META2010META2010
Application of metaheuristics through MATLAB optimization
toolboxes for the design of coupled resonator filters
José-Ceferino OrtegaDomingo Giménez
University of Murcia
Alejandro Álvarez-MelcónFernando D. Quesada
Polytechnic University of Cartagena
Content
Introduction
Synthesis of coupled resonator filters
MATLAB optimization toolboxes
Experimental results
Conclusions and future research
Design problems in telecommunications•Optimization of design parameters
Design of coupled resonator filters•Used in microwave-based communications•Several phases:
•Phase 1: obtain couplings matrix (design technology)•Phase 2: obtain geometry (physical design)
Hybridize local and global search methods Environment: MATLAB
Introduction
Content
Introduction
Synthesis of coupled resonator filters
MATLAB optimization toolboxes
Experimental results
Conclusions and future research
Synthesis of filters (I) Analysis of the problem of synthesis
of coupled resonators filters Filters based on coupled microwave
resonators Technological design: couplings matrix Characteristics of the filters:
Transfer function Topology (Kite, Transversal, 2-Trisection
ord. 3) Number of design parameters (8 or 9) Range of values (from -5 to 5)
Synthesis of filters (II)2 Trisection ord. 3, zeros
-5 and -3 Kite, zeros -3 and 3
-10 -8 -6 -4 -2 0 2 4 6 8 10-100
-90
-80
-70
-60
-50
-40
-30
-20
-10
0
In s11
In s21
-10 -8 -6 -4 -2 0 2 4 6 8 10-70
-60
-50
-40
-30
-20
-10
0
In s11
In s21
Synthesis of filters (III)Kite, Kite, fitnessfitness 10 10-13-13 Kite, Kite, fitness fitness 1010-5-5
Kite, Kite, fitnessfitness 10 10-1-1
Content
Introduction
Synthesis of coupled resonator filters
MATLAB optimization toolboxes
Experimental results
Conclusions and future research
MATLAB optimization toolboxes
MATLAB Optimization Toolboxes Optimization Toolbox
fmincon Genetic Algorithm and Direct Search
Toolbox Direct search (patternsearch)
Genetic algorithms (ga) Simulated annealing (simulannealbnd)
and Scatter Search
Content
Introduction
Synthesis of coupled resonator filters
MATLAB optimization toolboxes
Experimental results
Conclusions and future research
Experimental results: fmincon
fmincon Part of the MATLAB Optimization Toolbox Local search Parameters to study:
LargeScale Algorithm
Experimental results: patternsearch
patternsearch (Direct search) MATLAB Direct Search and Genetic
Algorithm Toolbox Local search Parameters to study:
InitialMeshSize MeshContraction MeshExpansion ScaleMesh PollMethod CompletePoll PollingOrder SearchMethod CompleteSearch
Experimental results: genetic algorithm
ga (Genetic algorithms) MATLAB Direct Search and Genetic
Algorithm Toolbox Global search Parameters to study:
PopulationSize and Generations EliteCount and CrossoverFraction FitnessScalingFcn and SelectionFcn CrossoverFcn and MutationFcn CreationFcn and HybridFcn
Experimental results: genetic algorithmpersonalized functions
CreationFnc CrossoverFnc
MutationFnc HybridFnc
Experimental results: genetic algorithmpersonalized functions exec. time–
HybridFnc
CreationFnc
MutationFnc
Experimental results: simulated annealing
simulannealbnd (Simulated annealing) MATLAB Direct Search and Genetic
Algorithm Toolbox Local search Parameters to study:
AnnealingFcn InitialTemperature ReannealInterval TemperatureFcn HybridFcn and HybridInterval
Experimental results: simulated annealingf i tness
AnnealingFnc TemperatureFnc
HybridFnc & HybridInterval
Content
Introduction
Synthesis of coupled resonator filters
MATLAB optimization toolboxes
Experimental results
Conclusions and future research
Conclusions
Evaluated the application to the design of coupled resonator filters of available tools in the toolboxes of MATLAB
Local and global search methods hybridation, with Genetic algorithms and Scatter Search
The best: ga (Genetic algorithm) personalized
Future research Application to the physical design (2nd
phase), with more computational cost. The 1st phase simplifies the physical design.
Application of other metaheuristics and implementation in MATLAB.
Study of relation between technological and physical design, to divide the physical design in smaller problems.
Application of parallelism, specially in the 2nd phase: parallel metaheuristics and parallelism in the computation of the fitness function (matricial computation).