For a description of the general AMR process, see “Using Adaptive Mesh Refinement (AMR)”. RF3p supports two AMR modes: Ports Only, in which the simulation is confined to extracting modes at wave ports, and Full, in which the simulation is performed on the total volume of the problem. Independent control of error tolerance and number of iterations for each mode is provided. The ports only AMR process is only required for wave ports; lumped ports are defined according to a simple analytical model and don't require a separate eigensolution. For wave ports, the field solution on the port is the source for the volumetric simulation; it is therefore important to ensure the ports only AMR proceeds appropriately. The same methods are used to ensure performance of the ports only AMR and the full solve AMR, as follows.
When a simulation spans multiple frequencies, the most fastidious AMR process would compute the solution for all frequencies in each AMR step. Usually, however, this is both unacceptably cumbersome and wholly unnecessary. In general, the mesh obtained from a single-frequency AMR process will yield accurate results over a spectrum of frequencies. In the simulation properties dialog box under the AMR Sequence tab, you may choose the AMR frequency by choosing a value for Frequency Modifier, which allows selection of predefined options. Most of the time the default selection, Mid, is sufficient, which corresponds to a frequency in the middle of the band. If widely varying or discontinuous behavior is expected over the range of simulation frequencies for a particular system, other choices may be more appropriate. In addition, you can choose a custom frequency from the set of frequencies at which your problem is to be solved. As an example, the simulation of a band pass filter may require use of a custom frequency, if the High, Low, and Mid frequencies are not in the band pass region. If All is chosen, the AMR sequence will usually take significantly longer to converge.
In the ports only AMR, the convergence criterion is based upon the characteristic impedance Zc for each of the modes.
In the full solve AMR, you can choose to base convergence on the maximum change in the magnitude of the scattering matrix S alone, or you can choose to base convergence on the maximum change in the magnitude and the phase of S (across all AMR frequencies and all terms in S). In the latter case you can set a cutoff magnitude below which the phase change is ignored.
At each new AMR step, RF3p calculates an estimate of the memory needed for the step. If it determines that further iterations will cause you to run out of memory, it aborts this phase with an appropriate message and moves to the final solve, described in “Running”. You should get results if this occurs, but the results are not fully converged and they should be used with caution.