# Designing an amplification circuit (amplifying small sensor signals)

#### KeepItSimpleStupid

Joined Mar 4, 2014
5,088

I was just saying is that if you need 0.5mV out of 20 mV, then you only need to count to 20/0.5 or 40.
And at 40 counts, the +- error is a lot less than at 1 count.

The PGA can more effectively resolve 0.5 mV when say 1 mV is 400 counts. (made up numbers)

I guess, in the process world say 0-1000 deg. +-0.5 degrees is the accuracy of say the thermocouple, but maybe we only care how close it is to 300 deg,
12 bits might be able to resolve an instrument that can measure -200 to 2000 degrees.

A blood pressure cuff doesn;t need a gazillion digits. Neither does it make any sense generally to know if its 25 or 25.00003 degrees. Is it 13 kV or 13.1 kV? We don't care if it's 13,0001.001 V

Then there was the vacuum gauge with decade outputs 0-1, 1-2, 2 to 3 up to 10 V out.
was like 1e-3, 1e-4, 1e-5, 1e-6 torr.

Last edited:

#### Henry603

Joined Nov 19, 2018
69
@danadak : ok thank you, I started reading all that info you gave me.
Also was doing further research and some questions came up:

1. Most instrumentation amplifiers (=INA) also if they are rail to rail, they will not swing completly to top/bottom rails as previously mentioned.
But for my sensor (as it outputs a signal from 0 to 20mV) I also need to detect the zero output from the sensor. So the idea is to set an offset voltage for the INA (using the reference pin). Could I do it like this:
If the output voltage swing of the INA is e.g. V_GND + 0.1V (for bottom rail), I will set the reference pin of the INA to 0.1V. This way now the INA will output 0.1V in case the sensor outputs '0' and also the smallest changes in the sensor can be detected. Is that common practice?

2. Then the next issues comes up: How do I get this kind of low reference voltage? What is the common practice for this kind of application? I can not drive the ref pin with a resistive divider (I had to add a voltage follower as well). And there are no voltage reference ICs that provide such a low voltage. What solutions are there for this issue? Or how would you amplify my sensor signal (0-20mV) with an INA (from a single supply)? I would really appreciate some help here. Im running in circles for too long now...

3. I need absolute accuracy and you mentioned some of the main error sources. BUt after some research I think i can calibrate most of them out? E.g. offset input voltage, gain error. I can short the inputs of the INA and check for the initial offset voltage. After I can set a known voltage on the inputs and with a known gain I will check for deviation. Then I can also compensate for it in software as well (as i know deviation and factor/gain). So I should mainly check for an INA with lo voltage/temperature drift and good linearity?

thank you for your help so far.

#### KeepItSimpleStupid

Joined Mar 4, 2014
5,088
Here is another idea. Use the LM7705. http://www.ti.com/product/LM7705

Then set Vref of the IA to zero.

I can short the inputs of the INA
That sort of stuff can get you in trouble. A tiny CURRENT across a nearly zero resistance means a large voltage. I learned he hard way.

Many times sensors have application notes with suggested interfacing.

Joined Mar 10, 2018
4,057
1) There are parts that actually go below the negative rail. Go here and look
at the selector table and you will see a number of offerings -

https://www.analog.com/en/parametri...4359|4051|4050|2687|4123|s3|s5|5042|2839|2840

Look at the headroom specs for the negative input, several can handle < 100 mV below ground.

2) You will not need a low reference V for issue in 1)

3) Yes to both questions

Regards, Dana.,

#### KeepItSimpleStupid

Joined Mar 4, 2014
5,088
@Henry603

I think you've made a lot of progress in your understanding. When I worked with an IA for the first time, it was full of surprises. What Rail to rail, over the top and what non-ideal means are now important concepts.
it's also important to consider what happens when the power supply is zero and you have an external voltage source. You can't forget bypassing. If you ever decide to use a poke in place breadboard, each connection adds a few pf of capacitance. When transferring what you think is the "same" circuit, it isn't because all of the parasitics are different.

Bypassing is an "art" in itself.

Some concepts at high frequency can bite you: three 1K resistors in series is not 3K. In some circuits three 1K resistors will work, a 3K resistor will not.

Surface mount components have lower parasitics when mounted on their edge.

I know one area I'd like to really understand is very low current differential amplifier front-ends.

#### Henry603

Joined Nov 19, 2018
69
@danadak: thank you for the answer and the listing, found some interesting components using it!

@KeepItSimpleStupid: thank you!
Yes, to understand INAs is much more difficult for a beginner than i initially thought!
And Im also reading a lot of stuff regarding bypassing currently, thank you.

The suggestion with the LM7705 is really helpful! I was searching for smth. like this but did not find anything suitable so I gave up thinking that this idea is not used in practice. But seems that there are ICs for just this purpose. This idea was kind of a little breakthrough for me so thank you very much for that!

I did a lot of searching/comparing and found a new device, the AD8237 Instrumentation amplifier.
I plan to set a gain of 102 to boost my 0-20mV sensor signal to max. 2,048V output signal (this is the reference voltage my ADC will use).

CMRR seems to be good, max. gain error outstanding (when comparing to most other INAs I saw so far), gain drift, Gain Nonlinearity and Maximum offset voltage drift also good.
Input offset voltage is not so good but I could calibrate that out using my µC I suppose.
Regarding the noise Im not sure, especially the Voltage noise (spectral density) seems to be pretty bad compared to other INAs.
Is that smth. I have to really worry about in my DC application (I think this spec dominates at lower frequencies but can you tell me if that is really smth. that would hurt my accuracy very much in my case?)

Using your idea with the LM7705, output swing would also fit my application.
Regarding Differential Input Operating Voltage and Input Operating Voltage (+IN, −IN, or REF), that should also be ok.
Could you be so kind and check the first 3-4 pages of the datasheet above in order to confirm my statements? Because if im mistaken on some of my assumptions I would be very happy If I could correct them before I have to find out the hard way .

Differential Input Impedance is specified with 100MOhm, that should also be ok to connect my sensor directly, as in the sensor datasheet they say the load should have an Input Impedance of >1MOhm.

In addition I used the diamond plot tool by Analog Devices to check for valid common mode/output signal range for my application for the AD8237:https://www.analog.com/designtools/...&r=0.02&sl=0.02&tab=1&ty=1&vn=-0.23&vp=5&vr=0

So with VDD 5V and GND -0.23V using the LM7705 this should be fine regarding the tool.

In conclusion:
I would use your suggested LM7705 to supply a negative GND to the AD8237 and power the AD8237 with a positive supply of 5V.
This way the AD8237 can swing down to completely zero Volts but also output the max. voltage for my 2.048V reference of the ADC (so I could use the whole ADC range).
Still have to find a good ADC that I can use but do you see issues so far with this approach (also regarding required accuracy and using the AD8237 and LM7705)?
(As before the amplification of the sensor signal in 0-20mV range I need 5µV step size/resolution, that makes a required 0.5mV step size/resolution after amplification with gain 100)

I feel I learned so much mainly because you guys always pointed out critical aspects and always pushed me into the right direction. I hope my understanding so far is not too far off again and that you still can find some motivation to help me out on this. I do really appreciate your help so much!
Thank you.

Last edited:

#### MisterBill2

Joined Jan 23, 2018
14,721
I did not have much to say in this discussion, but your conclusion about the resources here are certainly correct! There is a whole lot of expertise available, especially if one asks the right questions and adequately explains what they need.

Joined Mar 10, 2018
4,057
If you use a DelSig you might be able to eliminate need for IA.

For example, this is setup to digitize the 0 - 20 mV range directly,
to 20 bits, for a resolution of 122 nV / bit. Its total range for this
setup is +/- .064V.

Will you actually get absolute accuracy of 20 bits over T and V, short
answer is NO. You have to do the end to end error budget of the signal
path, A/D. But part has onboard ref of +/- .1%. If you needed better
Vref then it can work with external Vref.

Just a thought.

Regards, Dana.

Last edited:

#### KeepItSimpleStupid

Joined Mar 4, 2014
5,088
Been busy.

1. I would have not thought to use the bias generator on a single supply OP Amp. Makes sense though. You might want to check with Analog Devices to see if it's really OK.

2. You might have to worry about your power supplies too. The LT3045 might be worth looking into. https://www.analog.com/en/products/lt3045.html

#### Henry603

Joined Nov 19, 2018
69
@KeepItSimpleStupid:
So except the two points you mentioned you do not see any 'obvious' issues with the plan i proposed?

1.
I contacted AD in order to check if the INA can handle the negative GND power supply, thanks for pointing that out.
I did not really think about that might be an issue as from the description of the Bias Generator it seems that this is the intended use of this device (to enable true ground swing for single supply amplifiers).

2.
I was planning to use this LDO for the supply (LT1962):
https://www.analog.com/media/en/technical-documentation/data-sheets/1962fb.pdf
It is also a low noise LDO.
But the one you suggested is ultralow noise (much better regarding the noise specs).
Do you think in my application it is worth it to invest in this kind of ultralow noise LDOs or might it be a bit overkill?
(as more external components are required and layout in general is more complex)

Thanks a bunch!

#### Henry603

Joined Nov 19, 2018
69
#1. Did you get an answer?

Found these recently:

Single-supply IA amplifier
https://www.renesas.com/doc/datasheet/isl71218m.pdfm%2Fdoc%2Fdatasheet%2Fisl71218m.pdf

Precision single supply amplifier
https://www.mouser.com/pdfdocs/isl28108-208-408.pdf

Any more progress?
@KeepItSimpleStupid :

I want to prototype the circuit using the instrumentation amplifier AD8237 and the negative bias chip to feed the small negative voltage to the AD8237's negative supply pin (to get my true ground/zero output swing).
Also I will take the ADC MCP344 and hope that it will work as intended.

An Analog devices engineer said that the idea should work.
But he warned me about potential ground noise that might or might not be present so I should keep enough headroom (Could you maybe elaborate on this one if you can to make sure I understood it correctly?).

How could I minimize the potential ground noise issue?
I will pay attention to a proper layout and bypassing/decoupling of the power supply pins of all components in my circuit.
That should really minimize the ground noise to a minimum I guess?
This and choosing a proper, low noise LDO (very important I guess).

I want a 3.3V supply, so I chose the LT1763 3.3V LDO.
In the datasheet they mention that it is intended for 'Noise-Sensitive Instrumentation Systems'.
So I guess I should be ok with that one? Or do you see issues with that (or can propose better ones?).

Also regarding the ground noise, are there other things I should pay attention to in order to not get any ground noise issues?
Do you see a problem here?

Thank you so much for your help, really appreciate it!
Best.

#### Henry603

Joined Nov 19, 2018
69
@KeepItSimpleStupid : Did some research myself but that is also a very interesting post, thanks for sharing!
BTW, I do not know if you have experience in filtering but if you look at the following datasheet of the instrumentation amplifier, there will be noise at the output, that will decrease my ENOB when attaching it to the ADC directly.
In case you have experience it this field, what kind of filter would you recommend?
At what frequency I should start to attenuate the signal to effectively filter out noise by this particular instrumentation amp?
Dataheet of the instrumentation amp.
@danadak : Maybe that is also a question that you in particular can answer?

Thank you.

#### Henry603

Joined Nov 19, 2018
69
@KeepItSimpleStupid :
The input bias current needs a place to go. (transistors, FETS, OP-amps etc.) it's effectively a leakage current and it has to go somewhere. It will develop an offset voltage across some resistor. Even if it's 1 pA, it has to go somewhere. A 10 M resistor from each input to ground is fine. It can generate (1 pA * 10 M) or not a lot of input voltage and you have the same, but opposite on the other input.
One concern left regarding the bias current return path:
My sensor has a very low impedance (10-100 Ohm) and the INA-Inputs are high impedance (100 MOhm).
Will I get problems here if I use one 10 MOhm resistor to ground on every INA-Input to create the bias current return path (like you suggested)?
Would I need to add series resistors as well for every input to increase the resistance here to avoid problems?
(because the current will return through the path of least impedance so do I need series resistance here in order to prevent that)?

Hope you can clear my doubts here, thank you.

#### MisterBill2

Joined Jan 23, 2018
14,721
@KeepItSimpleStupid :

One concern left regarding the bias current return path:
My sensor has a very low impedance (10-100 Ohm) and the INA-Inputs are high impedance (100 MOhm).
Will I get problems here if I use one 10 MOhm resistor to ground on every INA-Input to create the bias current return path (like you suggested)?
Would I need to add series resistors as well for every input to increase the resistance here to avoid problems?
(because the current will return through the path of least impedance so do I need series resistance here in order to prevent that)?

Hope you can clear my doubts here, thank you.
I have done exactly that, used high value resistors to hold isolated differential inputs well within the allowable voltage range, and it works perfectly.The sensors were thermocouples, about as low-impedance as things can be.. I believe that we used 4.7 Megohm resistors in the one watt size, because they were on hand and available instantly.

#### Henry603

Joined Nov 19, 2018
69
@MisterBill2 :
Thank you very much for your input!

So you did not add any series resistors but only the 4.7 MOhm resistor to ground? (one resistor for every amplifier input?)
What is your thought on using the 10MOhm resistor to ground on only one input?
Does that even make a difference?

#### MisterBill2

Joined Jan 23, 2018
14,721
@MisterBill2 :
Thank you very much for your input!

So you did not add any series resistors but only the 4.7 MOhm resistor to ground? (one resistor for every amplifier input?)
What is your thought on using the 10MOhm resistor to ground on only one input?
Does that even make a difference?
The only current flowing through the resistor to ground is the amplifier bias current, and the difference in voltage at the two inputs was that current 's voltage drop flowing through a very few ohms. Much less than the quantization error due to it having only a 12 bit A/D converter.

#### Henry603

Joined Nov 19, 2018
69
The only current flowing through the resistor to ground is the amplifier bias current, and the difference in voltage at the two inputs was that current 's voltage drop flowing through a very few ohms. Much less than the quantization error due to it having only a 12 bit A/D converter.
So it is better to add the resistor to ground on both inputs to keep the differential voltage unaltered (in relation)?

#### MisterBill2

Joined Jan 23, 2018
14,721
So it is better to add the resistor to ground on both inputs to keep the differential voltage unaltered (in relation)?
It did not make any difference with thermocouples.