Help me choose an Op amp current booster stage

[stcoso]

Hi... It's not the first time I open a thread about this topic but i need some help in this new design. I've begin this journey a couple years ago and used many different circuit and now i'd like to put everything toghether.


Okay so... I'm trying to design a fast LED driver. The LED that i want to drive has a Vf of about 3.5V and a max current of 600mA. When in pulsed operation it can withstand a lot more (over 2A).

First thing i tried was using a LT1210 CFB op amp as a current source/sink. It works well (except for input offsets that needed compensation) but it's somewhat slow for my purposes.


When driven with an op amp current sink with a low side n-FET i could achieve very good rising edges (7-8ns measured with fast photodiode) but very poor falling edges (100ns). Obviously the single n-FET is not able to pass current both ways... 

So i started looking around and found that i could use very fast GaN FET driver directly connected to the LED to achieve very fast rise/fall edges.
I've made a prototype that works really well (using a LMG1020) achieving sub 5-ns rise-fall. The problem here is that i've got no control on the luminous output other than phisically swapping the current limiting resistor.


So i'd like to tie all this togheter designing an op amp based current source/sink. I'm probably gonna use something like this:




or a difference op amp (like a current source it's built in "the art of electronics").

But i've got a problem choosing the right output stage. Let's say i'd like to have 3A peak current with 5ns rise/fall edges.
Should i parallel fast op amps with current sharing resistors (like a ths3491) ?
Could this be done with a BJT or FET push-pull? And in this case, how can i choose the right parts?


Thank you in advance.

[Vovk_Z]

According to datasheet LT1210 has up to 10 ns rise or fall time with 10R load. What was your time? And what circuit?

[stcoso]

My circuit with the lt1210 was not so good. 

I didn't use a R(iso) between the output and my load so i couldn't use the full bandwidth to avoid oscillation (Rf=750Ohm G=1).

Anyway those 10ns are for a purely resistive load and the LED that i'm driving has more or less 1nF Cj.

[stcoso]

I'm wondering... which kind of FETs are used in those GaN driver? Can i found a discrete N/P MOS comparable to them?

[Doctorandus_P]

Why so complicated?
How fast do you need it to get?

Why not simply use a resistor to limit the LED current to 2A and then use a simple switch to turn it on / off.

[stcoso]

I'd like to achieve 5ns rise/fall. It's for a research project... i need to be able to drive this LED from DC to some MHz (with as square as possible waves) AND to control the luminous output (without swapping current limiting resistors between measurements  ).

I know that it's possible drive them at that speed because i can do that using the LMG1020 GaN driver... unfortunately this IC as logic supply and driver supply on the same pin and has got UVLO.

[Zero999]

I'd like to achieve 5ns rise/fall. It's for a research project... i need to be able to drive this LED from DC to some MHz (with as square as possible waves) AND to control the luminous output (without swapping current limiting resistors between measurements  ).

Vary the voltage to the LED and resistor, to control the intensity. It's very difficult to build a current regulator that fast, which won't oscillate.

I know that it's possible drive them at that speed because i can do that using the LMG1020 GaN driver... unfortunately this IC as logic supply and driver supply on the same pin and has got UVLO.

Then run the LMG1020 off a separate regulated power supply and only change the voltage to the LED & resistor.

Another option is to use a constant current source and switch the LED by short circuiting it. If it's a linear current source, you'll probably need to turn it on for a millisecond before flashing the LED and turn it of a millisecond later, otherwise it'll get hot. A switched mode regulator might be a better idea. Adding an inductor in series with the current source output will improve the transient response, but could leat to oscillation, especially with a high speed linear regulator.

[stcoso]

Varying current limiting resistor and/or voltage is what i'm doing with the LMG1020 . Being forced to stay over UVLO and under the max Vdd gives me little control over the current.  And there is another problem... rise/ fall edges change dramatically varying Vdd and/or CRL.

Then run the LMG1020 off a separate regulated power supply and only change the voltage to the LED & resistor.

I'm not sure if i've understood what you're saying but i'm not using a FET to drive the LED... I'm using the lmg1020 output stage directly to drive the LED.


A thing that i've never tried with this setup (at this speeds a new PCB layout is mandatory) is using a different (variable) ground for the LED.

Another option is to use a constant current source and switch the LED by short circuiting it. If it's a linear current source, you'll probably need to turn it on for a millisecond before flashing the LED and turn it of a millisecond later, otherwise it'll get hot. A switched mode regulator might be a better idea. Adding an inductor in series with the current source output will improve the transient response, but could leat to oscillation, especially with a high speed linear regulator.

I'll look into this option...

[David Hess]

I'd like to achieve 5ns rise/fall. It's for a research project... i need to be able to drive this LED from DC to some MHz (with as square as possible waves) AND to control the luminous output (without swapping current limiting resistors between measurements .

Separate the controlled current source from the switching.

Implement the current source of your choice, and maybe cascode it to raise AC output impedance, and then current switch the output of the current source using diodes or a differential pair driven by the fast pulse generator of your choice.

Think of it as a differential pair with the tail current controlled by your current source, one input fixed, the pulse generator driving the other input, and the load in series with the collector/drain of one of the differential pair transistors.

[stcoso]

Correct me because there's a big chance that I'm wrong in this... If I do like like you described I could achieve good turn on but poor turn off. The problem here is that big LED capacitance

[Zero999]

Varying current limiting resistor and/or voltage is what i'm doing with the LMG1020 . Being forced to stay over UVLO and under the max Vdd gives me little control over the current.  And there is another problem... rise/ fall edges change dramatically varying Vdd and/or CRL.

Then run the LMG1020 off a separate regulated power supply and only change the voltage to the LED & resistor.

I'm not sure if i've understood what you're saying but i'm not using a FET to drive the LED... I'm using the lmg1020 output stage directly to drive the LED.


A thing that i've never tried with this setup (at this speeds a new PCB layout is mandatory) is using a different (variable) ground for the LED.

You didn't post a schematic, so I assumed you used the one on the data sheet.

I don't see why varying the ground voltage for the LED wouldn't work. It might even speed up the turn off time, since the LED will become reverse biased and most parts can tolerate 5V reverse bias. Just check the LED doesn't have an anti-parallel diode, in which case you'll need to add a Schottky diode in series.

[David Hess]

Correct me because there's a big chance that I'm wrong in this... If I do like like you described I could achieve good turn on but poor turn off. The problem here is that big LED capacitance

If that is an issue, I might use a capacitor and step waveform to remove charge out of the LED at turn-off, or use a smaller reverse current from a separate source to do the same thing.

[stcoso]

Correct me because there's a big chance that I'm wrong in this... If I do like like you described I could achieve good turn on but poor turn off. The problem here is that big LED capacitance

If that is an issue, I might use a capacitor and step waveform to remove charge out of the LED at turn-off, or use a smaller reverse current from a separate source to do the same thing.

You're suggesting to AC couple a different source/sink (maybe a logic IC??) synced with the main current source?

[David Hess]

Correct me because there's a big chance that I'm wrong in this... If I do like like you described I could achieve good turn on but poor turn off. The problem here is that big LED capacitance

If that is an issue, I might use a capacitor and step waveform to remove charge out of the LED at turn-off, or use a smaller reverse current from a separate source to do the same thing.

You're suggesting to AC couple a different source/sink (maybe a logic IC??) synced with the main current source?

I am suggesting using a capacitor to couple charge into the output which drives the LED off as the current source is switched off.  The charge is equal to the change in voltage across the capacitor.