Lowpass filter at the output of AD8421 (Instrumentational Amplifier)

Hi Kenny,

Good day! I will be working with you on your situation using part. But first, just to confirm, are you measuring the green trace at the output of the AD8421?

Best regards,

Franz

Hello Franz,

I am measuring the net "AMP_OUT" which is the blue trace, it is the output of the AD8421 after it passed the LPF at the very right of the schematic. The green trace is there just to compare the blue trace against to understand effect of the added LPF. 

Thank you!

Kenny

Hello Kenny,

your spike has a risetime (from 10% to 90%) of approximately 85µs. This risetime of the signal is sigificantly reduced by your low pass filter. From theory one would expect a 10%-90% risetime of 120µs for your low pass filter. And your simulation shows this effect clearly: the blue trace rises much slower than the green one.

But the length of your spike is much longer than 120µs - the spike takes several ms to decay. And since the duration of the spike (several ms) is much longer than the time constant of your low pass filter (55µs), the shape of the spike is hardly changed (except for the fast rising edge).

best regards

Achim

Hello Achim,

Thank you for sharing your finding, increasing the time constant of the filter solves it then; however, the duration of the spike is much larger which is less ideal for my application... The two seem inversely proportional but is there a way to suppress this spike with low settling time?

Regards,

Kenny 

Hello Kenny,

your distortion pulse consists of a broad range of spectral frequencies (including a DC-component). There is no simple way to "remove" the spike by low pass filtering. Better spend your efforts in avoiding the spike in the beginning and not in filtering it afterwards.

The jump in common mode voltage will of course always cause some spike, and the best way to overcome this is an amplifier with high common mode rejetion at the desired frequency. Another option could be to simply "ignore" the output of the amplifier some ms after the common mode jump.

But in your case the spike is not caused by insufficient CMR, it is caused by inappropriate input circuit. The filtering of the input signal of your amplifier is not symmetrical: R3 is not equal to R5, C3 is not equal to C2. This seemingly slight assymmetry converts the common mode jump at in+ and in- to a spike in differential voltage Filt+ - Filt-

That is the real reason for the huge spike, and that is the point, where you should optimize your circuit. (at the output of the amplifier it's too late...)

Avoid this assymetry, and make your input filter as symmetrical as possible. Rerun your simulation and compare the difference signal V(filt+,filt-) with symmetrical and non symmetrical filter. In the real pcb take the two filter capacitors C2 and C3 from the identical batch (you will anyways not be able to buy capacitors, which differ just in the sixth significant digit).

Also try shorter time constants for your input filtering - your values for R3, R5, C1, C2, C3 are comparatively huge, and this causes the long duration of the spike. Optimize this filtering of the differential input of the amplifiert. This will be much more effective to reduce the spike than any attempt to remove the spike afterwards at the amplifier output.

best regards

Achim

Hello.

Thank you for your response and the suggestions. The large time constant of the input filter is part of the noise requirements, we're interested only in the low frequency range signals... I agree that the input circuit needs an improvement so I'll implement the recommended changes accordingly; however, is it possible to still retain that low frequency filtering by adding a low cutoff single-ended LPF at the output of the amplifier? What are some drawbacks of such design?

Best Regards

KennyWizard said:

The large time constant of the input filter is part of the noise requirements, we're interested only in the low frequency range signals...

Well, with regard to the spike, your input filter does not solve any problem, it causes the problem. Have a look at the attached simulations:

- on top you see your original circuit. The waveform window shows the output of the AD8421 and its differential input multiplied by 100. You can see, that the spike is not caused by the AD8421, it is already present at the input of the AD8421 (due to your asymmetric input filter). The AD8421 just amplifies the input signal just as it should.

- in the middle you see the results, if the input filter has the identical asymmetry (the spike has the identical height), but the time constants of the input filter are one order of magnitude smaller (the spike is much narrower). This narrower spike could be filtered more effectively at the amplifier output.

- on bottom you see the results with a complete symmetrical input filter (which is easyer to implement in simulation than in real life). On the scale shown, no spike can be seen. (If one would zoom in, the effect of non perfect common mode rejection would show up, but it's negligible compared to the spike of your input filter.)

KennyWizard said:

however, is it possible to still retain that low frequency filtering by adding a low cutoff single-ended LPF at the output of the amplifier? What are some drawbacks of such design?

Of course a LPF at the output of the AD8421 is possible. But it is not effective in suppressing the spike for which you started this thread. And a simple RC filter has quite "unhurried" filter characteristics. It attenuates the frequencies above the cutt-off frequency only relatively weakly.