EEVblog #567 - Precision 1A Current Source

[EEVblog]

Dave breadboards a precision 1A current source circuit and finds that even the application reference circuit doesn't work as expected. Showing that you can't just blindly trust that datasheet application circuit will work first go.
LTC6655 http://cds.linear.com/docs/en/datasheet/6655fc.pdf

[pickle9000]

This is my favorite type of video. Very enjoyable.

[Wytnucls]

Now the 1000$ question is, what does Dave need a precision 1A source for?

[robrenz]

Maybe testing the µcurrents

[lewis]

Try a big electrolytic in parallel with the input bypass capacitor, about 100-470uF should do it. Inductance of the leads from your PSU to the breadboard can upset loop stability and 100nF/4.7uF (or whatever it was) is nowhere near enough to compensate with that much series inductance at any significant load.

Try substituting the ceramic caps for electrolytics, or add a very small series resistance (0.5-1R) in series with the ceramics. Sometimes their ESR is TOO LOW, forms an underdamped LC tank circuit with the PCB trace/wire/package/whatever inductance, causing ringing on transients and oscillation. Some stuff here: http://cds.linear.com/docs/en/application-note/an88f.pdf.

10-22pF between force and sense (very close to the IC) might help too, as may a 22R resistor in series with the pass transistor base.

[Rufus]

The emitter follower doesn't increase the loop gain but it does lower the output impedance by a factor of 100 or whatever so a capacitor stuck on the output needs to be 100 (or whatever) times bigger for the same roll off. A diode in series isn't helping either (the capacitor isn't there if the diode is not conducting so it might make it prone to low frequency motor boating type oscillation).

I think you will have to stick enough capacitance on the chip output to make it stable and then maybe some more on the emitter to improve output noise and high frequency output impedance.

Now the 1000$ question is, what does Dave need a precision 1A source for?

A uohm meter adapter probably.

[Frank]

In my old job we had a precision current source made with the Analog Devices's AD584 reference, using a topology very similar to this one. 
 

Maybe testing the µcurrents

+1 for the new test jig for the µcurrents 

[rolandpenplotter]

Now the 1000$ question is, what does Dave need a precision 1A source for?

To power the circuit that powers the relay that powers his editing computer that creates the "TV station teardown" video!

[Marco]

If the sense input was high impedance this IC would be a lot nicer for this application.

The emitter follower doesn't increase the loop gain but it does lower the output impedance by a factor of 100 or whatever so a capacitor stuck on the output needs to be 100 (or whatever) times bigger for the same roll off.

In a different part they recommend 2.7 uF minimum for normal use, so the 4.7 uF for use with the 2n2222 in the datasheet was just hopelessly too small then? Could you just put an additional capacitor across the base emitter junction to stabilize the loop instead? Regardless, AC impedance of the current source is going to be pretty poor when all is said and done ... the whole concept of a high side current source with a NPN output transistor is a bit alien to me.

[Rerouter]

its a glorified op amp, (that constantly sinks 2mA on its sense line  ) so perhaps add a 125 ohm resistance on the sense line, a 1K resistance between the VF pin and the transistor base, and a 1nF cap between the transistor base and the VS pin (after the resistor)

this will give mean you would need to change to a 1.5 ohm shunt, but give it a shot.

[TheGreatGooglyMoogly]

Say Dave,

I think the culprit may lay in the ESR of the output cap (just a wild guess based on the video) - the ESR is too low.
I would try using a capacitor with a 0.5 - 2.0-ohm ESR -or- use a small resistor in series with the ceramic.
I've run into stability issues before while working on regulators and references used in Nano-positioning controllers. A quick peek at this data sheet in the bypass section will give you an idea what I am on about.

http://www.onsemi.com/pub_link/Collateral/NCP1117-D.PDF (page 9)

Great video as always! (I hate that regulator also :p )


EDIT: the data sheet for the LT6655 claims use a 0.1-ohm or less ESR cap. So my first guess is technically wrong, but I have my doubts still. The behavior looks exactly like a mis-bypassed NPC1117 regulator. Is it possible the breadboard leads have too much ESR or ESL? Or that this is one of those chips that never works? Puzzling indeed.

[Dave]

This video comes to mind. An LM399 is a bit of an overkill here, but you may get some ideas for your circuit.

[EEVblog]

this will give mean you would need to change to a 1.5 ohm shunt, but give it a shot.

I've go a couple of kick-arse 1.25ohm 4 terminal shunts on the way from Vishay, so need to stick with this value.

[Rerouter]

then try the base resistance as a starting point, i can understand when things are already ordered,

other than that the only solutions i can think of is to
A. find a better chip
B. measure the voltage on the shunt differentially through a low passed op amp with a slew rate limiting cap and drive the VS pin,  as opposed to floating the ground :/ yes i agree it may hurt absolute accuracy

[Marco]

This video comes to mind. An LM399 is a bit of an overkill here, but you may get some ideas for your circuit.

Seems like it could work fine with the LTC6655.

Could someone explain the trarlington to me, why is the extreme beta necessary?

[Rufus]

In a different part they recommend 2.7 uF minimum for normal use, so the 4.7 uF for use with the 2n2222 in the datasheet was just hopelessly too small then?

The output impedance (open loop) and the output capacitor need to form a dominant pole which reduces the open loop gain to less than unity before other circuit phase shifts add up to 90 degrees and the -ve feedback becomes +ve. The output impedance is likely quite low which is why you need a big capacitor with low ESR. It is a crappy way to form a dominant pole except for voltage regulators and references where a big capacitor on the output is a good thing reducing noise and improving high frequency transient response. The emitter follower still has some output impedance (re/hfe ?) so how hopelessly too small depends on how that compares with the chip output impedance.

[Rufus]

Could someone explain the trarlington to me, why is the extreme beta necessary?

Because the circuit regulates current in the sense resistor while the load gets that current less any base drive current for the pass transistor. With a tralington? gain of 1 million the load gets 1uA less than it should.

[Jay_Diddy_B]

Hi Dave,

This circuit will oscillate even in the simplest simulation:



The 4.7u capacitor current flows through the output of the op-amp A2.

If I remove the capacitor and put a 10uH inductor in series with the load, the circuit will also oscillate:



It doesn't matter if the inductance is in series with the supply or the load resistor, because they are in series, I have a two terminal current limiter. This circuit will also oscillate:



Adding a RC damping network will stabilize the circuit for inductance in the load circuit:



To stabilize for input inductance, an electrolytic should be added on the input:




I have attached a zip file with the zip models if you want to play along. There is no LTspice model for the LTC6655.

Jay_Diddy_B

[ScribblesOnNapkins]

This is crying out to be constructed dead bug style on a sheet of copper clad.

[EEVblog]

This circuit will oscillate even in the simplest simulation:

Thanks, but I haven't even gotten there yet, the basic app note circuit oscillates without the ground biasing opamp.
I haven't tried the PNP solution yet either.

[owiecc]

Regarding your initial circuit. I would remove the diode and the opamp from the circuit if you can have the output floating. Then you need to move the sense resistor to the current return path. Your LT is then ground referenced.

[Sarowie]

What I do not like about that "typical application circuit" is, that  the LTC6655 can source and sink.
With the NPN Transitors Base-Emiter Junction in Series to the forced output, that doesn't make sense:
If the sense pin is ever so slightly above the set point, the LTC6655 will try to sink current over the forced out put. But with the Base-Emiter junction in series, the LTC6655 is just not able to do so.
The forced output will then go more and more toward the Ground, until the sense pins becomes to low due to the load (which will be at least 2mA from the sense voltage divider). Then the forced output has then to climp back up from whatever voltage it was to VRef + VBE (1.25V+0.6V = 1.85V). As the NPN Transistor has an finite speed, it will ever so slightly over shoot and guess what: The LTC6655 forced output tries again to sink current, sucks out all the charge out of the base-emiter junction and goes toward GND. After some time, the sense input becomes to low and the error amp starts to increase the forced output voltage and will again overshoot ever so slightly.

I do not get, why they recommend a circuit which has an issue with the sink capability of the voltage reference.
Beeing able to source and sink is the main difference between a voltage regulator and a voltage reference.

[jesuscf]

Hi Dave,

This circuit will oscillate even in the simplest simulation:
...
Jay_Diddy_B

Jay_Diddy_B,

I think the oscillations you observe in the simulations are not caused  by the circuit topology itself but by the choice of integration method. The problem shows when solving the differential equations for inductors and capacitors if the trapezoidal integration method is used.   In most SPICE based simulators the trapezoidal method is selected by default. I got rid of the oscillations in the first circuit you posted by changing the integration method from trapezoidal to Gear.

As for the oscillations in Dave's circuit, they have to be caused by some kind of positive feedback.  I would check the main DC power source first and then check what is the effect of the typical 50pF capacitance between breadboard pins.  I can not see clearly from the video, but adjacent breadboard pins may be a problem in some circuits when one is an input and the other is an output.

Jesus

[riconette]

imho the scopes display "error" is the result of the persistence feature.
i've fed a scrambled signal in my scope (2072 fw 00.01.01, hw 1.0, fully hacked) and observed the same behaviour. the life waveform shows (persistence-?)artifacts of old waveforms when pressing <stop>. they vanish when changing the timebase of the stopped signal (forth->back->forth for example), OR when changing the waveform display from vectors to dots (and back).
so everytime the scope has to repaint the screen it resorts to the saved waveform data and it dumps the persistence.

might it be the persistence buffer is in fact the video memory where everytime before a new waveform is painted the video-mem content is dimmed by x% ?

whatever… it's no show stopper

[nitro2k01]

Maybe testing the µcurrents

Or maybe he's testing the new and completely revolutionary no-prefix-current, which can measure currents in the range 1 mA - 10 A. Just like with the µCurrent, the user can plug in a test current on one side and get a voltage reading on the two other  terminals, with the scale 1 V/A. I hear the no-prefix-current features a passive design, which has a heavily reduced complexity compared to the µCurrent and doesn't even require a battery.

[Sam__]

To remove any possible noise from the PSU you could test it with a battery. Then if it still oscillated you'd know the error was in the topology of the breadboarded circuit.

[Barny]

Sorry my english is realy bad.
And I'm a noob to.

Is it possible that the light (LED and fluorescent tube) makes some nasty stuff?
There are some long unshildet wires flapping around.

[gszo123]

Maybe testing the µcurrents

Or maybe he's testing the new and completely revolutionary no-prefix-current, which can measure currents in the range 1 mA - 10 A. Just like with the µCurrent, the user can plug in a test current on one side and get a voltage reading on the two other  terminals, with the scale 1 V/A. I hear the no-prefix-current features a passive design, which has a heavily reduced complexity compared to the µCurrent and doesn't even require a battery.

Hah, youre right, I ve got leaked schematics freshly stolen from EEVBlog Corp. for the new no-prefix-current. Seems like very robust mil-spec design. Dave's always worked on serious stuff

[TiN]

Dave butchery with poor TSSOP. Need some TSSOP-8 adapters? We have plenty of them here in Taiwan on local radio hobbyists shop.

I had built multi-channel current loads, even with simple LM324 usually things work pretty okay, but to be exact, i was using NFETs (IRF3205 for "low" power channels 0-30A and FB180SA10P for 0-50A). I was doing that for computer power supply ATE, driving 12 channels from 8bit DAC. Caps and proper passives are critical tho, but even on breadboard all should work fine.



Later i got rid off sense resistors, and was just using cable from power input board to FET (around 20-40cm AWG12) as a sense resistors. After calibration DAC/ADC, got accuracy around 1% at most of levels over 3A (again, that was designed for 20A range).



If need, i can post thread covering this ATE build somewhere in forum, just let me know where

P.S. Long time wanted ask you, from previous videos you have all nice gear, Agilent scopes, but mostly using Rigol gear. Why so, is there some deal with vendor, so u tied to that when doing videos? I'd use best available tools for experiments, if there is a choice. Just curious..

[kg4arn]

Thanks Dave,

I really like this kind of video.
Stirs up a lot of thought and insightful discussion.
Stretches my brain to think about what is going on and how to fix it.

Ed

[Frant]

I had built multi-channel current loads, even with simple LM324 usually things work pretty okay, but to be exact, i was using NFETs (IRF3205 for "low" power channels 0-30A and FB180SA10P for 0-50A). I was doing that for computer power supply ATE, driving 12 channels from 8bit DAC. Caps and proper passives are critical tho, but even on breadboard all should work fine.

I don't see any advantage of the circuit shown in the video compared to the well-known one with a MOSFET.

[TiN]

Well, having precision vref may help to have stable current over time, but i think still need opamp to drive FET/BJT from voltage reference.
Main point is not to keep fixed voltage over current sensor, but to keep fixed current flow thru sensor.
I had same mistake before, trying to fix voltage, but quicky find out that when resistor/FET heating up - resistance changes as well, and current drifts, if drive curcuitry have only
voltage feedback.

But issues Dave had on video are not related to this issue.

[jesuscf]

Dave,

After looking carefully at the breadboarded circuit I believe pin 8 of the LTC6655 is floating.  Shouldn't it be connected to GND?  The data sheet says: "GND (Pins 3, 5, 8 ) : Internal Function. Ground these pins."

Jesus

[LaurenceW]

Hmm, at the risk of appearing to give up on the LTC6655 device a little early, what about the MAX6225? I have a couple of these rigged as precision and untrrimmed voltage references. They are ROCK STEADY even on a breadboard, AND are available in good 'ole 8 pin DIPs

[smashIt]

ha!  i'm building a current-source for a power-LED with a max output of 1,4A and 5V supply-voltage at the moment   

the idea was simple:
led on +, then a mosfet (1010N in my case) then a shunt (0,22 ohms) to ground
voltage across the shunt goes into the inverting input of an opamp (lm358) and the noninverting comes from a voltag-divider
all neatly distributed across my breadboard

boy was it a failure 

when set to 0 the the led was lit
when set to max it didn't deliver the 1,4A
the values of the voltage-divider didn't fit the output
it even seemed that the opamp generated an offsetvoltage at its inputs 

what were the problems:
breadboards don't like high currents!
under load ground of the opamp und ground of the shunt drifted apart (~0,1V)
so i soldered the high-current stuff on a separate board  and kept the rest on the breadboard

the output of the opamp couldn't go low enough to switch the fet off
and it didn't go high enough to drive 1,4A through the fet
so i build up a voltage doubler and inverter powered by a single 555 
now the opamp is sourced by -3V and +7V

finaly the damn thing does what it's supposed to do 

[clifford]

Dave: I think a bypass cap between the force and the sense pins should do the trick. (The control loop becomes unstable because the transistor is to slow.) I've tried it in the simulation provided by Jay_Diddy_B and it seems to works there:

[ivan747]

Maybe testing the µcurrents

Or maybe he's testing the new and completely revolutionary no-prefix-current, which can measure currents in the range 1 mA - 10 A. Just like with the µCurrent, the user can plug in a test current on one side and get a voltage reading on the two other  terminals, with the scale 1 V/A. I hear the no-prefix-current features a passive design, which has a heavily reduced complexity compared to the µCurrent and doesn't even require a battery.

Sounds terrific! I'm gonna start saving for the Kickstater campaign... oh wait... 

[jesuscf]

Dave: I think a bypass cap between the force and the sense pins should do the trick. (The control loop becomes unstable because the transistor is to slow.) I've tried it in the simulation provided by Jay_Diddy_B and it seems to works there:

Once again.  The oscillations may be caused by the SPICE engine in LTspice and not the circuit topology itself.   Try this:

a) Right click on the line ".tran 10m startup", un-check "Start External DC voltages at 0V", click Ok.  The line should be now ".tran 10m".  If the power supplies start at 0V shouldn't the initial conditions for inductors and capacitors be also zero?
b) Click "Tools"->"Control Panel"->"SPICE" and change the default integration method to "Gear"; click Ok.
c) Run the simulation again.  Presto! No more oscillations.

Always beware of oscillations in SPICE simulations!

Jesus

[Jay_Diddy_B]

Hi Jesus and group,

While is possible that the oscillations result from the LTspice simulation algorithm.

If you take a circuit like a Wein bridge oscillator, in the absence of noise, initial conditions and transients the oscillation will not start.

If I take my model and I add the SPICE directive: .options method=gear

And change the transient simulation to .tran 10m



No oscillations are seen.

If I take the same model and then add a small step, only 10mV at time = 1ms, the circuit bursts into oscillations:






Would you expect this circuit to work? Running the decoupling capacitors into an op-amp configured for unity gain. Assume the op-amp is connected to suitable rails.



Jay_Diddy_B

[cellularmitosis]

Try substituting the ceramic caps for electrolytics, or add a very small series resistance (0.5-1R) in series with the ceramics. Sometimes their ESR is TOO LOW, forms an underdamped LC tank circuit with the PCB trace/wire/package/whatever inductance, causing ringing on transients and oscillation.

I was just reading about that earlier today!  http://www.acoustica.org.uk/t/3pin_reg_notes3.html

[nitro2k01]

Dave butchery with poor TSSOP. Need some TSSOP-8 adapters? We have plenty of them here in Taiwan on local radio hobbyists shop.

I'm pretty sure a certain PO box in Baulkham Hills wouldn't say no to a little loot.

http://www.eevblog.com/about/ (Scroll down a bit for the address.)

P.S. Long time wanted ask you, from previous videos you have all nice gear, Agilent scopes, but mostly using Rigol gear. Why so, is there some deal with vendor, so u tied to that when doing videos? I'd use best available tools for experiments, if there is a choice. Just curious..

Maybe he thinks Rigol gear is more expendable in case he would goof up completely? Just a guess.

[EEVblog]

Hmm, at the risk of appearing to give up on the LTC6655 device a little early, what about the MAX6225? I have a couple of these rigged as precision and untrrimmed voltage references. They are ROCK STEADY even on a breadboard, AND are available in good 'ole 8 pin DIPs

The LTC6655 is also rock stable on the breadboard as just a voltage reference, no problems there.

[EEVblog]

P.S. Long time wanted ask you, from previous videos you have all nice gear, Agilent scopes, but mostly using Rigol gear. Why so, is there some deal with vendor, so u tied to that when doing videos? I'd use best available tools for experiments, if there is a choice. Just curious..

I have explained this several times. There is no deal, or any pressure from the manufacturers at all. My Agilent scope sits on my working bench, which is an absolute mess and no room for shooting nice videos. My Rigol sits on my cleaner teardown/general video bench where I shot this video. So whatever bench I work at I use the nearest scope to hand. Same with the meters.

[TiN]

Ok, sounds fair, thanks for reply. I'm still new here..

[Oracle]

Rigol have some firmware issues... It's a trap for young players, that's why I invest lot of money on test gears... at least I can say that my instruments are telling me the truth.

[clifford]

Dave: I think a bypass cap between the force and the sense pins should do the trick. (The control loop becomes unstable because the transistor is to slow.) I've tried it in the simulation provided by Jay_Diddy_B and it seems to works there.

Once again.  The oscillations may be caused by the SPICE engine in LTspice and not the circuit topology itself.   Try this:

Fair enough. (I have only tried the spice simulation because it was already there and have not followed the discussion regarding it or have given it much thought.)

I nevertheless think that bypassing the force->sense path will fix the problem. (I'm not sure how much current the force output can source (could look it up in the datasheet) and what frequency/amplitude the oscillation was. It must source enough current to provide the necessary small signal feedback using the 1.25 Ohm as working resistor via the cap.. Its worth a try.)

[SeanB]

IIRC they were a mailbag thing he got. They are pretty common, you get them from all the major component suppliers and places like ADafruit and Sparkfun. Look for SMD breakout boards.

[trackman44]

Would't it be fun if there was a SMD breadboard? Just for SMD parts? That would be awesome. It would solve the SMD to DIP package woes. I'm just throwing an idea out there, hopefuly someone will pick it up and run with it

Will

[Rutger]

Hello Dave - I am interested to know where you source those little SMD breakout boards from on which the LTC6655 was soldered ?

Search for 'MSOP-8' on ebay and you will find plenty, most boards can be used for both MSOP-8 & TTSOP-8

[TiN]

Was making these couple years ago for own use at prototyping:)

SOT23-8 + SO8 + TSSOP8, but wider than DIP

[jesuscf]

If I take the same model and then add a small step, only 10mV at time = 1ms, the circuit bursts into oscillations:
Jay_Diddy_B

I agree.   The problem I was having is that the same LTspice circuit was not oscillating when simulated using Circuitmaker or Multisim.  If you simplify the current source to the circuit shown in "simple_ideal_circuit.png" you'll get the unrealistic oscillations shown in "simple_ideal_wave.png".

After I changed the ideal opamp with the LT1006A then I got the results as shown in "simple_LT1006A_circuit.png" and "simple_LT1006A_wave.png".  The results are very similar to the simulations with Circuitmaker and Multisim.  So there may be something going on in the way ideal opamps are modelled in LTspice.

I included the two LTSpice circuits in "current_source.zip".  By the way I changed the values of the components because my next step would have been breadboarding the circuit with parts I have at hand an see what is going on.  I may still assemble and test the circuit even though I have used this current source many times in the past (no inductor though) without any problems.

Jesus

[mkwired]

That distorted waveform bug on the Rigol scope occurs on earlier scopes as well.  I see it all the time on my DS1102E among other problems.

[calexanian]

This is how I would do it if I had the precision 1.25 ohm resistors.

[jesuscf]

I breadboarded the simple current source I simulated with LTspice (and Circuitmaker and Multisim) as shown in the post above.  I used a LM385-1.2 voltage reference.  Since I don't have 12.5 ohms resistors I made one by putting four 10 ohms resistors in parallel (2.5 ohms) with this parallel connected in series with a 10 ohm resistor.  I used an LM358 because I don't have a LT1006A opamp.  The big choke inductor is around 1 mH.  The transistor is a 2N3904.  The 5V comes out of a LM7805.  Here is a picture of the assembled circuit:



Here is a picture of the oscilloscope showing the voltage at the load after power on.  It flattens at approximately 1V.



Here is the same voltage from the Circuitmaker simulation.



Here is the same voltage from the LTspice simulation.



Once again, always be suspicious of simulations that oscillate just like that.

Jesus

[EEVblog]

This is how I would do it if I had the precision 1.25 ohm resistors.

That will have too much error. There is 2mA flowing in the sense pin, that's 0.2% error at 1A

[Jay_Diddy_B]

Hi,
I have put a designed a version of the current source. I have the parts on order. At this time the design is untested.

LTspice Model

Normally, I would do some LTspice modelling. Unfortunately, there is no model for the LTC6655 reference chip, so I have  gone straight to hardware.

Hardware Design

Schematic:



Board Design:

 The board is designed to replace the lid on a Hammond 1591B box. I will make this board on my LPKF Protomat.





4-Wire Resistor

The sense resistor is a critical component in this circuit. I have designed the board to take a Vishay VPR221Z series part. These are expensive. So I have designed a work around for the prototype. I will epoxy this resistor to a heatsink:






I will keep the group informed with progress, good or bad.

Jay_Diddy_B

[robrenz]

I have found that when the resistance of the connections used to parallel very low ohm resistors is not several orders of magnitude lower than the resistor values it pays to model the connection resistances also. IMO you need to model that resistor layout on spice with the full current and the all the actual trace resistances. When you look at the power dissipated in the trace resistances and the fact that they are copper with a terrible tempco I think it is going to give you poor precision.

[Rufus]

I will keep the group informed with progress, good or bad.

It will be bad.

You should add pads for capacitors between the chip OUT_F and chip GND and chip VIN and chip GND.

[Jay_Diddy_B]

Rufus,

I understand what you are saying. It certainly makes sense to add some extra pads for parts.

By doing the design in the open, I can bring the experience of the group to the problem.

My concern is that a capacitor from chip Vin to Chip ground, presents U2 with a capacitive load via C1. C1 and the capacitor from chip Vin to Chip ground would be in series. This may take some additional parts to make U2 work nicely with a capacitive load. I will check the datasheet.

With my LPKF machine I can make new boards very quickly.

This circuit is quite tricky....

Jay_Diddy_B

[Galaxyrise]

Since I only have experience with the Rigol scope, I thought that preserving the look of the screen when you hit run/stop was a feature.  If you want the result of just one waveform, use the Single button.  If you want to pause the screen, virtual phosphors and all, hit run/stop. 

Do other scope brands with virtual phosphors provide no way to stop the screen with the virtual phosphor effect still active?

[Rufus]

My concern is that a capacitor from chip Vin to Chip ground, presents U2 with a capacitive load via C1.

To be stable the LTC6655 requires a large low ESR capacitor on its output to ground. In normal circuits OUT_F and OUT_S are shorted, a capacitor between them isn't going to make it stable.

+ve currents into such a capacitor have to flow through the op-amp output and so the capacitor has the output impedance of the op-amp in series (and it is the output impedance at probably around 100kHz which matters). A capacitor from chip VIN to chip GND bypasses that but you are right it presents a large capacitive load to the op-amp. You might be able to feed the chip VIN through maybe 100R to present a load the op-amp can drive.

Taking the output capacitor all the way to power ground makes -ve currents flow through the op-amp and has the same problem + additional feedback from the load voltage.

I am doubtful that jacking up the ground of this voltage reference to make a current source like this is going to work out.

[Bored@Work]

Can the LTC2050 sink several mA? If not ...

[Rufus]

This with an LT6650 seems to behave and with some value adjustments (and 1 less diode) would probably work with an LTC6655.

[Jay_Diddy_B]

Rufus,

That will almost certainly work. There is also opportunity for controlling the loop gain.

Jay_Diddy_B

[cosmos]

This is how I would do it if I had the precision 1.25 ohm resistors.

That will have too much error. There is 2mA flowing in the sense pin, that's 0.2% error at 1A

I like calexanians proposal, perhaps one 1.25ohm resistor would be enough.
How about you buffer the feedback to OUT_S with one of your MAX4239s (or other low offset low noise op-amp) and let it handle the 2mAs?

Also, will there not be problems with the high side regulation when disconnecting the load?
Leading to the LT6655 floating with too low Vcc to operate just barely below the VIN rails? (can it start again when connecting the load?)

[calexanian]

This is how I would do it if I had the precision 1.25 ohm resistors.

That will have too much error. There is 2mA flowing in the sense pin, that's 0.2% error at 1A

I like calexanians proposal, perhaps one 1.25ohm resistor would be enough.
How about you buffer the feedback to OUT_S with one of your MAX4239s (or other low offset low noise op-amp) and let it handle the 2mAs?

Also, will there not be problems with the high side regulation when disconnecting the load?
Leading to the LT6655 floating with too low Vcc to operate just barely below the VIN rails? (can it start again when connecting the load?)

WOW!! 2ma draw on the sense pin? haven't they ever heard of a fet input? May as well use a 741 and a zener while they are at it!

[dac]

If you have an oscillator you need to reduce the amplification or add some phase shift! Add some RC filter in the amp input and/or output. In the input it will also protect the amplifier.

Which direction does the current flow in Australia? Here in Europe current flows from + to minus and electrons the reverse. Old US videos seem to flow the current reverse?!

Here is an informative video about how to make an oscillator (LC). You just do the reverse to get rid of it if an oscillator is not wanted.

--> http://youtu.be/a8casNPDvI4

[lewis]

WOW!! 2ma draw on the sense pin? haven't they ever heard of a fet input? May as well use a 741 and a zener while they are at it!

They probably use very low value feedback resistors internally to keep Johnson noise to a minimum.

[calexanian]

True.. Ground carbon shunt time! 

[EEVblog]

How about you buffer the feedback to OUT_S with one of your MAX4239s (or other low offset low noise op-amp) and let it handle the 2mAs?

Yes, that would be required to make this configuration a precision source.

[Rerouter]

and if you went one step further and made that buffer a differential amplifier configuration, bam your right back with what i suggested 2 pages ago :/

[nitro2k01]

Which direction does the current flow in Australia? Here in Europe current flows from + to minus and electrons the reverse. Old US videos seem to flow the current reverse?!

My theory is that we're converging toward a world where there's a relevant xkcd comic for every situation.

[Jay_Diddy_B]

Hi,

I think you are getting close the answer. It would be much easier to design and stabilize a current sink.

But, because this is going to be used in the Southern Hemisphere we need a current source.

Everybody knows that conventional current flows from the north pole to the south pole. Electrons must flow from the south pole to the north pole.

Or did I miss something when I was at skool? 

Jay_Diddy_B

[Napalm2002]

Did I miss something or am I just stupid. Why would it matter where the said item would be used. Or are you guys just messing around. Why would any electrical item care what part of the world it was used in. Besides the difference in mains voltage and frequency.

[EEVblog]

This is how I would do it if I had the precision 1.25 ohm resistors.

FYI, that solution works a treat, it you add a gate resistor, some decent capacitance across the shunt, and buffer the sense current.
Already shot a video on this, coming in due course. Supporters have already seen it.

[ali80]

well, my guess is too much gain on the darlington along with phase shifts causes instability in the control loop. maybe lowering the high frequency gain by connecting a 2.2n (or maybe bigger) across the force and sense pins could help.

[David Hess]

The output bypass capacitor, C2, needs to go to the same ground that the input bypass capacitor, C1, goes to and not the operational amplifier output.  The simplified LTC6655 schematic shows a common source output stage which is going to be picky about decoupling.

Adding a base-emitter shunt resistor or pulling the LTC6655 force pin down may be needed to force the output stage into class A and prevent the sink output transistor from becoming active because the LTC6655 ground is not going to be low impedance with the operational amplifier driving it.  Even that may not work.  The circuit may need to be changed so the operational amplifier is driving a low impedance capacitance to decouple the LTC6655 ground.  In the video I suspect that is where the low frequency oscillation with the larger capacitor originated.

I would decouple the collector of the transistor separately from the reference Vin and isolate them somewhat.  This would not be the first time I have seen the simple current gain stage couple back into the reference supply causing stability problems.  The relatively high impedance power and ground connections of the solderless breadboard are going to exasperate this issue.

I would be surprised if SPICE models the above problems well without doing enough work understanding the circuit to not need to use SPICE and I see no LTC6655 macromodel anyway.

It may be simpler to use a more traditional precision current source design with a separate reference, error amplifier, and single ended output stage.  The LTC6655 is going to be finicky because of its common source output stage.