are time-varying systems laplace-transformable?
TRANSCRIPT
University of Windsor Electrical & Computer Enginee ring Department
Are Time-Varying Systems Laplace-Transformable?
Shervin ErfaniResearch Centre for Integrated MicroelectronicsElectrical and Computer Engineering Department
University of WindsorWindsor, Ontario N9B 3P4, Canada
University of Windsor Electrical & Computer Enginee ring Department
Outline of the Presentation
• Objectives• From LTV Elements to LTV Systems• Observations on LTV Systems
– Bivariate Time and Bifrequency Characterization– Generalized -Delay System Representation
12/2/2009 2
– Generalized -Delay System Representation– Circularly Symmetric Functions
• Two-Dimensional Laplace Transform (2DLT)– Laplace –Carson Transform– Transformation of Nonanticipative Systems– Transformation of Synchronized Systems
• Conclusions• References
University of Windsor Electrical & Computer Enginee ring Department
Objectives � There has been an increasing interest in realizatio n and
implementation of linear time-varying (LTV) and ada ptive systems.
� The rather limited use of LTV networks in analog si gnal processingand communication fields is due largely to lack of powerful meansfor their analysis, synthesis and performance evalu ation.
� The theory of LTV systems has been widely based on the time - domain
12/2/2009 3
� The theory of LTV systems has been widely based on the time - domain approach.
� The main objective of this project is to provide a unified treatment for the analysis and synthesis of LTV systems.
� The approach to be described consists of extension of Laplace transform techniques commonly used for linear time- invariant (LTI) systems.
University of Windsor Electrical & Computer Enginee ring Department
Fundamental Input-Output Representation
• The classical theory of variable systems is based o n the solutions of linear ordinary differential equat ions with varying coefficients. – The varying coefficients are functions of an indepe ndent
variable, conveniently called the time .– The time is assumed to be real for physical systems.
12/2/2009 4
– The time is assumed to be real for physical systems.
∑∑==
=m
k
kk
n
i
ii txtbtyta
0
)(
0
)( )()()()(
University of Windsor Electrical & Computer Enginee ring Department
Observations on LTV Dynamic systemsObservation 1 – In general, time variables of the signal and system do not have to be synchronized; i.e., the (time) variables of the signal and system are independentof each other.
)(),()(),( txDKtyDL ττ =Observation 2 – At any instant of “t” there is a response, which is a specified function of “τ”.Observation 3 – At any fixed “τ” there is a response, which is a specified function of “t”.Observation 4 – The system response is a function of variations of observation
12/2/2009 5
Observation 4 – The system response is a function of variations of observationparameter “t” and application parameter “τ”.Observation 5 – A zero-input, SISO LTV system described by:
0)(),( =⋅yDL τis a linear system that its natural frequencies are varying with “τ”. In other words, solutions of this equation are exponential functions of time with varying natural frequencies, as given by:
∑=
−=⋅n
i
ti
iecy0
)()( τα
where )(τα i is a function of variable coefficients of the fundamental equation of the system under consideration.
University of Windsor Electrical & Computer Enginee ring Department
Linear Time Varying Elements• A single-input single-output (SISO) dynamic system element of
finite order characterized by its input-output rela tionship is said to be linear if the following holds for each t ≥0:
)()()( txthty =
12/2/2009 6
� Where h(t) is the system function defines the response at time t, denotes the slope of the y-x curve in a rectangular coordinates system.
h(t1)
h(t2)h(t3)
x(t)
y(t)
University of Windsor Electrical & Computer Enginee ring Department
Linear Time Varying Operator• A SISO dynamic system operation is shown symbolical ly by:
)}({)( txty ο=• The system operator is linear if and only if the fo llowing
relation holds:
12/2/2009 7
)()()}({)}({)}()({212121
tttttt yyxxxx βαβοαοβαο +=+=+
• The system input can be any function including an impulse or a delta function:
)()();( τδτδ
−= tthty
University of Windsor Electrical & Computer Enginee ring Department
Observing an Impulse Response Function (1)
• Observation 6 – The following symbolic identity holds:
)()()()( thtth −=− τδττδ
• Observation 7 - The product is different fro m zero at the point t=τ.
)()( τδ −tth
12/2/2009 8
at the point t=τ.
• Observation 8 - The impulse response of the system has a circular symmetric property with respect to its arg uments tand τ.
);();( tt yy ττδδ
=
University of Windsor Electrical & Computer Enginee ring Department
Observing an Impulse Response Function (2)
• Observation 9 – In the (t,τ)-plane , due to the circular symmetryand because delta function is an even function, we can define a bivariate response function as:
|)(|),(),();( τδτττ −= ttxthty
• Observation 10 - The ordinary output response at the point t=τ
12/2/2009 9
• Observation 10 - The ordinary output response at the point t=τis:
or, equivalently:
)()(|)(|),(),()(0
τττδτττ xhdtttxthy =−= ∫+∞
−
)()(|)(|),(),()(0
txthdttxthty =−= ∫+∞
−
ττδττ
Question – Can a system function be equal to an input functio n?Is the system output considered to be still linear?
University of Windsor Electrical & Computer Enginee ring Department
Frequency-Domain Characterization
• A linear time-invariant dynamic system described by a homogeneous n th-order differential equation:
• Using the operator d/dt ���� s, this equation can be
0)(0∑
=
=n
i
i
ii tydt
da
12/2/2009 10
0)(0∑
=
=n
i
ii tysa
• Using the operator d/dt ���� s, this equation can be written as:
• Thus, the homogenous solution is a linear combination of exponentials:
∑=
−=n
i
tsi
ieaty0
)( Roots of CharacteristicsOperator Equation
University of Windsor Electrical & Computer Enginee ring Department
Frequency-Domain Characterization (Cont.)• A linear time-invariant dynamic system with the
applied input x(t):
• The system is turned on at t, the impulse-response at t-τ is obtained :
)()(00
txdt
dbty
dt
da
m
kk
k
k
n
i
i
ii ∑∑==
=
12/2/2009 11
t-τ is obtained :
• We observe:• The response is a bivariate function of t and• Characteristics roots are a function of τ
∑=
−=n
i
tsi
ieaty0
)(),( ττ Roots of CharacteristicsOperator Equation
)();(00
τδτ −=∑∑==
tdt
dbty
dt
da
m
kk
k
k
n
i
i
ii
τ
University of Windsor Electrical & Computer Enginee ring Department
• The ordinary unilateral Laplace transform of is obtained as:
• This is a function of the variable application time τ.
Laplace Transform of the Impulse Function
)( τδ −t
ττδτδ 11
0
)()}({ sts edtettL −+∞
−
− =−=− ∫
12/2/2009 12
• This is a function of the variable application time τ.
• A second transformation yields:
210
2
1)}({ 21
ssdeetL ss
D +==− ∫
+∞
−
−− ττδ ττ
University of Windsor Electrical & Computer Enginee ring Department
Definition of 2DLT – The ordinary unilateral 2DLT is defined as:
Laplace-Carson Transform
20 10 21212211),(),( dtdteetthssH tsts
∫ ∫+∞ +∞ −−=
12/2/2009 13
Inverse Transformation - The inverse 2DLT is given by:
{ }
( )2 1
1 1 2 2
2 1
12 1 2 1 2
1 2 1 22
( , ) ( , )
1( , )
2
D
j j s t s t
j j
L H s s h t t
H s s e e ds dsj
σ σ
σ σπ
−
+ ∞ + ∞
− ∞ − ∞
=
= ∫ ∫
University of Windsor Electrical & Computer Enginee ring Department
Observation 11 – For conformal transformation, it is required that the unit function u(t, τ) transforms into itself:
u (t,τ) is equal to 1 when both t and τ are positive, and is equal to zero when at least one of the arguments is negat ive.
Two-Dimensional Laplace Transform (2DLT)
1),(),( 21 =⇔ ssUtu τ
12/2/2009 14
Observation 12 - Based on the above observation we modify the 2DLT is given by:
τττ τ dtdeethssssHth sts 21
0 0
2121 ),(),(),( −+∞+∞
−∫ ∫=⇔
University of Windsor Electrical & Computer Enginee ring Department
� The unit–step function u(t, τ) is defined as:
The Laplace-Carson transform of u(t- τ) is
Two-Dimensional Step Function
)()(),( ττ ututu =
22121
21)(),(ss
sdtdeetussssU sts
+=−= −−
+∞+∞
∫ ∫ ττ τ
12/2/2009 15
Similarly, the L-C transform of u( τ-t) is
Observation 13 - We can write:
The 2DLT transform of is .
≠=
=−−ττ
ττt
ttutu
0
1)()(
210 0
2121 )(),(ss
dtdeetussssU+
=−= ∫ ∫ ττ
21
1
ss
s
+
2
1
s)()( tutu −− ττ
University of Windsor Electrical & Computer Enginee ring Department
� The unit–impulse function δ(t, τ) is defined as:
The Laplace-Carson transform of δ(t-τ) is
Two-Dimensional Impulse Function
)()(),( τδδτδ tt =
212121
21)(),(ss
ssdtdeetssss sts
+=−=∆ −−
+∞+∞
∫ ∫ ττδ τ
12/2/2009 16
Similarly, the L-C transform of δ(t,τ) is
Observation 14 - We can obtain the L-C transform of as:
)()( τδ −tth
210 0
2121 )(),(ss
dtdeetssss+
=−=∆ ∫ ∫ ττδ
21),( sst ⇔τδ
∫+∞
+− +=⇔−0
2121)(
21 )()()()( 21 ssHssdehsstth ss τττδ τ
University of Windsor Electrical & Computer Enginee ring Department
Impulse-Response System Representation
• The more familiar impulse response, using sifting property of
• The system response for a circularly symmetric syst em function can equivalently be written as:
)(),(),( τδττδ −= tthty
12/2/2009 17
• The more familiar impulse response, using sifting property of the delta function will be:
)()(),()( ττδτττ
τδ hdttthy
t
t
=−= ∫+
−
=
=
� The limits of integration can be extended to infinit y.
University of Windsor Electrical & Computer Enginee ring Department
� Define the instant at which the input is applied to the system as the origin for time “ t.”
� The nonanticipative condition implies:
Nonanticipative System Function
ττ <≡− tforotutxth )()()(
12/2/2009 18
� Then, we may define h(.) to be zero for negative values of its argument:
τττ >≡− tfortytutxth ),()()()(
University of Windsor Electrical & Computer Enginee ring Department
� Let us define:
� The 2DLT is:
� Similarly, we define :
Frequency-Domain Representation of Nonanticipative System Functions
<>−
=τ
τττ
tfor
tforthth
0
)(),(1
21
1
0
211)()(),( 12
sssHdtthedessH tss
+=−= ∫ ∫+∞ +∞
−− τττ
τ
>− tforth ττ )(
12/2/2009 19
� Adding together, we obtain:
<>−
=tfor
tforthth
τττ
τ0
)(),(2
21
2212
)(),( sssHssH +=
{ }21
21212
)()(),(|)(|
ss
sHsHssHthL D +
+==−τ
University of Windsor Electrical & Computer Enginee ring Department
Symbolic Bifrequency Input-Output System Representation
h(t2)
Black box representation of circularly symmetric li near systems
12/2/2009 20
h(t2)
H(s2)y(t1) Y(s1)
x(t1) X(s1) t1 = t2-
University of Windsor Electrical & Computer Enginee ring Department
The 2DLT of General LTV SystemsConsider a SISO LTV system, initially at rest, described by:
An input x(.) is applied to the system at time ξ =t- τ
∑∑==
=m
kk
k
ki
in
ii dt
txdtb
dt
tydta
00
)()(
)()(
12/2/2009 21
Taking a 2DLT, we obtain:
∑∑==
−=m
kk
k
ki
in
ii dt
txdtb
dt
tydta
00
)()(
),()(
ττ
ττττ τ
τξ
τ dtdeedt
txdtbdtdee
dt
tydta sts
m
kk
k
k
tsts
i
in
ii
1212
0 00 0 0
)()(
),()( −−
=
∞
=
−−
=
∞ ∞
∑∫ ∫∑∫ ∫−=
−
This may demand more initialconditions that the problem requires!
University of Windsor Electrical & Computer Enginee ring Department
Synchronized System Representation
h(t2)
t1 is the observation time of signal and t 2 is the application time to the system.
12/2/2009 22
h(t2)
H(s2)y(t1, t2 )x(t1)
t2=0
t2=t1=0
t1
University of Windsor Electrical & Computer Enginee ring Department
The 2DLT of Synchronized LTV Systems� Consider a SISO LTV system described by:
� Taking Laplace transform with respect to τ , we obtain:
∑∑==
=m
kk
k
ki
in
ii dt
txdb
dt
tyda
00
)()(
)()( ττ
12/2/2009 23
� Taking Laplace transform with respect to τ , we obtain:
ττττ ττ ddt
txdebd
dt
tydea
m
kk
ks
ki
is
n
ii ∑ ∫∑ ∫
=
−+∞
∞−
−
=
+∞
∞−
=00
)()(
)()( 22
∑∑==
=m
kk
k
ki
in
ii dt
txdsB
dt
tydsA
02
02
)()(
)()(
University of Windsor Electrical & Computer Enginee ring Department
The 2DLT of Synchronized LTV Systems (Cont.)
� Taking a second transform with respect to t:
� If the system is initially at rest then for i =0, 1, 2, . . . , n-1.
dtedt
txdsBdte
dt
tydsA ts
m
kk
k
kts
i
in
ii
11
02
02
)()(
)()( −
=
−+∞
∞−=∑∫∑ =
0)( =
i
i
dt
tyd
12/2/2009 24
dt
∑ ∑
∑∑
=
−
=
−
=
−−
=
−=
=
−−−−=
m
k
k
j
jjkkk
m
k
kkkkk
in
ii
xssXssB
xxsxssXssBsYssA
0
1
0
)(1112
0
)()1(21
1111211
02
)0()()(
)0(...)0()0()()()()(
∑∑ ∑
=
=
−
=
−
−= n
i
ii
m
k
k
j
jjkkk
ssAxssXssBsY
012
0
1
0
)(11121
)()0()()()(� Thus
University of Windsor Electrical & Computer Enginee ring Department
Real-Variable Function Representationin ( t,τ)-Plane
f(p)
τo
12/2/2009 25
• A system function, which has the simple rotational property of circular symmetry is shown in this figure.
• Figure shows conversion of a one-dimensional profile of a system function of p to a two-dimensional function of a complex variable of t and τ.
• The common views of the independent signal and system functions are the projection over t-axis (pure real) and τ-axis (pure imaginary), respectively.
t
τ
p
o
θ
University of Windsor Electrical & Computer Enginee ring Department
Circularly Symmetric System Transformation
t1
τ
t + τ = t1
Region over which the impulse-response h(t, τ) of a nonanticipative system is defined is the shaded area.
12/2/2009 26
t1t
0
τττ τ dtdexthsY ts )()(),()( +−
∆∫∫=
Iterated LaplaceTransform
University of Windsor Electrical & Computer Enginee ring Department
Circularly Symmetric Functions (1)
h(t)
t2
θ
h(t)
t2
θ
12/2/2009 27
t1
L tt1
L t
Rotational property of a circularly symmetric funct ion
)],(),([2
1)()sin,cos(),( 1221 tthtththtthth sym +=≡= θθθ
)],(),([2
1)(),( 1221 ssHssHsHsH sym +=≡φ
University of Windsor Electrical & Computer Enginee ring Department
Other Observations on LTV Systems
• Mathematically speaking, t and τ represent time variables of an applied signal and the corresponding system.
• Variations of an input signal and an autonomous syste m are independent of each other.
12/2/2009 28
• For circularly symmetrical systems, with no loss of generality, we can rewrite the response as:
)(),(),(ln
txty eth ττ −=
University of Windsor Electrical & Computer Enginee ring Department
Generalized-Delay System Representation
• If h(t) is a (piecewise) continuous function and bounded by a finite number, its 1 st -order and higher-order derivatives exists.
• The system response can equivalently be written as:
12/2/2009 29
• The system response can be written more compactly a s a generalized-delay operator:
)()(),(
txty etg τ−=
)()( )(
)('
txty et
h
dh∫= −τ
ξξξ
University of Windsor Electrical & Computer Enginee ring Department
� The Hankel transform is compatible with LTV systems described by a general Bessel equation given as:
The Hankel Transform
)()(1 2
2
2
2
2
txtyat
n
dt
d
tdt
dN
=
±
−+
12/2/2009 30
� The Hankel transform pairs are symmetric because it deals with symmetric functions.
� The 2DLT of a circularly symmetric function with the property lim t�∞ h(t) �0 that is a Hankel transform of order zero.
� This property is quite useful in application of Hankel transforms to LTV systems.
tdttdt
University of Windsor Electrical & Computer Enginee ring Department
� The Mellin transform is compatible with LTV systems characterized by a general Euler-Cauchy equation given as:
� The impulse response of this nonanticipative Euler-Cauchy LTV system is:
The Mellin Transform
)()(
0
txdt
tydta
i
ii
n
ii =∑
=
)()(1
),( ττ −= tut
gth
12/2/2009 31
� where g(.) is the impulse response of a prototype LTI system obtained by changing the time scale t → - ln t
� The 2DLT in this case becomes the following Mellin Transform pairs:
)()(1
),( ττ
τ −= tugt
th
{ } dttththM s 1
0
)()( −+∞
∫= { } dstsHj
sHM sj
j
−∞+
∞−
−∫=
σ
σπ)(
2
1)(1
University of Windsor Electrical & Computer Enginee ring Department
Conclusions• The 2DLT techniques are applicable to LTV systems.
• This investigation justifies application of 2DLT as an operational calculus for system characterization, es pecially for analog signal processing problems.
• This approach allows, in effect, two-dimensional t ransform techniques to be used for the time -varying systems in the
12/2/2009 32
techniques to be used for the time -varying systems in the same manner that the conventional frequency-domain techniques are used in connection with fixed system s.
• The 2DLT, Mellin transform, and Hankel transform ca n be derived from the two-dimensional Fourier transform.
• The work presented here opens several areas for fur ther investigations in theory of variable systems.
University of Windsor Electrical & Computer Enginee ring Department
For Further Information1. S. Efani, “Extending Laplace and Fourier transforms and the case of variable systems: A
personal perspective,” Presentation to IEEE Signal Processing Long Island Section, NY, May 15, 2007.
2. S. Erfani and N. Bayan , “On Linear Time-Varying System Characterizations,” Proc. IEEE EIT 2009, Windsor, ON, Jun. 7-9, 2009, pp. 207-210.
3. S. Erfani and N. Bayan, “Laplace, Hankel, and Mellin Transforms of of Linear Time-Varying Systems,” Proc. IEEE 52th MWSCAS, Cancun, Mexico, Aug. 2-9, 2009, pp. 794-799.
4. V. A. Ditkin and A. P. Prudnikuv, Operational Calculus in Two Variables and Its Applications,Pergaman Press, 1962.
5. D. Voelker and G. Doetsch, Die Zweirdimensional Laplace Transformation, Birkhauser
12/2/2009 33
Verleg, Basel, 1950.
6. A. H. Zemanian, Generalized Integral Transformations, Dover Publications, New York, NY, 1987.
7. I. N. Sneddon, Fourier Transformations, Dover Publications, New York, NY, 1995.8. N. Bayan and S. Erfani, “Frequency-domain realization of linear time-varying systems by
two-dimensional Laplace transformation,” Proc. NEWCAS, Montreal, Que., June 10-14, 2008, pp. 213-216.
9. A. D. Poularikas, (ed.), The Transforms and Applications Handbook, 2nd Ed., CRC Press, Boca Raton, FL, 2000.
10. L. A. Zadeh, “Time-varying networks, I,” Proc. IRE, vol. 49, October 1961, pp. 1488-1503.
University of Windsor Electrical & Computer Enginee ring Department
12/2/2009 34
University of Windsor Electrical & Computer Enginee ring Department
Shervin ErfaniShervin Erfani is a professor and former Electrical and Computer Engineering Department Head at the University of Windsor, Windsor, Ontario, Canada. His experience spans over 30 years with AT&T Bell Labs, Lucent Technologies, University of Puerto Rico, University of Michigan-Dearborn, Stevens Institute of Technology, and Iranian Naval Academy. His expertise expands over many areas from System Theory to Digital Signal Processing to Network Security Management. He has been a consultant to the industry for a number of years.
Dr. Erfani has published more than 70 technical papers, holds three patents, and is the Senior Technical Editor of the Journal of Network and Systems Management and an associate editor for Computers & Electrical Engineering: An International Journal.
He received a combined B.Sc. and M.Sc. degree in Electrical Engineering from the University of Tehran in 1971, and M.Sc. and Ph.D. degrees, also in Electrical Engineering, from Southern Methodist University in 1974 and 1976, respectively. He was a Member of Technical Staff at Bell Labs of Lucent Technologies in Holmdel, New Jersey, from 1985 to 2001.