Gilbert Cell (or Multiplier) for noise removal

[ricko_uk]

Hi,
I have an Instrumentation Amplifier measuring a Wheatsone bridge imbalance. The instrumentation amplifier output is a 250KHz sine wave signal and the amplitude varies from 0 to 0.2V reflecting any bridge imbalance during measurement (I can scale it if required). The 250KHz frequency is fixed for the duration of the system usage but can be set anywhere between 20KHz to 250KHz at startup.
The reason why the Wheatstone bridge is AC-excited is because often (not always) two elements of the bridge are capacitors and/or inductors.

Because of long cables and noisy environment I want to remove as much noise as possible while obviously maintaining the low frequency amplitude variation caused by any bridge imbalances (i.e. bridge measurements).

I am thinking of using a Gilbert cells (NE602 / NE612) as a multiplier by feeding into it both the clean original 20-250KHz signal (any amplitude as I can scale it up/down as required) as well as the signal from the intrumentation amplifier.

Questions:
1) Would the output either remove noise and/or provide a much greater signal to noise ratio? If yes, how would I best configure it?
2) Are there other better ways to do that at low cost (around 1-2 USD)? Maybe op-amp multiplers or other demodulators or discretes?

Thank you

[langwadt]

a couple of analog switches to swap the input to the instrumentation amp using the reference signal ?

[unitedatoms]

Sorry if it is off-topic. If I may ask, why cost is so restrictive ($2!) ?

I ask because I can only think of cost being a matter, when something is mass produced. What can be possibly so mass produced, which involves remote sensing with bridge, instrumentation amplifier, AM modulator with preselectable frequency, lengthy cables and noisy environment and so on. Which already sounds like costly lab setup for some material science made in quantity of 3 for whole global market.

I also imagine that problem is possibly real, technical and challenging and obviously costly. Two industry magnates meet in a bar and after few drinks go into arguments about the problem. In heat of debate one of them exclaims, "I will freaking solve it with $1 solution, just watch". Next day the problem show up in forums

[Yansi]

Use a proper lock-in amplifier.

Not a shite RF mixer.

[Kleinstein]

The multiplier / mixer is converting from an AC signal to a DC signal. So at least a large part of the amplification should be before the mixer, as DC amplification is more difficult.  Especially the gilbert cell mixers are not very good with DC performance at the output. CMOS switches like HC4053 are lower drift, but should still have amplification first. If the drive signal is sine, there ideally would be a bandpass filter before the mixer, to limit something like 3 times to modulation frequency. If thr drive signal is square and the amplifier and bridge has little phase shifts it may work better without the filter.


The mixer itself does not improve the SNR, but a low pass filter after the mixer can. The low pass filter is what effectively reduces the bandwidth on the AC side and this way allows for a good SNR.

Depending on how the signal is later used, one could also do some bandpass filtering and than sample the AC signal with sufficient speed (e.g. 1 MSPS ADC in ARM µC) and than do demodulation and averaging in software. This can also give low bandwidth and similar SNR and may be even simpler. However the tolerance to high noise levels is not as good, as the ADC resolution tends to be limited.

[unitedatoms]

Use a proper lock-in amplifier.

Not a shite RF mixer.

Perhaps ADA2200 made by ADI is best fit. Also not very expensive

[Yansi]

Or, if one wants to go cheap within a custom application specific hardware, the easiest way is to use a beefy enough MCU a sample both the REF (or generate REF) and process within digital domain.

There are even high range PGA equipped audio ADC for couple bucks, if high input dynamic range is concerned.

[David Hess]

Use a proper lock-in amplifier.

Not a shite RF mixer.

Perhaps ADA2200 made by ADI is best fit. Also not very expensive

Or keep the instrumentation amplifier and synchronously demodulate the output with a couple of analog switches.  I do not know why you would even use AC with a Wheatstone bridge except as part of a lock-in amplifier.

Personally I think better results are available at much lower frequencies where reactance and other AC effects do not cause as much phase error.

Incidentally, Wheatstone transducers generally have more inherent error than simple DC measurement or chopped DC measurement to remove thermocouple effects so AC excitation and synchronous demodulation should not be required.

[ricko_uk]

Thank you all I will address each one below

First of all one additional information, the reason why the Wheatstone bridge is AC-excited is because often (not always) two elements of the bridge are capacitors and/or inductors.

Kleinstein, David Hess and langwadt, could you please explain (or let me know where I can find infos about) using analog switches/multiplexers to remove noise? How? Do such solution have a special name I can google resources for?

Yansi, could you suggest a low-cost lock-in amplifier solution/schematic/IC? As mentioned I looked at lock-in amplifiers extensively but they seem very expensive and complex (we have 20 to 40 channels in every instrument and would need one for each making it very expensive).


Thank you

[Yansi]

Then just get all the signals digitized and processed within a DSP. That is probably the cheapest way of obtaining a 40 channel lock-in amplifier core for cheap.

[ogden]

Then just get all the signals digitized and processed within a DSP. That is probably the cheapest way of obtaining a 40 channel lock-in amplifier core for cheap.

Be careful with such statement in case of 40 (!) channels. 250KHz signal @ 40 channels may lead to significant ADC and CPU resource requirement, thus cost. IMHO analog mux as mixer + LP filter +  SD ADC is solution here. Relevant article:

https://www.analog.com/en/analog-dialogue/articles/synchronous-detectors-facilitate-precision.html

[Yansi]

So you you think you will make it cheaper with dedicated analog ICs like the ADA2000? No way in hell I'd say.
Or make it cheaper with almost/discrete electronics? That would be one hell of a large design, to have a 40 channel lockin that way.

I am confident that there is no way to make a 40 channel lockin cheaper than using a DSP.

Also, things may get significantly simplified, when all channels do not need to be sampled at the same time and interleaving may be used then.  You know, 250kHz (not sure if that much needed either) is jsut the sensor excitation and the output integration time constant may be much larger, so there is no point in getting for example more than a couple samples per second, maybe.

Of course lock-in may not be needed at all and a simple synchronous detector with stupid polarity switching may be what is needed.

This is the typical X and Y problem thread. Exact requirements must be stated clearly, what the goal is. So far we are only throwing ideas at the OP.

[ogden]

So you you think you will make it cheaper with dedicated analog ICs like the ADA2000? No way in hell I'd say.

No. I will make it using analog switches and low cost opamps, slow ADC's
[edit] By referring to article that talks about synchronous detection and uses ADA2200 I did not suggest ADA2200 but synchronous detection  sorry that it was not obvious (for you)

[unitedatoms]

The application is a mystery. One can only guess with few clues available:


1. The $1-$2 budget per channel

2. A lot of channels

3. A lot of noise


So the thing is for one-time use, disposable ? Could it be


1. Underground geophysical microphone array to detect nuclear events ?
2. Disposable remoting costume for workers in pleasure industry ?
3. School lab demo equipment for a class overstaffed with low income students ?

My bet is 3.

[ogden]

The application is a mystery. One can only guess with few clues available:

Challenge of most efficient solution for given task stays no matter - it is state-owned nuclear lab or understaffed class.

[unitedatoms]

Well, challenge accepted. The discretes and low cost opamps win. They can cost below 1 cent each in expected quantities. So there are some solutions, mostly around synchronous detection.

However the board space will become a dominating portion of cost. Consider that 100 cm^2 of board is way over $2 already even if the parts are zero cents each. I myself can relate to the sinking feel of creeping cost when solution turns closer to realization. Usually all my cost studies of my own imaginary designs end in disappointment with cost of things like screws, enclosures and wires even.

[erikka]

Use a sa612 per bridge to downmix to the lowest possible frequency Use a cheap micro with adc to sample the downmix and filter in dsp. One bin fft on 10 periods of the downmix should work
How fast does the measurement have to be?

[Yansi]

So you you think you will make it cheaper with dedicated analog ICs like the ADA2000? No way in hell I'd say.

No. I will make it using analog switches and low cost opamps, slow ADC's
[edit] By referring to article that talks about synchronous detection and uses ADA2200 I did not suggest ADA2200 but synchronous detection  sorry that it was not obvious (for you)

Where did I say ADA2200 was your idea?

I like the idea of a lock-in amplifier since I have firs encountered them at my first sort-of job as a high school student. Hence talking about lockins and not synchronous detectors. Designing a lock-in from scratch using just cheap analog switches, opamps and stuff certainly will not be fun. But multiplying a couple of samples and filtering the result within a DSP is quite fun.

//One bin of fft ... look for Goertzel algorithm.

[David Hess]

Also, things may get significantly simplified, when all channels do not need to be sampled at the same time and interleaving may be used then.  You know, 250kHz (not sure if that much needed either) is just the sensor excitation and the output integration time constant may be much larger, so there is no point in getting for example more than a couple samples per second, maybe.

If the excitation frequency is higher, then flicker noise below the Nyquist frequency can be rejected.  But this is only an advantage where the sample rate is low enough to take advantage of it so maybe up to 10 Hz.  Notice that noise specifications for chopper stabilized amplifiers are specified a full decade below conventional amplifiers.  At higher frequencies, they are actually noisier so provide no advantage.

[Kleinstein]

The probably simplest version of synchronous detector is with CMOS switches (e.g. HC4053) similar to a Tayloe  mixer: So switch (SPDT) the amplified signal to either of 2 storage capacitors and have a low frequency differential amplifier between the 2. So it would be amplifier +1/3 HC4053 + 4 caps + 5 resistors + 1 OP (low cost) per channel.

A gilbert cell mixer could also be an option as the mixer input could also provide gain - possibly enough. Just find a cheap source for MC1496, NE612 or the like. Not mixing down to zero would eliminate the DC weakness and allow to still get the phase with only 1 mixer per channel.
However it still needs sampling the AC signal to a µC or the like, with maybe 8 channels to a µC.

I would still not design it as 40 channels all in one, more like 3 - 8 and than use several identical units. This would also apply to a all digital solution. 

If possible I would chose a slightly lower frequency, more like the audio range.  Parasitic effects usually get smaller and amplifiers are closer to the simple ideal picture.

Anyway for the beginning get one low frequency channel running first and than think about more, especially in the analog version.

[ricko_uk]

Thank you again for all the answers!

I was not clear in my cost statement. The 2 USD rough target figure is not per channel but for the IC (or noise removal part of the circuit)! If it does the job nicely then we can push it to maybe 3-3.5 USD but that is pushing it. 

Regarding the synchronous detector with a analog switch, I don't understand the principle. Is it as simple as the schematic attached? What would the output be exactly? Sorry but zero experience of RF design.

The application is a "universal" black box for various inductive, resistive and capacitive sensors placed in one or two arms of a wheatsone bridge. Up to 40 sensors arranges in linear or "square/rectangular" array are used. The application is different from client to client. They are often used in analysing thickness of coating or of extruded metals/plastics. Every client use their sensor and in their own way. That is why we need to provide a "black box" which then interfaces with out 3D mapping software displaying the acquired.

The DSP route is not an option for a number of reasons including the fact that it is already quite powerful but overloaded with lots of other real time tasks. Not to mention that neither I nor my colleague have a clue how to implement AM demodulation in software.

Thank you again

[David Hess]

First of all one additional information, the reason why the Wheatstone bridge is AC-excited is because often (not always) two elements of the bridge are capacitors and/or inductors.

I forgot about that.  I tend to think in terms of bridge transducers when noise is a problem.

Kleinstein, David Hess and langwadt, could you please explain (or let me know where I can find infos about) using analog switches/multiplexers to remove noise? How? Do such solution have a special name I can google resources for?

Yansi, could you suggest a low-cost lock-in amplifier solution/schematic/IC? As mentioned I looked at lock-in amplifiers extensively but they seem very expensive and complex (we have 20 to 40 channels in every instrument and would need one for each making it very expensive).

I can do both.  Check out figure 4 on page 4 of Linear Technology application note 3.  (1) There one part of an LTC1043 switched capacitor building block is used as a precision SPDT analog switch in the synchronous demodulator for a lock-in amplifier but any SPDT analog switch would work.  In the past this was commonly done with a pair of JFETs.

Below is an example where JFETs are used.

(1) The same application is shown in figure 27 on page 24 of Linear Technology application note 43.

[langwadt]

Regarding the synchronous detector with a analog switch, I don't understand the principle. Is it as simple as the schematic attached? What would the output be exactly? Sorry but zero experience of RF design.

the switch toggles the opamp between inverting and non-inverting, i.o.w. multiply with a +/- square-wave

[Kleinstein]

The version toggling the OP between +1 and -1 amplification needs a relatively fast OP.

The Tayloe type detector does not need a fast OP and thus may be simpler for a relatively high speed.
The 2 switches in the circuit could be 1/3 of an 4053 or similar.

I have not run the simulation - somewhat trouble with the switch model.

[ricko_uk]

Thank you Kleinstein, langwadt and David!! Much appreciated feedback and infos!!