How accurate are Coilcraft's real inductor models in LTSpice?

[NQ21HT449]

Hello, I'm trying to simulate real inductors in my filter circuits using Coilcraft's library for LTSpice but I'm getting unacceptable losses. Here's an example:
First, I synthesize a Chebyshev BPF at 25 MHz and it shows an ideal curve.

I then perform a parametric search on Coilcraft's website and I select ceramic inductor with high Q and low ESR.

Then I replace ideal inductor with their frequency model inductor and I get -0.925 dB from just a single inductor. In this example, there's only one inductor, if I simulate higher order circuits, with multiple inductors, I get even more loss, which is not going to work for our application.

I'm trying to understand, firstly, whether my setup in LTSpice is correct and secondly, what is the source of the loss. It seems that DCR is the culprit.

If that's the case, is it possible to design a high order band pass filter working in low MHz range with real inductors or is there a fundamental limitation of low Q at those frequencies?

 

[BradtheRad]

Coilcraft has been a leading name in electronics for years and is likely to make correct models.

By testing various L:C ratios you can see how it affects the rolloff curve. Anyway that's the theory for my simulation. At top are your LC values. Results of a sine sweep are the uppermost scope traces.

The other scope traces display that by increasing Farad value and decreasing Henry value you can obtain different shapes for your bandpass curve. The resistor values make a difference as well. By minimizing resistance you can increase Q.

 

[FvM]

Problem is that Q numbers listed in comparison don't match SPICE model https://www.coilcraft.com/en-us/models/spice/?seriesName=1008CS&partNumber=1008CS-820
The latter results in considerably higher ESR at 25 MHz, presume LTspice library uses the same model. Table Q of 31 produces about -0.5 dB S21 for your test resonator. Unfortunately I don't know which value describes part behaviour at 25 MHz better. The table suggests that higher Q is achievable with other (bigger) parts.

 

[FvM]

Q of frequency dependant model at 25 MHz shows as 16.4
You may want to ask Coilcraft which values are right.

 

[D.A.(Tony)Stewart]

Does LTspice compute the skin effect resistance from k in Coilcraft's model?

Shouldn't your filter have > |-6dB| loss?

 

[FvM]

Does LTspice compute the skin effect resistance from k in Coilcraft's model?

Yes, Rvar= k*sqrt(f) is modelling skin effect.

Shouldn't your filter have > |-6dB| loss?

It's LTspice s-parameter simulation mode (.net command), compensating port matching loss. The setup calculates s21 correctly.

 

[BradtheRad]

Shouldn't your filter have > |-6dB| loss?

I agree it's automatic because the input has a given series value of ohms while the output has an equal value of ohms to ground. Thus the filter taps at the middle value of a resistive divider. My simulated output (post #2) is one-half the amplitude of the input. So that can be seen as equivalent to automatic 6dB loss.

Performance (rolloff curve) changes shape when these resistor values change.

 

[NQ21HT449]

From Coilcraft:

"Since data sheet specifications are based on typical production measurements, and the SPICE models are based on de-embedded measurements as described below, the model results may be different from the data sheet specifications. Therefore, the online tool (1-port Q) will typically not match the model (2-port, air dielectric fixture) result for Q. The dependence of SRF and Q on the substrate is very sensitive."

I think I should be using their two-port Q SPICE model for filter design.

 

[FvM]

The SPLCE model provided by Coilcraft is no 2-port model, it has no common mode elements. In so far I don't understand the statement. More interesting question is, which Q value represents better behaviour e.g. in a LC filter.

 

[BigBoss]

I recommend you to use equivalent circuit of the coils of Coilcraft. They are more accurate.

 

[D.A.(Tony)Stewart]

If you want a high-order low-loss Chebychev BPF at 25 MHz you would approach the component choice differently with higher Q's for each LC pole.

You stated a 0.1 dB BW from 24 to 26 dB yet plotted the -6dB normalized attenuation instead of showing the -3dB BW.


What are your passband & stopband design specs?

 

[FvM]

I recommend you to use equivalent circuit of the coils of Coilcraft. They are more accurate.

Equivalent circuit with skin effect loss term is exactly reproduced in LTspice model. Basic question is why parametric search table gives considerably higher (about doubled) Q numbers.

 

[NQ21HT449]

If you want a high-order low-loss Chebychev BPF at 25 MHz you would approach the component choice differently with higher Q's for each LC pole.

You stated a 0.1 dB BW from 24 to 26 dB yet plotted the -6dB normalized attenuation instead of showing the -3dB BW.


What are your passband & stopband design specs?

This example was selected for simplicity, the filters are already synthesized with the ideal inductors.

 

[FvM]

O.k., let's return to the original question.

is it possible to design a high order band pass filter working in low MHz range with real inductors or is there a fundamental limitation of low Q at those frequencies?

There is a practical limitation, in case of air core inductors mainly set by DCR + frequency dependant skin effect resistance. No matter if SPICE model Q of about 16 or comparison table Q of 31 gives the right picture, it's most likely not good enough for your purposes. According to comparison table, other (larger) air core types have up to fourfold Q. For range up to a few MHz, closed magnetic path ferrite core inductors with air gap promise Q numbers up to several hundred.

 

[D.A.(Tony)Stewart]

When choosing higher-order sharp filters the max Q of each stage must increase and be staggered to minimize the passband ripple and steepness of the skirt to bandstop.

Also in realizing high Q = X/R resonant LC impedance ratios the circuit Q must be much higher than the component Q.

As already shown higher X and L values may lead to higher Q values if ESR is the same. This proves why I stated previously to use much higher (>> 100x) L values for high-order low-loss RF filters at 25 MHz.

I would suggest component Q must be 10x bigger to easily realize high Q LC stages in high Nth-order LC filters.

For a simple 2nd order filter with Q each component affects the result like parallel impedance.

Ref: Wiki

Conclusion:

82 nH is not a reasonable choice for a high-order 25 MHz filter but it is for much higher f since ESR from skin effect only increases by the sqrt root of f while XL is linear.

In active high Q filters the GBW constant rule and must be multiplied by the square of Q.

G*BW of circuit * Q² = GBW required of Op Amp. for Q>1.

 

[FvM]

In a passive LC filter with defined impedance, L can't be chosen freely.
The trivial 1st order example presented in post #1 clarifies the point, center frequency and bandwidth determine L. Only alternative topologies (e.g. first element series or shunt) gives a degree of freedom.

To check implementation options, we need to know the filter specs.
Good filter design software can deal with individual inductor and capacitor Q, but of course only calculate feasible implementations, said fundamental limit.

 

[D.A.(Tony)Stewart]

I got a prompt reply from Coilcraft which was already in the datasheet.

The datasheet values are based on 1-port measurements on an impedance analyzer. The models are based on wideband measurements using a special (air dielectric) 2-port fixture and a network analyzer to cover a wide range of frequencies.

The most accurate models are the substrate-dependent global models offered by Modelithics.

Thank you for choosing Coilcraft.

Best regards,

Chris / Christopher Hare | Technical Marketing | USA

Coilcraft, Inc.

--- Updated Aug 2, 2024 ---

I circled the L values in ceramic which yield the highest Q, consistent with my initial advice.

Ferrite inductors would give a different spectrum of Q's as would air-core.