|
WIPL-D Pro 8.0
WIPL-D Pro 8.0 provides significant benefits in simulation speed on
multi-core computers, modeling productivity and convergence control.
The main new features:
Improvements of old features:
-
Automatic switch from in-core to
out-of-core solver with automatic choice of optimum data block size,
-
Check of memory resources availability
for all types of matrix solution,
-
Increased resolution of near-field
results (precise calculation of very low fields inside cavities),
-
Improved handling of objects,
manipulations and/or RV-entities in complex scenarios,
-
Improved far-field approximation in MLFMM
(improving accuracy of results for the same distance threshold)
-
Automatic removal and replacement of a
domain with another, selected domain,
-
Saving all units on project level
(previously saved on application level, except for length units).
|
 |
Parallelization of All
Simulation Phases
Prior to release of 8.0 the in-core and
out-of-core direct solvers were parallelized. With 8.0, all the other
solvers and stages of simulation are parallelized using OpenMP in order
to fully exploit the multi-core CPU performance. The speed-up on a
quad-core CPU is:
-
Matrix fill-in stage - about
3 times,
-
Iterative solver - close to
4 times,
|
|
The new version provides significant
speed-up for many examples having rather low number of unknowns but a
large number of mesh elements. This is the case with all structures
containing lots of electrically small parts. Examples requiring
intensive far field and near field computations benefit from almost 4
times faster post-processing.
|
|
|
|
The model of the compact
multi-band antenna according to [1], was drawn and meshed in AW
Modeler before it was simulated in the frequency range from 0.5
GHz to 3 GHz. Simulated S-parameters match excellently
with the measured and simulated from [1]. The new version is
3.3 times faster than 7.0.
|
Number of unknowns |
Time [sec] –v7.0 |
Time [sec] – v8.0 |
|
4,273 |
121 |
37 |
[1] A. A. Eldek, “Analysis and
Design of a Compact Multi-band Antenna for Wireless
Communications Applications”, Microwave Journal, vol. 51, no. 5,
pp. 218-230, May 2008. |
A microstrip patch antenna has
been placed on a roof of a generic car model. The antenna was
designed for the working frequency of 3 GHz, at which the entire
model, with smart reduction applied, requires 17,548 unknowns
with one symmetry plane applied and PEC plane as model of
ground.
|
Number of unknowns |
Time [min] – v7.0 |
Time [min] – v8.0 |
|
17,548 |
78 |
21 |
Simulation was performed with the
direct solver, on a Core2 Quad CPU.
The new version is 3.7
times faster. |
Structure Tree
|
Structure Tree is a new tool in
WIPL-D Pro graphical user interface which facilitates modeling
operations.
-
The tree displays all the
model parts in a tree-like control, which enables easy
navigation through the model.
-
Selection of a model part is
possible from the tree, which also highlights the part in
Preview window.
-
The history of model creation
and relations between objects, manipulations, wires and
plates in the model is inherently present in the tree, thus
facilitating changes in the model.
-
Structure Tree offers
commands for viewing separate history trees for each object
or manipulation in the model and direct access to commands
for creating new model parts or for modifying the existing
ones.
Structure Tree does not replace
the GUI commands from WIPL-D Pro 7.0, rather it offers an
alternative approach. Therefore, it is possible to model by not
using the tree at all. However we consider that using the
Structure Tree brings a significant productivity benefit. |
 |
Optimized Out-of-core
Solver
|
The out-of-core solver is often
used for the largest examples, when there is not enough RAM to
solve them in-core. Having in mind that these simulations often
last for hours, special attention was devoted to optimizing the
out-of-core algorithm in this release.
The changes include:
-
Matrix fill-in stage on a
single-core CPU accelerated about 40%,
-
Inversion of diagonal matrix
blocks is now also running in parallel on multi-core CPUs
with speed-up of about 3 times on a quad-core CPU,
-
Larger data blocks can now be
used in order to maximize the speed of the algorithm.
Overall, combined with the effect
of multi-core parallelization, simulation of large
examples with out‑of‑core solver is accelerated roughly 2 times
on a quad-core computer. |
|
|
Number of
unknowns |
Time [h]
v7.0 |
Time [h]
v8.0 |
|
78,300 |
18.9 |
7.8 |
|
Frigate warship model, 117 meters
long and up to 12.6 meters wide, was placed above a PEC plane
that models the influence of the sea. A monopole antenna was
placed on top of a 24 meters high communication tower. The
simulation frequency was 240 MHz, which made the frigate about
94 wavelengths long. Rigorous MoM requires 78,300 unknowns,
although by using adaptive reduction of expansion orders this
can be halved without loss of accuracy.
Simulation of the model with
78,300 unknowns was performed with the out-of-core solver. The
new version is 2.4 times faster than version 7.
|
New Tools to
Control Accuracy /
Convergence
The reference frequency influences mesh
subdivisions and calculation of the electrical size of mesh elements
during simulation. The reference frequency is equal to the highest
simulation frequency, by default, but can now be set independently if
results verification is needed.
WIPL-D Pro determines the order of
current approximation on each mesh element according to its electrical
size. In the new version, by setting the maximum element electrical
size in the model, the user controls the maximum order of
approximation of currents, which is particularly useful for the MLFMM
solver.
Three sets of higher order basis
functions have been incorporated into WIPL-D Pro numerical kernel. With
the introduction of parallel matrix fill-in stage, basis functions were
modified in order to increase stability of results in single precision
when higher orders of approximation were used. Although the selection of
basis functions is automatic by default, the users can now alter the
automatic settings if additional convergence tests are required.
|
