LED driver with cascoded optocoupler feedback has any disadvantage?

[Flyback]

Hello,

The follwing two PDF Schematics (also attached as .txt LTspice simulations) are both LED drivers, with this spec…

Spec:
Buckboost
Vin = 12-100V
Vout = 40 to 64V
Switching frequency = 200KHz
LED current = 100mA
Discontinuous Mode.

One of the examples has cascaded optocoupler feedback, and the other has plain optocoupler feedback.

When trying to stabilise them, it is far far far easier to stabilise the one with cascaded optocoupler feedback…
Stabilising the one without the cascode is like trying to balance on the point of a pin.
Stabilising the one with the cascode is so simple its like falling out of bed.

Given the ease of stabilising the one with cascaded optocoupler feedback, why would anyone not use the cascaded optocoupler feedback? The extra components are very cheap, and give extra robustness in terms of improved stability.

Also, the one without the cascode has a longer start up time, and overshoots considerably more.

Whats the disadvantge of the cascoded opto feedback?

 

[crutschow]

Feedback stability is determined by the response of the compensation network not whether it is a cascode configuration or not. If the cascade network is more stable its because it provides better compensation or better frequency response then the one without the cascode circuit.

 

[Flyback]

If the cascade network is more stable its because it provides better compensation or better frequency response then the one without the cascode circuit.

Yes, but for any given crossover frequency, if you use the cascode network , you will achieve improved gain and phase margins....you will make your power supply stable in spite of future changes in component parameters due to aging, operation at high temperature etc.

The top post provides the simulations of cascode, and no cascode....if you download these, and try a bit of empirical feedback compensation with them, you will find the cascode one MASSIVELY EASIER to make stable.

......This tells you that those few extra cascode components are well worth it.

 

[crutschow]

Yes, but for any given crossover frequency, if you use the cascode network , you will achieve improved gain and phase margins....you will make your power supply stable in spite of future changes in component parameters due to aging, operation at high temperature etc.

The top post provides the simulations of cascode, and no cascode....if you download these, and try a bit of empirical feedback compensation with them, you will find the cascode one MASSIVELY EASIER to make stable.

......This tells you that those few extra cascode components are well worth it.

My point was that whether the circuit is cascode or not is not critical. What's important is the overall frequency response of the circuit, whatever its topology.

I see some text files but no simulation in your top post.

 

[ccurtis]

......This tells you that those few extra cascode components are well worth it.

If items are made in mass quantities a few extra components add up to big money. Parts counts are keep to a minimum.

 

[Flyback]

I see some text files but no simulation in your top post.

...if you change the .txt files to .asc, then you can run them in LT Spice.

My point was that whether the circuit is cascode or not is not critical. What's important is the overall frequency response of the circuit, whatever its topology.

...yes you right, but if you use a non cascoded circuit, then your circuit will have a poorer gain and phase marging than if you used the cascode..........your circuit design is almost always a compromise, and you never have as much phase margin as you like.......so the cascode offers you a better robustness for you design......just see for yourself the difference between the two circuits simulations posted....delete the feedback compensation components, and then try and redo the feedback.....i gaurantee you you will find the cascdoded one very easily...the non cascode will he hard.......please try it....i gaurantee you will be converted....

If items are made in mass quantities a few extra components add up to big money. Parts counts are keep to a minimum.

PNP SOT23's cost about 1 cent, even if your quantitys are 100,000, then thats $1000.....ok its a decent sum of money, but no more than a cleaners wages for a month.
And your test & development time will be saved by using the cascode, so you'll get that $1000 back.

Are you aware of the poor-ness of the tolerance on optocoupler parameters (eg CTR etc)?...believe me, robust optocoupler feedback, covering all variations on opto tolerance, is a nightmare!!....specially because the opto's collector capacitance tolerance is not even given by manufacuterer of the opto.....Explain to us how you're going to do a robust opto feedback compensation if you dont even know the variation on the opto's pole frequency?....(because the manufacturer will not provide opto collector capacitance parameter tolerance values)

 

[ccurtis]

.

PNP SOT23's cost about 1 cent, even if your quantitys are 100,000, then thats $1000.....ok its a decent sum of money, but no more than a cleaners wages for a month.
And your test & development time will be saved by using the cascode, so you'll get that $1000 back.

Are you aware of the poor-ness of the tolerance on optocoupler parameters (eg CTR etc)?...believe me, robust optocoupler feedback, covering all variations on opto tolerance, is a nightmare!!....specially because the opto's collector capacitance tolerance is not even given by manufacuterer of the opto.....Explain to us how you're going to do a robust opto feedback compensation if you dont even know the variation on the opto's pole frequency?....(because the manufacturer will not provide opto collector capacitance parameter tolerance values)

It's not only the cost of the part, but the space on the board and the stocking and handling during assembly. I'm just making the point that mass produced designs keep parts counts to a minimum and I suppose the designers factored in variances, perhaps even purchasing an opto with guaranteed parameters in accordance with a specification, or using a different technique altogether.

 

[Flyback]

perhaps even purchasing an opto with guaranteed parameters

Opto's for use in smps feedback do not come with "gauranteed parameters"....i used to work for one of the biggest TelCo's on the planet, and even they couldnt get tolerance spreads on base collector capacitance of the opto.....the manufacturers of opto's wouldnt even give us a typical value.
Given this fact, how are you going to do a robust design unless you push the opto pole high in frequency by cascoding?

I gaurantee you that you cannot buy opto's with the mains standard 3750VRMS isolation that have a stated base collector capacitance.

The base collector capacitance helps form the opto pole frequency, and what we are saying here, is that you have a parameter in your feedback loop equation that you dont even know what it is....you can do a lab measurment, but is the part you measured a "Typical" part?...and what's the spread over any particular batch?.....lets face it, we dont know.....therefore, we must use the cascode....unless you just want to "wing" it.

 

[schmitt trigger]

I think the cascode is a good idea, but would like to prototype a circuit and make some measurements.....not rely solely in simulations......but again, I think this idea has merit to investigate further.

 

[Flyback]

I think the cascode is a good idea, but would like to prototype a circuit and make some measurements.....not rely solely in simulations......but again, I think this idea has merit to investigate further.

Schmitt, yes i know what you mean.....prototypes must always be made...sims can never be ultimately relied upon...though in this case, the cascode thing is so simple that we can trust the simulator is right about this............simulators are good at these simple things....and a cascode stage is merely just shoving a common base stage between the opto and the feedback point........you can bet that the LTspice sim doesnt get this kind of thing wrong.

To appreciate this argument, you need to have worked with feedback optocouplers in SMPS's and then when you tried to calculate feedback gain and phase margins etc, you will have realised just what an absolute nightmare optocouplers are in the feedback path.

I am speaking of those opto's with 3750VRMS isolation voltage between optodiode and optotransistor.....there are no good ones...they are all hideous

 

[schmitt trigger]

I worked designing power supplies for 15 years. Although I'm doing something else right now, I'm thoroughly familiar with optocoupler quirks...I have several horror stories of PSUs related to optocouplers. That is why I latched immediately into this idea.

Having said that I also know, again by hard-won experience, that relying solely on simulation is also a recipe for trouble.

 

[ronsimpson]

What application needs 12V to 100V input voltage?

 

[Flyback]

that relying solely on simulation is also a recipe for trouble

.....agree entirely, simulation is just a part of the prep......a prototype would alwyas be made and tested.

What application needs 12V to 100V input voltage?

Many LED specs are this wide, because its cheaper to have one lamp that can be used by many different applications with different supply voltages.

However, in this case, the 12-100V input voltage is either a car battery (12V) , or a fork lift truck battery (~80-90V).

....ie the same lamp can be used on the car or the forklift

I have several horror stories of PSUs related to optocouplers

Please share you SMPS optocoupler feedback horror stories...heres one.....
.......................................................................................
As an aside, I once worked at a huge company where 400 out of 1000 field prototype SMPS’s went unstable. The equipment that they were running was inoperable as a result of the SMPS’s instability.
The 400 unstable units were exactly the same as the stable ones except for the optocoupler……..their optocoupler was the exact same part number as the ones in the stable units, but was branded differently, being manufactured by Fairchild instead of Vishay. (so it was the same opto, but just manufactured by a different company)


Though it was the (slight) change in optocoupler that precipitated the instability, the fact was that the SMPS did not have enough margin designed in to its feedback loop in order to handle the “normal” optocoupler parameter variations between the differently branded opto’s, ….in conjunction with the extra capacitance of the load…..

All 1000 SMPS units had actually been stable when operated into electronic loads in the lab.
-It was the field connection of these 400 SMPS’s that precipitated the instability….that is, the load equipment included a fairly large input capacitor bank, and this precipitated the instability, in conjunction with the different optocoupler parameters.
(The 600 SMPS’s with the Vishay optocoupler were stable when connected to the load and its capacitor bank)

The SMPS had an output LC filter. As you know, the cut-off frequency of this filter should be well above the crossover frequency of the SMPS (2 to 3 times).
With the added capacitance of the load capacitor bank, the LC cut off frequency was then brought down too low when compared to the SMPS crossover frequency……that is, the crossover frequency of those SMPS’s that featured the Vishay optocoupler, with its slightly variant parameters.

Using the cascode connection of the optocoupler, one can make it easier to achieve stability with these situations. You will be able to more easily avoid being caught out.

 

[ronsimpson]

However, in this case, the 12-100V input voltage is either a car battery (12V) , or a fork lift truck battery (~80-90V).

How are you going to power the IC? Getting power for the IC will not be easy from 10 to 100V.

I would think the difference between car voltage and fork truck would require different parts. Example: at 12V use a high current low voltage MOSFET. At 85V use a low current high voltage MOSFET. A input voltage of 10 to 100V requires a high current, high voltage MOSFET.

 

[schmitt trigger]

My horror story is very similar, of several hundred very expensive power supplies used for mobile phone tower modulators. Not only the cost of the recall, but fines and penalties by the mobile operator because of downtime.

The cause? We were purchasing optos from Motorola, had worked fine for many years. Then Motorola sold the division to someone else. Even though they had indicated that nothing would change, they made the internal LED dimmer. To maintain the CTR, they widened the photobase region. Of course, that substantially increased the capacitance, which screwed the phase margin. The result? Oscillation at a narrow input voltage window.

 

[Flyback]

How are you going to power the IC?

Zener /BJT regulator.......triple BJT ........i forgot the name of it now, where one bjt emitter feeds the others base.

its 10 to 100v but only 7W. Its economical to have just the one lamp...good points though

The result? Oscillation at a narrow input voltage window.

...good story...i wish others would come out with their opto horrors too.

 

[ronsimpson]

IC voltage.

I think your 'bulb' needs to start working down to 10 volts not 12 volts. A car battery, with the engine off, is 12V and drops from there. Then there can be wire loss.

The LT1243 needs 7.8/8.4/9.0 volts to start working. min/typ/max
The LT1243 needs 7.0/7.6/8.2 volts to operate when it is working.
...................(Will 7.0V supply turn the MOSFET on hard with the high current needed at low voltage?)
So you need to make a linear regulator that:
....With a 10V input delivers 9.0V to the IC.
....With 100V input will not over heat and deliver 12-15V. What is the load? 1-2 watts of heat?

Do you see how much the duty cycle changes from 12V to 100V.

When a car battery is charging its voltage gets to 15V and with a bad battery can get to 20V.
When a 85V battery, what charge voltage do you see? Where does the voltage go when things go wrong?

 

[ronsimpson]

You have the current sense resistor not on ground there fore:
...Need a opto
...Dual amp not at ground needing supply.

Use a SEPIC converter. (Using a dual coupled inductor.)
...Current sense resistor on ground.

There are some "LED Drivers" that have a low voltage reference voltage. (0.1, 0.15 volts) They are built to sense current, at ground.

LT3758 Look at this one.

 

[Flyback]

ronsimpson these are very good points that you make.

SEPIC.....

https://www.edaboard.com/thread250056.html

....SEPIC with high vin gives problems in the limitation it puts on sepic cap value.

SEPIC above 2W power level needs coupled inductor else it rings like mad....and the coupled inductor sepic needs a high value cap, which is not easy at high vin.

HV9910B controller has a low bias current draw, of 3mA....so that'll just about be ok with vin = 100v and a Multi BJT/zener regulator.

 

[ronsimpson]

SEPIC or Flyback....I think you will have less parts if you can get the current sense resistor on ground. (you won't need a 12 volt supply that floats on top of the input voltage.) Your error amplifier will be simple.

CoilCraft has good coupled inductors that can be used in SEPIC and Flyback applications.

The HV9910 solves many of the supply voltage problems I have talked about. Also it usually is used where the LED current is not directly measures. It makes your circuit in post #1 totally useless. There is no need for SEPIC with the HV9910.