1

Signals Systems: Working Notes

Michael Miller

Abstract

I. T WO L INEARITY AND T IME I NVARIANT E XAMPLES ; C ONSTANT C OEFFICIENT D IFFERENTIAL AND D IFFERENCE E QUATIONS Given a rst order, constant coefcient differential equations with zero-initial condition y(t) + ay(t) = x(t) , y() = 0 . Let us call x(t) the input, y(t) the output, and the system transformation equation describing the relationship between input and output is given by the differential equation. Given is a rst order, constant coefcient difference equation with zero-initial condition

shifts the output the same amount of time with no other transformation. This rst order differential equation has these two properties because (i) derivatives appear linearly in the equation, and (ii) the coefcients are constant in time. II. L INEARITY Now let us return to the two important properties of linearity and time-invariance. Linearity is a very powerful property, it essentially allows for us to divide complex signals into simpler ones and solve more complex signals by solving the simpler parts. Let us dene linearity as follows.

Denition II.1. A continuous system with input x(t) with output y(t) is linear if (i) x(t) creates y[n] = y[n 1] + x[n] , y[] = 0 . y(t) then ax(t) has output ay(t), and (ii) if x1 (t) creates y1 (t) and x2 (t) creates y2 (t), then Let us call x[n] the input, y[n] the output, and a1 x1 (t) + a2 x2 (t) creates a1 y1 (t) + a2 y2 (t) for the system transformation equation describing any x1 , x2 , a1 , a2 . the relationship between input and output is A discrete time system with input x[n] with given by the differential equation. output y[n] is linear if (i) x[n] creates y[n] Such systems enjoy two of the most imthen ax[n] has output ay[n], and (ii) if x1 [n] portant properties we shall study through this creates y1 [n] and x2 [n] creates y2 [n], then class, the properties of linearity and timea1 x1 [n] + a2 x2 [n] creates a1 y1 [n] + a2 y2 [n] for invariance. (i) Linearity: We will dene this any x1 , x2 , a1 , a2 . subsequently, but for now, linearity corresponds to the the very straightforward notion that if you scale the input x it scales the output y, and if you add inputs, the outputs add as well. (ii) Time-Invariance: This corresponds to the notion that shifting the input in time, simply
M. Miller is with the Department of Biomedical Engineering at the Johns Hopkins University. Prepared from 2000-2010

There is a second property which this rst order differential equation enjoys, the time invariance property coming from the constant coefcient properties. As for the time invariance, simply stated, an input shifted results in the d same output just shifted in time. (1 y1 (t) + 2 y2 (t)) +a (1 y1 (t) + 2 y2 (t)) dt Denition III.1. A continuous time system with y3 (t) y3 (t) = 1 (y1 (t) + ay1 (t)) + 2 (y2 (t) + ay2 (t)) input x(t) and output y(t) is time-invariant if x(t t0 ) creates output y(t t0 ), for all inputs = 1 x1 (t) + 2 x2 (t) . x and shifts t0 . x3 (t) A discrete time system with input x[n] and Example II.2. Let the input be x(t) = eit , output y[n] is time-invariant if x[nn0 ] creates 1 then y(t) = a+i eit . output y[n n0 ], for all inputs x and shifts n0 . i To see this, y(t) = a+i eit and then Example III.1. Let the input be x2 (t) = a i it it it e + e = e . ei(tt0 ) = x1 (t t0 ), then we have to show y(t) + ay(t) = a + i a + i (1) 1 y2 (t) = y1 (t t0 ) = ei(tt0 ) . a + i Example II.3. So now lets use the linearity i property. Suppose we change the input from To see this, y2 (t) = a+i ei(tt0 ) and then eit to cos t. What is the output of the differi i(tt0 ) a + ei(tt0 ) ential equation. Well, LTI systems allow us to y2 (t) + ay2 (t) = a + i e a + i solve for this change in input by rewriting the = ei(tt0 ) = x2 (t) . cosines in terms of the complex exponentials, and then adding the solutions: eit + eit cos t = , 2 implying that for x(t) = cos t then y(t) becomes y(t) = 1 1 1 1 eit + eit 2 a + i 2 a i 1 cos(t arctan/a) . a2 + 2

Example II.1. The rst order differential equation with zero initial conditions is linear. Assume x1 creates y1 and x2 creates y2 , y1 () = y2 () = 0, then we have to prove x3 = 1 x1 + 2 x2 creates y3 = 1 y1 + 2 y2 . Then we have

III. T IME I NVARIANCE

This reduces to y(t) =

Thus we see that a simple shift of the complex exponential, the output is shifted in the same manner. Demonstrate to yourself that the same is true for arbitrary periodic inputs represented via the Fourier series x(t) = k ak eik t . Example III.2. Let y[n] = nx[n]. Then we can see this is not time-invariant. Pick a signal x1 [n] = 0 for n = 0, and x1 [0] = 1. Then for x1 [n] we have y1 [n] = 0. For x2 [n] = x1 [n 1] we have y2 [1] = 1 and y2 [n = 1] = 0. Notice not a simple shift so that y2 [nn0 ] = y1 [nn0 ] IV. S UMMARY So now we have examined several simple properties that the differential equation satises, linearity and time-invariance. So we have arrived at the second major component of our study of linear, time-invariant systems. To understand the outputs of LTI systems to arbitrary inputs, one needs to understand the convolution integral. The remaining 12 lectures work to generalize and strenghten the these very notions.

V. U NIT I MPULSE S AMPLE AND U NIT S TEP A. The unit sample and unit step Lets examine some special signals, rst in discrete time, then in continuous time. Denition V.1. The discrete time unit step is given by u[n] = 1 , n 0 ; u[n] = 0 , n < 0 . The discrete unit sample or impulse is dened as [n] = u[n] u[n 1] . The unit sample sifts signals. Proposition V.1. The unit sample has the sampling or sifting property, picking off values of signals that it sums aginst:

x[n] =
k=

x[n k][k] .

This is true for all signals, implying for x[n] = u[n], then
n

u[n] =
k=

[k]

Denition V.2. The continuous time version has a similar form: u(t) = 1 , t 0 ; u(t) = 0 , t < 0 . The differentiated form is the continuous unit impulse: (t) = u(t) In particular, dening x(t) = u(t), then we see the integrated property follows
t

u(t) =

( )d .

Proposition V.2. The sampling or sifting property follows for smooth functions with

x() = x() = 0:

B. The Unit Sample Response of LTI Systems Now we dene the unit sample and unit impulse responses of our systems. Denition V.3. Dene y[n] = hk [n] to be the unit sample response of a system with input x[n] = [n k], the unit sample shifted to time k. If the system is time invariant, then dene h0 [n] = h[n], and hk [n] = h[n k]. Dene y(t) = h (t) to be the unit impulse response of a system with input x(t) = (t ), the unit impulse shifted to time . If the system is time invariant, then dene h0 (t) = h(t), and h (t) = h(t ).

x(t) =

x(t )( )d .

To prove this we use integration by parts. Also assume the functions x() = x() = 0, vanishing at . Integration by parts using the total differential gives for the product f ( )g( ) with f () = f () = 0: 0 = f ()g() f ()g()

d(f ( )g( ))

df ( )g( ) +

f ( )g( )d .

Dening f ( ) = x(t ), g( ) = u( ), then

Given a linear system, then the unit sample and unit impulse responses determine the out dx(t )u( ) put of these linear systems. If the systems are x(t )u( )d = also time invariant, then there is only one im= dx(t ) = (x(t ) x(t)) pulse response and it just gets shifted around. 0 = x(t) . Example V.1. Return to the simple example y[n] = nx[n]. Then h0 [n] = 0, with h1 [n] = 1[n 1], and h2 [n] = 2[n 2], ...., hk [n] = k[n k]. We see that there are many different impulse responses, which are not simply shifts of each other. So now we can write down the general time solution for arbitrary inputs. It involves the convolution integral or sum of the input and the response of the system. Denition V.4. Dene the convolution of two ccontinuous time signals f (t) g(t) as

f (t) g(t) =

f ( )g(t )d ;

dene their discrete analogue f [n] g[n] to be

f [n] g[n] =
k=

f [k]g[n k] .

Proposition V.3. Let x[n] be rewritten as x[n] = k x[k][n k]. Then the output of a

linear system with unit sample responses hk [n] Example V.2. Let system be a constant coefcient difference equation with zero initial to input is given by condition y[] = 0 and with input x[n] and y[n] = x[k]hk [n] ; output y[n] given by system equation
k

if the system is time-invariant, with h0 [n] = h[n], then y[n] =

k x[n]h[n]

x[n] = y[n] y[n 1] . Lets show that it is linear, time-invariant with unit sample response h[n] = n u[n] .

x[k]h[n k] .

Let x(t) be rewritten as x(t) = Linearity: Let x1 [n] give y1 [n], and x2 [n] x( )(t )d . Then the output of give y2 [n], then we have a linear system with unit sample responses (ay1 [n] + by2 [n]) (ay1 [n 1] + by2 [n 1]) h (t) to input is given by = ay1 [n] ay1 [n 1] + by2 [n] by2 [n 1] . y(t) = x( )h (t)d ; ax1 [n] bx2 [n]

if the system is time-invariant, with h0 (t) = Time invariant: Let x1 [n] produce y1 [n], then introducing shift m0 we must prove x2 [n] = h(t), then x1 [n m0 ] produces y2 [n] = y1 [n m0 ]. Given y(t) = x( )h(t )d . x1 [n] = y1 [n] y1 [n 1], then we have
x(t)h(t)

x1 [n m0 ] = y1 [n m0 ] y1 [n m0 1] ;
x2 [n] y2 [n] y2 [n1]

thus y2 [n] = y1 [n m0 ]. To solve for the unit-sample response, let [n] = h[n] h[n 1] . Then h[n] = 0, n 1, then h[0] = [0] = 1 h[1] = h[0] = h[2] = h[1] = 2 h[n] = n , n 0 .

Example V.3 (Differential Equation Integrating Factor). Now we turn our focus to a pure time domain analysis, understanding the response of the differential equation directly in terms of its time domain inputs. For this we explore the convolution integral. We do this by solving the rst-order differential equation directly using integrating factors. Proposition V.4. The output of this rst order differential equation x(t) = y(t) + ay(t), y() = 0 , with input x(t) is given according to y(t) = x(t) eat u(t) , where the impulse response of the differential equation is h(t) = eat u(t). For this, examine the differential equation and introduce the integrating factor f (t) which has the property that it makes one side of the equation into a total differential. Starting with x(t) = y(t)+ay(t), then dene the integrating factor f to satisfy f (t)x(t) = f (t)y(t) + f (t)ay(t) d (f (t)y(t)) . = dt For such a function f , this allows us to solve the differential equation directly according to
t

We have almost arrived at our convolution formula. For this introduce the unit step function, and the denition of the convolution formulation. The unit-step function is zero to the left of the origin, and 1 elsewhere: u(t) = 1 , t 0 , u(t) = 0 , t < 0 . Now using the unit-step rewrite the solution according to

y(t) =

x( ) ea(t ) u(t )
h(t )

= x(t) h(t) .

f (t)y(t)f ()y() =

f ( )x( )d .

Solving the integrating factor gives f(t)y(t) + f (t)y(t) = f (t)y(t) + af (t)y(t) . Thus the integrating factor is f (t) = eat , and using the boundary condition y() = 0 the total differential is solved giving
t

eat y(t) =

ea x( )d .

Example V.4. Here is an integrating factor for the non time-invariant case. One of the wonderful students pointed this out. Let y(t0 ) = A an initial condition with input applied after t0 with differential equation which has a non constant coefcient: x(t) = y(t) + a(t)y(t), y(t0 ) = A , then the integrating factor satises f(t) = a(t)f (t) . This gives f (t) = e tion takes the form y(t) = Ae

t t0 t t0

We make the following comment. Notice the output is a function of the input convolved with a property of the system, eat u(t). This property we call the impulse response of the system and we will study it extensively. For LTI systems this will always be true, although the property of the system will change depending on the system. Now we solve for the impulse response directly. Example V.5. Let us return to our favorite differential equation: x(t) = y(t) + ay(t) , y() = 0 ,

a()d

implying the soluwith impulse response h(t) = eat u(t) . To see this, let us now show for the differential equation the impulse response eat u(t) solves the differential equation: (-5) (t) = h(t) + ah(t) , h() = 0 . t a()d d For this we will use the fact that u(t) = (t). Assuming h(t) = eat u(t), we have h(t) + ah(t) d at e u(t) + aeat u(t) = dt d = eat u(t) aeat u(t) + aeat u(t) dt d = u(t) = (t) . dt

t a( )d

+
t0

x( )e

a()d

d .

To see this satises the differential equation just calculate. At t0 , y(t0 ) = A, then differentiating gives y(t) + ay(t) = a(t)Ae

t t0

t a( )d

a(t)
t0 a(t)y(t)

x( )e

+x(t) + a(t)y(t) = x(t) .

Example V.6. Examine an RLC circuit, take the output across the capacitor. Then y(t) is taken as the output, with input the voltage source x(t). To derive the input-output relationship, lets use one of the two stawlwarts of circuit theory - Kirchoffs voltage law. The voltage across the capacitor output y(t) when added to the voltage across the resistor must add to the input voltage x(t). The voltage across the resitor is RI(t), I the derivative of the current given by C y(t). Thus we have the equation x(t) = RC y(t) + y(t) , y() = 0 . Notice the initial condition corresponds to no voltage across the capacitor - system at rest. We have our LTI system (rst-order constantcoefcient differential equation), with impulse t 1 response h(t) = RC e RC u(t) (notice, the 1/RC scaling results from the equation). The output becomes

Example V.7. Adding initial condition so the system is not at rest make it a very different story for the linear difference and differential equations. Lets see that the system is not linear when we add an initial condition. Let x(t) = y(t) + ay(t), y(t0 ) = A , then the solution becomes
t

y(t) = Aea(tt0 ) +
t0

x( )ea(t ) d .

Check this solve the differential equation with initial condition.

y(t) =

x( ) 1 RC
t

1 t e RC u(t )d RC x( )e RC d .
t

10

VI. U NIT S TEP R ESPONSES We can derive the impulse response from the unit step response by using the fact that derivative and differences of step response gives the impulse response.

Example VI.1. So now we have our general solutions for the difference equations and differential equations, so we can look at unit-step responses. For the difference equation x[n] = y[n] y[n 1] ,

Proposition VI.1. Given a continuous time we have LTI system with input x(t) = u(t) the unit step, then the derivative of the response is the y[n] = x[n] h[n] = impulse response: y(t) = h(t) . Given a discrete time LTI system with input x[n] = u[n] the unit step, then the rst difference of the response is the impulse response: y[n] y[n = 1] = h[n] . y[n] = Proof: y(t) = d d u(t) h(t) = h(t) u(t) dt dt d = h( )u(t )d dt d = h( ) u(t )d dt = h( )(t )d = h(t) . we have
k=

x[k]h[n k]
k=

=
k= n

x[k]nk u[n k] x[k]nk .

k=

For an input of x[n] = u[n], then we have

n n

u[k]

nk

=
k=0

nk u[n]

= n

1 u[n] 1 1 n+1 1 1 n+1 = u[n] = u[n] 1 1

(n+1)

For the differential equation x(t) = y(t) + ay(t)

y[n] y[n 1] = u[n] h[n] u[n 1] h[n 1] = h[n] u[n] h[n 1] u[n 1] =
k

y(t) = x(t) h(t) =

t

x( )ea(t ) u(t )d

x( )ea(t ) d .

h[k](u[n k] u[n 1 k]) h[k][n k] = h[n] .

k

For x(t) = u(t), then

t

y(t) =
0

1 ea(t ) d u(t) = eat ea |t u(t) 0 a

1 = (1 eat )u(t) a

11

Denition VII.2. A system is BIBO stable (bounded-input, bounded output), if a bounded Proposition VII.1. Convolution has the prop- input implies a bounded output. erty that Here is a sufcient condition for LTI sysVII. P ROPERTIES : C OMMUTATIVE , C AUSALITY, BIBO S TABILITY x(t) h(t) = h(t) x(t) ; x[n] h[n] = h[n] x[n] . Show this by redening the limits of the integral. tems. Proposition VII.3. If |h(t)|dt < then the system is BIBO stable. Let |x(t)| < B, t, then

h( )x(t )d Denition VII.1. A system is causal, if the y(t) = x(t) h(t) = h(t) x(t) = output y(t) at time t is not a function of future |h( )||x(t )|d B |h( )|d < C inputs.

Here is a sufcient condition for LTI systems.

Example VII.2. Let h(t) = u(t), then system not BIBO stable. Let input x(t) = u(t), then Proposition VII.2. If an LTI system is causal, |x(t)| 1 bounded, and t then this implies h(t) = 0, t < 0. For discrete d = t . u( )u(t ) = y(t) = x(t)u(t) = systems, h[n] = 0, n < 0. 0 To see this,

There does not exist a bound C for all time t. x( )h(t )d

y(t) = x(t) h(t) =

t

x( )h(t )d

Example VII.3. Take our favorite differential equation, with 0. Then h(t) = et u(t) and for 0 we have

+
t

x( )h(t )d

1 et dt = (e 1) = .

= f (x( ), > t) , if h(t) = 0 f or t < 0 . Example VII.1. Let h(t) = u(t + 3), then system is not causal.

Notice for = 0, you have to be more careful (differentiate numerator and denominator).

y(t) = x(t) h(t) =

t+3

x( )u(t + 3 )d

x( )d

= f (x(s), s > t) .

12 1 . a+i2f

VIII. C OMPLEX E XPONENTIALS ARE E IGENFUNCTIONS OF LTI S YSTEMS So now we have examined several simple properties that systems satisfy including linearity and time-invariance, and the implication that the outputs for LTI systems are given by convolution integrals and sums. Now we will see that the complex exponential have the special property that it passes through changed only by a complex number the differential equation. We will see the roll of tansforms, as representing arbitrary inputs via the superpositions of complex exponentials. This discussion is often called a frequency domain analysis. Frequency domain analysis studyies the outputs of linear and time-invariant systems via their response to complex exponentials. A. The Eigenfunction property The differential equation is a beautiful example of a linear and time-invariant system, denoted LTI. All LTI system enjoy the following exquisite property which lays the basis for our use of Transforms and the special roll of complex exponentials, cosines and sines in LTI system. Example VIII.1. Let us see that this is true for our differential equation. Let x(t) = ei2f t , then the output to the differential equation with boundary condition y() = 0 is simply a scaled version of the input: y(t) = H(f )ei2f t = H(f )x(t) . The complex coefcient H(f ) is independent 1 of time t and given by H(f ) = a+i2f . To see this, simply plug it into the differential equation, giving for x(t) = ei2f t then ei2f t = i2f H(f )ei2f t + aH(f )ei2f t ,
x(t) y(t) ay(t)

which implies H(f ) =

13

Proof: To prove it, simply do the continExample VIII.2. Not all inputs have this property. Let us look at a particular example, ex- uous time: amine the special function the sinc function, y(t) = x(t) h(t) = h(t) x(t) very important in communciations engineering h( )x(t )d = sin 2t . x(t) = sinc(t) = 2t h( )ei2f (t ) d = Let us demonstrate that y(t) = Hx(t) for any complex constant H. Assume it is, then solving h( )ei2f d . = ei2f t the differential equation gives sin 2t sin 2t cos 2t sin 2t H = . +aH 2 2t 2t 2t 2t Example VIII.3. Given the differential equa1 Choosing t = 2 implies H = 0, which is a tion contradiction. x(t) = y(t) + ay(t) , H2 So, the fact that complex exponentials pass through this differential equation unchanged is in fact a general property of all LTI systems, and is an important result. It forms the basis for the application of Fourier and Laplace transforms to LTI systems. The general result is that LTI systems transform complex exponentials with a simple complex number; this complex number turns out to be the Fourier transform of the system impulse response. Proposition VIII.1. For input x(t) = ei2f t to the LTI system with impulse response h(t) and output y(t), then

then h(t) = eat u(t) giving

h( )ei2f d =
0

e(a+i2f ) d e(a+i2f ) | 0 . a + i2f

For a > 0, then the upper limit is zero and the result becomes

H(f ) =
0

e(a+i2f ) d =

1 . a + i2f

y(t) = ei2f t

h( )ei2f d ;
F ouriertransf orm

for input x[n] = ei2f n and discrete LTI system then output

y[n] = e

i2f n k=

h[k]ei2f k
discreteF ouriertransf orm

We will call these complex numbers the Fourier tranform H of the impulse response h.

14

Example VIII.4. Given the difference equation with || < 1, then x[n] = y[n] y[n 1] , then

IX. T HE T RANSFER F UNCTION OF THE S YSTEM We have already seen the special property of complex exponentials: they pass through relatively unchanged LTI systems, certainly the differential equation. Laplace and Fourier transforms exploit this idea, realizing that there are many signals that can be represented via complex exponentials. So, let us start with the Fourier series, a simple idea: suppose arbitrary signals could be represented according to the complex exponential series given by x(t) =
k

H(f ) =
k=

h(k)ei2f k (ei2f )k =
0

1 . (1 ei2f )

CHECKING OUR ANSWERS: Notice, complex exponentials pass right through this system as well with just a complex number added. guess y[n] = H(f )ei2f n , then e
i2f n

Xk ei2fk t , Xk complex .

= H(f )e

i2f n

H(f )e

i2f (n1)

implying H(f ) = 1 1 ei2f

Notice, if T is the period of each of the frequencies so that fk = k/T (which will be the case when we look at Fourier series), then the signal x(t) in this expansion is periodic x(t) = x(t mT ). The special property of LTI systems is that using the divide and conquer strategy of LTI systems, the output of such a series is a linear superposition of complex sinusoidal inputs. Example IX.1. For y(t) related to input x(t) with differential equation at rest y(t) + ay(t) = x(t), y() = 0, then for x(t) = i2fk t the output becomes k Xk e y(t) =
k

Xk

1 a + i2fk
F ouriertransf ormH(f )

ei2fk t .

15

X. P ERIODIC S IGNALS - F OURIER S ERIES For an LTI system, x(t) = ei2f t , then the E XPANSIONS complex number determining the output y(t) = i2f t H(f )e is given by the Fourier transform of the impulse response, also called the transfer Many, many signals have complex sinusoidal function of the system; representations.

H(f ) =

h(t)ei2f t dt .

The transfer function formula for arbitrary inputs is given as follows. Proposition IX.1. Given inputs x(t) =
k

First we notice that series representations 1 built from harmonics of frequencies f0 = T are periodic on T ! Proposition X.1. Let x(t) be a superposition k of harmonics fk = T = kf0 , so that x(t) =
k

Xk ei2fk t ,

Xk ei2 T t ,

to LTI system with impulse response h(t), then notice, outputs of LTI systems y(t) always take the form: y(t) =
k

then, x(t) = x(t mT ). Let us now throughout denote periodic signals with period T via the notation xT (t), explicitly indicating the period. The Fourier series representations gives a formula for the complex numbers in the representations Xk . Proposition X.2. Let xT (t) be periodic with k period T , so that the frequencies fk = T = kf0 ,f0 = 1/T and SYNTHESIS x (t) =
k T

Xk H(fk ) ei2fk t ,
systemtransf er

where

H(fk ) =

h(t)ei2fk t dt .

This is why transform representations are so valuable.

Xk ei2 T t =
k

Xk ei2kf0 t

Then, the complex coefcients Xk are given by ANALYSIS k 1 T T = x (t)ei2 T t dt T 0 1 T /2 T x (t)ei2kf0 t dt . = T T /2

Xk

We denote these as a Fourier series pair: xT (t) Xk . Proof: Let prove the analysis equation for the complex coefcients. Use the property that

16 T 0
(mn) t T

ei2

dt = T [m n]. Then we have

T 0 T

xT (t)ei2 T t dt =
0 k T

OK, so how do we use this. Well, for periodic signals with period T , then we just have to evaluate the Fourier series coefcients Xk . Example X.1. 1) x(t) = constant, then a0 = constant and Xk = 0, k = 0 for any period T . 2) xT (t) = ei2f0 t with T = f10 , a1 = 1, Xk = 0, k = 1. 3) xT (t) = cos 2f0 t, then T = f10 , a1 = a1 = 1 , Xk = 0, k = 1. 2 1 4) xT (t) = sin 2f0 t, then T = f10 , a1 = 2i , 1 a1 = 2i , Xk = 0, k = 1. Example X.2. Let xT (t) = B, t [ A + 2 A mT, 2 + mT ] and 0 otherwise, with m an integer. Then Xk = 1 T
T /2 T /2
A 2

Xk ei2 T t ei2 T t dt Xk
k 0

= =
k

ei2

(km) t T

dt

Xk T [k m]

xT (t)ei2 T t dt ei2 T t dt
k A k

B = T =

A 2

i2 T t 2 | A Be
2

k i2 T
k k

k B sin T A B ei T A ei T A = = k T k i2 T

17

XI. N ON - PERIODIC F OURIER T RANSFORM So, Fourier series is for periodic signals; the Fourier transform is for non-periodic signals. Lets examine and construct the Fourier transform by allowing the period of the periodic signals to go to , see what we get. Lets dene xT (t) to be the periodic version of x(t), where x(t) has nite support x(t) = 0, |t| T /2. Thus, xT (t mT ) = xT (t), t [T /2, T /2] . Denition XI.1. Dene the Fourier transform of x(t) to be

Example XI.1. Let x(t) = B, t [ A , A ], and 2 2 0 otherwise. Then

X(f ) =

x(t)ei2f t dt
A 2

= B
A 2

ei2f t dt
A

i2f B sin f A = f

= a

2 ei2f t | A 2

=B

eif A eif A i2f

Let xT (t mT ) = x(t), t [T /2, T /2]. Then xT (t) =

k

X(f ) =

x(t)ei2f t dt .

Xk ei2kf0 t where f0 =

1 T

Proposition XI.1. Then for signals x(t) which are zero outside of T /2, T /2, then we have the relationship between FT and FS. Xk = 1 1 X(f )|f =kf0 f0 = T T T Xk ei2kf0 t . where x (t) =
k

Xk =

B sin kf0 A 1 X(kf0 ) = T T kf0 k B sin T A = k

Proof:
T /2

Xk =
T /2

xT (t)ei2 T t dt x(t)ei2 T t dt

T k

= X(f )|f = k .

18

OK, so we see that the Fourier transform can be used to dened the Fourier series. Now what we would like to do is understand how to represent the periodic signals when the period goes to innity T , so that the we can have a synthesis pair.

Example XI.2. Let x(t) = B, t [ A , A ], and 2 2 0 otherwise, then the Fourier transform

X(f ) =
A

x(t)ei2f t dt Bei2f t dt =
A

B sin f A . f

Proposition XI.2. Then the Fourier transform Then the synthesis shows is given by B sin f A i2f t e df . x(t) = f ANALYSIS

X(f ) =

x(t)ei2f t dt ,

with synthesis equation SYNTHESIS

x(t) =

X(f )ei2f t df .

Proof: Lets remind ourselves that xT (t) is the periodic version of x(t), where x(t) has nite support x(t) = 0, |t| T . Let 2 xT (t) be periodic with period T , and x(t) = limT xT (t). Then x(t) = = = =
T

lim xT (t) = lim lim

Xk ei2kf0 t
k

1 X(kf0 )ei2kf0 t T X(kf0 )ei2kf0 t f0

f0 0

lim

X(f )ei2f t df

19

Example XI.3. Another example, go back to the impulse response of the differential equations h(t) = eat u(t) H(f ) = 1 . a + i2f

XII. E VEN -O DD P ROPERTY Lets remind ourselves of some properties. X(f ) = x(t) (cos 2f t i sin 2f t)dt ,

To see this the Fourier transform

H(f ) =

h(t)ei2f t dt eat u(t)ei2f t dt =

1 a + i2f

then if x(t) = real, then RealX(f ) is even, then ImageX(f ) is odd. Let x(t) = evensignal, so that x(t) = x(t), then X(f ) = = x(t) (cos 2f t i sin 2f t)dt x(t) sin 2f t dt
odd

The synthesis becomes

x(t) cos 2f t dt i
even

h(t) =

H(f )e

+i2f t

df

1 ei2f t df a + i2f

Thus if x(t) is real and x(t) is even, then X(f ) is real. Example XII.1. B sin f A A A x(t) = B, t [ , ] X(f ) = 2 2 f
Real,Even

Example XI.4. Let x(t) = (t), then

X(f ) =

(t)ei2f t dt = 1

x(t) = (t) 1 XIII. T HE T IME S HIFT P ROPERTY The time shift property is important. It says that the magnitude of the Fourier transform doesnt change - just the relative phases of each of the components. For pure time shift this corresponds to linear phase. Proposition XIII.1. If x(t) X(f ), then y(t) = x(t t0 ) Y (f ) = X(f )ei2f t0 Proof: y(t) = = = Y (f )ei2f t df = x(t t0 ) X(f )ei2f (tt0 ) df X(f )ei2f t0 ei2f t df
Y (f )

20

Example XIII.1. (t t0 ) ei2f t0

XIV. T HE C ONVOLUTION P ROPERTY AND S YSTEM T RANSFER P ROPERTY A celebrated and important property, showing the role of Fourier transforms and LTI systems. Proposition XIV.1. Convolution Property for Fourier Series Let xT (t) be periodic, period T , 1 f0 = T xT (t) = Xk ei2kf0 t
k

then the output of an LTI system y T (t) = i2kf0 t with impulse response h(t) is k bk e given by y T (t) = xT (t) h(t) =
k

Xk H(kf0 ) ei2kf0 t
bk

H(kf0 ) =

h(t)ei2kf0 t dt

Proposition XIV.2. Convolution Property for Fourier Transform Let period go to innity x(t) = limT xT (t), and

y(t) = x(t) h(t) =

X(f )H(f ) ei2f t df .

Y (f )

y(t) = = = = =

lim xT (t) h(t) lim lim Xk H(kf0 )ei2kf0 t

k

1 X(kf0 )H(kf0 )ei2kf0 t T X(kf0 )H(kf0 )ei2kf0 t f0

f0 0

lim

X(f )H(f )ei2f t df

21

Example XIV.1. Examine the RLC circuit, out- Example XIV.3. Another proof of put across the capacitor, then x(t) X(f ) , h(t) H(f ) , 1 1 y(t) + x(t) , y(t) = RC RC then
1 with impulse response h(t) = RC e RC u(t). Then assume input xT (t) a periodic waveform xT (t) = B, t [ A , A ], periodic xT (t mT ) = 2 2 xT (t). Then the output y T (t) in Fourier series is given by
t

y(t) = x(t) h(t) Y (f ) = X(f )H(f ) .

y(t) = x(t) h(t) =

x(s)h(t s)ds

Y (f ) = x(s)h(t s)ds ei2f t dt y T (t) = k 1 1 B sin f A y(t)(r=ts) |f = k |f = k ei2 T t . T 1 + i2f RC T T f k = x(s)h(r)ei2f (r+s) dsdr T Notice, as T , then y (t) y(t) with = X(f )H(f ) 1 B sin f A ei2f t y(t) = df f 1 + i2f RC Example XIV.2. LTI system y(t) = ay(t) + at x(t), then h(t) = e u(t) and let input x(t) = B, t [ A , A ]. Then 2 2

y(t) =

X(f )H(f )ei2f t df B sin f A 1 ei2f t df f a + i2f

Y (f )

= x(t) h(t)

22

XV. R EMIND O URSELVES OF W HAT W E K NOW Lets collect some results.

XVI. T HE D UALITY P ROPERTY Let us try to understand the following relationships. Suppose we know that

B sin f A A A B sin f A A A xT (t) = B, t [ , ], 0otherwise, , x(t) = B, t [ , ] 2 2 f 2 2 f B sin f A i2f t e df x(t) = then what is B sin tA ? t f A A xT (t) = B, t [ , ], xT (t) = xT (t + mT )Proposition XVI.1. 2 2 1 B sin f A x(t) X(f ) then |f =kf0 ei2kf0 t xT (t) = T f k X(t) x(f ) . 1 at x(t) = e u(t) a + i2f 1 Proof of : X(t) x(f ) ei2f t df eat u(t) = a + i2f Beginning with x(t) X(f ) implies y(t) = ay(t) + ei2f0 t at rest then 1 y(t) = ei2f0 t x(t) = X(f )ei2f t df = X(r)ei2rt dr. a + i2f0 (t) 1 x(t t0 ) x(t t0 ) = (t) = 1ei2f t df Now substituting f for the variable t gives x(f ) = X(r)ei2rf dr ,

X(f )e2f t0 X(f )e2f t0 ei2f t df

and changing the dummy variable r to t gives the Fourier transform pair X(t) x(f ) B sin f A i2f A 2 e x(t t0 ) = B, t [0, A] according to f B sin f A i2f A i2f t 2 e df e x(t) = x(f ) = X(t)ei2tf dt . f Example XVI.1. 1 A A B sin f A x(t) = B, t [ , ] T 2 2 f B sin tA A A x(f ) = B, f [ , ] Xk H(kf0 )ei2kf0 t t 2 2 Xk H(kf0 ) f0 = e.g. x(t) = B, t [0, A] B sin tA t
A

Periodic with period T: y T (t) = xT (t) h(t)

k

y (t) = x (t) h(t) =

k

y(t) = x(t) h(t) X(f )H(f ) y(t) = x(t) h(t) = X(f )H(f )ei2f t df

B sin f A i2f A 2 e f

ei2t 2 x(f ) = B, f [0, A]

23

Example XVI.2. x(t) = eat u(t) 1 a + i2f

XVII. T HE M ODULATION P ROPERTY Now we have already seen convolution results in multiplication in the frequency representations. How about multiplication in time? Proposition XVII.1. x(t) X(f ) x(t) ei2f0 t X(f f0 ) Proof: x(t)ei2f0 t ei2f t dt = x(t)ei2(f f0 )t

1 x(f ) = eaf u(f ) a + i2t (t) 1 1 (f ) (t + t0 ) e+i2f t0 e

+i2tf0

((f + f0 )) = (f f0 )

(t t0 ) ei2f t0 ei2tf0 ((f f0 )) = (f + f0 )

= X(f f0 ) . Proposition XVII.2. x(t) X(f ) , y(t) Y (f )

then z(t) = x(t) y(t) Z(f ) = X(f ) H(f ) Proof: z(t) = = = X(r)Y (f r)drei2f t df X(r)Y (s)drei2(r+s)t ds X(r)ei2r dr Y (s)ei2s ds = x(t) y(t) .