Author Topic: Discrete Capacitive Divider-Circuits  (Read 3536 times)

0 Members and 1 Guest are viewing this topic.

Offline Echo88Topic starter

  • Frequent Contributor
  • **
  • Posts: 821
  • Country: de
Discrete Capacitive Divider-Circuits
« on: December 10, 2018, 05:15:39 pm »
Since the LTC1043 has its limitations in reality and as a model in LTSpice, i begun to simulate discrete capacitive dividers.
My discrete capacitive-divider simulations, which you can find attached, work quite well. Attachment contains /2, /3, /4 and /10-dividers. However: i cant get the newest circuit on the right side to work. Maybe someone can help?
Description:
/2, /3 and /4-Dividers are copied from the LTC1043-examples. The /10-Divider contains 10 caps, after charging and equalizing they are sequentially paralleled to the cap C13. The nonworking /10-divider should do the same more elegant, but it seems i built it wrong.
 

Online Andreas

  • Super Contributor
  • ***
  • Posts: 3222
  • Country: de
Re: Discrete Capacitive Divider-Circuits
« Reply #1 on: December 10, 2018, 06:34:41 pm »
The nonworking /10-divider should do the same more elegant, but it seems i built it wrong.

Its not more elegant.
The first capacitor sees 9 switches (RDS,ON) when paralleled to the last capacitor.
the pre-Last sees only one Switch.
So the time constant / settling time for each capacitor is different -> reduced accuracy or longer settling time.

I guess that the "working" problem has to do with the threshold voltages of the switches and the wiring of the clock signals.

with best regards

Andreas




 

Offline Echo88Topic starter

  • Frequent Contributor
  • **
  • Posts: 821
  • Country: de
Re: Discrete Capacitive Divider-Circuits
« Reply #2 on: December 10, 2018, 08:31:01 pm »
Thanks for the hint. The settling-time approach sounds correct, however in the simulation the newer circuit has way faster settling. Maybe the simulation-comparison isnt correct, because the paralleling switches in the old circuit could run faster. Anyway: Found the problem and corrected the circuit, the 10V-input should of course be switched away when the caps are equalized. Otherwise it will disturb the process. Attached the new corrected version.
Old approach: 32switches/11caps New approach: 28 Switches/10caps  :)
 

Offline SilverSolder

  • Super Contributor
  • ***
  • Posts: 6126
  • Country: 00
Re: Discrete Capacitive Divider-Circuits
« Reply #3 on: December 11, 2018, 03:48:01 am »
What makes a capacitive divider better than a resistive one?
 

Online Andreas

  • Super Contributor
  • ***
  • Posts: 3222
  • Country: de
Re: Discrete Capacitive Divider-Circuits
« Reply #4 on: December 11, 2018, 05:29:01 am »
Hello,

See here a comparison of ratio stability of a DSMZ (Vishay Z-Foil) resistive divider and a LTC1043 2:1 divider over temperature:

https://www.eevblog.com/forum/metrology/t-c-measurements-on-precision-resistors/msg521240/#msg521240

and another here:

https://www.eevblog.com/forum/metrology/t-c-measurements-on-precision-resistors/msg528192/#msg528192

with best regards

Andreas
 

Offline Echo88Topic starter

  • Frequent Contributor
  • **
  • Posts: 821
  • Country: de
Re: Discrete Capacitive Divider-Circuits
« Reply #5 on: December 11, 2018, 10:04:39 am »
I found your temperature measurements of the LTC1043 very interesting Andreas.
Did you also try to compensate the offset error of a few ten µV, which is typical for the LTC1043, for example by shielding the capacitors or varying the Vcc?
I havent shielded the caps in my LTC1043-/2-Testdivider yet, but varying the Vcc couldnt compensate the about -40µV offset which i had.
 

Offline branadic

  • Super Contributor
  • ***
  • Posts: 2378
  • Country: de
  • Sounds like noise
Re: Discrete Capacitive Divider-Circuits
« Reply #6 on: December 11, 2018, 10:17:38 am »
Even though I'm not Andreas, you can find a picture of the divider on the web, which shows one of the caps being shielded.



-branadic-
Computers exist to solve problems that we wouldn't have without them. AI exists to answer questions, we wouldn't ask without it.
 

Offline Echo88Topic starter

  • Frequent Contributor
  • **
  • Posts: 821
  • Country: de
Re: Discrete Capacitive Divider-Circuits
« Reply #7 on: December 11, 2018, 10:36:00 am »
Yeah, question is: Can the shielding and/or varying Vcc completely compensate the offset-error?
 

Online Andreas

  • Super Contributor
  • ***
  • Posts: 3222
  • Country: de
Re: Discrete Capacitive Divider-Circuits
« Reply #8 on: December 11, 2018, 01:45:07 pm »
I havent shielded the caps in my LTC1043-/2-Testdivider yet, but varying the Vcc couldnt compensate the about -40µV offset which i had.

Hello,

Shielding of the capacitors has only minor effects. (In my newer designs I only have a copper area under the floating capacitor.)

40uV offset is neither the LTC1043 nor the buffer op-amp.
I guess you are having either a common conductor (star grounding) problem or a EMI-problem.

In my early designs I had problems by the used ADC (LTC2400).
The switching noise from the input of the ADC has been rectified on the output stage (ESD protection diodes?) of the buffer amplifier. 
This gave around 20-30 uV negative "offset".
With a adapted R/C low pass between Buffer and ADC you should have less than 10 uV "offset".   (typically -4 .. +4 uV)
(be carefully with the R/C to keep gain and linearity errors low see Data Sheet).

Of course the whole is also dependant on good decoupling of the LTC2400.

And the LTC1043 also needs at least 1 nF at the input.

with best regards

Andreas


 

Offline Echo88Topic starter

  • Frequent Contributor
  • **
  • Posts: 821
  • Country: de
Re: Discrete Capacitive Divider-Circuits
« Reply #9 on: December 11, 2018, 02:34:41 pm »
Thanks for the detailed description. I miss such details in the LTC1043-Datasheet.
Since i use the LTC2057 as the output buffer, it may be the culprit.
Makes sense that the input current peaks of the buffer influence the rather small 1µF-caps without the RC-filter.
I will change the LTC1043/LTC2057-circuit, measure it again with my 5440B/3458A and report back.  :-+
 

Online Andreas

  • Super Contributor
  • ***
  • Posts: 3222
  • Country: de
Re: Discrete Capacitive Divider-Circuits
« Reply #10 on: December 11, 2018, 03:59:41 pm »
Hello,

The LTC2057 is better than the LTC1050 together with the LT1043 + LTC2400.
With the LT1043 / LTC1050 you have temperature dependant clock interferences which increase output noise.
https://www.eevblog.com/forum/metrology/ltc2508-32-bit-adc/msg1728077/#msg1728077

The LTC2057 shows much lower interference in combination.

With best regards

Andreas
 

Online dietert1

  • Super Contributor
  • ***
  • Posts: 2021
  • Country: br
    • CADT Homepage
Re: Discrete Capacitive Divider-Circuits
« Reply #11 on: September 04, 2019, 07:14:27 pm »
I'd like to give a little push to this thread. Have been thinking about the same question: how a discrete circuit may outperform the LTC 1043.

One limitation of the LTC1043 i found is the high resistance of its switches. They made the switches small in order to work with small capacitors. But for a low noise voltage divider/multiplier i want a small resistance, i mean much smaller than the usual resistive divider output impedance (1 kOhm typical). So if i want to reach for example 50 Ohm, i can't work with the LTC1043. I will have to use discrete mosfets with about 5 to 10 Ohm RDSon, together with larger  capacitors, like 22 or 47 uF.

Another disadvantage of the LTC1043 is that it is very general and based on bidirectional switches. This is not necessary in a voltage divider/multiplier. All switches can be unidirectional (simple mosfets). I guess a unidirectional switch helps to lower charge injection. One might even use photo mosfet switches to eliminate charge injection effects.

I think the "voltage loss" in those circuits mainly originates from the switch parasitic (output) capacitances, that generate small currents across the capacitors.

What other differences could one think of? Have there been any experimental results?

Regards, Dieter
 

Offline razvan784

  • Regular Contributor
  • *
  • Posts: 90
  • Country: ro
Re: Discrete Capacitive Divider-Circuits
« Reply #12 on: September 04, 2019, 07:47:25 pm »
But for a low noise voltage divider/multiplier i want a small resistance, i mean much smaller than the usual resistive divider output impedance (1 kOhm typical). So if i want to reach for example 50 Ohm, i can't work with the LTC1043.
Can you please give a more detailed explanation? Isn't noise kT/C, independent of resistance? Won't there be a buffer anyway after the switched capacitor circuit?
 

Offline Kleinstein

  • Super Contributor
  • ***
  • Posts: 14076
  • Country: de
Re: Discrete Capacitive Divider-Circuits
« Reply #13 on: September 04, 2019, 08:09:50 pm »
Having CMOS switches and the NMOS and PMOS combined in part compensates charge injection. So this is not that bad. There are only few higher voltage simple switch chips available with only NMOS switches. The DG611-DG613 would be such a choice - the charge injection performance is not that exceptionally good.

It also depends a lot on the voltage needed. The LTC1043 is limited to some 18 V.  Also for much lower voltages (e.g. 5 V) other low voltage switches may be better than higher voltage ones.

The error in a capacitive divider is from the switch internal charge injection and external parasitic capacitance at different points. This can also give a divider ration that is slightly off the nominal value, but still quite stable.

The switch resistance also contributes to noise, though usually at a low level. The resistance also kind of limits the maximum useful frequency, as settling would take longer.
 

Online dietert1

  • Super Contributor
  • ***
  • Posts: 2021
  • Country: br
    • CADT Homepage
Re: Discrete Capacitive Divider-Circuits
« Reply #14 on: September 04, 2019, 08:23:50 pm »
I made a circuit with some photo mosfet switches we had. They are bidirectional switches with 2x RDSon = 6 Ohm. They are slow, so the chopper runs at 50 Hz. As a first test i made a voltage doubler 10 V ->  20 V using two 100 uF film capacitors. The circuit works and looses only 60 uV = 3 ppm on the doubled voltage. Stability is better than i can measure in this simple setup (definitely better than 1 ppm).
Output impedance is about 400 Ohm, as measured by observing the 40 uV voltage drop with a 100 MOhm load on the extra 10V. This impedance is not much lower than the output impedance of a resistor network i could use to do something similar. As far as i understand the impedance will multiply by the buffer amplifier input current noise. I am interested in 0.1 to 10 Hz noise or even below.
With a LTC1043 the output impedance would be somewhat higher just considering the channel resistance.

Another consideration i remember when using discretes: One may drive gate voltages negative to reduce residual current through off switches.

Regards, Dieter
« Last Edit: September 04, 2019, 08:26:26 pm by dietert1 »
 

Offline Kleinstein

  • Super Contributor
  • ***
  • Posts: 14076
  • Country: de
Re: Discrete Capacitive Divider-Circuits
« Reply #15 on: September 04, 2019, 08:53:19 pm »
Using discrete switches (e.g. single MOSFETs) would only make sense for low on resistance. It adds quite some change injection and parasitic capacitance. So it would need large capacitors. There is also a very limited choice of higher resistance MOFETs.

The switch resistance adds to the output impedance, but often she relatively high output impedance of the circuit is due to the low switching frequency and size of the caps.
Some 400 Ohms of output impedance would still not be that high impedance.
 

Online dietert1

  • Super Contributor
  • ***
  • Posts: 2021
  • Country: br
    • CADT Homepage
Re: Discrete Capacitive Divider-Circuits
« Reply #16 on: September 05, 2019, 05:40:12 am »
In a certain circuit layout charge injection is about proportional to the mosfet area, while channel resistance RDSon is inversely proportional to the mosfet size. Now this is not a CPU controlled switch but a chopper circuit, so charge injection means current and the voltage error created by the charge injection current is proportional to the product of capacitance and resistance. So in the first place the error is independent of the mosfet size.

Until now my conclusion would be: It is fairly easy to reach the LTC1043 performance level with a discrete circuit. In order to reduce the output impedance i need to increase chopper frequency. Will use another method of gate drive isolation, probably capacitive.

Regards, Dieter
 


Share me

Digg  Facebook  SlashDot  Delicious  Technorati  Twitter  Google  Yahoo
Smf