range. How are you matching this to the
regulator's 800mV internal reference?

Fast-rising and falling current transients will radiate, even in an
800hZ pwm cicuit. The resonant frequency is determined by your
battery, lamp and switch wiring loops reacting with the fet's output
capacitance. Check it out with a scope. At this low frequency, you can
probably be generous with snubbers.

RL

In the design I had working with a 40kHz PWM controller, I level shifted the
RMS output using a resistor divider by +750mV because the Feedback voltage of
the TL5001 is 1V. I can do the same with this design. I can just level shift
the output by 600 mV. That puts the RMS converter output (in regulation) at
200mV. Right in the middle of that linear range.
Good catch on the linear range. I'll make a note.

Please give a little more detail regarding the use of snubbers. Snubbers
around the FET? In a previous design of a simple PWM soft-start, scope trace
of the output under a 12 Amp load was pretty clean. It runs at 175 Hz.

 

[HarryD]

I'm building a PWM regulator for an incandescent bulb. Some of thos was

Cut lots of bad ideas!

You are trying to control a resistive load so a low side N-MOSFET to ground
is the simplest configuration. No need to look at the current waveform it is
just Vin/Rl. You just have to control the pulse width as the input voltage
changes, that is "Voltage Feed Forward Control". A simple comparator whose
output is beefed up to drive the FET. One input has a cap to ground and
charged thru a resistor by Vin. The cap is reset in the Toff time. The other
comparator input is your DC control voltage. This voltage must have a soft
start at power up. You might preheat the bulb with a resistor in parallel
with the FET prior to soft start. Most of the above can be found in one
controller chip or the ubiquitous UC42XX controllers can easily be
configured into Voltage Feed Forward control.

Cheers,
Harry

 

[nospam]

Cut lots of bad ideas!

Cut another bad idea.

He needs to regulate the power in a resistive load. The required PWM duty
cycle is inversely proportional to the square of the supply voltage. You
don't square anything with an RC network.

You could get a square law by making the PWM on time inversely proportional
to the supply voltage *and* the PWM period proportional to the supply
voltage but there are simpler solutions.
--

 

[JMini]

Cut lots of bad ideas!

You are trying to control a resistive load so a low side N-MOSFET to
ground is the simplest configuration. No need to look at the current
waveform it is just Vin/Rl. You just have to control the pulse width as
the input voltage changes, that is "Voltage Feed Forward Control". A
simple comparator whose output is beefed up to drive the FET. One input
has a cap to ground and charged thru a resistor by Vin. The cap is reset
in the Toff time. The other comparator input is your DC control voltage.
This voltage must have a soft start at power up. You might preheat the
bulb with a resistor in parallel with the FET prior to soft start. Most of
the above can be found in one controller chip or the ubiquitous UC42XX
controllers can easily be configured into Voltage Feed Forward control.

Cheers,
Harry

You make all of this sound so easy. It may be for you.
I'm going to readup on "Voltgae Feed Forward" control, but could you
elaborate on how the cap resets during Toff. If the cap is charged through
the resistor by Vin, its voltage will rise until it's at Vin. Or is the
resistor connected between the bulb and the FET drain?
Are there some shematic/examples on-line I can reference. Maybe I've just
been over complicating things. Regarding the laready available PWM contriller
(UC42XX included). They operate at very high frequencies. I don't need high
frequencies. Thanks for your input.

 

[JMini]

Cut another bad idea.

He needs to regulate the power in a resistive load. The required PWM duty
cycle is inversely proportional to the square of the supply voltage. You
don't square anything with an RC network.

You could get a square law by making the PWM on time inversely
proportional to the supply voltage *and* the PWM period proportional to
the supply voltage but there are simpler solutions.

I'm all ears. I'm really looking for all input here.
If someone has simple solution to keep the RMS voltage constant into a
resistive load using PWM while the input voltage drops, I'd love to hear
them. I showed you mine, now you show me yours. TIA, guys.

 

[nospam]

I'm all ears. I'm really looking for all input here.
If someone has simple solution to keep the RMS voltage constant into a
resistive load using PWM while the input voltage drops, I'd love to hear
them. I showed you mine, now you show me yours. TIA, guys.

For less that $1 you can get a PIC in SO8 (and smaller) which has built in
5v regulator, built in voltage reference, built in oscillator, 10 bit ADC,
10 bit PWM generator, and enough processing power to measure the supply
voltage, calculate and apply the required PWM duty at several hundred Hz
(and still have 4 pins left over).

If you can't do software the lamp filament *is* a thermistor providing
direct feedback of what you are actually trying to control (the filament
temperature). You can connect the lamp as one leg of a wheatstone bridge
and make a bistable circuit controlling the MOSFET from a comparator which
flips off when the filament resistance exceeds a value set by the rest of
the bridge resistors. A second comparator can make a timer to flip the
bistable on again after a fixed dead time or an oscillator which flips the
bistable on at constant frequency.

Both are one chip (+ maybe something to drive the MOSFET hard) solutions.


--

 

[HarryD]

Cut another bad idea.

He needs to regulate the power in a resistive load. The required PWM duty
cycle is inversely proportional to the square of the supply voltage. You
don't square anything with an RC network.

You could get a square law by making the PWM on time inversely
proportional
to the supply voltage *and* the PWM period proportional to the supply
voltage but there are simpler solutions.
--

He needs to regulate luminous intensity, controlling power may be one way.
A better way may be controlling current thru the lamp. VFF is a simple
current control. We need a plot of lamp current and power vises luminous
intensity. I'm betting on current being the more linear and best to control
intensity. Where is Don Klipstein when you need him? Sorry if this is a
rehash.

Cheers,
Harry

 

[JMini]

For less that $1 you can get a PIC in SO8 (and smaller) which has built in
5v regulator, built in voltage reference, built in oscillator, 10 bit ADC,
10 bit PWM generator, and enough processing power to measure the supply
voltage, calculate and apply the required PWM duty at several hundred Hz
(and still have 4 pins left over).

If you can't do software the lamp filament *is* a thermistor providing
direct feedback of what you are actually trying to control (the filament
temperature). You can connect the lamp as one leg of a wheatstone bridge
and make a bistable circuit controlling the MOSFET from a comparator which
flips off when the filament resistance exceeds a value set by the rest of
the bridge resistors. A second comparator can make a timer to flip the
bistable on again after a fixed dead time or an oscillator which flips the
bistable on at constant frequency.

Both are one chip (+ maybe something to drive the MOSFET hard) solutions.

I know the uC path is the lowest component count path. I just have ZERO
knowledge with regard to uCs. This would be a simple thing if I had the first
idea what I was doing with those things. I've been pointed in the direction
of the ATTiny85 AVR as well as the AVRFreaks forum, but I need a lot more
learning. I don't even know what to buy to connect my PC to the board to
program the chip.

Also, the circuit really needs to control RMS current or RMS voltage. The
bulb mary vary by application, so it's resistance would be different from
bulb to bulb.