The settings tab provides you with options on basic filter characteristics and more advanced options for configuring the optimizer. Tap to edit an option, for instance tap the name of your design to rename it.
The matrix can be represented as either physical or normalized. A physical matrix representation provides the designer with actual physical couplings, external qualities and resonator frequencies which is preferred in most situations. A normalized matrix is independent of the fractional bandwidth and center frequency. The chosen matrix representation is then applied to the coupling matrix view and tuning controls.
As a side note, the foundations of Couplings Designer uses the normalized coupling matrix since it’s more efficient to compute. The center frequency and fractional bandwidth is therefore used to determine how the coupling matrix is scaled and visualized. Keep in mind, filter synthesis is not dependent on these options but the realization is!
Configuring the optimizer
Getting the best performance out of the optimizer depends on the situation. In all optimization problems a compromise has to be made between accuracy and run time. To improve the speed try using as few points as possible, in most situations even 101 points provides sufficient accuracy. If few points are used the accuracy gets highly dependent on the frequency range that is being optimized. The range is defined as the global minimum and maximum frequency of all goals in the optimization. The frequency range divided by the number of points gives the step size, or accuracy. It is therefore always a good habit to minimize the global frequency range by defining your goals only where they make sense. For instance to define a goal that meets your demand on rejection below 900 MHz, a goal ranging from 900 MHz to 850 MHz is equivalent to a goal ranging from 900 MHz to 100 MHz, given the response is well behaved below 900 MHz. If there were for instance a second goal reaching 1100 MHz the global range would become 1000 MHz in the first case and 250 MHz in the second. With 101 points that would translate to a frequency resolution of 10 MHz and 2.5 MHz respectively. Please don’t confuse the global goals range used during optimization with the current frequency range in the response plot. That is, you don’t have to zoom out to include every goal in the optimization.
The speed of the optimizer can also be improved by adjusting the allowed limits of maximum physical couplings, minimum external qualities and maximum resonator frequency deviation. The time saving comes from the fact that the optimizer is based on a quasi-newton algorithm that involves a line-search check. Basically it instantly discards every optimization-sample that is not compliant to these limitations and thus avoids performing heavy calculations that would have produced non-physical or at least non-wanted couplings and frequencies in the first place.
Keeping the number of matrix indices, or variables, taking part in the optimization low will reduce the run time. The optimizer can also perform a lot faster if couplings are marked as symmetrical since every pair of symmetrical couplings are treated as one variable. It is therefore recommended to use symmetry whenever appropriate.
Monte Carlo settings
Tolerance analysis using Monte Carlo is a great way to check for yield problems in addition to if and how tuning might be required. A space of coupling matrix samples that are randomly distributed and uniform are calculated. Each physical coupling, external quality and resonator frequency marked for Monte Carlo analysis is adjusted with an offset that can range from -X to +X where X is the tolerance times the nominal value. This set of random coupling matrices are plotted on top of each other to give the designer a sense of the expected sensitivity to manufacturing tolerances and tuning.
Back to the Couplings Designer documentation.