Ebay Rogowski current probe

[moffy]

I just received the following item from Ebay: https://www.ebay.com.au/itm/224784699913
and after some tests have come up with the following equivalent circuit see JPG.

Most of the capacitance appears to be the cable and shield and if I put 250 ohm resistors between OUTA and shield and OUTB and shield the response is fairly flat up to about 700kHz.
The output needs to be followed by an integrator of some description. Haven't measured the sensitivity yet, except to notice that loading the output reduces sensitivity, but that makes sense as the load current will tend to cancel the input current and reduce sensitivity.

[moffy]

Update on sensitivity, the signal input is 100ma p-p:

1. No load: 20mv p-p @ 200kHz, 9mv p-p @ 100kHz, 4.5mv p-p @ 50kHz
2. 1kR between OUTA - OUTB: 19mv p-p @200kHz, 9mv p-p @ 100kHz, 4mv p-p @50kHz
3. 500R between OUTA - OUTB: 16mv p-p @200kHz, 8mv p-p @ 100kHz, 4mv p-p @50kHz bandwidth confirmed as approximately 600kHz

I have to take it back, loading does not appear to significantly change sensitivity. The differences could be due to measuring a moderately noisy scope trace by eye and the differences to the Q of the LC with and without loading and of course the 50R series resistance. 500R also produces a well damped response when a square wave current is used as the input signal.

[moffy]

Just an update, I built my first version of the Rogowski coil amplifier and it seems to work but looks like it is dominated by 1/f noise. Since the signal is differential, P2:3-1, U1B inverts one side which is fed into U1A and summed with the other side, this helps eliminate common mode noise. Because of the nature of the Rogowski coil, sensitivity increases with frequency, U1A is an integrator. U2, the OP07, sets the output of U1A near to zero, a few mv doesn't matter. R7, R14/C3 form the feedback, connecting to R6 and the +ve input of U1A. C3 stops the DC control loop from oscillating by providing some feed forward at the low end and R14 limits the noise injected from U2 into U1A. The traces show the 1/f noise on the output. Trace is 5mv/div and excitation is 100ma p-p @ 500Hz. The response looks reasonably flat as a 1kHz square wave looks reasonably good. Not sure how to deal with the 1/f noise, except maybe a sharpish high pass filter @ 10Hz. Any suggestions would be appreciated.

[Slh]

The high pass filter works well in my experience. I've built a few DC coupled ones and they've always had a lot of low frequency noise.

The best one that I've made used the cheapest, most rubbish amplifier available https://hackaday.io/project/190985-tl431-rogowski-coil-current-sensor and has the cleanest output signal thanks to the basic rc high pass filter.

[moffy]

The high pass filter works well in my experience. I've built a few DC coupled ones and they've always had a lot of low frequency noise.

The best one that I've made used the cheapest, most rubbish amplifier available https://hackaday.io/project/190985-tl431-rogowski-coil-current-sensor and has the cleanest output signal thanks to the basic rc high pass filter.

It looks like about a 3rd order high pass, and like we have similar projects, but I am impressed you wound your own! Thanks for sharing.

[Slh]

Thank you. Winding the coils isn't too hard. Making good, high frequency, flat output ones probably is but I haven't figured out a test set up for them yet so they clearly must be perfect

I'm not great with noise calculations etc as the stuff I work on typically has big enough signals that the amplifier noise doesn't crop up. Naïvely, looking at your circuit, it looks like there are three op amps adding all of their 1/f noise. is there any way that you can amplify it before going into the integrator? I guess it would have to be something AC coupled anyway so maybe you wouldn't gain anything but possibly worth a shot.

[moffy]

Thank you. Winding the coils isn't too hard. Making good, high frequency, flat output ones probably is but I haven't figured out a test set up for them yet so they clearly must be perfect

I'm not great with noise calculations etc as the stuff I work on typically has big enough signals that the amplifier noise doesn't crop up. Naïvely, looking at your circuit, it looks like there are three op amps adding all of their 1/f noise. is there any way that you can amplify it before going into the integrator? I guess it would have to be something AC coupled anyway so maybe you wouldn't gain anything but possibly worth a shot.

Only 2 of the amplifiers really contribute, U1A and U1B, the noise from U2 is pretty attenuated with respect to U1. At the low frequencies the integrator gain is around 15,000, so 5mv output noise is about 0.3uv of input noise. Amplifying before the integrator wont help as the amplifier will still amplify its 1/f noise, it won't make any real difference. The best way might be to use a very low noise differential pair but even then it might improve only by a factor of 3 or 4, the MC33078 is a surprisingly low noise opamp with 4.5nV/(Hz)^0.5.

[moffy]

It is possible that using the AD8599, though expensive, could improve the noise by a factor of 5.

[Slh]

I'm probably talking rubbish here but one approach for discrete parts is to put a bunch in parallel. I feel like opamps might not like that too much but it's all I've got.

[TimFox]

I'm probably talking rubbish here but one approach for discrete parts is to put a bunch in parallel. I feel like opamps might not like that too much but it's all I've got.

Connecting discrete components in parallel (with appropriate biasing circuits) is a common technique to reduce the input noise voltage, but it increases the input noise current.
This is useful when the input comes from a low-impedance source, and is used with (for example) moving-coil phono cartridges.
For identical components, the voltage noise density (V/Hz^1/2) scales down with the square root of the number of devices, and the current noise density (A/Hz^1/2) scales up with the same square root.

[Conrad Hoffman]

Circuit seems overly complicated and it always makes me nervous when opamps are combined within one feedback loop. There are tons of Rogowski coil amplifier/integrators online and none of them take this approach, save for a patent. You may have already see most of this but some interesting stuff from TI here- https://www.ti.com/tool/TIDA-01063 Patent here- https://patentimages.storage.googleapis.com/c5/5b/38/a9adf29825509d/US9588147.pdf Plus a mess of stuff right here- https://www.eevblog.com/forum/projects/rogowski-coil-with-integrator/

IMO, the LF noise problem is going to require a good choice of opamps, as it can't be filtered out.

[moffy]

Circuit seems overly complicated and it always makes me nervous when opamps are combined within one feedback loop. There are tons of Rogowski coil amplifier/integrators online and none of them take this approach, save for a patent. You may have already see most of this but some interesting stuff from TI here- https://www.ti.com/tool/TIDA-01063 Patent here- https://patentimages.storage.googleapis.com/c5/5b/38/a9adf29825509d/US9588147.pdf Plus a mess of stuff right here- https://www.eevblog.com/forum/projects/rogowski-coil-with-integrator/

IMO, the LF noise problem is going to require a good choice of opamps, as it can't be filtered out.

It's not overly complicated, just two opamps to convert differential to single ended as well as an integrator, that is quite efficient, and one opamp to zero the DC around the integrator. I have seen the TI app notes before. The differential to single ended circuit I use is not original, but a slight reversal of the single ended to differential drivers that some high precision differential ADCs use as the input drivers. Thanks for the other references I'll have a look.
As I mentioned previously the AD8599 might be a better fit but it is expensive, but don't discount the MC33078 it is a very decent low noise, low distortion opamp and is excellent value for money. It is possible that the noise could be thermally related i.e. the resistor to solder to pad induced thermal emfs, that would be hard to distinguish from 1/f noise.

[moffy]

I'm probably talking rubbish here but one approach for discrete parts is to put a bunch in parallel. I feel like opamps might not like that too much but it's all I've got.

Connecting discrete components in parallel (with appropriate biasing circuits) is a common technique to reduce the input noise voltage, but it increases the input noise current.
This is useful when the input comes from a low-impedance source, and is used with (for example) moving-coil phono cartridges.
For identical components, the voltage noise density (V/Hz^1/2) scales down with the square root of the number of devices, and the current noise density (A/Hz^1/2) scales up with the same square root.

Thanks, I am aware of the practice, but as you mentioned it improves as the square root, not a lot of value for complexity unless you have a target noise figure that you have to meet. If it is 1/f noise then the AD8599 is probably the best choice, but if thermally induced emfs then boxing the board and layout/symmetry would be important. Guess I'll start with a simple 3rd or 4th order highpass filter might solve all the issues.

[moffy]

I thought I would include some 1kHz square wave responses @ 100ma p-p as they reveal a lot about the transfer function.
The yellow trace is the current output, and the blue trace is the voltage across the resistor whose current is being sensed.

[moffy]

To test whether the noise was due to induced emf or thermal imbalances, I wrapped the circuit board in a nice blanket of gaffer tape then sealed it in with grounded aluminium foil. There was a slight reduction in noise with grounding the shield aluminium, but the majority of the noise remained so I believe most of the noise is 1/f from the opamps. There is a slight peak around 18Hz because of the OP07 feedback which also gets excited by the noise and is noticeable. I also realised that U1b has a noise gain of 2 but a signal gain of -1 so the next circuit will also address that. I have gone with the OPA1612, which has an input noise figure of 1.1nV/(Hz)^0.5 and low 1/f noise. The integrator amp will have less gain but a simple 1Meg resistor across the feedback capacitor for DC, that should remove any low frequency peaks. The integrator will be followed by an OP37 with a gain of 100 that is also part of a HP filter. I have included the rough schematic, values might change somewhat but should be reasonably close to final values.

P.S. I also used a battery supply which made no difference.

[Conrad Hoffman]

Progress!

[moffy]

Progress!

I hope so, guess we'll see when I get the new opamps.

[moffy]

I have finished laying out the PCB for the next version. The integrator will be followed by a fourth order HP Butterworth filter @ 20Hz, with a gain of 100. All four opamps will be the OPA1612 for optimal noise performance, but I will be able to interchange them with the MC33078 to see what the performance differences are like. It will have 10x more gain than the first version, and will be BW limited to 1MHz. Should get the OPA1612s sometime next week. Included is the revised schematic.

[moffy]

I have received the OPA1612 opamps and initial results are looking much better. I hacked the original MC33078 circuit, to use a 10n capacitor with a 1Meg resistor in parallel for the integrator, giving a 15.9Hz corner frequency and 1/10th the gain used by the MC33078. I then repurposed the OP07 to be a 10x gain amplifier after the integrator to give a similar overall response. The following captures show the reduced noise.

[moffy]

Sorry for the long delay but there have been complications with regards to phase distortion, the problem of increasing signal level with frequency and the limited HF bandwidth of the coil. Because I have been looking to see how effective this Ebay coil could be as a current probe I ran into the issue of uneven phase distorting the shape of the signal e.g. a square wave, even just 5 degrees between fundamental and the harmonics can cause it to look differentiated. This meant that the high pass filters that were needed to filter out the 1/f noise had to be shifted down and phase compensated, making them less effective. They also limit the LF performance to well above their 3db point. I tried Bessel/linear phase filters but they made little difference. This is an ongoing task.
The increasing signal level with frequency makes it impractical to put a fixed amplifier like a MIC preamp as the first stage as it would run out of headroom at higher frequencies for any meaningful dynamic range. So an integrator needs to be the first stage with the problems of noise introduced by the integrating resistor.
Finally the limited BW because of the capacitance, probably due to the cable, introduces again phase distortion and limits the usefull BW from the top. I could of course cut the cable and see if that extends the upper BW but I have constrained myself at present just to working with the coil as purchased.
There is also the issue of 50Hz common mode rejection which requires a differential approach. I have gone through two revisions of boards, different approaches but the 1/f noise is always dominant. Pity more to follow.

[moffy]

This is the latest schematic which involves 2 stages of integration, the LF is handled by the U1A as a classic integrator. The second higher frequency part is handled by the inductance of the Rogowski coil and R2/VR1 are meant to handle the crossover frequency, but at present that is not perfect due to some last minute mods to deal with common mode noise, hence the spike at each transition. The caps C10/C11 are not fitted, they provide some LF phase compensation but also introduce a LF peak that about doubles the noise waveform, so the square waves have more distortion than desired. The yellow traces are the output of the amplifier, the blue trace is the voltage across the 50R resistor passing through the Rogowski coil. There are some noise traces with the coil fitted but no signal and a 50R resistor fitted in its place as a dummy noise measurement.

[moffy]

Just a final note, I think I have gotten close to the best noise performance without heroic efforts like multiple opamps paralleled. The gain at 10Hz is around 180,000 so the 15mV p-p output noise is equivalent to about 83nv p-p noise on the input which is very respectable. Output sensitivity is around 1.2mV/mA or 1.2V/A from around 50Hz to around 300kHz, limited more by the phase response. The circuit would need fixing for the uncompensated part of the 2 stage integrator, because of a last minute change to pseudo differential input, due to minute levels of 50Hz and its harmonics. I hope you enjoyed the journey and might have even learned something, as I surely did during the process.