1.

Signal Processing Fundamentals

1.1 Continuous and discrete signals
Fundamental to the study of radio science and techniques for space exploration is the
concept of a signal and how the signal is represented and processed. In this course we
want to use the word “signal” in the broadest possible sense.

A signal is the variation of a physically measurable quantity. Variations of the quantity

are manifestations of a particular process under study and have this process as their
cause.

Noise, in contrast, is a variation of the physically measurable quantity. However,

variations of the quantity are not a manifestation of a particular process under study and
have a frequently unknown process as their cause.

Examples of signals:

• The voltage fluctutions at the focus of a radio antenna tracking the Cassini
spacecraft.
• The current across a 1 Ohm resistor.
• The beam pattern of a radar antenna
• The ground displacements in continental drift

In electrical engineering a signal is often a function of time and/or space and can be
functionally represented.

Example: a continuous time signal:

y(t)

If we assume that the function y(t) has been sampled at regular intervals, Dt, then the set
of samples is called a time series. It consists of N known values, yk, k=1,2,3,.., N

1
yk
or y[k]

k Dt

In the following we want to use parenteses for continuous functions and subscripts or
brackets for discrete functions. For instance:

y(t): continuous function

yk : discrete function
y[k]; discrete function

Some very fundamental continuous and discrete functions are:

Sign function

1, t ³0 1
sgn(t) ={
-1, t<0 t
-1

1
1, k ³0
sgn[k] ={
-1, k<0 k
-1

2
Step function

1, t ³0 1
u(t) ={
0, t<0 0 t

1, k ³0 1
u[k] ={
0, k<0 0 k

Ramp function

t, t ³0 1
r(t) ={
0, t<0 0 t

k, k ³0 1
r[k] ={
0, k<0 0 k

3
Delta function
d(t)
1
¥, t =0
d(t) ={
0, t¹0 0 t
!"
! 𝛿 (𝑡 ) = 1
#"

1
1, k =0
d[k] ={
0, k ¹0 0 k

Some important relations are:

u(t) = ½ (1+sgn(t))
u[k]=1/2(1+sgn[k])

r(t) = t u(t)
r[k]=k u[k]

d(t) = d/dt u(t)

d[k] =u[k] – u[k-1], k£0
t
u(t) = ò d(t) dt
-¥
k
u[k] = å d[k]
n=-¥

4
 1.2 Fourier Series
The representation of a function over a certain interval by a linear combination of
mutually orthogonal functions is called a Fourier series representation of a function.
Examples of sets of orthogonal functions are:

Trigonometric functions: sin nw0t, cos nw0t

Exponential functions: exp(i nw0t)
Legrendre polynomials: Pn(t)=1/(2n n!) dn/dtn (t2-1)n
+p
Bessel functions: Jn(b) = 1/(2p) ò exp(i b sin x – nx) dx
-p

One particular Fourier series we will use in this course is the exponential Fourier series. It
uses the functions { exp(i nw0t) } n=0, ±1, ±2, ±3,… which are orthogonal over the
interval (t0 ,t0 +T).

+∞
y(t) = ∑c e n
inω 0 t

n=−∞
t 0 +T
1 −inω 0 t
cn =
T
∫ y(t)e dt
t0

The magnitude and phase of the nth harmonic are

€

cn = Re 2 [cn ]+ Im 2 [cn ]
Im[cn ]
∠cn = tan −1 ( )
Re[cn ]

5
 Example: Rectangular pulse train with period T

y(t)
A

-t/2 0 t/2 T 2T t

We can now compute the coefficients, cn, by writing the function y(t) in the form;

+∞
y(t) = ∑c e n
inω 0 t
𝜔! =
"#
$
= 2𝜋𝜈! fundamental frequency
n=−∞

The coefficients are then given as:

€ 1 +τ / 2 −inω 0 t
cn = ∫
T −τ / 2
Ae dt

A +τ / 2 −inω 0 t
cn = ∫ e dt
T −τ / 2
A 1 +τ /2
cn = e−inω 0 t
T −inω 0 −τ /2
A 1 +τ /2
cn = (sin nω 0 t + icos nω 0 t
T nω 0 −τ /2
A nπτ
cn = (sin )
nπ T
A nπτ
c 0 = lim (sin )
n →0 nπ T

€
 Aτ
c0 =
T

And the magnitude and phase part of the coefficients are:

€
A nπτ
cn = (sin )
nπ T
0
∠cn = tan −1 ( )
cn

Let us now look at a specific example, for instance a pulsed radar signal for range and
range rate determinations, and assume that:

T = 4ms
t = 1ms.

We get discrete values for cn with an envelope that peaks at n=0 and has zero-crossings at

npt/T = p
nt/T = 1
nn0t = 1 with T=2p/w0 = 1/n0

nn0 = 1/t
= 1 kHz

n0 = 1/T
= 250 Hz

n =4

Plugging in several values of n, we get:

n 0 ±1 ±2 ±3 ±4 ±5 ±6 ±7 ±8 ±9
|cn|/A 0.250 0.225 0.159 0.075 0 0.045 0.053 0.032 0 0.025
< cn 0 0 0 0 - p p p - 0

7
 |cn|/A
0.25

-4w0 -2w0 -w0 0 w0 2w0 3w0 4w0 kw0

p < cn

-4w0 -2w0 -w0 0 w0 2w0 3w0 4w0 kw0

Note: The phases at -3w0 to 3w0 are zero, but the phases at -4w0 and 4w0 are
not defined. There is a difference!
Plots of the magnitude and phase of cn are called magnitude and phase spectra.
Now we are in a good shape to do some Fourier series mental gymnastics.

Here are some questions:

How do the magnitude and phase spectra change when we
• Increase A to 2A
• Increase T to 2T but leave t/T =1/4
• Increase t/T from ¼ to 1/2
• Decrease t/T from ¼ to 1/8

One of these gymnastic exercises is of particular interest:

Example: Unit impulse train

t®0, At = 1 , T=const.

y(t)

-2T -T 0 T 2T 3T t

8
 +∞
y(t) = ∑δ (t − nT)
n=−∞

A nπτ
cn = lim
τ →0,Aτ =1 nπ
(sin
T
)

A nπτ
c n = lim ( )
τ →0,Aτ =1 n π T
1
cn =
T
+∞
1
y(t) = ∑ e inω 0 t
T n=−∞

|cn|
€
1/T

-2w0 -w0 0 w0 2w0 kw0

< cn

Note:
For a periodic signal, the Fourier series is an accurate expression for all time even though
the integration for the computation of the coefficients is carried out over only one period.

… ……..

T
For a non-periodic signal, the fourier series is an accurate expression only over the time
interval assumed to be one period.

9
1.3 Fourier transform

The exponential Fourier series (FS) is an extremely useful technique for the
representation of periodic signals. They are also used for non-periodic signals for specific
time intervals, or more generally, variable intervals, outside which the accuracy of the
representation is unimportant.

The Fourier transform (FT) is used for the representation of a non-periodic signal that is
valid for all time, or more generally, for the whole range of the variable.

The FT is obtained from the exponential FS via a limiting argument.

Let us assume that we have an arbitrary function given below.

y(t)

This is a non-periodic signal. To derive the FT from the FS we want to first consider the
periodic version of this signal, yT(t), with period T, and its (assumed) spectrum, cn,T as
sketched below.

yT(t)

t
T

cn,T

0 w0 kw0

Hopefully we are sufficiently fit through our previous Fourier gymnastics that we can
now answer the following question:

10
 What happens with the spectrum if we increase the period T but leave the shape of the
pulses unchanged?

If the period T is increased, the fundamental frequency, w0, decreases, the spectrum
becomes denser, but the shape of the envelope of the spectrum remains (save for a scaling
factor) unchanged.

In the limit:
T®¥,
fT(t) ® f(t)
cn,T ® Y(w)
discrete spectrum ® continuous spectrum

y(t)

Y(w) t

0 w

1 +∞
y(t) = = ∫ Y (ω )e iωt dω
2π −∞
+∞
Y (ω ) = ∫ y(t)e −iωt
dt
−∞

y(t) and Y(w) are called: FT pair : y(t)« Y(w)

€
Alternative expressions, with w= 2pn, are:
+∞
y(t) = ∫ Y (ν )e i2 πνt
dω
−∞
+∞ (should be d𝜈)
Y (ν ) = ∫ y(t)e −i2 πνt
dt
−∞

11
€
 In order for the FT to exist, we must have Y(w) < ¥.
How can we find out?
1) Evaluate the integral
2) Consider Dirichlet conditions

Dirichlet Conditions:
If
1)
+∞

∫ y(t) dt < ∞
−∞
or
+∞
2
∫ y(t) dt < ∞
€ −∞
and

2) y(t) has a finite number of maxima and minima in any finite interval, and
€ 3) y(t) has a finite number of discontinuities in any finite interval

then FT exists.

Note: If Dirichlet conditions are not fulfilled, then FT could perhaps still exist.

Examples

Here are four examples that do and do not fulfill the Dirichlet conditions:

y(t) 1, |t| £1/2

1 y(t)=P(t) ={
0, |t| >1/2

y(t)=u(t)e-at, a>0

y(t)=cos w0t

y(t)=u(t)

12
 Which do and which do not fulfill the Dirichlet conditions? Here is the answer: The first
two do and the last two do not.

Now, let us compute the FTs.

1st example: FT of the gate function

y(t) A, |t| £t/2

A y(t)=AP(t/t) ={
0, |t| >t/2

-t/2 0 t/2 t

+τ / 2
Y (ω ) = A ∫e −iωt
dt
−τ / 2

A −iωt +τ / 2
Y (ω ) = − e
iω −τ / 2

τ τ
A iω 2 −iω 2
Y (ω ) = (e − e )
iω
2A ωτ
Y (ω ) = sin
ω 2
τ
sin ω
Y (ω ) = Aτ 2
τ
ω
2
Y(w) is a real function and has zero-crossings at wt/2 = ±p, ±2p, ±3p, …
That means that we have zero-crossings at w=±2p/t, ±4p/t, ±6p/t, …
€
Re{Y(w)}

-6p/t -4p/t -2p/t 0 2p/t 4p/t 6p/t w

Im{Y(w)}

13
 We can also plot the magnitude and phase spectra:

|Y(w)|

-6p/t -4p/t -2p/t 0 2p/t 4p/t 6p/t w

ÐY(w)
p

-6p/t -4p/t -2p/t 0 2p/t 4p/t 6p/t w

sin x
The function of the form plays an important role in signal theory.
x
sin πx
sinc(x) =
πx
Sinc function:€ t ωτ
Π( ) ↔ τ sinc( )
τ 2π
1 sinc(x)

-3 -2 -1 0 1 2 3 x

t
For the gate function: Π( ) we can now write:
τ
t ωτ
Π( ) ↔ τ sinc( )
τ 2π
€

14
 2nd example: FT of the exponentially decaying function

A y(t)=Au(t)e-at, a>0

0 t

+∞
Y(ω ) = A ∫ e−( a +iω )t )dt
0

A +∞
Y(ω ) = − e−( a +iω )t
a + iω 0

A
Y(ω ) =
a + iω
A
Y(ω ) = 2 (a − iω )
a + ω2

A 1
Y (ω ) = 2 2
(a 2 + ω 2 ) 2
a +ω
A
Y (ω ) = 1
(a 2 + ω 2 ) 2

−ω
∠Y (ω ) = tan−1
a
ω
∠Y (ω ) = −tan−1
a

|Y(w)|
A/a
€

0 w

ÐY(w)
p/2

-p/2

15
 The two other functions do not fulfill the Dirichlet conditions, however, their FT exists
anyway. For the computation of the FT of these two other functions we need the
d-function which is also called the unit impulse function:

+∞

∫ δ(t)dt = 1
−∞

Properties of the d-function

1) Shifting property or sampling property:

€ +∞

∫ f (t)δ (t)dt = f (0)

−∞
+∞

∫ f (t)δ (t − t 0 )dt = f (t 0 )
−∞
+∞

∫ f (t − t1 )δ (t − t 2 )dt = f (t 2 − t1 )
−∞

2) Scaling property:
+∞
€ 1
∫ f (t)δ (at)dt =
a
f (0)
−∞

3) derivative property:
+∞
€
∫ f (t)δ (n ) (t − t 0 )dt = (−1) n f (n ) (t) t= t 0
−∞

Now we can compute the FT of the d-function:

€ +∞
FT{δ (t)} = ∫ δ (t)e −iωt
dt
−∞

FT{δ (t)} = 1

And since FT is unique, d(t) is the inverse FT of 1

+∞
1
€ δ (t) =
2π
∫ 1e iωt
dω
−∞

δ (t) ↔1

€ 16
 d (t) FT{d (t)}
1

0 t 0 w

Note: This result can also be obtained through a limiting argument;

P (t) Y(w) = FT{P (t)}

A At=1 At=1

-t/2 0 t/2 t 0 w

FT of a constant:
+∞
FT{1} = ∫ 1e−iωt dt
−∞

this is not absolutely integrable. d-function is needed. Also we have to do a change of

variables to be able to use a previous equation.
€ t = - x => dt = -dx, interval boundaries need a sign change too.
−(+∞)
FT{1} = ∫ 1e −iω (−x )
(−dx)
−(−∞)
+∞
FT{1} = ∫ 1e iωx dx
−∞

FT{1} = 2πδ (ω )
1↔ 2πδ (ω )

y(t) Y(w)
1 2pd(w)
€

0 t 0 w

Now it gets really exciting since we can now compute the FTs of the other two example
functions.

17
 3rd example: FT of the cos function
1
FT{cosω 0 t} = FT{e iω t + e−iω t } 0 0

2
1 +∞ −i(ω −ω )t
FT{cosω 0 t} = [ ∫ e dt + e−i(ω +ω )t dt]
0 0

2 −∞
And again with change of variables: t = -x
1 +∞ i(ω −ω 0 )x
FT{cosω 0 t} = [ ∫ e dx + e i(ω +ω 0 )x dx]
€
2 −∞
FT{cosω 0 t} = πδ (ω − ω 0 ) + πδ (ω + ω 0 )

y(t)=cos w0t Y(w)

€

t -w0 0 w0

Note:
The FT was originally defined for non-periodic signals. Now we find FT’s also for
certain periodic signals through the use of the d-function. In fact we can find the FT for
any general, periodic function! The procedure is to write down the complex exponential
FS for a function f(t) and then take the FT of the series on a term to term basis.

+∞
FT{y(t)} = FT{ ∑ c n e inω 0 t }
n=−∞
+∞
FT{y(t)} = ∑ c FT{e
n
inω 0 t
}
n=−∞
+∞
FT{y(t)} = 2π ∑ c δ(ω − nω )
n 0
n=−∞

Example: Train of unit impulses

+∞

€ y(t) = ∑δ (t − nT)
n=−∞
+∞
1
y(t) = ∑
T n=−∞
e inω 0 t (see p.9)

€ +∞
2π
Y (ω ) =
T
∑δ(ω − nω ) 0
€ n=−∞
18
 +∞
Y (ω ) = ω 0 ∑δ(ω − nω ) 0
n=−∞
y(t) Y(w)

€
0 t 0 w

4th example: FT of the step function

1 1
u(t) = + sgn(t)
2 2

First we compute the FT of the sign function through a limiting argument.

sgn(t)

e-atu(t), a³0
t
-eatu(-t)

sgn(t) = lim[e−at u(t) − e at u(−t)]

a →0
∞ 0
FT{sgn(t)} = lim[ ∫ e−at e−iωt dt − ∫e at −iωt
e dt]
a →0 0 −∞
∞ 0
FT{sgn(t)} = lim[ ∫ e−(a +iω )t dt − ∫e +( a−iω )t
dt]
a →0 0 −∞

1 ∞ 1 0
FT{sgn(t)} = lim[− e−( a +iω )t − e(a−iω )t ]
a →0 a + iω 0 a − iω ∞

19
 1 1
FT{sgn(t)} = lim[ − ]
a →0 a + iω a − iω
2
FT{sgn(t)} =
iω
2
sgn(t) ↔
iω

Now we can compute the FT of the step function.

€ 1 1
FT{ } = 2πδ (ω )
2 2
1
FT{u(t)} = πδ (ω ) +
iω
ℜeFT{u(t)} = πδ (ω )
1
ℑmFT{u(t)} = -
ω

u(t) |U(w)
€

0 t 0 w
ÐU(w)
p/2

-p/2

1.4 Properties of the Fourier transform

Symmetry property:

If f(t) ) « F(w) then

F(t) ) «2pf(-w)

Example:
sgn(t) «2/(iw)
20
 2/(it) «2p sgn(-w)
i/(pt) «sgn(w)

Linearity property:

If f1(t) « F1(w) and

f2(t) « F2(w), then for arbitrary constants a, b
af1(t) + b f2(t) « aF1(w) + b F2(w)

Note: This is very useful, since it means that you can compute the FT in steps.

Time-shifting property:

If
f (t) ↔ F(ω )
then
f (t − t 0 ) ↔ F(ω )e−iωt 0

Note: If
F(ω ) = F(ω ) e iθ (ω )
€
f (t − t 0 ) ↔ F(ω ) e i(θ (ω )−ωt0 )

A shift of t in the time domain leaves the magnitude spectrum unchanged,

but the phase spectrum acquires an additional term - wt0, see, for instance:
€
cosω (t − t 0 ) = cos(ωt − ωt 0 )
Example:

€ f(t) ÐF(w) |F(w)|

ÐF(w)=q(w)
=0

f(t-t)0 ÐF(w) |F(w)|

0 t0 0 ÐF(w)=
-wt0
21
 Frequency-shifting property:

If
f (t) ↔ F(ω )
then
f (t)e iω 0 t ↔ F(ω − ω 0 )

Example 1:
€ f(t) F(w)

t -wm wm w

iω 0 t
f (t)e
w0 -wm w0 w0 +wm

Example 2:
f(t) F(w)

t w

f (t)cos(ω 0 t)
-w 0 w0

In radio technology you often need to translate a spectrum to a different frequency range,
e.g. baseband ® IF ® RF or RF ® IF ® baseband. This is achieved through the use of
up and down-converters, or mixers.

22
 mixer
cosw0t f(t) cosw0t

f(t)

Time-differentiation property

If
f (t) ↔ F(ω )
then
d
f (t) ↔ iωF(ω )
dt
€
dn n
n
f (t) ↔ (iω ) F(ω )
dt

€ Time-integration property

If
f (t) ↔ F(ω )
then
t
1
∫ f (x)dx ↔
iω
F(ω ) + πF(0)δ (ω )
−∞
€

Scaling property
€
If
f (t) ↔ F(ω )
then for a real constant b

1 ω
f (bt) ↔ F( )
€ b b
23
 Example:

f(t) F(w)

-t/2 0 t/2 t 0 w
2p/t
1/t n

f (2t) 1/|b| F(w/b)

-t/4 0 t/4 t 0 w
4p/t

2/t n

1.5 The two-dimensional Fourier transform

So far we have used the variables, t and w or n in the context of FTs. These variables,
time and frequency (radians per second and cycles per second) stand for physical
quantities that are one-dimensional. However, in cases which are two-dimensional, an
antenna , arrays of antennas, brightness distributions at the sky, pictures on a TV screen,
etc., variables that describe two-dimensional quantities need to be used.

A two-dimensional function f(x,y) has a two-dimensional Fourier transform F(u,v) with

+∞ +∞
f (x, y) = ∫ ∫ F(u,v)e i2 π (ux +vy )
dudv
−∞ −∞
+∞ +∞
F(u,v) = ∫∫ f (x, y)e−i2 π (ux +vy )dxdy
−∞ −∞

If x, y are spatial coordinates like angles, then u, v are called spatial frequencies.

24
 Example (only qualitative):

Y v

f(x,y) ÐF(u,v)
|F(u,v)|

X u

Properties of the two-dimensional Fourier transform

The properties of the two-dimensional FT are very similar to those of the one-
dimensional FT. The most important for our purposes are:

Linearity property:
af1 (x, y) + bf 2 (x, y) ↔ aF1 (u,v) + bF2 (u,v)

Shifting property:
€ f (x − a, y − b) ↔ F(u,v)e−i2 π (au+bv )

Modulation property:
€ ω0
f (x, y)e iω 0 x ↔ F(u − ,v)
2π
Scaling property:
1 u v
f (ax,by) ↔ F( , )
€ ab a b

The extension to more than two dimensions is straight forward.

25