Transcript ANNs - Fake Engineer

APPLICATIONS OF ANN IN MICROWAVE
ENGINEERING
www.fakengineer.com
Introduction

ANNs are neuroscience -inspired computational tools.
Learn from experience/examples (training) & not the example
itself.

Generalize automatically as a results of their structure (not by
using human intelligence embedded in the form of ad hoc
computer programs).

Used extensively for visual pattern recognition, speech
understanding, and more recently, for modeling and simulation of
complex processes.

Recently it has been applied to different branches of Microwave
Engineering

www.fakengineer.com
When To Apply ANN



When the problem is poorly understood
When observations are difficult to carry out using noisy or
incomplete data
When problem is complex, particularly while dealing with
nonlinear systems
www.fakengineer.com
Feedforward Neural Model
Output
lines
Hidden
layer
Input
lines
www.fakengineer.com
Topics Covered


Smart antennae modeling
Demand node concept
1.
2.
3.
Initialization & selection
Adaptation
Optimization
www.fakengineer.com
Smart Antenna Modeling
•A smart antenna consists of an antenna array combined with
signal processing in both space and time.
•These systems of antennas include a large number of techniques
that attempt to enhance the received signal, suppress all
interfering signals, and increase capacity, in general.
www.fakengineer.com
ANN Model for Resonant Frequency
Rectangular Patch Antenna
www.fakengineer.com
Training/Network Parameters






Network size:
Learning Rate:
Momentum:
Time Step for integration:
Training Time:
No. of Epochs:
5  40  1
0.08
0.205
5  10-10
6.4 min.
15000
www.fakengineer.com
Bandwidth of Patch Antenna
Rectangular Patch Antenna
www.fakengineer.com
Rectangular Patch Antenna
Algorithm’s used
• Back Propagation
• Delta – Bar – Delta
(DBD)
• Extended DBD (EDBD)
• Quick Propagation
Other Details
•ANN structure:
3481
•Max. no. of iterations:
5,00,000
•Tolerance (RMS Error):
0.015
www.fakengineer.com
Network Parameters
BP Parameters
• Learning Coefficients:
– 0.3 for the 1st hidden
layer
– 0.25 for the 2nd hidden
layer
– 0.15 for the output layer
• momentum coefficient : 0.4
DBD Parameters
• k = 0.01,  = 0.5,  = 0.7, a
= 0.2
• Momentum coefficient = 0.4
• The sequential and/or
random training procedure
follows
EDBD Parameters
• k = 0.095, k = 0.01,
gm = 0.0, g = 0.0
• m = 0.01,  = 0.1,
 = 0.7, l = 0.2,
• The sequential and/or random
training procedure follows
QP Parameters
•  = 0.0001
• a = 0.1
•  = 1.0
• m = 2.0
www.fakengineer.com
Demand Node Concept
Demand Node Concept
www.fakengineer.com
Input
Geographical map
Morphology model
Step
Output
Radio network
definition
Estimated
tx location
Propagation analysis
Coverage
Land use categories
interference
distance
Frequency allocation
Freq plan
Stochastic channel
characteristics
Radio network analysis
Network performance
Mobile network
www.fakengineer.com
Initialization
&
Selection
Start
•Distribute sensory neurons.
•Place transmitting stations
•Determine initial temperature.
Determine supplying areas.
Random selection of a
Sensory neuron
N
N
N
No supply?
Multiply supplied?
No.of selection
Y
Y
Values=preset
Change position for
Change position for
Val.?
Y
attraction
repulsion or
Or increasing power.
Decreasing power.
www.fakengineer.com
Adaptation
E1=Energy of current system
State z1
Determine transmitting
Station tworst
Displace
T
Y
Change position
Determine supplying areas
www.fakengineer.com
N
Change
Power
Optimization
E2=Energy of current
System state z2
E1—e2<0 ?
N
Choose random
Number r
P:=prob(znew=zp)
Y
N
P<r ?
Y
Regenerate state
Z1
N
Steady state
System ?
Reduce temperature
Y
End
www.fakengineer.com
Displacement:Case Of Attraction
D1 D2 D3 D4
D5
D6
Base station
Area of coverage
www.fakengineer.com
Sensory neuron
Displacement:Case Of Repulsion
Base station
locations
BEFORE
Sensory
neurons
Borders of
supplying areas.
www.fakengineer.com
AFTER
Power Enhancement
Sensory neurons.
Base station
locations
BEFORE
Borders of the
supplying areas.
www.fakengineer.com
AFTER
Power Decrement
Borders of the supplying areas
Sensory neurons
Base station
After
Before
www.fakengineer.com
Emerging Trends / Future Applications




To find the optimized compact structures for low-profile
antennas
Applications in reconfigurable antennas/arrays
Applications in fractal antennas
To increase the efficiency of numerical algorithms used in
antenna analysis like MoM, FDTD, FEM etc.
www.fakengineer.com
Conclusion
Neural networks mimics brain’s problem solving process
& this has been the motivating factor for the use of ANN
where

huge amount of data is involved.

the sources vary.

decision making is critical.

environment is complex.
www.fakengineer.com
REFERENCES
[1]Haykin, S., 1999.Neural Networks A Comprehensive
Foundation, 2nd edition, Pearson Education.
[2]Freeman James A. & Skapura David M., Neural Networks,
Pearson Education.
[3]Yuhas, Ben & Ansari Nerman. Neural Networks in
Telecommunications.
[4]B.Yegnanarayana. 1999.Artificial Neural Networks. Prentice
Hall of India.
[5]G.A. Carpenter and S.Grossberg, ‘The ART of adaptive pattern
recognition by a self-organization neural network’, IEEE
Computer, vol. 21, pp. 77-88, 1988.
[6]N.K. Bose and P.Liang, Neural Network Fundamentals with
Graphs, Algorithms and Applications,McGraw-Hill,Int.
www.fakengineer.com
Thank You
www.fakengineer.com