Instrumentation Amplifier Modification or Replacement

[loop123]

I'd like to change the AMP01 instrumentation amplifier to the top one with best spec ever. First some technical discussions. This is its datasheet and its spectral density graph.

https://www.analog.com/media/en/technical-documentation/data-sheets/amp01.pdf



I know that to convert to Voltage in RMS. Multiply nV/Sqrt(Hz) to the square root of the bandwidth.

Let's use 100Hz bandwidth as example (actual example).

from the graph with spectral density 5 nV/Sqrt (Hz).   V(rms) = 5nV/Sqrt (Hz)  multiply by Sqrt (100) = 5nV * 10 = 0.000000005 x 10 = 0.000000050 V or 50nV.

So the noise is like V(rms) = 50nV.

In my application. I need to run it at 10uV (microVolt) input using 1000Hz-3000Hz  frequency with good noise profile.

What instrumentation amplifier to replace the AMP01 that is the top or the best out there that can use 10uV input with 1000Hz-3000Hz frequency with not visible or minimal noise? 

If possible. The frequency can be 10,000 Hz instead of 1000Hz. Just give the instrumentation amplifier with the highest bandwidth that can do it.

Thank you.

[moffy]

In such an application, shielding, both thermal and electrical is very important. I found a ground plane top and bottom as well as a shielded box for both electrical and thermal effects is essential. You also need to determine if your PSU is inducing the noise, use a battery supply as a reference to check the difference.

[loop123]

Ok.

Say. If two instrumentation amplifiers have same nV/sqrt (Hz) lets say 5 nV/Sqrt (Hz). It means they have similar noise at say 1000Hz or 2000Hz bandwidth?

Know the top Instru Amps (INA) model out there that accept differential signal with 1uV to 10uV 1000Hz-2000Hz with very clean output with noises in the nV instead of uV?  What is the lowest spectral density figure available? like is there a 1nV/sqrt (Hz) or even 0.5nV/sqrt (Hz) Instru amps?

[moffy]

Ok.

Say. If two instrumentation amplifiers have same nV/sqrt (Hz) lets say 5 nV/Sqrt (Hz). It means they have similar noise at say 1000Hz or 2000Hz bandwidth?

Know the top Instru Amps (INA) model out there that accept differential signal with 1uV to 10uV 1000Hz-2000Hz with very clean output with noises in the nV instead of uV?  What is the lowest spectral density figure available? like is there a 1nV/sqrt (Hz) or even 0.5nV/sqrt (Hz) Instru amps?

Depends on the input impedance of the circuit. The THAT1512 can have 1nV/hz^0.5 but that is with zero input impedance and 60dB of gain, but current noise can easily become the dominant factor at around 1k ohm perhaps lower, in which case a JFET input amp is a better choice. What you are seeing though is induced pickup and you won't even get close to the performance of a good instrument amp if you don't pay serious detailed attention to what was mentioned previously, both electrical and thermal shielding as well as power supply noise. You should also aim for a true differential input to try and maximise CMRR.

[loop123]

No problem with power supplies as I used batteries only. The circuit has metal cover on all corners and I may use Faraday cage (know where to get these to enclose projects?) About thermal shielding.  Did you mean putting it in very cold aircon? How to do thermal shielding?

It is true differential input with about 10kOhm source impedance. So what state of the art INA (Instrumentation Amp) that can accept high input impedance, etc. It should be something that is significantly better than the AMP01. Actually it will be used as EEG with microVolt input but with 1000Hz frequency so I can lessen the gain when time to switch to milliVolt EMG and still have the 1000Hz-2000Hz capability. I already have the finished product but it uses AMP01 so for my project. I'll replace the amps to the best one in the world. I have 2 units. One is for experiment. Don't worry about Galvanic isolation because it is all purely powered by 1.2V rechargeable Eneloop batteries.

[CaptDon]

You are getting so bogged down in numbers and math and 'best in the world' opamps. I worked in the hospital biomed O.R. field for around 5 years and simply stated "You don't need the best in the world opamps" to accomplish your tasks. EEG machines with perfectly usable results have been around for decades. EEG brainwave monitors and ECG heart monitors have relatively standard run of the mill parts. The key ingredient is shielding not only of the equipment and its connections but also of the other equipment within the room including the wiring in the walls. The thermal noise of the resistors in your bio-amplifier circuit may well make the need for a 'super noise free opamp' pointless. Take a case in point, the 20 meter ham band at 14MHz. A receive preamplifier can help, however, the atmospheric noise floor is so high that an 'ultra low noise preamplifier' is not needed and even a noisy poorly designed tube preamplifier gives as good of results as can possibly be obtained. It seems you are chasing a problem in mathematical details and not a problem that exists in the real world which every manufacturer of biomedical equipment has already solved. A better antenna beats a preamp every time!! Maybe there is a better way to capture your signals like a pre-amplified probe?

[Andy Chee]

On a slightly different tangent, I understand that you will be using your amplifier for a brain computer interface.  I hope you understand that EEG signals are inherently noisy (especially in unknown/uncontrolled physical environments), and much of the clean up occurs in signal processing software rather than hardware.

Of course it helps to have clean hardware in the first place, but there will be a point of diminishing returns.

[Kleinstein]

For a 10 K range source impedance the amp01 looks not that bad. The noise resistance (= voltage noise divided by current noise density) is at some 35 Kohms and thus not that far off the source impedance.
In some cases the current noise can be actually a bit larger, as not all datasheets handle the correlation between the input right.
The very low voltage noise types (e.g. AD8428) are not suitable.

Just have a look at the AD webside to get an idea, what is available.

With some 5 nV/Sqrt(Hz) to beat, there could also be JFET based amplifiers as an option. If one really wants to and the main interest with higher frequencies, one could consider a discrete JFET version. This way could offer less noise, but is quite some effort.

[loop123]

Ignore the initial message because I interchanged the reference and ground. You see. In EMG, the reference is the real ground. While in EEG the so called "reference" is connected to V- of the differential amplifier. So in the Netech EEG simulator. I connected the amplifier ground to the REF socket when it should be the V- of the amplifier that should be connected to the Netech REF.

With this correction. I got the following waveform with 10uV, 60Hz, 100Hz bandwidth selectded at main amp with 50000 gain



The following with 1000Hz bandwidth selected:



So my simple goal is to replace the amplifier to get rid of the noises at 1000Hz.

Usually in EEG, the bandwidth used is only 40Hz. In EMG, it's 1000Hz. In my Brain Computer Interface project. I just want not to keep adjusting the bandwidth switch when adjusting and to zoom the milliVolt to microVolt also at will so I need to upgrade the amplifier.

So please give those amps suggestions that is much better than the AMP01 in the unit. Thanks.

[loop123]

For a 10 K range source impedance the amp01 looks not that bad. The noise resistance (= voltage noise divided by current noise density) is at some 35 Kohms and thus not that far off the source impedance.
In some cases the current noise can be actually a bit larger, as not all datasheets handle the correlation between the input right.
The very low voltage noise types (e.g. AD8428) are not suitable.

Where did you get 35Kohm exactly?  Also in the AMP01 datasheet below. Why is the noise greater at Gain=1 compared to say Gain=100? Is it not as the gain goes higher, there should be more noise?

And why should the source impedance be close to the noise resistance?

Just have a look at the AD webside to get an idea, what is available.

With some 5 nV/Sqrt(Hz) to beat, there could also be JFET based amplifiers as an option. If one really wants to and the main interest with higher frequencies, one could consider a discrete JFET version. This way could offer less noise, but is quite some effort.

What you mean quite effort? is there no single discrete JFET that is inside single chip so you dont have to buld the transistors from scratch. So without going into discrete JFET. There is no way to  get lower than noise of 5nV(sqrt(Hz))

[moffy]

The AMP01 really is quite good. It has 5nV/sqrt(Hz) voltage noise and 0.15pA/sqrt(Hz) current noise which means that if the input impedance is 33.33k then the current noise and voltage noise are equivalent. Excellent CMRR and true differential inputs. The quoted noise of the opamp is referred to the input, so you multiply the input noise by the gain to get the output noise. The input referred noise drops at higher gains because the low noise input stage dominates (where most of the gain occurs) and the noiser later stages become less significant.
Nice results.

[Kleinstein]

BJT based amplifiers are somewhat limited in the noise figure they can reach. When they are made for lower voltage noise the current noise goes up. The noise figure is best when the source impedance matched the ration of voltage noise to current noise (provided the 2 are not correlated).

JFET amplifiers have much lower current noise and can reach better noise figures. However it takes some effort to get low voltage noise and the low noise figures are for higher impedance.  There are integrated FET based instrumentation amplifiers, but they may not be better than 5 nV/sqrt(Hz). The best bet here seems to be the LTC6373 with some 8 nV/sqrt(Hz).
Using discrete FETs could get lower noise, though with extra effort. There may be ready made hybirds or modules available, though usually not cheap.


p.s.
The AD8421 is another good BJT based INA, more suitable for low source impedance. Depending on the details (the specs for the current noise are sometimes a bit fishy) it could be slightly lower noise than the amp01.

[loop123]

In skin electrode application, you abrade the skin to 5kOhm.

Why is it best to match source impedance with voltage/current noise? It is not about the concept of voltage divider where it is better to match input and source resistance so either will take half the voltage, is it?

Lets take the actual example of the skin needing to be 5kOhm. But the voltage and current noise resistance of the AMP01 is really 33kOhm as one of you computed. So why is the source impedance better at 33kOhm instead with 5kOhm (which should really be better so the voltage would be taken up more by the input (related to the concept where input impedance) but must be much higher. Pls give actual computations of 5kOhm vs 33kOhm.

I need to understand this bec my next project is building an active amplifier so you dont have to abrade the skin and it can have good noise even at higher resistance.

Also its needed to find upgrade to the AMP01 to match 5kOhm source impedance where you cant see any noise at 10 microVolt and 1000Hz.

Without abraiding the skin, there would be more noise in principle, is the noise that significant? Lets see in your computations.

Thanks a lot.

[moffy]

Why is it best to match source impedance with voltage/current noise? It is not about the concept of voltage divider where it is better to match input and source resistance so either will take half the voltage, is it?

Lets take the actual example of the skin needing to be 5kOhm. But the voltage and current noise resistance of the AMP01 is really 33kOhm as one of you computed. So why is the source impedance better at 33kOhm instead with 5kOhm (which should really be better so the voltage would be taken up more by the input (related to the concept where input impedance) but must be much higher. Pls give actual computations of 5kOhm vs 33kOhm.

33k is just the impedance point where current noise equals the voltage noise, it isn't an optimal point. Below the 33k voltage noise dominates, above 33k current noise will dominate. Better than the AMP01 for your application would take some work because you need very high input CMRR of a true differential input and that can be challenging using discretes like the JFE2140: https://www.ti.com/product/JFE2140
which have excellent voltage and current noise but it is getting the high CMRR of the AMP01 as well for what at the end might only be a modest improvement in noise.

[Kleinstein]

The matching to the voltage noise to current noise ratio is about the point where the amplifier has the best noise figure. It is not about adding resistance to the input to match, but about where the amplifier would be at it's best. One would match something like a transformer if one uses it, as there no exra noise is added. The amp01 may still be OK with a 5 K source impedance.  Because of the 2 inputs to current noise part would ideally need more than just 1 number for a full charactirization and both sides would contribute. There is a tendency for datasheets to be a bit optimistic with the current noise.
It is just that if one is well off, chances are to find a better other amplifier. If below 70% the noise resistance could use 2 amplifiers in parallel could reduce the noise.

The AD8421 would be a solution that has it's best noise figure at some 16 K, kind of close to 2 x amp01 in parallel. The noise figure is similar, just made for a difference source impedance.

With 33 K or source impedance the amp01 would give 5 nV from the voltage noise and a similar noise from the current noise. So some 7 nV/sqrt(Hz) total. With 5 K source impedance the current noise part would be smaller and the total noise more like 5.5 nV/sqrt(Hz). In addition there is also just the noise from the source resistance (e.g. some 10 nV/sqrt(Hz) for 6 K).  So the overall noise can still to a large part from from the source impedance and not from the amplifier. So the possible advantage from a better amplifier is limited.

[Zero999]

All your threads appear to be about the same project. You'll get much better quality replies if you keep it all in one thread, rather than starting so many. I've completely lost track of it.

It appears as though you lack adequate filtering for mains frequencies. The input should have a passive, low pass filter, to get rid of RF and mains harmonics, followed by active filtering to get rid of the interference from the mains itself.

[loop123]

The matching to the voltage noise to current noise ratio is about the point where the amplifier has the best noise figure. It is not about adding resistance to the input to match, but about where the amplifier would be at it's best. One would match something like a transformer if one uses it, as there no exra noise is added. The amp01 may still be OK with a 5 K source impedance.  Because of the 2 inputs to current noise part would ideally need more than just 1 number for a full charactirization and both sides would contribute. There is a tendency for datasheets to be a bit optimistic with the current noise.
It is just that if one is well off, chances are to find a better other amplifier. If below 70% the noise resistance could use 2 amplifiers in parallel could reduce the noise.

The AD8421 would be a solution that has it's best noise figure at some 16 K, kind of close to 2 x amp01 in parallel. The noise figure is similar, just made for a difference source impedance.

With 33 K or source impedance the amp01 would give 5 nV from the voltage noise and a similar noise from the current noise. So some 7 nV/sqrt(Hz) total. With 5 K source impedance the current noise part would be smaller and the total noise more like 5.5 nV/sqrt(Hz). In addition there is also just the noise from the source resistance (e.g. some 10 nV/sqrt(Hz) for 6 K).  So the overall noise can still to a large part from from the source impedance and not from the amplifier. So the possible advantage from a better amplifier is limited.

I spent many hours reading about current noise vs voltage noise. The dogma seems to be "thermal noise increases with a larger resistance value, whereas current noise decreases when resistance increases."

A web site mentioned this is just a myth. Maybe there are circuits where the above applies and there are other circuits where the following applies? How to know which circuit applies to the dogma and myth belows? Thanks.

https://www.analog.com/en/resources/technical-articles/11-myths-about-analog-noise-analysis.html

"
8. The Amplifier with the Lowest Voltage Noise Is the Best Choice

When choosing an op amp, the voltage noise is often the only noise specification considered by the designer. It is important not to overlook the current noise as well. Except in special cases such as input bias current compensation, the current noise is typically the shot noise of the input bias current: in = √2 × q × IB. The current noise is converted to a voltage via the source resistance, so when there is a large resistance in front of the amplifier input, the current noise can be a larger noise contributor than the voltage noise. The typical case where current noise is a problem is when a low noise op amp is used with a large resistance in series with the input. For example, consider the ADA4898-1 low noise op amp with a 10 kΩ resistor in series with the input. The voltage noise of the ADA4898-1 is 0.9 nV/√Hz, the 10 kΩ resistor has 12.8 nV/√Hz, and the 2.4 pA/√Hz current noise times the 10 kΩ resistor is 24 nV/√Hz, the largest noise source in the system. In cases like this, where the current noise dominates, it is often possible to find a part with lower current noise and thereby reduce the noise of the system. This is especially true for precision amplifiers, but there are high speed FET input op amps that can help in high speed circuits as well. For example, instead of choosing the ADA4898-1 and not getting the benefit of the 0.9 nV/√Hz voltage noise, a JFET input amplifier such as the AD8033 or the ADA4817-1 could have been chosen.".


What kind of circuit where as the resistance increases, the current noise decreases?

What kind of circuit where as the resistance increases, the current noise also increases?

[Terry Bites]

If you can live with switching noise, a "zero drift" amp offers the lowest 1/f noise. For very low level measurments you can't polish off all the common mode with the INA. You boost the apparent cmrr by floating the input. That keeps cm noise from the measurement side out of your front end. Iso amps are not noted for their DC precsion so ideally you digitize before the isolation. I made just such an amp with an icoupler, an AD7400 and a home brew zero drift INA. These days I'd just chuck in a 32bit ADC, eg ADS126x

[loop123]

If replacing the AMP01 won't make significant improvement in the remaining noises. Will using active electrodes where the amplifier is in the electrode itself offer much more improvement?

I don't want to abrade skin before use. They say when you use active electrodes, you don't have to abrade the skin to lower resistance to 5kOhm from 30kOhm?

Btw.. I use only batteries in my units so no need for any passive, low pass filter, to get rid of RF and mains harmonics, followed by active filtering to get rid of the interference from the mains itself. The only interference is Radio frequency from air, but can this get into the unit when there are already bandpass filters in the unit such as when selecting the 1000Hz filter switch should filter the higher frequency already. For power line capacitive coupling interference. I always unplug all cables near the area and I don't see any interference from capacitive coupling at all.

I got the units at bargain because the company which uses them think it's defective. I spent the last threads fixing it (which is not connected to the topic here). And now it's fully fixed. So, I'm now focusing on the use and replacing the AMP01 if possible, because of noises at 10microVolt input at 1000Hz frequency.

[moffy]

I don't know how well the skin is modeled by a resistor but at 25C a resistor of 5k has a spectral noise of 9nV/sqrt(Hz) and 30k has a spectral noise density of 22nV/sqrt(Hz) both well above the voltage noise of 5nV/sqrt(Hz) of the AMP01. Resistors generate thermal noise, hence the temperature dependence, even when nothing is attached. But it depends on what the noise model of skin looks like, there probably are capacitances at least that limit bandwidth.

[loop123]

Most explanations about needing low impedance in the skin is so that there will not be voltage divider where the electrode would soak up some voltage making the signal going to amplifier lower. So the negative effect is not more about getting lesser voltage but about the noise?



I used an El-check electrode checker to measure my skin impedance. The led that lights up is about 10k-20k Ohm.

I read about Active Electrodes in the following. How would getting active electrodes eliminate the 10k-20k Ohm impedance since the Active electrodes use electrodes too?  And the AMP01 has input impedance of greater than 20,000 MegaOhm as it the datasheet mentioned. So even if the skin impedance is 50kOhm. Voltage divider would still make the AMP01 take up 99% of the voltage. Isn't it?

https://zeto-inc.com/blog/active-and-passive-electrodes-what-are-they-pros-cons/



If the advantage is just the cable not picking  up intereference. Then just use notch filter or bandpass filter to handle it. Isn't it?

So what is the real advantage of active electrodes?

[Andy Chee]

If the advantage is just the cable not picking  up intereference. Then just use notch filter or bandpass filter to handle it. Isn't it?

No.  It completely depends on what causes the interference.

For example, if it is mains noise, then yes, a notch filter may work.  But if it is noise from a computer power supply, then probably not.  And what if it's noise coming unidentified the location? (e.g. motor interference from nearby tram/train power lines)

In summary, if you cannot identify where the interference is coming from, then active electrodes are a good solution.

[loop123]

If the advantage is just the cable not picking  up intereference. Then just use notch filter or bandpass filter to handle it. Isn't it?

No.  It completely depends on what causes the interference.

For example, if it is mains noise, then yes, a notch filter may work.  But if it is noise from a computer power supply, then probably not.  And what if it's noise coming unidentified the location? (e.g. motor interference from nearby tram/train power lines)

In summary, if you cannot identify where the interference is coming from, then active electrodes are a good solution.

The whole unit is powered by batteries. How can computer power supply cause interference? unless by means of EMI? Do you have illustrations of waveforms of common causes of the noises? They are easy to recognize. Even just running a sine wave generator will make them show  up, Isn't it?

How much would it cost to get a Faraday cage the size of person like a telephone booth?

[Kleinstein]

The loss in signal amplitude should not be an issue: the amplifier input impedance should be well larger than 10 K and also the capacitive loading from a shielded cable is not that much, unless one is interested in the really high frequencies.

EMI can be a point - it is a bit tricky to design a high impedance filter to keep EMI away from the amplfier.  Once the very high frequency EMI part hits an amplifier it can get demodulated and effect the frequency band of interrest. After that filtering can no longer separate it from the signal.

It may be relevant how good the amplifier can tolerate RF band interference (e.g. WLAN, mobile phones, TV signals, radio). Not all amplifiers react the same way in this respect.
If really deparate one could have an RF receiver to detect the interferening signal and than subtract it. Still tricky when the cables move.

I see an only moderate advantage for active electrodes - maybe a reason to charge more money for them  .

Adding shielding at least adds inconvenience - like move to a special room instead of moving the intrument to the patients room. A shilded room would also need filtering for the supplies and light. RF tight doors are also a thing that is not easy.

[Andy Chee]

unless by means of EMI? Do you have illustrations of waveforms of common causes of the noises? They are easy to recognize. Even just running a sine wave generator will make them show  up, Isn't it?

No.  Sometimes EMI has a known pattern.  But EMI can equally be random.

If you don't want to use a faraday cage, then you should analyse the surrounding EMI in your intended operating environment, so that either;

a) design the hardware to cope with the environment, or
b) program the EEG/EMG software with your EMI analysis, and the software will do the noise reduction for you.

https://eeglab.org/tutorials/ConceptsGuide/Setting_up_your_EEG_lab.html

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6312138/pdf/june-17-10.pdf