Nyquist Sampling, Pulse-Amplitude Modulation, and Time Division Multiplexing on Mac

12. Nyquist Sampling, Pulse-Amplitude Modulation, and TimeDivision Multiplexing

Many analogue communication systems are still in wide use today. These include AM,
FM, and PM systems. If analogue signals are to be transmitted digitally, they have to be
converted to discrete samples. The conversion of an analogue signal into a discrete-time
sampled signal is accomplished by sampling the analogue signal at regular time intervals
Ts. Ts is called the sampling period and fs = 1/ Ts is known as the sampling rate.
Definition [1]. A signal m(t) is called a band-limited signal if
M(f) = 0

for | f | > fm Hz

(12.1)

where fm is the highest-frequency spectral component of m(t).

Consider a periodic rectangular waveform s(t) of period Ts = 1/fs, unit amplitude, and
pulse width . The trigonometric Fourier series of s(t) is

s(t)

sin 2nfs / 2

= T + 2 (
) cos 2nfst
2nfs / 2
T s n =1
s
c
0
2
c cos 2nfst
=T +
T s n =1 n
s

sin nd
cos 2nfst
= d + 2d
nd
n =1

(12.2)
(12.3)
(12.4)

where c 0 = , c n = sin nd , and d =

T s . The corresponding exponential
nd
Fourier series is

s(t) = d
n =

sin nd j 2 nfs t
e
nd

(12.5)

If we multiply m(t) by s(t), we obtain

sc(t)

= m(t) s(t)

= m(t) (d + 2d
n =1

= dm(t) (1 + 2
n =1

sin nd
cos 2nfst )
nd
sin nd
cos 2nfst )
nd

(12.6)

s c (t) consists of the component m (t) and an infinite number of DSB signals at
12.1

Nyquist Sampling, Pulse-Amplitude Modulation, and Time Division Multiplexing on Mac

sampling frequencies fs, 2fs, 3fs, ...

Sc(f) = dM(f) + d
n =1

The Fourier transform of sc(t) is

sin nd
[M(f - nfs) + M(f + nfs)]
nd

(12.7)

Figure 12.1 shows the waveform and spectra associated with signal sampling.
Figure 12.1

Waveform and spectra associated with signal sampling.

Figure 12.2 shows what happen if fs = 2 fm and fs < 2 fm.

Figure 12.2 Signal spectra for (a) fs = 2fm and (b) fs < 2fm.
Since the bandwidth of m(t) is fm , we see that the spectra do no overlap if fs > 2 fm
and the spectrum associated with the signal m(t) can be separated from others using a
low-pass filter with a cutoff frequency of fm . When fs < 2 fm , the spectra overlap.
Since the frequency content in these regions of overlap adds, the signal is distorted. The
distortion is called aliasing and it is no longer possible to recover m(t) from its sample
values by low-pass filtering.
Sampling Theorem [2]
Let m (nT s ) be the sample values of m (t) where n is an integer. The sampling
theorem states that the signal m (t) can be reconstructed from m (nT s ) with no
distortion if the sampling frequency fs > 2f m . The minimum sampling rate 2f m is
called the Nyquist sampling rate.
Proof.
Since M(f) is a non-periodic bandlimited function, we can make a new function which is
periodic at frequency fs but not overlapping in the frequency-domain. Let the periodic
frequency function be M p(f), as shown in Figure 12.3. M(f) is a band-limited version
of M p(f),
Figure 12.3 M(f) represented as a periodic frequency function.
The exponential Fourier series of Mp(f) is
M p(f) =

cn ejnt0f
f s n =

(12.8)

where fs > 2fm and

12.2

Nyquist Sampling, Pulse-Amplitude Modulation, and Time Division Multiplexing on Mac

t0 = 2/fs

(12.9)

The coefficients cn are given by

f s /2
cn =

M p (f)e -jnt0 fdf =

f s / 2

f s /2

-jn(2
M p(f) e
f s / 2

/fs)fdf

(12.10)

However, the Fourier transform tells us that

f s /2

m(t) = F-1[M(f)] = M(f) e j2ftdf = M p (f) e j2ftdf

f s / 2
(12.11)
n
Comparing equations (12.10) and (12.11), we see that, if t = f , we obtain
s
n
cn = m( f )
s

(12.12)

This says that we can obtain each c n from the sample value of m(t) at time t =
n
f . Once c n is known, we can obtain M p (f) from equation (12.8), and once
s
Mp(f) is known, we can obtain m(t) from equation (12.11).
Substituting cn into equation (12.8), we get
M p(f) =

n
1

m( f ) ejnt0 f
f s n =
s

Substituting this expression for Mp(f) into equation (12.11), we get

f s /2

m(t)

n
1

m( f ) ejnt0 f] e j2ftdf
[f
s
f s / 2 s n =
f s /2

n
1
jnt f
=
m( f ) e 0 e j2ft df

f
n = s f / 2
s
s
sin[ fs ( t n )]

fs
n
m( f )
=
n
n =
fs ( t )
s
fs

= F-1[M(f)] =

12.3

(12.13)

Nyquist Sampling, Pulse-Amplitude Modulation, and Time Division Multiplexing on Mac

n
sin [ fs (t )]
fs
f s (t n )
fs

n
m( )
=
f
n =
s

(12.14)

Weighting factor
Equation (12.14) shows that each sample is multiplied by a weighting factor.
Signal Reconstruction [2, 3]
The process of reconstructing an analogue signal m(t) from its samples is known as
n
interpolation. How do we reconstruct m(t) from its samples m( f )? Consider
s
the sample signal of m(t) shown in Figure 12.4.
Figure 12.4 Samples of m(t).
n
Let M n (f) be the Fourier transform of the n-th sample m( f ). If << 1/ fs , m (t)
s
can be assumed to be constant over the sampling time and

/2
/2
n -j2f(n/fs)
n
m( f )e -j2ftdt m( f )e
dt
s
s
/ 2
/ 2
n -j2f(n/fs)
M n(f) m( f )e
s

M n(f) =

(12.15)

From our knowledge of basic PAM theory, we can recover the analogue signal using a
low-pass filter with a cutoff frequency of fs/2 (> fm ). Assume that we have an ideal
low-pass filter whose transfer function is H(f) = K e -j2 ftd , where K is a constant
and td is a time delay. Without loss of generality, we set the filter gain K = 1 and the
filter delay td = 0.
n
Let g n (t) be the filter output response to the n-th input sample m( f ). The Fourier
s
transform of gn(t) is
Gn(f) = H(f)M n(f) = M n(f)
and

12.4

Nyquist Sampling, Pulse-Amplitude Modulation, and Time Division Multiplexing on Mac

gn(t)

f s /2

j2ft
= G n(f) e
df =
Gn(f) ej2ft df

f s / 2
sin[ fs ( t n )]
fs
n
= m( f )fs
n
fs ( t )
s
fs

(12.16)

For a linear ideal filter, the filter output response to all input samples is just the sum of the
filter output to each input sample, or
g(t)

gn(t)
n =
sin[ fs ( t n )]

fs
n
= fs
m( )
f
n
n =
fs ( t )
s
fs
= fsm(t)
=

(12.17)
(12.18)

Equation (12.17) yields values of g(t) between samples as a weighted sum of all sample
values. g(t) is not only defined at the sampling instants, but it is proportional to m(t)
at all instants of time. This is shown in Figure 12.5.
Figure 12.5 Filter response to input samples.
Practical Sampling Frequency and Pulse-Amplitude Modulation
________________________________________________________________________
Table 12.1 Practical sampling frequency values for audio and broadcast signals
________________________________________________________________________
Signal
fm
fs > 2fm
Actual sampling frequency fs
________________________________________________________________________
Audio
3.3 kHz
> 6.6 kHz
8 kHz
> 40 kHz
44.1 kHz
Music
20 kHz
> 8 MHz
TV
4 MHz
________________________________________________________________________
The sampling theorem is very important because it allows us to replace an analogue signal
by a discrete sample and reconstruct the analogue signal from its sample values. It opens
doors to many new techniques of communicating analogue signal by samples. A system
transmitting sample values of the analogue signal is called a pulse-amplitude modulation
(PAM) system and is shown in Figure 12.6.
Figure 12.6 PAM system.
12.5

Nyquist Sampling, Pulse-Amplitude Modulation, and Time Division Multiplexing on Mac

Time Division Multiplexing (TDM)

One of the basic problems in communication engineering is the design of a pulse
communication system which allows signals from many users to be transmitted
simultaneously over a single communication channel. We see from the sampling process
that, with << Ts, there is a time gap between two consecutive samples in a single-user
PAM system. Suppose that we have several different signals of the same or different
bandwidth. If we sample the signals in a sequential manner, we can put the samples in the
time gaps. All these signal samples can now be transmitted along a single communication
channel. At the receiving end, the signals can be separated and recovered. We now have a
time-multiplexed system. Such a multiplexing technique is called time division
multiplexing (TDM). It permits the simultaneous transmission of several signals on a
time-shared basis. Figure 12.7 shows the transmitter, the receiver, and the spectrum of a
5-user TDM PAM system with fs = 8000 samples/s.
Figure 12.7 TDM system.
References
[1]

H. P. Hsu, analog and Digital Communications, McGraw-Hill, 1993.

[2]

M. Schwartz, Information Transmission, Modulation, and Noise, 4/e,

McGraw-Hill, 1990.

[3]

M. S. Roden, Analog and Digital Communication Systems, 3/e, Prentice Hall,

1991.

12.6

Nyquist Sampling, Pulse-Amplitude Modulation, and Time Division Multiplexing on Mac

m (t )

M (f )
t

Envelope

s (t )
1
...

...

-1

-T s Ts
2 2

1
Ts
0 fs

2 fs

Sc ( f )

M (0)
Ts

sc ( t )

0 fm
cn

-1

0f

2 fs

Figure 12.1 Waveform and spectra associated with signal sampling.

12.7

1
Ts
Guard band

Envelope

Nyquist Sampling, Pulse-Amplitude Modulation, and Time Division Multiplexing on Mac

Sc ( f

Envelope

M (0)
Ts

...
1
-

...
0

2f s
f s =2 f m

(a)

Sc ( f )

M (0)
Ts

Envelope

...
1
-

...

2f s
f
Regions of overlap s < 2 f m
0

(b)
Figure 12.2 Signal spectra for (a) fs = 2fm and (b) fs < 2fm.

Mp ( f )
M (f )
...

...

- fs

f s -f m 0
2

fm

fs
2

f
fs

Figure 12.3 M(f) used to make a periodic frequency function.

12.8

Nyquist Sampling, Pulse-Amplitude Modulation, and Time Division Multiplexing on Mac

m (t )

m (t )
Samples of m ( t )
<< T s
t

Ts

Figure 12.4 Samples of m(t).

g n (t )
f m (n )
s
fs
Reconstructed signal
g n +1 ( t )

g n (t )
f m (n )
s
fs
1

0 n -2
fs

n -1
fs

n +1
fs

n
fs

f m (n )
s
fs
t
n +2
fs

Figure 12.5 Filter response to input signal samples.

LPF

m (t )

sc ( t )

Sampler

PAM signal

//

Transmitter
Figure 12.6 PAM system.

12.9

~
~

Recovered
signal
Receiver

Nyquist Sampling, Pulse-Amplitude Modulation, and Time Division Multiplexing on Mac

m 1(t )

^ (
LPF m
1 t )
~
~ m^ ( t )
2
~

f s = 8000 samples/s
sc ( t )
//
Composite
PAM signal

m (t )
2

Source

~
~

...

...
m (t )
5

^ (t )
m
5

Destination
(a) Transmitter and receiver

sc ( t )

125 s
25 s

...
0

...

3
1 2

4 5 1 2
(b) Waveform of TDM signal
Figure 12.7 TDM system.

12.10