N5625 2

2
Chapter
Signal and Linear System Analysis

Contents
2.1
Signal Models . . . . . . . . . . . . . . . . . . . . . .
2-3
2.1.1
Deterministic and Random Signals . . . . . . . .
2-3
2.1.2
Periodic and Aperiodic Signals . . . . . . . . . .
2-3
2.1.3
Phasor Signals and Spectra . . . . . . . . . . . .
2-4
2.1.4
Singularity Functions . . . . . . . . . . . . . . .
2-7
2.2
Signal Classifications . . . . . . . . . . . . . . . . . . 2-11
2.3
Generalized Fourier Series . . . . . . . . . . . . . . . 2-14
2.4
Fourier Series . . . . . . . . . . . . . . . . . . . . . . 2-20
2.5
2.4.1
Complex Exponential Fourier Series . . . . . . . 2-20
2.4.2
Symmetry Properties of the Fourier Coefficients
2.4.3
Trigonometric Form . . . . . . . . . . . . . . . 2-25
2.4.4
Parsevals Theorem . . . . . . . . . . . . . . . . 2-26
2.4.5
Line Spectra . . . . . . . . . . . . . . . . . . . 2-26
2.4.6
Numerical Calculation of Xn . . . . . . . . . . . 2-30
2.4.7
Other Fourier Series Properties . . . . . . . . . . 2-37
2-23
Fourier Transform . . . . . . . . . . . . . . . . . . . . 2-38

2.5.1
Amplitude and Phase Spectra . . . . . . . . . . 2-39
2-1
CHAPTER 2. SIGNAL AND LINEAR SYSTEM ANALYSIS
2.6
2.7
2.5.2
Symmetry Properties . . . . . . . . . . . . . . . 2-39
2.5.3
Energy Spectral Density . . . . . . . . . . . . . 2-40
2.5.4
Transform Theorems . . . . . . . . . . . . . . . 2-42
2.5.5
Fourier Transforms in the Limit . . . . . . . . . 2-51
2.5.6
Fourier Transforms of Periodic Signals . . . . . 2-53
2.5.7
Poisson Sum Formula . . . . . . . . . . . . . . 2-59
Power Spectral Density and Correlation . . . . . . . . 2-60

2.6.1
The Time Average Autocorrelation Function . . 2-61
2.6.2
Power Signal Case . . . . . . . . . . . . . . . . 2-62
2.6.3
Properties of R./ . . . . . . . . . . . . . . . . 2-63
Linear Time Invariant (LTI) Systems . . . . . . . . . 2-74

2.7.1
Stability . . . . . . . . . . . . . . . . . . . . . . 2-76
2.7.2
Transfer Function . . . . . . . . . . . . . . . . . 2-76
2.7.3
Causality . . . . . . . . . . . . . . . . . . . . . 2-77
2.7.4
Properties of H.f / . . . . . . . . . . . . . . . . 2-78
2.7.5
Input/Output with Spectral Densities . . . . . . . 2-82
2.7.6
Response to Periodic Inputs . . . . . . . . . . . 2-82
2.7.7
Distortionless Transmission . . . . . . . . . . . 2-83
2.7.8
Group and Phase Delay . . . . . . . . . . . . . . 2-84
2.7.9
Nonlinear Distortion . . . . . . . . . . . . . . . 2-87
2.7.10 Ideal Filters . . . . . . . . . . . . . . . . . . . . 2-89

2.7.11 Realizable Filters . . . . . . . . . . . . . . . . . 2-91
2.7.12 Pulse Resolution, Risetime, and Bandwidth . . . 2-101
2.8
Sampling Theory . . . . . . . . . . . . . . . . . . . . . 2-106
2.9
The Hilbert Transform . . . . . . . . . . . . . . . . . 2-106
2.10 The Discrete Fourier Transform and FFT . . . . . . . 2-106
2-2
ECE 5625 Communication Systems I
2.1. SIGNAL MODELS
2.1
Signal Models
2.1.1
Deterministic and Random Signals
Deterministic Signals, used for this course, can be modeled as

completely specified functions of time, e.g.,
x.t / D A.t / cos2f0.t /t C .t /
Note that here we have also made the amplitude, frequency, and phase functions of time
To be deterministic each of these functions must be completely specified functions of time
Random Signals, used extensively in Comm Systems II, take
on random values with known probability characteristics, e.g.,
x.t / D x.t; i /
where i corresponds to an elementary outcome from a sample
space in probability theory
The i create a ensemble of sample functions x.t; i /, depending upon the outcome drawn from the sample space
2.1.2
Periodic and Aperiodic Signals
A deterministic signal is periodic if we can write

x.t C nT0/ D x.t /
for any integer n, with T0 being the signal fundamental period
2-3
A signal is aperiodic otherwise, e.g.,

(
1; jt j 1=2
.t / D
0; otherwise
(a) periodic signal, (b) aperiodic signal, (c) random signal
2.1.3
Phasor Signals and Spectra
A complex sinusoid can be viewed as a rotating phasor

x.t
Q / D Ae j.!0tC/;
1<t <1
This signal has three parameters, amplitude A, frequency !0,

and phase
The fixed phasor portion is Ae j while the rotating portion is
e j!0t
2-4
2.1. SIGNAL MODELS
This signal is periodic with period T0 D 2=!0

It also related to the real sinusoid signal A cos.!0t C / via
Eulers theorem

x.t / D Re x.t
Q /

D Re A cos.!0t C / C jA sin.!0t C /
D A cos.!0t C /
(a) obtain x.t/ from x.t/,

Q
(b) obtain x.t/ from x.t/
Q
and xQ .t/
We can also turn this around using the inverse Euler formula
x.t / D A cos.!0t C /
1
1
D x.t
Q / C xQ .t /
2
2
j.!0 t C/
Ae
C Ae
D
2
j.!0 tC/
The frequency spectra of a real sinusoid is the line spectra plotted in terms of the amplitude and phase versus frequency
2-5
The relevant parameters are A and for a particular f0 D

!0=.2/
(a) Single-sided line spectra, (b) Double-sided line spectra
Both the single-sided and double-sided line spectra, shown

above, correspond to the real signal x.t / D A cos.2f0t C /
Example 2.1: Multiple Sinusoids

Suppose that
x.t / D 4 cos.2.10/t C =3/ C 24 sin.2.100/t
=8/
Find the two-sided amplitude and phase line spectra of x.t /

First recall that cos.!0t
=2/ D sin.!0t /, so
x.t / D 4 cos.2.10/t C =3/ C 24 cos.2.100/t
5=8/
The complex sinusoid form is directly related to the two-sided

line spectra since each real sinusoid is composed of positive
and negative frequency complex sinusoids
h
i
j.2.10/tC=3/
j.2.10/tC=3/
x.t / D 2 e
Ce
h
i
j.2.100/t 5=8/
j.2.100/t 5=8/
C 12 e
Ce
2-6
2.1. SIGNAL MODELS
Amplitude
12
2
f (Hz)
-100
-10
100
10
5/8
Phase
/3
f (Hz)
-/3
-5/8
Two-sided amplitude and phase line spectra
2.1.4
Singularity Functions
Unit Impulse (Delta) Function

Singularity functions, such as the delta function and unit step
The unit impulse function, .t / has the operational properties
Z t2
.t t0/ dt D 1; t1 < t0 < t2
t1
.t
t0/ D 0; t t0
which implies that for x.t / continuous at t D t0, the sifting

property holds
Z 1
x.t /.t t0/ dt D x.t0/
1
Alternatively the unit impulse can be defined as

Z 1
x.t /.t / dt D x.0/
1
2-7
Properties:
1. .at / D .t /=jaj
2. . t / D .t /
3. Sifting property special cases
8
t1 < t0 < t2
Z t2
<x.t0/;
x.t /.t t0/ dt D 0;
otherwise
t1
:undefined; t0 D t1 or t0 D t2
4. Sampling property
x.t /.t
t0/ D x.t0/.t
t0/
for x.t / continuous at t D t0

5. Derivative property
Z t2
x.t / .n/.t
t1
t0/ dt D . 1/nx .n/.t0/
d
x.t
/
D . 1/n
n
dt tDt0
Note: Dealing with the derivative of a delta function requires care

A test function for the unit impulse function helps our intuition
and also helps in problem solving
Two functions of interest are
(1
;
1
t
.t / D
D 2
2
2
0;

1
t 2
1 .t / D
sin
t

2-8
jt j
otherwise
2.1. SIGNAL MODELS
Test functions for the unit impulse .t/: (a) .t/, (b) 1 .t/
In both of the above test functions letting ! 0 results in a

function having the properties of a true delta function
Unit Step Function
The unit step function can be defined in terms of the unit impulse
8
t <0
Z t
<0;
u.t /
. / d D 1;
t >0
:undefined; t D 0
also
.t / D
du.t /
dt
2-9
Example 2.2: Unit Impulse 1st-Derivative

Consider
x.t / 0.t / dt
Using the rectangular pulse test function, .t /, we note that

1
t also 1
.t / D
D
u.t C / u.t /
2
2
2
and
1
d .t /
D
.t C /
dt
2
.t

/
Placing the above in the integrand with x.t / we obtain, with

the aid of the sifting property, that
Z 1

1
x.t / 0.t / dt D lim
x.t C / x.t /
!0 2
1

x.t / x.t C /
D lim
!0
2
D x 0.0/
2-10
2.2. SIGNAL CLASSIFICATIONS
2.2
Signal Classifications
From circuits and systems we know that a real voltage or current waveform, e.t / or i.t / respectively, measured with respective a real resistance R, the instantaneous power is
P .t / D e.t /i.t / D i 2.t /R W
On a per-ohm basis, we obtain
p.t / D P .t /=R D i 2.t / W/ohm
The average energy and power can be obtain by integrating
over the interval jt j T with T ! 1
Z T
i 2.t / dt Joules/ohm
E D lim
T !1
T
Z T
1
P D lim
i 2.t / dt W/ohm
T !1 2T
T
In system engineering we take the above energy and power
definitions, and extend them to an arbitrary signal x.t /, possibly complex, and define the normalized energy (e.g. 1 ohm
system) as
E D lim
T
2
jx.t /j dt D
T !1
T
Z T
1
P D lim
jx.t /j2 dt
T !1 2T
T
jx.t /j2 dt
2-11
Signal Classes:
1. x.t / is an energy signal if and only if 0 < E < 1 so that
P D0
2. x.t / is a power signal if and only if 0 < P < 1 which
implies that E ! 1
Example 2.3: Real Exponential

Consider x.t / D Ae
u.t / where is real
For > 0 the energy is given by
Z 1
2 2t 1

A
e
2
ED
Ae t dt D
2 0
0
A2
D
2
For D 0 we just have x.t / D Au.t / and E ! 1
For < 0 we also have E ! 1
In summary, we conclude that x.t / is an energy signal for >
0
For > 0 the power is given by
1 A2
P D lim
1
T !1 2T 2
D0
For D 0 we have
1
A2
2
P D lim
A T D
T !1 2T
2
2-12
2.2. SIGNAL CLASSIFICATIONS
For < 0 we have P ! 1

In summary, we conclude that x.t / is a power signal for D 0
Example 2.4: Real Sinusoid

Consider x.t / D A cos.!0t C /;
1<t <1
The signal energy is infinite since upon squaring, and integrating over one cycle, T0 D 2=!0, we obtain
Z N T0=2
A2 cos2.!0t C / dt
E D lim
N !1
D lim N
N !1
N T0 =2
Z T0=2
A2 cos2.!0t C / dt
T0 =2
2 Z T0 =2

A
1 C cos.2!0t C 2 / dt
N !1
2
T0 =2
A2
D lim N
T0 ! 1
N !1
2
D lim N
The signal average power is finite since the above integral is

normalized by 1=.N T0/, i.e.,
A2
1
A2
P D lim
N
T0 D
N !1 N T0
2
2
2-13
2.3
Generalized Fourier Series
The goal of generalized Fourier series is to obtain a representation of

a signal in terms of points in a signal space or abstract vector space.
The coordinate vectors in this case are orthonomal functions. The
complex exponential Fourier series is a special case.
Let AE be a vector in a three dimensional vector space
Let aE1; aE2, and aE3 be linearly independent vectors in the same
three dimensional space, then
c1aE1 C c2aE2 C c3aE3 D 0 .zero vector/
only if the constants c1 D c2 D c3 D 0
The vectors aE1; aE2, and aE3 also span the three dimensional
space, that is for any vector AE there exists a set of constants
c1; c2, and c3 such that
AE D c1aE1 C c2aE2 C c3aE3
The set fE
a1; aE2; aE3g forms a basis for the three dimensional
space
Now let fE
a1; aE2; aE3g form an orthogonal basis, which implies
that
aEi aEj D .E
ai ; aEj / D hE
ai ; aEj i D 0; i j
which says the basis vectors are mutually orthogonal
From analytic geometry (and linear algebra), we know that we
can find a representation for AE as
E
E
E
.E
a2 A/
.E
a3 A/
.E
a1 A/
E
C
C
AD
jE
a1j2
jE
a 2 j2
jE
a 3 j2
2-14
2.3. GENERALIZED FOURIER SERIES
which implies that

AE D
3
X
ci aEi
i D1
where
aEi AE
; i D 1; 2; 3
ci D
jE
ai j2
is the component of AE in the aEi direction
We now extend the above concepts to a set of orthogonal functions f1.t /; 2.t /; : : : ; N .t /g defined on to t t0 C T ,
where the dot product (inner product) associated with the ns
is
Z t0CT

m.t /; n.t / D
m.t /n.t / dt
t0
(
cn; n D m
D cnmn D
0; n m
The ns are thus orthogonal on the interval t0; t0 C T
Moving forward, let x.t / be an arbitrary function on t0; t0CT ,
and consider approximating x.t / with a linear combination of
ns, i.e.,
x.t / ' xa .t / D
N
X
Xnn.t /; t0 t t0 C T;
nD1
where a denotes approximation

2-15
A measure of the approximation error is the integral squared

error (ISE) defined as
Z
N D
x.t / xa .t / dt;
T
where
R
T
denotes integration over any T long interval
To find the Xns giving the minimum N we expand the above

integral into three parts (see homework problems)
Z
jx.t /j2 dt
N D
T
N
X
nD1
cn Xn
2
Z
N
X
1
.t
/
dt
x.t
/
n
c
n
T
nD1
2
Z
1

x.t /n .t / dt
cn T
Note that the first two terms are independent of the Xns
and the last term is nonnegative (missing steps are in text
homework problem 2.14)
We conclude that N is minimized for each n if each element
of the last term is made zero by setting
Z
1
Xn D
x.t /n.t / dt Fourier Coefficient
cn T
This also results in
N
2-16

min
Z
D
T
jx.t /j2 dt
N
X
cnjXnj2
nD1
Definition: The set of of ns is complete if

lim .N /min D 0
N !1
for
R
T
jx.t /j2 dt < 1
Even if though the ISE is zero when using a complete

set of orthonormal functions, there may be isolated points
where x.t / xa .t / 0
Summary
x.t / D l.i.m.
1
X
Xnn.t /
nD1
Xn D
1
cn
Z
T
x.t /n.t / dt
The notation l.i.m. stands for limit in the mean, which is

a mathematical term referring to the fact that ISE is the
convergence criteria
Parsevals theorem: A consequence of completeness is
Z
1
X
jx.t /j2 dt D
cnjXnj2
T
nD1
2-17
Example 2.5: A Three Term Expansion

Approximate the signal x.t / D cos 2 t on the interval 0; 1
using the following basis functions
1(t)
x(t)
0.75
0.75
0.5
0.5
0.25
0.25
0.2
0.4
0.6
0.8
-0.25
-0.25
-0.5
-0.5
-0.75
-0.75
-1
-1
2(t)
0.4
0.6
0.8
0.2
0.4
0.6
0.8
3(t)
0.75
0.75
0.5
0.5
0.25
0.25
0.2
0.2
0.4
0.6
0.8
-0.25
-0.25
-0.5
-0.5
-0.75
-0.75
-1
-1
Signal x.t/ and basis functions i .t/; i D 1; 2; 3
To begin with it should be clear that 1.t /; 2.t /, and 3.t /

are mutually orthogonal since the integrand associated with the
inner product, i .t / j.t / D 0, for i j; i; j D 1; 2; 3
2-18
Before finding the Xns we need to find the cns

Z
Z 1=4
c1 D
j1.t /j2 dt
j1j2 dt D 1=4
0
ZT
c2 D
j2.t /j2 dt D 1=2
ZT
c3 D
j3.t /j2 dt D 1=4
T
Now we can compute the Xns:

Z
X1 D 4 x.t /1.t / dt
ZT 1=4
1=4
2
2
D4
cos.2 t / dt D sin.2 t / D
0

Z0 3=4
3=4
1
2
X2 D 2
1=4

1=4
Z 1
1
2
2
X3 D 4
3=4

3=4
1
x(t)
0.75
2/
xa(t)
0.5
0.25
0.2
0.4
0.6
0.8
-0.25
-0.5
-2/
-0.75
-1
Functional approximation
2-19
The integral-squared error, N , can be computed as follows:

3
X
x.t /
N D
nD1
2
jx.t /j dt
T
1
D
2
1
D
2
2.4
Xnn.t / dt
3
X
cnjXnj2
nD1

1 2 2 1 2 2
4
2
2
2
D 0:0947

1 2 2
4
Fourier Series
When we choose a particular set of basis functions we arrive at the

more familiar Fourier series.
2.4.1
Complex Exponential Fourier Series
A set of ns that is complete is

n.t / D e j n!0t ; n D 0; 1; 2; : : :
over the interval .t0; t0 C T0/ where !0 D 2=T0 is the period
of the expansion interval
2-20
2.4. FOURIER SERIES
proof of orthogonality
Z t0CT0
Z t0CT0
t
2 t

j m 2
j
n
j 2
.m n/t
T0
T0
m.t /; n.t / D
e
e
dt D
e T0
dt
t0
8t0R
t0 CT0
dt;
mDn
<Rt0

t0 CT0
D
cos2.m n/t =T0
t
0
: Cj sin2.m n/t =T0 dt; m n

(
T0 ; m D n
D
0; m n
We also conclude that cn D T0
Complex exponential Fourier series summary:
x.t / D
1
X
Xne j n!0t ; t0 t t0 C T0
nD 1
1
where Xn D
T0
Z
x.t /e
j n!0 t
T0
The Fourier series expansion is unique
Example 2.6: x.t / D cos2 !0t

If we expand x.t / into complex exponentials we can immediately determine the Fourier coefficients
1 1
x.t / D C cos 2!0t
2 2
1 1
1
D C e j 2!0t C e j 2!0t
2 4
4
2-21
The above implies that

8
1
<2;
Xn D 14 ;
:0;
nD0
n D 2
otherwise
Time Average Operator

The time average of signal v.t / is defined as
Z T
1
hv.t /i D lim
v.t / dt
T !1 2T
T
Note that
hav1.t / C bv2.t /i D ahv1.t /i C bhv2.t /i;
where a and b are arbitrary constants
If v.t / is periodic, with period T0, then
Z
1
hv.t /i D
v.t / dt
T0 T 0
The Fourier coefficients can be viewed in terms of the time
average operator
Let v.t / D x.t /e
that
j n!0 t
using e
j
D cos
j sin , we find
Xn D hv.t /i D hx.t /e j n!0t i

D hx.t / cos n!0ti j hx.t / sin n!0t i
2-22
2.4. FOURIER SERIES
2.4.2
Symmetry Properties of the Fourier Coefficients
For x.t / real, the following coefficient symmetry properties

hold:
1. Xn D X
2. jXnj D jX nj
3. Xn D
proof

Z
1
x.t /e j n!0t dt
Xn D
T0 T0
Z
1
D
x.t /e j. n/!0t dt D X
T0 T0

since x .t / D x.t /

Waveform symmetry conditions produce special results too
1. If x. t / D x.t / (even function), then

Xn D Re Xn ; i.e., Im Xn D 0
2. If x. t / D
x.t / (odd function), then

Xn D Im Xn ; i.e., Re Xn D 0
3. If x.t T0=2/ D
x.t / (odd half-wave symmetry), then

Xn D 0 for n even
2-23
Example 2.7: Odd Half-wave Symmetry Proof

Consider
1
Xn D
T0
t0 DT0 =2
t0
1
x.t /e j n!0t dtC
T0
t0 CT0
j n!0 t 0
x.t 0/e
dt 0
t0 CT0 =2
In the second integral we change variables by letting t D t 0

T0=2
1
Xn D
T0
x.t /e
j n!0 t
dt
t0
1
C
T0

D 1
t0 CT0 =2
tCT0 =2
x.t T0=2/ e

t0
j n!0 .t CT0 =2/
dt
x.t/
j n!0 T0 =2
t0 CT0 =2
1 Z
T0
x.t /e
j n!0 t
dt
t0
but n!0.T0=2/ D n.2=T0/.T0=2/ D n, thus

(
1
j n
2;
n odd
0;
n even
We thus see that the even indexed Fourier coefficients are indeed zero under odd half-wave symmetry
2-24
2.4. FOURIER SERIES
2.4.3
Trigonometric Form
The complex exponential Fourier series can be arranged as follows

1
X
x.t / D
Xne j n!0t
nD 1
D X0 C
1
X
Xne j n!0t C X ne
j n!0 t
nD1
For real x.t /, we may know that jX nj D jXnj and Xn D

X n, so
x.t / D X0 C
1
X
jXnje
nD1
1
X
D X0 C 2
j n!0 t CXn
C jXnje
jXnj cos n!0t C Xn
j n!0 t CXn
nD1
since cos.x/ D .e jx C e
jx
/=2
From the trig identity cos.u C v/ D cos u cos v

follows that
x.t / D X0 C
1
X
nD1
An cos.n!0t / C
1
X
sin u sin v, it
Bn sin.n!0t /
nD1
where
An D 2hx.t / cos.n!0t /i
Bn D 2hx.t / sin.n!0t /i
2-25
2.4.4
Parsevals Theorem
Fourier series analysis are generally used for periodic signals,

i.e., x.t / D x.t C nT0/ for any integer n
With this in mind, Parsevals theorem becomes
1
P D
T0
jx.t /j2 dt D
T0
D X02 C 2
1
X
jXnj2
nD 1
1
X
jXnj2
.W/
nD1
Note: A 1 ohm system is assumed
2.4.5
Line Spectra
Line spectra was briefly reviewed earlier for simple signals

For any periodic signal having Fourier series representation we
can obtain both single-sided and double-sided line spectra
The double-sided magnitude and phase line spectra is most
easily obtained form the complex exponential Fourier series,
while the single-sided magnitude and phase line spectra can be
obtained from the trigonometric form
1
X
Double-sided
()
Xne j 2.nf0/t
mag. and phase
nD 1
X
Single-sided
() X0 C 2
jXnj cos2.nf0/t C Xn
mag. and phase
nD1
2-26
2.4. FOURIER SERIES
For the double-sided simply plot as lines jXnj and Xn

versus nf0 for n D 0; 1; 2; : : :
For the single-sided plot jX0j and X0 as a special case
for n D 0 at nf0 D 0 and then plot 2jXnj and X0 versus
nf0 for n D 1; 2; : : :
Example 2.8: Cosine Squared

Consider

A A
x.t / D A cos .2f0t C / D C cos 2.2f0/t C 21
2
2
A
A A
D C e j 21 e j 2.2f0/t C e j 21 e j 2.2f0/t
2
4
4
2
DoubleSided
f
-2f0
-2f0
f
2f0
2f0
SingleSided
f
2f0
f
2f0
2-27
Example 2.9: Pulse Train

x(t)
A
...
...
t
-2T0
-T0
T0 T0 +
Periodic pulse train
The pulse train signal is mathematically described by

1
X
t nT0 =2
x.t / D
A

nD 1
The Fourier coefficients are
Z
A e j 2.nf0/t
1
j 2.nf0 /t
Ae
dt D

Xn D
T0 0
T0
j 2.nf0/ 0
A 1 e j 2.nf0/

D
T0
j 2.nf0/
A e j.nf0/ e j.nf0/
D

e j.nf0/
T0
.2j /.nf0/
A sin.nf0/
D

e j.nf0/
T0
.nf0/
To simplify further we define
sinc.x/ D
2-28
sin.x/
x
2.4. FOURIER SERIES
Finally,
Xn D
A
sinc.nf0 /e
T0
j.nf0 /
; n D 0; 1; 2; : : :
To plot the line spectra we need to find jXnj and Xn

A
jsinc.nfo/ j
T
80
sinc.nfo / > 0
< .nf0/;
Xn D
.nf0/ C ; nf0 > 0 and sinc.nf0 / < 0
: .nf0/ ; nf0 < 0 and sinc.nf0 / < 0

jXnj D
Pulse train double-sided line spectra for D 0:125 using Python,

specifically use ssd.line_spectra()
As a specific example enter the following at the IPython command prompt
n = arange(0,25+1) # Get 0 through 25 harmonics

tau = 0.125; f0 = 1; A = 1;
Xn = A*tau*f0*sinc(n*f0*tau)*exp(-1j*pi*n*f0*tau)
figure(figsize=(6,2))
f = n # Assume a fundamental frequency of 1 Hz so f = n
ssd.line_spectra(f,Xn,mode='mag',fsize=(6,2))
xlim([-25,25]);
ssd.line_spectra(f,Xn,mode='phase',fsize=(6,2))
xlim([-25,25]);
2-29
f0 = 1, t = 0.125
At f0 = 0.125
1 t =8
Phase slope = -p f t
= -0.125 p f
2.4.6
Numerical Calculation of Xn
Here we consider a purely numerical calculation of the Xk coefficients from a single period waveform description of x.t /
In particular, we will use MATLABs fast Fourier transform
(FFT) function to carry out the numerical integration
By definition
1
Xk D
T0
2-30
Z
x.t /e
j 2kf0 t
dt; k D 0; 1; 2; : : :
T0
2.4. FOURIER SERIES
A simple rectangular integration approximation to the above

integral is
N 1
1 X
Xk '
x.nT /e
T0 nD0
j k2.nf0 /T0 =N
T0
; k D 0; 1; 2; : : :
N
where N is the number of points used to partition the time

interval 0; T0 and T D T0=N is the time step
Using the fact that 2f0T0 D 2, we can write that
N 1
1 X
x.nT /e
Xk '
N nD0
j 2k n
N
; k D 0; 1; 2; : : :
Note that the above must be evaluated for each Fourier coefficient of interest
Also note that the accuracy of the Xk values depends on the
value of N
For k small and x.t / smooth in the sense that the harmonics rolloff quickly, N on the order of 100 may be adequate
For k moderate, say 550, N will have to become increasingly larger to maintain precision in the numerical
integral
Calculation Using the FFT
The FFT is a powerful digital signal processing (DSP) function, which is a computationally efficient version of the discrete Fourier transfrom (DFT)
2-31
For the purposes of the problem at hand, suffice it to say that

the FFT is just an efficient algorithm for computing
X k D
N
X1
xne
j 2k n=N
; k D 0; 1; 2; : : : ; N
nD0
If we let xn D x.nT /, then it should be clear that

Xk '
N
1
X k; k D 0; 1; : : : ;
N
2
To obtain Xk for k < 0 note that

X
N 1
1
1 X
x.nT /e
' X k D
N
N nD0
N 1
1 X
x.nT /e
D
N nD0
since e
j 2N n=N
De
j 2 n
j 2.N k/n
N
j 2. k/n
N
D X N
D1
In summary
(
Xk '
X k=N;
X N
k=N;
0 k N=2
N=2 k < 0
To use the Python function fft.fft() to obtain the Xk we

simply let
X D fft.fft(x)
where x D fx.t / W t D 0; T0=N; 2T0=N; : : : ; T0.N
1/=N g
Unlike in MATLAB X 0 is really found in X[0]

2-32
2.4. FOURIER SERIES
Example 2.10: Finite Rise/Fall-Time PulseTrain

x(t)
1
Pulse width =
Rise and fall time = tr
1/2
tr
+ tr
t
T0
Pulse train with finite rise and fall time edges
Shown above is one period of a finite rise and fall time pulse
train
We will numerically compute the Fourier series coefficients of
this signal using the FFT
The Python function trap_pulse was written to generate one
period of the signal using N samples
def trap_pulse(N,tau,tr):
"""
xp = trap_pulse(N,tau,tr)
Mark Wickert, January 2015
"""
n = arange(0,N)
t = n/N
xp = zeros(len(t))
# Assume tr and tf are equal
T1 = tau + tr
# Create one period of the trapezoidal pulse waveform
for k in n:
if t[k] <= tr:
xp[k] = t[k]/tr
elif (t[k] > tr and t[k] <= tau):
2-33
xp[k] = 1
elif (t[k] > tau and t[k] < T1):
xp[k] = -t[k]/tr + 1 + tau/tr;
else:
xp[k] = 0
return xp, t
We now plot the double-sided line spectra for D 1=8 and two
values of rise-time tr
# tau = 1/8, tr = 1/20

N = 1024
xp,t = trap_pulse(N,1/8,1/20)
Xp = fft.fft(xp)
plot(t,xp)
grid()
title(r'Spectra of Finite Risetime Pulse Train: $ D 1=8$ $t _r D 1=20$0 /
ylabel(r'$x(t)$')
xlabel('Time (s)')
f = arange(0,N/2)
ssd.line_spectra(f[0:25],abs(Xp[0:25])/N,'magdB',floor_dB=-80,fsize=(6,2))
ylabel(r'$|X_n| = |X(f_n)|$ (dB)');
#% tau = 1/8, tr = 1/10
xp,t = trap_pulse(N,1/8,1/10)
Xp = fft.fft(xp)
plot(t,xp)
grid()
%title(r'Spectra of Finite Risetime Pulse Train: $ D 1=8$ $tr D 1=10$0 /
ylabel(r'$x(t)$')
xlabel('Time (s)')
ssd.line_spectra(f[0:25],abs(Xp[0:25])/N,'magdB',floor_dB=-80,fsize=(6,2))
2-34
2.4. FOURIER SERIES
1/20
1/8
Sidelobes smaller
than ideal pulse train
which has zero rise
time
1/ = 8
Signal x.t/ and line spectrum for D 1=8 and tr D 1=20
The spectral roll-off rate is faster with the trapezoid

Clock edges are needed in digital electronics, but by slowing
the edge speed down the clock harmonic generation can be
reduced
The second plot is a more extreme example
2-35
1/10
1/8
Sidelobes smaller
than with tr = 1/20
case
~10dB
1/ = 8
more
Signal x.t/ and line spectrum for D 1=8 and tr D 1=10
2-36
2.4. FOURIER SERIES
2.4.7
Other Fourier Series Properties
Given x.t / has Fourier series (FS) coefficients Xn, if

y.t / D A C Bx.t /
it follows that
(
Yn D
A C BX0;
nD0
BXn;
n0
proof:
Yn D hy.t /e
j 2.nf0 /t
D Ahe j 2.nf0/t i C Bhx.t /e

1; n D 0
C BXn
DA
0; n 0
j 2.nf0 /t
QED
Likewise if
y.t / D x.t
t0 /
it follows that
Yn D X n e
j 2.nf0 /t0
proof:
Yn D hx.t
Let D t
t0/e
j 2.nf0 /t
t0 which implies also that t D C t0, so

Yn D hx./e
D hx./e
D Xn e
j 2.nf0 /.Ct0 /
j 2.nf0 /
ie
j 2.nf0 /t0
j 2.nf0 /t0
QED
2-37
2.5
Fourier Transform
The Fourier series provides a frequency domain representation

of a periodic signal via the Fourier coefficients and line spectrum
The next step is to consider the frequency domain representation of aperiodic signals using the Fourier transform
Ultimately we will be able to include periodic signals within
the framework of the Fourier transform, using the concept of
transform in the limit
The text establishes the Fourier transform by considering a
limiting case of the expression for the Fourier series coefficient
Xn as T0 ! 1
The Fourier transform (FT) and inverse Fourier transfrom (IFT)
is defined as
Z 1
X.f / D
x.t /e j 2f t dt (FT)
1
Z 1
X.f /e j 2f t df (IFT)
x.t / D
1
Sufficient conditions for the existence of the Fourier transform

are
R1
1. 1 jx.t /j dt < 1
2. Discontinuities in x.t / be finite
R1
3. An alternate sufficient condition is that 1 jx.t /j2 dt <
1, which implies that x.t / is an energy signal
2-38
2.5. FOURIER TRANSFORM
2.5.1
Amplitude and Phase Spectra
FT properties are very similar to those obtained for the Fourier

coefficients of periodic signals
The FT, X.f / D Ffx.t /g, is a complex function of f
X.f / D jX.f /je j.f / D jX.f /je j X.f /
D RefX.f /g C j ImfX.f /g
The magnitude jX.f /j is referred to as the amplitude spectrum
The the angle X.f / is referred to as the phase spectrum
Note that
Z
RefX.f /g D
x.t / cos 2f t dt

Z
1
1
ImfX.f /g D
x.t / sin 2f t dt

1
2.5.2
Symmetry Properties
If x.t / is real it follows that

Z 1
X. f / D
x.t /e j 2. f /t dt
Z11

D
x.t /e j 2f t dt D X .f /
1
thus
jX. f /j D jX.f /j (even in frequency)
X. f / D X.f / (odd in frequency)
2-39
Additionally,
1. For x. t / D x.t / (even function), ImfX.f /g D 0
2. For x. t / D
2.5.3
x.t / (odd function), RefX.f /g D 0
Energy Spectral Density
From the definition of signal energy,

Z 1
jx.t /j2 dt
ED
1

Z 1
Z 1
X.f /e j 2f t df dt
x .t /
D
1

Z 1
Z 1
1
x .t /e j 2f t dt df
D
X.f /
1
but
Z
x .t /e j 2f t dt D
Z
x.t /e
j 2f t

dt
D X .f /
Finally,
Z
ED
jx.t /j dt D
1
jX.f /j2 df
which is known as Rayleighs Energy Theorem

Are the units consistent?
Suppose x.t / has units of volts
jX.f /j2 has units of (volts-sec)2
2-40
In a 1 ohm system jX.f /j2 has units of Watts-sec/Hz =

Joules/Hz
The energy spectral density is defined as
G.f / D jX.f /j2 Joules/Hz

It then follows that
Z
ED
G.f / df
1
Example 2.11: Rectangular Pulse

Consider

x.t / D A
t0
FT is
Z
t0 C=2
X.f / D A
j 2f t
dt
t0 =2
t0C=2
DA
j 2f
t =2
jf 0 jf
e
e
D A
e
.j 2/f
D A sinc.f /e j 2f t0
j 2f t

A
t0

! Asinc.f /e
j 2f t0
j 2f t0
2-41
Plot jX.f /j, X.f /, and G.f /

A1
X(f) 3
Amplitude
Spectrum
|X(f)| 0.8
Phase
Spectrum
0.6
?3
0.4
-2/? 2
?2
-2/
?1
-1/
f
0
1/1
2
2/
?3
/2
? 1-1/
?1
/2
?2
t0 = /2
0.2
?3
11/
2/
2
slope = -f/2
(A)12
G(f) = |X(f)|2
Energy
Spectral
Density
0.8
0.6
0.4
0.2
?3
?2
-2/
?1
-1/
1/1
2
2/
Rectangular pulse spectra
2.5.4
Transform Theorems
Be familiar with the FT theorems found in the table of Appendix G.6 of the text
Superposition Theorem
a1x1.t / C a2x2.t /
! a1X1.f / C a2X2.f /
proof:
2-42
Time Delay Theorem

x.t
t0 /
! X.f /e
j 2f t0
proof:
Frequency Translation Theorem

In communications systems the frequency translation and modulation theorems are particularly important
x.t /e j 2f0t
proof: Note that
Z 1
x.t /e j 2f0t e
j 2f t
! X.f
dt D
f0/
x.t /e
j 2.f f0 /t
dt
so

F x.t /e j 2f0t D X.f
f0 /
QED
Modulation Theorem
The modulation theorem is an extension of the frequency translation theorm
x.t / cos.2f0t /
1
! X.f
2
1
f0/ C X.f C f0/
2
2-43
proof: Begin by expanding

1
1
cos.2f0t / D e j 2f0t C e
2
2
j 2f0 t
Then apply the frequency translation theorem to each term separately
X(f)
signal
multiplier
A
x(t)
Y(f)
y(t)
A/2
f
0
-f0
cos(2f0t)
f
f0
A simple modulator
Duality Theorem
Note that
Z
FfX.t /g D
X.t /e
j 2f t
dt D
X.t /e j 2.
f /t
dt
which implies that

X.t /
2-44
! x. f /
Example 2.12: Rectangular Spectrum

X(f)
1
-W
Using duality on the above we have

t
F
X.t / D
! 2W sinc.2Wf / D x. f /
2W
Since sinc( ) is an even function (sinc.x/ D sinc. x/), it follows that

f
F
2W sinc.2W t / !
2W
Differentiation Theorem
The general result is
d nx.t / F
! .j 2f /n X.f /
n
dt
proof: For n D 1 we start with the integration by parts formula,
R
R
u dv D uv
v du, and apply it to
Z 1
dx
dx j 2f t
F
D
e
dt
dt
1 dt
Z 1
1
D x.t /e j 2f t Cj 2f

x.t /e j 2f t dt
1
X.f /
2-45
alternate From Leibnitzs rule for differentiation of integrals,

Z 1
Z 1
@F .f; t /
d
F .f; t / df D
df
dt 1
@f
1
so
Z 1
dx.t /
d
D
X.f /e j 2f t df
dt
dt
Z 1 1
@e j 2f t
df
D
X.f /
@t
Z 1
1
j 2f X.f /e j 2f t df
! j 2f X.f /
dx=dt
QED
Example 2.13: FT of Triangle Pulse

1
Note that
1/
1/
-1/
2-46
-2/
Using the differentiation theorem for n D 2 we have that

n1
o
t
1
2
1
F
D
F .t C /
.t / C .t /

.j 2f /2

2
1 j 2f
e
C 1 e j 2f

D
.j 2f /2
2 cos.2f / 2
D
.2f /2
4 sin2.f /
2
D
D
sinc
.f /
4.f /2

t
F
! sinc2.f /
Convolution and Convolution Theorem

Before discussing the convolution theorem we need to review
convolution
The convolution of two signals x1.t / and x2.t / is defined as
Z 1
x.t / D x1.t / x2.t / D
x1./x2.t / d
1
Z 1
D x2.t / x1.t / D
x2./x1.t / d
1
A special convolution case is .t t0/

Z 1
.t t0/ x.t / D
. t0/x.t / d
1
D x.t /Dt D x.t t0/

0
2-47
Example 2.14: Rectangular Pulse Convolution

Let x1.t / D x2.t / D .t = /
To evaluate the convolution integral we need to consider the
integrand by sketching of x1./ and x2.t / on the axis for
different values of t
For this example four cases are needed for t to cover the entire
time axis t 2 . 1; 1/
Case 1: When t < we have no overlap so the integrand is
zero and x.t / is zero
x2(t - )
x1()
No overlap for t +
/2 < -/2 or t <
t - /2
t + /2
/2
/2
Case 2: When < t < 0 we have overlap and

Z 1
x1./x2.t / d
x.t / D
1
Z t C=2
t C=2
D
d D
=2
=2
D t C =2 C =2 D C t
x2(t - )
x1()
Overlap begins when t
+ /2 = -/2 or t = -
0
/2
t + /2
2-48
/2
Case 3: For 0 < t < the leading edge of x2.t / is to the

right of x1./, but the trailing edge of the pulse is still overlapped
Z =2
x.t / D
d D =2 t C =2 D t
t =2
x2(t - )
x1()
Overlap lasts until t =
/2
/2
t - /2
t + /2
Case 4: For t > we have no overlap, and like case 1, the

result is
x.t / D 0
x2(t - )
x1()
No overlap for t >
/2
/2
t - /2
t + /2
Collecting the results

8
0;
t<
< C t;
t <0
x.t / D
t; 0 t <
:0;
t
(
jt j; jtj
D
0;
otherwise
2-49
Final summary,

t
t
t

D

Convolution Theorem: We now consider x1.t /x2.t / in terms
of the FT
Z 1
x1. /x2.t / d
1Z
Z 1

1
D
x1. /
X2.f /e j 2f .t / df d
1

Z 11
Z 1
x1. /e j 2f d e j 2f t df
X2.f /
D
1
1
Z 1
D
X1.f /X2.f /e j 2f t df
1
which implies that

x1.t / x2.t /
! X1.f /X2.f /
Example 2.15: Revisit .t = / .t = /

Knowing that .t = /.t = / D .t = / in the time domain,
we can follow-up in the frequency domain by writing

2
F .t = / F .t = / D sinc.f /
We have also established the transform pair

t
F
2
2
2
! sinc .f / D sinc .f /

2-50
or

t
! sinc2.f /
Multiplication Theorem
Having already established the convolution theorem, it follows
from the duality theorem or direct evaluation, that
x1.t / x2.t /
2.5.5
! X1.f / X2.f /
Fourier Transforms in the Limit
Thus far we have considered two classes of signals

1. Periodic power signals which are described by line spectra
2. Non-periodic (aperiodic) energy signals which are described
by continuous spectra via the FT
We would like to have a unifying approach to spectral analysis
To do so we must allow impulses in the frequency domain by
using limiting operations on conventional FT pairs, known as
Fourier transforms-in-the-limit
Note: The corresponding time functions have infinite energy, which implies that the concept of energy spectral
density will not apply for these signals (we will introduce
the concept of power spectral density for these signals)
2-51
Example 2.16: A Constant Signal

Let x.t / D A for 1 < t < 1
We can write
x.t / D lim A.t =T /
T !1
Note that

F A.t =T / D AT sinc.f T /
Using the transform-in-the-limit approach we write
Ffx.t /g D lim AT sinc.f T /
T !1
?3
?2
?1
AT1
0.8
0.8
0.6
0.6
0.4
0.4
0.2
0.2
? 0.2
3 ?3
?2
?1
AT2
T2 >> T1
? 0.2
Increasing T in AT sinc.f T /
Note that since x.t / has no time variation it seems reasonable

that the spectral content ought to be confined to f D 0
Also note that it can be shown that
Z 1
AT sinc.f T / df D A;
8T
Thus we have established that

A
2-52
! A.f /
As a further check
Z

F 1 A.f / D
A.f /e
j 2ff t
df D Ae
j 2f t
f D0
DA
As a result of the above example, we can obtain several more

FT-in-the-limit pairs
F
Ae j 2f0t
! A.f f0/
A j
F
e .f f0/ C e
A cos.2f0t C / !
2
F
A.t t0/ ! Ae j 2f t0
j

.f C f0/
Reciprocal Spreading Property: Compare

F
! A and A
A.t /
! A.f /
A constant signal of infinite duration has zero spectral width,

while an impulse in time has zero duration and infinite spectral
width
2.5.6
Fourier Transforms of Periodic Signals
For an arbitrary periodic signal with Fourier series

x.t / D
1
X
Xne j 2 nf0t
nD 1
2-53
we can write
1
X
"
X.f / D F
#
Xne j 2 nf0t
nD 1
D
D
1
X
nD 1
1
X
n
o
j 2 nf0 t
XnF e
Xn.f
nf0/
nD 1
using superposition and FfAe j 2f0t g D A.f
f0 /
What is the difference between line spectra and continuous

spectra? none!
Mathematically,
Convert to
time domain
Line
Spectra
Sum phasors
Convert to
time domain
Continuous
Spectra
Integrate impulses to
get phasors via the
inverse FT
The Fourier series coefficients need to be known before the FT

spectra can be obtained
A technique that obtained the FT directly will be discussed
later
2-54
Example 2.17: Ideal Sampling Waveform

When we discuss sampling theory it will be useful to have the
FT of the periodic impulse train signal
1
X
ys .t / D
.t
mTs /
mD 1
where Ts is the sample spacing or period

Since this signal is periodic, it must have a Fourier series representation too
In particular
1
Yn D
Ts
Z
.t /e
j 2.nfs /t
dt D
Ts
1
D fs ; any n
Ts
where fs is the sampling rate in Hz

The FT of ys .t / is given by
Ys .f / D fs
1
X
1
X
j 2 nf /t
0
F e
D fs
.f
nD 1
nfs /
nD 1
Summary,
1
X
.t
mD 1
mTs /
! fs
1
X
.f
nfs /
nD 1
2-55
ys(t)
...
...
-Ts
Ts
4Ts
Ys(f)
fs
...
...
f
-fs
fs
4fs
An impulse train in times is an impulse train in frequency
Example 2.18: Convolve Step and Exponential
Find y.t / D Au.t / e
u.t /, > 0
For t 0 there is no overlap so Y .t / D 0
No overlap
t
2-56
0
For t > 0 there is always overlap

t
Z
y.t / D
Ae
.t /
d
D Ae
D Ae
e t
0
e t 1
For t > 0 there is

always overlap
Summary,
y.t / D
A
1
u.t /
A/
y(t)
2-57
Direct Approach for the FT of a Periodic Signal

The FT of a periodic signal can be found directly by expanding
x.t / as follows
" 1
#
1
X
X
x.t / D
.t mTs / p.t / D
p.t mTs /
mD 1
mD 1
where p.t / represents one period of x.t /, having period Ts

From the convolution theorem
( 1
X
X.f / D F
.t
)
mTs / P .f /
mD 1
1
X
D fs P .f /
.f
nfs /
nD 1
1
X
D fs
P .nfs /.f
nfs /
nD 1
where P .f / D Ffp.t /g
The FT transform pair just established is
1
X
mD 1
2-58
p.t
mTs /
1
X
fs P .nfs /.f
nfs /
nD 1
Example 2.19: p.t / D .t =2/ C .t =4/, T0 D 10

x(t)
2
...
...
t
-2
-1
T0 = 10
Stacked pulses periodic signal
We begin by finding P .f / using Ff.t = /g D sinc.f /

P .f / D 2sinc.2f / C 4sinc.4f /
Plugging into the FT pair derived above with nfs D n=10,

1
n
1 X
2n
n
X.f / D
2sinc
C 4sinc
f
10 nD 1
5
5
10
2.5.7
Poisson Sum Formula
The Poisson sum formula from mathematics can be derived

using the FT pair
e
j 2.nfs /t
by writing
( 1
X
1
F
fs P .nfs /.f
nD 1
! .f
nfs /
1
X
nfs / D fs
P .nfs /e j 2.nfs /t
nD 1
2-59
From the earlier developed FT of periodic signals pair, we

know that the left side of the above is also equal to
1
X
p.t
also
mTs / D fs
mD 1
1
X
P .nfs /e j 2.nfs /t
nD 1
We can finally relate this back to the Fourier series coefficients,

i.e.,
Xn D fs P .nfs /
2.6
Power Spectral Density and Correlation
For energy signals we have the energy spectral density, G.f /,

defined such that
Z 1
ED
G.f / df
1
For power signals we can define the power spectral density

(PSD), S.f / of x.t / such that
Z 1
S.f / df D hjx.t /j2i
P D
1
Note: S.f / is real, even and nonnegative

If x.t / is periodic S.f / will consist of impulses at the
harmonic locations
For x.t / D A cos.!0t C /, intuitively,
1
S.f / D A2.f
4
2-60
1
f0/ C A2.f C f0/
4
2.6. POWER SPECTRAL DENSITY AND CORRELATION
since S.f / df D A2=2 as expected (power on a per ohm

basis)
R
To derive a general formula for the PSD we first need to consider the autocorrelation function
2.6.1
The Time Average Autocorrelation Function
Let . / be the autocorrelation function of an energy signal

. / D F 1 G.f /

1

D F X.f /X .f /

1
1
D F X.f / F X .f /
F
! X .f / for x.t / real, so

Z 1
x.t /x.t C / d
. / D x.t / x. t / D
but x. t /
or
Z
x.t /x.t C / d
. / D lim
T !1
Observe that

G.f / D F . /
The autocorrelation function (ACF) gives a measure of the
similarity of a signal at time t to that at time t C ; the coherence between the signal and the delayed signal
2-61
X(f)
G(f) = |X(f)|2
x(t)
() =
Energy spectral density and signal relationships
2.6.2
Power Signal Case
For power signals we define the autocorrelation function as

Rx . / D hx.t /x.t C /i
Z T
1
D lim
x.t /x.t C / dt
T !1 2T
Z T
if periodic 1
x.t /x.t C / dt
D
T0 T0
Note that
Z
Rx .0/ D hjx.t /j i D
Sx .f / df
1
and since for energy signals . /

assumption is that
Rx . /
2-62
! G.f /, a reasonable
! Sx .f /
A formal statement of this is the Wiener-Kinchine theorem (a

proof is given in text Chapter 7)
Z 1
Sx .f / D
Rx . /e j 2f d
1
x(t)
Rx()
Sx(f)
Power spectral density (PSD) and signal relationships
2.6.3
Properties of R. /
The following properties hold for the autocorrelation function

1. R.0/ D hjx.t /j2i jR. /j for all values of
2. R. / D hx.t /x.t
/i D R. / ) an even function
3. limjj!1 R. / D hx.t /i2 if x.t / is not periodic

4. If x.t / is periodic, with period T0, then R. / D R. CT0/
5. FfR. /g D S.f / 0 for all values of f
The power spectrum and autocorrelation function are frequently
used for systems analysis with random signals
For the case of random signal, x.t /, the Fourier transform
X.f / is also a random signal, but now in the frequency domain
The autocorrelation and PSD of x.t /, under typical assumptions, are both deterministic functions; this turns out to be vital
in problem solving in Comm II and beyond
2-63
Example 2.20: Single Sinusoid

Consider the signal x.t / D A cos.2f0t C /, for all t
Z
1 T0 2
Rx . / D
A cos.2f0t C / cos.2.t C / C / dt
T0 0
Z

A2
D
cos.2f0 / C cos.2.2f0/t C 2f0 C 2 / dt
2T0 T0
A2
cos.2f0 /
D
2
Note that

A2
F Rx . / D Sx .f / D
.f
4

f0/ C .f C f0/
More Autocorrelation Function Properties

Suppose that x.t / has autocorrelation function Rx . /
Let y.t / D A C x.t /, A D constant
Ry . / D hA C x.t /A C x.t C /i
D hA2i C hAx.t C /i C hAx.t /i C hx.t /x.t C /i
D A2 C 2Ahx.t /i CRx . /
const. terms
Let z.t / D x.t
t0 /
Rz . / D hz.t /z.t C /i D hx.t t0/x.t t0 C i

D hx./x. C /i; with D t t0
D Rx . /
2-64
The last result shows us that the autocorrelation function is

blind to time offsets
Example 2.21: Sum of Two Sinusoids

Consider the sum of two sinusoids
y.t / D x1.t / C x2.t /
where x1.t / D A1 cos.2f1tC1/ and x2.t / D A2 cos.2f2tC
2/ and we assume that f1 f2
Using the definition
Ry . / D hx1.t / C x2.t /x1.t C /x2.t C /i
D hx1.t /x1.t C /i C hx2.t /x2.t C /i
C hx1.t /x2.t C /i C hx2.t /x1.t C /i
The last two terms are zero since hcos..!1 !2/t /i D 0 when
f1 f2 (why?), hence
Ry . / D Rx1 . / C Rx2 . /; for f1 f2
A21
A22
cos.2f1 / C
cos.2f2 /
D
2
2
Example 2.22: PN Sequences

In the testing and evaluation of digital communication systems
a source of known digital data (i.e., 1s and 0s) is required
(see also text Chapter 10 p. 524527)
2-65
A maximal length sequence generator or pseudo-noise (PN)

code is often used for this purpose
Practical implementation of a PN code generator can be accomplished using an N -stage shift register with appropriate
exclusive-or feedback connections
The sequence length or period of the resulting PN code is M D
2N 1 bits long
Clock
Period = T
C
C
Q1
D1
D2
C
Q2
D3
x(t)
Q3
M = 23 - 1 = 7
x(t)
+A
t
-A
one period = MT
Three stage PN (m-sequence) generator using logic circuits
PN sequences have quite a number of properties, one being that

the time average autocorrelation function is of the form shown
below
2-66
Rx()
A2
MT
...
-T
...
MT
-A2/M
PN sequence autocorrelation function
The calculation of the power spectral density will be left as a

homework problem (a specific example is text Example 2.20)
Hint: To find Sx .f / D FfRx . /g we use
X
X
F
p.t nTs / ! fs
P .nfs /.f
n
nfs /
where Ts D M T
One period of Rx . / is a triangle pulse with a level shift
Suppose the logic levels are switched from A to positive levels of say v1 to v2
Using the additional autocorrelation function properties
this can be done
You need to know that a PN sequence contains one more
1 than 0
Python code for generating PN sequences from 2 to 12 stages
plus 16, is found ssd.py:
def PN_gen(N_bits,m=5):
"""
Maximal length sequence signal generator.
2-67
Generates a sequence 0/1 bits of N_bit duration. The bits themselves are obtained
from an m-sequence of length m. Available m-sequence (PN generators) include
m = 2,3,...,12, & 16.
Parameters
---------N_bits : the number of bits to generate
m : the number of shift registers. 2,3, .., 12, & 16
Returns
------PN : ndarray of the generator output over N_bits
Notes
----The sequence is periodic having period 2**m - 1 (2^m - 1).
Examples
------->>> # A 15 bit period signal nover 50 bits
>>> PN = PN_gen(50,4)
"""
c = m_seq(m)
Q = len(c)
max_periods = int(np.ceil(N_bits/float(Q)))
PN = np.zeros(max_periods*Q)
for k in range(max_periods):
PN[k*Q:(k+1)*Q] = c
PN = np.resize(PN, (1,N_bits))
return PN.flatten()
def m_seq(m):
"""
Generate an m-sequence ndarray using an all-ones initialization.
Available m-sequence (PN generators) include m = 2,3,...,12, & 16.
Parameters
---------m : the number of shift registers. 2,3, .., 12, & 16
Returns
------c : ndarray of one period of the m-sequence
Notes
----The sequence period is 2**m - 1 (2^m - 1).
Examples
------->>> c = m_seq(5)
"""
# Load shift register with all ones to start
sr = np.ones(m)
# M-squence length is:
Q = 2**m - 1
c = np.zeros(Q)
2-68
if m == 2:
taps = np.array([1, 1, 1])
elif m == 3:
taps = np.array([1, 0, 1, 1])
elif m == 4:
taps = np.array([1, 0, 0, 1, 1])
elif m == 5:
taps = np.array([1, 0, 0, 1, 0, 1])
elif m == 6:
taps = np.array([1, 0, 0, 0, 0, 1, 1])
elif m == 7:
taps = np.array([1, 0, 0, 0, 1, 0, 0, 1])
elif m == 8:
taps = np.array([1, 0, 0, 0, 1, 1, 1, 0, 1])
elif m == 9:
taps = np.array([1, 0, 0, 0, 0, 1, 0, 0, 0, 1])
elif m == 10:
taps = np.array([1, 0, 0, 0, 0, 0, 0, 1, 0, 0, 1])
elif m == 11:
taps = np.array([1, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 1])
elif m == 12:
taps = np.array([1, 0, 0, 0, 0, 0, 1, 0, 1, 0, 0, 1, 1])
elif m == 16:
taps = np.array([1, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 1, 1])
else:
print 'Invalid length specified'
for n in range(Q):
tap_xor = 0
c[n] = sr[-1]
for k in range(1,m):
if taps[k] == 1:
tap_xor = np.bitwise_xor(tap_xor,np.bitwise_xor(int(sr[-1]),
int(sr[m-1-k])))
sr[1:] = sr[:-1]
sr[0] = tap_xor
return c
R. /, S.f /, and Fourier Series

For a periodic power signal, x.t /, we can write
1
X
x.t / D
Xne j 2.nf0/t
nD 1
There is an interesting linkage between the Fourier series representation of a signal, the power spectrum, and then back to
the autocorrelation function
2-69
Using the orthogonality properties of the Fourier series expansion we can write
* 1
!
!+
1
X
X
R. / D
Xne j 2.nf0/t
Xme j 2.mf0/.tC/
D
nD 1
1
1
X X
mD 1
XnXm
nD 1 mD 1
D
D
1
X
nD 1
1
X
j 2.nf t/ j 2.mf /.tC/

0 e
0
e
nm terms are zero, why?
jXnj2 e j 2.nf0/t e
j 2.nf0 /.tC/
jXnj2e j 2.nf0/
nD 1
The power spectral density can be obtained by Fourier transforming both sides of the above
S.f / D
1
X
jXnj2.f
nf0/
nD 1
Example 2.23: PN Sequence Analysis and Simulation

In this example I consider an N D 4 or M D 15 generator
from two points of view:
1. First using the FFT to calculate the Fourier coefficients
Xn using one period of the waveform (in discrete-time I
use 10 samples per bit)
2. Second, I just generate a waveform of 10,000 bits (again
at 10 samples per bit) and use standard signal processing
2-70
tools to estimate the time average autocorrelation function

and the PSD
Numerical Fourier Series Analysis: Generate one period of
the waveform using ssd.m_seq(4) and then upsample and interpolate to create a waveform of 10 samples per bit
The waveform amplitude levels are 0; 1 so there is a large
DC component visible in the spectrum (why?); eight ones and
seven zeros makes the average value X0 D 8=15 D 0:533
The function ssd.m_seq() returns an array of zeros and ones
of length 15, i.e., len(x_PN4) = 15
To create a waveform I upsample the signal by 10 and then
filter using a finite impulse response of exactly 10 ones
To plot the line spectrum I use the ssd.line_spectra() function used earlier
x_PN4 = ssd.m_seq(4)
x = signal.lfilter(ones(10),1,ssd.upsample(x_PN4,10))
t = arange(0,len(x))/10
plot(t,x);
title(r'Time Domain and PSD of $M=4$ PN Code with $T = 1$')
xlabel(r'Time (s)')
ylabel(r'x(t)')
axis([0,15,-.1,1.1]);
grid()
X = fft.fft(x,10*15) # 10 samples/bit so 150 samples/period
f = arange(0,7*10)/15 # harmonics spaced by 1/(15*T) = 1/15
ssd.line_spectra(f[0:45],abs(X[0:45])/150,'magdB',floor_dB=-50,fsize=(6,2))
xlim([-3,3])
2-71
15 bit
period
Line spacing =
1/(MT) = 1/15 Hz
DC line 0.533 (-5.46 dB)

f = 1/T = 1 Hz
sinc 2 ( x ) spect. shape
Fourier series based spectral analysis of PN code

import digitalcom as dc
y_PN4_bits = ssd.PN_gen(10000,4) # 10000 bit waveform
# Convert to waveform level shifted to +/-1 amplitude
y = 2*signal.lfilter(ones(10),1,ssd.upsample(y_PN4_bits,10))-1
# Find the time averaged autocorrelation function normalized
# to have a peak amplitude of 1
Ry,tau = dc.xcorr(y,y,200)
tau_s = tau/10
plot(tau_s,Ry)
title(r'Autocorrelation and PSD Estimates for $M=4$ with $T = 1$')
%xlabel(r'Autocorrelation Lag $$ .s/0 /
%ylabel(r'$R_y( /$0 /
grid();
psd(y,2**12,10)
xlabel(r'Frequency (Hz)')
ylabel(r'$S_y(f)$ (dB)')
xlim([0,3]);
ylim([-30,20]);
2-72
...
...
MT = 15T = 15
-1 -1
=
M
15
1
Same spectral shape and line spacing

as direct FS analysis; scaling with
PSD function is different however
DC different
as waveform
values are
[-1,1]
Using simulation to estimate Ry ./ and Sy .f / of a PN code
In generating an estimate of the autocorrelation function I use

the FFT to find the time averaged autocorrelation function in
the frequency domain
In generating the spectral estimate, the Python function psd()(from
matplotlib) function uses Welchs method of averaged periodograms 1
Here the PSD estimate uses a 4096 point FFT and assumed
sampling rate of 10 Hz; the spectral resolution is 10/4096 =
0.002441 Hz
1
http://en.wikipedia.org/wiki/Periodogram
2-73
2.7
Linear Time Invariant (LTI) Systems

x(t)
y(t) =
operator
Linear system block diagram
Definition
Linearity (superposition) holds, that is for input 1x1.t /C2x2.t /,
1 and 2 constants,

y.t / D H 1x1.t / C 2x2.t /

D 1H x1.t / C 2H x2.t /
D 1y1.t / C 2y2.t /
A system is time invariant (fixed) if for y.t / D Hx.t /, a
delayed input gives a correspondingly delayed output, i.e.,

y.t t0/ D H x.t t0/
Impulse Response and Superposition Integral
The impulse response of an LTI system is denoted

h.t / D H .t /
assuming the system is initially at rest
Suppose we can write x.t / as
x.t / D
N
X
n.t
tn /
nD1
2-74
2.7. LINEAR TIME INVARIANT (LTI) SYSTEMS
For an LTI system with impulse response h. /

N
X
y.t / D
nh.t
tn /
nD1
To develop the superposition integral we write

Z 1
x.t / D
x./.t / d
1
' lim
N !1
N
X
x.nt /.t
nt / t; for t 1
nD N
Rectangle area is approximation
x(t)
...
...
2t
3t
4t
5t
6t
Impulse sequence approximation to x.t/
If we apply H to both sides and let t ! 0 such that nt !

we have
y.t / ' lim
N
X
N !1
nD N
x./h.t
Z
nt / t
D
or
x.nt /h.t
/ d D x.t / h.t /
1
1
x.t
/h. / d D h.t / x.t /
1
2-75
2.7.1
Stability
In signals and systems the concept of bounded-input boundedoutput (BIBO) stability is introduced
Satisfying this definition requires that every bounded-input (jx.t /j <
1) produces a bounded output (jy.t /j < 1)
For LTI systems a fundamental theorem states that a system is
BIBO stable if and only if
Z 1
jh.t /j dt < 1
1
Further implications of this will be discussed later
2.7.2
Transfer Function
The frequency domain result corresponding to the convolution

expression y.t / D x.t / h.t / is
Y .f / D X.f /H.f /
where H.f / is known as the transfer function or frequency
response of the system having impulse response h.t /
It immediately follows that
h.t /
! H.f /
and

1
y.t / D F X.f /H.f / D
X.f /H.f /e j 2f t df
1
2-76
2.7.3
Causality
A system is causal if the present output relies only on past and

present inputs, that is the output does not anticipate the input
The fact that for LTI systems y.t / D x.t / h.t / implies that
for a causal system we must have
h.t / D 0; t < 0
Having h.t / nonzero for t < 0 would incorporate future
values of the input to form the present value of the output
Systems that are causal have limitations on their frequency response,
in particular the PaleyWiener theorem states that for
R1
2
1 jh.t /j dt < 1, H.f / for a causal system must satisfy
Z
j ln jH.f /jj
df < 1
2
1
C
f
1
In simple terms this means:

1. We cannot have jH.f /j D 0 over a finite band of frequencies (isolated points ok)
2. The roll-off rate of jH.f /j cannot be too great, e.g., e k1jf j
2
andpe k2jf j are not allowed, but polynomial forms such
as 1=.1 C .f =fc /2N , N an integer, are acceptable
3. Practical filters such as Butterworth, Chebyshev, and elliptical filters can come close to ideal requirements
2-77
2.7.4
Properties of H.f /
For h.t / real it follows that

jH. f /j D jH.f /j and H. f / D
H.f /
why?
Input/output relationships for spectral densities are
Gy .f / D jY .f /j2 D jX.f /H.f /j2 D jH.f /j2Gx .f /
Sy .f / D jH.f /j2Sx .f / proof in text chap. 6
Example 2.24: RC Lowpass Filter

R
x(t)
X(f )
vc(t)
y(t)
ic(t)
Y(f )
h(t), H(f)
RC lowpass filter schematic
To find H.f / we may solve the circuit using AC steady-state

analysis
1
1
Y .j!/
j!c
D
D
1
X.j!/
1 C j!RC
R C j!c
so
H.f / D
2-78
Y .f /
1
D
; where f3 D 1=.2RC /
X.f /
1 C jf =f3
From the circuit differential equation

x.t / D ic .t /R C y.t /
but
ic .t / D c
y.t /
dvc .t /
Dc
dt
dt
thus
RC
dy.t /
C y.t / D x.t /
dt
FT both sides using dx=dt
! j 2f X.f /
j 2f RC Y .f / C Y .f / D X.f /
so again
Y .f /
1
D
X.f /
1 C jf =f3
1
Dp
e j tan
1 C .f =f3/2
H.f / D
1 .f =f /
3
The Laplace transform could also be used here, and perhaps is

preferred, we just need to substitute s ! j! ! j 2f
2-79
f3
-f3
/2
-/2
RC lowpass frequency response
Find the system response to

x.t / D A
.t
T =2/
T
Finding Y .f / is easy since

1
Y .f / D X.f /H.f / D AT sinc.f T /
e
1 C jf =fs
jf t
To find y.t / we can IFT the above, use Laplace transforms, or

convolve directly
From the FT tables we known that
h.t / D
2-80
1
e
RC
t=.RC /
u.t /
In Example 2.18 we showed that

Au.t / e
A
1
u.t / D
u.t /
Note that

A
T =2
T

D Au.t /
u.t
T /
and here D 1=.RC /, so

A
t=.RC /
u.t /
RC 1 e
y.t / D
RC

A
.t T /=.RC /
RC 1 e
u.t
RC
T/
RC =
|X(f)|, |H(f)|, |Y(f)|
T/1
0
T/
5
y(t)
0.8
T/
2
0.8
0.6
0.6
0.4
0.4
0.2
2T
0.2
0.5
1.5
2.5
t/T
-3
-2
-1
0.8
0.6
0.6
RC = T/2
0.4
0.2
-1
fT
|X(f)|, |H(f)|, |Y(f)|

0.8
0.4
-2
|X(f)|, |H(f)|, |Y(f)|
-3
RC = 2T
RC = T/10
0.2
1
fT
-3
-2
-1
fT
Pulse time response and frequency spectra with A D 1
2-81
2.7.5
Input/Output with Spectral Densities
We know that for an LTI system with frequency response H.f /

and input Fourier transform X.f /, the output Fourier transform is given by Y .f / D H.f /X.f /
It is easy to show that in terms of energy spectral density
Gy .f / D jH.f /j2 Gx .f /
where Gx .f / D jX.f /j2 and Gy .f / D jY .f /j2
For the case of power signals a similar relationship holds with
the power spectral density (proof found in Chapter text 7, i.e.,
Comm II)
Sy .f / D jH.f /j2 Sx .f /
2.7.6
Response to Periodic Inputs
When the input is periodic we can write

1
X
x.t / D
Xne j 2.nf0/t
nD 1
which implies that

1
X
X.f / D
Xn.f
nf0/
nD 1
It then follows that

Y .f / D
1
X
XnH.nf0/.f
nf0/
nD 1
2-82
and
y.t / D
D
1
X
nD 1
1
X
XnH.nf0/e j 2.nf0/t
jXnjjH.nf0je j 2.nf0/t CXnCH.nf0/
nD 1
This is a steady-state response calculation, since the analysis

assumes that the periodic signal was applied to the system at
tD 1
2.7.7
Distortionless Transmission
In the time domain a distortionless system is such that for any

input x.t /,
y.t / D H0x.t t0/
where H0 and t0 are constants
In the frequency domain the implies a frequency response of
the form
H.f / D H0e j 2f t0 ;
that is the amplitude response is constant and the phase shift is
linear with frequency
Distortion types:
1. Amplitude response is not constant over a frequency band
(interval) of interest $ amplitude distortion
2. Phase response is not linear over a frequency band of interest $ phase distortion
2-83
3. The system is non-linear, e.g., y.t / D k0 C k1x.t/ C

k2x 2.t / $ nonlinear distortion
2.7.8
Group and Phase Delay
The phase distortion of a linear system can be characterized

using group delay, Tg .f /,
Tg .f / D
1 d.f /
2 df
where .f / is the phase response of an LTI system

Note that for a distortionless system .f / D
Tg .f / D
1 d
2 df
2f t0, so
2f t0 D t0 s;
clearly a constant group delay

Tg .f / is the delay that two or more frequency components
undergo in passing through an LTI system
If say Tg .f1/ Tg .f2/ and both of these frequencies are
in a band of interest, then we know that delay distortion
exists
Having two different frequency components arrive at the
system output at different times produces signal dispersion
An LTI system passing a single frequency component, x.t / D
A cos.2f1t /, always appears distortionless since at a single
2-84
frequency the output is just

y.t / D AjH.f1/j cos 2f1t C .f1/

.f1/
D AjH1.f /j cos 2f1 t
2f1
which is equivalent to a delay known as the phase delay
.f /
2f
Tp .f / D
The system output now is
y.t / D AjH.f1/j cos 2f1.t
Tp .f1//
Note that for a distortionless system

Tp .f / D
1
. 2f t0/ D t0
2f
Example 2.25: Terminated Lossless Transmission Line

Rs = R0
R0, vp
x(t)
RL = R0
y(t)
1
y.t / D x t
2
L
vp
Lossless transmission line

2-85
We conclude that H0 D 1=2 and t0 D L=vp

Note that a real transmission line does have losses that introduces dispersion on a wideband signal
Example 2.26: A Fictitious System

Ampl.
Radians
1.5
H(f)
|H(f)|
1
1.5
0.5
1
-20
-10
10
20
f (Hz)
-0.5
0.5
-1
-20
-10
10
20
f (Hz)
-1.5
Time
Tg(f)
Time
Tp(f)
0.015
0.0125
0.015
0.01
0.014
0.0075
0.013
0.005
0.012
0.0025
-20
0.016
-10
0.011
f (Hz)
10
20
-20
-10
10
20
f (Hz)
No distortion on |f | < 10 Hz band
Amplitude, phase, group delay, phase delay
The system in this example is artificial, but the definitions can

be observed just the same
For signals with spectral content limited to jf j < 10 Hz there
is no distortion, amplitude or phase/group delay
2-86
For 10 < jf j < 15 amplitude distortion is present

For jf j > 15 both amplitude and phase distortion are present
What about the interval 10 < jf j < 15?
2.7.9
Nonlinear Distortion
In the time domain a nonlinear system may be written as

y.t / D
1
X
anx n.t /
nD0
Specifically consider
y.t / D a1x.t / C a2x 2.t /
Let
x.t / D A1 cos.!1t / C A2 cos.!2t /
Expanding the output we have

y.t / D a1 A1 cos.!1t / C A2 cos.!2t /

2
C a1 A1 cos.!1t / C A2 cos.!2t /

D a1 A1 cos.!1t / C A2 cos.!2t /
na
a2 2
o
2
2
2
2
A CA C
A cos.2!1t / C A2 cos.2!2t /
C
2 1 2
2 1

C a2A1A2 cos.!1 C !2/t C cos.!1 !2/t
The third line is the desired output
2-87
The fourth line is termed harmonic distortion

The fifth line is termed intermodulation distortion
Input
Output
a1A1
a2A1
A1
NonLinear
0
f1
Input
f1
A2
2f1
a1A1A2
Output
NonLinear
A1
a2A1
a1a2A1
a2A1
a1A1
a1A2 2
a2(A1 + A2)
a2A2
2
0
f1
f2
0 f2-f1
f1
f2
2f1
2f2
f1+f2
One and two tones in y.t/ D a1 x.t/ C a2 x 2 .t/ device
In general if y.t / D a1x.t / C a2x 2.t / the multiplication theorem implies that
Y .f / D a1X.f / C a2X.f / X.f /

In particular if X.f / D A f =.2W /

f
Y .f / D a1A
2W
2-88
f
C a22WA2
2W
Y( f ) =
2Wa2A2
a 1A
-W
+
f
-2W
2W
a1A + 2Wa2A2
a1A + Wa2A2
Wa2A2
=
-2W -W
W 2W
Continuous spectrum in y.t/ D a1 x.t/ C a2 x 2 .t/ device
2.7.10
Ideal Filters
1. Lowpass of bandwidth B

f
HLP.f / D H0
e
2B
|HLP(f)|
H0
-B
j 2f t0
slope =
-2t0
HLP(f)
f
-B
2. Highpass with cutoff B

HHP.f / D H0 1
|HHP(f)|
H0
-B

.f =.2B// e
slope =
-2t0
j 2f t0
HHP(f)
f
-B
2-89
3. Bandpass of bandwidth B

f0/ C Hl .f C f0/ e
HBP.f / D Hl .f
j 2f t0
where Hl .f / D H0.f =B/
|HBP(f)|
H0
B
slope = -2t0
HBP(f)
f
-f0
f
f0
-f0
f0
The impulse response of the lowpass filter is

1
hLP.t / D F H0.f =.2B//e

D 2BH0sinc2B.t t0/
j 2f t0
Ideal filters are not realizable, but simplify calculations and

give useful performance upper bound results
Note that hLP.t / 0 for t < 0, thus the filter is noncausal
and unrealizable
From the modulation theorem it also follows that
hBP.t / D 2hl .t t0/ cos2f0.t t0/
D 2BH0sincB.t t0/ cos2f0.t
2-90
t0/
hLP(t)
hBP(t)
2BH0
2BH0
t0
t0
t0 - 1
2B
t0 + 1
2B
t0 - 1
2B
t0 + 1
2B
Ideal lowpass and bandpass impulse responses
2.7.11
Realizable Filters
We can approximate ideal filters with realizable filters such as

Butterworth, Chebyshev, and Bessel, to name a few
We will only consider the lowpass case since via frequency
transformations we can obtain the others
Butterworth
A Butterworth filter has a maximally flat (flat in the sense of
derivatives of the amplitude response at dc being zero) passband
In the s-domain (s D Cj!) the transfer function of a lowpass
design is
HBU.s/ D
.s
!cn
s1/.s s2/ .s
sn/
where

1 2k 1
sk D !c exp
C
; k D 1; 2; : : : ; n
2
2n
2-91
Note that the poles are located on a semi-circle of radius !c D

2fc , where fc is the 3dB cuttoff frequency of the filter
The amplitude response of a Butterworth filter is simply
jHBU.f /j p
1
1 C .f =fc /2n
Butterworth n D 4 lowpass filter
Chebyshev
A Chebyshev type I filter (ripple in the passband), is is designed to maintain the maximum allowable attenuation in the
passband yet have maximum stopband attenuation
The amplitude response is given by
jHC.f /j D p
1
1 C 2Cn2.f /
where
(
Cn.f / D
2-92
cos.n cos 1.f =fc //;
0 jf j fc
cosh.n cosh 1.f =fc //;
jf j > fc
The poles are located on an ellipse as shown below
Chebyshev n D 4 lowpass filter
Bessel
A Bessel filter is designed to maintain linear phase in the passband at the expense of the amplitude response
HBE.f / D
Kn
Bn.f /
where Bn.f / is a Bessel polynomial of order n (see text) and

Kn is chosen so that the filter gain is unity at f D 0
2-93
Amplitude Rolloff and Group Delay Comparision

Compare Butterworth, 0.1 dB ripple Chebyshev, and Bessel
n D 3 Amplitude response
n D 3 Group delay
2-94
Despite DSP being pervasive in communication systems, there

is still a great need for analog filter design and implementation
technologies
The table below describes some of the well known construction
techniques
Filter Construction Techniques
Construction
Type
LC (passive)
Description of El- Center Freements or Filter

quency Range
lumped elements
DC300 MHz
or higher in integrated form
Active
R, C , op-amps
DC500 kHz
or higher using
WB op-amps
Crystal
quartz crystal
1kHz 100
MHz
Ceramic
ceramic disks with 10kHz 10.7
electrodes
MHz
Surface acoustic interdigitated fin- 10-800 MHz,
waves (SAW)
gers on a Piezoelectric substrate
Transmission line quarterwave stubs, UHF and miopen and short ckt crowave
Cavity
machined
and Microwave
plated metal
Unloaded
(typical
100
Filter Application
Audio, video,
IF and RF
200
Audio and low

RF
100,000
IF
1,000
IF
variable
IF and RF
1,000
RF
10,000
RF
Example 2.27: Use Python to Characterize Standard Filters

You can use the filter design capability of Python with Scipy
or MATLAB with the signal processing to study lowpass, bandpass, bandstop, and highpass filters
Here I will consider two functions written in Python, one for
digital filters and one for analog filters, to allow plotting of gain
in dB, phase in radians, and group delay in samples or seconds
2-95
def freqz_resp(b,a=[1],mode = 'dB',fs=1.0,Npts = 1024,fsize=(6,4)):

"""
A method for displaying digital filter frequency response magnitude,
phase, and group delay. A plot is produced using matplotlib
freq_resp(self,mode = 'dB',Npts = 1024)
A method for displaying the filter frequency response magnitude,
freqs_resp(b,a=[1],Dmin=1,Dmax=5,mode = 'dB',Npts = 1024,fsize=(6,4))
b
a
Dmin
Dmax
mode
=
=
=
=
=
ndarray of numerator coefficients

ndarray of denominator coefficents
start frequency as 10**Dmin
stop frequency as 10**Dmax
display mode: 'dB' magnitude, 'phase' in radians, or
'groupdelay_s' in samples and 'groupdelay_t' in sec,
all versus frequency in Hz
Npts = number of points to plot; defult is 1024
fsize = figure size; defult is (6,4) inches
"""
f = np.arange(0,Npts)/(2.0*Npts)
w,H = signal.freqz(b,a,2*np.pi*f)
plt.figure(figsize=fsize)
if mode.lower() == 'db':
plt.plot(f*fs,20*np.log10(np.abs(H)))
plt.xlabel('Frequency (Hz)')
plt.ylabel('Gain (dB)')
plt.title('Frequency Response - Magnitude')
elif mode.lower() == 'phase':
plt.plot(f*fs,np.angle(H))
plt.ylabel('Phase (rad)')
plt.title('Frequency Response - Phase')
elif (mode.lower() == 'groupdelay_s') or (mode.lower() == 'groupdelay_t'):
"""
Notes
----Since this calculation involves finding the derivative of the
phase response, care must be taken at phase wrapping points
2-96
and when the phase jumps by +/-pi, which occurs when the
amplitude response changes sign. Since the amplitude response
is zero when the sign changes, the jumps do not alter the group
delay results.
"""
theta = np.unwrap(np.angle(H))
# Since theta for an FIR filter is likely to have many pi phase
# jumps too, we unwrap a second time 2*theta and divide by 2
theta2 = np.unwrap(2*theta)/2.
theta_dif = np.diff(theta2)
f_diff = np.diff(f)
Tg = -np.diff(theta2)/np.diff(w)
max_Tg = np.max(Tg)
#print(max_Tg)
if mode.lower() == 'groupdelay_t':
max_Tg /= fs
plt.plot(f[:-1]*fs,Tg/fs)
plt.ylim([0,1.2*max_Tg])
else:
plt.plot(f[:-1]*fs,Tg)
if mode.lower() == 'groupdelay_t':
plt.ylabel('Group Delay (s)')
else:
plt.ylabel('Group Delay (samples)')
plt.title('Frequency Response - Group Delay')
else:
s1 = 'Error, mode must be "dB", "phase, '
s2 = '"groupdelay_s", or "groupdelay_t"'
print(s1 + s2)
def freqs_resp(b,a=[1],Dmin=1,Dmax=5,mode = 'dB',Npts = 1024,fsize=(6,4)):
"""
A method for displaying analog filter frequency response magnitude,
freqs_resp(b,a=[1],Dmin=1,Dmax=5,mode='dB',Npts=1024,fsize=(6,4))
b
a
Dmin
Dmax
mode
=
=
=
=
=
ndarray of numerator coefficients

ndarray of denominator coefficents
start frequency as 10**Dmin
stop frequency as 10**Dmax
display mode: 'dB' magnitude, 'phase' in radians, or
'groupdelay', all versus log frequency in Hz
Npts = number of points to plot; defult is 1024
2-97
fsize = figure size; defult is (6,4) inches

"""
f = np.logspace(Dmin,Dmax,Npts)
w,H = signal.freqs(b,a,2*np.pi*f)
plt.figure(figsize=fsize)
if mode.lower() == 'db':
plt.semilogx(f,20*np.log10(np.abs(H)))
plt.ylabel('Gain (dB)')
plt.title('Frequency Response - Magnitude')
elif mode.lower() == 'phase':
plt.semilogx(f,np.angle(H))
plt.ylabel('Phase (rad)')
plt.title('Frequency Response - Phase')
elif mode.lower() == 'groupdelay':
"""
Notes
----See freqz_resp() for calculation details.
"""
theta = np.unwrap(np.angle(H))
# Since theta for an FIR filter is likely to have many pi phase
# jumps too, we unwrap a second time 2*theta and divide by 2
theta2 = np.unwrap(2*theta)/2.
theta_dif = np.diff(theta2)
f_diff = np.diff(f)
Tg = -np.diff(theta2)/np.diff(w)
max_Tg = np.max(Tg)
#print(max_Tg)
plt.semilogx(f[:-1],Tg)
plt.ylabel('Group Delay (s)')
plt.title('Frequency Response - Group Delay')
else:
print('Error, mode must be "dB" or "phase or "groupdelay"')
2-98
Case 1: A 5th-order Chebyshev type 1 digital bandpass filter,

having 1 dB ripple and passband of 250; 300 Hz relative to a
sampling rate of fs D 1000 Hz
5th-OrderCheby1 BPF
Digital
fs = 1000 Hz
Highly peaked
near the bandedge, even with
0.1 dB rippler
Digtal bandpass with fs D 1000Hz (Chebyshev)
The Chebyshev in both analog and digital forms still has a large
peak in the group delay, even with a small ripple (here 0.1 dB)
2-99
Case 2: A 7th-order Bessel analog bandpass filter, having passband of 10; 50 MHz
7th-Order
Bessel BPF
Analog
Not 3dB bandwidth
Group delay relatively flat

in passband, compared
with Chebyshev
Analog bandpass (Bessel)
The Bessel filter has a much lower and smoother group delay,
but the magnitude response is rather sloppy
The filter passband is far from being flat and the roll-off is
gradual considering the filter order is seven
2-100
2.7.12
Pulse Resolution, Risetime, and Bandwidth
Problem: Given a non-bandlimited signal, what is a reasonable estimate of the signals transmission bandwidth?
We would like to obtain a relationship to the signals time duration
Step 1: We first consider a time domain relationship by seeking
a constant T such that
Make areas
equal via T
T x.0/ D
x(0)
jx.t /j dt
|x(t)|
t
-T/2
Note that
Z 1
x.t / dt D
1
and
Z
x.t /e j 2f t dt
T/2
D X.0/
f D0
jx.t /j dt
1
x.t / dt
1
which implies
T x.0/ X.0/
Step 2: Find a constant W such that
2-101
Make areas
equal via W
2W X.0/ D
X(0)
jX.f /j df
|X(f)|
f
-W
Note that
Z 1
X.f / df D
1
and
j 2f t
X.f /e
df
D x.0/
t D0
jX.f /j df
X.f / df
1
which implies that

2W X.0/ x.0/
Combining the results of Step 1 and Step 2, we have
2W X.0/ x.0/
or
2W
1
T
or
1
X.0/
T
W
1
2T
Example 2.28: Rectangle Pulse

Consider the pulse x.t / D .t =T /
We know that X.f / D T sinc.f T /
2-102
x(t)
-1
|X(f)|/T
-0.5
0.8
0.8
0.6
0.6
0.4
0.4
0.2
0.2
0.5
t/T
-3
-2
-1
Lower
bound for
W
fT
f
-1/(2T) 1/(2T)
Pulse width versus Bandwidth, is W 1=.2T ?
We see that for the case of the sinc. / function the bandwidth,
W , is clearly greater than the simple bound predicts
Risetime
There is also a relationship between the risetime of a pulse-like
signal and bandwidth
Definition: The risetime, TR , is the time required for the leading edge of a pulse to go from 10% to 90% of its final value
Given the impulse response h.t / for an LTI system, the step
response is just
Z 1
ys .t / D
h./u.t / d
Z t1
Z t
if causal
D
h./ d D
h./ d
1
0
2-103
Example 2.29: Risetime of RC Lowpass

The RC lowpass filter has impulse response
h.t / D
1
e
RC
t=.RC /
u.t /
The step response is

ys .t / D 1
t=.RC /
u.t /
The risetime can be obtained by setting ys .t1/ D 0:1 and ys .t2/ D

0:9

t1
0:1 D 1 e t1=.RC / ) ln.0:9/ D
RC

t2
0:9 D 1 e t2=.RC / ) ln.0:1/ D
RC
The difference t2
TR D t 2
t1 is the risetime
t1 D RC ln.0:9=0:1/ ' 2:2RC D
0:35
f3
where f3 is the RC lowpass 3dB frequency
Example 2.30: Risetime of Ideal Lowpass

The risetime of an ideal lowpass filter is of interest since it is
used in modeling and also to see what an ideal filter does to a
step input
2-104
The impulse response is

f
h.t / D F 1
D 2Bsinc2Bt
2B
The step response then is
Z t
ys D
2Bsinc2B d
1
Z
1 2Bt sin u
du
D
1
u
1
1
D C Si2Bt
2
where Si( ) is a special function known as the sine integral
We can numerically find the risetime to be
TR '
Step Response of RC Lowpass
0:44
B
Step Response of Ideal Lowpass
1
1
0.8
0.8
0.6
0.4
0.6
2.2
0.4
0.2
0.44
0.2
1
t RC
-2
-1
t/B
RC and ideal lowpass risetime comparison
2-105
2.8
Sampling Theory
Integrate with Chapter 3 material.
2.9
The Hilbert Transform
Integrate with Chapter 3 material.
2.10
The Discrete Fourier Transform and

FFT
2-106
2.10. THE DISCRETE FOURIER TRANSFORM AND FFT
2-107

N5625 2

Hochgeladen von

Dokumentinformationen

Originaltitel

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

N5625 2

Hochgeladen von

Copyright:

Verfügbare Formate

2

Signal and Linear System Analysis

Deterministic and Random Signals . . . . . . . .

Periodic and Aperiodic Signals . . . . . . . . . .

Phasor Signals and Spectra . . . . . . . . . . . .

Signal Classifications . . . . . . . . . . . . . . . . . . 2-11

Generalized Fourier Series . . . . . . . . . . . . . . . 2-14

Fourier Series . . . . . . . . . . . . . . . . . . . . . . 2-20

Complex Exponential Fourier Series . . . . . . . 2-20

Symmetry Properties of the Fourier Coefficients

Trigonometric Form . . . . . . . . . . . . . . . 2-25

Parsevals Theorem . . . . . . . . . . . . . . . . 2-26

Line Spectra . . . . . . . . . . . . . . . . . . . 2-26

Numerical Calculation of Xn . . . . . . . . . . . 2-30

Other Fourier Series Properties . . . . . . . . . . 2-37

Fourier Transform . . . . . . . . . . . . . . . . . . . . 2-38

Amplitude and Phase Spectra . . . . . . . . . . 2-39

CHAPTER 2. SIGNAL AND LINEAR SYSTEM ANALYSIS

Symmetry Properties . . . . . . . . . . . . . . . 2-39

Energy Spectral Density . . . . . . . . . . . . . 2-40

Transform Theorems . . . . . . . . . . . . . . . 2-42

Fourier Transforms in the Limit . . . . . . . . . 2-51

Fourier Transforms of Periodic Signals . . . . . 2-53

Poisson Sum Formula . . . . . . . . . . . . . . 2-59

Power Spectral Density and Correlation . . . . . . . . 2-60

The Time Average Autocorrelation Function . . 2-61

Power Signal Case . . . . . . . . . . . . . . . . 2-62

Properties of R./ . . . . . . . . . . . . . . . . 2-63

Linear Time Invariant (LTI) Systems . . . . . . . . . 2-74

Transfer Function . . . . . . . . . . . . . . . . . 2-76

Properties of H.f / . . . . . . . . . . . . . . . . 2-78

Input/Output with Spectral Densities . . . . . . . 2-82

Response to Periodic Inputs . . . . . . . . . . . 2-82

Distortionless Transmission . . . . . . . . . . . 2-83

Group and Phase Delay . . . . . . . . . . . . . . 2-84

Nonlinear Distortion . . . . . . . . . . . . . . . 2-87

2.7.10 Ideal Filters . . . . . . . . . . . . . . . . . . . . 2-89

Sampling Theory . . . . . . . . . . . . . . . . . . . . . 2-106

The Hilbert Transform . . . . . . . . . . . . . . . . . 2-106

2.10 The Discrete Fourier Transform and FFT . . . . . . . 2-106

ECE 5625 Communication Systems I

2.1. SIGNAL MODELS

Deterministic and Random Signals

 Deterministic Signals, used for this course, can be modeled as

Periodic and Aperiodic Signals

 A deterministic signal is periodic if we can write

CHAPTER 2. SIGNAL AND LINEAR SYSTEM ANALYSIS

 A signal is aperiodic otherwise, e.g.,

(a) periodic signal, (b) aperiodic signal, (c) random signal

Phasor Signals and Spectra

 A complex sinusoid can be viewed as a rotating phasor

 This signal has three parameters, amplitude A, frequency !0,

ECE 5625 Communication Systems I

2.1. SIGNAL MODELS

 This signal is periodic with period T0 D 2=!0

(a) obtain x.t/ from x.t/,

CHAPTER 2. SIGNAL AND LINEAR SYSTEM ANALYSIS

 The relevant parameters are A and  for a particular f0 D

(a) Single-sided line spectra, (b) Double-sided line spectra

 Both the single-sided and double-sided line spectra, shown

Example 2.1: Multiple Sinusoids

 Find the two-sided amplitude and phase line spectra of x.t /

x.t / D 4 cos.2.10/t C =3/ C 24 cos.2.100/t

Properties of R./ . . . . . . . . . . . . . . . . 2-63

Deterministic Signals, used for this course, can be modeled as

A deterministic signal is periodic if we can write

A signal is aperiodic otherwise, e.g.,

A complex sinusoid can be viewed as a rotating phasor

This signal has three parameters, amplitude A, frequency !0,

This signal is periodic with period T0 D 2=!0

The relevant parameters are A and for a particular f0 D

Both the single-sided and double-sided line spectra, shown

Find the two-sided amplitude and phase line spectra of x.t /

x.t / D 4 cos.2.10/t C =3/ C 24 cos.2.100/t

The complex sinusoid form is directly related to the two-sided

In both of the above test functions letting ! 0 results in a

Using the rectangular pulse test function, .t /, we note that

Placing the above in the integrand with x.t / we obtain, with

For > 0 the energy is given by

For < 0 we have P ! 1

The signal average power is finite since the above integral is

A measure of the approximation error is the integral squared

To find the Xns giving the minimum N we expand the above

Definition: The set of of ns is complete if

Signal x.t/ and basis functions i .t/; i D 1; 2; 3

To begin with it should be clear that 1.t /; 2.t /, and 3.t /

Before finding the Xns we need to find the cns

Now we can compute the Xns:

The integral-squared error, N , can be computed as follows:

A set of ns that is complete is

: Cj sin2.m n/t =T0 dt; m n

The Fourier series expansion is unique

The above implies that