[SOLVED] Core material for high frequency current transformer

[Shuo__]

Hello everyone, after reading a lot of information on the internet, I am now confused and need your help.
The question is how should I choose the core material for high frequency current sensor (1MHz-20MHz). I have gathered two views here :

1. Selection of materials with high magnetic permeability. Materials with high permeability are selected, and their higher permeability usually means that it has less impedance to high frequency signals. This is because the permeability of a core material is the ratio between the magnetic flux density and the magnetic field strength, indicating the responsiveness of the magnetic field in the material. In high frequency applications, the greater the permeability of a core material, the faster the material can respond to changing magnetic fields, and therefore the lower the impedance to high frequency signals.
2. Selection of materials with low magnetic permeability. "With most ferrites, the permeability doesn't just drop at high frequency, the material also becomes absorbent; At "low" frequency you will have real permeability - the coil will work like an inductor, store and release the magnetic field; but at high frequency, you get imaginary permeability, the coil will work like a resistor - the magnetic field will be turned to heat; this is the principle on which ferrite beads work."--From FesZ. According to my understanding, in order to minimize high frequency losses, low imaginary permeability materials should be used. and the materials with low permeability have smaller imaginary permeability at high frequencies than materials with high permeability.

The following plots show the variation of composite permeability with frequency for material 3E6 (MnZn, ui=10000) and material K1 (NiZn, ui=60), respectively.
I would be very grateful for your advice!

 

[Easy peasy]

Have you calculated the flux levels in your CT for worst case ?

If you can keep these to +/- 5mT or ideally less - this will help you a lot

you should choose a modern high freq power ferrite with high resistivity - 3C98 stands out

3F34 is another

--- Updated Apr 5, 2023 ---

Just had a quick look at the latest offerings:

3F46

https://download.ferrite.de/pdf/3f46.pdf

--- Updated Apr 5, 2023 ---

dB / dt = V / N Ae - so it is easy to work out the Bpp flux swing

 

[Shuo__]

Have you calculated the flux levels in your CT for worst case ?

If you can keep these to +/- 5mT or ideally less - this will help you a lot

you should choose a modern high freq power ferrite with high resistivity - 3C98 stands out

3F34 is another

--- Updated Apr 5, 2023 ---

Just had a quick look at the latest offerings:

3F46

https://download.ferrite.de/pdf/3f46.pdf

--- Updated Apr 5, 2023 ---

dB / dt = V / N Ae - so it is easy to work out the Bpp flux swing

I use the current transformer to measure the common mode noise currents on the DC supply line, which only appear when the half-bridge is commutated. The following figure shows the measurement configuration. Because both DC+ and DC- flow through the magnetic ring, I do not need to consider the effect of DC current.
Here is a waveform of the noise current I measured on the LISN.

As you can see from the waveform, the voltage peak is about 2,5V, so the peak noise current on the DC line is 2,5V/50 Ohm = 50mA(max). They flow through the ferrite ring on the DC lines.

I think the flux in the core is very small, but I'm not sure how I should calculate it in this case. Can you give me more advice, thank you very much!

 

[mtwieg]

I recall you were asking about this some months ago. Did you try any experiments based on the advice there?

 

[Shuo__]

I recall you were asking about this some months ago. Did you try any experiments based on the advice there?

Right, thanks for your previous advice. I haven't tried low permeability materials, since the 3E6 material worked very well for me.
As you can see from the Wave I uploaded above (yellow: LISN 50 ohm output, white: CT 50 Ohm output), the CT basically replicates the noise currents very well, and these are in the 1-10 MHz range.
But I can't understand why it works, judging from the complex permeability of the 3E6, it doesn't look right for this frequency range. I thought the loss at high frequencies would absorb the noise current signals

Above is the measurement of the bandwidth of the CT (3E6 primary 1 turn secondary 10turn). This looks bad in the 1-10MHz range, and that spike I think is caused by the stray capacitance between the primary and secondary coils. Phase also looks very strange, it should be gently straight in useful areas.

But when I reversed the CT measurement, things changed. That is, 10turn as the primary coil and 1turn as the secondary coil. Gain in the 1-10MHz range looks much better, and the Phase is also flat. This might explain why it can measure noise currents well, but I'm not sure.
If I use materials with lower permeability, i.e. lower imaginary permeability at high frequencies, is it possible to increase the bandwidth of CT, by which I mean increase the high frequency corner?

 

[Easy peasy]

That is not a LISN - but a scope.

Given the fairly low volts out and 200nS width - you will be fine for peak B

you need a core with higher resistivity for such high frequencies ( ideally > 10 )

Keeping the turns to 10:1 will keep capacitive effects down - but a 5 ohm burden R will be better than

50 ohms,

from the numbers you give, 50 ohm, 0.5V peak, 10 turns the CM current is 100mA peak

depending on the application - this may be quite a bit

 

[mtwieg]

As you can see from the Wave I uploaded above (yellow: LISN 50 ohm output, white: CT 50 Ohm output), the CT basically replicates the noise currents very well, and these are in the 1-10 MHz range.
But I can't understand why it works, judging from the complex permeability of the 3E6, it doesn't look right for this frequency range. I thought the loss at high frequencies would absorb the noise current signals

High frequency loss doesn't necessarily translate to bad performance (but might mean you could get equivalent performance with a smaller core). What are the dimensions of the core?

View attachment 182165
Above is the measurement of the bandwidth of the CT (3E6 primary 1 turn secondary 10turn). This looks bad in the 1-10MHz range, and that spike I think is caused by the stray capacitance between the primary and secondary coils. Phase also looks very strange, it should be gently straight in useful areas.

View attachment 182166
But when I reversed the CT measurement, things changed. That is, 10turn as the primary coil and 1turn as the secondary coil. Gain in the 1-10MHz range looks much better, and the Phase is also flat. This might explain why it can measure noise currents well, but I'm not sure.

The first plot doesn't look abnormal to me, including the phase response (except I would expect the phase to rise at the lower corner frequency instead of drop).

The second plot being very different might be reasonable too, depending on your measurement setup. If you were measuring it with a VNA then it should look identical in both directions (S21=S12) but I'm assuming there's something about your method of measurement which explains this.

If I use materials with lower permeability, i.e. lower imaginary permeability at high frequencies, is it possible to increase the bandwidth of CT, by which I mean increase the high frequency corner?

Upper frequency limit is likely dominated by parasitic capacitance and burden resistance, so just changing core material won't have much effect. Try first decreasing the burden resistance.

 

[Shuo__]

That is not a LISN - but a scope.

Given the fairly low volts out and 200nS width - you will be fine for peak B

you need a core with higher resistivity for such high frequencies ( ideally > 10 )

Keeping the turns to 10:1 will keep capacitive effects down - but a 5 ohm burden R will be better than

50 ohms,

from the numbers you give, 50 ohm, 0.5V peak, 10 turns the CM current is 100mA peak

depending on the application - this may be quite a bit

The channel with the yellow waveform is connected to one side of the LISN, and the input resistance is selected to be 50 ohms.

The output of the CT is connected to the amplifier circuit via a coaxial cable, so I used a Feed-through Terminator (50 ohms) as my measurement resistor, which I think is a better impedance match.

The 100mA peak you mentioned is actually the sum of the CM currents on DC+ and DC-, so the real CM current is 50mA peak.

I'm not quite sure what you mean by "higher resistivity for such high frequencies", do you mean there should be high imaginary permeability at high frequencies? "ideally > 10" Which parameter does this 10 correspond to?

 

[Shuo__]

(but might mean you could get equivalent performance with a smaller core)

Can you please explain what you mean by that?

What are the dimensions of the core?

I don't know the exact size right now, but the size in that picture below should be pretty close to what I'm using

The first plot doesn't look abnormal to me, including the phase response

The sudden change in phase angle at 300kHz is abnormal for me, do you know what causes that?

I am using a VNA Bode 100, and the source power is the same for both measurements. I really don't know why the results are different

Upper frequency limit is likely dominated by parasitic capacitance and burden resistance, so just changing core material won't have much effect. Try first decreasing the burden resistance.

You are right, I have also tested the magnetic ring of nanocrystal material (u ~ 80000), number of turns and burden resistance remains the same, and the upper frequency basically does not change.

The output of the CT is connected to the amplifier circuit via a coaxial cable, so I used a Feed-through Terminator (50 ohms) as my burden resistor, look at the pictures below

I did this because I thought it would be a better output impedance match. And the 50ohm resistor inside the Terminator has better performance at high frequencies, i.e., lower parasitic parameters.

But I don't know if this is necessary, maybe I can use normal SMD resistors and then smaller resistance values to get higher upper frequency.

But there is a trade-off here, the amplifier circuit will get a lower input voltage, so I need to turn up the gain of the amplifier, this will also be a significant reduction in the bandwidth of the amplifier. Maybe I just need a higher bandwidth amplifier

 

[Easy peasy]

The total CM current is 100mA - you do not know exactly the split

Higher resistivity of the core - in ohm.meters - this is given on most data sheets

There is a jump in resistivity from MnZn to NiZn ferrites, with the latter being much higher and therefore less lossy at higher frequencies - but their Ur is generally lower ...

for short coax you will only need to terminate one end correctly to avoid reflections

keeping the volts down on the termination resistor on the CT will help with BW and core losses.

 

[mtwieg]

Can you please explain what you mean by that?

Basically, flat frequency response requires that the magnetizing impedance Zm of the secondary be much higher than the impedance ZL loading it (the burden resistor, plus leakage impedance). Like your 3E6 plot shows above, effective permeability (and thus magnetizing impedance) starts to drop off as frequency rises. But this doesn't actually affect performance so long as the Zm >> ZL. Which means even a poor material can get high frequency response if you just increase Zm (add secondary turns, increase core area, etc) or decrease ZL (decrease burden resistor).

The sudden change in phase angle at 300kHz is abnormal for me, do you know what causes that?

Whenever you see a jump from -180 to +180 degrees, you can assume that it's just an artifact of how the software is plotting phase. It will take any value outside of the -180/+180 range and map it back inside that range, meaning any time the real phase crosses those limits it will be displayed as wrapping to the other extreme. Maybe there's a setting you can change to make it unwrap the phase for you.

View attachment 182182

I am using a VNA Bode 100, and the source power is the same for both measurements. I really don't know why the results are different

Not very familiar with this instrument, but I'm assuming it's due to the two ports having different terminations. Can you share exactly how you're connecting your DUT to this device?

The output of the CT is connected to the amplifier circuit via a coaxial cable, so I used a Feed-through Terminator (50 ohms) as my burden resistor, look at the pictures below
View attachment 182184
I did this because I thought it would be a better output impedance match. And the 50ohm resistor inside the Terminator has better performance at high frequencies, i.e., lower parasitic parameters.

But I don't know if this is necessary, maybe I can use normal SMD resistors and then smaller resistance values to get higher upper frequency.

Cheap SMT resistors typically have very good performance, and can be used into the GHz range (with proper layout).

But there is a trade-off here, the amplifier circuit will get a lower input voltage, so I need to turn up the gain of the amplifier, this will also be a significant reduction in the bandwidth of the amplifier. Maybe I just need a higher bandwidth amplifier

Yes, there's always going to be a tradeoff between bandwidth and gain. Here's what the frequency response of an example CT can be expected to do when you change the burden resistance (red=50ohms, blue=5ohms, green=1ohm):

Basically you're just changing the frequency response in the center, flattening it. The transfer function beyond the passband is mostly unaffected.

Extending bandwidth without compromising gain is much harder. Need to reduce parasitics, especially secondary capacitance.

 

[Shuo__]

Basically, flat frequency response requires that the magnetizing impedance Zm of the secondary be much higher than the impedance ZL loading it (the burden resistor, plus leakage impedance). Like your 3E6 plot shows above, effective permeability (and thus magnetizing impedance) starts to drop off as frequency rises. But this doesn't actually affect performance so long as the Zm >> ZL. Which means even a poor material can get high frequency response if you just increase Zm (add secondary turns, increase core area, etc) or decrease ZL (decrease burden resistor).

I think you're saying that the CT lower cutoff frequency, which can be expressed simply with this formula:
fL=RBurden / 2*pi*Lm
If I only increase the number of turns, that is, raise the Lm, so that I can get a smaller lower cutoff frequency. But increasing turns will also introduce more parasitic capacitance, which will slightly reduce the upper frequency.
Since I'm not interested in signals below 1MHz, it looks like reduce the burden resistor will help me to get higher upper frequency. Or try to reduce the parasitic capacitance, but like you said, that would be difficult

Can you share exactly how you're connecting your DUT to this device?

Actually your question reminded me that my measurements seem to be wrong.
Channel 1 is connected to the CT input and Channel 2 is connected to the CT output. so the transfer function is gain(phase)=VCH2/VCH1. But I just realized that the inputs Channel 1 and Channel 2 are set to 1MΩ by default.
I should select 50Ω input impedance at CH2 or continue to use the default1MΩ but add 50Ω burden resistor at CT output. Did I think wrong?

--- Updated Apr 7, 2023 ---

Higher resistivity of the core - in ohm.meters - this is given on most data sheets

There is a jump in resistivity from MnZn to NiZn ferrites, with the latter being much higher and therefore less lossy at higher frequencies - but their Ur is generally lower ...

I'm a bit confused now, As shown above the impedance in the high frequency range is mainly composed of Rac which also represents the core loss. So higher impedance should mean more loss, not lower loss, or did I misunderstand.

When I look at most of Ferroxcube's materials, NiZn ferrites generally have very small permeabilities, and they also have smaller imaginary permeabilities at high frequencies compared to MnZn, which also means less core loss. So according to your opinion, NiZn should be a better choice, do I follow correctly?

 

[D.A.(Tony)Stewart]

You said "I use the current transformer to measure the common mode noise currents on the DC supply line,"

But what is your end goal? Increase CM suppression of noise of emissions? Then you want a lossy ferrite with high impedance not an ideal current transformer. You may want to raise the CM impedance relative to the often high impedance floating shunt impedance but also increase loss absorption.

You can measure Vcm, Icm and Zcm or attenuate Vcm by raising Zcm relative to the shunt impedance (e.g. ~ 1nF Line+Neutral to PNG.

Isolated DC converters are notorious for CM noise problems. Which ones are yours?

 

[Shuo__]

You said "I use the current transformer to measure the common mode noise currents on the DC supply line,"

But what is your end goal? Increase CM suppression of noise of emissions? Then you want a lossy ferrite with high impedance not an ideal current transformer. You may want to raise the CM impedance relative to the often high impedance floating shunt impedance but also increase loss absorption.

You can measure Vcm, Icm and Zcm or attenuate Vcm by raising Zcm relative to the shunt impedance (e.g. ~ 1nF Line+Neutral to PNG.

Isolated DC converters are notorious for CM noise problems. Which ones are yours?

I am working on a project about active CM noise cancellation, these CM noises come from the electric motor, which is controlled by the inverter. so I need a measuring circuit to detect Icm and process it as input to my control system. The measurement circuit should not attenuate the original Icm as much as possible.

 

[D.A.(Tony)Stewart]

There is a very wide range of CM chokes. It starts with your expectations. DCR, L, SRF or Z(f)
In SMD

Is this what you are doing?
'

--- Updated Apr 8, 2023 ---

https://www.researchgate.net/publication/337721665_Common-mode_EMI_Noise_Analysis_and_Reduction_for_AC-DC-AC_Systems_with_Paralleled_Power_Modules

 

[Shuo__]

There is a very wide range of CM chokes. It starts with your expectations. DCR, L, SRF or Z(f)
In SMD

Thank you very much for sharing. That paper looks very helpful. The noise currents I want to measure are in the 1-10MHz range, so the core material better have a small impedance in that area. The 3E6 material I'm using doesn't meet that requirement, but he's still working pretty well. As mtwieg said above, high frequency loss doesn't necessarily translate to bad performance. Maybe I'll try NiZn materials, they have less impedance at high frequencies

 

[D.A.(Tony)Stewart]

you can model the RLC values from the slope

Or better, choose the desired RLC value then the parts to match.

The rise in impedance slope is due to L and above this the lossy R from eddy currents. The falling slope is due to C and the peak is SRF.

 

[mtwieg]

I think you're saying that the CT lower cutoff frequency, which can be expressed simply with this formula:
fL=RBurden / 2*pi*Lm
If I only increase the number of turns, that is, raise the Lm, so that I can get a smaller lower cutoff frequency. But increasing turns will also introduce more parasitic capacitance, which will slightly reduce the upper frequency.
Since I'm not interested in signals below 1MHz, it looks like reduce the burden resistor will help me to get higher upper frequency. Or try to reduce the parasitic capacitance, but like you said, that would be difficult

If you really don't care about signal below 1MHz, you can reduce your turns and burden resistance together and this should significantly increase your upper cutoff frequency, but without compromising passband gain as much as decreasing burden resistance alone.

View attachment 182200
Actually your question reminded me that my measurements seem to be wrong.
Channel 1 is connected to the CT input and Channel 2 is connected to the CT output. so the transfer function is gain(phase)=VCH2/VCH1. But I just realized that the inputs Channel 1 and Channel 2 are set to 1MΩ by default.
I should select 50Ω input impedance at CH2 or continue to use the default1MΩ but add 50Ω burden resistor at CT output. Did I think wrong?

If you were to enable the 50ohm termination on CH2 but leave CH1 at 1Mohm, the instrument would present 50ohms to both sides of the DUT. That should result in the transfer function of the DUT being the same in both directions. That doesn't mean this is the "correct" way to characterize the DUT. That's a matter of interpretation.

 

[voyile3]

1. Permeability: As you mentioned, materials with high magnetic permeability can be beneficial for high frequency applications as they can respond faster to changing magnetic fields. However, it's worth noting that the relationship between permeability and frequency is not straightforward, and some materials may exhibit losses or other effects at high frequencies that could affect their performance. Therefore, it's important to consider the specific frequency range of your application and choose a material that is optimized for that range.
2. Imaginary permeability: This is a measure of how much energy is lost as heat when the magnetic field in the material changes. As frequency increases, many materials exhibit higher imaginary permeability, which can lead to significant losses. Therefore, choosing a material with low imaginary permeability can be beneficial for minimizing losses in high frequency applications.
3. Other factors: There are many other factors to consider when choosing a core material, including temperature stability, cost, availability, and compatibility with other components in your system. Therefore, it's important to evaluate all of these factors when making a decision or hire expert.

Regarding the specific materials you mentioned, it's difficult to make a recommendation without knowing more about your application requirements. However, based on the plots you provided, it appears that Material K1 (NiZn, ui=60) may have lower losses at higher frequencies than Material 3E6 (MnZn, ui=10000). However, it's worth noting that the plots only show one aspect of the materials' performance, and other factors may be more important in your specific application.

 

[Shuo__]

If you really don't care about signal below 1MHz, you can reduce your turns and burden resistance together and this should significantly increase your upper cutoff frequency, but without compromising passband gain as much as decreasing burden resistance alone.

If you were to enable the 50ohm termination on CH2 but leave CH1 at 1Mohm, the instrument would present 50ohms to both sides of the DUT. That should result in the transfer function of the DUT being the same in both directions. That doesn't mean this is the "correct" way to characterize the DUT. That's a matter of interpretation.

Thank you very much for your answer and help, I will let you know if I have any progress