How to calculate the bandwidth from simulation graph

[Monzerje]

I have designed a microstrip Chebyshev bandpass filter, and I got the simulation result as shown in the figure below. Can anyone explain to me how can I find the bandwidth from that graph in order to compare it with the theoretical value?

 

[niki]

In this case defining the bandwidth does not make much sense. First of all the filter should be properly tuned.
There is a reflection zero outside the passband (26GHz). For bandpassfilter you can take the corner frequencies with the the equal minima return loss, e.g. 20dB (

 

[BradtheRad]

You might look at your band filter as being divided into two sections: a low frequency section, and a high frequency. Run these individually through Chebyshev formulas.

Or, could this be a question whether you should start the rolloff curve at 3dB (as for a first order filter)? Or should it be 6 or 9 or12 dB, etc., as is considered an alternate way of specifying higher order filters which have a steeper curve?

 

[albbg]

Place a marker (let me say m1) to the minimum in-band attenuation, then place other two markers m2 and m3 having a delta of 3 dB with respect to m1.
You can the calculate the central frequency as Fc=[f(m2)+f(m3)]/2 and the 3dB bandwidth as BW=f(m3)-f(m2) (supposing you placed m3 at the higher frequency).

However the bandwkdth is usually specified at 3dB rolloff, but not always, sometimes it can be defined also as 1dB BW or, less frequently, as other deltas. In this case you simply have to place the two delta markers at the wanted delta and apply the same formulas.

Very roughly, from you graph would be: minimum insertion loss=0.5 dB @ 30GHz, then f(m2)=27.2GHz, f(m3)=31.4GHz from which:

Fc=29.3GHz
BW(-3dB)=2.1GHz

 

[Monzerje]

First of all the filter should be properly tuned. (

Why do I need to do tuning since my frequency band (27.5 - 31.2 GHZ), and how can that enhance my result?

- - - Updated - - -

You might look at your band filter as being divided into two sections: a low frequency section, and a high frequency. Run these individually through Chebyshev formulas.

Or, could this be a question whether you should start the rolloff curve at 3dB (as for a first order filter)? Or should it be 6 or 9 or12 dB, etc., as is considered an alternate way of specifying higher order filters which have a steeper curve?

Why do I need to divide it into two sections?
Regarding to the rolloff, the filter order has already been defined which is 9th order

- - - Updated - - -

Place a marker (let me say m1) to the minimum in-band attenuation, then place other two markers m2 and m3 having a delta of 3 dB with respect to m1.
You can the calculate the central frequency as Fc=[f(m2)+f(m3)]/2 and the 3dB bandwidth as BW=f(m3)-f(m2) (supposing you placed m3 at the higher frequency).

Very roughly, from you graph would be: minimum insertion loss=0.5 dB @ 30GHz, then f(m2)=27.2GHz, f(m3)=31.4GHz from which:

Fc=29.3GHz
BW(-3dB)=2.1GHz

I think in the example you provided, you wrongly divided the BW/2, so the answer should be 4.2GHz. Anyway, I have followed your method and I got BW=4.13GHZ and FC=29.3GHz, is there any possible thing that can I do to reduce the bandwidth?, where according to the theoretical calculation the bandwidth should be BW=3.72 GHz

 

[niki]

Why do I need to do tuning since my frequency band (27.5 - 31.2 GHZ), and how can that enhance my result?

- - - Updated - - -

Show me your design of the filter (ADS schematic). Maybe you can improve something with optimization.

 

[BradtheRad]

Why do I need to divide it into two sections?

It's merely another path to try looking into, that is, to break a big complicated puzzle down into smaller parts which are easier to solve individually. Of course I'm not an expert and I can't say which approach does or doesn't work. The high frequency rolloff may display a different curve from the low frequency rolloff curve. Thus they might need to be examined individually. Some filters have more than one stage to process a signal.

Regarding to the rolloff, the filter order has already been defined which is 9th order

Then the cutoff frequency probably is determined by a different measurement than for a first order filter. Here's a quote from the Wikipedia article:

The common practice of defining the cutoff frequency at −3 dB is usually not applied to Chebyshev filters; instead the cutoff is taken as the point at which the gain falls to the value of the ripple for the final time.

https://en.wikipedia.org/wiki/Chebyshev_filter

 

[D.A.(Tony)Stewart]

fo=29.31 GHz (mean of 27.09 + 31.53)
f-3dB=4.44 GHz
PB Ripple = <0.5dB error
-15dB BW = 6.2 GHz (stopband)
A11= <-10dB PB return loss

n=6 or 9 unique?

 

[Monzerje]

Show me your design of the filter (ADS schematic). Maybe you can improve something with optimization.

I did not use lumped elements, since they cannot be used for high frequency


- - - Updated - - -

fo=29.31 GHz (mean of 27.09 + 31.53)
f-3dB=4.44 GHz
PB Ripple = <0.5dB error
-15dB BW = 6.2 GHz (stopband)
A11= <-10dB PB return loss

n=6 or 9 unique?

Can you explain something?, and how did you get these values??

 

[volker@muehlhaus]

I did not use lumped elements, since they cannot be used for high frequency
View attachment 144702

This filter will not work properly in hardware.

You geometry parameters are too extreme, the schematic models are not accurate at these aspect ratios. Check the model parameter limitations in online help.

And even with proper MCLIN parameters, you need to include MSTEP for the steps in width and MOPEN for the open end effect.
After optimizing at schematic level, you need to check with EM (Momentum).

 

[Monzerje]

This filter will not work properly in hardware.

You geometry parameters are too extreme, the schematic models are not accurate at these aspect ratios. Check the model parameter limitations in online help.

Actually the pervious parameters are much better than the one in the figure which they have been obtained from the optimization. I checked about the limitation values, but I did not find anything about it. So what is your suggestion here?

And even with proper MCLIN parameters, you need to include MSTEP for the steps in width and MOPEN for the open end effect.
After optimizing at schematic level, you need to check with EM (Momentum).

I have seen many journals, but no one has mentioned anything about MSTEP or MLOC for this type of design. I looked again and I found that MSTEP used in circuits to model the junction effects (but I did not understand what that means), So can you elaborate more about them please? or if you have any relevant website.

 

[D.A.(Tony)Stewart]

fo=29.31 GHz (mean of 27.09 + 31.53)
f-3dB=4.44 GHz
PB Ripple = <0.5dB error
-15dB BW = 6.2 GHz (stopband)
A11= <-10dB PB return loss

n=6 or 9 unique?

Just by reading the graph m1,m3, m2 and choosing -15db for SB

 

[volker@muehlhaus]

Actually the pervious parameters are much better than the one in the figure which they have been obtained from the optimization. I checked about the limitation values, but I did not find anything about it. So what is your suggestion here?

It seems that you are using a thick substrate? That results in wide lines for a given line impedance. Your coupled lines are almost square (!) which results in poor accuracy of the model. MCLIN works best for lines that are long compared to the width. For your short lines, each segment has parasitic coupling to the next element, which is not included in your piecewise analysis. You will need EM simulation (Momentum) so see that parasitic coupling.

Also note the model frequency limit from MCLIN documentation: Simulation frequency ≤ 25/H[mm] (GHz)

I have seen many journals, but no one has mentioned anything about MSTEP or MLOC for this type of design.

You will not find this in journals because this is basic simulation technique known for 30 years. See ADS online help for these elements.

MSTEP is used to model the step in width (discontinuity effect) if you connect two MLIN with different width. This will introduce parasitic series inductance and/or shunt capacitance, which is modeled by the MSTEP. MSTEP has no layout, it just models the parasitics for more accurate results.

MLOC models the open end effect of a transmission line that is left open. Again, this has no layout, and just models the parasitics for more accurate results.

Instead of using MCLIN with MLOC, you can use MCFIL. These are coupled line segments for such coupled line filters, including the open end effect.

 

[Monzerje]

It seems that you are using a thick substrate? That results in wide lines for a given line impedance. Your coupled lines are almost square (!) which results in poor accuracy of the model. MCLIN works best for lines that are long compared to the width. For your short lines, each segment has parasitic coupling to the next element, which is not included in your piecewise analysis. You will need EM simulation (Momentum) so see that parasitic coupling.
Also note the model frequency limit from MCLIN documentation: Simulation frequency ≤ 25/H[mm] (GHz)

Yes i am using FR4 with 1mm thickness since it's the only available material here in Malaysia.
Here is the EM simulation for the design

Regarding for the MCLIN limitation, I think yeah it cannot be used for my design since Fc= 29.36 GHz

You will not find this in journals because this is basic simulation technique known for 30 years. See ADS online help for these elements.
MSTEP is used to model the step in width (discontinuity effect) if you connect two MLIN with different width. This will introduce parasitic series inductance and/or shunt capacitance, which is modeled by the MSTEP. MSTEP has no layout, it just models the parasitics for more accurate results.
MLOC models the open end effect of a transmission line that is left open. Again, this has no layout, and just models the parasitics for more accurate results.
Instead of using MCLIN with MLOC, you can use MCFIL. These are coupled line segments for such coupled line filters, including the open end effect.

I was looking in MCFIL limitation and it is similar to MCLIN Simulation frequency ≤ 25/H[mm] (GHz), even though I tried to use it but I did not find the place to add the even and odd characteristic impedances or even how to calculate the S, W, and L since MCFIL is not included in LineCala tool.
If I use MCFIL, I still need to add MSTEP between them right?

 

[volker@muehlhaus]

Yes i am using FR4 with 1mm thickness since it's the only available material here in Malaysia.

Then don't forget to put realistic values for loss tangent into your MSUB, so that you can see the effect of FR4 losses.

I was looking in MCFIL limitation and it is similar to MCLIN Simulation frequency ≤ 25/H[mm] (GHz), even though I tried to use it but I did not find the place to add the even and odd characteristic impedances or even how to calculate the S, W, and L since MCFIL is not included in LineCala tool.

You misunderstand the workflow. MCFIL is just a more accurate model for MCLIN used in these filters (two open ends). So the calculation Linecalc is the same, it is just another model for simulation.

If I use MCFIL, I still need to add MSTEP between them right?

For the large steps in width that you have: yes.

 

[Monzerje]

Then don't forget to put realistic values for loss tangent into your MSUB, so that you can see the effect of FR4 losses.

Yes, the tangent loss has already been considered. Thanks for the reminder

You misunderstand the workflow. MCFIL is just a more accurate model for MCLIN used in these filters (two open ends). So the calculation Linecalc is the same, it is just another model for simulation.

Yes, I got your point, I have to replace MCLIN with MCFIL, but my point is, should I use the same physical parameters of MCLIN that have been calculated by ADS before optimization?

 

[niki]

FR4-Substrate with a height of 1mm is not a good choice for a bandpass filter at 30 GHz. Look at the layout of the filter: the microstrip models are no longer accurate or fail completely (MCFIL, MSTEP,
MCLIN, MOPEN,...). You can see this in the EM-simulation, which delivers very different results than the circuit simulation with the microstrip models.

 

[Monzerje]

FR4-Substrate with a height of 1mm is not a good choice for a bandpass filter at 30 GHz. Look at the layout of the filter: the microstrip models are no longer accurate or fail completely (MCFIL, MSTEP,
MCLIN, MOPEN,...). You can see this in the EM-simulation, which delivers very different results than the circuit simulation with the microstrip models.

Yeah, I know FR4 is not a good material, and I prefer to chose roger duroid instead, but FR4 is the only option I have, and I am not expecting a good performance as the simulation results

 

[niki]

Do you only have 1mm substrates? Or can you get thinner substrates?

 

[niki]

In the attached pdf you see em simulations (full wave) of the bandpassfilter for fr4 substrate with
1mm and 0.5mm thickness of substrate.
With 0.5mm it looks like a fiter, but with 1mm. But with 1mm it's useless.
EM simulation was done with CST MWS.