Optimization Techniques
in
CST STUDIO SUITE
Vratislav Sokol, CST
[Link]
Agenda
Optimization algorithms in CST STUDIO SUITE
Classic Powell
Interpolated Quasi-Newton
Trust Region Framework
Nelder-Mead Simplex
Genetic Algorithm
Particle Swarm Optimization
General suggestions for optimizer setting
Examples
Waveguide corner
Dual-band matching circuit network
Planar filter tuning
Antenna array side lobes suppression
2
[Link]
Optimizer Window Overview
Solver Selection
Optimizer Choice
Automatically
choose parameter
boundaries
Parameter space
definition
Termination
criterion
Goal Function Overview (1)
An arbitrary number of goals
can be defined. The optimizer
will try to satisfy all goals.
You can choose to optimize
the sum of all goals or the
maximum of all goals
Goals are 0D, and can be derived from any 1D or 0D Result Template
Goal Function Overview (2)
A range of the
1D result can be
defined for goal
value calculation
The weight allows
you to give priority
to a goal over others
Possible operators are:
>, <, =, move min/max.
minimize and maximize
operators exist for Sparameter results
Optimizer - Goal Visualisation
Optimizer - Result Plots
also in DS
Local vs. Global Optimizers
Classic Powell
Interpolated Quasi Newton
Trust Region Framework
Nelder-Mead Simplex Algorithm
Particle Swarm Optimization
Genetic Algorithm
local
Initial parameters already
give a good estimate of the
optimum, parameter ranges
are small
global
Initial parameters give a
poor estimate of the
optimum, parameter
ranges are large
Example 1: Waveguide Corner
Classic Powell
Quasi Newton
Simplex (Nelder-Mead)
Genetic Algorithm
Particle Swarm
Trust Region Framework
Goal Minimize S11
9
Classic Powell
Optimization terminates
if two consecutive goal
values g1 and g2 yield
2( g1 g 2 )
g1 g 2
Accuracy
A local optimizer that robustly finds an optimum within the given
parameter bounds. Sometimes, many iterations are necessary when
closing in on the optimum. This algorithm is suitable for one-variable
problems.
10
Interpolated Quasi Newton
The optimizer allows a restart
of the algorithm within an
automatically chosen smaller
parameter range. This range
is determined by the previous
pass.
A Search algorithm for expensive problems: The parameter space is
sampled in each variable direction. EM simulations are only performed for
these discrete parameter space points. A model is created from these
evaluations and used for optimization. During the search, the model is
updated regularly by real evaluations.
11
Trust Region Framework
If a normalized variation of
the parameters becomes
smaller than this value, the
optimization terminates
A fast and accurate optimizer that converges robustly and finds an
optimum within the given parameter bounds using a low number of
evaluations. It is suitable for 3D EM optimization.
12
Trust Region Framework Algorithm (1)
Choose initial point x0
Create a local linear model around that point,
and define an initial trust region radius, an
area in which we think the model is good.
Repeat:
Go to the minimizer (predicted optimum) of
the model inside the trust-region
Verify: Does the error decrease?
If true, and if the model is very good, go
further until quality gets worse, take last
point as new center. Reduce trust region
radius and calculate new model
If just true, keep trust region radius and
calculate new model
If not true, reduce size of trust region.
13
Trust Region Framework Algorithm (2)
Choose initial point x0
Create a local linear model around that point,
and define an initial trust region radius, an
area in which we think the model is good.
Repeat:
Go to the minimizer (predicted optimum) of
the model inside the trust-region
Verify: Does the error decrease?
If true, and if the model is very good, go
further until quality gets worse, take last
point as new center. Reduce trust region
radius and calculate new model
If just true, keep trust region radius and
calculate new model
If not true, reduce size of trust region.
14
Trust Region Framework Algorithm (3)
Choose initial point x0
Create a local linear model around that point,
and define an initial trust region radius, an
area in which we think the model is good.
Repeat:
Go to the minimizer (predicted optimum) of
the model inside the trust-region
Verify: Does the error decrease?
If true, and if the model is very good, go
further until quality gets worse, take last
point as new center. Reduce trust region
radius and calculate new model
If just true, keep trust region radius and
calculate new model
If not true, reduce size of trust region.
15
Trust Region Framework Algorithm (4)
Choose initial point x0
Create a local linear model around that point,
and define an initial trust region radius, an
area in which we think the model is good.
Repeat:
Go to the minimizer (predicted optimum) of
the model inside the trust-region
Verify: Does the error decrease?
If true, and if the model is very good, go
further until quality gets worse, take last
point as new center. Reduce trust region
radius and calculate new model
If just true, keep trust region radius and
calculate new model
If not true, reduce size of trust region.
16
Trust Region Framework Algorithm (5)
Choose initial point x0
Create a local linear model around that point,
and define an initial trust region radius, an
area in which we think the model is good.
Repeat:
Go to the minimizer (predicted optimum) of
the model inside the trust-region
Verify: Does the error decrease?
If true, and if the model is very good, go
further until quality gets worse, take last
point as new center. Reduce trust region
radius and calculate new model
If just true, keep trust region radius and
calculate new model
If not true, reduce size of trust region.
17
Trust Region Framework Algorithm (6)
1
The algorithm will
be converged once the
trust region radius or
distance to the next
predicted optimum
becomes smaller than
the specified domain
accuracy.
0
18
Global Optimizer Overview
Nelder Mead
An optimizer for more
complex problem
domains with good
convergence behavior:
Uses relatively few
evaluations if the
problem has a low
number of parameters
(i.e., less than 5 ).
19
Particle Swarm
A global optimizer that
uses a higher number of
evaluations to explore
the search space, also
suited for larger
numbers of parameters
(hint: use distributed
computing).
Genetic Algorithm
A global optimizer that
uses a high number of
evaluations to explore
the search space, suited
for large numbers of
parameters or very
complex problem
domains (hint: use
distributed computing).
General Suggestions
1. Try to use a concise parameterization.
2. Try to keep the number of goal functions low.
3. Monitor parameter changes throughout optimization to gain
insight into convergence behavior.
4. Sometimes, re-formulating your goal function makes the
difference (e.g., min vs. move min).
5. You can use coarse parameter sweeps to determine good
initial values and boundaries, and to support the right choice
of optimization algorithm.
6. If possible, use face constraints together with sensitivities in
combination with the trust region optimizer.
20
Mobile Phone Antenna
Goal: Best impedance matching in bands 890-960 MHz and 1710-1880 MHz.
21
Optimisation in CST DS (1)
22
Optimisation in CST DS (2)
23
Trust Region Framework + Sensitivity
24
[Link]
TRF + Sensitivity: Results
25
[Link]
Post-Processing Optimisation
Radiation pattern of 8x1 antenna
array is constructed from the
farfield of one element by
applying so called array factor
using template based postprocessing (TBPP).
8x
26
In the second step the side lobe
level is minimized using pure
TBPP optimisation.
Optimisation of TBPP Steps
Optimise amplitudes and
phases of element excitations
as a post-processing step to
minimise vertical plane sidelobe levels.
vertical plane directivity
a1
a2
a3
a4
a5
a6
a7
original an=1 SLL = -13.7 dB
27
a8
Optimisation of Side Lobe Levels
Optimise amplitudes and
phases of element excitations
as a post-processing step to
minimise vertical plane sidelobe levels.
vertical plane directivity
a1
a2
a3
a4
a5
a6
a7
original an=1 SLL = -13.7 dB
optimised an SLL = -20 dB
28
a8
Optimisation of Side Lobe Levels
Optimisation
Method
3D
TBPP
Comparison
Time per 3D
simulation
Number of 3D
simulations
Time per
TBPP eval.
vertical plane directivity
a1
5 min
5 min
a2
40
a3
a4
30 sec 30 sec
a5
Optimisation
steps
40
40
Total
simulation
time
200
min
60
min
a6
a7
a8
original an=1 SLL = -13.7 dB
29
optimised an SLL = -20 dB
Summary
1. CST STUDIO SUITE 2011 offers a complete portfolio
of optimization methods for various application.
2. A new Trust Region Framework algorithm is very
efficient tool for a direct 3D EM optimization
especially in conjunction with the sensitivity
analysis.
3. New visualization of goals and parameter values
4. Post-processing optimization without a need of any
EM or circuit solver
5. A new Minimax goal function definition is now
available.
30
