Analysis of Signals and Characterization of Systems Chih-Ping Lin Geotechnical Engineering Group National Chiao Tung University Outline 1. Signals and Systems 2. Time Domain Analysis 3. Frequency Domain Analysis of Signals 4. Frequency Domain Analysis of Systems 5. Discrete Inverse Theory
34
Embed
Analysis of Signals and Characterization of Systemsgeo.cv.nctu.edu.tw/EngGeo/download/2-AnalysisOfSignalsAnd... · 4.Frequency Domain Analysis of Systems 5.Discrete Inverse ... •Continuous-time
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Analysis of Signals and Characterization of Systems
Chih-Ping LinGeotechnical Engineering GroupNational Chiao Tung University
Outline
1. Signals and Systems
2. Time Domain Analysis
3. Frequency Domain Analysis of Signals
4. Frequency Domain Analysis of Systems
5. Discrete Inverse Theory
Signals and Systems
1
2
3
4
5
Signals and Systems
1
2
3
4
5
Classification of Systems
Lumped vs. distributed parameter Causal vs. non-causal Linear vs. non-linear Time-invariant vs. time-variant
1
2
3
4
5
Combination of Systems
1
2
3
4
5
Combination of Systems
1
2
3
4
5
Classification of Signals
Multi-channel and multi-dimensional Continuous time vs. discrete time Continuous-valued vs. discrete-valued Deterministic vs. random Periodic vs. aperiodic
1
2
3
4
5
Elementary Signals
1
2
3
4
5
Decomposition Methods
1
2
3
4
5
Applications
Computation of the output response
Identification of similarities
Noise control
1
2
3
4
5
Time Domain Analysis
Convolution: Computing the output signal (in a LTI system)
Cross-Correlation: Identifying similarities
Noise control in time domain
1
2
3
4
5
Convolution
Decomposition in time domain
Convolution sum
k
knkxnx )()()(
kk
k
knhkxknTkx
knkxTnxTny
),()()()(
)()()()(
)(),( knhknh
)(*)()()()( nhnxknhkxnyk
Time invariant
1
2
3
4
5
Properties of Convolution
1
2
3
4
5
Graphical Computation of Convolution
1
2
3
4
5
Convolution sum in matrix form
1
2
3
4
5
Example of convolution sum
1
2
3
4
5
Determination of Impulse Response
Direct solution of the difference equation
Estimated experimentally by excite the system with a signal of short duration
1
2
3
4
5
Cross-Correlation
1
2
3
4
5
n
xy mnynxmr )()()(
n
xy nymnxmr )()()(
)(*)()()()( mymxmnynxmrn
xy
Definition:or
Computation:
)()( mrmr xyyx Properties:
Cross-Correlation
1
2
3
4
5
Cross-CorrelationTypical example
1
2
3
4
5
Application of Cross-Correlation
Identifying similarities Determination of travel time Identifying replicas in noisy signals Biases: Lossy media, dispersive and
multiple paths
1
2
3
4
5
Application of Cross-Correlation
1
2
3
4
5
Noise control in time domain
StackingIf xi's are independent and identical (i.i.d) Gaussian with mean and variance 2 (i.e. N(, 2) ) and y= mean(x), then y = N(, 2/N).
Smoothing with moving average
1
2
3
4
5
2/)1(
2/)1(
)()()(M
Mk
kixkwiy 1)(kwwhere
)(*)()( nwnxny
Noise control in time domainStacking
1
2
3
4
5
Noise control in time domainMoving average
1
2
3
4
5
Frequency Domain Analysis of Signals
1
2
3
4
5
The concept of frequency in continuous and discrete time
3
4
5
1
2 tFtAtAtxa ),2cos()cos()(
•Continuous-time sinusoidal signals
a. For every F, xa is periodicb. Xa distinct for distinct Fc. Rate of oscillation increases with F
Does the sinusoid offer any advantage in the study of system?
Impulse Inpulse response
Sinusoid Frequency Response
System
System
The sinusoid is an eigenfunction analogous to eigenvector in matrix form•If x is an eigenvector, then y = x•If x is a sinusoid, then y=H() xAnalytical and experimental determination of H(w)Are made easier because of the eigenfunction property
Convolution
3
4
5
1
2
Response of a LTI systemto periodic signals
3
4
5
1
2
Response of a LTI systemto periodic signals
3
4
5
1
2
Response of a LTI systemto aperiodic signals
3
4
5
1
2
t
x(t)
t
x(t)
Time Periodof Interest
PaddingZeros
Tr
2Tr
Response of a LTI systemto aperiodic signals
3
4
5
1
2
X(nt)
t0 T=1/ft
f0 f
X(nf)
TimeWindow
FrequencyWindow
F=1/t
tfF
12 max
ftT
12 max
Cross-correlation and cross-spectral density
3
4
5
1
2)()()(
)}(*)({))({
fYfXfC
nynxDTFTnrDTFT
xy
xy
)()()( fXfXfCxx
)()()()()()()()( fCfHfHfXfXfYfXfC xxyx
)()()()()()()()()(2
______________
fCfHfXfHfXfHfYfYfC xxyy
Cross-spectral density
Auto-spectral density (power spectral density)
Important relations:
Noise control in frequency domain
3
4
5
1
2)()()(' kWkXkX
Filters
Filters
3
4
5
1
2
Determination of frequency response
3
4
5
1
2
)(
)()(
)(
)()()()(
kX
kRkH
kX
kRkXkHkH noisy
)()()(
)()()]())()([(
)()]()()([
)()()(
kCkHkC
kXkRkHkXkX
kXkRkHkX
kXkZkCC
xrxx
xz
)(
)()(
kC
kCkH
xx
xz
)(
)()(
kX
kYkH
Measure of Noise
3
4
5
1
2Coherence:
)(
)()(2
kC
kCk
ZZ
YY
)()(
)()(
22
kCkC
kCk
ZZXX
XZ
)(
)()(
kC
kCkNSR
YY
NN
)(
)(1)(
2
2
k
kkNSR
Noise-to-signal ration (N/S):
Discrete Inverse Theory
3
4
5
1
2
ModelParameters
mMODEL g(.)
DataMeasurement
d
Methods of solution
Exact inversion Neural network Inversion based on a forward model
3
4
5
1
2
Neural network
3
4
5
1
21
2
3
4
1
3
2
1
2
3
Input layer Hidden layer Output layer
Inversion based on a forward model
3
4
5
1
2Method of Solution
Exact InversionInversion Based on a
Forward ModelNeural Network
StatisticalApproach
Linear AlgebraicApproach
ClassicalInterpretation
BayesianInterpretation
Least Squares Solution If the problem y=hx is over-determined and inconsistent, there
is no unique solution, and the estimate of the solution x will be a compromise.
One alternative is to minimize the L2-norm of the vector of residuals e=y-hx. The objective function to be minimized:
The solution is found by setting the derivative of Γw.r.t. x equal to zero. The solution is found as
hxhxyhxhxyyy
hxyhxyTTTTTT
T
)()(
yhhhx TTest
1)(
Inversion based on Optimization
Matlab function fmin() or fminsearch() Examples
2cxbxay
)/(1000 bxay
Bayesian Inversion
Thank You!
Chih-Ping LinGeotechnical Engineering GroupNational Chiao Tung University