The effect of cascading method on Butterworth performance

I'm designing a Butterworth microstrip bandpass filter, and I have calculated its order by using Bilinear method and I got it as 31st order. The frequency band is from 27.5 GHz to 31.225 GHz, and I have tried to simulate it on ADS by using cascading method (added 10th order 3 times with 1st order) but I'm getting unexpected results for insertion and return losses as shown in the figure, I even tried to do optimization but I couldn't get any enhancement for such a high order like that, and I showed my design to my supervisor and he said that the cascading method is correct but i have to work on the performance, otherwise I have to show him a research paper saying that the cascading method has a negative effect on filter's performance. So what is wrong with my design, why I'm not getting the desired results?

Your 10th order pieces are designed with 50 ohm source and load impedance. When you cascade the pieces, they no longer see the broadband 50 ohm source and load. Instead, they now see the impedance of your other 10th order filter piece. How good is that matched to 50 Ohm in the passband?

yep, very true. In the rejection band, the filter does NOT have 50 ohm input impedance, so it does not load the previous filter properly.

You CAN cascade two complete bandpass filters together, but you need to add a ferrite isolator, or perhaps a 3 dB pad between the two to get the simulated results. Also to get ACCURATE simulation results, the length of line connecting complete filters together is critical. A half wavelength line between two stop bands miraculously turns the combination into a bandpass where there should be high rejection.

BTW, my experience is that anything over say 13 element filters is just a pipe dream. Something more like a 9 element filter is about all i would seriously consider...expecially up at 24 GHz.

As biff44 wrote: Cascading passive filters is not a good idea.
Show your supervisor the book of Hong/Lancaster (Microstrip Filters for RF/Microwave Applications)
and he will hopefully understand that you don't cascade passive filters (very rarely). Do not confuse this with active filters, where decoupled blocks are often cascaded. Here is an example with 2 bandpass filters (n=4, RL=20dB)

Design Opportunities:

1) Tolerance errors on high Q of cascading filters will certainly destroy ideal RL of each cascade. This requires a Monte Carlo analysis to determine the max Q you can guarantee. Given tolerance of 10% on dielectric constant unless test with TDR at board shop. Butterworth high order will be terrible and Chebychev 0.1dB even worse.

2) You forgot to specify the Passband ripple, group delay error and Bandstop rejection specs.

To illustrate the Qmax needed for a 20th order filter;

Qmax
10 Bessel
18 Linear Phase 0.05'
23 Linear Phase 0.5'
54 Butterworth
75 Gaussian to 12dB
126 Gaussian to 6dB
300 Chebyshev 1dB ripple
100 Chebyshev 0.001dB ripple

Group Delay error , Q and Passband Ripple or Return Loss are tradeoffs with tolerance error.

In general, I would say don't attempt to make a high Q multistage filter with higher Q than you can make otherwise poor results may be expected.

Rather, choose a low Q design with more amplitude error but more linear phase shift or less group delay error. It depends on the sensitivity of the modulation scheme used to amplitude and phase errors.

If the skirts have excessive group delay or ripple you can widen the passband 5~10% to use up some margin. But all the above tradeoffs come from well defined must have filter specs and tolerance sensitivity to manufacturing.

These 3 stages cascaded 10th filter is not feasible at all in all aspects.If you type its' specifications, we can say something..

Yes, we discussed the problem before:
https://www.edaboard.com/showthread....=1#post1607352

The design will fail because Monzerje has no appropriate substrate available. And it seems that the supervisor doesn't do his job.

I have gone through all your comments, and Yes I already know that my design is incorrect and not logical at all. However, what is the most suitable way to overcome this problem since my project is only for simulation, and I am not required to fabricate it?. Should I take Niki's advice and inform my supervisor that this design can't be used or?

what is the filter for? It seems someone has done a very poor system design, and sent you on a hunt for some unobtainium.

It is up to you to figure out if your telling the emperor he has no clothes will either get you praised or fired.

I can send you a design and simulation that works, but why bother you have insufficient specs.

To make it, one needs ceramic or similar stable loss loss tangent substrate.