Author Topic: Solving the classic dynamic range problem  (Read 7898 times)

0 Members and 1 Guest are viewing this topic.

Offline jeremyTopic starter

  • Super Contributor
  • ***
  • Posts: 1079
  • Country: au
Solving the classic dynamic range problem
« on: July 24, 2014, 12:45:31 pm »
Hi all,

I have a signal which is 100uV or so. It's on top of a constant baseline on the order of mV to single digit Volts, however the baseline does change between experiments; so experiment 1 might have 500mV, experiment 2 might have 765mV depending on the DUT. Obviously, if I don't remove the offset extremely accurately then the gain stages for the 100uV signal will saturate. I'd like to be able to capture the small signal at as low a frequency as possible (less than 1Hz) with my +-10V ADC, so I'd rather not just high pass the signal. And this is a single ended signal, not differential.

Has anyone solved this sort of problem before? I'm open to ideas.

 

Offline fcb

  • Super Contributor
  • ***
  • Posts: 2117
  • Country: gb
  • Test instrument designer/G1YWC
    • Electron Plus
Re: Solving the classic dynamic range problem
« Reply #1 on: July 24, 2014, 12:56:24 pm »
You don't mention the required resolution of the 100uV signal and the desired settling time - that will determine the feasibility of various schemes (such as high resl. ADC).

One possible method is to use a low-pass filter (say 0.05Hz) to create a voltage equivalent to your baseline voltage, and then subtract it from the signal.

You could even do this in software much more quickly:
1. sample waveform to get background voltage.
2. set a DAC to this back-ground voltage.
3. use a differential amplifier on the DAC voltage & signal, sample with your ADC
4. use software to remove any underlying (but small) DC component.

The advantage of a software driven servo loop like this would be very fast settling (much faster than an RC filter).

https://electron.plus Power Analysers, VI Signature Testers, Voltage References, Picoammeters, Curve Tracers.
 

Offline jeremyTopic starter

  • Super Contributor
  • ***
  • Posts: 1079
  • Country: au
Re: Solving the classic dynamic range problem
« Reply #2 on: July 24, 2014, 03:53:24 pm »
Thanks for your response fcb. The system you described is actually what I was thinking of doing, but I didn't want to bias any answers ;) I did not think about doing it in software though.

The ADC is already chosen, it's an 18bit ADC over +-10V made by NI. Ideally, I would like to have +10V corresponding to 50uV and -10V corresponding to -50uV. But this is not the part I am worried about so much.

Settling time is a much more difficult one, I'm actually not sure. I would think it needs to be about 100us or less but I don't know if that is feasable.
 

Offline Richard Crowley

  • Super Contributor
  • ***
  • Posts: 4317
  • Country: us
  • KJ7YLK
Re: Solving the classic dynamic range problem
« Reply #3 on: July 24, 2014, 04:03:18 pm »
You did not mention the frequency range of the signal. That would be a factor in considering solutions.
You also mentioned the baseline/bias voltage changing between experiments. That that imply that the bias/baseline is constant during an experiment?  How long is each experiment?
 

Offline mathias

  • Regular Contributor
  • *
  • Posts: 59
  • Country: 00
Re: Solving the classic dynamic range problem
« Reply #4 on: July 24, 2014, 04:03:35 pm »
Depending on the situation, you can modulate the signal you want to measure, but keep the baseline in DC. For example measuring (very weak) laser light by turning on and off the laser makes it possible to separate this component from daylight normally flooding the sensor/detector/amplifier. If you do it in the audio range, 24-bit ADCs are cheaply available.
 

Offline ejeffrey

  • Super Contributor
  • ***
  • Posts: 3685
  • Country: us
Re: Solving the classic dynamic range problem
« Reply #5 on: July 24, 2014, 05:24:10 pm »
Can you just manually zero it at the beginning of an experiment?

Use an instrumentation amplifier for the first stage with a low-leakage capacitor on the (-) input.  Use a low-leakage analog switch or even a mechanical push-button to connect a buffered copy of the input to the capacitor.  Push the button to capture the baseline and your amplifier goes to zero (within the amplifier offsets).  You can polish off any residual offset in software if necessary.
 

Offline jeremyTopic starter

  • Super Contributor
  • ***
  • Posts: 1079
  • Country: au
Re: Solving the classic dynamic range problem
« Reply #6 on: July 24, 2014, 07:24:49 pm »
Hi all,

Sorry, I should have been more specific but I was typing on my phone.

System is to characterise laser dynamics for diode lasers, the large offset is due to the drive current so it should be constant for a given diode. Unfortunately I can't use modulation or a chopper as it will upset the laser dynamics I am trying to measure. The current system is a 5Hz-3MHz amplifier, but I would like to get well below 1Hz. I'd also like to push the low pass cutoff higher, but I can always do that with a separate amplifier/ADC and recombine the information later.

Thanks all for your suggestions so far. I am not so sure about using a DAC because I then have another dynamic range problem on the DAC. ejeffrey, I think I will try yours first seeing as it is the simplest.
 

Offline David Hess

  • Super Contributor
  • ***
  • Posts: 16545
  • Country: us
  • DavidH
Re: Solving the classic dynamic range problem
« Reply #7 on: July 24, 2014, 09:23:41 pm »
Old oscilloscopes use a special form of differential input amplifier called a differential comparator to do this; the differential amplifier adds or subtracts a precise internally generated offset voltage from the signal.  This can result in effectively having a position control with 10,000 divisions or more of range allowing small signals with large DC offsets to be displayed

Many oscilloscopes have a lesser version of this in the form of an offset control which may or may not be separate from the position control.

As far as a practical implementation in your case, adding or subtracting a 1 volt offset with a resolution of better than 100 microvolts and accuracy over time and temperature to match is certainly achievable as it is only requires control to better than 1 part in 10,000.  In an analog implementation, this would be a good place for one of those 10-turn wirewound potentiometers that cost under $20.  Many 16-bit DACs could do it as well.
 

Offline onlooker

  • Frequent Contributor
  • **
  • Posts: 395
Re: Solving the classic dynamic range problem
« Reply #8 on: July 25, 2014, 04:08:09 am »
You may google for level shifter ic/circuitry, with words like analog or precision or dc.
 

Offline vk6zgo

  • Super Contributor
  • ***
  • Posts: 7563
  • Country: au
Re: Solving the classic dynamic range problem
« Reply #9 on: July 25, 2014, 04:19:58 am »
Old oscilloscopes use a special form of differential input amplifier called a differential comparator to do this; the differential amplifier adds or subtracts a precise internally generated offset voltage from the signal.  This can result in effectively having a position control with 10,000 divisions or more of range allowing small signals with large DC offsets to be displayed

Many oscilloscopes have a lesser version of this in the form of an offset control which may or may not be separate from the position control.

As far as a practical implementation in your case, adding or subtracting a 1 volt offset with a resolution of better than 100 microvolts and accuracy over time and temperature to match is certainly achievable as it is only requires control to better than 1 part in 10,000.  In an analog implementation, this would be a good place for one of those 10-turn wirewound potentiometers that cost under $20.  Many 16-bit DACs could do it as well.

David,this is exactly what I was thinking!

I was not sure whether the OP meant a DC voltage or not when he referred to an "constant baseline",so I did not comment earlier.
What you describe was definitely an everyday method used with older 'scopes---something we would do without even thinking!
 

Offline David Hess

  • Super Contributor
  • ***
  • Posts: 16545
  • Country: us
  • DavidH
Re: Solving the classic dynamic range problem
« Reply #10 on: July 25, 2014, 03:32:46 pm »
I was not sure whether the OP meant a DC voltage or not when he referred to an "constant baseline",so I did not comment earlier.
What you describe was definitely an everyday method used with older 'scopes---something we would do without even thinking!

We would not normally think of microvolt signal levels though on an oscilloscope unless we had something like a Tektronix 7A22 differential amplifier.  The stand alone version of that vertical amplifier in the form of a AM502 will do almost exactly what is needed.

An easy way to do this with a single ended signal path and no differential amplifier is to simply insert a precision floating variable reference voltage in series with either the input signal or ground.  This will preserve the high input impedance of the first amplifier stage if necessary and does not require any other circuit additions.  Inserting it into the ground is advantageous as far as noise, bandwidth, and input capacitance but it could be done on the signal side as well.

I am a dubious about the original poster's noise and drift requirements because amplifying a 100 microvolt signal to 10 volts for an 18 bit ADC yields a least significant bit of 0.381 nanovolts which is close to state of the art.  The best integrated operational amplifiers have drift and noise which is a two or three orders of magnitude higher than that.
 

Offline Conrad Hoffman

  • Super Contributor
  • ***
  • Posts: 1928
  • Country: us
    • The Messy Basement
Re: Solving the classic dynamic range problem
« Reply #11 on: July 25, 2014, 05:07:50 pm »
You can servo it with a very low filter frequency just like they do with hifi amps. You use an integrator with any time constant you want to drive the signal back to zero DC.

My 1A7A Tek plug-in (ancient) has the manual offset control with coarse and fine adjustments. Works well, but a bit of a PITA compared to a filter solution.
 

Offline Richard Crowley

  • Super Contributor
  • ***
  • Posts: 4317
  • Country: us
  • KJ7YLK
Re: Solving the classic dynamic range problem
« Reply #12 on: July 25, 2014, 07:18:12 pm »
There isn't nearly enough detail about the requirements here to make any useful suggestions.  And Jeremy doesn't appear to understand what "chopping" means (i.e. that it does NOT affect the actual signal being measured, etc.)  And offers other seemingly arbitrary limitations without explaining WHY he thinks they are limits. 
 

Offline ejeffrey

  • Super Contributor
  • ***
  • Posts: 3685
  • Country: us
Re: Solving the classic dynamic range problem
« Reply #13 on: July 25, 2014, 07:35:21 pm »
The only thing a chopper will do is remove the room lights.  It won't do anything to address the variable baseline output of a laser.
 

Offline David Hess

  • Super Contributor
  • ***
  • Posts: 16545
  • Country: us
  • DavidH
Re: Solving the classic dynamic range problem
« Reply #14 on: July 25, 2014, 09:02:05 pm »
The only thing a chopper will do is remove the room lights.  It won't do anything to address the variable baseline output of a laser.

Is this signal from the laser or a receiver?  I thought he was measuring the voltage across a semiconductor laser for purposes of something like improving the coherence.

Chopper techniques are great when you can use them and given his requirements, they may be needed anyway for signal conditioning.
 

Offline David Hess

  • Super Contributor
  • ***
  • Posts: 16545
  • Country: us
  • DavidH
Re: Solving the classic dynamic range problem
« Reply #15 on: July 25, 2014, 09:07:01 pm »
You can servo it with a very low filter frequency just like they do with hifi amps. You use an integrator with any time constant you want to drive the signal back to zero DC.

My 1A7A Tek plug-in (ancient) has the manual offset control with coarse and fine adjustments. Works well, but a bit of a PITA compared to a filter solution.

I thought he wanted to get away from having a low frequency cutoff.  I have done the same thing in correcting baseline DC drift automatically using an integrator or even sampled system.

Tektronix made some amazing differential comparators for the 500 series of oscilloscopes.  I make do with a 7A13 which I really like.  I have a 7A22 but have only tested it to make sure it works.  It will probably get some use soon working on a microphone circuit.
 

Offline Richard Crowley

  • Super Contributor
  • ***
  • Posts: 4317
  • Country: us
  • KJ7YLK
Re: Solving the classic dynamic range problem
« Reply #16 on: July 25, 2014, 09:07:56 pm »
Is this signal from the laser or a receiver? 
Or is this simply the supply current into the laser?
That's the problem.  None of us really know what Jeremy is asking about because critical details are absent.
We all have different assumptions about what he is describing. Perhaps some of us are even correct. Who knows?   :-//
 

Offline David Hess

  • Super Contributor
  • ***
  • Posts: 16545
  • Country: us
  • DavidH
Re: Solving the classic dynamic range problem
« Reply #17 on: July 25, 2014, 09:39:50 pm »
Is this signal from the laser or a receiver? 
Or is this simply the supply current into the laser?
That's the problem.  None of us really know what Jeremy is asking about because critical details are absent.
We all have different assumptions about what he is describing. Perhaps some of us are even correct. Who knows?   :-//

I do not think it is quite that bad but the lack of information constrains suggestions for alternatives.

He has an 18 bit National Instruments ADC with an external amplifier to make a full scale range of hundreds of microvolts that he wants to use with inputs with up to a couple volts of offset.  That is certainly solvable but as I pointed out above, I think 18 bits of resolution is unrealistic unless the signal conditioning is state of the art or better.
 

Offline jeremyTopic starter

  • Super Contributor
  • ***
  • Posts: 1079
  • Country: au
Re: Solving the classic dynamic range problem
« Reply #18 on: July 26, 2014, 08:00:40 am »
Hi all,

Thanks for the constructive suggestions.

There isn't nearly enough detail about the requirements here to make any useful suggestions.  And Jeremy doesn't appear to understand what "chopping" means (i.e. that it does NOT affect the actual signal being measured, etc.)  And offers other seemingly arbitrary limitations without explaining WHY he thinks they are limits.

Richard, what information would you like? I am not sure what you mean by arbitrary limits. I have calculated the voltage magnitude of the information and the frequency range, now I want to try to measure it in real life?

I am very much aware of what a chopper does, and it will not help as outlined in ejeffrey's post. However, we are talking about optical chopping, do you mean chopping as in a chopper amplifier? Optical chopping definitely affects the signal, that is the point of using it?

The reason I said "no chopping" is because that is usually the first thing that people suggest in optical instrumentation problems (myself included).

You can servo it with a very low filter frequency just like they do with hifi amps. You use an integrator with any time constant you want to drive the signal back to zero DC.

My 1A7A Tek plug-in (ancient) has the manual offset control with coarse and fine adjustments. Works well, but a bit of a PITA compared to a filter solution.

I thought he wanted to get away from having a low frequency cutoff.  I have done the same thing in correcting baseline DC drift automatically using an integrator or even sampled system.


This is exactly what I am trying to do.

I am interested in looking into this servoed system, but also running the output through a sample and hold rather than continuously subtracting the filter output. Thanks for the idea.

I am a dubious about the original poster's noise and drift requirements because amplifying a 100 microvolt signal to 10 volts for an 18 bit ADC yields a least significant bit of 0.381 nanovolts which is close to state of the art.  The best integrated operational amplifiers have drift and noise which is a two or three orders of magnitude higher than that.

Yes, you are right to see them as dubious. I am only using the 18bit ADC because that is what I have available. I also have other techniques to reduce the noise (averaging, non-linear filtering). I do not actually expect to get less than nano volt resolution though. Even having one shot resolution of 1uV would be awesome as far as I am concerned!

--

I feel that this problem doesn't really need much of a context? But I am willing to learn if someone can point out the problems. I basically just have a very small signal with a big signal added. The big signal is dc, the little signal has some time varying information. I want the offset to go away without just highpassing but it is hard because I need the offset subtraction to be really precise and changeable between experiments. Even though I am measuring lasers, this would be the same if I was, for instance, measuring the magnetic field inside a coil, measuring conductivity of a solution undergoing a multistage reaction, measuring neuron impulses, etc. I completely understand that one can mathematically subtract an offset, but I was looking for perhaps someone with experience with doing this in the real world to give me some pointers on how to actually pull this off as I do not have much experience with very low drift/offset amplifiers.

And just to be super clear, there are no extra detectors or anything. I am just measuring the terminal voltage of the semiconductor laser.

Thanks :)
« Last Edit: July 26, 2014, 08:27:36 am by jeremy »
 

Offline David Hess

  • Super Contributor
  • ***
  • Posts: 16545
  • Country: us
  • DavidH
Re: Solving the classic dynamic range problem
« Reply #19 on: July 26, 2014, 01:15:37 pm »
Quote
I am a dubious about the original poster's noise and drift requirements because amplifying a 100 microvolt signal to 10 volts for an 18 bit ADC yields a least significant bit of 0.381 nanovolts which is close to state of the art.  The best integrated operational amplifiers have drift and noise which is a two or three orders of magnitude higher than that.

Yes, you are right to see them as dubious. I am only using the 18bit ADC because that is what I have available. I also have other techniques to reduce the noise (averaging, non-linear filtering). I do not actually expect to get less than nano volt resolution though. Even having one shot resolution of 1uV would be awesome as far as I am concerned!

Good integrated chopper stabilized amplifiers have a maximum input offset voltage drift on the order of 15 nV/C and maximum peak to peak low frequency noise of 200 nV.  That makes thermocouple effects a major source of error so achieving that level of performance will require careful attention to layout and thermal effects.  This is one of those places where a couple of well placed Dixie cups acting as air baffles may improve performance by an order of magnitude.

Similar attention will need to be paid to the offset voltage.  It needs to be low drift and have low low frequency noise.

If you are only interested in the DC level, another idea I would at least consider is using a switched capacitor front end in the form of an LTC1043 to subtract the offset voltage directly from the signal without relying on resistors.

Is the signal at a constant DC level or is it being switched?  If the later, the amplifier is going to be overloaded between measurements and recovery time will be important which will be a problem if chopper stabilized amplifiers are used.
 

Offline Richard Crowley

  • Super Contributor
  • ***
  • Posts: 4317
  • Country: us
  • KJ7YLK
Re: Solving the classic dynamic range problem
« Reply #20 on: July 26, 2014, 02:33:28 pm »
Richard, what information would you like?
For example: What frequency response do you need (i.e. what kind of AC-coupling and sample rate could be used). How long is each experiment (i.e. could you use simple AC-coupling for short-duration experiments. By sampling the baseline value at the beginning of the experiment, and then simply measure the delta, you effectively create a "floating-point" scheme with the baseline as the exponent, and the measurement over time as the mantissa, etc.
 

Offline David Hess

  • Super Contributor
  • ***
  • Posts: 16545
  • Country: us
  • DavidH
Re: Solving the classic dynamic range problem
« Reply #21 on: July 26, 2014, 04:25:53 pm »
It is probably worth pointing out that a 6.5 digit voltmeter can make DC measurements to 2 volts with 1 microvolt resolution directly.
 


Share me

Digg  Facebook  SlashDot  Delicious  Technorati  Twitter  Google  Yahoo
Smf