Why there is a huge spike on the peak on the instrument amplifier out put

[Albert6186]

Hi
    I use a TI ina333  to amplify the signal from a Rogowski coil. At the ina333 output (netlabel ADC_CURRENT), I got the waveform 1. There are huge periodic spikes on the peak of the wave.  The detail is in the waveform 2.
    This circuit is identical to the Ti design TIDA-01063. I just change the opamp and power converter for low power reasons.
    My test board is an empty PCB. Only the components in the schematic are soldered. So, there is no other noise source on the board. I used both a 3.6v battery and lab DC power supply as a power source, and the result is the same.
    Why there is a spike? How can I remove it?
    Thank you.

[RandallMcRee]

This circuit is identical to the Ti design TIDA-01063. I just change the opamp and power converter for low power reasons.

Well, no, I don't think so--it looks like you have a voltage divider as the input to Vref (pin 5).
Datasheet specifically says
REF 5 I Reference input. This pin must be driven by low impedance or connected to ground
Later:
The output of the INA333 device is referred to the output reference (REF) pin, which is normally grounded. This
connection must be low-impedance to assure good common-mode rejection. Although 15 Ω or less of stray
resistance can be tolerated while maintaining specified CMRR, small stray resistances of tens of Ωs in series
with the REF pin can cause noticeable degradation in CMRR.

in the aforementioned TI app note they are using a voltage reference to drive that input.

Another problem (in addition!) could be your CM filtering. It is probably not a coincidence that the spikes occur at the signal extremum. This is where common mode voltages bounds are typically broken. Check your diamond plot for this circuit.

[Albert6186]

Thank you RandallMcRee.
Yes, I neglected that I have changed the vref input.  This circuit is a little different from the TIDA-01063.
However, I directly connect the pin 5 of ina333 to the ground, and the result is the same.
To avoid my first test board may be damaged, I made a whole new test board.
The second board connects the ina333 pin 5 to the ground.  The waveform (wave 3) got at ina333 output is almost identical to the previous one. There are many spikes still at the peak.

Sorry for my limited knowledge,  I can't find the problem of the filter circuit. In fact, that is the TIDA-01063 original design.

I suspect the input signal which comes from a Rogowski coil may be wrong. But when I use another circuit ( two cascaded opamps) to amplify the same signal, the output wave is smooth, and there is no spike.

I can't figure out what happened. Please help.

[Albert6186]

 The second circuit output waveform is here. It's almost the same as the previous one.

[Albert6186]

I think the primary current signal which is measured by the Rogowski coil is correct.

The blue curve in the first picture is the primary current. I use a current clamp to get this signal.
The detail about the spike is also provided in the following picture. When spiking happens, there is no fault on the primary current.

[RandallMcRee]

Hi Albert,
I'm afraid that I cannot help you much further. I do think that you should do the diamond plot exercise that I mentioned in my first post. I believe that TI has a tool that you can use (it is specific to the IA amp that you are using).

If you can't find the tool at the TI site, Analog.com has a good one but only for their own amplifiers. You can try plugging in your parameters and if there are no recommendations that would be a real red flag since both Analog and TI have broadly similar offerings. Notice that the power supply value is a very important input here--I suspect, again, that you are running out of headroom...
https://tools.analog.com/en/diamond/#difL=-0.1&difR=0.1&difSl=-0.1&gain=100&l=-8&pr=AD8422&r=8&sl=-8&tab=1&ty=1&vn=-15&vp=15&vr=0

TI has some good suggestions about common-mode filtering. Where did you get the current filter recommendation?

Lastly, if the two-opamp configuration is working for you perhaps that is a signal from the universe to go in that direction?? 

[Andy Watson]

The output of the ina333 is also much noiser - it looks as if it only marginally stable. Try isolating the output, from whatever follows, with, say, a 100\$\Omega\$ resistor.

[StillTrying]

Check if the spikes are real, without moving the scope's GND clip and with the scope still on AC coupling check the +/- 3.3V supplies near the ina333's pins.

[David Hess]

What kind of power supply is being used?  Those spikes at the peaks of the AC waveform are exactly what happens when some rectifiers in a power supply "snap" back after charging the bulk input capacitance on every half-cycle.  Putting a couple hundred picofarads across each rectifier can suppress it until a better solution is found.

[Albert6186]

Thank you for all.

After I thoroughly check my test environment, there is another coil on the primary line. When I put the probe at the output of the second coil, I got this waveform. I think for some reason the second coil saturated, and this saturated coil made the primary current distorted. Although the distortion is very small, it really exists. So, the spike at the ina333 output is from the di/dt of the distortion.

The blue curve is the output of the second coil.

If the second coil can be removed, I believe the spike will disappear. But in my environment, that coil is permanent equipment.

I think maybe I can smooth the ina333 output waveform by post integrate circuit.

I also changed the filter parameter., R23 and R27 are 5k, and C54 and C56 are 10nF. The output waveform became more clear.  But the spikes are still there. As we now know, the spike is from the primary current distortion, so it can't be removed by the filter.

[Albert6186]

Hi RandallMcRee, the filter circuit is the Ti original design. I just copy it.
I think R23+C54 and R27+C56 are both low pass filters. They are just effective for the common-mode signals.
C56 maybe smooth the differential mode signal.
The R24 and R26 are required by the ina333 datasheet. These two resistors provide a return path for the input bias current.

[jbb]

That’s challenging

One thing to check - in case you’re lucky - is it a Differential Mode (DM) or Common Mode (CM) spike?

Suggest as a diagnostic tool that you temporarily disconnect INA -IN from coil and reconnect it to +IN. Output signal is DM and will go away. But will the nasty spike? If it goes away, it’s CM and could be reduced by CM choke.

Filtering of some kind is required, but filtering the opamp output makes me a little nervous. The nasty spike could be upsetting the innards of the INA (eg internal nodes might saturate) and impacting the response of the INA in the ‘non spike’ times too.

If it won’t wreck your Rogowski coil output, it’s probably better to filter most of the spike out on the input side.

[jbb]

Oops, just looked at your schematic again. I see you have filtering already. Sorry.

Suggest adding 2 more TVS diodes: one from AC+ to ground and one from AC- to ground.

[Albert6186]

Thank you jbb. I added 2 TVS at each input, and the result didn't change much. But I think they are useful, I'll add them in next version.

[splin]

[EDIT] INA118 should read INA188. Datasheet values are those of the 188 but the pricing was for the INA118. the 188 is < $2.

I can't help with the source of the spike but TIDUBV4A seems to be a way to demonstrate the design calculations use of an INA118  rather than a good design. It also seems to have a few errors.

Firstly,  why are they using an instrumentation amp given the Rogowski coil [EDIT] has a single ended output can be grounded to produce a single-ended output?  Presumably because the purpose of the design idea is to showcase the INA118?  In-amps are expensive ($7 for the INA118 @ 1K) and have some performance disadvantages compared to single ended opamps. In the case of the INA118 it is quite noisy at 12.5nV/rt(Hz). Its maximum gain drift @ gains > 1 is 50ppm/C  which is pretty poor. It doesn't include the gain setting resistors' tempco (25ppm in this design), which they have overlooked in their error analysis, taking overall gain drift to 75ppm/C.

A pair of 10ppm resistors costs less than $0.30 coupled with a  $0.22 TLV6741 should outperform the INA118 solution.

More importantly, by referencing the low impedance Rogowski coil to ground only via 100K input biasing resistors will make this design very vulnerable to picking up common mode noise which could easily exceed the supply rails of the amp. The document bizarrely assumes a common mode voltage of 0.005328V for the error analysis. I wonder where that number came from and why so precise? I guess it was measured in their test setup but is a very odd number to use for the error analysis. It doesn't matter much given the CM error contribution is insignificant.

Perhaps a more interesting question is why bother with an amplifier at all?  The ADS131 ADC they suggest has approx 760nVrms input noise with a 12X PGA setting and 4V reference which is less than the 2.9uV noise (273ppm) shown in Table 5. However, that number makes no sense to me given INA118 12.5nV/rt(Hz) noise (RTI) and 1500Hz bandwidth (plus a small but negligable contribution from output noise) which amounts to 484nV. The calculation shown multiplies this by 6 for some reason unknown to me. The same formula appears in the INA118 datasheet on page 26. Surely this is wrong?

As to errors, the document specifies 100x gain but the schematic and BOM show 500x (Rg =  100 ohms). Not really a problem but confusing.

In the error analysis they show .8ppm gain linearity error which should probably be 8ppm. Again no big deal. Most importantly though they state:

From Table 3 through Table 6, the worst case resolution error for the INA188 at a bandwidth of 1.5 kHz is 2.92μV, which is less than expected value.

That overlooks the gain drift with temperature which they calculated at 3000ppm (32uV) but didn't include for some reason. It is actually 4500ppm if you include the gain setting resistors 25ppm TC.

I've probably got some or all of the above wrong so would appreciate some feedback.

[splin]

Note that in my previous post I mistyped INA118 instead of the INA188 several times. I did actually refer to the INA188 datasheet.

This is table 3 from the INA188 datasheet showing the noise calculation:



Does anyone know where the 6x multiplier comes from in the noise calculation? This must be a mistake?

[MiDi]

Does anyone know where the 6x multiplier comes from in the noise calculation? This must be a mistake?

Sounds like the conversion factor from rms to peak to peak value.

[tszaboo]

So, you should test your circuit with proper testing equipment, like signal generator, instead of unknown sources.
Also, forget current clamps. They are slow, and they have false output. Even high end ones. IE, phase is delayed, on high end 500KHz Fluke clamps. Or they do weird things on anything not sinusoidal.

A pair of 10ppm resistors costs less than $0.30 coupled with a  $0.22 TLV6741 should outperform the INA118 solution.

The INA333 is very cheap. And sometimes you just want a working circuit.

[splin]

Does anyone know where the 6x multiplier comes from in the noise calculation? This must be a mistake?

Sounds like the conversion factor from rms to peak to peak value.

Of course, thanks. Don't know why I missed that.