Electronically divide voltage signal (sim to differential amplifier)

  1. I know that a differential amplifier is used to effectively subtract one voltage signal from another (and amplify).

    However, are there any devices or techniques that can be used to electronically divide two voltage signals?

    Thank you
     
  2. jcsd
  3. jim hardy

    jim hardy 4,668
    Science Advisor
    Gold Member
    2014 Award

    try a search first on "analog multiplier divider"
    then with modifiers
    'transconductance"
    and
    "time division"
     
  4. berkeman

    Staff: Mentor

    Can you say more about what you are trying to do? Is it anything like a voltage divider (2 resistors)?

    Switched capacitor circuits may be what you are looking for -- have you considered them?
     
  5. I thought a voltage divider simply divides a voltage signal by a constant value.

    I have x2 voltage signals, slowly varying between 0V and ~5V. I record these for about 1s at a sampling frequency of ~10^5 Hz. I then divide the digitized signals in software.

    What I would prefer to do is divide the signals electronically, and only record one signal - the ratioed one.
     
  6. berkeman

    Staff: Mentor

    And what happens when the denominator signal goes to zero? Or even gets small...
     
  7. berkeman

    Staff: Mentor

    Beyond the divide by zero and other problems, the typical way you make an analog divider is to take the log of each signal and subtract them. Look up log amplifier and analog divider circuits to get some ideas...
     
  8. jim hardy

    jim hardy 4,668
    Science Advisor
    Gold Member
    2014 Award

    that's what two resistors connected in series will do.

    I thought you were looking for an active circuit with two inputs and one output

    that does this function:

    [itex]output[/itex] = [itex]\frac{input1}{input2}[/itex]

    that would be a more complicated active divider circuit that divides one variable voltage by another. They're used a lot .

    http://www.datasheetcatalog.com/datasheets_pdf/M/P/Y/1/MPY100.shtml

    now observe in above transfer function
    [itex]output[/itex] = [itex]\frac{input1}{input2}[/itex]



    if you connect input2 directly to output

    [itex]output[/itex] = [itex]\frac{input1}{output}[/itex]

    or
    ([itex]output[/itex])^2 = [itex] {input1} [/itex]

    which means
    [itex]output[/itex] = √([itex] {input1} [/itex])

    which is handy for example maybe linearizing a fluid flow signal where flow of a liquid is measured by the pressure drop along a pipe or across an elbow ..
     
  9. dlgoff

    dlgoff 3,075
    Science Advisor
    Gold Member
    2014 Award

    This thread made me think of the old Analog computers. :tongue2:


     
  10. sophiecentaur

    sophiecentaur 13,571
    Science Advisor
    Gold Member

    This would seem to be the right approach. 'Hybrid' systems are usually a headache. Getting into digits as soon as possible is the answer, as long as your analogue signals are not too wide bandwidth. Digital processors are as cheap as chips and software is very easy to modify. Why would you want to use the creaky analogue route unless you real have to?
     
  11. jim hardy

    jim hardy 4,668
    Science Advisor
    Gold Member
    2014 Award

    Then there's cmos rate multipliers
    where each voltage is turned into a frequency

    and logic gates arranged to perform math functions..

    it's to me a hybrid - half analog half binary arithmetic.
    Schmidt Herman, a GE engineer, built his "Electronic digital slide rule" in the late sixties . He used nixie tubes and packaged it in a fishing tackle box. That was the beginning of the "pocket calculator".

    http://www.google.com/patents/US3676656

    http://www.mwftr.com/cneF12/Assign2_Reading.pdf
     
  12. sophiecentaur

    sophiecentaur 13,571
    Science Advisor
    Gold Member

    There's a Rave from the Grave!

    But, if he's going to "record" a signal (and he implied digital recording) then why not get into digits asap? It's just 'philosophy' (true but how relevant?) to consider all circuits as analogue, even when their functions and states are totally digital. As someone who managed to construct a 'Shannon Rack" DAC, in the days when they were not just off the shelf, I know all the problems associated with analogue summing of exponentially decaying signals, in the hope that the sum would follow a monotonic set of samples. That thing really WAS Hybrid. Within Months of my getting it to work, Texas and their ilk had produced integrated DACs that worked fine.
     
  13. Baluncore

    Baluncore 2,609
    Science Advisor

    You can do that by applying one signal to an A to D converter input and the other to the converter reference voltage input. Some converters are optimised for that conversion mode.

    The reference input will probably need to remain greater than the signal input. You may need to scale the signal to meet that requirement.

    The ratio technique is used for example, to digitise differential strain gauge bridge outputs in digital scales, where the voltage supply to the bridge is used as the reference.

    http://lpvo.fe.uni-lj.si/fileadmin/...110_Understanding_Ratiometric_Conversions.pdf
     
  14. sophiecentaur

    sophiecentaur 13,571
    Science Advisor
    Gold Member

    But why? (Except to prove that it can be done.) Engineering is about doing things the cheapest way, consistent with accuracy, reliability and performance. There was a time when component count was everything. Nowadays, the optimum solution to most problems is digital - with good reason. Using a variable reference for an ADC means that the multiplication process is asymmetrical. Phase and frequency response of the reference input is unlikely to be the same as that of the input and what about linearity? What is the specification for performance with an out of range reference input voltage?. I was just wondering about the lineup procedure for such a set up.
     
    Last edited by a moderator: Apr 30, 2014
  15. jim hardy

    jim hardy 4,668
    Science Advisor
    Gold Member
    2014 Award

    i think we answered his original question:

    His only problem statement was the bold part of this:
    i completely agree, it'd be best done by software
    unless he's running something like Interpreted Basic and is against the processor time wall.
    That is quite a few floating point divides.

    Maybe we need a better idea of what is the problem he's facing ?
     
  16. berkeman

    Staff: Mentor

    Thread closed temporarily for Moderation...

    EDIT -- Thread re-opened.
     
    Last edited: Apr 30, 2014
  17. berkeman

    Staff: Mentor

    Yeah, that's a useful trick to have in your EE bag of tricks. We used that technique to speed up the fine position calculation for a graphics tablet that I helped to design at HP many years ago... :smile:
     
  18. Baluncore

    Baluncore 2,609
    Science Advisor

    Ratiometric conversion is not a trick. It is a professional design technique that has progressively risen in importance over the years. It is now more important than ever. As an example, accurate high resolution digital scales are not practical without ratiometric conversion.


    Because ratiometric conversion is now so widely applied, support is provided by many microcontroller families, for example dsPIC. That makes it possible to program a single chip solution to the analogue, digital conversion, signal processing and I/O.


    Because ratiometric conversion uses a sampling converter there is no requirement to synchronously sample and hold two channels prior to performing two conversions. It therefore doubles the maximum data conversion rate while halving the power and eliminating sample jitter and phase problems.
    With ratiometric conversion, not only is cost lowest, but accuracy, reliability and performance are all significantly improved.


    There is always a fundamental problem with division by zero. The OP specified voltages between 0 and 5 volts, not –5V through zero to +5V, so the OP clearly does not require bipolar ratio computation. There will be no reference polarity reversal relative to zero volts, so a four quadrant multiplying converter is not needed. Where a ratiometric conversion is needed it is almost always the case that the reference does not change polarity and is greater than the signal.


    Frequency response does not need to be the same in this case as the OP specified slowly varying signals. The reference voltage input bandwidth on ratiometric converters now often matches the signal input bandwidth because they use identical input circuitry to cancel drift.


    What about it? Converters these days have linearity better than ½ LSB. That was not the case 50 or even 25 years ago.


    Reference input voltages are now often differential, by using mosfet rail to rail analogue design, both –Vin and +Vin will continue to function 0.3V beyond the supply rails. The reference and signal inputs are protected from more extreme voltages.


    I see no reason why it should not be self calibrating and need no adjustment. Any transducer errors are better corrected in software than with analogue trimming or adjustment.


    When a typical ADC generates two successive digital values, those values might be at the low end of the scale, say 11 lsb and 10 lsb, when the CPU divides them it gets 1.1 but with a +/–20% accuracy. That is not the case with the ratiometric converter where the output resolution and accuracy is independent of input signal voltage. A 10 bit ratiometric converter should preserve +/–0.1% across the input range.

    Ratiometric conversion also reduces the cost and component count by eliminating the need for a stable voltage reference. References can get very expensive for converters of 20 bits and above.

    So to sum it all up, a ratiometric converter, (probably implemented in a microcontroller), will double the data rate and offer a single chip solution without any external analogue components.

    It is an ideal digital solution.
     
  19. berkeman

    Staff: Mentor

    I didn't read through your links -- is the bandwidth of the Vref input specified in the datasheets?
     
  20. I design ADCs for a living. Most of the parts I design have the Vref inputs (they are almost always differential) heavily bypassed internally. Vref is designed to be as close to DC as possible to reduce noise (and improve vref stability). So, unless the signal connected to Vref were step-wise DC, it wouldn't work very well on a lot of modern ADC designed for high-speed signal processing. Other parts have a fast reference buffer on-chip that would have a much higher bandwidth. This is typically not specified on the datasheet.

    There are certainly ADCs that are designed for this kind of application, but buyer beware.

    I'd just sample the signals and do the subtraction in the digital domain.
     
  21. sophiecentaur

    sophiecentaur 13,571
    Science Advisor
    Gold Member

    It would be interesting to see some details (block diagram) of how both the cases of |VA|>|VB| and |VA|<|VB| are dealt with, when one of them is the Vref of an ADC. Is there no lower limit to the possible range of Vref for the device to work correctly? Is it possible to deal with that problem by offsetting reference and input volts and yet still get the 'right answer'? I can't see the arithmetic working properly.
    The link, provided by Baluncore seems to be for a device that will give the ratio of two Resistors and not voltages, in the way that is required by the OP. Have I got this wrong and how can the device be used for Voltage division?? [Edit: I didn't put that in the right way. I realise the device measures volts but still the ref voltage needs to be kept 'more' than the input]
    I feel inclined to go along with analogdesign's ("professional") opinion. He would appear to know a thing or two about the business - from first hand. There is something about Symmetry in design that is appealing and reassuring, which is why the use of two ADCs would seem to be the best way to go. Are they really that expensive (compared with the MSC1211, quoted)?
     
    Last edited: May 1, 2014
Know someone interested in this topic? Share a link to this question via email, Google+, Twitter, or Facebook

Have something to add?