Something is very wrong with the prediction - S11 mustn't be bigger than 0 dB. You might have the odd 0.1 dB due to calculation errors and the results might still be be helpful, but with S11 over 3 dB in lots of places this simulation is completely wrong. Are you plotting the right thing?

Yes,of course.

I expect it is a mistake in setting up the simulation.

Try simulating a microstrip on the same PCB, as long as the PCB with 50 ohms impedance, and if that works, then make it wider in stages, and move the excitation to the corner, and see where it goes wrong.

Yes, good suggestion. May be the issue is due to the technology definition where the reference plane is defined as conductor instead of cover.

Matrix size: 2596 (reduced: 2420)

--- WARNING -------------------------------------------------------------------

The size of the plus pin for S-parameter port 1 is electrically large

above 0.150588 GHz, S-parameters may become unphysical.

This is the message I get when starting the simulation of a very wide microstrip modeling the plane. Moving vertically the position of simulated port on the microstrip the result doesn't change and is not corresponding with the measurement even if now it is more reasonable.

This is a collage showing the simulation (up to10Ghz) vs measurement (up to 4Ghz)

A little step ahead: I used a dummy microstrip to place the injection port at x=0 y=52mm point of the upper plane. The simulation runs but the result seems not to be in line with the measurement (see attached collage).

I'd go back to the previous one, which is nearly right up to 1.33 GHz, and try changing something else, like add a transmission line to the left to feed it, that looks something like the real feed and that the software doesn't complain about.

I'm not sure if your feed was in the corner for the measurement. At low frequencies a corner feed and mid-edge feed may give similar results, but at higher frequencies there will be significant differences. That may be why it goes wrong above 1.33 GHz.

I don't think the reason of ADS's mistakes is the feed point position.

I performed the measurements at 3 different points and the results are

obviously different at frequencies above 100MHz. It interesting to see

what happens for the resonances due to reflections at the borders.

The maximum number of resonances happens when the feed is at the

corner and they are the union of resonances happening when the feed is placed

at the middle of the edges. ADS seems not to be able reproduce this behavior when I put the feed at the middle of 104mm edge as I tried in the last case.

See for reference: Research EXPERIMENTAL WIDEBAND CHARACTERIZATION OF A PARALLEL-PLATE C...

Working in the frequency domain is very critical when I try to reconstruct the time domain step response of S11. This applies for both the measurements and for simulations. The behavior at very low frequencies including the DC point is very difficult to be reproduced starting ffrom the fd data. A small error on S11 at low frequencies is able to greatly distort the td waveform which is roughly an exponential due to the charge of the parallel plate capacitor. Even applying Vector Fitting to the fd data can add a great distorion at low frequencies.

Even adding a short TL at the feeding point, the Momentum result is wrong.

At the moment I don't know what is going wrong. These programs are usually good. Things go wrong when we do not understand the question we have actually asked them. I would try to make sure that the feed and the boundary conditions are actually the same as in the measurements. When you say that your feed point is in the middle of the top plate that rings alarm bells for me because I would always expect the feed to be between two points, on the top and bottom plane. ("I used a dummy microstrip to place the injection port at x=0 y=52mm point of the upper plane.")

If I switch to Momentum RF the result changes completely and seems more realistic even if it doesn't match the measurement.

Here we are : previous stack up has cond. plane 1 non referenced to gnd. Fully redefining the stackup and defining cond 1 as COVER the results are in good agreement with measurements. Previous results are related to cond1 plane left FLOATING

Looks quite good. It is hard to get better with such a complicated response, apart from a slight shift and amplitude taper that can be corrected by changing material parameters. The second simulation probably has estimated the er for the board a bit too high, and both probably underestimate the loss.

Yes, I agree. The problem is still the same: to select the right input parameters to match the measurements. For this reason measurements are always needed.

Most codes require either experience of modelling similar kinds of problems, or a good understanding of what the code requires, for there to be confidence in the results. This isn't restricted to software - to get reliable measurements often requires similar levels of experience or understanding.

Congratulations on finding the solution. I commend your approach to this problem, and your response to comments. I am not sure the comments helped, but they may have made you annoyed enough to increase your determination to get there.

Discussion is always important and helps to find the solution of problems. This is particularly true in the research activity. In this case I have a very little experience in using ADS. My general goal is to compare the simulated results with actual measurements carried out using a TDR (CSA 803C) or a very low cost VNA (nanoVNA V2.2, a 70Euro USB instrument). The parallel plate configuration is a very significant benchmark even if very simple to implement. It applies to both Power Distribution Network (PDN) of pcbs and Patch Antenna design. About thirty years ago I developed a very simple algorithm to simulate this configuration within the DWS general purpose Digital Wave Simulator usinga 2D mesh of lossy Transmission Lines modeled witha extremely simple PWL approximation (2-segment) of S21. The results were very good in terms of accuracy and speed. A TLM-PWL model of such configuration requires about 100 millisecond to be simulated. The method was integrated in the post-layout pcb simulator PRESTO produced by the company HDT founded in 1988. The results were excellent as confirmed by several benchmarks including radiation patterns carried out also by industrial customers.

Technical Report MODELING AND SIMULATION OF P.C.B. POWER AND GROUND DISTRIBUTION PLANES

Presentation PRESTO POWER

This is what happens in time domain as step response calculated by MAUI application of Teledyne Lecroy applied to the spectra. There is a progressive shift of reflections related to the ADS simulation with respect the measured one. This can be due to an overstimation of dielectric constant (4.6) used in ADS. The DC capacitance on the contrary is bigger in the measurement, that is in contrast with a lower dielectric constant. This discrepancy can be due to an evaluation error of S11 on both the measurement and simulation. The TDR measurement is more reliable than the VNA onefrom this point of view.

Zoomed view:

The underevaluation of losses is evident in the simulation. Losses are very difficult to predict because are strictly dependent on the manufacturing process of the pcb board.

The dielectric constant varies with frequency, so there may be one error in its value at DC and the opposite at RF.

Yes, that's true. But in this case there is a significant difference in low frequency behavior between measurement and simulation. Momentum fits the simulation data using a vector-fitting algorithm. Using the adaptive option to cut the time required by calculation the number of simulated frequency can be very low (in the order of 50). This means that the result of VF interpolation can be affected by a non negligible error. On the other hand, the VNA measurement can be also affected by an offset error when S11 is approaching the unity value. This explain the significant difference between the calculated step responses when the time increases. The only reliable method to get a "true" step response is a true TDR measurement. In the attached plot it is possible to point out the differences in frequency domain between measurement and simulation.

Here a comparison of step responses among the actual TDR measurement (CSA 803C) and three different simulation models (ADS Momentum RF, PEEC and TLM-PWL). It can noticed that both PEEC and and TLM-PWL (both processed by DWS) match the measuremt while ADS Momentum model behaves not so well, especially for steady state. The simplest and fastest model is TLM-PWL that runs in hundreds of milliseconds compared to minutes required by ADS.

A lot of useful comparisons here. Codes are often best suited to one type of problem and not others because of the method they use.