Op Amp output noise problem

[mmuj]

Hi All

Design Overview:

I am designing precision, very low input signal (in micro volts), low frequency (0.5Hz to 20Hz) amplifier in In Amp configuration to drive ADC. Source Impedance is unknow and will vary. Input signal is AC coupled with 40uF (4x10uF) capacitors then biasing resistor 100K then 50/60Hz notch filters with dc resistance is 116K6 (31K6x2 + 26K7x2) (notch filter is tested circuit and had been used in many other places). then OPA376 (2 Op Amp for differential signal) feedback resistor is 22k while gain setting resistor is 1K (Gain = 22x2+1 = 45). dc resistance on inverting input is corrected by 220K series resistor to minimize DC offset. then there are (OPA211 driven from opposite signal) stage with a gain of 240 (together). I isolated OPA211 stage for troubleshooting and found that it is input (OPA376) stage causing trouble. Supply voltage =5v, mid rail = 2.5 (using OP AMP design used in many other circuits) overall gain is 45x120x2 = 10800, around 80dB.

Problem:

Output is swinging (some time more than +/-2.5V which is saturation for ADC) with or without input signal applied. I captured the output signal and looked though LabVIEW application (inhouse made) to see if there is any frequency component but it looks like a 1/f noise. I also powered and repeated test with battery to avoid power supply noise but same results.

Observation:

Noise seems like directly proportional to biasing resistor values, higher the value higher the noise (understandable because of thermal noise but its way out of calculation). bypassing the notch filter didn't make any difference. I even tried 1K biasing resistor (i know its ridiculous in terms of corner frequency) noise gone down to 3mVp-p.

I have already spent a lot of time trying to get to this point and now running out of ideas. Although I have order few other Op Amps (OPA333, OP196, OPA132 and LMP2021) to replace and compare the results. but meanwhile any suggestion will be very much appreciated. or if explanation is not clear enough, please ask.

Thanks in advance.

Mujahid

 

[FvM]

In a short, you shouldn't use any RC signal filters without previous signal amplification. You also should try to avoid AC coupling at the lowest signal level, if ever possible. A low noise amplifier should be selected according to the source impedance, performing noise matching. You didn't tell the source impedance, so I won't guess about a better amplifier now. OPA376 isn't bad as such, but you can get amplifiers with a factor ten lower low frequency noise, if the source impedance fits. Most of the observed noise is however apparently brought up by your circuit.

 

[vinoth8051]

Hi,
High resolution ADCs are available these days (more than 20,24,32 bit). Signal is not really fast changing. so i don't think amplifier would be needed for conversion.

 

[mmuj]

Thanks for reply

Intially amp was designed with OPA211 with low value biasing resistance but later some one sitting above me wanted me to add notch filter so it increased source impedance and i have to change to OP376. because OPA211 is bipolar OP amp and would result in high dc offset because of Ios.

I can not change design with out scraping PCBs so i am trying to find some compromise.

Now about your suggestion, I am not using any RC filter, its just decoupling capacitor at the input as there will be dc component in the input signal.

I agree that for low noise, impedance should be low but it is out of my control, signal source impedance is variable (from few ohms to k ohms).

But in this case source impedance at op amp input is set by the biasing resistance and and notch filter resistance and it is ac coupled so i think source impedance become bit of a irrelevant here.


Mujahid

 

[FvM]

several kohms are still low impedance, so a bipolar OP would be in fact preferred for a low noise design. You didn't tell yet if AC coupling is strictly required at the input. Obviously it would be the case, if the input voltage range can't kept otherwise. But a coupling capacitor with a higher bias resistance (e.g. 100 k) isn't a problem as long as the cut-off frequency is considerably below signal frequency range, respectively, the voltage drop at the capcitor generated by the input noise current can be neglected.

With a 50 Hz notch filter, you have the opposite situation, it's effective impedance is mainly resistive in the signal frequency range of interest, so the resistors are fully acting as noise sources. Remember, that a few kohm are able to score off a good OP in terms of noise voltage. But these effects can be easily calculated, so I'm a bit surprized that the results are "a way out of calculation". A parameter, that can be measured without a LF spectral analyzer (or a computer) is the 0.1 to 10 Hz Vpp often specified in OP datasheets. A 0.5 to 20 Hz value would be closely related to it. Or calculate total input referred noise density at 1 and 10 Hz and check for the dominant contributions. A Spice noise analysis is quite helpful to take care of all relevant noise sources.

In some cases, you have additional sources of "noise" respectively low frequent fluctuations. Thermolectric generated voltages can be a serious problem, you would want to shield the sensitive circuit part against moving air at least for test. I also presume, that the signal source is disconnected and replaced by a short or a equivalent resistance, unless you are sure, that it doesn't involve unknown noise.

High resolution ADCs are available these days (more than 20,24,32 bit). Signal is not really fast changing. so i don't think amplifier would be needed for conversion.

Even the best ADC have CMOS input stages, and low frequency noise considerably above a good bipolar or JFET OP.

P.S.: If you look to OPA211 datasheet Figure 39, you see that it's optimal below 1 or 2 k source impedance and OPA227 preferable for upto 10 or 20k.

 

[mmuj]

Noise is enterying from input and get worst by thermoelectric effects. If for example if PCB is covered with bubble wrap (ESD protected), input wires removed and have aircon switched OFF it significantly reduces the noise. I will do some calculation tomorrow and will post here.

Source impedance is even higher than OPA227 recommendation that is why I order OPA132.

One very similar design with less demanding restriction has used OP196 so to get comparison I will try that.

Is there anyway of stoping noise enterying from input ?

 

[FvM]

If the source itself is noisy, you can't but performing optimal filtering, depending on the application requirements. If the input signal noise is high, amplifier noise may be almost neglectable in some cases.

I'm under the impression, that you understand the key parameters for OP selection, so I don't need to assist in this regard. You should however understand the significance of 1/f noise for a low frequency application, and in case of doubt calculate the expectable noise with your actual application parameters. The basic design of a µV amplifier is simple: have a first stage with at least G=10 or 20, with low impedance feedback network, DC coupled if possible. All filtering (except for input RF blocking) should be applied after the first stage.

Regarding thermal effects, consider the simple truth that a µV-amplifier can't be operated with good results in "free air", an enclosure will be needed not only as an electrostatical shield but also to fight thermal fluctuations. Electromagnetic interferences can be a problem as well, but they are mostly mains frequency components.

 

[mmuj]

Hi FvM

I started this design with virtually no analog experiance and after making few (rather quite a few) mistakes i learned a lot but still I have to make this work with minimul changes otherwise I will end up scraping PCBs. Things i learned and your suggestions will definately help me next time.

Noise calculation that i am working around are simple (may be too simple)

OPA376 voltage noise for 0.1Hz to 10Hz = 0.8uV
first stage OpAmp voltage noise * gain = 0.8uV * 45 = 1.125mV (at the output of firststage)
output stage OpAmp voltage noise 1.125*120*2 = 270mV (noise just caused by OPA376)

Current noise is given for 1Khz which is 2fA/rtHz. so I made assumption and multiplied it with 100 for 0.1 to 10Hz. voltage noise caused by current noise = 1pA*560k*45 = 25.2uV

this simple figure is more than what i get in quite envirnoment (noise goes down to 25mV)

One thing that is still not clear to me is that how and why biasing resistor is controlling noise while thermal noise of biasing resistor (FOR 560k noise = 0.3uv) is much less than OPA376 so we should be able to ignore it (root of square sum...)

 

[FvM]

OPA376 voltage noise for 0.1Hz to 10Hz = 0.8uV
first stage OpAmp voltage noise * gain = 0.8uV * 45 = 1.125mV (at the output of firststage)

Something's wrong in the calculation, I think.

To see the specified wide band noise voltage in a measurement, you need a bandpass, of course.

Current noise is given for 1Khz which is 2fA/rtHz. so I made assumption and multiplied it with 100 for 0.1 to 10Hz. voltage noise caused by current noise = 1pA*560k*45 = 25.2uV

The noise bandwidth calculation isn't correct. The noise density would be multiplied by √10, not 10². 560K has to be put in, if it's the effective source impedance. Hopefully, the source impedance is considerably lower, however. As I mentioned before, you get a high source impedance for the OP if a notch filter is connected to it's input.

One thing that is still not clear to me is that how and why biasing resistor is controlling noise while thermal noise of biasing resistor (FOR 560k noise = 0.3uv) is much less than OPA376 so we should be able to ignore it (root of square sum...)

Depends on the circuit. If the bias resistor is connected in parallel to the signal source (AC wise) and the source impedance is considerably lower, than the bias resistor can be ignored in noise caculation.

 

[mmuj]

I just got to the bottom of the problem, it is actually ac coupling capacitor causing problem. I was using ceramic X7R capacitor for ac coupling. ceramic capacitors are have piezoelectric effect that makes them very sensitive to any vibration around.

Moral of the story, with low voltage signals and precision required, never use ceramic capacitors for ac coupling.

Although it has been very frustrating but i learned a lot

thanks FvM for your help

Mujahid

 

[mmuj]

Yes you are right, my noise calculations were wrong. now i have simulated in LTSPICE and compared with real results they reasonably match .