Download PowerPoint **

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts

Microevolution wikipedia , lookup

Gene expression programming wikipedia , lookup

Transcript
Implementation of a genetic algorithm-based
decision making framework for opportunistic
radio
Authors: Soamsiri Chantaraskul, Klaus Moessner
Source: IET Commun., Vol.4, No.5, 2010, pp.495 - 506
Presenter:Ya-Ping Hu
Date: 2011/12/23
Outline
 Introduction
 Decision making framework for the OR
 GA approach for the decision making engine
 Development of demonstration platform for the decision
making engine
 Test cases and engine performance observation
 Conclusion
2
2011/12/23
Outline
 Introduction
 Decision making framework for the OR
 GA approach for the decision making engine
 Development of demonstration platform for the decision
making engine
 Test cases and engine performance observation
 Conclusion
3
2011/12/23
Introduction
 Innovation of the cognitive radio (CR) concept with the
aim to enhance SDRs by exploiting the environmental
awareness and intelligent adaptation
 The opportunistic radio (OR) is proposed, which only
confines its knowledge to the spectrum awareness
4
2011/12/23
Introduction (cont.)
 Two major aspects of the OR technology
 Determine the best opportunity
 Define an optimum usage of such opportunity
 Soft-computing methods used in the proposed
framework
 Rule-based reasoning and case-based reasoning
 Genetic algorithm
5
2011/12/23
Outline
 Introduction
 Decision making framework for the OR
 GA approach for the decision making engine
 Development of demonstration platform for the decision
making engine
 Test cases and engine performance observation
 Conclusion
6
2011/12/23
Decision making framework for OR
 Key components and their relationships
7
2011/12/23
Decision making engine
 Filtering engine
 Take into account the context information, the related policies
and profiles
 Discards ineffective solutions
 Filters the available solutions
 Provides a smaller range of possible set of configurations
 Reasoning engine
 The operational configuration is taken into account
 The preferred solution is obtained
8
2011/12/23
Filtering mechanism
 Concerned value (CV) algorithm
 Collaborative filtering or social filtering
 Case-based reasoning (CBR)
9
2011/12/23
Reasoning engine
 The operational configuration is taken into account and
the preferred solution is obtained
 The approach is based on the GA
10
2011/12/23
Outline
 Introduction
 Decision making framework for the OR
 GA approach for the decision making engine
 Development of demonstration platform for the decision
making engine
 Test cases and engine performance observation
 Conclusion
11
2011/12/23
Genetic algorithm (GA)
 One of the artificial intelligent techniques
 Find an optimal radio solution in response to the
dynamic spectrum usage environment
 Solve the multi-objective optimization problem
 Define each of the major components
 Encoding method
 Recombination operators
 Fitness evaluation and selection
12
2011/12/23
Algorithms for the reasoning
engine
 Encoding the radio solution
 Fitness function
 Elitism
 Selection process and GA operations
 Termination of algorithm
13
2011/12/23
GA-based reasoning engine
14
2011/12/23
Encoding the radio solution
 GA works with the chromosomes, which are the
representations of solutions
 Map the radio adaption into a chromosome
 An initial population or a set of chromosomes is
randomly generated from the operational context
 Use binary encoding technique
15
2011/12/23
OR chromosome structure
16
2011/12/23
Fitness function
 Calculate the fitness for each chromosome
 The fittest chromosome will have the highest fitness
value
 Optimization problem
 Single-objective optimization
 Multi-objective optimization
17
2011/12/23
Fitness assignment mechanisms
 Weighted-sum
 Vector evaluation
 Pareto-ranking
 Rank-based
 Compromise
 Goal programming
18
2011/12/23
Objective functions and the
formulas
19
2011/12/23
Fitness formula
𝑘
𝑖=1 𝜔𝑖
±𝑓𝑖 ′ (𝑥)
𝐹=
∙𝑒
′
 𝑓𝑖 (𝑥) is the normalized objective function𝑓𝑖 (𝑥)
 𝜔 = (𝜔1 , 𝜔2 , … , 𝜔𝑘 ) is non-negative weight vector
 𝜔𝑖 = 1
 The relationship between 𝑥 and the resulting fitness is used to
identify plus or minus sign of 𝑓𝑖 ′ (𝑥)
20
2011/12/23
Elitism
 Retains the best chromosome(s) at each generation and
carries over to the next population
 Guarantee that the chromosome(s) with the best fitness
value(s) will not be lost during the selection process
 In this work, each new generation carries two
chromosomes from the previous generation to the next
one
21
2011/12/23
Selection process
 Produce the offspring from the selected parents
 Existing methods for the parental selection
 Roulette wheel sampling
 Boltzmann selection
 Rank selection
 Tournament selection
 Steady-state selection
22
2011/12/23
GA operations
 Crossover
 Two new chromosomes are created by swapping section(s) of
genes from parents according to the position(s) determined by
the crossover points
 Mutation
 The random modification is performed to the randomly
selected gene(s) to introduce new candidate(s) to the
population
23
2011/12/23
GA operations (cont.)
 The important parameters
 Crossover rate
 Mutation rate
 In this work
 Crossover rate is 0.6
 Mutation rate is 0.001
24
2011/12/23
Termination of algorithm
 The search continues until the termination criterion is
met
 Existing criteria
 Maximum number of generations
 Stability of the fitness of best individual
 Convergence of population
 Online and off-line performances
25
2011/12/23
Termination of algorithm (cont.)
 Convergence of population
 A gene is said to have converged when 95% of the population
share the same value. The population is said to have converged
when all of the genes have converged
 The value of genes is expressed as follows:
1
𝑛
𝐶 𝑖 = max[
𝑛
𝑘=1 𝑏(𝑘, 𝑖, 1) ,
𝑛
𝑘=1 𝑏(𝑘, 𝑖, 0)]
1, if the 𝑖th gene of the
𝑏 𝑘, 𝑖, 𝑣 =
𝑘th individual is 𝑣
0, else
𝐶1 =
26
1
𝐿
𝐿
𝑖=1 𝐶(𝑖)
2011/12/23
System evaluation via MatLab
simulations
 Grefenstette studied the variation of six GA parameters
 Population size
 Crossover rate
 Mutation rate
 Generation gap
 Scaling window
 Selection strategy
27
2011/12/23
System evaluation via MatLab
simulations (cont.)
 Settings for GA parameters
 Population size = 50
 Crossover rate = 0.6
 Mutation rate = 0.001
 Selection strategy = Elitist strategy
 Number of generation = 1000
28
2011/12/23
Convergence and fitness plots

Average convergence and
fitness
29
vs.
Example plot for fitness and
convergence from a single test
2011/12/23
Outline
 Introduction
 Decision making framework for the OR
 GA approach for the decision making engine
 Development of demonstration platform for the decision
making engine
 Test cases and engine performance observation
 Conclusion
30
2011/12/23
Implementation platform for the OR
 Hardware
 USRP motherboard
 Four ADC and four DAC
 Fit with different daughterboards to cover different frequency ranges
 RFX2400 transceiver daughterboard
 Cover the frequency range from 2.3 to 2.9 GHz
 Support the maximum transmit power of 50mW
 Quad Patch 2.4GHz antenna
31
2011/12/23
Implementation platform for the OR
(cont.)
 Software
 GNU Radio
 An open-source software
 Interface between the OR decision making engine and the RF frontend
 Use a combination of Python and C++ programming models
 Provides a library of signal processing blocks
32
2011/12/23
Implementation platform for the OR
(cont.)
33
2011/12/23
Outline
 Introduction
 Decision making framework for the OR
 GA approach for the decision making engine
 Development of demonstration platform for the decision
making engine
 Test cases and engine performance observation
 Conclusion
34
2011/12/23
RF scanner in the 2.4 GHz ISM
band
 The sensing information is recorded and can be
processed to detect the spectrum opportunity
In an uncontrolled
environment
35
vs.
The other RF frontend
transmits data at a center
frequency of 2.442 GHz
2011/12/23
MatLab screen shot as decision being
made by the OR engine
36
2011/12/23
Power spectrum observed by spectrum
analyzer
 The spectrum analyzer is used to capture the power
spectrum of the entire 2.4 GHz band
Other device starts data
transmission using the channel vs.
currently used by OR terminal
37
OR terminal switches to
reallocated channel
2011/12/23
Outline
 Introduction
 Decision making framework for the OR
 GA approach for the decision making engine
 Development of demonstration platform for the decision
making engine
 Test cases and engine performance observation
 Conclusion
38
2011/12/23
Conclusion
 Present the system architecture and algorithms of the
proposed OR decision making framework
 The GA-based reasoning engine is developed under
MatLab environment, where the system stability is
observed through the simulation
 The benefit of the proposed GA-based approach is in its
capacity to cope with multiple objectives simultaneously
39
2011/12/23
Thanks for your listening