This is my first post here.

I just made this a while ago and haven't test it yet.
My main intention was to use this circuit to power my 3W LED for my project.

The goal is to keep component count minimal and no special-chip being use.

Top red trace is the input power from battery, and bottom green one is output at the LED.

The circuit is configured as constant current SMPS, with V2 as Vref.

A Sziklai-pair is being use as switcher.

I might use a common LM311 comparator for the circuit.

The problem is, the efficiency is not as high as I expected.
Well, maybe 90%?

Is there anything that I can tweak?

Thanks...

Afdhal Atiff Tan

 

90% is pretty good! How did you measure and calculate it? Perhaps it's better than you think, and anyway others will be able to better help if they know details of what you are talking about.

 

You have listed a dual op-amp and each half uses 4 ma. 1.6% of 3 watts
You have no limiting resistor on the base of the transistor pair allowing more drive current than necessary?

 

The problem is, the efficiency is not as high as I expected.
Well, maybe 90%?

Looking at your graphs you seem to be obtaining ~73%.

Rather than the Sziklai-pair you could use a single power mosfet.

Have you checked the LED current is within rated value? You may have to adjust the voltage feedback resistor [say] to ensure this is the case.

If you stay with the Sziklai-pair, you will need to provide an additional negative supply to your comparator rather than tying the negative supply terminal to ground - per bountyhunter's warning about Q2 turn of in post #5. Also use something like an LM339 rather than the LT1720. I'd be very surprised if you get anywhere near 90% efficiency with this configuration.

 

That schematic pretty much defines the problem of using simulators to do designs. It won't work. If the simulator says it will, it's feeding you baloney.

One of many problems:

What will turn off Q2?

Hint: absolutely nothing...... and that's a problem.


BTW: the absolute maximum (destruction) voltage of the LT1720 comparator shown in the circuit is 7V.

 

What is meant by me in the first post is, I'm targeting much higher efficiency than the one I just constructed.

I don't know how to calculate the efficiency from the simulation result. Taking the peaks, worst-case might be ~60% efficient.

I'm targeting at least ~90% for my project.

Well yes, the comparator used here might not be suitable, i.e. operating voltage is way too high. But, hey, it's a simulation, it suppose to be a good substitute for LM311 with a pull-up resistor.

There's no current-limiting resistor at the base because it's suppose to act like an emitter-follower input, the base-current is low enough for the comparator to switch it, depending on the transistor's gain of course.

I tried using a MOSFET, but it would need much higher voltage to turn-off, hence the Sziklai-pair, i.e. preferring current-controlled device rather than voltage-controlled ones.
Else, I might need a high-side driver, which is something I want to avoid.

I just added a turn-off resistor at the power-PNP transistor, makes almost no different. This might indicate the Sziklai-pair turn-off properly.

Sorry for the fuss, I just too curious...

 

That schematic pretty much defines the problem of using simulators to do designs. It won't work. If the simulator says it will, it's feeding you baloney.

One of many problems:

What will turn off Q2?

Hint: absolutely nothing...... and that's a problem.

Certainly simulators have their limitations. But they are a very useful tool in the initial phase of a design. What are these "many problems"?

I don't understand your problem with Q2. When the Q1 base goes to ground there is no base current for Q1, so it's off and thus there is also no base current for Q2. So why do you think Q2 won't turn off?

To improve efficiency, use an P-MOSFET to replace the Sziklai-pair. That doesn't require a high side-driver. But it will reverse the polarity of the loop, so the connections to the comparator input must be reversed to compensate.

 

To improve efficiency, use an P-MOSFET to replace the Sziklai-pair. That doesn't require a high side-driver. But it will reverse the polarity of the loop, so the connections to the comparator input must be reversed to compensate.

Just replaced Q1 and Q2 with a P-MOS, and it seems to improved at least 10%! Thanks!

But, I guess P-MOS is pretty hard to find and MOSFETs would require higher voltage to turn-on, i.e. higher than 700mV to saturate.

If possible, I want to run this off a supercapacitor, which might go down to 6V.

Thanks again!

 

Certainly simulators have their limitations. But they are a very useful tool in the initial phase of a design. What are these "many problems"?

I don't understand your problem with Q2. When the Q1 base goes to ground there is no base current for Q1, so it's off and thus there is also no base current for Q2. So why do you think Q2 won't turn off?

It's a switcher. You need the switching devices to turn off sharply or the transition power dissipation will smoke them. There is nothing to remove the stored charge in Q2 when it turns off. At minimum, it needs a pull up resistor from base to emitter..

And I don't get the point of using a comparator for the error amplifier. It's output is just going to bang around in response to the switching noise.... and what in the circuit determines the switching frequency and generates the desired switching?

All I can say is, whoever thinks this design is feasible should build it up and see.

And the problems with sims is when the person using them doesn't know they are feeding the user BS which can't be trusted. Sims have uses, but this isn't one that is doing anything good.

 

....................
But, I guess P-MOS is pretty hard to find and MOSFETs would require higher voltage to turn-on, i.e. higher than 700mV to saturate.

If possible, I want to run this off a supercapacitor, which might go down to 6V.

P-MOSFETS are quite common.

You can buy "logic level" MOSFETs that fully turn on with 3 to 5V Vgs, which would work fine with a 6V supply.

 

I did not know until know a switcher can be built primitively like that.

If it works in reality, why not use it for 3W LED?

 

It's a switcher. You need the switching devices to turn off sharply or the transition power dissipation will smoke them. There is nothing to remove the stored charge in Q2 when it turns off. At minimum, it needs a pull up resistor from base to emitter..

And I don't get the point of using a comparator for the error amplifier. It's output is just going to bang around in response to the switching noise.... and what in the circuit determines the switching frequency and generates the desired switching?

All I can say is, whoever thinks this design is feasible should build it up and see.

And the problems with sims is when the person using them doesn't know they are feeding the user BS which can't be trusted. Sims have uses, but this isn't one that is doing anything good.

You point is well taken about the switching speed. I thought you meant it won't turn off at all. But I believe the sim will show the slow turn-off if you closely examine the transistor switching speed, which the op does not show.

The circuit is a simple hysteretic type switching regulator which is also available in commercial ICs. When the output voltage drops below the switch point, the transistor is switched on and the voltage starts to rise with a rate determined by the inductor and capacitor values. When it reaches the switch point the transistor turns off. At the point the output voltage continues to rise some from the stored energy in the inductor. The cycle then repeats. The switching frequency varies and is determined by the value of the inductor and capacitor, and the load current. Hysteretic regulators are simple, with the disadvantage that they do have an inherent small amount of ripple at the output. You might try building one yourself to see it work.

The may be some BS in this thread but it isn't necessarily coming from the sim.

 

I did not know until know a switcher can be built primitively like that.

If it works in reality, why not use it for 3W LED?

Why not indeed. You could configure it as a current regulator, which LEDs need, by adding a small resistor in series with the LED to ground and using the voltage across the resistor as the sense trip voltage for the comparator.

 

Why not indeed. You could configure it as a current regulator, which LEDs need, by adding a small resistor in series with the LED to ground and using the voltage across the resistor as the sense trip voltage for the comparator.

Isn't that what the OP is simulating? 80mV + feedback from 0.1 Ohms applied to the OpAmp.

I have most of this stuff here except P-ch digital MOSFET but PNP transistor also would do. Some 3W LEDs survived previous testing
But currently I work at TFT LCD circuit.

I would be interested to know about this circuit, if it works in reality.

Would it be easier eventually to use MC34063? Or is it overload condition at 12v input, 2.5v output? Since it has explicit current limiting.

 

The circuit is a simple hysteretic type switching regulator which is also available in commercial ICs. When the output voltage drops below the switch point, the transistor is switched on and the voltage starts to rise with a rate determined by the inductor and capacitor values. When it reaches the switch point the transistor turns off. At the point the output voltage continues to rise some from the stored energy in the inductor. The cycle then repeats. The switching frequency varies and is determined by the value of the inductor and capacitor, and the load current. Hysteretic regulators are simple, with the disadvantage that they do have an inherent small amount of ripple at the output. You might try building one yourself to see it work.

I know what a hysteretic is, I have built a few in the 25 years I designed power supplies. Look at the schematic: the comparator has no hysteresis, it's in a switcher circuit.

Do you seriously think it's going to switch accurately? It will have hundreds of millivolts of noise induced into it.

The circuit is junk. It makes me cringe when I see sim data from circuits that can't possibly work as drawn. I suggest you build it up, and if you eventually get it to work, post the actual schematic. I guarantee it will have a lot more parts than shown.

Once upon a time designers actually did that.

 

Isn't that what the OP is simulating? 80mV + feedback from 0.1 Ohms applied to the OpAmp.

I have most of this stuff here except P-ch digital MOSFET but PNP transistor also would do. Some 3W LEDs survived previous testing
But currently I work at TFT LCD circuit.

I would be interested to know about this circuit, if it works in reality.

Would it be easier eventually to use MC34063? Or is it overload condition at 12v input, 2.5v output? Since it has explicit current limiting.

Yes, it is what he is simulating. I didn't take a close look at the feedback circuit.

An IC dedicated as a switcher is likely better than the circuit shown since they do have addition functions built in. But for a simple switching current regulator for an LED, that circuit may be adequate.

 

I know what a hysteretic is, I have built a few in the 25 years I designed power supplies. Look at the schematic: the comparator has no hysteresis, it's in a switcher circuit.

Do you seriously think it's going to switch accurately? It will have hundreds of millivolts of noise induced into it.

The circuit is junk. It makes me cringe when I see sim data from circuits that can't possibly work as drawn. I suggest you build it up, and if you eventually get it to work, post the actual schematic. I guarantee it will have a lot more parts than shown.

Once upon a time designers actually did that.

Forgive my lack of knowledge, I know I would get into trouble with this simplistic circuit. As you said, the circuit might not work in real life.

That's why I post my question to this forum; requesting advice.

You point is well taken about the switching speed. I thought you meant it won't turn off at all. But I believe the sim will show the slow turn-off if you closely examine the transistor switching speed, which the op does not show.

The circuit is a simple hysteretic type switching regulator which is also available in commercial ICs. When the output voltage drops below the switch point, the transistor is switched on and the voltage starts to rise with a rate determined by the inductor and capacitor values. When it reaches the switch point the transistor turns off. At the point the output voltage continues to rise some from the stored energy in the inductor. The cycle then repeats. The switching frequency varies and is determined by the value of the inductor and capacitor, and the load current. Hysteretic regulators are simple, with the disadvantage that they do have an inherent small amount of ripple at the output. You might try building one yourself to see it work.

The may be some BS in this thread but it isn't necessarily coming from the sim.

Wow, thanks! That's a real helpful explanation. Now I'm learning something new.

P-MOSFETS are quite common.

You can buy "logic level" MOSFETs that fully turn on with 3 to 5V Vgs, which would work fine with a 6V supply.

I never knew such component exist before. Thanks, I'll look it up.


I guess the circuit already reached it dead-end, maybe no more improvement could be made.

 

I know what a hysteretic is, I have built a few in the 25 years I designed power supplies. Look at the schematic: the comparator has no hysteresis, it's in a switcher circuit.

Do you seriously think it's going to switch accurately? It will have hundreds of millivolts of noise induced into it.

The circuit is junk. It makes me cringe when I see sim data from circuits that can't possibly work as drawn. I suggest you build it up, and if you eventually get it to work, post the actual schematic. I guarantee it will have a lot more parts than shown.

Once upon a time designers actually did that.

It's great that you know what a hysteretic is. Then you will see that the hysteresis is provided by the lag due to the output inductor and capacitor. The output will overshoot the set point due to the energy stored in the inductor when the transistor is switched off. When the voltage is drops below the set point the transistor turns back on. The voltage difference between those two points is the hysteresis. My simulations for the given circuit shows that the voltage across the 0.1Ω resistor goes from about 75mV to 96mV, so the hysteresis is about 16mV above the threshold point. You can certainly use resistors to add some additional hysteresis to the comparator if you like, but the circuit as shown does have hysteresis and will switch with the required duty-cycle to give the desired output current.

Why would this circuit have any more noise than any other hysteretic circuit? Of course you need careful layout and grounding to minimize noise coupled into the feedback loop, but that's true for any switcher.

Certainly you have to build the circuit as a breadboard after simulating it and before you commit to any type of production. No one has ever implied that you don't. And you likely will have to add parts like decoupling capacitors and perhaps beef up the drive to the MOSFET. But I see no fundamental reason the circuit won't work basically as shown. You declaring the circuit "junk" doesn't necessarily make it so.

You and the late Bob Pease share a certain disdain for simulators and I realize there are some great designers who can design circuits near perfectly without needing to use a simulator, and that's fine. But most of us can use a little help, and a simulator is the best tool available to assist in developing the circuit design before you build it. Slide rule anyone?

 

It's great that you know what a hysteretic is. Then you will see that the hysteresis is provided by the lag due to the output inductor and capacitor.

Hysteretics have "built in" hysteresis (notice the name?) in the comparator to keep the thing from endlessly chattering. It's one of the main disadvantages of a hyst design, since it's output voltage ripple is higher due to the fact the output goes up and down a specific amount of voltage between the design set points where the oscillator turns on (Vlow) and shuts off (Vhigh).

I think you are ignoring my point of posting: IMHO the circuit as shown won't work, and it's ridiculous when a simulator shows performance data for a circuit that can't work since it is LYING and telling the user it does work. The schematic doesn't even include a pull-up resistor to turn off the switching transistor. As I said: if you think that dog can run, build it up. Then when you get it working, post the final schematic and compare to the one posted in the OP.

You and the late Bob Pease share a certain disdain for simulators and I realize there are some great designers who can design circuits near perfectly without needing to use a simulator, and that's fine. But most of us can use a little help,

It's true Bob and I were skeptical of sims, and with good reason: we kept running into a whole batch of idiots who use them as a crutch. Our customers used sims to leapfrog the design phase and then howled when their crap didn't work. No person should ever be allowed to use a sim unless and until they are experienced enough to know where the sim is lying. That sentence can not be emphasized enough. Even some of our "best" designers fell into the lazy trap and got burned using sims for IC design. The so-called best models don't work in many cases and designers get bit.

I could post about ten thousand incidents on the subject , but don't care to invest the time.

In this thread, the problem is pretty clear: the OP doesn't know enough about a buck converter to know where the power losses are and is using a sim to give performance data. I don't see that as helpful.

Certainly you have to build the circuit as a breadboard after simulating it

Then exactly what good did simulating it do? The "data" from the sim can't be taken as accurate, you still have to take real data. The design is not accurate, you still have to make it work. The designer should read an app note, build the circuit, modify until performance is correct, then they are done. In the process they gain the understanding of how it actually works.

IMHO, the only thing sims do is make it cheaper and faster for people to pretend that they did a design when they did not, and in the VAST MAJORITY of cases today, they are using sims to go direct to "final design". As I said, thousands of cases of how that train rolled off the cliff.


BTW: even the famed "Simple Switcher" modeling software which is pretty good (and took more than five years to get into usable condition) has some well defined "blind spots" where it doesn't give accurate data. And the point is that the reason it gives reasonable accurate data at any places is because people like me built up the circuits, took the data, AND THEN THE SOFTWARE GUYS TWEAKED THE MODELS UNTIL THEY ALIGNED WITH THE DATA. Something to always consider......... the people who create models know the sequence is:

1) Take the data

2) Fix the model to agree with the data

3) Repeat until you give up and have to release the model with whatever problems it still has.

The old joke about never watching sausage being made applies to software models. It also explains why people who were involved know their limitations.

 

Geez, you guys makes me feeling guilty now.
I never thought it could be this complicated.

I just built the first circuit I posted, and here are some pics.

And it works. My hand covering the LED.

Close-up.

Scope-shot at the comparator output. Low-power output.

Higher-power output, with the same timebase. Frequency seems to increase.

----------------------------------------------------------------------
Using the same configuration, I use LM393 comparator, with 1kR pull-up resistor at the output. 12V voltage-source, those two light bulbs are as my R2, just in case it would blow-up.

300uH inductor, Q1 is 2N2222, Q2 is BD140 with 4.7kR pull-up resistor at the base for turn-off enhancement, and the fast diode is MBR1635.

1N4148 as my Vref for non-inverting input, connected to a voltage divider, adjustable from 0V to ~80mV. The Rsense is 150mR.
----------------------------------------------------------------------
I made some rough measurements...

The powerIn is 12.15V*340mA=4.13W, powerOut is 3.3V*250mA=825mW.
Efficiency approximately 20%.

That is so poor. The Q2 gets warm, without heatsink.
----------------------------------------------------------------------
I guest the best way is still by using crutschow suggestion, which is to replace the Sziklai-pair with a PMOS.
----------------------------------------------------------------------
I have to admit, maybe the circuit design is already flawed from the beginning. I'm still learning...

I never meant to cause troubles...


Thanks a lot for the feedbacks you guys.