Signal Processing in

Wireless Systems 14

14.1 Introduction
ignal processing techniques such as diversity, equalization, speech, and channel coding

S have been successfully used in communication systems to improve the quality of

communications. With the recent exploding research interest in wireless communications,
the application of signal processing to this area is becoming increasingly important. Indeed, it is the
advances in the signal processing technology that make most of today’s wireless communications
possible and hold the key to future services. The application of signal processing techniques to
wireless communications is an emerging area that has recently achieved dramatic and important
results. It holds the potential for even greater results in the future as an increasing number of
researchers from the signal processing and communications areas participate in this expanding
field.
From an industrial viewpoint, advanced signal processing technology can not only
dramatically increase wireless system capacity, but also can improve communication quality
including the reduction of the effects of all types of interference. In this chapter, we first define
diversity as a commonly used technique in mobile radio systems to combat signal fading to
improve the signal to noise ratio (SNR) of the system. We then discuss the basic principle of
diversity, different diversity schemes; space, polarization, frequency and time diversity, etc.; and
the roles of different diversity signal combining techniques.
We then examine the concept of equalization. Equalization means equalizing all parts of the
received signal (with respect to time delay and frequency) to the same attenuation level. This helps
mitigate the inter-symbol interference (ISI) and fading effect. Then we focus on the different equalization
techniques (linear and non-linear). Equalization techniques which can combat and/or exploit the
frequency selectivity of the wireless channel are of enormous importance in the design of high data rate
wireless systems. Finally, the detailed descriptions of different speech and channel coding schemes with
reference to the Global System for Mobile Communications (GSM) are also included in this chapter.
Equalization, diversity, and channel coding are three techniques that can be used
independently to improve the received signal quality.

14.2 Diversity
Diversity refers to a technique for improving the transmission of a signal, by receiving and
processing multiple versions of the same transmitted signal. The multiple received versions can
be the result of signals following different propagation paths (spatial diversity), being transmitted
at different times (time diversity) or frequencies (frequency diversity).
332 Mobile Cellular Communication

Diversity is a commonly used technique in mobile radio systems to combat signal fading. If
several replicas of the same information carrying signal are received over multiple channels with
comparable strengths which exhibit independent fading, then there is a good likelihood that at
least one or more of these received signals will not be faded at any given instant. This makes it
possible to deliver adequate signal level to the receiver. Without diversity techniques, in noise-
limited conditions, the transmitter would have to deliver a much higher power level to protect
the link during the short intervals when the channel is severely faded. In mobile radio, the power
available on the reverse link is severely limited by the battery capacity of hand-held subscriber
units. Diversity methods play a crucial role in reducing transmitted power needs. Also, cellular
communication networks are mostly interference limited and mitigation of channel fading
through use of diversity can translate into reduced variability of carrier to interference ratio (C/I),
which in turn means lower C/I margin and hence better reuse factors and higher system capacity.

14.2.1 Principle of diversity

The principle of diversity systems is to send copies of the same information signal through several
different channels. Performance enhancement is achieved as these channels fade independently;
thus, fading will affect only a part of the transmission. The diversity techniques operate over
time, frequency, or space, but the basic idea remains the same. By sending signals that carry
the same information through different paths, multiple independently faded replicas of data
symbols are obtained at the receiver end and more reliable detection can be achieved.
Diversity exploits the random nature of radio propagation by finding independent (or at
least highly uncorrelated) signal paths for communication.
There are many techniques for obtaining independently fading branches and these can be
subdivided into two main classes. The first are explicit techniques where explicit redundant
signal transmission is used to exploit diversity channels. An example of explicit diversity is
the use of dual polarized signal transmission and reception in many point-to point radios.
Clearly, such redundant signal transmission involves a penalty in frequency spectrum or
additional power.
In the second class, there are implicit diversity techniques: the signal is transmitted only
once, but the decorrelating effects in the propagation medium such as multipaths are exploited
to receive signals over multiple diversity channels. A good example of implicit diversity is the
RAKE receiver (discussed in Chapter 20) in code division multiple access (CDMA) systems, which
uses independent fading of resolvable multipath to achieve diversity gain. Figure 14.1 illustrates
the principle of diversity where two independently fading signals are shown along with the
diversity output signal (thick line) which selects the stronger signal. The fades in the resulting
signal have been substantially smoothed out while yielding higher average power.
The envelope cross-correlation r between these signals is a measure of their independence.

E ⎣⎡[ r1 − r1 ][ r2 − r2 ]⎦⎤
r = (14.1)
E [ r1 − r1 ] E [ r2 − r2 ]
2 2

where E is the expectance and r1 and r2 represent the instantaneous envelope levels of the
normalized signals at the two receivers, and r1 and r2 are their respective means.
It has been shown that a cross-correlation of 0.7 between signal envelopes is sufficient to
provide a reasonable degree of diversity gain. Depending on the type of diversity employed, these
diversity channels must be sufficiently separated along the appropriate diversity dimension. For
 Signal Processing in Wireless Systems 333

–90 3

–95

–100
Signal level in dB

–105

–110 2 r2(t)

–115 1
r1(t)
–120

–125

–130

0 50 100 150 200 250 300 350 400 450

Time in milliseconds

Figure 14.1 Example of diversity combining. The two independently fading signals are 1 and 2.
Signal 3 is the result of selecting the strongest signal

spatial diversity, the antennas should be separated by more than the coherence distance to ensure
a cross-correlation of less than 0.7.

14.2.2 Diversity schemes

Diversity techniques are widely used to improve wireless link performance and received signal
quality. In wireless systems, the most effective method of counteracting the effects of channel
fading is to use diversity schemes in the transmission and reception of the signal. Diversity is
based on the fact that individual channels experience independent fading patterns. The short-
term multipath fading can severely reduce transmission accuracy.
There are many ways to obtain diversity. Diversity over time can be obtained via coding and
interleaving: information is coded and the coded symbols are dispersed over time in different
coherence periods so that different parts of the code words experience independent fades.
Analogously, one can also exploit diversity over frequency if the channel is frequency selective.
In a channel with multiple transmitting or receiving antennas spaced sufficiently, diversity can be
obtained over space as well. In a cellular network, macro diversity can be exploited by the fact that
the signal from a mobile can be received at two base stations. Since diversity is such an important
resource, a wireless system typically uses several types of diversity. There are several techniques for
obtaining diversity branches. The most important of these are discussed in the following section.

Space diversity
Space diversity, also known as antenna diversity, is one of the most popular forms of receiver
diversity schemes widely used in wireless systems. It is easy to implement and does not require
additional frequency spectrum resources. Space diversity is exploited on the reverse link at
the base station receiver by spacing antennas apart so as to obtain sufficient decorrelation.
334 Mobile Cellular Communication

TX and RX TX and RX

0.7l

RX only
RX only

Figure 14.2 Space diversity topology

The key for obtaining minimum uncorrelated fading of antenna outputs is adequate spacing
of the antennas. The required spacing depends on the degree of multipath angle spread. For
example, if the multipath signals arrive from all directions in the azimuth, as is usually the
case at the mobile, antenna spacing (coherence distance) of the order of 0.5l to 0.8l is quite
adequate.
On the other hand, if the multipath angle spread is small, as in the case of base stations,
the coherence distance is much larger. Also, empirical measurements show a strong coupling
between antenna height and spatial correlation. Larger antenna heights imply larger coherence
distances. Typically, 10l–20l separation is adequate to achieve r = 0.7 at base stations in suburban
settings when the signals arrive from the broadside direction. The coherence distance can be
3–4 times larger for end fire arrivals. The end-fire problem is averted in base stations with tri-
sectored antennas as each sector needs to handle only signals arriving ±60° off the broadside.
The coherence distance depends strongly on the terrain. The space diversity topology is shown
in Figure 14.2. There is a statistical relationship between the spatial separation of two receiving
antennas and the likelihood of both receiving faded signals simultaneously.

Polarization diversity
In polarization diversity, horizontally and vertically polarized signals or left and right circular polarized
waves are transmitted simultaneously. Since these are uncorrelated in the mobile radio path, one of
these signals will provide strong received level after fading. In mobile radio environments, signals
transmitted on orthogonal polarizations exhibit low fade correlation and, therefore, offer potential
for diversity combining. Polarization diversity can be obtained in two ways:

• Explicit polarization diversity – In explicit polarization diversity, the signal is

transmitted and received in two orthogonal polarizations.
• Implicit polarization diversity – In the implicit polarization technique, the signal is
launched in a single polarization, but is received with cross-polarized antennas.

Frequency diversity
Another technique to obtain decorrelated diversity branches is to transmit the same signal
over different frequencies. The frequency separation between carriers should be larger than the
coherence bandwidth. The coherence bandwidth, of course, depends on the multipath delay
spread of the channel. The larger the delay spread, the smaller the coherence bandwidth and
more closely we can space the frequency diversity channels. Clearly, frequency diversity is
an explicit diversity technique and needs additional frequency spectrum. A common form of
frequency diversity is multicarrier (also known as multitone) modulation.
 Signal Processing in Wireless Systems 335

This technique involves sending redundant data over a number of closely spaced carriers to
benefit from frequency diversity, which is then exploited by applying interleaving and channel
coding/forward error correction (FEC) across the carriers. Another technique is to use frequency
hopping wherein the interleaved and channel coded data stream is transmitted with widely
separated frequencies from burst to burst. The wide frequency separation is chosen to guarantee
independent fading from burst to burst. Some examples of systems employing frequency diversity
include spread spectrum systems, such as frequency-hopping spread spectrum (FHSS) and multi-
carrier spread spectrum (MCSS) systems.

Time diversity
In time diversity, the desired message is transmitted repeatedly over several time periods having
separation between adjacent time slots larger than the channel coherence time. In mobile
communications channels, the motion of the mobile together with scattering in the vicinity of
the mobile causes time selective fading of the signal with Rayleigh fading statistics for the signal
envelope. Signal fade levels separated by the coherence time show low correlation and can be
used as diversity branches, if the same signal can be transmitted at multiple instants separated
by the coherence time.
The coherence time depends on the Doppler spread of the signal, which in turn is a function
of the mobile speed and the carrier frequency. Time diversity is usually exploited via interleaving,
FEC coding, and automatic request for repeat (ARQ). These are sophisticated techniques to exploit
channel coding and time diversity. One fundamental drawback with time diversity approaches
is the delay needed to collect the repeated or interleaved transmissions. If the coherence time is
large, as for example, when the vehicle is slow moving, the required delay becomes too large to
be acceptable for interactive voice conversation.

14.2.3 Micro diversity

In Micro diversity or antenna diversity the signal from antennas mounted at separate locations are
combined. Typically, these antennas are located on the vehicle or at the same base station tower
and their spacing are few wavelengths. The received signal amplitude is correlated for antennas
separated by a distance d. The received multipath signal becomes practically uncorrelated if
antennas at the mobile are spaced by more than, say, half a wavelength.
In the analysis of this correlation, it is assumed that the mobile antenna is mounted at
low height and close to all kinds of reflecting and scattering objects. The base station antenna,
however, is mostly located well above such obstacles. Hence, at the base station all multipath
waves arrive from approximately the same direction. If the antenna is moved over a certain small
distance d, the phase shift is almost identical for all arriving waves. This is in sharp contrast to the
situation at the mobile where motion over half a wavelength leads to almost uncorrelated signal
phases. To ensure effective antenna diversity at the base station, antennas must be separated
much farther than the fraction of the wavelength required for diversity at the mobile.

14.2.4 Macro diversity

In the field of wireless communication, macro diversity or site diversity is a kind of space diversity
scheme using several receiver antennas and/or transmitter antennas for transferring the same
signal. The distance between the transmitters is much longer than the wavelength as opposed to
micro diversity where the distance is in the order of or shorter than the wavelength. In a cellular
network or a wireless LAN, macro diversity implies that the antennas are typically situated
in different base station sites or access points. Receiver macro diversity is a form of antenna
336 Mobile Cellular Communication

combining and requires an infrastructure that mediates the signals from the local antennas or
receivers to a central receiver or decoder.
Transmitter macro diversity may be a form of simultaneous broadcasting, where the same
signal is sent from several nodes. If the signals are sent over the same physical channel (e.g., the
channel frequency and spreading sequence), the transmitters are said to form a single frequency
network, a term used especially in the broadcasting world. The aim is to combat fading and to
increase the received signal strength and signal quality in exposed positions in between the
base stations or access points. Macro diversity may also facilitate efficient broadcasting and
multicasting services, where the same frequency channel can be used for all transmitters sending
the same information. The diversity scheme may be based on transmitter (downlink) macro
diversity and/or receiver (uplink) macro diversity.

14.2.5 Diversity signal combining techniques

The use of diversity at receiver needs combining the outputs of statistically independent fading
channels, in accordance with a criterion that leads to improved receiver performance. It is
assumed that the wireless communication channel is described by a frequency flat, slow fading
Rayleigh channel. It implies that all the frequency components constituting the transmitted
signal are characterized by the same random attenuation and phase shift. The fading remains
essentially unchanged during the transmission of each symbol and the fading phenomenon is
described by Rayleigh distribution. Several techniques have been studied for diversity combining.
Some of these techniques are selection combining, maximal ratio, equal gain, and square law
combining. They can be used with each of the aforementioned diversity schemes.

Selection combining
Selection combining is the simplest and perhaps the most frequently used form of diversity
combining. With selective diversity, one best signal is chosen based on the received signal
strengths from the set of diversity branches. The receiver monitors the SNR value of each diversity
branch and selects the one with the maximum SNR value for signal detection. It is much easier to
implement without much performance degradation. It is generally employed for the reverse link
transmission where the diversity branches can be physically located in different base stations.
Figure 14.3 depicts a simple block diagram of the selective diversity combining technique.
Consider M number of independent fading signals received by multiple receiver antennas.
There are M-branch receivers comprising of coherent demodulators. The output of demodulators
is presented to a logic circuit, which selects the particular branch receiver output having the largest
SNR value of the received signal. Conceptually, selection diversity combining is the simplest

Receiver 1

Receiver 2 Output
Logic
circuit

Receiver M

Figure 14.3 Selective diversity combining technique

 Signal Processing in Wireless Systems 337

form of “space diversity on receive” technique. The selection combining procedure requires that
the receiver outputs are monitored continuously. At each instant of time, the receiver with the
largest instantaneous SNR value is selected. From a practical implementation point of view, such
a selective combining procedure is difficult.
There is an alternate method by adopting a scanning version of the selective combining
procedure. Firstly, the receiver with the strongest output signal is selected and maintained till its
instantaneous SNR value does not drop below a pre-defined threshold SNR value. Then, a new
receiver that offers the strongest output signal is selected and the selection procedure is repeated.
This method has an advantage that it requires only one receiver and is very simple to implement.
The performance of the selective combining technique is not optimum because it ignores the
information available from all the diversity branches. But there is exception for the particular
selected receiver branch that produces the instantaneous SNR value greater than the pre-defined
threshold SNR value. The probability, PM (Si), that all independent diversity branches receive
signals, which are simultaneously less than specified threshold SNR, Si, is given by

(
PM (Si ) = 1 − e si / sr )
M

(14.2)
where
Sr is the average received SNR value
M is the number of independent diversity branches

It can be seen that selective diversity can greatly improve the performance of the bit error rate
(BER). The performance improvement is more significant when M is increased from 1 to 2 than it
is increased from 2 to 4 and 4 to 8.

Maximal ratio combining

Maximal ratio combining is considered to be the optimum technique of combining in which
the diversity branches are weighted before summing them, each weight being proportional to
the signal strength of the received branch. This technique assumes that the receiver is able to
accurately estimate the amplitude fading and carrier phase distortion for each diversity channel.
With knowledge of the complex channel gains, the receiver coherently demodulates the received
signal from each branch. After removing the phase distortion, the coherently detected signal is
then weighted by the corresponding amplitude gain. The weighted received signals from all the
M branches are then summed and applied to the decision device. Figure 14.4 describes the block
diagram of a maximal ratio combiner.
Consider the transmission of a digital modulated signal over flat slow Rayleigh fading
channels using coherent demodulation with Mth order diversity. It is assumed that the channel
fading processes and the additive white Gaussian noise (AWGN) processes are mutually statistically
independent. For a slow fading channel, the complex gain can be assumed to be a complex
constant over each symbol interval. The demodulator in each channel is optimum for an AWGN
channel using filters matched to the orthonormal functions.
Theoretically, the maximal ratio combiner is the optimum among linear diversity combining
techniques in the context that it produces the largest possible value of instantaneous SNR value.
In fact, the instantaneous output SNR value of the maximal ratio combiner can be large even
when the SNR values of the individual branches are small. Practically, it is difficult to achieve
the exact setting, and significant instrumentation is needed to adjust the complex weighting
parameters to their exact values.
338 Mobile Cellular Communication

Receiver 1

Receiver 2
Output
Linear
combiner

Receiver M

Figure 14.4 Maximal ratio diversity combining technique

Equal gain combining

In this, all the signals are weighted equally after coherent demodulation which removes the phase
distortion. All the weighting parameters have their phase angles set opposite to those of their
respective multipath branches, and their magnitudes are set equal to some constant value. The
coherently detected signals from all the M branches are simply added and applied to the decision
device. As the receiver does not need to estimate the amplitude fading, the receiver design is not
complex. Equal gain diversity combining also relies on the ability to estimate the phase of the
different diversity branches and to combine the signals coherently. Owing to hardware limitations
or physical separation of the diversity receivers, it is difficult to implement it practically. The
performance of an equal gain combiner is only marginally inferior to a maximal ratio combiner
and superior to a selection diversity combiner. Among the four linear combining techniques,
maximal ratio combining offers the best performance, followed by equal gain combining.

Square law non-linear combining

Square law combining offers the opportunity to obtain an advantage in diversity without requiring
phase estimation. Unlike maximum ratio combining, square law combining is applicable to
certain modulation schemes such as orthogonal modulations. This includes direct sequence
CDMA signals or frequency shift keying in which different but approximately orthogonal
sequences or frequencies are used to represent different data symbols.

14.2.6 Transmit diversity

Transmit diversity (TD) is one of the key contributing technologies to define the International
Telecommunications Union (ITU) endorsed 3G systems, WCDMA and CDMA2000. Spatial
diversity is introduced into the signal by transmitting through multiple antennas. The antennas
are spaced far enough so that the signals emanating from them can be assumed to undergo
independent fading. In addition to diversity gain, antenna gain can also be incorporated through
channel state feedback. This leads to the categorization of TD methods into open loop and closed
loop methods.
Most mobile communication channels must combat the effects of fading caused by multipath
propagation. An important way of quantifying fading is in terms of a measure called the coherence
 Signal Processing in Wireless Systems 339

bandwidth Bc, which indicates the amount of bandwidth that will fade in a correlated fashion at
any instant of time. To define this correlation, consider the delay spread of the channel, Td, which
is the maximum duration of the mobile communication channel. The time index is a measure
of the time of arrival relative to the first multipath component at time 0. Often, the “direct
path” arrives first and subsequent paths represent paths reflected at increasing distances from the
receiver. The coherence bandwidth is given by

1
Bc ≈ (14.3)
Td

The coherence bandwidth is approximated at typically path powers less than 5–10 per cent of
the total power. Considering a communication system with bandwidth Bw. If Bc > Bw, the channel
between the transmitter and receiver is called a flat fading channel, and if Bc < Bw the channel
is called a frequency selective channel. Flat fading channels are problematic for systems without
time diversity, because a deep fade can result in a received signal that is below the background
noise level, making communication unreliable. The worst types of channel conditions for many
communication systems are slowly changing flat fading channels. This is due to the length of
time the receiver cannot reliably demodulate the bits sent by the transmitter.
TD can improve the receiver performance in the presence of flat fading. It reduces the
impact of fading by offering multiple independent copies of the digitally modulated waveform
at the receiver, where the chance that all copies are simultaneously in a fade is very small. TD
in radio communication using signals that originate from two or more independent sources has
been modulated with identical information bearing signals. It may vary in their transmission
characteristics at any given instant. It can help overcome the effects of fading, outages, and
circuit failures. When using diversity transmission and reception, the amount of received signal
improvement depends on the independence of the fading characteristics of the signal as well as
circuit outages and failures.

14.3 Equalization
Equalization means equalizing all parts of the received signal with respect to time delay and
frequency to the same attenuation level. This is because of the fact that the various components
of the signals suffer different levels of attenuation or fading when travelling through the medium.
This helps mitigate the ISI, fading effect, and so on. An equalizer is usually implemented at the
baseband or at the IF section in a receiver. In a typical wireless system, the RF communication
occurs in a pass band of bandwidth B with a centre frequency fc. Most of the processing such
as coding/decoding, modulation/demodulation, and synchronization are performed in the
baseband.
Equalization techniques which can combat and/or exploit the frequency selectivity of the
wireless channel are of enormous importance in the design of high data rate wireless systems.
The purpose of an equalizer is to reduce the ISI as much as possible to maximize the probability
of correct decisions.

14.3.1 Fundamentals of equalization techniques

Equalization techniques have found wide use in wireless systems for combating the effect of ISI.
The general idea of equalization is to predict the ISI that will be encountered by a transmission
and accordingly modify the signal to be transmitted so that the signal reaching the receiver will
340 Mobile Cellular Communication

Equalizers

Linear Non-linear

DFE MLSE
MAP
Transversal Lattice symbol
detector
Zero-forcing Gradient
LMS RLS
RLS Transversal Lattice
Fast RLS
Square root RLS LMS
Gradient Transversal channel
RLS
RLS estimator
Fast RLS
Square root RLS
LMS
RLS
Fast RLS
Square root RLS

Figure 14.5 Types of equalizers

Where LMS - Least Mean Square
RLS - Recursive Least Square
DFE - Decision Feedback Equalizer
MLSE - Maximum Likelihood Sequence Estimation
MAP - Mobile Application Part

represent the information the transmitter wants to send. Equalization techniques can be broadly
classified as linear and non-linear (see Fig. 14.5). These categories are determined from output of
an adaptive equalizer used for subsequent control of the equalizer. In general, the analogue signal
d(t) is processed by the decision-making device in the receiver. The decision maker determines
the value of the digital data bit being received and applies a slicing or thresholding operation (a
non-linear operation) in order to determine the value of d(t). If d(t) is not used in the feedback
path to adapt the equalizer, the equalization is linear. On the other hand, if d(t) is fed back
to change the subsequent outputs of the equalizer, the equalization is non-linear. Many filter
structures are used to implement linear and non-linear equalizers.
Equalization is used to overcome ISI due to channel time dispersion, and diversity is used to
overcome flat fading.

14.3.2 Linear equalization

A linear equalizer is of the simplest type and can be implemented as a finite impulse response (FIR)
filters. In this, the present and the past values of the received signal are linearly weighted by the filter
coefficient and summed up to produce the output. The linear equalizer can be implemented either as
the simple transversal filter or as a complicated lattice filter. Linear equalizers increase the noise present
 Signal Processing in Wireless Systems 341

in near vicinity of the spectrum, so they are not very effective on channels having severe distortion.
The ISI can be completely removed, without taking into consideration the resulting noise enhacement.
Symbol spaced equalizers and fractionally spaced equalizers are examples of linear equalizers.
A symbol spaced linear equalizer (SSLE) consists of a tapped delay line that stores samples from
the input signal. Once per symbol period, the equalizer outputs a weighted sum of the values
in the delay line and updates the weights to prepare for the next symbol period. This class of
equalizer is called “symbol spaced” because the sample rates of the input and output are equal.
The algorithms for the weight setting and error calculation blocks are determined by the adaptive
algorithm chosen. The new set of weights depends on the current set of weights, the input
signal, the output signal, and for adaptive algorithms other than constant modulus algorithm, a
reference signal whose characteristics depend on the operation mode of the equalizer. In typical
applications, the equalizer begins in training mode to gather information about the channel, and
later switches to decision-directed mode.
A fractionally spaced equalizer (FSE) is a linear equalizer that is similar to a symbol spaced
linear equalizer. By contrast, however, a fractionally spaced equalizer receives Z input samples
before it produces one output sample and updates the weights, where Z is an integer. In many
applications, Z is 2. The output sample rate is 1/Tb, while the input sample rate is Z/Tb, where Tb
is the bit duration. The weight updating occurs at the output rate, which is slower. Figure 14.6
illustrates a general approach to implement an equalizer using a linear equalizer circuit. The
linear equalizer consists of a tapped delay line, which stores the input samples and then outputs
a weighted sum of the stored values once per symbol interval.
Some algorithms like Wiener algorithm are used to minimize the error, and the tap coefficients
are updated in preparation for the next symbol interval. If the time delay t is equal to symbol
interval TS (Symbol rate = RS), then the linear equalizer is called the symbol spaced equalizer. Such
equalizers are optimum if preceded by receiving filters matched to the operating characteristics
of the channels. However, in such practical cases, the channel characteristics are unknown and
consequently the equalizers are sub-optimum.

x(t) x(t+t) x(t+2t) x(t+3t) x(t+4t)

Unequalized Delay Delay Delay Delay
input (t) (t) (t) (t)

Equalized
output
Σ

Tap gain
adjustment

Figure 14.6 Block diagram on linear equalizer

342 Mobile Cellular Communication

14.3.3 Non-linear equalization

Linear equalizers have the drawback of enhancing channel noise while trying to eliminate ISI, a
characteristic known as noise enhancement. As a result, satisfactory performance is unattainable
with linear equalizers for channels having severe amplitude distortion. Linear equalization
techniques are not preferred for wireless communication systems; whereas non-linear techniques,
such as DFE, data directed estimation (DDE), and MLSE, are commonly used for wireless systems.
Of the non-linear techniques, the choice for use in wireless systems is usually DFE since MLSE
requires an increased computational complexity and knowledge of the channel characteristics.
When channel distortion is too severe, then non-linear equalizers are used. The basic
limitation of a linear equalizer such as transversal filter is the poor performance on the channel
having spectral nulls. These equalizers do not perform well on channels that have deep spectral
nulls in the pass band. The most commonly used non-linear equalizers are DFE andMLSE equalizer.

Decision feedback equalization

A decision feedback equalizer (DFE) is a non-linear equalizer that contains a forward filter and a
feedback filter. The forward filter is similar to the symbol spaced equalizer while the feedback
filter contains a tapped delay line whose inputs are the decisions made on the equalized signal.
The purpose of a DFE is to cancel ISI while minimizing noise enhancement. By contrast, noise
enhancement is a typical problem with the aforementioned linear equalizers.
Figure 14.7 depicts a typical structure of DFE comprising of two filters, referred to as the
forward and the feedback equalizers. The received signal is the input to the forward equalizer. The
input to the feedback equalizer is the stream of the detected symbols. The tap gains of this section
are the estimates of the channel sampled impulse response, including the forward equalizer. Due
to past samples, this section cancels the ISI. Decision directed mode means that the equalizer uses
a detected version of its output signal when adapting the weights. Adaptive equalizers typically
start with a training sequence and switch to decision-directed mode after exhausting all symbols
in the training sequence.
A delay can be taken into account by padding or truncating data appropriately. The DFE is
particularly useful for channels with severe amplitude distortions and has been widely used in
wireless communications. There is an improved performance since the addition of the feedback
filter allows more freedom in the selection of feed forward coefficients. The exact inverse of the

Feedback
filter
Sampled
received signal
+ −
Feed forward Decision
∑ device
filter Deleted
symbol output

∑
Error signal to
− + feedback filter

Figure 14.7 Block diagram of DFE

 Signal Processing in Wireless Systems 343

channel response need not be synthesized in the feed forward filter, therefore excessive noise
enhancement is avoided and sensitivity to sampler phase is decreased.
The advantage of a DFE implementation is the feedback filter, which is additionally working
to remove ISI, which operates on noiseless quantized levels and thus its output is free from
channel noise. A drawback of the DFE structure surfaces when an incorrect decision is applied
to the feedback filter. The DFE output reflects this error during the next few symbols as the
incorrect decision propagates through the feedback filter. Under this condition, there is a greater
likelihood of more incorrect decisions following the first one, producing a condition known as
error propagation. On most channels of interest, the error rate is so low enough that the overall
performance degradation is less significant.

Maximum likelihood sequence estimation equalizer

Theoretically, an MLSE equalizer yields the best possible performance but is computationally
intensive. In this equalizer, the training sequence is used to estimate the channel multipath
characteristics and then this estimate of the channel is used to estimate the effects of ISI. Given
the estimates of the discrete channel impulse response, an MLSE receiver uses a trellis diagram
with the Viterbi algorithm to obtain maximum likelihood estimates of the transmitted symbols.
It receives a baseband linearly modulated input signal and outputs the maximum likelihood
sequence estimate of the signal, using an estimate of the channel modelled as a FIR filter.
The Viterbi algorithm, introduced by Viterbi in 1967, reduces the complexity of maximum
likelihood decoding by systematically removing paths from consideration that cannot achieve
the highest path metric.
The equalizer decodes the received signal first by applying the FIR filter, corresponding to
the channel estimate, to the symbols in the input signal. Then it uses the Viterbi algorithm to
compute the trace back paths and the state metric, which are the numbers assigned to the symbols
at each step of the Viterbi algorithm. The metrics are based on Euclidean distance. The output
is the maximum likelihood sequence estimate of the signal, which is a sequence of complex
numbers corresponding to the constellation points of the modulated signal. A functional block
schematic of the adaptive MLSE receiver is shown in Figure 14.8. It consists of two main parts,
the adaptive channel estimator and the MLSE algorithm.

Channel estimator
Sampled
received
signal
Delay Delay Delay

Correlator Correlator Correlator

Detected
MLSE using viterbi algorithm symbols

Figure 14.8 Functional schematic of adaptive MLSE receiver

344 Mobile Cellular Communication

The sampled impulse response of the channel, taken at symbol intervals, is measured with
the adaptive channel estimator. It is then compared with the sequence of the sampled received
signal with all possible received sequences and determines the most likely transmitted sequence
of symbols. The MLSE is the optimal method of cancelling the ISI. However, the complexity of
MLSE receiver grows exponentially with the length of the channel impulse response.

14.3.4 Adaptive equalization algorithms

The mobile fading channel is random and varies with time, so the equalizers must track the time-
varying characteristics of the mobile channel and thus are called adaptive equalizers. Since an
adaptive equalizer compensates for an unknown and time-varying channel, it requires a specific
algorithm to update the equalizer coefficients and track the channel variations.
Three classic equalizer algorithms are primitive for most of today’s wireless standards.
These include the zero-forcing (ZF) algorithm, the least mean square (LMS) algorithm, and the
normalized LMS (NLMS) algorithm.

Zero-forcing algorithm
The zero-forcing equalizer applies the inverse of the channel to the received signal, to restore the
signal before the channel. It has many useful applications. The name zero forcing corresponds
to bringing down the ISI to zero in a noise-free case. This will be useful when ISI is significant
compared to noise.
For a channel with frequency response F(f ), the zero-forcing equalizer C(f ) is constructed by
C(f ) = 1/F(f ). Thus, the combination of channel and equalizer gives a flat frequency response and
linear phase F(f ) C(f ) = 1.
In reality, zero-forcing equalization does not work in most applications, for the following
reasons:
• Even though the channel impulse response has finite length, the impulse response of the
equalizer needs to be infinitely long.
• The channel may have zeroes in its frequency response that cannot be inverted.
• At some frequencies, the noise may be very small. To compensate, the equalizer amplifies
the noise making it a large value. As a consequence, any noise added after the channel gets
boosted by a large factor and destroys the overall SNR.

The third reason is often the most important.

If the channel response (or channel transfer function) for a particular channel is H(s), then
the input signal is multiplied by the reciprocal of this. This is intended to remove the effect of
channel from the received signal, in particular the ISI. The zero-forcing equalizer removes all ISI,
and is ideal when the channel is noiseless. However, when the channel is noisy, the zero-forcing
equalizer will amplify the noise greatly at frequencies f where the channel response H ( j2pf ) has a
small magnitude (i.e., near zeroes of the channel) in an attempt to invert the channel completely.
A more balanced linear equalizer in this case is the minimum mean square error (MSE) equalizer,
which does not usually eliminate ISI completely but instead minimizes the total power of the
noise and ISI components in the output.

Least mean square algorithm

The LMS is an important member of the family of stochastic gradient algorithms. A significant
feature of the LMS algorithm is its simplicity. The LMS algorithm adjusts the weights and biases
of the linear network so as to minimize the MSE. LMS algorithm is the simplest algorithm based
 Signal Processing in Wireless Systems 345

on minimization of the MSE between the desired equalizer output and the actual equalizer
output. The LMS algorithm is a linear adaptive filtering algorithm that consists of two basic
processes.

• Filter processes, which involve (1) computing the output of a transversal filter produced
by a set of tap inputs; and (2) generating an estimation error by comparing this output to a
desired response.
• An adaptive process, which involves the automatic adjustment of the tap weights of the
filter in accordance with the estimated error.

Thus, the combination of these two processes working together constitutes a feedback
loop around the LMS algorithm, as illustrated in the block diagram of Figure 14.9(a), and LMS
algorithm filter is shown in Figure 14.9(b).
First, we have a transversal filter, around the LMS algorithm. This component is responsible
for performing the filtering process. Second, we have a mechanism for performing the adaptive
control process on the tap weight of the transversal filter, hence the designation adaptive weight
control mechanism.

∧
u (n) d (n/un)
∧
Transversal filter w (n)

−
Adaptive weight-control e (n) ∑
mechanism
+
d (n)

Figure 14.9(a) Block diagram of adaptive transversal filter

x (n) Z −1 Z −1 Z −1

WN−1
W0 W1

y (n)
−
e (n) d (n)
+

Figure 14.9(b) LMS algorithm filter

346 Mobile Cellular Communication

In practice, the minimization of the MSE is carried out recursively and may be performed by
use of the stochastic gradient algorithm. It is the simplest equalization algorithm and requires
only (2N + 1) operations per iteration. The filter weights are updated by the update equation.
Let variable n denote the sequence of iteration. LMS is computed iteratively by considering
wk ( n ) to denote the tap-weight vector of LMS filter, computed at iteration (time step) n.
The adaptive linear operation of the filter is completely described by the recursive equation
(assuming complex data).
Filter output:
H

y( n ) = w
ˆ (n) x (n − i)
Estimation error:

e( n ) = d( n ) − y( n )

Tap-weight adaptation:

wk ( n + 1) ≡ wk ( n ) + m ek ( n ) x ( n − k ) (14.4)

where k is the kth delay stage in the equalizer

μ is the step-size parameter
wk ( n ) is the tap-weight vector of LMS filter
n is the iteration time step
ek ( n ) is the the tap-input vector
x(n − k) is the input vector

The LMS equalizer maximizes the signal-to-distortion ratio at its output within the constraints
of the equalizer filter length. If an input signal has a time dispersion characteristic that is greater
than the propagation delay through the equalizer, then the equalizer will be unable to reduce
distortion. The convergence rate of the LMS algorithm is slow due to the fact that there is only
one parameter, the step size that controls the adaptation rate.
In reality, however, exact measurement of the gradient vector is not possible since this
would require prior knowledge of both the correlation matrix R of the tap inputs and the cross-
correlation vector r between the tap inputs and the desired response. Consequently, the gradient
vector must be estimated from the available data. To prevent the adaptation from becoming
unstable, the value of m is chosen from
N
0<m <2 ∑l
i ≡1
i (14.5)

where li is the ith Eigen value of the covariance matrix R.

Normalized LMS algorithm

In the LMS algorithm, the correction that is applied to wk(n) is proportional to the input
sample x(n − k). Therefore, when x (n – k) is large, the LMS algorithm experiences gradient noise
2
amplification. This problem is eliminated with the normalization of the LMS step size by x( n )
2
in the NLMS algorithm. Only when x(n − k) becomes close to zero, the denominator term x( n )
in the NLMS equation becomes very small and the correction factor may diverge. So, a small
 Signal Processing in Wireless Systems 347

positive number ε is added to the denominator term as the correction factor. Here, the step size
is time varying and is expressed as

b
m (n) = (14.6)
x ( n )) + e
2

where
b is the positive step size
m(n) is the step-size sequence
Therefore, the NLMS algorithm update equation takes the form of

b
wk ( n + 1) = wk ( n ) + ek ( n ) x ( n − k ) (14.7)
x ( n )) + e
2

The NLMS algorithm is utilized, as its convergence speed is, generally, superior to that of
LMS. The NLMS algorithm differs from the LMS algorithm in the incoming reference variable and
is normalized by the magnitude of the incoming reference signal.

14.4 Review of speech coding

The efficient utilization of the allocated spectrum is the prime objective in the design of digital
wireless communication systems. The systems rely on the use of speech coding to remove almost
all the natural redundancy inherent in an analogue speech signal, while ensuring a high quality
reproduction of the original speech signal at the receiver. The speech coding, also called source
coding, makes the information signal compatible with digital processing. Encoding analogue
voice into digital format enhances the quality of voice, improves the overall performance of the
system in terms of spectral efficiency, and increases the system capacity.
Digital modulation techniques combined with robust speech coding schemes, which can
inherently withstand higher BERs needing less channel coding, will be the ultimate choice for
a spectrum-efficient mobile communication system. The carrier modulation and speech coding
techniques are logically independent processes but are strongly interrelated. The improvement in
either of them is towards achievement of a common goal of higher spectral efficiency.
Higher level modulation offers higher spectral efficiency but higher SNR is required to achieve
a given BER that is difficult to achieve in a mobile environment. The required S/N can be reduced
by using low bit rate speech coding and efficient error correction techniques prior to modulation.
The important parameters of a speech coder are the transmitted bit rate, the speech quality, the
robustness in the presence of fading and interference, and the complexity of implementation.
Various speech coding techniques are mainly based on linear predictive coding (LPC) strategy.

14.4.1 Selection of speech coders for mobile communication

Speech coder can be defined as a hardware circuit that represents analogue waveforms with a sequence
of binary digits. The fundamental idea behind coding schemes is to take advantage of the special
features of human speech system, the statistical redundancy, and the shortcomings in human capability
to receive sounds. The speech signal varies quite infrequently resulting in a high degree of correlation
between consecutive samples. This short-term correlation is due to the nature of the vocal tract.
348 Mobile Cellular Communication

There also exists a long-term correlation due to the periodic nature of the speech. This
statistical redundancy can be exploited by introducing prediction schemes, which quantize the
prediction error instead of the speech signal itself. The shortcomings in human capability to
receive sounds, on the other hand, lead to the fact that a lot of information in the speech signal is
perceptually irrelevant. The perceptual irrelevancy means that the human ear cannot differentiate
between changes of magnitude below a certain level and cannot distinguish frequencies below
16 Hz or above 20,000 Hz. This can be exploited by designing optimum quantization schemes,
where only a finite number of levels are necessary.
Speech coding is the process for reducing the bit rate of digital speech representation for
transmission or storage, while maintaining a speech quality that is acceptable for the application.
There are three different speech coding methods, which use different features in different ways:

• Waveform coding
• Source coding
• Hybrid coding

The following sections will explain the speech coding methods in detail. Figure 14.10 shows
the plot between the bit rate (Kbps) on a logarithmic axis and the speech quality classes of “poor
to excellent” corresponding to the five point MOS scale values of 1 to 5, defined by the ITU. It
may be noted that for low complexity and low delay, a bit rate of 32–64 Kbps is required. This
suggests the use of waveform codecs. However, for low bit rate of 4–8 Kbps, hybrid codecs should
be used. These types of codecs tend to be complex with high delay.

14.4.2 Speech codec attributes

The speech codec is able to achieve a much higher compression ratio, which results in a smaller
amount of digital data transmission. Speech quality as produced by a codec is a function of
transmission bit rate, complexity, delay, and quality. Therefore, it is mandatory to consider all

Speech
quality

Excellent Waveform
codecs
Hybrid
Good codecs

Fair

Poor

Bad Vocoders

1 2 4 9 16 32 64
Bit rate (Kbps)

Figure 14.10 Quality of service versus bit rate

 Signal Processing in Wireless Systems 349

of these attributes when considering speech codecs. To have more delay, the codecs with lower
bit rate can be considered as compared to the higher bit rate codecs. They are generally more
complex to implement and often have lower speech quality than the higher bit rate codecs. The
following are the speech codec attributes.

Transmission bit rate

Since the speech codec shares the communication channel with other data, the peak bit rate
should be as low as possible so as not to use a disproportionate share of the channel. The codecs
below 64 Kbps are primarily developed to increase the capacity of circuit multiplication equipment
used for narrow bandwidth links. For the most part, they are fixed bit rate codecs, meaning they
operate at the same rate regardless of the input. In the variable bit rate codecs, network loading
and voice activity determine the instantaneous rate assigned to a particular voice channel. A two-
state variable bit rate system is obtained by combining any of the fixed rate speech codecs with
a voice activity detector (VAD). The lower rate could be either zero or some low rate needed to
characterize slowly changing background noise characteristics. Either way, the bandwidth of the
communications channel is only used for active speech.

Delay
The delay of a speech codec can have a great impact on its suitability for a particular application.
The components of total system delay include the following:
• Algorithmic delay
• Processing delay
• Transmission delay

Algorithmic delay is mainly due to the delay introduced by frame size, look ahead, and
multiplexing. Most low bit rate speech codecs process a frame of speech data at a time. The
speech parameters are updated and transmitted for every frame. In addition, to analyse the data
properly it is sometimes necessary to analyse data beyond the frame boundary. Hence, before
the speech can be analysed, it is necessary to buffer a frame’s worth of data. The resulting delay
is referred to as algorithmic delay. This delay component cannot be reduced by changing the
implementation, but all other delay components can.
The processing delay is the second major contribution for delay, which comes from the time
taken by the encoder to analyse the speech and the decoder to reconstruct the speech. It depends
on the speed of the hardware used to implement the coder. The sum of the algorithmic and
processing delays is called the one-way codec delay.
The transmission delay, third component of delay, is the time taken for an entire frame of
data to be transmitted from the encoder to the decoder. The total of the three delays (algorithmic,
processing, and transmission delay) is the one-way system delay. In addition, frame interleaving
delay adds an additional frame delay to the total transmission delay. Frame interleaving is
necessary to combat channel fading and is part of the channel coding process.

Complexity
Special purpose hardware is used to implement the speech codecs, such as digital signal processing
(DSP) chips. The following are the main attributes of DSP:
• The computing speed, in millions of instructions per second (MIPS)
• Random-access memory (RAM)
• Read-only memory (ROM)
350 Mobile Cellular Communication

To design a speech codec, the system designer should make estimation about how much
of these resources are to be allocated. Here we have two types of speech codecs based on the
complexity:
• Low complexity codecs, which use less than 15 MIPS
• High complexity codecs, which require 30 MIPS or more

More complexity results in higher costs and greater power usage. For portable applications,
greater power usage means reduced time between battery recharges or use of larger batteries,
which means more expense and weight.

Quality
Quality is the main attribute among all the speech codec attributes. In many applications,
along with the general-purpose noise, that is, thermal noise, shot noise, etc., there is a large
amount of background noise like car noise, street noise, and office noise. How well does the
codec perform under these adverse conditions? What happens when there are channel errors
during transmission? Are the errors detected or undetected? If undetected, the codec must
perform even more robustly than when it is informed that entire frames are in error. How good
does the codec sound when speech is encoded and decoded twice? All these questions must
be carefully evaluated during the testing phase of a speech codec. The speech quality is often
based on the 5-point MOS scale as defined by the International Telecommunication Union-
Technical (ITU-T).

14.4.3 Linear-prediction based analysis-by-synthesis

In linear-prediction based analysis-by-synthesis (LPAS), coding predicting the present (or future)
value of a discrete-time signal is done from a given set of past samples of the signal. The smaller
the prediction error in a statistical sense, the more reliable the model will be. The prediction
error is defined as the difference between the actual future value of the signal and the predicted
value produced by the model. To be specific, consider a discrete-time signal represented by the
set of samples x (t ) , x (t − TS ) ,..., x (t − NTS ) , where Ts denotes the sampling period. The sampling
rate fs is related to the sampling period as fs = 1/Ts. In the one-step form of linear prediction,
the requirement is to estimate the present value of the signal x(t) given the N past samples
x (t − TS ) , x (t − 2TS ) ,..., x (t − NTS ) .
Let x̂ ( t ) be expressed as a linear combination of the N past samples through the formula
N
xˆ = ∑ an x (t − nTS ) = aT X (14.8)
n =1

where superscript T denotes matrix transposition and the N-by-1 signal vector x and parameter
vector a are, respectively,

X = ⎡⎣ x (t − TS ) , x (t − 2TS ) ,..., x (t − NTS )⎤⎦

T
(14.9)

and

a = [a1 , a2 ,..., aN ]
T
(14.10)

The parameters a1, a2,…,aN in effect define the N degrees of freedom available to us in designing
the predictive model. Equation (14.8) readily suggests the tapped delay line (TDL) shown in
 Signal Processing in Wireless Systems 351

Input x (t −Ts) x (t −2Ts) x (t −NTLTs) x (t −NTs)

Ts Ts Ts
x(t)

aN −1 aN
a a

∑ Prediction
∑ ∑
x(t)

Figure 14.11 Structure of linear predictor

Figure 14.11 as the structure for the predictive model. The key question is how do we determine
the filter coefficients? To answer this question, we need a statistical criterion for optimizing the
design of the filter. A criterion widely used in practice is the MSE criterion, defined by

MSE = E[( x ( t ) − xˆ ( t ))2 ] (14.11)

where E denotes the statistical expectation operator and the difference x ( t ) − xˆ ( t ) stands for the
prediction error. It turns out that if x ( t ) is the sample value of a stationary random process, then
the optimum value of the parameter vector a is given by

a = R−1r (14.12)

where the N × N matrix R is the correlation matrix of the tap inputs of the predictive model
and the N × 1 vector r has its elements as the autocorrelation of the input signal x(t) for lags
−1
TS ,2TS ,..., NTS . The symbol R in Equation (14.12) stands for the inverse of the correlation
matrix. If, however, the physical process responsible for the generation of the signal x(t) is non-
stationary (i.e., its statistics vary with time ), then we require the use of an adaptive procedure
whereby the model parameters are allowed to vary with time.

14.4.4 Waveform coding

In general, waveform codecs are designed to be independent of signal. They map the input
waveform of the encoder into a facsimile-like replica of it at the output of the decoder. Coding
efficiency is quite modest. The coding efficiency can be improved by exploiting some statistical
signal properties, if the codec parameters are optimized for most likely categories of input signals,
while still maintaining good quality for other types of signals as well. The waveform codecs are
further subdivided into time-domain waveform codec and frequency-domain waveform codec.

Time-domain waveform coding

Waveform codes attempt without using any knowledge of how the signal to be coded is generated
to produce a reconstructed signal whose waveform is as close as possible to the original signal.
That is, they should be signal independent. They are low complexity codes, which produce
high quality speech at rates above 16 Kbps. Types of modulations used in waveform coding are
pulse code modulation (PCM), adaptive PCM, differential PCM, adaptive differential PCM, delta
modulation (DM), and adaptive DM.

Frequency-domain waveform coding

In the frequency-domain waveform codecs, the input signal undergoes some short-time spectral
analysis. The signal is split into a number of frequency-domain sub-bands. The individual sub-band
352 Mobile Cellular Communication

signals are then encoded by using different numbers of bits to fulfil the quality requirements of that
band based on its prominence. The various schemes differ in their accuracies of spectral analysis and
in the bit allocation principle (fixed, adaptive, and semi-adaptive). Two well-known representatives
of this class are sub-band coding (SBC) and adaptive transform coding (ATC).

14.4.5 Vocoders
The source coding is also called a voice codec or vocoders. It is a hardware circuit that converts
the human speech into a digital code and vice versa. Speech coding differs from other forms of
audio coding, as speech is a much simpler signal than most other audio signals, a speech signal
is limited to a bandwidth of 300–3,400 kHz (whereas audio signal is limited to a bandwidth of
0–20 kHz, e.g., audible range). There is a lot of statistical information available about the properties
of speech. The pressure of sound waves radiated from the lips produces the human speech,
although, with some sounds, significant energy emanates also from the nostrils, throat, etc.
In human speech, the air compressed by the lungs excites the vocal cord in two typical
modes. When generating voice sounds, the vocal cord vibrates and generates quasi-periodic voice
sounds. In the case of lower energy unvoiced sounds, the vocal cord does not participate in
voice production and the source acts like a noise generator. The excitation signal is then filtered
through the vocal apparatus, which behaves like a spectral shaping filter. This can be described
adequately by an all pole transfer function that is constituted by the spectral shaping action of
the vocal tract, lip radian characteristics, etc. In the case of vocoders, instead of producing a close
replica of an input signal at the output, an appropriate set of source parameters is generated to
characterize the input signal sufficiently close for a given period of time.
The following steps are used in this process:

1. First, the speech signal is partitioned into segments of 5–20 ms.

2. To minimize the prediction residual energy, the speech segments are subjected to spectral
analysis to produce the coefficients of the all zero analysis filters. This process is based on
the computation of the speech autocorrelation coefficients and then using either matrix
inversion or iterative scheme.
3. Then the corresponding source parameters, that is, the excitation parameters and filter
coefficients, are specified. These parameters are quantized and transmitted to the decoder to
synthesize a replica of the original signal by exciting the all pole synthesis filter.

The quality of this type of scheme is predetermined by the accuracy of the source model
rather than the accuracy of the quantization of the parameters. The speech quality is limited
by the fidelity of the source model used. The main advantage of vocoders is their low bit rate,
with the penalty of relatively low, synthetic speech quality. Vocoders can be classified into the
frequency-domain and time-domain subclasses. When compared with the time-domain vocoders,
the frequency-domain vocoders are more effective.

14.4.6 Hybrid coding

Hybrid coding method constitutes an attractive trade-off between waveform coding and
source coding, both in terms of speech quality and transmission bit rate, although usually
at the price of higher complexity. Combing waveform and source coding methods in order
to improve the speech quality and reduce the bit rate falls into this broad category of speech
coding method. The most important family of hybrid codecs, often reffered to as analysis-by-
synthesis (Abs) codecs.
 Signal Processing in Wireless Systems 353

14.5 Review of channel coding

Channel coding is effective in combating independent random transmission errors over a noisy
channel. Channel coding is a common strategy to make digital transmission more reliable, or,
equivalently, to achieve the same required reliability for a given data rate at a lower power level
at the receiver. This gain in power efficiency is called coding gain. For mobile communication
systems, channel coding is often indispensable.
Channel coding adds redundancy information to the information data at the transmitter
in a deterministic manner, following some logical relation with the original information.
The receiver receives the encoded data with transmission errors. At the receiver, the original
information can be decoded with the controlled redundancy based on the same logical
relationship between the original information and redundant information. The redundancy
causes channel coding to consume additional frequency bandwidth during transmission and
may seem to be a waste of system resources. However, if an error correction code is designed
properly using the frequency bandwidth and transmission power, the coded sequences can be
transmitted at a faster rate.
When the transmission rate for the information bits remains the same as in the uncoded
system, the transmission accuracy for the coded system is higher. This results in the coding gain,
which translates to higher transmission accuracy, higher power, and higher spectral efficiency.
In cellular systems, the traffic consists of compressed data and is very sensitive to transmission
errors. Therefore, channel coding can be defined as the processing of coding discrete digital
information in a form suitable for transmission, with an objective of enhanced reliability.
Channel coding is applied to ensure adequacy of transmission quality in terms of BER
and frame error rate. Thus, channel coding provides excellent BER performance at low SNR
values at the expense of reduction in the bandwidth efficiency of the wireless link in high
SNR conditions. To improve the reliability of a digital transmission system further without
increasing the required bandwidth or transmit power, channel coding can be combined with
trellis-coded modulation. This is achieved by combining convolution code with a higher order
modulation scheme at the transmitter and by combining the channel decoder and demodulator
together at the receiver.
Figure 14.12 shows the classical channel coding setup for a digital transmission system.
The channel encoder adds redundancy to digital data bi from a data source. For simplicity,
we will often speak of data bits bi and channel encoder output bits ci, keeping in mind that
other data symbol alphabets than binary ones are possible and the same discussion applies
to that case.
Speech coding scheme is used to save the bandwidth and improve the bandwidth efficiency,
whereas channel coding is employed to improve the signal quality and to reduce the BER.
Owing to the inevitable presence of noise in the wireless channel, the transmitted data
sequences are corrupted. This increases the bit error probability at the receiver. The design goal
of channel coding is to detect and correct the bit errors in the received data sequence by adding
some extra redundant bits into the transmitted data sequence. This is done by error control
coding. Error control coding is the process of adding redundant information to a message to be
transmitted that can then be used at the receiving end to detect and possibly correct errors in the
transmission. There are two different approaches for error control coding
• Automatic repeat request (ARQ)
• Forward error correction (FEC)
354 Mobile Cellular Communication

bi ci s (t )
Data Channel
source encoder Modulator

Transmit
channel

bˆ i
ˆc i r (t )
Data Channel
sink decoder Demodulator

Interface

Figure 14.12 Block diagram for a digital transmission setup with channel coding

The error correction and detection codes can be broadly classified into three categories:

• Block codes; Hamming code, Bose–Chaudhuri–Hocquenghem (BCH) code, Reed–Soloman

code, etc.
• Convolution codes; Viterbi code
• Turbo codes

14.5.1 Error correcting codes

Error controlling codes are used for correcting errors when messages are transmitted over a
noisy channel or a stored data is retrieved. Error correcting codes are a kind of safety net– the
mathematical insurance against the vagaries of an imperfect digital world. The main idea behind
error correcting codes is to add some redundancy. This redundancy is added in a controlled
manner. The encoded message when transmitted might be corrupted by noise of the channel.
The original signal at the receiver’s end can be recovered from the corrupted signal if the numbers
of errors are within the limit for which the code has been designed.
The objectives of a good error control coding scheme include the following:

1. Error correcting capability in terms of the number of errors that it can correct
2. Fast and efficient encoding of the message
3. Fast and efficient decoding of the received message
4. Maximum transfer of information bits per unit time

14.5.2 Block codes

A block code operates on fixed length input blocks of information bits, which are known as
message blocks. Here the general principle is to segment the information into blocks and add a
 Signal Processing in Wireless Systems 355

parity check number, which is normally the product of information bits contained in the block.
Block codes are FEC codes that help to detect and correct a limited number of errors. In a block
encoder, k is the input information bits to the encoder unit and the encoder then uses different
generator polynomials and adds r numbers of extra bits to k bits.
So the total number of output bits from the encoder will be n = r + k. The block code is
referred to as an (n, k) code, and the rate of the code is defined as Rc = k/n. With a k input bit
sequence, 2k distinct code words can be transmitted. For each code word, there is a specific
mapping between the k message bits and the r check bits. The code is systematic because a part
of the sequence in the code word coincides with the k message bits. As a result, it is possible to
make a clear distinction in the code word between the message bits and the parity bits. The code
is binary as these are constructed from bits and linear as each code word can be created by a linear
modulo-2 addition of two or more code words.
To generate an (n, k) block code, the channel encoder accepts information data in successive
k-bit blocks and adds (n − k) redundant bits to each block that are algebraically related to the k
message bits, where k < n, thereby producing an overall encoded block of n bits. The data rate at
which the block encoder produces bits is given by

Ro = ( n / k )Rs (14.13)
Where Ro is the channel data rate at the output of block encoder
Rs is the source information data rate

Example problem 14.1

In GSM cellular communication system, a data block of 180 bits is encoded into 220 bits of code
word on the control channel before sending it to a convolution encoder. Determine the number
of parity check bits added and the code rate of the block encoder used.

Solution

Number of information bits, k = 180 bits (given)

Number of encoded bits, n = 220 bits (given)
Step 1: To determine the number of parity check bits added
Number of parity check bits = n − k = 220 − 180 = 40 bits
Step 2: To determine t he code rate of the block encoder used
The code rate of block encoder, r = k/n = 180/220 = 0.82

Block codes use algebraic technique properties to encode and decode blocks of data bits or
symbols. The code words are generated in a systematic form such that the original k number of
source information data bits are retained as it is and the (n − k) parity check bits or redundant bits
are either appended or prepended to information data bits. A simple operation of a block code is
depicted in Figure 14.13.
Mostly, operation in a block encoder are based on linear feedback shift registers that are easy
to implement and inexpensive. The parity check bits are generated using a generator polynomial
or matrix. Codes generated by a polynomial are called cyclic codes, and the resultant block codes
are called cyclic redundancy check (CRC) codes. The code word comprising of n bits of encoded
356 Mobile Cellular Communication

n-bit code word

Block
encode k-bit data block
k-bit data block

(n−k) parity check bits

Figure 14.13 A simple block code operation

block of data is a transmitter over the channel. The received code word may be error free or
modified due to channel error. Sometimes it may result in another valid code word, which cannot
be detected as error, with a probability given by

PFD ≤ 2–(n−k) (14.14)

where PFD is the probability of false detection.

It can be minimized by designing block codes in such a way that different code words have a
large code distance. Block codes have certain limitations such as the following:

• Block codes have limited ability to handle large numbers of distributed errors in a wireless
channel.
• Block codes use hard decisions that tend to destroy information. They do not achieve the
performance as obtained through the use of soft decisions.
• The block encoder accepts a k-bit information data block and generates an n-bit code word.
Thus, code words are generated on a block-by-block data basis. Clearly, a provision must be
made in the block encoder to buffer an entire information data block before generating the
associated code word.

14.5.3 Convolutional codes

Convolutional coding is a special case of error control coding. There are applications in which
the information bits come in serially rather in large blocks, in which case the use of a buffer as in
case of a block encoder may be undesirable. As for block codes, the convolution codes divide the
bit stream from the source into k-bit blocks. Each k-bit block is then encoded into an n-bit block,
but unlike block codes, the values of n bits depend not only on the values of the k bits in the
corresponding source block but also on the values of the bits in the previous k-bit source blocks.
Although convolution code is more complex, these are more powerful than block codes as they
exploit past history. The idea is to make every code word symbol to be the weighted sum of the
various input message symbols.
For a convolutional code at every moment k, the encoder delivers a block of N binary symbols
2 ck = (ck,1, ck,2,…, ck,N ), a function of the block of K information symbols dk = (dk,1,dk,2,…, dk,K)
present at its input along with m preceding blocks. Convolutional codes consequently introduce
a memory effect of the order m. The quantity ν = m + 1 is called the constraint length of the code and
 Signal Processing in Wireless Systems 357

Blocks of K Register with (m + 1) levels

binary Blocks of N
symbols binary
Input symbols

Convertor Output
Combinatory logics parallel
serial

Figure 14.14 General diagram of a convolution encoder with an output of K/N and memory m

the ratio R = K/N is called the code rate. If K information symbols at the encoder input are found
explicitly in the coded block ck, that is,

ck = (dk,1,…,dk,K,ck,K+1,…,ck,N)

Then the code is known as systematic. In the contrary case, it is known as non-systematic. The
general diagram of an encoder with output K/N and memory m is represented in Figure 14.14.
At every moment k, the encoder has m blocks of K information symbols in memory. These m
K binary symbols define the Sk state of the encoder:

Sk = (dk,dk−1,…,dk−(m+1))

If the input of the encoder is permanently fed by blocks of K information symbols, then the
encoder output consists of N infinite sequences of coded symbols, which, for the output i, have
the form:

(c1, i, c2, i,…,ck, i,…) i = 1,…,N

Let us note that convolutional codes are well adapted to code transmissions with continuous
flow of data. Indeed, the sequences of data to be coded can have any length. To each coded
sequence i = 1, …, N
Generally, convolution coders are more powerful than block codes in terms of providing
FEC, but are not useful for detection or ARQ schemes. At the receiver, FEC is performed using a
maximum likelihood decoding algorithm that determines what data sequence would have been
most likely transmitted, given the received sequence of bits. VLSI implementation of the Viterbi
algorithm is the most common algorithm used for this purpose.

14.5.4 Turbo codes

Turbo codes may use serial (concatenated) and/or parallel recursive convolution codes.
Traditionally, the design of good codes has been tackled by constructing codes with a great
deal of algebraic structure, for which there are feasible decoding schemes. Such an approach is
exemplified by the linear block codes and convolutional codes discussed in preceding sections.
The difficulty with these traditional codes is that, in an effort to approach to the theoretical
limit for Shannon’s channel capacity, we need to increase the code word length of a linear block
code or the constraint length of a convolutional code, which in turn causes the computational
complexity of a maximum likelihood decoder to increase exponentially. Ultimately, we reach a
point where complexity of the decoder is so high that it becomes physically unrealizable.
358 Mobile Cellular Communication

Parity check
Encoder 1 bits z1

Input data
stream Parity check
Interleaver Encoder 2
bits z2

Systematic bits x

Figure 14.15 Block diagram of turbo encoder

Various approaches have been proposed for the construction of powerful codes with large
equivalent block lengths structured in such a way that the decoding can be split into a number
of manageable steps. Building on these previous approaches, the development of turbo codes
and low-density parity-check codes has been by far most successful. Indeed this development
has opened a brand new and exciting way of constructing good codes and decoding them with
feasible complexity.

Turbo coding
In its most basic form, the encoder of a turbo code consists of two constituent systematic encoders
joined together by means of an interleaver, as illustrated in Figure 14.15.
An interleaver is an input output mapping devise that permutes the ordering of a sequence
of symbols from a fixed alphabet in a completely deterministic manner, that is, it takes the
symbols at the input, produces identical symbols at the input, and produces identical symbols at
the output but in a different temporal order. The interleaver can be of many types, of which two
of them are periodic and pseudorandom. Turbo codes use a pseudorandom interleaver, which
operates only on the systematic bits.
There are two reasons for the use of an interleaver in a turbo code:

• To tie together errors that is easily made in one-half of the turbo code to errors that are
exceptionally unlikely to occur in the other half. This is indeed the main reason why the
turbo code performs better than a traditional code.
• To provide robust performance with respect to mismatched decoding, this is a problem that
arises when the channel statistics are not known or have been incorrectly specified.

Turbo decoding
Turbo codes derive their distinctive name from analogy of the decoding algorithm to the turbo
engine principle. Figure 14.16 shows the basic structure of the turbo decoder. It operates on noisy
versions of the systematic bits and two sets of parity check bits in two decoding stages to produce
an estimate of the original message bits.
The first decoder, the interleaver, the second decoder, and the de-interleaver constitute a
single hop feedback system. This arrangement makes it possible to iterate the decoding process in
the receiver many times so as to achieve satisfactory performance. The inputs to the first decoder
are the channel samples corresponding to the systematic information bits, the channel samples
corresponding to the parity bits of the first encoder, and the extrinsic information about the
systematic bits that were determined from the second decoder.
 Signal Processing in Wireless Systems 359

De-interleaver

Noisy
Decoder Random Decoder Hard
systematic bits 2 De-interleaver
1 interleave limite
Noisy parity bits x1

Decoded
bits x
Noisy parity bits x2

Figure 14.16 Block diagram of turbo decoder

Each decoding process yields a soft output decision for the next decoder. The key component
is the soft-input, soft-output (SISO) decoder. To achieve the benefits of turbo code, several
iterations are provided. Thus, the turbo codes are processing intensive and are applied to less
delay sensitive applications such as data. They have been implemented in both software and
hardware as a single integrated circuit, and theydo not suffer from the error at low BERs that have
been attributed to other codes. Turbo codes are capable of providing high coding gain, even with
high code rates.

14.5.5 Comparison between convolution and turbo codes

A turbo code is applicable where digital data is transmitted over a noisy channel because it spreads
error uniformly over the interleaved duration. Owing to their strength in non-linear coding/
decoding and feedback property, turbo codes are better than convolutional codes. Turbo codes
can support the data rates achieved with other error correction codes while offering improved
correction capability.
Figure 14.17 provides a comparison of a turbo code having constraint length K = 4
decoded with 4 iterations with a convolutional code of constraint length K = 9 decoded with a
Viterbi decoder. Figure 14.17 shows the performance of a rate 1/3 turbo code compared to the
corresponding rate 1/3 convolutional code. The results indicate that the turbo code outperforms
the corresponding convolutional code with the same decoding complexity for data rates larger
than 9.6 Kbps; the improvement in performance increases with the data rate for a fixed frame
length of 20 ms.
The performance improvement is due to the fact that the number of bits in the 20 ms frames
increases with the data rate and the performance of a turbo code improves with the number of
bits in the frame. With a large number of bits in a frame, the interleaver separating the codes can
randomize the errors more effectively.

• A turbo code is closer to the random code because it uses a pseudorandom interleaver to
separate its own two convolutional encoders.
• In higher code rates or low SNR value conditions, turbo codes exhibit better performance
than traditional convolutional codes.
• Convolutional codes and turbo codes perform better with soft decisions. However,
convolutional codes can also work with hard decisions.
• Convolutional codes do not have an error floor, whereas turbo codes do have an error floor.
It means that in turbo codes, the BER drops very quickly in the beginning, but eventually
settles down and decreases at a much slower rate.
360 Mobile Cellular Communication

10−1

Turbo code (9.6 Kbps)

10−2
Convolution code
Bit error rate (BER)

(9.6 Kbps)
10−3

10−4 Turbo code

(76.8 Kbps)

10−5
Turbo code
(480.8 kbps)
10−6
0 0.5 1.0 1.5 2.0 2.5 3.0 3.5
EB /No(dB)

Turbo K = 4.4 iterations. 480.8 Kbps

Turbo K = 4.4 iterations. 76.8 Kbps
Turbo K = 4.4 iterations. 9.6 `Kbps
Convolution K = 9

Figure 14.17 Performance of turbo code compared with convolutional code

• Both convolutional and turbo codes require the use of flush bits to initialize them to state
0 at the end of the incoming source information bits sequence. However, due to parallel
encoding structure of turbo codes, it is not straightforward to flush the second encoder.
• Turbo codes are decodable, and hence they are of more practical importance.

Turbo codes are inherently block codes with the block size determined by the size of the
turbo interleaver. These codes are used in 3G cellular technology for high-speed data rate
applications.

14.6 Summary
• In this chapter, different signal processing techniques such as diversity, equalization, speech,
and channel coding are introduced. They have been successfully used in communication
systems to improve the quality of communications.
• Diversity is a commonly used technique in mobile radio systems to combat signal fading to
improve the SNR of the system. The basic principle of diversity, different diversity schemes,
and roles of different diversity signal combining techniques are discussed.
• Equalization is used to overcome ISI due to channel time dispersion.
• Equalization techniques are widely used to improve wireless link performance and received
signal quality.
• Equalization techniques which can combat and/or exploit the frequency selectivity of
the wireless channel are of enormous importance in the design of high data rate wireless
systems.
 Signal Processing in Wireless Systems 361

• The efficient utilization of the allocated spectrum is the prime objective in the design of
digital wireless communication systems. The systems rely on the use of speech coding to
remove almost all the natural redundancy inherent in an analogue speech signal, while
ensuring a high quality reproduction of the original speech signal at the receiver.
• Speech coding is the process for reducing the bit rate of digital speech representation for
transmission or storage, while maintaining a speech quality that is acceptable for the
application. Speech coding methods can be classified as waveform coding, source coding,
and hybrid coding.
• Channel coding is a common strategy to make digital transmission more reliable, or,
equivalently, to achieve the same required reliability for a given data rate at a lower power
level at the receiver.
• There are two different approaches for error control coding: ARQ and FEC. ARQ is detection-only
type coding in which transmission errors can only be detected by the receiver but not corrected.
FEC allows not only detection of errors at the receiving end but correction of errors as well.

Review questions
1. What is the basic principle of diversity? Explain different diversity schemes.
2. What is diversity? How is it provided in a communication system?
3. Among the selection, equal gain, and maximal ratio combining, which scheme is the best
and why?
4. What is meant by equalization? Explain adaptive linear equalization.
5. Explain the LMS algorithm.
6. List the main attributes of a speech coding.

Objective type questions and answers

1. The relation between the coherence bandwidth (Bc ) and delay spread (Td ) is_______
(a) Bc ≈ 1/Td (b) Bc ≈ Td (c) Bc ≈ 2Td (d) Bc ≈ 4Td
2. The condition for a flat fading channel with respect to coherence bandwidth (Bc) and system
bandwidth (Bw) is_______
(a) Bc > Bw (b) Bc = Bw (c) Bc < Bw (d) Bc = 1/Bw
3. The condition for a frequency selective channel with respect to coherence bandwidth (Bc)
and system bandwidth (Bw ) is_______
(a) Bc > Bw (b) Bc = Bw (c) Bc < Bw (d) Bc = 1/Bw
4. In a speech codec, the delay introduced by frame size, look ahead, and multiplexing is called
the _______
(a) processing delay (b) transmission delay
(c) algorithmic delay (d) none
5. The advantage of adaptive multi rate codec (AMR) over enhanced full rate (EFR) is _______
(a) greater spectral efficiency (b) better voice quality
(c) operates under much worse conditions (d) all of the above
6. The waveform codecs use _______ approach in coding the speech signal.
(a) time domain (b) frequency domain
(c) either time or frequency domain (d) none
362 Mobile Cellular Communication

7. The channel coding approach which detects and corrects the transmission errors is _______
(a) automatic repeat request (ARQ) (b) forward error correction (FEC)
(c) Both (a) and (b) (d) none
8. Which of the following is the convolutional code _______
(a) Hamming code (b) Viterbi code
(c) Reed–Solomon code (d) BCH code

Answers: 1. (a), 2. (a), 3. (c), 4. (c), 5. (d), 6. (a), 7. (a), 8. (c), 9. (b).

Open book questions

1. What are the different diversity signal combining techniques?
2. What are the different factors that influence the equalizer algorithm?
3. What are the different speech coding methods?

Key equations
1. The coherence bandwidth is given by
1
Bc ≈
Td

2. The adaptive linear operation of the filter is completely described by the recursive equation
wk ( n + 1) ≡ wk ( n ) + m ek ( n ) x ( n − k )

3. The NLMS algorithm update equation takes the form of

b
wk ( n + 1) = wk ( n ) + ek ( n ) x ( n − k )
x ( n )) + e
2

4. Mean square error (MSE) criterion is defined by

MSE = E[( x ( t ) − xˆ ( t ))2 ]

5. The data rate at which the block encoder produces bits is given by

R0 = ( n/k )Rs

Further reading
Atal, B. S. “Predictive Coding of Speech Signals at Low Bit Rates,” IEEE Transactions on
Communications, Vol. 30, No. 4, 1982, pp. 600–614.
Atal, B. S., and Schroeder, M. R. “Stochastic Coding of Speech at Very Low Bit Rate,” Proceedings of
International Conference in Communications, Amsterdam, 1984, pp. 1610–1613.
 Signal Processing in Wireless Systems 363

Bahl, L. R., Cocke, J., Jelinek, F., and Raviv, J. “Optimal Decoding of Linear Codes for Minimizing
Symbol Error Rate,” IEEE Transactions on Information Theory, Vol. IT-20, No. 2.
Chen, J., Cox, R., Lin, Y., Jayant, N., and Melchner, M. “Coder for the CCITT 16 kbps Speech
Coder Standard,” IEEE Journal of Selected Areas of Communications, Vol. 6, 1988, pp. 353–363.
Furuskar, A., et al. “System Performance of EDGE: A Proposal for Enhanced Data Rates in Existing
Digital Cellular System,” IEEE VTC 98, pp. 1284–1289.
Hess, W. Pitch Determination of Speech Signals. Berlin: Springer Verlag, 1983.
Jarvinen, K., et al. “GSM Enhanced Full Rate Speech Codec,” IEEE GLOBECOM ’97.
Kleijn, W. B., Ramachandran, R. P., and Kroon, P. “Generalized Analysis by Synthesis Coding
and Its Application to Pitch Prediction,” International Conference on Acoustics. Speech Signal
Processing, San Francisco, 1992, pp. 1337–1340.
Kroon, P., and Deprettere, E. F. “A Class of Analysis by Synthesis Prediction Coders for High
Quality Speech Coding at Rates Between 4.8 and 16 kbps,” IEEE Journal of Selected Areas of
Communications, Vol. 6, 1988, pp. 353–363.