KING ABDULLAH UNIVERSITY OF SCIENCE AND TECHNOLOGY

CEMSE DIVISION ELECTRICAL ENGINEERING PROGRAM

EE 242 Digital Communications and Coding
Pulse Amplitude Modulated (PAM) Signals

In digital PAM, the signal waveforms can be expressed as

sm (t) = Am g(t) 0≤t≤T (1)

where g(t) is the pulse shape of duration T and Am is one of M = 2k possible amplitudes of the
signal. A typical 4-PAM signal is shown in Figure 1.

Figure 1: 4-PAM Signal

The signal described above is called baseband PAM. In many cases we modulate the waveform by
multiplying it by a cosine function with carrier frequency fc . The resulting signal is called bandpass
digital PAM or amplitude-shift keying (ASK). A typical ASK signal is shown in Figure 2.

Figure 2: ASK Signal

1
 In this section we will focus on ASK modulation. Here, the waveforms are given by

xm (t) = Am g(t) cos(2πfc t) 0≤t≤T (2a)

= Re Am g(t)ej2πfc t

(2b)

where, g(t) : a pulse with duration T; we mainly use a rect function.

fc : carrier frequency, the baseband signal will be shifted by it.
Am : m = 1, . . . , M.

If M = 2k , each symbol represents k bits. To represent ASK signals, we only need one basis
function in the form:
φ0 (t) = g(t) cos 2πfc t (3)
Next, we need to normalize this function, so we have to find its energy

1 T 2 1 T 2
Z Z Z
02
φ (t)dt = g (t)dt + g (t) cos 2π(2fc t)dt (4a)
2 0 2 0
Eg
= (4b)
2
(we don’t want to have analysis dependent on corner frequency)
⇒ where the 2nd part of RHS of Equation (4a) is ' 0, since fc  BW of g(t). So the final form of
the base function is given by the next equation

s
2
φ(t) = g(t) cos 2πfc t (5)
Eg

Using this basis we can represent each symbol with the following vector,
r r
Eg Eg
xm = Am since xm (t) = Am φ(t) (6)
2 2
Now we need to know how to choose the Am ’s. Usually we center them around the origin and make
them equi-distant. This makes the best use of energy. So we have (assuming M is even)

Am = (2m − 1 − M ) m = 1, 2, · · · , M

In the spacial case of M = 2, the PAM waveforms are s1 (t) = −s2 (t), so they have the same
energy. This type of modulation is called antipodal.

ˆ What is A m if M is odd?

2
 Figure 3: PAM Constellations

Distribution of the bits

Since M = 2k , each symbol is represented by k bits. How do we assign the bits to the symbols?
We use Gray coding (Two adjacent symbols differ by one bit only)

Figure 4: PAM Constellations with Gray Coding

Gray Coding: adjacent symbols are different in one bit only.

⇒ since most errors happen between adjacent symbols with high probability
⇒ Each symbol error results in one bit error

3
Demodulation
r
Eg
xm (t) = Am φ(t)
s 2
2
φ(t) = g(t) cos 2πfc t
Eg
Y (t) = xm (t) + n(t)
↓ find ID of Y (t)
Z ∞
Y = Y (t)φ(t)dt
−∞

Figure 5: PAM Demodulation

The scheme described in Fig. 5 will do the job, but it has one drawback, which is that we need to
design a receiver for each carrier frequency fc . The way around this is to downconvert the received
signal to an intermediate frequency fI , then demodulate as described in the Figs. 5 and 6.

4
 r ! s !
Eg 2
YI (t) = Am g(t) cos 2πfI t +n(t) (7)
2 Eg
| {z }
φ(t)

Figure 6: PAM downconversion followed by demodulation

Performance analysis:
ˆ Probability of error )
ˆ Rate (bits/symbol) Satisfaction
ˆ Average energy Resource
ˆ Minimum distance: d min )

ˆ Average number of neighbors at d min

Quick measure of prob. of error

ˆ Bandwidth Resource

5
Performance analysis of ASK:
Studying the performance of a modulation scheme involves characterizing the resources it needs in
terms of average energy and bandwidth and the quality of service it guarantees in terms of rate and
probability of error. Since the later might be difficult to characterize, we could study things like
minimum distance and average number of neighbors which acts as a surrogate of the probability of
error.

Average energy:

Assuming that the symbols are equiprobable, the average energy per symbol is given by,
M
1 X
Eavg = Em (8)
M m=1
M
1 X Eg 2
= A
M m=1 2 m
M
Eg X
= (2m − 1 − M )2
2M m=1
2Eg 2
= (1 + 32 + · · · + (M − 1)2 )
2M
Eg
= (M 2 − 1)
6
The average energy per bit is given by

Eavg (M 2 − 1)Eg
Eb = = (9)
k 6 log2 M

Minimum distance:

dmin = distance between any two adjacent points

= |xm+1 − xm |
r
Eg
= |Am+1 − Am |
2
p
= 2Eg
r
12 log2 M
= Eb
M2 − 1

6
Bit rate:
1
Let Rb = bit rate, Tb = bit interval, then the symbol rate Rs is given by
Rb
Rb
Rs = (10)
k

Average number of nearest neighbors:

The number of neighbors for the internal points is two and it is one for the two external points. So,
the average number of neighbors is given by

1 2(M − 1)
N= (2(M − 2) + 1 ∗ 2) = (11)
M M

BW requirement of PAM:

PAM waveform

xm (t) = Am g(t) cos 2πfc t DSB-SC signal

BW xm (t) = 2BW g(t)

⇒ BW efficiency = 50%
We can improve the BW efficiency by utilizing one of the following 3 approaches:

1) Use SSB Modulation

xm (t) = Am g(t) cos 2πfc t + Am ĝ(t) sin 2πfc t

where ĝ(t) is the Hilbert Transform

7
 xm (t) = Am g(t) cos 2πfc t − Am ĝ(t) sin 2πfc t

BWxm (t) = BWg(t) ⇒ BW efficiency= 50%

2) Use Baseband Transmission

xm (t) = Am g(t)
BW xm (t) = BW g(t)

Disadvantage is that we can only use it at a particular frequency band [0,BW g(t) ], i.e. we can not
multiplex signals.
BW efficiency = 100%

3) Use Quadrature Amplitude Modulation

φQAM (t) = g1 (t) cos wc t − g2 (t) sin wc t

BW g1 (t) = BW g2 (t) = B BW QAM = 2B
send two signals of BW = B using a BW = 2B
⇒ BW efficiency = 100%
More on this modulation scheme will be discussed later.

Decision regions:

Assuming that the symbols are equiprobable and AWGN channel, MAP rule reduces to a Minimum
Distance detector. Since the symbols have different energies we can’t use correlation to decide which
symbol was transmitted. Based on these assumption the decision boundaries will be the mid points
between the adjacent symbols.

Error Probability for PAM

For many modulation schemes, the distance between constellation points turns out to be a very
important quantity to calculate the error probability. For instance, the distance between two

8
constellation points xm and xn in PAM is given by :
dmn = kxm − xn k
r
Eg
= |Am − An | (12)
r 2
Eg
= |m − n|
2

R1 R2 RM −1 RM

x1 x2 xM −2 xM −1 xM

Figure 7: Decision regions for PAM

The received base-band signal is given by :

r = xm + n
r
Eg (13)
= Am + n
2
where n ∼ N 0, N20 .

Decisions regions are decided by the Euclidean distance between inner points ((M − 2) of them).
Hence, the error probability counting for inner points can be derived as follows
 
(in) 1
Pe = P |n| > dist(adjacent points)
2
 
dmin
= 2P n >
2
!
dmin /2
= 2Q p (14)
N0 /2
 
dmin
= 2Q √
2N0
r !
Eg
= 2Q
N0

For external points (2 of them), the error probability can be expressed as

 
(ex) dmin
Pe = P n >
2
r ! (15)
Eg
=Q
N0

9
Therefore, the error probability is given by :
1
2Pe(ex) + (M − 2) Pe(in)


Pe =
M
2 (M − 1)
r !
Eg (16)
= Q
M N0

d2 Eg
Using then fact that the constellation’s average energy is Eavg = 6
(M 2 − 1), we have :
s !
2 (M − 1) 6Eavg
Pe = Q
M (M − 1) N0
2
s !
2 (M − 1) 6 log2 (M ) Eb
= Q (17)
M (M 2 − 1) N0
s !
6 log2 (M ) Eb
' 2Q
(M 2 − 1) N0

where Eb = Eavg
k
= logEavg is the average energy per bit.
2 (M ) p
Another way of writing (16) is in terms of dmin = 2Eg and the average number of nearest
neighbors N = 2(MM−1) and we can write
 
dmin
Pe = N Q √
2No
This is an exact result. We will soon see however that this is a good approximation for most
modulation schemes.

10