</table>
By default, [[EM.Tempo]]'s mesh generator tries to place grid points at the corners of each graphic object's bounding box, and also at any internal boundaries any object may have. For models with a large number of complex geometric objects, this could drive the typical mesh cell size toward the "Absolute Minimum Grid Spacing", and would result in a much denser mesh than is required. Since the Golf model has more than 2000 distinct graphic objects, we will turn off some of these adaptive mesh options. A mesh density of 18 cells per effective wavelength is chosen for this structure with the absolute minimum grid spacing parameter set equal to 0.75mm. The figures below show the Yee mesh of the overall whole vehicle structure as well as the portion of the roof in the proximity of the installed patch antenna. The overall mesh involves <b><u>220 million</u></b> cells.
<table>
</table>
The FDTD simulation of the vehicle structure was run on [https://aws.amazon.com/ Amazon Web Services]. For the purpose of this project, we logged into an Amazon instance via Remote Desktop Protocol (RDP) and used a c4.4xlarge instance running Windows Server 2012. This instance had 30 GB of RAM memory, and 16 virtual CPU cores. The CPU for this instance was an Intel Xeon E5-2666 v3 (Haswell) processor. The thread factor setting essentially tells the FDTD engine how many CPU threads to use during [[EM.Tempo]]'s time-marching loop. For a given system, some experimentation may be needed to determine the best number of threads to use. Eight thread factors were used for this simulation, with a total computation time of 285 minutes.
First, we consider the electric field distributions at four planes: two vertical E and H planes coincident with the principal YZ and ZX planes, respectively, and two horizontal planes, one on the reflector aperture and the other slightly above the horn aperture.
<table>
<tr>
<td>
[[Image:Roof field.png|thumb|left|640px| The dB-scale electric field distribution of the vehicle-antenna combination structure in the vertical ZX plane.]]
</td>
</tr>
</table>
The figure below shows the 3D far-field radiation pattern of the installed patch antenna on the vehicle roof. For this simulation, the far-field angular resolution was set to 2.5° along both azimuth and elevation directions.
The Mirage III CAD model As you can see from the figure, the value of the directivity has an approximate length not changed significantly from the previous case due to the presence of 15mthe horn feed. However, a wingspan the perfect circular symmetry of 8m, and an approximate height the pattern in the case of 4the short dipole feed is now all but gone.5m <table><tr><td>[[Image:Roof pattern. Expressed in freepng|thumb|left|640px| 3D far-space wavelengths at 850 MHzfield radiation pattern of the vehicle-antenna combination structure, with the approximate dimensions of patch antenna installed on the aircraft model are 42vehicle roof.5 λ]]<sub/td>0</subtr> x 22</table> The figures below show the 2D polar radiation patterns of the parabolic reflector with the pyramidal horn feed in the principal E and H planes.66 λ <subtable>0</subtr> x 12<td>[[Image:Roof yz cut.75 λpng|thumb|left|480px| 2D linear-scale polar radiation pattern of the roof-mounted patch antenna.]]<sub/td>0</subtr>. Thus, for the purposes of <tr><td>[[EMImage:Roof zx cut.Tempo]], we need to solve a region of about 12,279 cubic wavelengths. For problems of this size, a very large CPU memory is needed, and a highpng|thumb|left|480px| 2D linear-performance, multi-core CPU is desirable to reduce scale polar radiation pattern of the simulation timeroof-mounted patch antenna.]]</td></tr></table>
[https://aws.amazon.com/ Amazon Web Services] allows one to acquire high-performance compute instances on demand, and pay on a per-use basis. To be able to log into an Amazon instance via Remote Desktop Protocol (RDP), the [[EM.Cube]] license must allow terminal services. For the purpose of this project, we used a c4.4xlarge instance running Windows Server 2012. This instance has 30 GB of RAM memory, and 16 virtual CPU cores. The CPU for this instance is an Intel Xeon E5-2666 v3 (Haswell) processor.
Thread Factor: 8
The thread factor setting essentially tells the FDTD engine how many CPU threads to use during [[EM.Tempo]]'s time-marching loop. For a given system, some experimentation may be needed to determine the best number of threads to use. In many cases, using half of the available hardware concurrency works well. This comes from the fact that many modern processors often have two cores per memory port. In other words, for many problems, the FDTD solver cannot load and store data from CPU memory quickly enough to use all the available threads or hardware concurrency. The extra threads remain idle waiting for the data, and a performance hit is incurred due to the increased thread context switching. [[EM.Cube]] will attempt use a version of the FDTD engine optimized for use with Intel's AVX instruction set, which provides a significant performance boost. If AVX is unavailable, a less optimal version of the engine will be used alternatively.
[[Image:Roof field.png|thumb|left|400px|]]