Hi

I am a mechanical engineer so, please, read what I write with a grain of salt.

For what it is worth, I know from measurement equipment and my HiFi hobby that OP amps like also stable thermal conditions as resistance variation affects amplification and similar.

Perhaps you should take a look at thermal fluctuation. As long as fluctuation exceeds, say 50 - 100 mK, you should be able to use a an InfraRed camera to spot where such fluctuation arises.

Some people actively cool electronics to get high performance. Some IR cameras have sensors cooled to 70 k.For simpler ways, take a look at thermal mass and other cooling tricks that are used for PC/GPU processors.

Just my 2 cents.

Sincerely

Claes

Dear Rujun,

The ADA4898 is stable even at unity gain. I suspect the circuit surrounding the opamp is causing the problems. Give us some more information so we can help you find the problem, ideally send the circuit diagram:

What do you mean by unstable? is it a fixed frequency, or drift? can you send a picture of the waveform.

What overall gain are you using?

What resistors? (carbon are very noisy and wire-wound inductive)

What capacitors? (electrolytic are noisy)

What bandwidth are you measuring the noise? and with what input?

Best Regards

Paul

I agree with Paul. and add to him, There is a device noise and circuit noise. In order to verify the performance parameters of a device you have to stick strictly to the given instructions in the manufacturer data sheet for measuring such parameters specially the noise performance.

The noise given in the data sheet is the root of the spectral noise power density in nV/ square root of Hz.  So if you measure the total noise you have to integrate the spectral noise density over the amplifier bandwidth.

You have to take into consideration the noise generated by the external input resistances or input sources. As Paul; said you have to choose metal film resistors which is less noise that the carbon film resistor. Also, the external resistance at the amplifier input must be as small as possible. Also, it may be very helpful to restrict the bandwidth of the amplifier to just the required frequency range.

Concerning the ripples you must suppress then by good regulation and filtering.

One reaming  note, what do you mean by the stability? is it the frequency stability?

or what? or noise? 

wish you success

Dear Rujun,

I do agree with Professor Paul,

The stability depends on the environment, type of capacitor/resistor used in the circuits.

Regards

Hi, Claes Richard Fredo,

Your suggestion may be right. The unstable condition happens when the input is large. The temperature may rise greatly.  I'll adjust the load resistance to watch what happen. 

Dear Paul Howard Riley,

Many thanks for your help. I use ADA4898 in unity gain buffer and noninverting amplifier. The circuit is very sample. If I use ADA4898 as unity gain buffer and the input and output is open, the output of ADA4898 will be oscillating. If I use AD549 in same condition and same circuit, the output of AD549 is good. It seems that ADA4989 have very strict requirement to the input resistance and output resistance. Parasitic capacitance will cause strong oscillating in ADA4898.

For noninverting amplifier, I use relays to control gains. The gains are 1,5,25. I use good resistors and COG capacitors, the bandwidth of my circuit is about 3 MHz,and I use one-order RC filter to limit the bandwith because I use a 24 bit ADC (AD7760) to convert conditioned signal to digital signal. My problem is that if the input is saturated about 3 us several times by a 50 Hz impulse, the drift of ADA4898 will exceed 1 mV. It seems that the drift of ADA4898 is out of control.

To my opinion, there exist very strict requirements to the usage of ADA4898. These requirements are not exist for AD549. 

Rujun 

Dear Adbelhalim Zekry,

I use 1% COG capacitor and 0.1% metal film resistors in the cicuit.

It seems that the requirement to use ADA4898 is very strict. We study the datasheet reference board carefully, and design our PCB board several times, but we can not achieve simulated noise performance. The noise is about two times of simulated result. The parasite capacitance in the circuit may cause oscillating and damage the noise performance. I use ADC code from AD7760 to calculate to input noise. When the gain is less than 25, the noise is mainly caused by ADC. The noise is mainly caused by amplifier cicuit when the gain is great than 25.

Furthmore, the transient signal is not satisfying, the bandwidth for large signal and small signal is different. I find the following from datasheet.

The dynamic performance (+ - 15V supply, Rl = 1 kΩ):

-3 dB bandwidth is 65 MHz for 100 mV p-p Vout,

-3 dB bandwidth is only 14 Mhz for 2 V p-p Vout.

Rujun Chen

Hi

Again, I advice a grain of salt as I do not have any electronics backround. Some thoughts for what they are worth

A) The 100mV/2V amplitude behavior.

i) Narrow PCBs conduits and high load introduce additional resistance. On the same note, a 'cold soldering', i.e. soldering that did not cure properly, may introduce similar phenomena.

ii) Have you excluded grounding problems that cause DC shift across the PCB and offset amplifier zero points?

iii) I have never tried it but, as a thought, cooling spray that you use to test soldering - why not try spraying individual components with it when operating at high load? If there is a power related change, it would be logical for it to change when greatly cooled. This way you may get lucky and find if one or more components affect your results.

B) On a more general note as theory and practice are off.

Have you eliminated the influence from exterior excitation? To exemplify, capacitors act as poor microphones and pick up sound. I imagine that resistors react to mechanical load.

Claes

Dear Rujun,

You compare the noise from the simulation with noise from the measurement.  It is so that the noise model in the simulator is not accurate enough to give values approaching that measured. One of the major discrepancies may arise from the different temperatures of the two circuits. Normally the practical circuit may be at an appreciably higher temperature because of the self heating  especially at large signal operation.This will cause more thermal noise. Also, the large current causes larger shot noise. 

You have to consider that in the simulation. 

Concerning the bandwidth of the op amp, it decreases with the signal amplitude. This is okay. In order to stabilize the amplifier you have to kill the oscillations by adding a compensating capacitor  at the intended node if any other wise you can add it at the output.

best wishes

The input impedance of the ADA4898 is not infinite so you have to take account of it when running simulations, or in specific circuits. (All real world amplifiers are imperfect) Here are some examples that you may find useful.

Input voltage offset: zero input volts will not give zero output volts; the higher the gain the more obvious this will be.

Input current offset: the current coming out of the inputs are not zero. So if you put an infinite input resistance the output will vary. Have a look at the spec sheet for variation of input offset current with temperature. It is relatively flat at room temperature, but can become very significant at higher temperatures.

If you want to measure signals from a high impedance source, you have to take special precautions; get back to me if that is the case and I can elaborate.

As mentioned above the input resistance has Nyquist noise, are you taking both voltage and current input resistance noise into account?

https://en.wikipedia.org/wiki/Johnson%E2%80%93Nyquist_noise

Vasile is taking a very purist view when he says "..nonsense to simulate noise.. ". Although by definition, one cannot predict the noise voltage at any instant, as long as averaging over time is used, noise parameters can be predicted. Indeed, there are whole scientific areas dedicated to this. Here is one example in RF amplifiers:

https://en.wikipedia.org/wiki/Noise_figure

As for oscillation stability. Although the amplifier is unity gain stable, that does not mean it will not oscillate, give the right parameters. Here is an example of a simple oscillator http://wackyvorlon.typepad.com/old_school_hacker/2007/10/opamp-oscillato.html

For high frequency op amps, (even unity gain ones) parasitic circuit elements can cause oscillation with poor circuit layouts. Are you using a PCB, or some experimental board? Experimental boards can induce extra capacitance and inductance that cause oscillation. This is why most high frequency PCBs have a ground plane around tracks. In this way parasitic elements are referred to ground, rather than another unwanted part of the circuit.

Best Regards

Paul