Wireless Communication Systems

Module 4: Digital modulation and pulse shaping

techniques

Part 1
Main Reference: Rappaport Chapter 6

Introduction

Modulation: process of encoding information from a message

source in a manner suitable for transmission. This involves a
carrier signal, a parameter of which is varied in proportion to the
amplitude of the message signal (also referred to as the modulating
signal). The resulting signal is called the modulated signal.
The carrier signal is commonly either a single (very high)
frequency sinusoidal waveform, or a periodic pulse waveform. We
focus on the first….often referred to as continuous wave (CW)
modulation. It is suitable for passband systems such as wireless
communication systems. [The second is known as pulse
modulation. It is suitable for baseband systems such as Ethernet
cabled systems.]

CW modulation leads to the (baseband) message signal being

translated in frequency up to the (high frequency) spectral region
of the carrier signal. The carrier frequency is chosen to suit the
characteristics of the transmission channel….the aim is to have the
modulated signal placed within a low attenuation, minimal
distortion spectral region of the channel.

The carrier signal in CW modulation has the form

c(t)=acos(2πft+φ).
The message signal may be coded into the (i) amplitude a, (ii) the
frequency f, or (iii) the phase φ.
If the message signal is analogue then (i)=amplitude modulation
(AM), (ii)=frequency modulation (FM), (iii)=phase modulation
(PM).
If the message signal is digital then (i)=amplitude shift keying
(ASK), (ii)=frequency shift keying (FSK), (iii)=phase shift keying
(PSK). A fourth type of digital modulation is also used … the
digital message is coded into (amplitude,phase) pairs. This is
known as quadrature amplitude modulation (QAM).

Finally, demodulation is the process (carried out within the

receiver) of extracting the baseband message from the modulated
signal.

Because of its many advantages, digital modulation has become

the standard method used in modern communication systems,
including wireless communication systems. For this reason we
restrict our attention to digital modulation.Digital modulation –
an overview
Consider a bit stream b1,b2,b3,… where each bit can take on one of
two values (e.g. 0,1 or –1,+1). Digital modulation involves coding
this bit stream into a stream of symbols, e.g. frequency, phase
and/or amplitude symbols (corresponding to different frequency,
phase and/or amplitude values of the carrier signal).
The size M of the symbol alphabet is finite. Each symbol
represents n=log2M bits For example: M=8 frequency symbols …
each symbol represents 3 bits.
Choice of digital modulation scheme

A desirable modulation scheme for wireless communication

provides:
Low bit error rates at low received signal-to-noise ratios
Performs well in multipath and fading environments
Occupies a minimum of bandwidth
Easy and cost effective to implement.
In practice, there is a trade-off between these desired qualities …
the different modulation schemes provide different trade-offs. The
modulation scheme chosen is that which gives the best trade-off
for the particular application being considered.

Two standard measures of performance:

Power efficiency (ηP): ability of modulation scheme to preserve the
fidelity of the digital message at low power levels. PE is usually
expressed as a ratio of the signal energy per bit to noise power
spectral density (Eb/N0) required at the receiver input for a certain
probability of error (say 10-5).
Bandwidth efficiency (ηB): ability of modulation scheme to
accommodate data within a limited bandwidth. In general,
increasing the data rate implies decreasing the duration width of a
digital symbol, which increases the bandwidth of the signal. BE
reflects how efficiently the allocated bandwidth is utilized and is
defined as the ratio of the throughput data rate per Hertz.
ηB=R/B where R=data bit rate (bits per sec) and
B=bandwidth occupied by the modulated signal (Hertz).

Shannon’s Theorem: for an arbitrarily small probability of error,

the maximum possible data bit rate C=RMAX (known as the channel
capacity) is
C = Blog2(1+S/N)
Where S/N=signal-to-noise ratio.
E.g. S/N(dB)=10dB S/N=100, B=200kHz
C=200 log2(1+10)kbps=691.886kbps.
Note: C=maximum possible bit rate. In practice, in order to get R
close to C we must use channel coding (error control coding).

In the design of a digital communication system, very often there is

a trade-off between BE and PE. For example, adding channel
coding to a message increases the bandwidth occupancy (and thus
decreases the BE), but at the same time reduces the required
received power for a particular bit error rate (and thus increases the
PE).
On the other hand, digital modulation schemes which use a larger
symbol alphabet size M have a smaller bandwidth occupancy
(better BE), but require a larger received power for a particular bit
error rate (worse PE).

Bandwidth of digital signals

There are several definitions of bandwidth. All based on the power

spectral density (PSD)… signal power at each frequency.

Absolute bandwidth: range of frequencies over which the PSD is

non-zero. Not particularly useful since all signals in practice are
time restricted, which implies all signals have an infinite absolute
bandwidth.

Null-null bandwidth: width of the main PSD lobe.

Half power (or 3dB) bandwidth: interval between frequencies at

which the PSD has dropped to half that of peak value (or –3dB
below the peak PSD(dB) value).

X-dB bandwidth: interval between frequencies at which the PSD

has dropped to –XdB below peak value. [Note: Half power
bandwidth corresponds to X=3 dB bandwidth.]
Pulse Shaping techniques
Suppose we use a rectangular pulse (baseband or bandpass) to
represent each data symbol. The rectangular pulse has an infinite
bandwidth. So, when a stream of such pulses (corresponding to a
stream of data) is passed through a bandlimited channel, each of
the pulses is smeared in time. This causes the pulse for each
symbol to spread into the time interval of succeeding symbols.
This is known an inter-symbol interference (ISI) and, in the
presence of noise, leads an increased probability of the receiver
making an error in detecting(recovering) a symbol.

One approach to minimizing ISI is to increase the channel

bandwidth, but this is not desirable … since we aim to minimize
channel bandwidth.
Another approach is to use pulse shaping techniques … which
shape the rectangular pulse into a pulse shape which is less
susceptible to channel smearing, that is to ISI.
The aim of such techniques is to minimize ISI while also keeping
the signal bandwidth low.

Nyquist criterion for ISI cancellation

As proved by Nyquist, if heff(t) is the impulse response of the

overall communication system (transmitter, channel, receiver), the
condition of communication without ISI is
heff(nTS) = A for n=0 (*1)
= 0 for n=nonzero integer
where TS=symbol interval, n=integer, A=non-zero constant.
Physically, this means that the (smeared) pulses of all
neighbouring symbols are zero at the particular time instant kTS of
the current symbol.
The effective impulse response is
heff(t)= δ(t)*p(t)*hC(t)*hR(t)
where δ(t) = the impulse function representing our symbol, p(t)=
pulse shape of the symbol, hC(t)=channel impulse response,
hR(t)=receiver impulse response, *=convolution.

The condition of equation (*1) is met by

heff(t)=sin(πt/TS)/(πt/TS) (*2)
So, if overall communication system can be modeled as a filter
with the impulse response of (*2) then no ISI occurs … if we
sample our received waveform exactly at the symbol time instants
nTS.

The transfer function corresponding to (*2) is

Heff(f) = (1/fS)rect(f/fS)
= (1/fS) -fS/2<=f<=fS/2
=0 elsewhere
where fS=1/TS. This is the rectangular `brick wall’ filter, with
absolute bandwidth fS/2.
There are very serious practical problems in the implementation of
this zero-ISI filter.
(i) The impulse response heff(t) of (*2) is nonzero for t<0.
This means it is noncausal … current output depends on
future inputs….not realistic.
(ii) The function of (*2) is zero only at exactly t=nTs and that
the value of the function increases/decreases quickly away
from zero as we move away from t=nTs (the function has a
relatively large waveform slope (1/t) at each zero
crossing). Thus any deviation(error) in the sampling time
from t=nTS leads to significant ISI.
Raised cosine filter
The raised cosine filter, like the rectangular ‘brick wall’ filter,
satisfies the condition of (*1) … and as such is a zero-ISI filter.
See equations (6.48), (6.49) and figures 6.16-6.18 of Rappaport for
a detailed definition of this filter.

The raised cosine filter is far more popular than the brick wall filter
because the impulse response function has a much smaller
waveform slope at the zero crossings … and therefore produces
much less ISI (than the brick wall filter) when sampling does not
occur exactly at t=nTS.

The raised cosine filter includes a roll-off parameter r (Rappaport

uses α) where 0<=r<=1.

The transfer function HRC(f) of the raised cosine filter is zero for
|f|>(1+r)/(2TS).
Thus, the absolute bandwidth B=RS(1+r)/2 where RS=1/TS =
symbol rate.
The raised cosine filter could also be implemented in a bandpass
region of the communication system. In this case, HRC(f)=0 for |f-
fc|>(1+r)/(2TS). The absolute bandwidth is then B=(1+r)/TS
=(1+r)RS.

If roll-off r=0, then we have the rectangular brick wall filter.

As r increases:
Require larger absolute bandwidth for given symbol rate
(disadvantage)
Slope of impulse response function at zero crossing decreases
(advantage).
Implementation
The raised cosine filter is noncausal. In practice, this is avoided by
using a time-windowed time delayed version of hRC(t) , e.g. hRC (t-
6Ts) and hRC (t)=0 for |t|>6Ts .

Recall
Heff(f)=P(f).HC(f).HR(f).
Typically the receiver includes an equalizer, so that
HR(f)=HEQ(f).HRR(f).
Assuming that HEQ(f)=1/HC(f) … that is the distortions introduced
by the channel are completely nullified by using an equalizer …
then the overall system transfer function is
Heff(f)=P(f).HRR(f).
Thus, for raised cosine based zero-ISI filtering we require
P(f).HRR(f) = HRC(f).

One popular approach is to use P(f)=HRR(f)= sqrt{HRC(f)}. This

corresponds to the matched filter implementation. Such an
implementation is optimal in that it leads to a maximum signal to
noise ratio within the recovered signal (after sampling, but prior to
detection … application of a threshold to decide which symbol was
transmitted). This, is turn, leads to minimum probability of noise
induced errors.
Gaussian pulse shaping filter
Non-Nyquist techniques are also used in practice for pulse
shaping. The most popular of these is the Gaussian pulse shaping
filter.

This filter, although it does not satisfy the zero ISI condition
specified earlier, typically introduces only small amounts of ISI.
The main advantage is the absence of sidelobes in the impulse
response….minimal increased ISI when errors occur in sampling
time.

See equations (6.52), (6.54) and figure 6.20 of Rappaport.

The Gaussian filter has an infinite absolute bandwidth. It has a

(baseband) 3dB bandwidth of B=0.5887/α where the parameter
α>0. As α decreases, the amount of ISI decreases, while the 3dB
bandwidth increases.