Turki Aommran submitted the September question of the month:
Dear Harlan,
I have designed microwave filters with Agilent ADS and the simulation results were perfect, but I have run into a problem with the measured results, which displayed huge differences. What are the effects from this huge impact in my measured results? What caused the mismatching in the reciprocity? S11, s22 should be equal, but in reality s11 is not equal to s22, whereas they have the same ship. What caused the resonance frequency shifting and what causes the drastic insertion losses between the measured and the results? Notice that my microwave filters are a parallel-coupled filter and planar lumped-element filters
The winning response to the September Question of the Month is from Peter Militch, Honeywell Technology Solutions Inc. Congratulations, Peter!
From: Peter Militch, Honeywell Technology Solutions Inc.
Dear Turki,
The design will make several assumptions and one very important one will be that the actual filter, when physically realized, employs components with electrical properties that match the model within the range of assumed model tolerances. If a lumped element filter was constructed, it is safe to assume that the key parameters (reactance, coupling to other components, impedance of the connecting links) did not precisely match the values within the model's assumed tolerances. As to S11 and S22 not matching, I assume this is a symmetric layout that should have reciprocal properties at the input and output ports. Here, it is most likely that the input and output matching impedances are not equivalent and not correct. Thus, when terminating the output and performing an S11 measurement, and then terminating the input and performing an S22 measurement, different results are obtained. Bottom line: It sounds (without seeing the final result) that the filter was not built with the level of control needed on device values, coupling, matching impedances, etc. This would shift the actual behavior away from the expected results.
From: Satyajit Chakrabarti, SAMEER Kolkata Centre
Dear Turki,
The following problems can occur: (1) You may not have considered the effect of connector lead inductance while simulating in ADS. When one connects the connector there can be overlap on the PCB. (2) Maybe, in the circuit that you have fabricated, there could be a hair crack in the PCB that is not visible to the naked eye. This hair gap will give some additional capacitance that may be the cause of your frequency shift and impedance mismatch at the ports. This can also cause drastic change in the insertion loss. (3) If you have put the connector inside a box, maybe the ratio of the connector radius and drill hole diameter is not proper. As a result, it is not matched with the connector impedance or the input impedance of the PCB. (4) You may replace the connector and see whether the connectors are all ok.
From: Milan Motl, Flextronics Design
Dear Turki,
It could have two reasons. Firstly, you mentioned that s11 and s22 should be equal. I expect that the filters are designed as symetrical. Are the input and output connections the same? Secondly, the tolerance of the layout of planar lumped-elements could cause the differences. Have you done a tolerance analysis?
From: Dan Woodward, Tektronix Inc.
Dear Turki,
The differences in S11 and S22 are probably caused by multiple reflections from the discrete elements that make up the filter. These reflections are never equal from both ends, plus the losses in the discrete elements are not the same. This makes position important, which creates different reflections from each end.
From: Greg Milford, UNSW
Dear Turki,
In essence, what you have built is not exactly the same as the circuit you have simulated. Small variations in component values - slightly inaccurate microstrip track widths and coupled line spacings or lumped element component tolerances, for example - can have a significant de-tuning effect on resonant structures such as filters. Similarly, uncertainty in the planar substrate materials dielectric permittivity (i.e. differences between the value you used in the simulation and the board's actual value) will have the effect of scaling electrical line lengths, thereby causing a shift in the filter's resonant frequencies. A de-tuned filter will have a larger insertion loss than the tuned response. Insertion loss in the correctly tuned planar filter is most likely due to substrate losses or losses (i.e. finite Q) in lumped element components. The bottom line: know your components!
From: Lester Cheung, Blessed Electronics
Dear Turki,
As with any work that attempts to correlate simulation and fabricated hardware design, the criteria is how well the software is able to simulate the reality or practical conditions of the fabricated hardware. In this case, if there is a huge difference between simulated and fabricated (real) design, the following questions must be asked: 1) How robust was the design? How far did the designer consider the effects to the fabricated design when he simulated it? Did he consider possible fabrication errors of say 5 percent, 10 percent or 20 percent? Was the "perfect" simulation results the product of only considering "perfectly" fabricated hardware? 2) What was the extent of the difference between the dimensions of the simulation design and the dimensions of the fabricated design? Unfortunately, fabrication will never be perfect, more so if high frequency designs that require smaller line widths are considered. Thus, if there are line shifts in fabricated hardware, this may explain differences and phase shifts in measured S11 and S22 as well as the other differences between simulated and fabricated measured results. 3) How accurate were the assumptions used for the simulation? During the initial phase of any simulation work, various variables to properly represent the materials used in the fabrication must be input. The accuracy of the variables (e.g. permeabiilty of the material used) input should also be studied. 4) Lastly, of course, this may be due to environmental conditions during measurement and testing. As a possible methodology to study the reason behind the differences seen by Mr. Aommran, it may be worth measuring the dimensions of the fabricated design for comparison against the simulation design. Secondly, it may also be worth checking specification data of the fabrication material to see what is the manufacturing error or uncertainty for the material (5%? 10%? 20%?). If necessary, the designer may need to simulate for various nodes of possible variable shifts. If after modifying the simulation design and variables to match that of the fabricated design still results in insufficient accuracy, then the designer may have to check the environment in which the measurements were taken.
From: Sandeep Chaturvedi, GAETEC
Dear Turki,
A very common reason for shift in resonance frequency and mismatch in S11, S22 is the fabrication tolerance. If the fabrication tolerances are not taken care of during design, the fabricated parallel coupled filter strips will have different line gaps and widths, which will lead to frequency shift and S11, S22 mismatch. Another factor that plays a role in shift in resonant frequency is the consideration of open ends of the parallel sections. It is well known that due to fringing at the open ends of a transmission line the effective length of the line increases. This increase has to be offset during design to avoid the resonant frequency shift.
