best way to create a 100mV negative rail ?

[rob77]

Hi Guys,

probably someone has the solution for this   i need to extend a R2R buffer opamp for an ADC to true 0V output. so i need an approx 100mV negative rail (low noise) for that opamp. voltage needs to be low enough to not exceed the ADC's input range (max 300mV below ground) but high enough to overcome the opamp's saturation voltage (15mv) + to have some headroom.
there is a lm7705 out there , designed exactly for this job, but available only in vssop package and i really don't like that
i was thinking of a 7660 charge-pump making a -5V rail  and then create the -100mV with an opamp ( inverting amp making the -100mV from a TL431 divided down to 100mV)
if there is a simple solution involving "normal" sized packages (soic) then i'm more than happy to listen

[Macbeth]

Schottky diode?? 

[rob77]

yes that was my first thought but that would mean the whole circuit is "floating" on top of that shottky - so the current through it would be varying a lot, and the voltage as well.. i have a feeling it would introduce some noise throwing away a extra bit or two from the ADC readings..

[Andreas]

Hello,

I would use a better RRIO Op-Amp (eg. OPA340) which is good for a 12 Bit ADC and 5V supply.

With best regards

Andreas

[fcb]

I've done this before on a design that needed to sense to 0V, used a spare pin on a micro to drive a very basic chargepump, then BAS16 diode to clamp the rail to -0.35v for the OPAMP.

[rob77]

Hello,

I would use a better RRIO Op-Amp (eg. OPA340) which is good for a 12 Bit ADC and 5V supply.

With best regards

Andreas

HI, thanks for the suggestion. but in my case it's either a  LTC2400 ADC or second option would be MCP3550 ADC and considering opamps LTC2050 or AD8628. it would be no problem to change the opamp (it's a one-off project) but i'm afraid there is no opamp on this planet going close enough to ground for 20+ bit resolution.

[dadler]

Mount the lm7705 on an adapter board?

VSSOP is not difficult to solder.

There is also the LM2776.

[technotronix]

Buck Converter?

[Siwastaja]

VSSOP is not difficult to solder.

+1. Just do it OP, if that part is suitable otherwise.

[Kleinstein]

One could still float the whole circuit on a diode (no need to be schotky) and than add an OP for regulating the negative supply (e.g. -200 mV).

The diode would also work as reverse voltage protection (at least for most of the circuit).

[Zero999]

With many CMOS op-amps, the output voltage will be zero, providing the output stage doesn't sink any current.

For an op-amp with a CMOS output stage, the saturation voltage at low currents can be predicted if you know the on resistance of the output MOSFETs. Unfortunately this isn't given on the datasheets. Out of the op-amps mentioned in this thread, the OPA340 is the only one which gives a clear graph showing the output saturation, when the MOSFETs are operating in the Ohmic region, when the voltage drop is directly proportional to the output current.

When operating off 5V, the N-MOSFET on the OPA340 has an on resistance of between 7.5 and 30, depending on the temperature.

20 bits of accuracy with a full scale of 5V, is 4.77µV per count. Assuming a typical output resistance of 15R, the maximum current which the op-amp can sink and still remain under one count is 318nA, so the maximum resistor you can connect between the output and 5V is 15.73M.

However the OPA340 isn't suitable for this application, as the offset voltage is typically 150µV.

[richard.cs]

For <100 uA and no more than a few volts I have used small solar cells illuminated with LEDs. Efficiency is terrible but it works reasonably well and is cheap enough (I used cells from a poundland garden light).

[Zero999]

What about just using a small battery? At 100µA it may never need replacing, especially if it's in a sealed container with a sachet of silica gel to keep it dry.

[danadak]

If you use a UP then Cypress PSOC has a DelSig A/D that can handle as
its conversion range 100 mV outside either rail.

One way is to use a pair of Rs in a divider at A/D input terminated to Vdd.
See attached for a calculator.

Another way is to use a clk to gen a - supply V, and then an OpAmp
to scale/offset to whatever you need. See techniques here -

https://www.dropbox.com/s/k1yn4m7heaixvmg/ADinput%20negative%20inputs.zip?dl=0


Regards, Dana.

[rob77]

thank you for all the suggestions, i really appreciate all the input from you !

considering all the facts i think i'll stick to that lm7705 in vssop package... it seems to be the best solution apart from that tiny package..  and the reason why i hate those tiny packages is not the soldering - i have no problem with that.. but using such small packages means to switch to 0402 (possibly 0603) passives around it and that's what i would like to avoid. usually i'm working with 1206 and 0805 components.

[edavid]

Why does the ADC input have to go down to 0?   You could offset the input and then subtract the offset from the ADC output.

[rob77]

Why does the ADC input have to go down to 0?   You could offset the input and then subtract the offset from the ADC output.

offsetting the input would mean another problem - how to create the offset voltage stable enough to not degrade the ADC readings. we're talking about 20+ bits of resolution that's down to 4uV resolution (but ltc2400 is technically a 24bit delta sigma ADC going down to sub uV resolution). it's far easier to provide a little negative juice to the buffer opamp than offsetting the input without introducing errors.

[krivx]

Can you use the charge pump and add a clamping circuit to limit signals to -100mv pk?

[edavid]

Why does the ADC input have to go down to 0?   You could offset the input and then subtract the offset from the ADC output.

offsetting the input would mean another problem - how to create the offset voltage stable enough to not degrade the ADC readings. we're talking about 20+ bits of resolution that's down to 4uV resolution (but ltc2400 is technically a 24bit delta sigma ADC going down to sub uV resolution). it's far easier to provide a little negative juice to the buffer opamp than offsetting the input without introducing errors.

Just add an autozero switch at the input so you can measure the offset.  (This is always a good idea since you will certainly have more than 4uV offset between your input ground and ADC ground.)

Anyway, don't you have the same problem with the reference?

[rob77]

Why does the ADC input have to go down to 0?   You could offset the input and then subtract the offset from the ADC output.

offsetting the input would mean another problem - how to create the offset voltage stable enough to not degrade the ADC readings. we're talking about 20+ bits of resolution that's down to 4uV resolution (but ltc2400 is technically a 24bit delta sigma ADC going down to sub uV resolution). it's far easier to provide a little negative juice to the buffer opamp than offsetting the input without introducing errors.

Just add an autozero switch at the input so you can measure the offset.  (This is always a good idea since you will certainly have more than 4uV offset between your input ground and ADC ground.)

Anyway, don't you have the same problem with the reference?

no, the reference is max6241AC - that should be stable enough 1ppm/C typ. (2ppm/C max) - that would mean approx 4uV (8uV max) drift per deg C.

to create that stable offset voltage for the input i would have to use extremely expensive resistors (tempco , noise) , and basically over-complicating the whole thing...

this way it's simple and nice - chopper/zero-drift opamp on the input feeding the ADC... the only thing it needs to go down to zero is a little of negative bias

[Marco]

yes that was my first thought but that would mean the whole circuit is "floating" on top of that shottky - so the current through it would be varying a lot, and the voltage as well.. i have a feeling it would introduce some noise throwing away a extra bit or two from the ADC readings..

How many bits ADC are you using that the Schottky diode with a capacitor in parallel could punch through the PSRR of the opamp?

If whatever you are measuring is floating relatively to your normal power supply you can't really go simpler than this, one Schottky and one decoupling cap.

[rob77]

yes that was my first thought but that would mean the whole circuit is "floating" on top of that shottky - so the current through it would be varying a lot, and the voltage as well.. i have a feeling it would introduce some noise throwing away a extra bit or two from the ADC readings..

How many bits ADC are you using that the Schottky diode with a capacitor in parallel could punch through the PSRR of the opamp?

If whatever you are measuring is floating relatively to your normal power supply you can't really go simpler than this, one Schottky and one decoupling cap.

you're right !   the LTC2050 has a typical PSSR of 130dB  (worst case 115dB over whole temp range) a 100mV jump on the supply rail will translate to approx 0.2uV change on signal when considering 115dB PSSR.

but the lm7705 chips are already traveling from farnell and i have a lot of PCB space to spare ,  so going to use them.

[David Hess]

If your current requirements are 10s of microamps, then I can do it with a single bipolar transistor and a resistor.  This method makes for a good but obscure electronics engineering puzzle.

For slightly higher currents of 100s of microamps, an optocoupler and resistor can be used in place of the solar cell and light source which richard.cs suggested.  If you have a suitable bias supply, then no resistor is needed.

If you have a clock source somewhere, then buffer it and use a charge pump with a schottky diode shunt regulator.  Adjust the charge pump impedance for best efficiency.