EE306: Digital Communication

Prabhat Kumar Upadhyay

Professor
Department ofElectrical Engineering
Indian Institute of Technology Indore

Major Topics
Towards 6G Networks: Motivation and Challenges
Block Diagram ofa Digital Communication System
Signal Space Representations
1 Vector Space Concepts
2 Signal Space Concepts
3 Gram-Schmidt Orthonormalization Procedure
Block Diagram ofa Digital Communication System

Transmitter
Transmitter

Information
Information
Source
Source Source
Source Channel Digital
&& Input Coding Coding Modulation
Transducer

Noise Channel
(Noisy)

Output Source Channel Digital

Output Transducer Decoding Decoding Demodulation
Signal

Receiver
Receiver Discrete
Channel

(II T I iicl c›r o) January Z, 202a 4 / 29

Signal Space Representations (Signals as Vectors)
It is possible to represent the signals ina digital communication system as
vectors ina linear space or vector space, much like the familiar 3-D
vectors in our physical world.
We demonstrate that signals have characteristics that are similar to
vectors and developa vector representation for signal waveforms.
We begin with some basic definitions and concepts involving vectors.

Vector Space Concepts:-

A vector X in an n-dimensional space (Euclidean space) is characterized

by its n components: N = [ml, 2 i - , nl -
It may also be represented asa linear combination of unit vectors or basis
vectors e;,1 i n.

(i)
i=l
Where, by definition,a unit or basis vector has length unity and z; is the
projection of the vector X onto the unit vector e;.
The inner product of two n-dimensional vectors, Hi = i 11i 12 lnl
and +2 = 1 21 22 2.l^ is defined as

_ (2)

where A is conjugate transpose (Hermitian) of A;.

 Two vectors Hi and <2 are orthogonal if Al <2 0
A set of m vectors <#.$ = *.- .. , m, are orthogonal if
X¡ - X ——0 for all1 i, m, and i j

The norm ofa vector N is defined as

(3)

which is simply its length.

A set of m vectors is said to be orthonormal if the vectors are orthogonal
and each vector hasa unit norm.
A setof m vectors is said to be linearly independent if no one vector can
be represented asa linear combination of the remaining vectors.
Two n-dimensional vectors% and <2 satisfy the triangle inequality

(4)
with equality if Hi and +2 are in the same direction
i.e., Al = ++2› wherek isa positive scalar.
From thetriangle inequality, there follows the Cauchy-Schwartz inequality

with equality if Hi => 2-

The norm square of sum of two vectors is

(6)

2 2 2 (y)
n —h - | n || || h|| — 2 Re(n“b}
(g)

IfHi and +2 areorthogonal, then Hi <2 - 0 and

(9)
Signal Space Concepts

As in the case of vectors, we may developa parallel treatment fora set of

signals.
The inner product of two complex-valued signals >1 I/) and +2 I/) is
defined as

(10)

=0 if they are orthogonal (zero correlation)

It is also called “Correlation” of «u (/) with +2(/) . Intuitively, the

correlation measures how closely the two signals resemble one another.
The norm ofa signal is defined as

|s(t)| 2dt —— , (jj)

Hence, the length ofa signal is the square root of its energy.
Example

2
II=i(/)II = I=i(/)| dt
1 2 1/2
2 2
2 1/2 || z2() || - 1 d/ —i— 1 d/
2 0 1
- 1 dt
0
 1 2
1$): 2( )) = 1(1)dt+ 1(—1)dt
0 1 (12)
1 - 1 —1 -0
So, <1(/) and <2(/) are orthogonal.

Energy ofa signal z(I) is

|x:(/)|2d/ (13)

i.e., when mlI ) and +2() areidentical( z(I)), the correlation reduces to the
energy.

The Cauchy-Schwartz inequality for two signals is

1/2 1/2
2 2
|s1(')| dt |s2(')| dt (14)
Correlation coefficient between mlI/) and >2( ) is defined as

( 1 (*)› 2 (*)

where8 cos 1 I#12) is angle between »i( ) and +2I ) -

Geometrical interpretation of inner product in the plane spanned by >1 and +2-

(16)
Projection onto the Signal Space

Whena digital transmitter transmits one of M possible signals

1$ )› , sM[!)I, it is common forthereceiver to project the received
signal r(I) onto the signal space spanned by the M signals. In this
setting, the signal spaceN is defined as the span of the M signals:

N = span (siI ),s2I ). , sM )l

Inother words,N is the set of all signals that can be expressed asa linear
combination of (siI ),s2( ). , sM ))-

An orthogonal basis for the signal spaceN = span (s1 I/)› , sM ) ÎS fi

minimal setofN orthonormal functions(81 I/)› 2 I )› › Q (t) with the
property that every element s(t)C N can be expressed asa linear
combination of basis functions:

(18)
i=l
 Since the m—th signalN (I) is an element of N, it too can be expanded
in terms ofa basis:
zr
s.-(•)= y s... «.‹•) ‹'9)
i—1
This expansion is most useful when theno.ofbasis functions is smaller
than the size of the original signal set i.e., N M. (As such, complexity
of receiver reduces...N correlators or MFs required for demodulation).
Thereby, any N-dimensional signal waveform6 (I) can be represented
geometrically asa point in the signal space spanned by the N
orthonormal basis functions (Q;(I)} 1.
Note thata set of functions(81 I/)› › Q (I)} is said to be orthonormal
when each signal has unit energy, and distinct signals are orthogonal, so
that
1, t = A
$,(t)$k(t)dt —— 6,d —— (20)
0, i 2^A
Naturally, there are two pertinent questions:
(1) How small can N be?
(2) What arethecorresponding N basis functions?
The answers to both questions can be found using the Gram-Schmidt
orthonormalization procedure.
i. L'i,i1a1i,it l\. t/ a,ir1liv,i RR3tJ(i Loc'tiiio /l1‹1e>. Hi' 2tJ25
Gram-Schmidt Orthonormalization Procedure:-

It isa systematic method forconvertinga set of M finite-energy signal

waveforms (st(t), , sM 1!) into an orthonormal basis(8i( ). . 8r( ))
of size N for their span, with N as small as possible.

The first orthonormal basis function is simply constructed as

(21)

Thus, 8i( ) is simplysi( ) normalized to unit energy.

This definesa one-dimensional subspace Si span {8i(/)}-

If we lets2 I/) denote the projection ofs2 I/) onto this subspace, then the
projection error will be orthogonal to £i› i.e.,s2 I/) — s2(/) -I AlI )

Thus, if we normalize the projection error to have unit energy, we

producea second basis function.
 To illustrate, the second basis function is constructed froms2 I/) by first
computing the projection ofs2 I/) onto 8i I/)›
which is
(22)
where /'21 = (s2I ). 8iI )) (23)
Then,s2I )= 2181() is subtracted froms2( ) to get projection error
S2 (*)' 2 t*)
— 21 1 t*) (24)

This waveform is orthogonal to 8iI ) but it does not have unit energy.
Illustration of Projection:-

A projection is illustrated in Fig. for the 3-D Euclidean space, where the
subspaceN is the plane formed by z-axis and p-axis and N is an arbitrary
vector. The projection is the result of droppinga line from X down tothe
plane (this is the dashed line). The resulting vector (N — X) is the vector
shown parallel to the dashed line. It is orthogonal to the plane N, and hence
to every vector in N.

The vector inN closest to A is evidently N; any other vector inN is farther
from A.
 2
If 2 I*) have the energy A 2 J dt then the normalized waveform
that is orthogonal to 81() is

(25)
S2 t*)— S2 t*)
2

Now, 81(/) and 82 I/) are orthogonal and have unit-energy.

Similarly, the k—th basis function can be obtained as

(26)

(27)

#‹ = Projection ofsk 1!) onto Q;(t) = (Sk ) 8: )

and
2
E k
—— Energy ofsk I) = |st (t)| dt

The procedure is continued until all the M signal waveforms (s;(t) have been
exhausted and N M orthonormal waveforms have been constructed.
If the numerator in (25) is zero, then we infer thats2(/) * NJ and so the
projection error is zero. We then renumber the signals by movings2 I/) to
the end of the list.

The dimensionality N of the signal space will be equal to M if all the

signal waveforms are linear independent i.e., none of the signal waveforms
isa linear combination of the other signal waveforms.

Ifall the M — 1 projection errors are non-zero, then the signals are
linearly independent and the final number ofbasis functions N will be the
same as thenumber ofsignals, N —— M. Each zero projection error would
decrease the number ofbasis functions by one, yielding N M. The
dimension ofN is the number ofN basis functions.

The basis functions themselves are not unique, but the dimension N is
fixed and independent of the basis.
Example

Consider the four waveforms as shown inFig.

Determine the dimensionality of the waveforms anda setof basis
functions.
Use the basis functions to represent the four waveforms ina signal-space
diagram.

0
—1

O s ”t
 (28)

(29)

(30)
(3î)

Where, +'21 = 2 ) 1$ )dt (32)

1 2
dt —i— (—1) dt (33)
0

|0] = 0 (34)
 |S2(t)| dt (35)
2
—— dt 2 (36)
0

S2(/)_ 1
2 (*)
' (37)

r. P r ab h at K . U p ad hyay E R 3 0 6 L e ct ure S 1id es, H pri ng 2 0 2 5

 (38)

31' 3 (*)1 (*) (39)

2 i 1
= 1x dt—— (2) - v'Ž (40)
o 2
32 3 (*) 2 (*) 0 (41)

S3(‹) S3(t)— v'ŽØ1(‹) (42)

2
ńî3 = |S3(t)| dt (43)
3
- 12dt ——1 (44)
2

Dr. P r ab h at K . U p ad hyay E R3 0 6 L e ct ure S 1id es, H pri ng 2 0 2 5

 3 (*) ' (45)

'4(') ' 4 ) 41 1 ) 42 2() 43 3() (46)

2 i 3
dt—I— 0 dt (47)
o 2 2
(48)

r. Pr abh at K . Up ad hyay E R306 L ectureS1id es

, H pr
i ng 2025
 (49)
3
43' (—1)(—1)d/ - 1 (50)
2

(52)
Since £4() isa linear combination of 81(/) and 83(*)› 4() = 0.
So, the three orthonormal functions are:

4›t‹)

( I I T I itetcr o) January A 202a 2?/20

Once we have constructed the set of orthonormal waveforms( (t) , we can
express the M signals $k I!) as a weighted linear combinations of the( (I)}
Thus,

(53)
n=l
(T4)

2 2
(55)
— on n=l

Thus, from (53), each signal may be represented by the vector

(56)

Or equivalently, asa point in the N- dimensional signal space with

co-ordinates($#:.' 1, 2, , N . The energy in the \-th signal is simply the
square of the length of the vector or, equivalently, the square of the Euclidean
distance from the origin to the point in the N- dimensional space.
Thus, any signal can be represented geometrically asa point in the signal
space spanned by the orthonormal functions( (I)}.

i(*)' 2$i(*)
* Si= (v', 0, 0)

S2 ' (0,V !, 0,)

S3 (v', 0, 1)

Their lengths are -

||Si II = 1, ||S2 ' 2, ||S3 = V'2 —F1 - v', ||S4 ' 3 (57)
2
and the corresponding signal energies are +# = || S# 11 k 1, 2, 3, 4.
Important Points

The orthonormal basis functions (Qq(t)} obtained from the GS-procedure

are not unique.

Ifwe change the order (e.g., if we start with $3 I*) instead of $1() ), the
orthonormal basis functions (Qq(t)} will be different and the
corresponding vector representation will depend on the choice of (Qq(I)}.
But the vectors (S#) will retain their geometrical confirmation and their
lengths will be invariant to the choice of (Qq(t)}.

Differences with FS
(1)Exact instead of FS approx (first few terms dominate)
(2) Q, can be any type of functions but in FS, it is sinusoidal ore "

phasors.
Parseval's relation:
Energy remain same ineither form(+#(*) orS#)-