10.5 Forecasting the computing time
After the simulations are started we can watch the iteration counter to assess how many FDTD iterations are made in one second. The assessment of the needed number of iterations would give us a precise approximation of the total expected computing time. In general it is rather difficult to give clear and general indications about the number of iterations required for electromagnetic simulation. However good understanding of the physics of the simulated process and experience with a particular type of simulation usually allows quite precise estimation of the expected simulation time. Let us formulate here some general remarks about the required number of iterations:
The FDTD method reflects the natural process of wave simulation with a pulse excitation. We can watch this process and note when most of the energy injected into the structure has been dissipated inside the structure (when it is lossy) or at the ports. This is the time when we can consider the S-parameters. Note that the duration of this process depends on the loaded Q-factor of the considered structure. Typically the user does not directly control the time step of the FDTD process (which is set by the software to comply with the CFL stability conditions). However he can read the time step from the Simulation Info tab of Simulator Log window of the QW-Simulator. Knowing the physical reality of the considered project we can estimate the physical time needed to obtain the steady state in the structure. Dividing that time by dt brings the estimation for the needed number of FDTD iterations.
If we are not sure about the number of iterations needed it is better to set as a parameter a very high number. During the simulation process we will be able to watch how the S-parameters versus frequency change with the number of iterations. When these changes become negligible we can terminate the simulation. It is also worth noting that the frequency domain characteristics obtained with insufficient number of iterations have typical ripples, which are due to truncation of the Fourier transformation process. These ripples vanish after more iterations.
A quantitative measure of convergence is available in the case of shielded lossless circuits. This measure is power balance, which QW-Simulator automatically produces by integrating the power entering and leaving through all the ports. Refer to the User Guide 3D: S‑parameter extraction problems or User Guide V2D: A circular waveguide discontinuity (for QW-V2D) for illustration of such an approach.
Another quantitative measure is available in the case of lossless antennas, excited by a single feed. This measure is radiation efficiency, which should approach 100% in steady state.
In a particular circuit the number of iterations required is inversely proportional to the step in time dt. On the other hand dt is enforced by the stability condition of the FDTD process which makes it proportional to the smallest cell size in the circuit. Note that in the Mesh Parameters window the user defines only the maximum cell size along the particular coordinate. Much smaller cell size can be produced during the meshing procedure if two mesh snapping planes are placed very close to each other. That is why the user should watch the smallest cell size (in the Mesh/Splanes Info dialogue) and take it into account when deciding about the number of iterations for S-parameters calculations.