Structure and Interpretation of Signals and Systems by Edward Ashford Lee and Pravin Varaiya - HTML preview

PLEASE NOTE: This is an HTML preview only and some elements such as links or page numbers may be incorrect.
Download the book in PDF, ePub for a complete version.

Exponential

{s | s − a ∈ RoC(x)}

scaling

(Exercise 11)

t

y(t) = R x(τ)dτ

ˆ

Y (s) = ˆ

X (s)/s

RoC(y) ⊃

Integration

−∞

RoC(x) ∩ {s | Re{s} > 0} (Section 13.3.1)

ˆ

Y (s) = s ˆ

X (s)

RoC(y) ⊃ RoC(x)

Differentiation

(page 13.5.2)

d

y(t) =

x(t)

dt

Table 13.4: Properties of the Laplace transform. In this table, a, b are complex

constants, c and τ are real constants.

552

Lee & Varaiya, Signals and Systems

13. LAPLACE AND Z TRANSFORMS

13.3.1

Integration

Let y be defined by

t

Z

∀ t ∈ R, y(t) =

x(τ)dτ.

−∞

The Laplace tranform is

∀ s ∈ RoC(y), ˆY(s) = ˆX(s)/s,

where

RoC(y) ⊃ RoC(x) ∩ {s | Re{s} > 0}.

This follows from the convolution property in table 13.4. We recognize that

y(t) = (x ∗ u)(t),

where u is the unit step. Hence, from the convolution property,

ˆ

Y (s) = ˆ

X (s) Û (s)

and

RoC(y) ⊃ RoC(x) ∩ RoC(u).

Û and RoC(u) are given in example 12.15, from which the property follows.

13.3.2

Sinusoidal signals

Sinusoidal signals have Laplace transforms with poles on the imaginary axis, as illustrated

in the following example.

Example 13.10: Let the causal sinusoidal signal y be given by

∀ t ∈ R, y(t) = sin(ω0t)u(t),

where ω0 is a real number and u is the unit step. Euler’s relation implies that

1

y(t) =

[eiω0t u(t) − e−iω0tu(t)].

2i

Lee & Varaiya, Signals and Systems

553

13.3. PROPERTIES OF THE LAPLACE TRANSFORM

Using (12.18) and linearity,

1

1

ˆ

Y (s)

=

1

2i

s + iω0

s − iω0

ω0

=

.

s2 + 2

ω0

This has no finite zeros and two poles, one at s = iω0 and the other at s = −iω0.

Both of these poles lie on the imaginary axis, as shown in Figure 13.3. The region

of convergence is the right half of the complex plane. Note that if this were the

impulse response of an LTI system, that system would not be stable. The region of

convergence does not include the imaginary axis.

13.3.3

Differential equations

We can use the differentiation property in table 13.4 to solve differential equations with

constant coefficients.

Example 13.11: In the tuning fork example of example 2.16, the displacement y

of a tine is related to the acceleration of the tine by

¨

y(t) = − 2

ω0y(t),

where ω0 is a real constant. Let us assume that the tuning fork is initially at rest,

and an external input x (representing say, a hammer strike) adds to the acceleration

as follows,

¨

y(t) = − 2

ω0y(t) + x(t).

We can use Laplace transforms to find the impulse response of this LTI system.

Taking Laplace transforms on both sides, using linearity and the differentiation

property,

∀ s ∈ RoC(y) ∩ RoC(x), s2 ˆY(s) = − 2

ω ˆ

0Y (s) + ˆ

X (s).

From this, we can find the transfer function of the system,

ˆ

Y (s)

1

ˆ

H(s) =

=

.

ˆ

X (s)

s2 + 2

ω0

554

Lee & Varaiya, Signals and Systems

13. LAPLACE AND Z TRANSFORMS

Comparing this with example 13.10, we see that this differs only by a scaling by ω0

from the Laplace transform in that example. Thus, the pole-zero plot of the tuning

fork is shown in Figure 13.3, and the impulse response is given by

∀ t ∈ R, h(t) = sin(ω0t)u(t)/ω0.

Interestingly, this implies that the tuning fork is not stable. This impulse response

is not absolutely integrable. However, this model of the tuning fork is idealized. It

fails to account for loss of energy due to friction. A more accurate model would be

stable.

The above example can be easily generalized to find the transfer function of any LTI

system described by a differential equation. In fact, Laplace transforms offer a powerful

and effective way to solve differential equations.

In the previous example, we inverted the Laplace transform by recognizing that it matched

the example before that. In the next chapter, we will give a more general method for

inverting a Laplace transform.

13.4

Frequency response and pole-zero plots

Just as with Z transforms, the pole-zero plot of a Laplace transform can be used to get a

quick estimate of key properties of an LTI system. Consider a stable continuous-time LTI

system with impulse response h, frequency response H, and rational transfer function ˆ

H.

We know that the frequency response and transfer function are related by

∀ω ∈ R, H(ω) = ˆH(iω).

That is, the frequency response is equal to the Laplace transform evaluated on the imag-

inary axis. The imaginary axis is in the region of convergence because the system is

stable.

Assume that ˆ

H is a rational polynomial, in which case we can express it in terms of the

first-order factors of the numerator and denominator polynomials,

(s − q

ˆ

1) · · · (s − qM )

H(s) =

,

(s − p1) · · · (s − pN)

Lee & Varaiya, Signals and Systems

555

13.4. FREQUENCY RESPONSE AND POLE-ZERO PLOTS

with zeros at q1, · · · , qM and poles at p1, · · · , pN. The zeros and poles may be repeated

(i.e., they may have multiplicity greater than one). The frequency response is therefore

(iω − q1) · · · (iω − qM)

ω ∈ R,

H(ω) =

.

(iω − p1) · · · (iω − pN)

The magnitude response is

|

iω − q1| · · · |iω − qM|

ω ∈ R,

|H(ω)| =

.

|iω − p1|···|iω − pN|

Each of these factors has the form

|iω − b|

where b is the location of either a pole or a zero. The factor |iω − b| is just the distance

from iω to b in the complex plane.

Of course, iω is a point on the imaginary axis. If that point is close to a zero at location

q, then the factor |iω − q| is small, so the magnitude response will be small. If that point

is close to a pole at p, then the factor |iω − p| is small, but since this factor is in the

denominator, the magnitude response will be large. Thus,

The magnitude response of a stable LTI system may be estimated from the pole-zero

plot of its transfer function. Starting at iω = 0, trace upwards and downwards along the

imaginary axis to increase or decrease ω. If you pass near a zero, then the magnitude

response should dip. If you pass near a pole, then the magnitude response should rise.

Example 13.12: Consider an LTI system with transfer function given by

s

s ∈ {s | Re{s} > Re{a}},

H(s) =

.

(s − a)(s − a∗)

Suppose that a = c + iω0. Figure 13.4 shows three pole-zero plots for ω0 = 1

and three values of c, namely c = −1, c = −0.5, and c = −0.1. The magnitude

frequency responses can be calculated and plotted using the following Matlab code:

1

w = [-10:0.05:10];

2

a1 = -1.0 + i;

3

H1 = i*w./((i*w - a1).*(i*w-conj(a1)));

4

a2 = -0.5 + i;

556

Lee & Varaiya, Signals and Systems

13. LAPLACE AND Z TRANSFORMS

5

H2 = i*w./((i*w - a2).*(i*w-conj(a2)));

6

a3 = -0.1 + i;

7

H3 = i*w./((i*w - a3).*(i*w-conj(a3)));

8

plot(w, abs(H1), w, abs(H2), w, abs(H3));

The plots are shown together at the bottom of Figure 13.4. The plot with the higher

peaks corresponds to the pole-zero plot with the poles closer to the imaginary axis.

13.5

The inverse transforms

There are two inverse transforms. The inverse Z transform recovers the discrete-time sig-

nal x from its Z transform ˆ

X . The inverse Laplace transform recovers the continuous-time

signal x from its Laplace transform ˆ

X . We study the inverse Z transform in detail. The

inverse Laplace transform is almost identical. The general approach is to break down

a complicated rational polynomial into a sum of simple rational polynomials whose in-

verse transforms we recognize. We consider only the case where ˆ

X can be expressed as a

rational polynomial.

13.5.1

Inverse Z transform

The procedure is to construct the partial fraction expansion of ˆ

X , which breaks it down

into a sum of simpler rational polynomials.

Example 13.13: Consider a Z transform given by

1

1

1

z ∈ RoC(x),

ˆ

X (z) =

=

+

.

(13.6)

(z − 1)(z − 2)

z − 1

z − 2

This sum is called the partial fraction expansion of ˆ

X , and we will see below how

to find it systematically. We can write this as

∀ z ∈ RoC(x),

ˆ

X (z) = ˆ

X1(z) + ˆ

X2(z),

Lee & Varaiya, Signals and Systems

557

13.5. THE INVERSE TRANSFORMS

where ˆ

X1(z) = −1/(z − 1) and ˆ

X2(z) = 1/(z − 2) are the two terms.

To determine the inverse Z transforms of the two terms, we need to know their

regions of convergence. Recall from the linearity property that RoC(x) includes the

intersection of the regions of convergence of the two terms,

RoC(x) ⊃ RoC(x1) ∩ RoC(x2).

(13.7)

Once we know these two regions of convergence, we can use table 13.1 to obtain

the inverse Z transform of each term. By the linearity property the sum of these

inverses is the inverse Z transform of ˆ

X .

ˆ

X given by (13.6) has one pole at z = 1 and one pole at z = 2. From section 12.2.3

we know that RoC(x) is bordered by these poles, so it has one of three forms:

1. RoC(x) = {z ∈ C | |z| < 1},

2. RoC(x) = {z ∈ C | 1 < |z| < 2}, or

3. RoC(x) = {z ∈ C | |z| > 2}.

Suppose we have case (1), which implies that x is anti-causal. From (13.7), the

region of convergence of the term −1/(z − 1) must be {z ∈ C | |z| < 1}. The

only other possibility is {z ∈ C | |z| > 1}, which would violate (13.7) unless the

intersection is empty (which would not be an interesting case). Thus, from table

13.1, the inverse Z transform of the first term must be the anti-causal signal x1(n) =

u(−n), for all n ∈ Z.

For the second term, 1/(z − 2), its region of convergence could be either {z ∈

C | |z| < 2} or {z ∈ C | |z| > 2}. Again, the second possibility would violate

(13.7), so we must have the first possibility. This results in x2(n) = −2n−1u(−n),

from the last entry in table 13.1. Hence, the inverse Z transform is

∀ n ∈ Z, x(n) = u(−n) − 2n−1u(−n).

If RoC(x) is given by case (2), we rewrite (13.6) slightly as

z

1

ˆ

X (z) = −z−1

+

.

z − 1

z − 2

The inverse Z transform of the first term is obtained from table 13.1, together with

the delay property in table 13.2. The inverse Z transform of the second term is the

558

Lee & Varaiya, Signals and Systems

13. LAPLACE AND Z TRANSFORMS

same as in case (1). We conclude that in case (2) the inverse Z transform is the

two-sided signal

∀n, x(n) = −u(n − 1) − 2n−1u(−n).

In case (3), we write (13.6) as

z

z

ˆ

X (z) = −z−1

+ z−1

,

z − 1

z − 2

and conclude that the inverse Z transform is the causal signal

∀n, x(n) = −u(n − 1) + 2n−1u(n − 1).

We can generalize this example. Consider any strictly proper rational polynomial

A(z)

a

ˆ

M zM + · · · + a1z + a0

X (z) =

=

.

B(z)

zN + bN−1zN−1 + · · · + b1z + b0

The numerator is of order M, the denominator is of order N. ‘Strictly proper’ means that

M < N. We can factor the denominator,

a

ˆ

M zM + · · · + a1z + a0

X (z) =

.

(13.8)

(z − p1)m1(z − p2)m2 · · · (z − pk)mk

Thus ˆ

X has k distinct poles at pi, each with multiplicy mi. Since the order of the denomi-

nator is N, it must be true that

k

N = ∑ mi .

(13.9)

i=1

The partial fraction expansion of (13.8) is

k

R

R

R

î1

i2

im

X (z) =

i

+

+ · · · +

.

(13.10)

(z − p

(z − p

(z − p

i=1

i)

i)2

i)mi

A pole with multiplicity mi contributes mi terms to the partial fraction expansion, so the

total number of terms is N, the order of the denominator, from (13.9). The coefficients Ri j

are complex numbers called the residues of the pole pi.

We assume that the poles p1, · · · , pN are indexed so that |p1| ≤ · · · |pN|. The RoC(x) must

have one of the following three forms:

Lee & Varaiya, Signals and Systems

559

13.5. THE INVERSE TRANSFORMS

1. RoC = {z ∈ C | |z| < |p1|},

2. RoC = {z ∈ C | |p j−1| < |z| < |p j|}, for j ∈ {2, · · · , k}, or

3. RoC = {z ∈ C | |z| > |pk|} .

As in example 13.13, each term in the partial fraction expansion has two possible regions

of convergence, only one of which overlaps with RoC(x). Thus, if we know RoC(x), we

can determine the region of convergence of each term of the partial fraction expansion,

and then use table 13.1 to find its inverse.

The following example illustrates how to find the residues.

Example 13.14: We will find the inverse Z transform of

2z + 3

R

R

ˆ

1

2

X (z) =

=

+

.

(z − 1)(z + 2)

z − 1

z + 2

The residues R1, R2 can be found by matching coefficients on both sides. Rewrite

the right-hand side as

(R1 + R2)z + (2R1 − R2) .

(z − 1)(z + 2)

Matching the coefficients of the numerator polynomials on both sides we conclude

that R1 + R2 = 2 and 2R1 − R2 = 3. We can solve these simultaneous equations to

determine that R1 = 5/3 and R2 = 1/3.

Alternatively, we can find residue R1 by multiplying both sides by (z − 1) and eval-

uating at z = 1. That is,

2z + 3

5

R1 =

=

.

z + 2

3

z=1

Similarly, we can find R2 by we multiplying both sides by z + 2 and evaluating at

z = −2, to get

2z + 3 |

= R2,

z − 1 z=−2

so R2 = 1/3. Thus the partial fraction expansion is

5/3

1/3

ˆ

X (z) =

+

.

z − 1

z + 2

RoC(x) is either

560

Lee & Varaiya, Signals and Systems

13. LAPLACE AND Z TRANSFORMS

1. {z ∈ C | |z| < 1},

2. {z ∈ C | 1 < |z| < 2}, or

3. {z ∈ C | |z| > 2}.

Knowing which case holds, we can find the inverse Z transform of each term from

table 13.1. In the first case, x is the anti-causal signal

∀n, x(n) = −5u(−n) − 1(−2)n−1u(−n).

3

3

In the second case it is the two-sided signal

5

n,

x(n) =

u(n − 1) − 1 (−2)n−1u(−n).

3

3

In the third case it is the causal signal

5

1

n,

x(n) =

u(n − 1) + (−2)n−1u(n − 1).

3

3

If some pole of ˆ

X has multiplicity greater than one, it is slightly more difficult to carry out

the partial fraction expansion. The following example illustrates the method.

Example 13.15: Consider the expansion

2z + 3

R

R

R

ˆ

1

21

22

X (z) =

=

+

+

.

(z − 1)(z + 2)2

z − 1

z + 2

(z + 2)2

Again we can match coefficients and determine the residues. Alternatively, to ob-

tain R1 we multiply both sides by (z − 1) and evaluate the result at z = 1, to get

R1 = 5/9. To obtain R22 we multiply both sides by (z + 2)2 and evaluate the result

at z = −2, to get R22 = 1/3.

To obtain R21 we multiply both sides by (z + 2)2,

2z + 3

(z + 2)2R1

=

+ R21(z + 2) + R22,

z − 1

z − 1

Lee & Varaiya, Signals and Systems

561

13.5. THE INVERSE TRANSFORMS

and then differentiate both sides with respect to z. We evaluate the result at z = −2,

to get

d 2z + 3

= R21.

dz z − 1 z=−2

Hence R21 = −5/9. So the partial fraction expansion is

2z + 3

5/9

1/3

=

− 5/9 +

.

(z − 1)(z + 2)2

z − 1

z + 2

(z + 2)2

Knowing the RoC, we can now obtain the inverse Z transform of ˆ

X . For instance,

in the case where RoC = {z ∈ C | |z| > 2}, the inverse Z transform is the causal

signal

5

1

n,

x(n) =

u(n − 1) − 5 (−2)n−1u(n − 1) + (n − 1)(−2)n−2u(n − 2).

9

9

3

In example 13.15, we used the next to the last entry in table 13.1 to find the inverse transform of the term (1/3)/(z + 2)2. That entry in the table is based on a generalization

of the geometric series identity, given by (12.9). The first generalization is

1

∑ (n + 1)an = ( ∑ an)2 =

.

(13.11)

(1 − a)2

n=0

n=0

The series above converges for any complex number a with |a| < 1 (see Exercise 3 of

Chapter 12). The broader generalization, for any integer k ≥ 1, is

1 ∞

1

∑ (n + k)(n + k − 1) · · · (n + 1)an =

,

(13.12)

k!

(1 − a)k+1

n=0

for any complex number a with |a| < 1.

Consider then a Z transform ˆ

X that has a pole at p of multiplicity m and no zeros. Since

the pole p cannot belong to RoC, the RoC is either

{z ∈ C | |z| > |p|} or {z ∈ C | |z| < |p|}.

562

Lee & Varaiya, Signals and Systems

13. LAPLACE AND Z TRANSFORMS

In the first case we expand ˆ

X in a series involving only the terms z−n, n ≥ 0,

1

ˆ

X (z)

=

(z − p)m

z−m

=

(1 − pz−1)m

1

=

z−m

(m + n − 1) · · · (n + 1)(pz−1)n, using (13.12)

(

m − 1)! n=0

1

=

(k − 1) · · · (k − m + 1)pk−mz−k, defining k = n + m,

(

m − 1)! k=m

and the series converges for any z with |z| > |p|. We can match the coefficients of the

powers of z in the Z transform definition,

ˆ

X (z) = ∑ x(k)z−k,

k=−∞

from which we can recognize that

0,

k < m

k ∈ Z,

x(k)

=

1

(k − 1) · · · (k − m + 1)pk−m, k ≥ m

(m−1)!

1

=

(k − 1) · · · (k − m + 1)pk−mu(k − m).

(m − 1)!

(13.13)

In the second case, RoC