low drift low noise current source

[Andree Henkel]

I want to get some hints on my low drift + low noise current source. See attached PDF.
It is not yet final schematic. I know there are probably at least two flaws. Discussed below.
Purpose of circuit is to drive a laser-diode, that is wavelength stabiliced using an external grating, and temperatur satbilized using peltier. I´m somewhat space constraint, so the circuit drawn roughly already uses up the pcb estate I got.

Reference selected is LTC6655, non LN (Low Noise) type, output voltage and package selection is not yet final. For now I go for 5V for maximum signal level across sense resistor. And LS Packacke for better stability.
I decided not to use LN version, as by filtering output with lowpass (R2+C1), I also filter widedand output noise of reference output buffer. User Andreas here wrote something of higher noise out of LTC6655 or LTC6655LN at higher supply voltages, but I couuld not find it. I might do something to account for that.

Potentiometer P1 adjusts Current setpoint Zero to Vref/Shuntresistor R1. Optionally U3 sets a minimum current above Zero.

Lowpass R2+C1 filter noise coming from Reference, Pot, optional U3 and intrinsic noise of R2, It also should lowpass filter current noise of noninverting input of U2. C13 is used to double Value of C2.
Value of R2 is a tradeoff. Higher Value gives lower Noise from U1, P1, U3, R2 as long as R2´s noise does not approach Noise of U1 which is the dominating source. Higher Values give higher noise from 1/f current noise of U2 and higher drift for U2 input bias drift * R2. Voltage noise from Wideband current noise of U2 should roughly be independent of R2 as R2, C1 act as lowpass for this current noise.
In libbrecht Hall circuit https://www.google.com/search?client=firefox-b-d&q=libbrecht-hall+current+source Resistance looking from noninverting Input of Control-OPV is 11k assuming Poti is in center position, so with 10k for R2 seems a good starting point for me as long as I stick with LT1128.

I choose LT1128 as I don´t need the speed of the LT1028.
However there is probably a flaw with with the negative Supply at Ground.
I don´t know yet at which current the diode will have to be driven, so it might actually work. But for lower currents it most probably won´t.
I do have a -5V Supply available from a DC/DC-Converter at a connector of the board, and could change that to a postregulated -4.1V with some manual rework on other board that is connected there (it contains the temperature control and other necessary stuff).
I would actually prefer to get rid of the DC/DC-Converter completely for this project, to not get interference from this source, but keep unipolar supply for the lasermodul.
Maybe there is a reasonable good Replacement for U2 with input range closer to negative supply?

Current measurement shunt R1. Currently maximum Power will be 1,8W. I probably will use a metalfoil 4wire TO222 type.
However as can be seen, it is currently used in a three wire configuration.
I want to avoid extra components for differential shunt sensing feedback.
Is there a way to accomplish that and still use the fourth wire?

R8, R15, R16, R15, C29, C17 are there to help in case of stability problems.

So the noise at +Input of U2 is kept as small as resonable.
Wideband Noise of current source mainly is from input noise voltage of U2 and from Sense Resistor R1. R16 should be kept zero for Noise purpose, or better shorted in Layout for low drift.
For 360mA R1 is maxed out, at 13,9Ohm Voltage Noise of U2 dominates. So roughly wideband current noise schould be U2 Voltage Noise divided by R1.
This should be valid roughly for the regulation bandwidth of the current source.

The purpose of the components inductor L5, Ferrit L3, capacitors C22..C25 is to steer the current noise above corner frequency set by L5 and C22+C23 around the laser-diode.

The second Flaw of the circuit: 11.3V are probaly short of what is necessary. Depends on beeing lucky of having a diode specimen with low voltage drop. I want to create th 11.3V using a LT3045 out of 12V. I´think, I´ll either have to switch to 15V or reduce the Reference Voltage.

[Kleinstein]

The 5 V range for the maximum voltage drop at the current setting resistor is a bit on the high side for a 350 mA range current. That is just too much heat loss to be practical for good stability.
The question is more if 2.5 V is still high. It depends a bit on the impoartance of noise at the different frequency ranges. This also effect the amplifier choice.

The LT1128 is not single supply. So one would need an extra negative supply. It may anyway not be the best choice (high power, quite some current noise, high speed).
There are not that many single supply precision OP-amps, except for most of the Auto zero types, but these have there own noise spike porblems. So good for the low frequency part, but not that great for the higher frequencies. There is the LT1006, but a bit noisy, though still not too bad compared to the reference.

The ground connection with a 4 wire resistor would also need some care - e.g. the reference and pot etc. connected to the lower voltage sense path. Ideally one would have current compensation and thus need a negative supply anyway.

The pot for setting the current could be a problem, as pots are usually not that stable.

One would definitely want some protection for the current regulator not going all the way on, when the diode is not connceted. This would lead to overcurrent when connecting the diode, and possibly if the supply comes back after a short brown out.

[Andreas]

User Andreas here wrote something of higher noise out of LTC6655 or LTC6655LN at higher supply voltages, but I couuld not find it.

Hello,
My measurements for 1/f noise (0.1-10 Hz) are here:
https://www.eevblog.com/forum/metrology/diy-low-frenquency-noise-meter/msg986718/#msg986718

with best regards

Andreas

[NWerner]

Hello,

As Kleinstein pointed out: 5V is rather a high voltage level.  I guess that thermal poblems will dominate over stochastic noise.

Some additional points:
- I don't see any connection between AGND and LGND. Watch out for common impedance coupling! Check for unwanted resistance between sense and force pins of resistor.
- Laserdiodes are easily destroyed (even by short) overcurrent events. Some protection may be needed. I would be uncomfortable with this potential LC tank circuit.  An accidental connection between laserdiode cathode and GND will likely destroy laserdiode.
- laser output and laser current are not strictly proportional. You may end up with an ultrastable current which results in lousy laser output.
 

[Andree Henkel]

The 5 V range for the maximum voltage drop at the current setting resistor is a bit on the high side for a 350 mA range current. That is just too much heat loss to be practical for good stability.

For now I´ll go down to 3.3V for reference, and down to 11V for supply, so I should have enough margins in making the 11V from 12V and for voltage drops within the current source. As reult I can also go with off the shelf 10Ohm Resistor with 330mA current. 

The LT1128 is not single supply. So one would need an extra negative supply. It may anyway not be the best choice (high power, quite some current noise, high speed).
There are not that many single supply precision OP-amps, except for most of the Auto zero types, but these have there own noise spike porblems. So good for the low frequency part, but not that great for the higher frequencies. There is the LT1006, but a bit noisy, though still not too bad compared to the reference.

For now I will use LT1128 and negative supply

The ground connection with a 4 wire resistor would also need some care - e.g. the reference and pot etc. connected to the lower voltage sense path. Ideally one would have current compensation and thus need a negative supply anyway.

Thank you, I did redraw my schematic to do so, introducing additional net "RGND" and wiring it that R1 Pins 3,4 provide starground with net LGND. I also added Resistor R25 between negative Supply and RGND to compensate for current flowing into Pin 3 of Shunt R1.

The pot for setting the current could be a problem, as pots are usually not that stable.

I did consider using a foil type pot. Are they worth it, regarding Resolution and Stability? Price for me is less important, I got more a problem with PCB-Space and especially through hole parts.

One would definitely want some protection for the current regulator not going all the way on, when the diode is not connceted. This would lead to overcurrent when connecting the diode, and possibly if the supply comes back after a short brown out.

The diode will be permanently wired to pcb, not attached via connectors.

[DavidAzulay5000]

combo of LT1634 AND LT3092
For example in the project I designed and built a milliohm meter

[Andree Henkel]

Some additional points:
- I don't see any connection between AGND and LGND. Watch out for common impedance coupling! Check for unwanted resistance between sense and force pins of resistor.

Connection between AGND and LGND is at another Page of schematic at LT3045 generating (now) 11V from 12V.

- Laserdiodes are easily destroyed (even by short) overcurrent events. Some protection may be needed. I would be uncomfortable with this potential LC tank circuit.  An accidental connection between laserdiode cathode and GND will likely destroy laserdiode.
- laser output and laser current are not strictly proportional. You may end up with an ultrastable current which results in lousy laser output.

I use the current limit functionalitiy of LT3045.
I´ll have to check circuit for unwanted Oszillation of the LC-Filter.
The laser diode will be temperature stabilized with Peltier and wavelength+linewidth stabilized with external grating.

Indeed one tricky point will be to tune current + temperature to get operation at a single mode.

[dietert1]

Just to mention it: Replace the potentiometer by a DAC for better repeatability and reliability. If you use a MCU for current monitoring, you can use it to control the current, too. In the end you may want an automated control loop to maintain the laser in single mode.

Regards, Dieter

[n_haku]

 - as already mentioned by dietert1, do not use cermet or wirewound pots like in your schematic, at least metal foil, but dac is much better choice, there are a lot of that will which match ltc6655 performance nowdays. If you feel they a bit pricely, you always have this sync dac schematic, its pretty nice, as can deliver 1E6 steps in 32Hz with small filter cap.
 - you are welcomed to read AN90 by Jim Williams ( and all rest of his AN's ), lasers are not easy load, do not understimate anything.

[r6502]

hi Andree,

the choosen MOSFET DMG6402LVT seems to be optimised for switching applications, and not for linear operation. Data sheet tells you in Applications section: DC-DC Converters,  Power Management Functions, Backlighting. Try to find one type designed for linear applications.

Shunt resistor do I understand it correct, 13,9 Ohms? That would generate a voltage drop of about 5V resulting in a loss of 1.8W. That is a lot for for a shut resistor. Go to resistance of below 1Ohm to get as low power in the shunt as possible.

With the actual setup there is not a big safety margin for current regulation for the MOSFET. 5V at the shunt + 5.5V at the laser diode are 10.5V that results in 0.8V max voltage drop for the MOSFET. This does not take into account voltage drop from PCB traces and other components in the signal path.

The positive supply of 11.3V looks like, you are generating a negative supply with the voltage drop from a diode from 12V supply. Also take into account, that a  voltage drop on d diode can be greater than 0.7V in reality.

Guido

[r6502]

Hello again,

one thing I forgot to address laser diodes are more sensitive to ESD. So adding TSV diodes to your schematic would be a good idea and will help to let the laser diodes survive.

Do you knew IC-Haus, they have a lot of integrated laser drivers. and thy have integrated protection for the laser diodes.

Guido

[Andree Henkel]

Just to mention it: Replace the potentiometer by a DAC for better repeatability and reliability. If you use a MCU for current monitoring, you can use it to control the current, too. In the end you may want an automated control loop to maintain the laser in single mode.

Regards, Dieter

I have decided to add a miniature 16Bit DAC with SPI Interface. I do have a free SPI Port, but µC is on other board that is to be reused during prototype phase, unfortunately it is not on the (Legacy Connector)Interface between the two boards. So I´ll need to add some small 4 Pin for the 3 Pin interface and GND

I still need the pot for testing the current source without doing firmware (I´m Hardware Engineer), DAC Control can be only later step and would be done by one of my coworkers.

[Andree Henkel]

hi Andree,

the choosen MOSFET DMG6402LVT seems to be optimised for switching applications, and not for linear operation. Data sheet tells you in Applications section: DC-DC Converters,  Power Management Functions, Backlighting. Try to find one type designed for linear applications.

I´ll look for another MOSFET choice, also I´m not yet sure, if I go for SMD or Trough-Hole + wires -> to reduce power entry into board

Shunt resistor do I understand it correct, 13,9 Ohms? That would generate a voltage drop of about 5V resulting in a loss of 1.8W. That is a lot for for a shut resistor. Go to resistance of below 1Ohm to get as low power in the shunt as possible.

I reduced shuntsize to 9R17, I´ll most probably go down to 6.94.
To minimize current-noise level it is not the correct choice to minimize the shunt

With the actual setup there is not a big safety margin for current regulation for the MOSFET. 5V at the shunt + 5.5V at the laser diode are 10.5V that results in 0.8V max voltage drop for the MOSFET. This does not take into account voltage drop from PCB traces and other components in the signal path.
The positive supply of 11.3V looks like, you are generating a negative supply with the voltage drop from a diode from 12V supply. Also take into account, that a  voltage drop on d diode can be greater than 0.7V in reality.

See post #4

Guido

Thank you, Andree 

[Andree Henkel]

I did contact Powertron (Vishay), I used their Resistors before.
They will quote resistors trimmed to tbd. custom value / 1% in very small quantities, via their Distributor Bader.

Resistor Types see datasheets attached.

[Andree Henkel]

Hello again,

one thing I forgot to address laser diodes are more sensitive to ESD. So adding TSV diodes to your schematic would be a good idea and will help to let the laser diodes survive.

this is an integrated laser module, no connectors to the diode, diode is mounted into temperture controlled aluminium block, diode wires are extended with wires, those wires are soldered to current source pcb. Soldering is done in ESD control environment.

Do you knew IC-Haus, they have a lot of integrated laser drivers. and thy have integrated protection for the laser diodes.

Guido

yes I know ICHaus 

[Kleinstein]

With only a rather low voltage (0.5-2 V range) at the MOSFET even the moder MOSFETs designed for switching should be OK. The fet choice gets tricky with higher voltages, like > 20 V.  Maybe still avoid low voltage types with extra low capacitance.

The resistor noise gets more important with a small shunt. Still the voltage reference noise is usually still way larger than the noise of the amplifier and the resistor. With the bandgap type references they are internally usually 1.2 V (maybe 2.4 V in a few cases) and a higher ref. voltage would be with chip internal gain. So noise wise a higher voltage does not help much.

[Andree Henkel]

With only a rather low voltage (0.5-2 V range) at the MOSFET even the moder MOSFETs designed for switching should be OK. The fet choice gets tricky with higher voltages, like > 20 V.  Maybe still avoid low voltage types with extra low capacitance.

Originally with the higher shunt voltage I had to shoot for kind a low treshold voltage. Gate Capacity I did not consider important in my selection process.

The resistor noise gets more important with a small shunt. Still the voltage reference noise is usually still way larger than the noise of the amplifier and the resistor. With the bandgap type references they are internally usually 1.2 V (maybe 2.4 V in a few cases) and a higher ref. voltage would be with chip internal gain. So noise wise a higher voltage does not help much.

yes at the reference output the ppm noise. but then there is my external lowpass. still after it the ppm noise is roughly independent of voltage level.
But: With my current values the BW of LP after ref is about 1/1000 of the current source internal LC LP. So I should get the reference noise down to levels where the current source intenal noise sources (LT1028, Shunt) get relevant.
And to my understanding I do reduce the noise current density level of the current source, as long as I increase shunt, at least as long as voltage noise density of shunt is not yet same as of LT1028. Actually even beyound as noise voltage rms then scales ~sqrt(R), but resulting Inoise goes ~sqrt(R)/R

But ofcourse as you rightly point out: this is academic as long it´s not thermal drift stable feasible.

[Andree Henkel]

- as already mentioned by dietert1, do not use cermet or wirewound pots like in your schematic, at least metal foil

I´ll change pot

but dac is much better choice

yea, I´ll add a 16Bit SPI type miniature DAC

you are welcomed to read AN90 by Jim Williams ( and all rest of his AN's )

always nice to remember.

lasers are not easy load, do not understimate anything.

well, yea I know this is not my first laserdriver design. 

[maat]

I´ll look for another MOSFET choice, also I´m not yet sure, if I go for SMD or Trough-Hole + wires -> to reduce power entry into board

Get a beefier FET, you are awfully close to the SOA limit. See Fig. 11 of the datasheet. And no, MOSFETs are not immune to thermal runaway, when used in saturation, only when used in the ohmic region. As soon as you go above 25 °C ambient, that poor little FET will likely be toast. And as Kleinstein said, stay away from 'logic level' FETs. Their gate capacitance is enormous. Make sure you the FET is in saturation at 350 mA and 2V. Otherwise the output impedance of the CCS will go down the drain and the current noise will go through the roof. The latter requirement is tough with that little headroom (11V - 5.5V - 3.3V - x).

The LT1128 is imho a poor choice. There is an ugly voltage noise bumb out at around 400 kHz. The LT1x28 is nice for audio, but not here. Try the AD797 instead.

Noise wise, the limit is likely the mechanical noise of the external cavity, not the current source unless you totally screw it up. Gratings are nice to tune, but horrible in stabilty. What linewidth are you aiming for? Or just 'single mode'?

[Kleinstein]

The linked FET is indeed a poor choice, rather high in capacitance and low R_on, but a tiny case. So it can not get rid well of heat.
There is nothing really wrong with a logic level FET, the main problem is the small case.

I would more look at a SOT223 case and with less capacitance and higher on resistance (e.g. 0.5-1 ohm range). E.g. ZVN4206GV should be OK if there is enough headroom for the gate voltage.

[Andree Henkel]

Get a beefier FET, you are awfully close to the SOA limit. See Fig. 11 of the datasheet. And no, MOSFETs are not immune to thermal runaway, when used in saturation, only when used in the ohmic region. As soon as you go above 25 °C ambient, that poor little FET will likely be toast.

I´ll change FET to other package for higher Power handling

And as Kleinstein said, stay away from 'logic level' FETs. Their gate capacitance is enormous.

I think that problem would be managable there are components to play with in case OPA is troubled by FET gate capacity, but ofcourse it´s nicer to don´t have to deal with that.

Make sure you the FET is in saturation at 350 mA and 2V. Otherwise the output impedance of the CCS will go down the drain and the current noise will go through the roof. The latter requirement is tough with that little headroom (11V - 5.5V - 3.3V - x).

yes the 2V should be there. I´ll have already decided to go down 3.3V->2.5V as i plan to add DAC and now selected type: AD5683(R), which only has available 2.5V or 5V output range

The LT1128 is imho a poor choice. There is an ugly voltage noise bumb out at around 400 kHz. The LT1x28 is nice for audio, but not here. Try the AD797 instead.

I did look at the AD797, and did not choose it because of it´s much higher input bias current -(drift) which is problematic in combination with the RC lowpass between reference and the IC.
the 400kHz should hopefully not be a problem, while it wll be well inside regulation bandwidth of current source, the LC-lowpass should steer it around the laser diode.

Noise wise, the limit is likely the mechanical noise of the external cavity, not the current source unless you totally screw it up. Gratings are nice to tune, but horrible in stabilty. What linewidth are you aiming for? Or just 'single mode'?

Just single mode operation. Grating is a volume bragg grating VBG. VBG is not tunable. Target specification for laser linewidth is 0,1pm.

[Andree Henkel]

The linked FET is indeed a poor choice, rather high in capacitance and low R_on, but a tiny case. So it can not get rid well of heat.
There is nothing really wrong with a logic level FET, the main problem is the small case.

I would more look at a SOT223 case and with less capacitance and higher on resistance (e.g. 0.5-1 ohm range). E.g. ZVN4206GV should be OK if there is enough headroom for the gate voltage.

with the other changes: added DAC+Conector to tunnel SPI to µC, change POT to through hole I´m basically already screwed PCB space wise 
So I´ll have to find a way for more PCB real estate.
Then I´ll will be also able to use SOT223 case

[Andree Henkel]

I choosed 16Bit SPI DAC:
AD5542A because of low low frequency and wideband noise, unbuffered.
AD5541A is onobtanium
5542 additional has matched resistor network for bipolar output (with ext. OP)
I´ll repurpose to create neg. reference voltage. so far not needed, but that may change

not shure yet, if I´ll use the lower pin count or version with force/sense Pins

[maat]

The LT1128 is imho a poor choice. There is an ugly voltage noise bumb out at around 400 kHz. The LT1x28 is nice for audio, but not here. Try the AD797 instead.

I did look at the AD797, and did not choose it because of it´s much higher input bias current -(drift) which is problematic in combination with the RC lowpass between reference and the IC.
the 400kHz should hopefully not be a problem, while it wll be well inside regulation bandwidth of current source, the LC-lowpass should steer it around the laser diode.

Noise wise, the limit is likely the mechanical noise of the external cavity, not the current source unless you totally screw it up. Gratings are nice to tune, but horrible in stabilty. What linewidth are you aiming for? Or just 'single mode'?

Just single mode operation. Grating is a volume bragg grating VBG. VBG is not tunable. Target specification for laser linewidth is 0,1pm.

I see. If you consider the bias current change an issue (~2 nA/K), then do make sure, you are not using tantalum capacitors for the filter, because their change in leakage is typicallly an order of magnitufe higher. The same goes for the leakage in absolute numbers.

The singlemode 0.1 pm target is easy to hit. I will look up the numbers on our ECDLs to give you a ballpark figure

[Andree Henkel]

I see. If you consider the bias current change an issue (~2 nA/K), then do make sure, you are not using tantalum capacitors for the filter, because their change in leakage is typicallly an order of magnitufe higher. The same goes for the leakage in absolute numbers.

yes, I specified foil type capacitor, type Rubicon 16MU226MD35750 22µ/16V - mainly because of no piecoelectic noise (ceramic), but was aware to, that electrolytics would be also problematic because of leakage.

The singlemode 0.1 pm target is easy to hit. I will look up the numbers on our ECDLs to give you a ballpark figure

That would be very much appreciated.
ECDL: external cavity diode laser?