performance of m-qam over generalized mobile fading...
TRANSCRIPT
DEPARTMENT OF ELECTRICAL ENGINEERING
COLLEGE OF ENGINEERING KING SAUD UNIVERSITY
PERFORMANCE OF M-QAM OVER GENERALIZED
MOBILE FADING CHANNELS USING MRC DIVERSITY
BY
IBRAHIM AL-SHAHRANI
A DISSERTATION
SUBMITTED TO THE DEANSHIP OF GRADUATE STUDIES IN PARTIAL
FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF
MASTER OF SCIENCE
IN
ELECTRICAL ENGINEERING
SUPERVISED BY
PROF. ADEL AHMAD ALI
Feb 2007
II
We, the undersigned, have read the thesis and examined the candidate and do recommend the
thesis for acceptance.
Prof. Adel Ahmed Ali ----------------------------- (Supervisor)
III
ABSTRACT
Demands for faster data rates on wireless and cellular communication systems, such as
HSDPA and WiMax, have led to much current interest in the use of M-ary Quadrature
Amplitude Modulation (M-QAM) signaling formats due to its high spectral efficiency. In
wireless channel, it is well known that the fading phenomenon, which inherently exists in
most radio links, constitutes one of the boundary conditions of radio communications design.
A widely recognized practice for combating fading in digital communications over such a
time-varying channel is to use space diversity techniques. And the optimum liner diversity
combining technique is the Maximal-Ratio-Combining (MRC).
However, exact performance analyses of square coherent M-QAM in fading environments
have been reported recently in the literature, especially for MRC diversity systems. But all
the previous works were in integral form or in terms of special functions such as
hypergeometric function. In this thesis, new analytical, simple expressions for the exact
average symbol error rates (ASER) for M-QAM transmitted over slow, flat, identically
independently distributed (i.i.d) fading channels using MRC. Three types of fading channels
are considered: Rayleigh, Ricean, and Nakagami-m. These simple and efficient ASER
formulas make it possible for the first time to study, analyze and discuss the parameters of
various constellations of square M-QAM, diversity order and fading parameters precisely and
easily.
The obtained expressions are in the form of sum of exponentials where the number of terms
can be determined according to the required accuracy. Theoretically to get the exact solution,
the series must be infinite. But because the series converge rapidly, 10 terms are enough to
get an error less than 0.005 dB in the worst case which is at low order of modulation index
(M), diversity, and low fading parameter (K for Rician or m for Nakagami-m). The minimum
number of terms to approximate the ASER is two terms, so single term is also investigated
which is available in the literatures. The error for different number of terms are investigated
and tabulated.
IV
CONTENTS
Abstract III
Acknowledgements VI
1. Introduction 1
1.1 Background 1
1.2 Multipath Fading Environment 2
1.2.1 Rayleigh Fading 3
1.2.2 Ricean Fading 4
1.2.3 Nakagami-m Fading 5
1.3 Principles of Diversity Combining 6
1.3.1 Maximal Ratio Combining 7
1.3.2 Equal Gain Combining 9
1.3.3 Selection Combining 10
1.4 Literature Review 11
1.4.1 Ricean Fading channel 11
1.4.2 Nakagami-m Fading channel 13
2. Performance of M-QAM transmitted over AWGN Channel 15
2.1 Introduction 15
2.2 Exact representation of error function Q(x) in terms of a series of exponentials 17
2.3 Exact representation of error function squared Q2(x) in terms of a series of 22
exponentials
2.4 M-QAM Performance over AWGN channel using the exponential form 23
2.5 Concluding Remarks 25
3. Performance of M-QAM transmitted over Ricean Channel with MRC 33
3.1 Introduction 33
3.2 Performance of M-QAM over single Ricean channel 33
3.3 M-QAM Performance over Ricean channel with MRC Diversity 38
3.3.1 Approximation of M-QAM Performance over Ricean channel with MRC 43
Diversity
V
3.3.2 Analysis of Approximation 45
3.4 Concluding Remarks 45
4. Performance of M-QAM transmitted over Nakagami-m Channel with MRC 63
4.1 Introduction 63
4.2 Performance of M-QAM over Nakagami channel with MRC Diversity 63
4.3 Results and Analysis of M-QAM over Nakagami channel with MRC Diversity 65
4.4 Approximation of M-QAM over Nakagami channel with MRC Diversity 69
4.4.1 Analysis of Approximation 71
4.5 Concluding Remarks 72
5. Conclusions and Future Work 84
5.1 Conclusion 84
5.2 Future Work 85
References 86 Appendices 89 Appendix I Appendix II
VI
ACKNOWLEDGEMENTS
All praises be to the Almighty ALLAH, whose uniqueness, oneness, and wholeness is
unchangeable, and Who gave me the confidence, courage, and patience during the course of
this work. All respects are for the prophet MUHAMMAD (P.B.U.H) who enabled us to
recognize our creator.
During the course of my graduate studies, I have benefited tremendously from my
interactions with many extraordinary individuals. Foremost, I thank my advisor Prof. Adel A.
Ali for his guidance and support. His unique blend of vision, technical knowledge and
generosity will be an inspiring role model for my future career.
I thank all faculty members of the Electrical Engineering department for being so nice and
cooperative to me, especially Prof. Abdulaziz Al-Ruwais and Prof. Abdulrahman Al-Jabri.
I would like to thank the staff of STC and Huawei Technologies for their co-operation and
support in completing this research.
Last, but not the least, I would like to thank my wife and daughter and my parents.
1
CHAPTER 1
INTRODUCTION 1.1 Background Wireless communications has been one of the fastest growing segments in the
telecommunications industry. In 2005 with 140 million wireless telephone subscribers in the
Middle East and Africa and more than 2 billion around the world, the future looks even
brighter with potential for growth. The current mobile network is a combination of 2G and
3G cellular systems. The various advantages of 3G over 2G such as higher speed as well as
increased system capacity have been major motivations to move to 3G. In addition to
wireless voice service, markets in wireless broadband services, cordless phones, and direct-
to-home satellite broadcasting have been expanding.
The trends described above are sufficient to sustain a strong demand for wireless digital
communication systems in the future. The major driving force for wireless in the 21st
century, however, will likely lie in the increasingly popular Internet. The 90s have witnessed
the emergence of the Internet as a source for information access, a way to communicate
(email, on-line chat rooms) and an opportunity for business. The impact of the Internet on the
traditional telephone industry is far-reaching. For some time, it has been apparent that data
and telecommunication technologies are converging. Today operator companies not only
provide voice but also data services. Future communication networks, using both wired and
wireless interconnection, will seamlessly support a variety of services, including voice, video
and data. Such huge information flow can only be realized with high speed transmission.
Since present phone lines support data transmission at fixed points, high-speed wireless
mobile networks are being developed as a possible choice for transmission from a central
office to the end user.
People want high transmission speeds and considerable effort has been made to raise the
transmission speed of networks. The increases in speed of mobile systems during the past
twenty five years have come about in the main with a corresponding increase in the number
of signals in a two-dimensional (2-D) modulation format, where M-QAM is the best
candidate today. Knowledge of the error performance of M-QAM signal set in additive white
2
Gaussian noise (AWGN) and in fading channel is very useful to a particular system design,
such as reduced implementation complexity. In the latter case, it is then desirable to know the
error performance sacrificed in exchange for other gains. Evaluation of the exact symbol
error rate (SER) of M-QAM already available in the literatures but not in easy formulas such
as hypergeometric functions.
In this thesis, we represent the Q(x) function or the error function in terms of a summation of
exponential functions to make the analytical solution doable and simple. Then this new form
is used to find the ASER of M-QAM over different fading channels and also when the MRC
diversity technique is used. The objective is to provide new and simple analytical tools and
results that can be used by system designers and students to determine and compare the
performances of M-QAM coherent modulation schemes in fading channel when MRC is
used. This is essential to the development of high speed wireless data systems. Of the same
importance, or maybe more important, is the fact that this method can be used for a wide
range of fading conditions and environments , and are not limited to a specific fading model,
diversity technique, or modulation scheme.
1.2. Multipath Fading Environment
The mobile radio propagation environment places fundamental limitations on the
performances of wireless radio systems. There are roughly three independent phenomena that
together create a hostile transmission environment: path loss variation with distance, short
term (fast) multipath fading and slow log-normal shadowing. The underlying physical
principles behind these three phenomenon are different. Path loss is due to the decay of
electro-magnetic wave intensity in the atmosphere. Multipath fading is caused by multipath
propagation, while slow shadowing is due to the topographical variations along the
transmission path. In this thesis, we will focus on multipath fading.
In wireless radio systems, several signals with different amplitudes, phases and delays
corresponding to different transmission paths arrive at the receiver. The different signal
components add at the receiver constructively or destructively to give the resultant signal.
This multipath fading results in rapid variations typically as much as 30 to 40 dB in the
envelope of the received signal over a distance corresponding to a fraction of a wavelength.
The velocity of the mobile station, v, and the carrier frequency, fc, determine the fading rate
or fading bandwidth, fD, that is, fD = (v/c) fc, where c is the speed of light. Hence, faster
motion leads to more rapid fading. Multipath also causes time dispersion, because the
3
multiple replicas of the transmitted signal propagate over different transmission paths and
reach the receiver antenna with different time delays. This is called frequency selective
fading. On the other hand, the fading is said to be non selective or flat if time delays in
distinct paths are not large enough to result in resolvable replicas of the transmitted signal at
the receiver antenna. From the frequency domain perspective, all frequency components of
the transmitted signal undergo the same attenuation and phase shift through the channel in a
frequency-flat fading channel. This thesis is concerned with frequency-flat multipath fading.
In a frequency-flat fading environment, the received signal is simply the transmitted signal
multiplied by a complex-valued random process which introduces a fading envelope and a
random phase to the transmitted signal. A number of different models have been proposed in
the literature to describe the statistical behavior of the fading envelope of the received signal
[14]-[16]. Well known models are the Rayleigh, Ricean and Nakagami distributions. Next we
discuss and review the theoretical origins and the characteristics of the three models.
1.2.1 Rayleigh Fading
In Rayleigh fading, the composite received signal consists of a large number of plane waves
resulting from scattering at surface elements [15]. Using a central limit theorem, the received
complex low-pass signal g(t) = α(t) exp(jφ(t)) = gI(t) + jgQ(t) can be modeled as a complex
Gaussian random process. In the absence of a line-of-sight (LOS) or specular component,
gI(t) and gQ(t) have zero mean. At any time t, gI and gQ are Gaussian random variables (RVs)
with
(1.1)
where E[x] denotes the expected value of x and Var[x], the variance of x. Therefore, the
envelope α of the received signal has a Rayleigh distribution given by
(1.2)
The Rayleigh fading mode1 agrees well with macrocellular field measurements over the
frequency range from 50 to 11,200 MHz at distances of a few tens of wavelengths or greater
where the mean signal is sensibly constant [17]. It usually applies to scenarios where there is
4
no LOS path between the transmitter and receiver antennas. The phase φ of the received
signal is uniformly distributed from 0 to 2π at any time t, and the amplitude and phase are
statistically independent. Define a new variable γs = α2 /N0 proportional to the squared
envelope α2, where No denotes the one-sided power spectral density of Gaussian noise.
Variable γs, denotes received signal-to-noise ratio per symbol. The probability density
function (PDF) of γs, is given by
(1.3)
where, Λ= σ2 /N0 is the average signal-to-noise ratio per symbol and u(x) is the unit step function. 1.2.2 Ricean Fading If there is a LOS or specular component between the transmitter and receiver, gI(t) and gQ(t)
have non-zero mean m1 and m2 and the envelope α is a Ricean RV with PDF given by
(1.4)
where µ2 = m12 + m2
2 is the non-centrality parameter, and I0(x) is the zero-order modified
Bessel function of the first End. Ricean fading is often observed in microcellular and satellite
applications where a LOS path exists [18], [19]. The Rice K factor is the ratio of the power in
the specular and scattered components, i.e., K = µ2/2σ2. For K = 0, the channel exhibits
Rayleigh fading, and for K = ∞, the channel has no fading. It was reported in [20] that a
typical value of K for practical microcellular channels is about K = 7 dB. Rice factor K = 12
dB was reported for a smaller number of cases. Values of Rice factor in outdoor and indoor
systems usually range from 0 to 25 [20]. The PDF of SNR variable γs, in a Ricean fading
channel can be expressed in terms of the Rice K factor as
(1.5)
where, Λ = (µ2 + 2σ2) /N0 is the average SNR in the Ricean fading.
5
The phase is no longer uniformly distributed but has a preferred value owing to the presence
of a dominant component. The PDF of the phase is given in [21, Eqn. (5.62)].
1.2.3 Nakagami-m Fading
The Nakagami-m distribution was developed by Nakagami in the early 1940's to characterize
rapid fading in long distance High-Frequency (HF) channels [14]. It was shown to sometimes
have greater flexibility and accuracy in matching some experimental data than either the
Rayleigh, Ricean, or log-normal distributions [22]-[24]. The Nakagami-m distribution
describes the received envelope amplitude by
(1.6)
where, Ω = E[α2] is the average power of Nakagami distributed α, Γ(x) is the gamma
function, and the parameter m is defined as the ratio of moments, called the fading parameter,
(1.7)
The Nakagami-m distribution is a generalized distribution that can be used to model different
fading environments by changing the value of m. Rayleigh fading is obtained
for m = 1, and a one-sided Gaussian RV is described by m = 0.5. The non-fading case
corresponds to m = ∞. For values of m in the range 1/2 ≤ m ≤ 1, (1.6) models fading
conditions more severe than Rayleigh fading. For values of m > 1, (1.6) models less severe
fading than Rayleigh. Furthermore, the Ricean distribution and log-normal distribution can
sometimes be closely approximated by Nakagami fading under certain conditions [14], [22],
[23]. A summary of Nakagami- m distribution properties can be found in a review paper by
Nakagami [14].
Using a transformation of random variables, the SNR γs, in Nakagami fading has PDF,
(1.8)
where, Λ = Ω /N0 is the average SNR.
6
1.3. Principles of Diversity Combining
In order to improve the reliability of transmissions on wireless radio channels, some measures
have to be employed to reduce the seventy of multipath fading. Diversity techniques have
been known to be effective in combating the extreme and rapid signal variations associated
with the wireless radio transmission path. Basically, the diversity method requires that a
number of transmission paths be available, al1 carrying the same message but having
independent fading statistics. The mean signal strengths of the paths should also be
approximately the same. Diversity can be achieved by methods that can be placed into seven
categories [25]. In this section, we are concerned with space diversity where the distance
between the receiving antennas is made large enough to ensure independent fading. Usually a
spatial separation of about a half-wavelength will suffice (typically less than 30cm for
frequencies above 500 MHz) [17]. Excellent references on the topic of diversity systems are
[17], [25]-[29].
Diversity combining refers to the method by which the signals from the diversity branches
are combined. There are several ways of categorizing diversity combining methods.
Predetection combining refers to diversity combining that takes place before detection, while
postdetection combining takes place after detection. For diversity strategies incorporating
signal summing, summing after detection can be either equal or inferior to summing before
detection because a nonlinear effect is often experienced in a detection process [26]. With
ideal coherent detection, there is no performance difference between predetection and
postdetection combining [25]. In any case, diversity combining methods include maximal
ratio combining (MRC), equal gain combining (EGC) and selection combining (SC).
The received signal from the l th diversity branch is represented by zl = glsi + nl = yl + nl (1.9)
where, gl = αl exp(jφl) is a complex channel gain, si is the transmitted signal, yl = glsi is the
faded signal, and nl is additive Gaussian noise.
7
1.3.1 Maximal Ratio Combining
In this method, the individual branches must be first co-phased and weighted proportionately
to their channel gain and then summed. This is equivalent to weighting each branch by the
complex conjugate of its channel gain, i.e., cl = gl* = αl exp(–jφl), where cl is the weighting
coefficient of the l th branch. It is well known that MRC results in a maximum likelihood
(ML) receiver [25] and gives the best possible performance among the diversity combining
techniques. Fig. 1.1 shows general block diagrams of a coherent predetection and
postdetection L-branch maximal ratio combiner.
The combined noiseless signal is
(1.10) The noise powers Pn in al1 branches are assumed to be equal. Likewise, the total noise power
is the sum of the noise powers in each branch, weighted by the branch gain factors,
(1.11) The total SNR is
(1.12)
the sum of the branch SNRs.
(a)
8
(b)
Fig.1.1. Block diagram of coherent maximal ratio combining, (a) predetection (b)
postdetection
When all the diversity branches provide the same average power and they are uncorrelated,
the PDF of the total SNR with MRC in Rayleigh fading is given by [17]
(1.13)
where, Λ is the average signal-to-noise ratio per branch.
In Ricean fading with MRC [30]
(1.14)
where, )2/( 22
1σµ==∑
=
L
llT KK , lK )2/( 22 σµl= , 2µ = ∑
=
L
ll
1
2µ , Λ = (µ2/ L + 2σ2) /(2Pn) is the
mean SNR per branch, and IL-1 (x) is the (L –1)th-order modified Bessel function of the first
kind.
9
In Nakagami-m fading with MRC [31],
(1.15)
where, mT = ∑=
L
llm
1 and ΛT = ∑
=
ΛL
ll
1, Λl is the mean SNR of the lth branch, and the ratio
lm /Λl is the same for all diversity branches. 1.3.2 Equal Gain Combining
Maximal ratio combining requires complete knowledge of channel branch gains. Equal gain
combining is similar to MRC because the diversity branches are co-phased, but simpler than
MRC as the gains are set equal to a constant value of unity. The block diagram for EGC is
the same as Fig.1.1 except for the weighting coefficients cl = exp(–jφl). That is, the channel
estimator in EGC only needs to estimate the channel phase but not the channel amplitude.
The performance of EGC is not as good as optimal MRC but is comparable to MRC. In
practice, coherent postdetection EGC is useful for modulation schemes having equal energy
symbols, such as M-ary phase shift keying (MPSK) because only channel phase information
is required. For signals of unequal energy, complete channel knowledge is required for
coherent detection and therefore postdetection MRC is usually used. Predetection EGC,
however, still has merits to be used with unequal energy signals because apart from the L
channel phase estimators, predetection EGC only requires one AGC to estimate channel
amplitude after the matched filter while predetection MRC would require L AGC's to obtain
L channel amplitude knowledge.
The combined noiseless signal is given by
(1.16) and the SNR of the combiner output is
(1.17)
To find the probability distribution function of an EGC output y which is a sum of Rayleigh
RV's is a difficult task. We will adopt Beaulieu's infinite series result for the distribution of
the combiner output signal in Rayleigh, Ricean and Nakagami-m fading [32]-[34].
10
1.3.3 Selection Combining Selection diversity is generally the simplest method of dl. Its performance suffers some loss
compared to MRC and EGC. Ideal selection combining chooses the branch giving the highest
SNR at any instant. Fig. 1.2 illustrates the principle of selection combining. In practice, the
branch with the largest (S +N) is usually selected, since it is difficult to measure SNR. For
radio links using continuous transmission, e.g., frequency division multiple access (FDMA)
systems, SC is not very practical, since it requires continuous monitoring of al1 the diversity
branches. If such monitoring is performed, it is probably better to use maximal ratio
combining since the implementation is marginally more complex and the performance is
better [25]. In time division multiple access (TDMA) systems, however, a form of SC can be
implemented where the diversity branch is selected prior to the transmission of a TDMA
burst. The selected branch is then used for the transmission of the entire burst [25]. In the
following analysis, we assume ideal continuous branch selection. As far as the statistics of the
output signal are concerned, it is immaterial where the selection is done. The antenna signals
could be sampled, for example, and the best one sent to the receiver.
Fig.1.2. Block diagram of predetection selection combining.
Assuming all L diversity branches are independent and identically distributed (iid), the PDF
of the signal amplitude at the output of a selection combiner in Ricean fading is given by [33]
(1.18)
11
where, Ω = E[α2] is the total signal power in each Ricean channel, and Q(a,b) =
dxaxIxeb
xa
∫∞ +−
)(02
22
is the Marcum-Q function.
The selection combining Nakagami-m distributed γs is derived for iid diversity branches as
(1.19)
where, γ(α,x) = dttex
t∫ −−
0
1α is the incomplete gamma function [35, 8.35] and Λ = E[γs].
For selection combining and Rayleigh fading, the PDF of γs, assuming iid diversity branches
is given by [17]
(1.20)
1.4 Literature Review In this section we will discuss the previous work done on the performance of square MQAM
of fading channels using Maximal Ratio Combining (MRC) reception. Both Ricean and
Nakagami-m channels will be analyzed and Rayleigh as a special case of either Ricean or
Nakagami-m channel.
1.4.1 Ricean Fading Channel Recently, Zhang et al [7] presented a closed form series expression for the error probability
of MRC multichannel reception of MQAM coherent systems over a flat Ricean fading
channels. The derived expression was
(1.21)
where,
and,
12
Seo et al [8] derived the exact SER expression of the coherent square M-QAM scheme with
an MRC diversity reception in Ricean fading channels. Their results were in terms of
confluent hypergeometric and hypergeometric functions.
Ps(e) = X1 – X2 (1.22)
Where,
Where,
denotes the confluent hypergeometric
function.
where,
2F1(a, b; c; z) is the hypergeometric function and A1 = 1–1/ M , A2 = 1.5 log2 (M)/(M–1).
Kl is the Ricean factor of the lth diversity channel.
is the average SNR per bit at the combiner output, is
the average SNR per bit on the lth diversity channel.
Manjeet et al [9] came up with a closed form expression for SER of MQAM in Ricean fading
environment under MRC diversity reception. The derived expression is in terms of a single
finite integral with an integrand composed of elementary functions.
(1.23)
13
where, g = 3/[2(M – 1)], B = 1–1/ M , sgkh γθθ += 2sin)( and k = K + L.
1.4.1 Nakagami-m Fading Channel Falujah et al [10] proposed a method for computing the exact average symbol error
probability (SEP) of the square MQAM with MRC diversity over independent Nakagami-m
fading channels with arbitrary fading index m. They proposed a closed form expression for
the average SEP over L independent diversity channels in terms of Gaussian hypergeometric
functions.
2122)( HHqqePM −+−= (1.24)
where,
(1.25)
q = 1–1/ M , p = 1.5 log2 (M)/(M–1) and 2F1(w, x; y; z) is the Gaussian hypergeometric
function.
(1.26)
The same authors [11] proposed another solution in terms of Gaussian hypergeometric
function and Appell's hypergeometric function. The total average SEP of MQAM is
2122)( HHqqePM −+−=
Where,
given that for Re(b) > 0, and |k| < |h|; (Re(.) denotes the real value). H2 is given by
14
(1.27)
given that F2 (d; a1, a2; b1, b2; x, y) is the Appell's hypergeometric function.
Manjeet et al [12] presented the symbol error rate (SER) performance of coherent square
MQAM with l th order diversity in frequency non-selective Nakagami-m fading environment
corrupted by additive white Gaussian Noise (AWGN). They derived an expression for SER in
terms of a single finite integral with an integrand composed of elementary function.
(1.28)
Annamalai et al [13] derived an exact integral expression for calculating the SER of MQAM
in conjunction with L-fold antenna diversity on Nakagami-m fading channel with MRC
reception. The solution was derived using Parseval's theorem. It was given in terms of Gauss
hypergeometric series.
21)( IIPES −=ε (1.29)
where,
(1.30)
(1.31)
where, )/( λλ +=∧
pp .
15
CHAPTER 2 PERFORMANCE OF M-QAM TRANSMITTED OVER AWGN CHANNEL 2.1. Introduction
M-ary Quadrature Amplitude Modulation (M-QAM) is the most spectrum efficient
modulation technique. The transmitted signal varies in both the carrier phase and carrier
amplitude between known constellation points in the (I, Q) plane. The information sequence
mj is separated into 2 branches Im, Qm. The second is modulated by a quadrature carrier
of the first. The signal waveforms can be expressed as [5]:
sm(t) = Im g(t) cos (2πfct) - Qm g(t) sin(2πfct,)
where, g(t) is the pulse shape signal, Im and Qm are the information-bearing signal amplitudes
corresponding to M = 2k possible k-bit blocks of symbols. Im and Qm take the discrete values
Im = Qm = (2m-1-M)d m=1,2, … M.
The signal waveforms have two terms, each one is M -PAM signal. For square M-QAM
constellation the k-bit per sample is even integer (i.e. M = 4, 16, 64 …).
Symbol Error Rate (SER) for square constellation M-QAM can be obtained from the SER of
M -PAM, where the SER for symmetric M-PAM is given by
( ) ⎟⎟⎠⎞
⎜⎜⎝
⎛
−⎟⎠
⎞⎜⎝
⎛−=
16112)( 2
0 MNE
QM
ESER sPAM (2.1)
Since QAM is composed of quadrature combination of 2 M −PAM, each with half the total
power and since a correct QAM decision is made only when a correct symbol decision is
made independently on each of these modulations, the probability of correct symbol decision
Ps(c) for QAM can be expressed as [5,2]:
16
[ ]
[ ]22
)(11)(
)()(
ESERESER
cPcP
PAMMQAM
PAMMsQAMs
−−=
=
⎟⎟⎠
⎞⎜⎜⎝
⎛
−⎟⎟⎠
⎞⎜⎜⎝
⎛−−⎟
⎟⎠
⎞⎜⎜⎝
⎛
−⎟⎟⎠
⎞⎜⎜⎝
⎛−=
)1(3114
)1(3114)(
0
22
0 MNE
QMMN
EQ
MESER ss
QAM
(2.2)
or
( ) ( )( )γγ bQabQaESER QAM24)( −=
(2.3)
where, ⎟⎠
⎞⎜⎝
⎛ −=M
a 11 ,)1(
3−
=M
b and0N
Es=γ
As can be seen from equation (2.3) we have two terms; the first one with Q(x) and the second
has Q2(x). For simplicity equation (2.3) can be approximated by the first term only so that the
Q2(x) is ignored since Q2(x) << Q(x) [5]. Fig.2.1 shows equation (2.3) with and without
Q2(x).
12 12.5 13 13.5 14 14.5 15 15.5 1610-2
10-1SER of 16-MQAM on AWGNC for (3) with & without Q2(x)
SNR per Symbol (dB)
Pro
babi
lity
of S
ymbo
l erro
r
(3) with squer Q(x)(3) without squer Q(x)
Fig.2.1. M-QAM over AWGN channel with and without Q2(x).
17
From Fig.2.1, we can see that at 10-2
, the difference is only 0.0025 dB, so the squared term in
equation (2.3) can be omitted without major loss in the accuracy. Then, equation (2.3) can be
approximated as [5]
⎟⎟⎠
⎞⎜⎜⎝
⎛
−⎟⎟⎠
⎞⎜⎜⎝
⎛−≈
13114)(,, M
QM
EP AWGNQAMsγ (2.4)
where, γ = Es/No.
An upper-bound on BER, which is good to within 1 dB for M > 4 is given by [6]:
( ), , 2( ) 0.2exp 1.5 /( 1) logs QAM AWGNP E M Mγ< − − (2.5)
Equation (2.5) doesn’t have the error function. It has important advantage; it consists of only
elementary functions, exponential and the log functions in particular. However, the
approximation error is in the order of 1 dB. We will call this approximation Goldsmith in the
rest of this thesis. In the next section we will consider an expansion of the error function in
terms of exponential function.
2.2 Exact representation of Error Function Q(x) in terms of a series of exponentials In communication theory problems we often face Q(x), where
∫∞
−=x
u duexQ 2/2
21)(π
(2.6)
Few years ago, the following form appeared [1]
∫ ⎟⎟
⎠
⎞⎜⎜⎝
⎛ −=
2/
02
2
sin2exp1Q(x)
π
θπx
(2.7)
This new form has the following advantages [2]:
1) The integration limits are independent of the argument of the function.
2) The integrand has a Gaussian form with respect to the function.
18
3) It has finite integration limits [0, π/2].
Both equations (2.6) and (2.7) can not be evaluated in closed form. While it is not possible to
get a solution for equation (2.7) in a simple closed form expression not involving infinite
sums, so a numerical approximation needs to be implemented. Equation (2.7) can be written
in the following improved exponential form by applying numerical integration using right
rectangular rule [1]:
∫∑−
⎟⎟⎠
⎞⎜⎜⎝
⎛ −≤
=
i
i
dxxQi
N
i
θ
θ
θθπ
1
2
2
1 sin2exp1)(
( )∑=
−≤N
iii xbaxQ
1
2exp)(
where, πθθ 1−−
= iiia ,
iib
θ2sin21
= andN
ii 2
πθ = .
or [3] ∑=
⎟⎟⎠
⎞⎜⎜⎝
⎛ −≤
N
i iNR
xN
xQ1
2
2
, sin2exp
21)(
θ (2.8)
where N
ii 2
πθ =
By increasing N, the bound tends to the exact value [1]. For example:
N=1, ⎟⎟⎠
⎞⎜⎜⎝
⎛ −≤
2exp
21)(
2
1xxQR (2.9)
N=2, ( ) ⎥⎦
⎤⎢⎣
⎡⎟⎟⎠
⎞⎜⎜⎝
⎛ −+−≤
2expexp
41)(
22
2xxxQR (2.10)
N=3, ( ) ⎥⎦
⎤⎢⎣
⎡⎟⎟⎠
⎞⎜⎜⎝
⎛ −+⎟⎟
⎠
⎞⎜⎜⎝
⎛ −+−≤
2exp
32exp2exp
61)(
222
3xxxxQR (2.11)
N=6, ( ) ( ) ⎥⎦
⎤⎢⎣
⎡⎟⎟⎠
⎞⎜⎜⎝
⎛ −+⎟⎟⎠
⎞⎜⎜⎝
⎛
+−
+⎟⎟⎠
⎞⎜⎜⎝
⎛ −+−+−+⎟⎟
⎠
⎞⎜⎜⎝
⎛
−−
≤2
exp32
2exp32expexp2exp
322exp
121)(
22222
2
6xxxxxxxQR
(2.12)
Fig.2.2 shows the plot of equations (2.7), (2.9), (2.10), (2.11), and (2.12) respectively. For the
case when N=1, we have the Chernoff-Rubin bound [1].
19
0 2 4 6 8 10 12 1410-6
10-5
10-4
10-3
10-2
10-1
x(dB)= 20 log (x)
log
Q(x
)
ExactN=1N=2N=3N=6
Fig. 2.2: Q(x) Exact given by equation (2.7) Vs. approximations given by (2.9), (2.10),
(2.11), & (2.12).
If we apply trapezoidal rule to equation (2.7), then we have [4]:
∑−
=
−
⎟⎟⎠
⎞⎜⎜⎝
⎛ −+=
1
12
22/
, sin2exp
21
4)(
2 N
i i
x
NTx
NNexQ
θ (2.13)
where, N
ii 2
πθ =
Substituting in equation (2.13) for different values of N
N=1, 2/1,
2
41)( x
T exQ −≤ (2.14)
N=2, 22
41
81)( 2/
2,xx
T eexQ −− +≤ (2.15)
N=3, ( )3/222/3,
222
61
121)( xxx
T eeexQ −−− ++≤ (2.16)
N=4, ( )2222 )22/(2)22/(22/4, 8
1161)( xxxx
T eeeexQ −+−−−− +++≤ (2.17)
20
N=6, ( )3/22)32/(2)32/(22/6,
222222
121
241)( xxxxxx
T eeeeeexQ −−−+−−−− +++++≤ (2.18)
For the case when N=2 for an arbitrary pointθ , we have
( )θπθθ
222 sin2/2/2, 4
124
1),( xxT eexgQ −− +⎟
⎠⎞
⎜⎝⎛ −=≈ (2.19)
For optimumθ , 3/πθ =opt [1], equation (2.19) is given by
3/22/2,
22
41
121 xx
optT eeQ −− +≈ (2.20)
For high order of N (N ≥ 3) the optimum θ is almost like the equispaced one. Fig. 2.3 shows
Q(x) exact and three approximations for N=1, 2, 2opt. Fig.2.4 is zoomed version of Fig. 2.3.
0 2 4 6 8 10 12 1410-6
10-5
10-4
10-3
10-2
10-1
x(dB)= 20 log (x)
log
Q(x
)
ExactQT1QT2
QT2op
Fig.2.3. Exact Vs. approximations of equations (2.14), (2.15) and (2.20)
21
0 1 2 3 4 5 6 7 8 9 1010-3
10-2
10-1
x(dB)= 20 log (x)
log
Q(x
)ExactQT1QT2
QT2op
Fig 2.4: Exact Vs. approximations of equations (2.14), (2.15) and (2.20)
From Fig 2.4, we can see that equation (2.20) gives a tight upper bound with maximum
difference at x = 5dB or 0.5dB at 10-1.4. Equation (2.15) gives good approximation but it is
under the exact for 3.5 < x < 9 dB and the difference is maximum at x = 14dB. Fig. 2.5 shows
that for N ≥ 3 the trapezoidal overlaps the exact.
0 2 4 6 8 10 12 1410-6
10-5
10-4
10-3
10-2
10-1
x(dB)= 20 log (x)
log
Q(x
)
ExactQT3QT4
QT6
Fig.2.5. Exact Vs. approximations of equations (2.16), (2.17) and (2.18)
22
Table 2.1 and 2.2 provides a summary of the differences between the exact and the
approximation using rectangular rule and Trapezoidal rule in dB for different number of
terms respectively.
Table 2.1: Rectangular rule
Difference in dB using Rectangular Rule Log Q(x) N=1 N=2 N=3 N=6
10-1 2.922 1.587 1.132 0.6375 10-2 1.601 0.8045 0.6135 0.3415 10-3 1.1445 0.634 0.4325 0.25 10-4 0.905 0.536 0.351 0.2035 10-5 0.7545 0.467 0.3065 0.174 10-6 0.6505 0.4145 0.277 0.1535
Table 2.2: Trapezoidal Rule
Difference in dB using Trapezoidal Rule Log Q(x) N=1 N=2 N=2opt N=3 N=4 N=6
10-1 0.476 0.118 0.73 -0.052 -0.004 0.0005 10-2 0.754 -0.079 0.2775 0.0125 -0.0025 0 10-3 0.6305 0.062 0.0995 -0.0085 0.001 0 10-4 0.536 0.134 0.0415 -0.005 0 0 10-5 0.467 0.159 0.0315 0.012 -0.0015 0 10-6 0.415 0.1655 0.0375 0.0295 -0.0005 0
Comparing tables 2.1 and 2.2, it is clear that trapezoidal rule converges faster hence it is
better to be used in evaluating Q(x) numerically. For N > 6, equation (2.13) represents Q(x)
nearly exactly using sum of exponentials.
2.3 Exact representation of Error Function Squared Q2(x) in terms of a series of
exponentials
Another function which needs to be evaluated in communication theory especially when we
deal with QAM is Q2(x) which is defined as [2] follows:
∫ ⎟⎟⎠
⎞⎜⎜⎝
⎛ −=
4/
02
22
sin2exp1(x)Q
π
θπx
23
Since Q(x) and Q2(x) are both monotonically increasing functions, we can apply trapezoidal
rule on Q2(x) as we did for Q(x) to get
∑−
=
−
⎟⎟⎠
⎞⎜⎜⎝
⎛ −+=
1
12
22
sin2exp
41
8)(
2 N
i i
x xNN
exQθ
(2.21)
where, N
ii 4
πθ =
In appendix I it is shown that the best form to approximate Q2(x) using the minimum number
of exponential is
( )22 exp81)( xxQ −≈ (2.22)
2.4. M-QAM Performance over AWGN Channel using the Exponential Form
Using the exponential form of Q(x) and Q2(x) equation (2.3) can be written as:
2 2
/ 2 1 1/ 2sin / 2sin
1 1
1( ) 44 2 8 4
i i
b bm nb b
QAMi i
e ae aSER E a e em m n n
γ γγ α γ β
− −− −− −
= =
⎧ ⎫= + − −⎨ ⎬
⎩ ⎭∑ ∑
(2.23)
m
ii 2
πα = n
ii 4
πβ =
where, ∑−
=
−
⎟⎟⎠
⎞⎜⎜⎝
⎛ −+=
1
12
22/
sin2exp
21
4)(
2 m
i i
x xmm
exQθ
, with m
ii 2
πθ =
and ∑−
=
−
⎟⎟⎠
⎞⎜⎜⎝
⎛ −+=
1
12
22
sin2exp
41
8)(
2 n
i i
x xnn
exQθ
, with n
ii 4
πθ =
Let m = 2n then equation (2.23) will be
2 2/ 2 2 1 1
/ 2sin / 2sin
1 1
1( ) 48 4 8 4
i i
b bn nb b
QAMi i
e ae aSER E a e en n n n
γ γγ θ γ θ
− −− −− −
= =
⎧ ⎫= + − −⎨ ⎬
⎩ ⎭∑ ∑
24
and n
ii 4
πθ =
2 2/ 2 2 1 1
/ 2sin / 2sin
1 1( )
2 2i i
b bn nb b
QAMi i
a e aeSER E e a en
γ γγ θ γ θ
− −− −− −
= =
⎧ ⎫= + − −⎨ ⎬
⎩ ⎭∑ ∑
2 2/ 2 1 2 1 1
/ 2sin / 2sin / 2sin
1 1( )
2 2i i i
b bn n nb b b
QAMi i n i
a e aeSER E e e a en
γ γγ θ γ θ γ θ
− −− − −− − −
= = =
⎧ ⎫= + + − −⎨ ⎬
⎩ ⎭∑ ∑ ∑
( ) 2 2/ 2 1 2 1
/ 2sin / 2sin
1( ) 1
2 2i i
b b n nb b
QAMi i n
a e aeSER E a e en
γ γγ θ γ θ
− − − −− −
= =
⎧ ⎫= − + − +⎨ ⎬
⎩ ⎭∑ ∑ (2.24)
where, ⎟⎠
⎞⎜⎝
⎛ −=M
a 11 ,)1(
3−
=M
b ,0N
Es=γ and n
ii 4
πθ =
The probability of symbol error of M-QAM with square constellation over AWGN channel in
exponential form is given by equation (2.24). Next equation (2.24) will be examined for
different values of n by comparing equation (2.24) with the exact value computed using the
built-in MATLAB function (erfc).
n =1, ( )/ 2( ) (2 )2
b bQAM
aSER E e a eγ γ− −≈ + − (2.25)
n = 2, ( )/ 2
2 /(2 2 ) 2 /(2 2 )(2 )( ) 12 2 2
b bb b
QAMa e a eSER E a e e
γ γγ γ
− −− − − +⎛ ⎞−
≈ + + − +⎜ ⎟⎝ ⎠
(2.26)
n = 3, ( )( )/ 2
2 /(2 3) 2 2 /3 2 /(2 3)(2 )( ) 12 2 2
b bb b b b
QAMa e a eSER E a e e e e
γ γγ γ γ γ
− −− − − − − +⎛ ⎞−
≈ + + − + + +⎜ ⎟⎝ ⎠
(2.27)
Fig.2.6 shows the exact and approximate SER given by equation (2.25) for M= 4, 16 and 64
respectively. We can see that even for n =1 we have excellent approximation with maximum
difference of 0.3 dB for M = 64 at SER = 10-1, 0.27dB and 0.2dB for M=16 and M=4,
respectively. If we plot equation (2.26) with n = 2 we can see that both exact and approximate
25
overlap as shown in fig.2.8. The maximum difference between equations (2.26) and (2.3) for
all M is at SER= 10-1, which are 0.04dB, 0.05dB and 0.06dB for M=4, 16 and 64
respectively. For n = 3, the largest difference is 0.025dB at M = 64 and SER = 10-1. For n =
6, the largest difference is 0.006dB at M = 64 and SER = 10-1. Making n=16 will lead to
difference less than 0.0008 dB for all M and for 0 < SNR< 30 dB.
Also the approximation proposed by Chiani [1] can be used to express the system
performance by using equations (2.20) and (2.22) in (2.3) to get what we will call Chiani
approximation in the rest of the thesis. Then, the SER of square M-QAM over AWGN
channel using Chiani approximation is
/ 2
2 / 3
3 2
b bb
AWGNe aeSER a e
γ γγ
− −−⎛ ⎞
≈ + −⎜ ⎟⎝ ⎠
(2.28)
2.5 Concluding Remarks
The essence of the above results is summarized below:
1. The error function can be represented by an infinite series of exponential functions
using numerical integration.
2. Numerical integration using trapezoidal rule provide much better results as compared
to rectangular rule.
3. Error rate for M-QAM over Gaussian channels can be represented as a sum of N
exponential functions.
4. The above error rate analysis can be obtained with any required accuracy by simply
increasing N.
26
0 5 10 15 20 25 3010-6
10-5
10-4
10-3
10-2
10-1
SNR per Symbol (dB)
Pro
babi
lity
of S
ymbo
l erro
rExactApp
M=4M=4
M=16
M=64
Fig.2.6. M-QAM on AWGN for both exact and approximate when n =1.
4 5 6 7 8 9 10 11 12 13 1410
-6
10-5
10-4
10-3
10-2
10-1
SNR per Symbol (dB)
Pro
babi
lity
of S
ymbo
l erro
r
ExactApp,n=1
Fig.2.7. 4-QAM on AWGN for both exact and approximate when n =1.
27
4 5 6 7 8 9 10 11 12 13 1410-6
10-5
10-4
10-3
10-2
10-1
SNR per Symbol (dB)
Pro
babi
lity
of S
ymbo
l erro
rExactApp
Fig.2.8. 4-QAM on AWGN for both exact and approximate when n = 2.
4 5 6 7 8 9 10 11 12 13 1410
-6
10-5
10-4
10-3
10-2
10-1
SNR per Symbol (dB)
Pro
babi
lity
of S
ymbo
l erro
r
ExactApp
Fig.2.9. 4-QAM on AWGN both exact & approximate when n = 3.
28
12 13 14 15 16 17 18 19 20 2110-6
10-5
10-4
10-3
10-2
10-1
SNR per Symbol (dB)
Pro
babi
lity
of S
ymbo
l erro
rExactApp
Fig.2.10. 16-QAM on AWGN for both exact and approximate when n = 1.
12 13 14 15 16 17 18 19 20 2110
-6
10-5
10-4
10-3
10-2
10-1
SNR per Symbol (dB)
Pro
babi
lity
of S
ymbo
l erro
r
ExactApp
Fig.2.11. 16-QAM on AWGN for both exact and approximate when n = 2.
29
12 13 14 15 16 17 18 19 20 2110-6
10-5
10-4
10-3
10-2
10-1
SNR per Symbol (dB)
Pro
babi
lity
of S
ymbo
l erro
rExactApp
Fig.2.12. 16-QAM on AWGN for both exact and approximate when n = 3.
19 20 21 22 23 24 25 26 27 2810
-6
10-5
10-4
10-3
10-2
10-1
SNR per Symbol (dB)
Pro
babi
lity
of S
ymbo
l erro
r
ExactApp
Fig.2.13. 64-QAM on AWGN for both exact and approximate when n = 1.
30
19 20 21 22 23 24 25 26 27 2810-6
10-5
10-4
10-3
10-2
10-1
SNR per Symbol (dB)
Pro
babi
lity
of S
ymbo
l erro
rExactApp
Fig.2.14. 64-QAM on AWGN for both exact and approximate when n = 2.
19 20 21 22 23 24 25 26 27 2810
-6
10-5
10-4
10-3
10-2
10-1
SNR per Symbol (dB)
Pro
babi
lity
of S
ymbo
l erro
r
ExactApp
Fig.2.15. 64-QAM on AWGN for both exact and approximate when n = 3.
31
19 19.5 20 20.5 21 21.5 22 22.5 23 23.5 2410
-3
10-2
10-1
SNR per Symbol (dB)
Pro
babi
lity
of S
ymbo
l erro
rExactApp
Fig.2.16. Zoomed Version of 64-QAM on AWGN for both exact & approximate when n = 3.
19 19.5 20 20.5 21 21.5 2210
-2
10-1
SNR per Symbol (dB)
Pro
babi
lity
of S
ymbo
l erro
r
ExactApp
Fig.2.17. Zoomed version of 64-QAM on AWGN for both exact & approximate when n = 3.
32
0 5 10 15 20 25 3010-6
10-5
10-4
10-3
10-2
10-1
SNR per Symbol (dB)
Pro
babi
lity
of S
ymbo
l erro
rExactApp
Fig.2.18. M-QAM on AWGN for both exact and approximate when n = 2
33
CHAPTER 3
PERFORMANCE OF M-QAM TRANSMITTED OVER RICEAN CHANNEL WITH MRC 3.1 Introduction
In this chapter, the average symbol error rate (ASER) performance of coherent square M-
QAM with L-th order diversity in frequency nonselective slowly Ricean fading environment
for single channel and multiple reception using MRC are presented. Ricean distribution is
considered here, because this distribution is not only the best fit for digital signals received in
line-of-sight (LOS) communication links [5], but also because there is a close agreement
between the Ricean distribution and variety of propagation paths spanning nearly all
frequency bands [2]. Moreover, like Nakagami distribution, Ricean distribution is also a two
parameter distribution, that provides more flexibility and accuracy in matching the observed
signal statistics. M-QAM considered here is the widely used modulation technique due to its
high spectral efficiency.
We assume that there are L diversity channels carrying the same transmitted signal. Each
channel is modeled as frequency non-selective slowly Ricean fading channel corrupted by
additive white Gaussian noise (AWGN) process. The fading processes among the L diversity
channels are assumed to be mutually statistically independent. The noise processes in the L
diversity channels are assumed to be mutually statistically independent, with identical
distribution (iid). For a M-QAM demodulator [5] which makes its decision based on the
output of a linear filter operating on an undistorted symbol waveform, it is well known that
the probability of symbol error caused by AWGN depends only on the instantaneous SNR (γ)
associated with each symbol.
3.2 Performance of M-QAM over Single Ricean Channel
From chapter 2, we have seen that the Symbol Error Rate (SER) for square constellation M-
QAM over AWGN channel is given by [5]
34
⎟⎟⎠
⎞⎜⎜⎝
⎛
−⎟⎠
⎞⎜⎝
⎛−−⎟
⎟⎠
⎞⎜⎜⎝
⎛
−⎟⎠
⎞⎜⎝
⎛−=
)1(3114
)1(3114
0
22
0 MNEQ
MMNEQ
MSER ss
QAM (3.1)
or ( ) ( )[ ]γγ bQabQaESERQAM24)( −= (3.2)
where, ⎟⎠
⎞⎜⎝
⎛ −=M
a 11 , )1(
3−
=M
b , 0N
Es=γ
Writing equation (3.1) in the exponential form we have,
( )⎭⎬⎫
⎩⎨⎧
+−+−= ∑ ∑−
=
−
=
−−−− 1
1
12//
2/
122
)(n
i
n
ni
SbSbbb
QAMii eeaaee
naESER γγ
γγ
(3.3)
where, ⎟⎠
⎞⎜⎝
⎛ −=M
a 11 , )1(
3−
=M
b , 0N
Es=γ , n
ii 4
πθ = 22sini iS θ=
To find the Average Symbol Error Rate (ASER) of QAM system over a single Ricean
channel we need first to determine the pdf of the instantaneous SNR, γ . In fading γ becomes
a random variable and for Ricean fading environment, the probability density function (pdf)
of instantaneous SNR γ at the output of channel is given as [2]
⎟⎟⎠
⎞⎜⎜⎝
⎛ +⎥⎦
⎤⎢⎣
⎡ +−−+=
γγ
γγ
γγγ
)1(20)1(exp1)( KKIKKeKp (3.4)
where, γ is the expected value ofγ , K is the Ricean parameter and 0I (.) is the zeroth-order
modified Bessel function of the first kind. Equation (3.4) can be written in a compact form as
[ ] ( )γγγγ rKIrKrep 20exp)( −−= (3.5)
where, γ
Kr +=
1
35
Once the statistics of the instantaneous SNR γ are determined as a function of the Ricean
parameter K and the average instantaneous SNRγ , the Average Symbol Error Rate of Ricean
channel (ASERR) as a function of K and γ of the system can be calculated by averaging the
conditional probability of SER over the pdf ofγ . i.e.,
γγγ dpEPASER S∫∞
=0
)()./( (3.6)
where, )/( γEPS is the conditional probability of symbol error which is the SER over AWGN
channel given by equation (3.3). Substituting equations (3.3) & (3.5) in equation (3.6) to
derive ASERR we have
( )⎪⎩
⎪⎨⎧ −×−−
= ∫∞
0
202
2/γγ
γγdrKI
reben
KareASERR ( )∫∞ −×−
−0
202γγ
γγdrKI
rebae
( )∑∫−
=
∞
××−+1
1 0
20/
)1(n
i
dIreiSbea γβγγγ
( )⎪⎭
⎪⎬⎫
××+ ∑∫−
=
∞12
0
20/n
ni
dIreiSbe γβγγγ
(3.7)
Using the following relationship [35]
( ) ( )∫∞
⎟⎠⎞
⎜⎝⎛=−
0exp120exp
ccdxxIcx ββ
Equation (3.7) can be written as
( )rbrb
Kra
rbrb
Kr
n
KareASERR +
⎟⎠⎞
⎜⎝⎛
+−
⎪⎪⎩
⎪⎪⎨
⎧
+
⎟⎠⎞
⎜⎝⎛
+−=
2
exp
22
2exp
( )∑−
= +
⎟⎟⎠
⎞⎜⎜⎝
⎛+
−+
1
1
exp1n
i i
i
ii
rSbrSb
KrSSa
⎪⎪⎭
⎪⎪⎬
⎫
+
⎟⎟⎠
⎞⎜⎜⎝
⎛+
+∑−
=
1
1
expn
i i
i
ii
rSbrSb
KrSS
36
( )rbrb
Kba
rbrb
Kb
narASERR +
⎟⎠⎞
⎜⎝⎛
+−
−
⎪⎪⎩
⎪⎪⎨
⎧
+
⎟⎠⎞
⎜⎝⎛
+−
=2
exp
22
exp
( )
⎪⎪⎭
⎪⎪⎬
⎫
+
⎟⎟⎠
⎞⎜⎜⎝
⎛+−
++
⎟⎟⎠
⎞⎜⎜⎝
⎛+−
−+ ∑∑
−
=
−
=
121
1
expexp1n
ni i
iin
i i
ii
rSbrSb
KbS
rSbrSb
KbSa (3.8)
( ) ( )( )
⎪⎪⎩
⎪⎪⎨
⎧
++
⎟⎟⎠
⎞⎜⎜⎝
⎛++
−+
=Kb
KbKb
nKaASERR 12
12exp
1γγ
γ
( )
( )( )KbKb
Kba
++
⎟⎟⎠
⎞⎜⎜⎝
⎛++
−
−12
1exp
γγ
γ
( ) ( )
( )( )
( )⎪⎪⎭
⎪⎪⎬
⎫
++
⎟⎟⎠
⎞⎜⎜⎝
⎛++
−
+++
⎟⎟⎠
⎞⎜⎜⎝
⎛++
−−
+ ∑∑−
=
−
=
121
1 11
exp
11
exp1n
ni i
iin
i i
ii
SKbSKb
KbS
SKbSKb
KbSa
γγ
γ
γγ
γ
(3.9)
Equation (3.9) represents the ASER of square QAM over single Ricean fading channel as a
function of K (Ricean parameter) and the average SNRγ . If we let K= ∞ in equation (3.9) we
will get equation (3.3) which is the SER of M-QAM over AWGN channel. By substituting
K=0 in equation (3.9) which represents Rayleigh fading channel, we get
⎩⎨⎧
+=
21γbn
aASERRay ( )12 +−
γba ( )
⎭⎬⎫
++
+−
+ ∑∑−
=
−
=
121
1
1 n
ni i
in
i i
i
SbS
SbSa
γγ
⎩⎨⎧
+=
21γbn
aASERRay ( )12 +−
γba
⎭⎬⎫
+−
++ ∑∑
−
=
−
=
1
1
12
1
n
i i
in
i i
i
SbaS
SbS
γγ (3.10)
Fig.3.1 shows 16-QAM over Ricean fading channel for different values of K that are 0, 2, 6,
and 12. These values usually used in the literatures because they close to the values observed
in the experimental work. From fig.3.1 we can see that small change in K leads to big change
in the system performance. Both figs.3.2 and 3.3 shows the performance of square M-QAM
37
over single Ricean channel. Where fig.3.2 is for both K= 0 and 2. And fig.3.3 for K=6 and
12.
0 5 10 15 20 25 30 35 4010-6
10-5
10-4
10-3
10-2
10-1
100
SNR per Symbol (dB)
Pro
babi
lity
of S
ymbo
l erro
r
K=0K=2K=5K=12AWGN
16-QAM
Fig.3.1. Exact 16-QAM over single Ricean with K= 0, 2, 6 & 12.
0 5 10 15 20 25 30 35 40
10-4
10-3
10-2
10-1
100
SNR per Symbol (dB)
Pro
babi
lity
of S
ymbo
l erro
r
K=0,RayleighK=2
+ 4 -QAM* 16-QAMo 64-QAM
Fig.3.2. Exact M-QAM over single Ricean with K= 0 & 2.
38
0 5 10 15 20 25 30 35 4010-6
10-5
10-4
10-3
10-2
10-1
100
SNR per Symbol (dB)
Pro
babi
lity
of S
ymbo
l erro
rK=6K=12
+ 4 -QAM* 16-QAMo 64-QAM
Fig.3.3. Exact M-QAM over single Ricean with K= 6 & 12.
3.3 M-QAM Performance over Ricean Channel with MRC Diversity.
In the previous section, the performance of square M-QAM over single Ricean channel was
obtained. In this section, the system performance will be considered when the Maximal Ratio
Combiner (MRC) is used. The MRC diversity system has L antennas. With an assumption
that the received signal from each antenna is subjected to slow and flat Ricean channel and
these channels are independent and identical distributed (iid). Then the probability density
function (pdf) of γ after MRC diversity combining, is given by [7]-[9].
( )⎟⎟⎠
⎞⎜⎜⎝
⎛ +−⎥
⎦
⎤⎢⎣
⎡ +−−−
⎟⎟⎠
⎞⎜⎜⎝
⎛ ++=
γγ
γγ
γγ
γγγ
)(21)(exp2
1
)( KLKLIKLKe
L
KKLKLp
where, L is the number of branches, γ is the expected value of γ (SNR), K is the Ricean
parameter and 1−LI (.) is the modified Bessel function of L-1 order.
39
Let γ
KLr += &
21−
=Lv
[ ] ( )γγγγγ rKvIrKev
Krrp v 22exp)( −−⎟⎠⎞
⎜⎝⎛= (3.11)
The ASER as functions of K, L, and γ of the system can be calculated by averaging the
conditional probability of error over the pdf ofγ . This could be done by substituting
equations (3.3) & (3.11) in equation (3.6)
( )⎪⎩
⎪⎨⎧ −−
⎟⎠⎞
⎜⎝⎛−
= ∫∞
0, 222
2/γγ
γγγ drKvIberevv
Kr
n
KareASER MRCRic
( )∫∞ −−
−0
222γγ
γγγ drKvIbereva
( ) ( ) γγγ γγ drKvIeean
i
Sbrv i 2211
1 0
/∑∫−
=
∞−−−+ ( )
⎭⎬⎫
+ ∑∫−
=
∞−− γγγ γγ drKvIee
n
ni
Sbrv i 2212
0
/
( )( )
⎪⎩
⎪⎨⎧ +−
⎟⎠⎞
⎜⎝⎛−
= ∫∞
0, 222
2/2γγ
γγ drKvIrbevv
Kr
n
KareASER MRCRic
( ) ( )∫
∞ +−
−0
222γγγ γ
drKvIea rbv
( ) ( ) ( ) γγγ γ drKvIean
i
SrSbv ii 2211
1 0
/∑∫−
=
∞+−−+ ( ) ( )
⎭⎬⎫
+ ∑∫−
=
∞+− γγγ γ drKvIe
n
ni
SrSbv ii 2212
0
/
(3.12)
Using the following relationship [35]
⎟⎟
⎠
⎞
⎜⎜
⎝
⎛∫∞
= +−−
αββαβα
2exp
0)2(2
2)12( vvxv dxxvIex
40
( )⎥⎦⎤
⎢⎣⎡
+−
⎪⎩
⎪⎨⎧ +
⎟⎠⎞
⎜⎝⎛
+⎟⎠⎞
⎜⎝⎛=
rbKbv
rb
v
Kr
n
vrKarASER MRCRic 2exp
12
22
21
,
⎥⎦⎤
⎢⎣⎡
+−
⎟⎠⎞
⎜⎝⎛
+−
+
rbKb
rba v
exp12
12
( )⎥⎥⎦
⎤
⎢⎢⎣
⎡
+−
⎟⎟⎠
⎞⎜⎜⎝
⎛+
−+ ∑−
=
+
irSbKb
rSbSa
n
i
v
i
i exp11
1
12
⎪⎭
⎪⎬⎫
⎥⎥⎦
⎤
⎢⎢⎣
⎡
+−
⎟⎟⎠
⎞⎜⎜⎝
⎛+
+ ∑−
=
+
irSbKb
rSbSn
ni
v
i
i exp12 12
(3.13)
⎥⎦⎤
⎢⎣⎡
+−
⎪⎩
⎪⎨⎧
⎟⎠⎞
⎜⎝⎛
+=
rbKbL
rbr
naASER MRCRic 2
exp2
221
, ⎥⎦⎤
⎢⎣⎡
+−
⎟⎠⎞
⎜⎝⎛
+−
rbKb
rbra L
exp2
( )⎥⎥⎦
⎤
⎢⎢⎣
⎡
+−
⎟⎟⎠
⎞⎜⎜⎝
⎛+
−+ ∑−
= irSbKb
rSbrSa
n
i
L
i
i exp11
1
⎪⎭
⎪⎬⎫
⎥⎥⎦
⎤
⎢⎢⎣
⎡
+−
⎟⎟⎠
⎞⎜⎜⎝
⎛+
+ ∑−
= irSbKb
rSbrSn
ni
L
i
i exp12
(3.14)
where, ⎟⎠
⎞⎜⎝
⎛ −=M
a 11 , )1(
3−
=M
b , 0N
Es=γ , n
ii 4
πθ = ,
γKLr +
= , Si=2sin2 (θi) , γ = mean(γ )
Equation (3.14) is the ASER of QAM using MRC of L branches over Ricean channels that is
iid. Then it is varied in γ (average signal-noise ratio), L (number of MRC branches), Ricean
parameter, ∑=
=L
jjKK
1, and M (the modulation index).
For L=1, equation (3.14) is reduced to equation (3.9). If K= ∞ then equation (3.14) will be
equation (3.3) where the MRC doesn't improve the system. When K=0 then equation (3.14)
represents a Rayleigh flat slowly fading channel and equation (3.14) is reduced to
LL
MRCRay LbLa
LbL
naASER ⎟⎟
⎠
⎞⎜⎜⎝
⎛+
−⎪⎩
⎪⎨⎧
⎟⎟⎠
⎞⎜⎜⎝
⎛+
=γγ 22
221
, ( )∑−
=⎟⎟⎠
⎞⎜⎜⎝
⎛+
−+1
1
1n
i
L
i
i
LSbLSa
γ
⎪⎭
⎪⎬⎫
⎟⎟⎠
⎞⎜⎜⎝
⎛+
+ ∑−
=
12n
ni
L
i
i
LSbLS
γ
(3.15)
Fig.3.4 shows the performance of Rayleigh channel using MRC for different values of L (1,
2, 3, 4, 8, 16) for 16-QAM and fig.3.5 shows 16-QAM over Ricean channel with K=8. In
41
fig.3.5 we can see 16-QAM with diversity over both Rayleigh and Ricean with K=2 and in
fig.3.7 the same system but the Ricean parameter K is 6 and 12. Form figs.3.4 – 3.7 we can
see that the improvement in the system performance decreases when the K increase and also
when the order of diversity increase (the number of used branches).
15 20 25 30 3510-6
10-5
10-4
10-3
10-2
10-1
SNR per Symbol (dB)
AS
ER
L=1L=2L=3L=4L=8L=16
Fig.3.4. Exact 16-QAM over Rayleigh for different L.
42
15 20 25 30 3510-6
10-5
10-4
10-3
10-2
10-1
SNR per Symbol (dB)
Pro
babi
lity
of S
ymbo
l erro
r
L=1L=2L=3L=4L=8L=16
Fig.3.5. Exact16-QAM with MRC of different L over Ricean with K = 8.
0 5 10 15 20 25 30 35 4010-6
10-5
10-4
10-3
10-2
10-1
100
SNR per Symbol (dB)
Pro
babi
lity
of S
ymbo
l erro
r
L1L=2L=4L=8
o K=0+ K=2* AWGN
16-QAM
Fig.3.6. Exact 16-QAM over Ricean with K=0, 2 and MRC with L = 1,2,4 and 8.
43
15 20 25 3010-6
10-5
10-4
10-3
10-2
10-1
SNR per Symbol (dB)
Pro
babi
lity
of S
ymbo
l erro
r
L1L=2L=4L=8
o K=6+ K=12* AWGN
16-QAM
Fig.3.7. Exact 16-QAM over Ricean with K=6, 12 and MRC with L = 1,2,4 and 8.
3.3.1 Approximation of M-QAM Performance over Ricean Channel with MRC
Diversity.
In the previous sections we discussed the exact system performance where we get equation
(3.14). This equation is good because it is a simple closed form but we need to make n very
large to get the exact solution. In this section the error between the exact and different
approximation formulas will be analyzed for square M-QAM over single Ricean channel and
when MRC reception is used. These approximated formulas are based on equation (3.14),
Goldsmith, and Chiani. Where Goldsmith, and Chiani were introduced in chapter 2.
The first approximation we will call it " n = 1 " which can be obtained by letting n=1 in
equation (3.14) we get
,2 exp
2 2 2
L
Ric MRCa r KbASER
b r b r−⎛ ⎞ ⎡ ⎤≈ ⎜ ⎟ ⎢ ⎥+ +⎝ ⎠ ⎣ ⎦
2
exp2
La r Kbab r b r
⎛ ⎞ −⎛ ⎞ ⎡ ⎤+ −⎜ ⎟⎜ ⎟ ⎢ ⎥+ +⎝ ⎠ ⎣ ⎦⎝ ⎠
(3.16)
44
where, ⎟⎠
⎞⎜⎝
⎛ −=M
a 11 , )1(
3−
=M
b , 0N
Es=γ , γ
KLr += , γ = mean(γ )
For single Ricean channel we let L=1 in equation (3.16) to have the approximate system
performance as
,2 exp
2 2 2Ric MRCa r KbASER
b r b r−⎛ ⎞ ⎡ ⎤≈ ⎜ ⎟ ⎢ ⎥+ +⎝ ⎠ ⎣ ⎦
2
exp2
a r Kbab r b r
⎛ ⎞ −⎛ ⎞ ⎡ ⎤+ −⎜ ⎟⎜ ⎟ ⎢ ⎥+ +⎝ ⎠ ⎣ ⎦⎝ ⎠
(3.17)
As a special case for single Rayleigh channel where K=0 and L=1 (3.16) reduced to
22 2Raya rASER
b r⎛ ⎞≈ ⎜ ⎟+⎝ ⎠
2
2a ra
b r⎛ ⎞⎛ ⎞+ −⎜ ⎟⎜ ⎟+⎝ ⎠⎝ ⎠
(3.18)
And for Rayleigh channel with MRC diversity the system performance can be approximated
when n=1 as
,2
2 2
L
Ray MRCa rASER
b r⎛ ⎞≈ ⎜ ⎟+⎝ ⎠
2
2
La rab r
⎛ ⎞⎛ ⎞+ −⎜ ⎟⎜ ⎟+⎝ ⎠⎝ ⎠ (3.19)
Of course by increasing n we will get better approximation as we will see later.
The second approximation that we consider is Goldsmith approximation. From chapter 2, we
have symbol error rate of M-QAM over AWGN channel is
AWGN 21.5SER 0.2exp log
1M
Mγ−⎡ ⎤≈ ⎢ ⎥−⎣ ⎦
(3.20)
By substituting equations (3.20) and (3.11) in equation (3.6) and using the relation [35]
⎟⎟
⎠
⎞
⎜⎜
⎝
⎛∫∞
= +−−
αββαβα
2exp
0)2(2
2)12( vvxv dxxvIex
We will get the second approximation of M-QAM over Ricean with MRC as
( )( )( )( ) ( )( )Ri-MRC 2
1 1.5ASER 0.2 log exp1 1.5 1 1.5
LM L K KM
M L K M L Kγ
γ γ
⎛ ⎞ ⎡ ⎤− + −≈ ⎜ ⎟ ⎢ ⎥⎜ ⎟− + + − + +⎢ ⎥⎝ ⎠ ⎣ ⎦
(3.21)
The last approximate expression is based on Chiani approximation. By going though the
same steps in the previous expressions but with SER over AWGN channel as
45
/ 2
2 / 3
3 2
b bb
AWGNe aeSER a e
γ γγ
− −−⎛ ⎞
≈ + −⎜ ⎟⎝ ⎠
Then approximate system performance can be expressed as
,1 2 exp3 2 2
L
Ric MRCr KbASER a
b r b r⎧ −⎪ ⎛ ⎞ ⎡ ⎤≈ ⎨ ⎜ ⎟ ⎢ ⎥+ +⎝ ⎠ ⎣ ⎦⎪⎩
3 2exp2 3 2 3
Lr Kbb r b r
−⎛ ⎞ ⎡ ⎤+⎜ ⎟ ⎢ ⎥+ +⎝ ⎠ ⎣ ⎦
exp2
La r Kbb r b r
⎫− ⎪⎛ ⎞ ⎡ ⎤− ⎬⎜ ⎟ ⎢ ⎥+ +⎝ ⎠ ⎣ ⎦⎪⎭
(3.22)
3.3.2 Analysis of Approximation
In fig. 3.8, we can see 16-QAM over single Rayleigh for both exact and approximate with
n=1 we find that the difference is 0.2 dB at ASER= 10-2. And in fig.3.9, we see zoom in of
fig. 3.8 but with Goldsmith and Chiani approximations where the differences are 0.2 and 0.6
dB for Chiani and Goldsmith respectively. These results are listed in more detailed in table
3.1.The figures 3.10 to 3.21 show different combination of K (Ricean parameter) and L
(diversity order) for exact and the approximations that presented in section 3.4.1. The tables
3.1 to 3.10 summarized the error difference between the exact and the approximation when
n= 1,2,3,4, and 10 also the Goldsmith and Chiani approximation for different M, L and K.
Where equations (3.21) and (3.22) are used for the approximation of Goldsmith and Chiani
respectively. And equation (3.14) was used in the approximation for n= 1,2,3,4, and 10 where
equation (3.16) is special case of (3.14).
3.4 Concluding Remarks
The essence of the above results is summarized below:
1. The system performance can be expressed by a simple finite series of elementary
functions that is the exponential.
2. The error decays rapidly with increasing values of n.
3. At fixed n, the error increases when the modulation index is increased.
4. At fixed n, K and L, the error decreases when SNR is increased.
46
15 20 25 30 3510-3
10-2
10-1
SNR per Symbol (dB)
AS
ER
n=1Exact
Fig.3.8. ASER of 16-QAM on Rayleigh channel for n=1 and Exact.
27.6 27.8 28 28.2 28.4 28.6 28.8 29 29.2
10-2
SNR per Symbol (dB)
Pro
babi
lity
of S
ymbo
l erro
r
16-QAM, Single Rayleigh
Exactn=1GoldsmithChiani
Fig.3.9. Exact and Approx. of 16-QAM over Single Rayleigh (K=0).
47
24 24.2 24.4 24.6 24.8 25 25.2 25.4 25.6
10-2
SNR per Symbol (dB)
Pro
babi
lity
of S
ymbo
l erro
r
16-QAM, Single Rician (K=2)
Exactn=1GoldsmithChiani
Fig.3.10 Exact and Approx. of 16-QAM over Single Ricean (K=2).
15 20 25 3010-4
10-3
10-2
10-1
SNR per Symbol (dB)
Pro
babi
lity
of S
ymbo
l erro
r
K=6,n=1K=12,n=1K=6,ExactK=12,Exact
Fig.3.11. ASER of 16-QAM on Ricean channel of K=6 &12 for n =1 and Exact.
48
24 24.2 24.4 24.6 24.8 25 25.2
10-3
SNR per Symbol (dB)
Pro
babi
lity
of S
ymbo
l erro
r
16-QAM, Single Rician( K=6)
Exactn=1GoldsmithChiani
Fig.3.12 Exact and Approx. of 16-QAM over Single Ricean (K= 6).
23.1 23.2 23.3 23.4 23.5 23.6 23.7 23.8 23.9 24
10-4
SNR per Symbol (dB)
Pro
babi
lity
of S
ymbo
l erro
r
16-QAM, Single Rician (K=12)
Exactn=1GoldsmithChiani
Fig.3.13 Exact and Approx. of 16-QAM over Single Ricean (K=12).
49
15 20 25 3010-4
10-3
10-2
10-1
SNR per Symbol (dB)
Pro
babi
lity
of S
ymbo
l erro
rAppr,n=1ExactL=1
L=2
L=3L=8
Fig.3.14. 16-QAM over Rayleigh for different L and for both exact and approximate, n = 1.
31 31.2 31.4 31.6 31.8 32 32.2 32.4 32.6 32.8
10-4
SNR per Symbol (dB)
Pro
babi
lity
of S
ymbo
l erro
r
16-QAM, MRC-L=2 over Rician with K=0
Exactn=1GoldsmithChiani
Fig.3.15 Exact and Approx. of 16-QAM over Rayleigh with MRC of L = 2.
50
27.2 27.3 27.4 27.5 27.6 27.7 27.8 27.9 28 28.1
10-5
SNR per Symbol (dB)
Pro
babi
lity
of S
ymbo
l erro
r
16-QAM, MRC-L=4 over Rician with K=0 (Rayleigh)
Exactn=1GoldsmithChiani
Fig.3.16 Exact and Approx. of 16-QAM over Rayleigh with MRC of L = 4.
23.3 23.4 23.5 23.6 23.7 23.8 23.9 24
10-5
SNR per Symbol (dB)
Pro
babi
lity
of S
ymbo
l erro
r
16-QAM, MRC-L=8 over Rician with K=0 (Rayleigh)
Exactn=1GoldsmithChiani
Fig.3.17 Exact and Approx. of 16-QAM over Rayleigh with MRC of L = 8.
51
20 21 22 23 24 25 26 27 28 29 30
10-4
10-3
SNR per Symbol (dB)
Pro
babi
lity
of S
ymbo
l erro
rExactExactExactExactn=1n=1n=1n=1
L=1
L=8
L=4
L=2
Fig.3.18. 16-QAM over Ricean (K=6) for different L and for both Exact and Approx., n =1.
30.1 30.2 30.3 30.4 30.5 30.6 30.7 30.8 30.9 31
10-5
SNR per Symbol (dB)
Pro
babi
lity
of S
ymbo
l erro
r
16-QAM, MRC-L=2 over Rician with K=6
Exactn=1GoldsmithChiani
Fig.3.19 Exact and Approx. of 16-QAM over Ricean (k = 6) with MRC of L = 2.
52
25.3 25.4 25.5 25.6 25.7 25.8 25.9 26 26.1 26.2
10-5
SNR per Symbol (dB)
Pro
babi
lity
of S
ymbo
l erro
r
16-QAM, MRC-L=4 over Rician with K=6
Exactn=1GoldsmithChiani
Fig.3.20 Exact and Approx. of 16-QAM over Ricean (k = 6) with MRC of L = 4.
22.8 22.9 23 23.1 23.2 23.3 23.4 23.5
10-5
SNR per Symbol (dB)
Pro
babi
lity
of S
ymbo
l erro
r
16-QAM, MRC-L=8 over Rician with K=6
Exactn=1GoldsmithChiani
Fig.3.21 Exact and Approx. of 16-QAM over Ricean (k = 6) with MRC of L = 8.
53
Table 3.1: Difference in dB between exact and approximate for M-QAM over single Rayleigh fading channel.
ASER for M-QAM over Single Rayleigh(K=0) 1.E-01 1.E-02 1.E-03 1.E-04 1.E-05 1.E-06
M n Difference (dB) 4 Goldsmith 0.9839 0.5891 0.5588 0.5558 4 Chiani 0.1796 0.2256 0.2285 0.2288 4 1 0.1879 0.1685 0.1665 0.1663 4 2 0.0449 0.0402 0.0397 0.0397 4 3 0.0198 0.0177 0.0175 0.0175 4 4 0.0111 0.01 0.0098 0.0098 4 10 0.0018 0.0016 0.0016 0.0015 16 Goldsmith 0.9194 0.9171 0.9166 16 Chiani 0.1188 0.1554 0.1582 16 1 0.2922 0.2682 0.2658 16 2 0.0693 0.0634 0.0629 16 3 0.0305 0.0279 0.0277 16 4 0.0171 0.0157 0.0155 16 10 0.0027 0.0025 0.0025 64 Goldsmith 2.2647 2.1368 64 Chiani 0.0809 0.1169 64 1 0.35 0.3231 64 2 0.0825 0.076 64 3 0.0363 0.0334 64 4 0.0204 0.0187 64 10 0.0033 0.0029
54
Table 3.2: Difference in dB between exact and approximate for 4-QAM over single Ricean fading channel (K = 2,6,&12).
ASER for 4-QAM over Single Ricean 1.E-01 1.E-02 1.E-03 1.E-04 1.E-05 1.E-06
K n Difference (dB) 2 Goldsmith 0.5633 0.4213 0.533 0.5531 2 Chiani 0.287 0.245 0.2309 0.229 2 1 0.1719 0.1581 0.1648 0.1662 2 2 0.0428 0.0376 0.0393 0.0396 2 3 0.0188 0.0166 0.0173 0.0175 2 4 0.0106 0.0093 0.0097 0.0098 2 10 0.0017 0.0014 0.0015 0.0016 6 Goldsmith 0.1069 0.227 0.1214 0.2235 0.4957 6 Chiani 0.4266 0.307 0.2786 0.2578 0.2342 6 1 0.1675 0.1143 0.1061 0.1407 0.1623 6 2 0.0428 0.0247 0.0255 0.0337 0.0387 6 3 0.0186 0.0109 0.0113 0.0148 0.0171 6 4 0.0104 0.0061 0.0064 0.0083 0.0096 6 10 0.0017 0.001 0.001 0.0013 0.0016
12 Goldsmith 0.0374 0.3892 0.4108 0.3952 0.3421 0.1688 12 Chiani 0.4619 0.2733 0.2048 0.1958 0.2213 0.2591 12 1 0.1821 0.0973 0.0298 0.0152 0.0368 0.0862 12 2 0.0422 0.0173 0.0108 0.0105 0.0142 0.0226 12 3 0.0185 0.0077 0.0049 0.0048 0.0063 0.01 12 4 0.0104 0.0043 0.0028 0.0027 0.0036 0.0056 12 10 0.0017 0.0007 0.0005 0.0004 0.0005 0.0009
55
Table 3.3: Difference in dB between exact and approximate for 16-QAM over single Ricean fading channel (K = 2,6,&12).
ASER for 16-QAM over Single Ricean 1.E-01 1.E-02 1.E-03 1.E-04 1.E-05 1.E-06
K n Difference (dB) 2 Goldsmith 0.9409 0.9147 0.916 2 Chiani 0.2114 0.1748 0.1606 2 1 0.2709 0.2537 0.2637 2 2 0.0635 0.0599 0.0624 2 3 0.0278 0.0263 0.0275 2 4 0.0156 0.0148 0.0154 2 10 0.0024 0.0023 0.0025 6 Goldsmith 0.9782 0.8834 0.8819 0.9067 6 Chiani 0.3418 0.2621 0.2341 0.1933 6 1 0.2613 0.1618 0.1706 0.2307 6 2 0.058 0.0367 0.0409 0.0547 6 3 0.0252 0.0162 0.018 0.0241 6 4 0.0141 0.009 0.0101 0.0135 6 10 0.0022 0.0014 0.0016 0.0022
12 Goldsmith 0.9829 0.8349 0.7755 0.7701 0.8052 0.8648 12 Chiani 0.3728 0.2372 0.1868 0.1829 0.2049 0.2235 12 1 0.2683 0.1088 0.0416 0.0356 0.073 0.1496 12 2 0.0554 0.0233 0.0158 0.0165 0.0232 0.0374 12 3 0.0241 0.0104 0.0072 0.0075 0.0104 0.0165 12 4 0.0134 0.0059 0.0041 0.0042 0.0058 0.0093 12 10 0.0021 0.001 0.0007 0.0007 0.0009 0.0014
56
Table 3.4: Difference in dB between exact and approximate for 64-QAM over single Ricean fading channel (K = 2,6,&12).
ASER for 64-QAM over Single Ricean 1.E-01 1.E-02 1.E-03 1.E-04 1.E-05 1.E-06
K n Difference (dB) 2 Goldsmith 2.0191 2.0549 2 Chiani 0.1727 0.137 2 1 0.3225 0.3063 2 2 0.0746 0.072 2 3 0.0327 0.0317 2 4 0.0183 0.0178 2 10 0.0029 0.0028 6 Goldsmith 1.6639 1.441 1.5996 6 Chiani 0.3061 0.2407 0.2096 0.1581 6 1 0.302 0.1875 0.2061 0.2801 6 2 0.0657 0.0431 0.0492 0.0661 6 3 0.0285 0.0189 0.0217 0.0291 6 4 0.0159 0.0106 0.0121 0.0163 6 10 0.0025 0.0017 0.0019 0.0026
12 Goldsmith 1.5467 1.1868 1.0881 1.108 1.2445 1.5539 12 Chiani 0.3386 0.223 0.1788 0.1759 0.1944 0.202 12 1 0.3005 0.1154 0.0485 0.0467 0.0925 0.1843 12 2 0.0617 0.0263 0.0183 0.0197 0.0281 0.0456 12 3 0.0267 0.0117 0.0083 0.0089 0.0125 0.0202 12 4 0.0149 0.0066 0.0047 0.005 0.0071 0.0113 12 10 0.0023 0.0011 0.0007 0.0008 0.0012 0.0018
57
Table 3.5: Difference in dB between exact and approximate for 4-QAM over Ricean fading channel (K= 0 and 2) with MRC of L branches.
ASER for 4-QAM over Ricean With MRC 1.E-01 1.E-02 1.E-03 1.E-04 1.E-05 1.E-06
K L n Difference (dB) 0 2 Goldsmith 0.308 0.1344 0.1997 0.2166 0.2215 0.2231 0 2 Chiani 0.392 0.3438 0.3224 0.3156 0.3134 0.3128 0 2 1 0.1527 0.1277 0.1127 0.1076 0.1061 0.1055 0 2 2 0.0435 0.03 0.0265 0.0255 0.0252 0.025 0 2 3 0.0191 0.0132 0.0117 0.0112 0.0111 0.011 0 2 4 0.0107 0.0074 0.0066 0.0063 0.0062 0.0062 0 2 10 0.0017 0.0012 0.0011 0.001 0.001 0.001 0 4 Goldsmith 0.0431 0.3533 0.4033 0.4151 0.4188 0.4201 0 4 Chiani 0.4521 0.3021 0.2337 0.2024 0.1868 0.1787 0 4 1 0.1695 0.1119 0.0553 0.0256 0.0102 0.0019 0 4 2 0.0438 0.0207 0.0134 0.0105 0.0092 0.0084 0 4 3 0.0187 0.0092 0.006 0.0047 0.0042 0.0038 0 4 4 0.0105 0.0051 0.0034 0.0027 0.0024 0.0022 0 4 10 0.0017 0.0008 0.0006 0.0004 0.0004 0.0003 0 8 Goldsmith 0.0711 0.4218 0.4429 0.4347 0.424 0.4149 0 8 Chiani 0.4705 0.2556 0.1566 0.1103 0.0868 0.0739 0 8 1 0.1876 0.0923 0.0065 0.0591 0.0874 0.1036 0 8 2 0.0418 0.015 0.0059 0.0028 0.0017 0.0013 0 8 3 0.0186 0.0066 0.0028 0.0015 0.0009 0.0007 0 8 4 0.0104 0.0037 0.0016 0.0009 0.0005 0.0004 0 8 10 0.0016 0.0006 0.0003 0.0001 0.0001 0.0001 2 2 Goldsmith 0.2127 0.1791 0.2182 0.2232 0.2238 0.2238 2 2 Chiani 0.4081 0.3346 0.3157 0.3128 0.3125 0.3125 2 2 1 0.1577 0.1227 0.1081 0.1056 0.1053 0.1053 2 2 2 0.0432 0.0281 0.0255 0.0251 0.025 0.025 2 2 3 0.0188 0.0124 0.0112 0.0111 0.011 0.011 2 2 4 0.0105 0.0069 0.0063 0.0062 0.0062 0.0062 2 2 10 0.0017 0.0011 0.001 0.001 0.001 0.001 2 4 Goldsmith 0.0236 0.3615 0.4066 0.4165 0.4193 0.4203 2 4 Chiani 0.4542 0.2966 0.2282 0.1984 0.1842 0.1772 2 4 1 0.1721 0.1091 0.0508 0.0219 0.0078 0.0004 2 4 2 0.0434 0.02 0.0129 0.0101 0.009 0.0084 2 4 3 0.0186 0.0088 0.0058 0.0046 0.0041 0.0038 2 4 4 0.0104 0.0049 0.0033 0.0026 0.0024 0.0022 2 4 10 0.0016 0.0008 0.0005 0.0004 0.0004 0.0003 2 8 Goldsmith 0.0745 0.4231 0.4432 0.4344 0.4235 0.4144 2 8 Chiani 0.4707 0.254 0.155 0.109 0.0858 0.0732 2 8 1 0.1884 0.0914 0.0082 0.0606 0.0885 0.1045 2 8 2 0.0417 0.0148 0.0057 0.0028 0.0017 0.0013 2 8 3 0.0187 0.0065 0.0027 0.0015 0.0009 0.0007 2 8 4 0.0104 0.0037 0.0016 0.0008 0.0006 0.0004 2 8 10 0.0017 0.0006 0.0002 0.0001 0.0001 0.0001
58
Table 3.6: Difference in dB between exact and approximate for 4-QAM over Ricean fading channel (K = 6 and 12) with MRC of L branches.
ASER for 4-QAM over Ricean with MRC 1.E-01 1.E-02 1.E-03 1.E-04 1.E-05 1.E-06
K L n Difference (dB) 6 2 Goldsmith 0.0384 0.3272 0.3474 0.3203 0.2791 0.2475 6 2 Chiani 0.4466 0.3008 0.2594 0.266 0.2864 0.3016 6 2 1 0.1716 0.108 0.0695 0.0707 0.0855 0.0971 6 2 2 0.0429 0.0213 0.0172 0.0184 0.0213 0.0235 6 2 3 0.0185 0.0094 0.0077 0.0082 0.0094 0.0104 6 2 4 0.0104 0.0053 0.0043 0.0046 0.0053 0.0059 6 2 10 0.0016 0.0008 0.0007 0.0007 0.0009 0.001 6 4 Goldsmith 0.0314 0.3927 0.4238 0.4255 0.424 0.4228 6 4 Chiani 0.4626 0.2758 0.1986 0.1695 0.1602 0.1593 6 4 1 0.1805 0.0998 0.0271 0.0045 0.0154 0.0173 6 4 2 0.0425 0.0174 0.0099 0.0076 0.007 0.0069 6 4 3 0.0185 0.0077 0.0045 0.0035 0.0033 0.0032 6 4 4 0.0103 0.0043 0.0025 0.002 0.0019 0.0018 6 4 10 0.0016 0.0007 0.0004 0.0003 0.0003 0.0003 6 8 Goldsmith 0.0886 0.4291 0.4443 0.4327 0.4203 0.4105 6 8 Chiani 0.4723 0.2471 0.146 0.1007 0.0791 0.068 6 8 1 0.1913 0.0874 0.0167 0.0694 0.0963 0.1109 6 8 2 0.0413 0.014 0.005 0.0023 0.0015 0.0011 6 8 3 0.0186 0.0061 0.0024 0.0013 0.0008 0.0006 6 8 4 0.0103 0.0035 0.0014 0.0007 0.0005 0.0003 6 8 10 0.0016 0.0005 0.0002 0.0001 0.0001 0.0001
12 2 Goldsmith 0.0547 0.4041 0.427 0.4219 0.4116 0.3925 12 2 Chiani 0.4656 0.2654 0.1859 0.1626 0.1684 0.194 12 2 1 0.1848 0.0945 0.0157 0.0127 0.0114 0.0093 12 2 2 0.042 0.0162 0.0088 0.0072 0.008 0.0106 12 2 3 0.0185 0.0072 0.004 0.0034 0.0037 0.0048 12 2 4 0.0103 0.004 0.0023 0.0019 0.0021 0.0027 12 2 10 0.0016 0.0007 0.0004 0.0004 0.0004 0.0005 12 4 Goldsmith 0.0792 0.4215 0.4397 0.4324 0.4253 0.4217 12 4 Chiani 0.4702 0.2529 0.1596 0.1221 0.1084 0.1068 12 4 1 0.1894 0.0894 0.0053 0.0491 0.0665 0.0703 12 4 2 0.0415 0.0147 0.0063 0.0038 0.0032 0.0032 12 4 3 0.0186 0.0065 0.003 0.0019 0.0016 0.0015 12 4 4 0.0103 0.0036 0.0017 0.0011 0.0009 0.0009 12 4 10 0.0016 0.0006 0.0003 0.0002 0.0002 0.0001 12 8 Goldsmith 0.1072 0.4376 0.4458 0.429 0.4129 0.4007 12 8 Chiani 0.4745 0.2369 0.1314 0.086 0.0659 0.0567 12 8 1 0.1954 0.0817 0.0306 0.0853 0.1117 0.125 12 8 2 0.0407 0.0129 0.0037 0.0015 0.0009 0.0008 12 8 3 0.0186 0.0056 0.0019 0.0009 0.0005 0.0004 12 8 4 0.0103 0.0032 0.0011 0.0005 0.0003 0.0003 12 8 10 0.0016 0.0005 0.0002 0.0001 0 0.0001
59
Table 3.7: Difference in dB between exact and approximate for 16-QAM over Ricean fading channel (K = 0 and 2) with MRC of L branches.
ASER for 16-QAM over Ricean with MRC 1.E-01 1.E-02 1.E-03 1.E-04 1.E-05 1.E-06
K L n Difference (dB) 0 2 Goldsmith 0.9815 0.92 0.9005 0.8944 0.8925 0 2 Chiani 0.3136 0.2909 0.2786 0.2745 0.2731 0 2 1 0.2626 0.1954 0.1733 0.1664 0.1642 0 2 2 0.0626 0.0449 0.0403 0.039 0.0386 0 2 3 0.0274 0.0197 0.0177 0.0171 0.0169 0 2 4 0.0153 0.011 0.0099 0.0096 0.0095 0 2 10 0.0024 0.0018 0.0015 0.0015 0.0015 0 4 Goldsmith 0.987 0.8618 0.8027 0.7728 0.757 0.7484 0 4 Chiani 0.3662 0.2625 0.2111 0.1865 0.1741 0.1676 0 4 1 0.2658 0.1368 0.0705 0.0386 0.0224 0.0138 0 4 2 0.0582 0.0288 0.0192 0.0154 0.0136 0.0126 0 4 3 0.0252 0.0127 0.0086 0.007 0.0062 0.0058 0 4 4 0.0141 0.0071 0.0048 0.004 0.0035 0.0033 0 4 10 0.0022 0.0011 0.0007 0.0007 0.0006 0.0005 0 8 Goldsmith 0.984 0.8163 0.7234 0.667 0.6305 0.6057 0 8 Chiani 0.3803 0.2201 0.1408 0.1022 0.0822 0.0711 0 8 1 0.2722 0.0907 0.0117 0.0622 0.0891 0.1045 0 8 2 0.0548 0.0189 0.0077 0.004 0.0025 0.0018 0 8 3 0.024 0.0085 0.0038 0.0021 0.0013 0.0009 0 8 4 0.0133 0.0048 0.0022 0.0012 0.0008 0.0005 0 8 10 0.0021 0.0008 0.0004 0.0002 0.0001 0.0001 2 2 Goldsmith 0.9801 0.9091 0.894 0.8919 0.8916 2 2 Chiani 0.3268 0.2854 0.2744 0.2728 0.2726 2 2 1 0.2607 0.1837 0.1662 0.1635 0.1633 2 2 2 0.0606 0.042 0.0389 0.0384 0.0384 2 2 3 0.0264 0.0184 0.0171 0.0169 0.0169 2 2 4 0.0148 0.0103 0.0096 0.0094 0.0095 2 2 10 0.0024 0.0016 0.0015 0.0015 0.0015 2 4 Goldsmith 0.9859 0.8565 0.7974 0.7688 0.7543 0.7467 2 4 Chiani 0.3676 0.2578 0.2065 0.1833 0.172 0.1663 2 4 1 0.266 0.1315 0.0648 0.0344 0.0197 0.0122 2 4 2 0.0575 0.0278 0.0184 0.0149 0.0133 0.0125 2 4 3 0.0249 0.0123 0.0083 0.0068 0.0061 0.0057 2 4 4 0.0139 0.0069 0.0046 0.0038 0.0034 0.0033 2 4 10 0.0022 0.0011 0.0007 0.0006 0.0005 0.0005 2 8 Goldsmith 0.9838 0.815 0.7216 0.6651 0.6287 0.6043 2 8 Chiani 0.3805 0.2187 0.1392 0.101 0.0812 0.0704 2 8 1 0.2724 0.0891 0.0135 0.0637 0.0902 0.1053 2 8 2 0.0546 0.0185 0.0075 0.0039 0.0025 0.0018 2 8 3 0.0239 0.0083 0.0037 0.002 0.0013 0.0009 2 8 4 0.0133 0.0046 0.0022 0.0012 0.0008 0.0006 2 8 10 0.0021 0.0007 0.0004 0.0002 0.0001 0.0001
60
Table 3.8: Difference in dB between exact and approximate for 16-QAM over Ricean fading channel (K = 6 and 12) with MRC of L branches.
ASER for 16-QAM over Ricean with MRC 1.E-01 1.E-02 1.E-03 1.E-04 1.E-05 1.E-06
K L n Difference (dB) 6 2 Goldsmith 0.9828 0.8649 0.834 0.8449 0.8668 0.8816 6 2 Chiani 0.36 0.2612 0.2337 0.2405 0.2556 0.2657 6 2 1 0.264 0.1387 0.1019 0.1124 0.1357 0.1522 6 2 2 0.0573 0.0304 0.0261 0.0288 0.0333 0.0364 6 2 3 0.0249 0.0134 0.0116 0.0128 0.0147 0.016 6 2 4 0.0139 0.0075 0.0065 0.0072 0.0082 0.009 6 2 10 0.0022 0.0012 0.001 0.0012 0.0013 0.0015 6 4 Goldsmith 0.9841 0.8363 0.7677 0.7379 0.7276 0.7267 6 4 Chiani 0.3738 0.2394 0.1805 0.1581 0.1513 0.1512 6 4 1 0.2685 0.1108 0.0338 0.0035 0.0061 0.007 6 4 2 0.0559 0.0234 0.014 0.0112 0.0105 0.0105 6 4 3 0.0243 0.0104 0.0064 0.0052 0.0049 0.0048 6 4 4 0.0135 0.0059 0.0036 0.003 0.0028 0.0027 6 4 10 0.0021 0.001 0.0006 0.0005 0.0005 0.0004 6 8 Goldsmith 0.9829 0.8086 0.7119 0.6538 0.6171 0.5932 6 8 Chiani 0.3813 0.2121 0.131 0.0935 0.0751 0.0656 6 8 1 0.2733 0.0821 0.0234 0.073 0.0981 0.1117 6 8 2 0.0542 0.0173 0.0064 0.0032 0.0021 0.0016 6 8 3 0.0237 0.0078 0.0032 0.0017 0.0011 0.0008 6 8 4 0.0132 0.0044 0.0019 0.001 0.0007 0.0005 6 8 10 0.0021 0.0007 0.0003 0.0002 0.0001 0.0001
12 2 Goldsmith 0.983 0.8266 0.755 0.7307 0.7385 0.7693 12 2 Chiani 0.3759 0.2299 0.1695 0.1529 0.1605 0.1842 12 2 1 0.2698 0.1005 0.0204 0.0033 0.0042 0.0338 12 2 2 0.0551 0.0213 0.0124 0.0109 0.0125 0.0169 12 2 3 0.024 0.0095 0.0057 0.0051 0.0057 0.0076 12 2 4 0.0134 0.0053 0.0032 0.0029 0.0032 0.0043 12 2 10 0.0021 0.0008 0.0005 0.0005 0.0005 0.0007 12 4 Goldsmith 0.9828 0.8143 0.7266 0.6814 0.6614 0.6578 12 4 Chiani 0.3795 0.2179 0.1444 0.1144 0.1038 0.1035 12 4 1 0.2721 0.0878 0.0082 0.0489 0.064 0.0659 12 4 2 0.0544 0.0185 0.0084 0.0055 0.0048 0.0049 12 4 3 0.0238 0.0083 0.0041 0.0027 0.0024 0.0024 12 4 4 0.0132 0.0047 0.0024 0.0016 0.0014 0.0014 12 4 10 0.0021 0.0007 0.0004 0.0002 0.0003 0.0002 12 8 Goldsmith 0.9819 0.7993 0.6956 0.6324 0.5922 0.5663 12 8 Chiani 0.3827 0.2023 0.1172 0.0797 0.0628 0.0551 12 8 1 0.2748 0.0717 0.0395 0.0898 0.1139 0.1261 12 8 2 0.0534 0.0153 0.0047 0.002 0.0013 0.0011 12 8 3 0.0234 0.0069 0.0025 0.0011 0.0007 0.0005 12 8 4 0.013 0.0039 0.0015 0.0007 0.0004 0.0003 12 8 10 0.002 0.0006 0.0002 0.0001 0 0.0001
61
Table 3.9: Difference in dB between exact and approximate for 64-QAM over Ricean fading channel (K = 0 and 2) with MRC of L branches.
ASER for 64-QAM over Ricean with MRC 1.E-01 1.E-02 1.E-03 1.E-04 1.E-05 1.E-06
K L n Difference (dB) 0 2 Goldsmith 1.815 1.5504 1.4865 1.4677 0 2 Chiani 0.276 0.265 0.2562 0.2531 0 2 1 0.3152 0.2306 0.2051 0.1972 0 2 2 0.0722 0.0525 0.0474 0.046 0 2 3 0.0315 0.023 0.0208 0.0202 0 2 4 0.0176 0.0129 0.0116 0.0113 0 2 10 0.0028 0.0021 0.0018 0.0018 0 4 Goldsmith 1.6123 1.2598 1.1348 1.0771 1.0476 1.0319 0 4 Chiani 0.3312 0.2461 0.2011 0.1793 0.168 0.1621 0 4 1 0.3052 0.1493 0.0786 0.0454 0.0288 0.0199 0 4 2 0.0654 0.0328 0.0222 0.0178 0.0159 0.0148 0 4 3 0.0283 0.0145 0.01 0.0081 0.0073 0.0068 0 4 4 0.0158 0.0081 0.0056 0.0046 0.0041 0.0039 0 4 10 0.0026 0.0013 0.0009 0.0007 0.0007 0.0006 0 8 Goldsmith 1.5185 1.1197 0.9542 0.8633 0.8071 0.7702 0 8 Chiani 0.3467 0.2071 0.1348 0.0991 0.0803 0.0699 0 8 1 0.302 0.0908 0.013 0.0631 0.0895 0.1046 0 8 2 0.0607 0.0206 0.0085 0.0044 0.0029 0.0021 0 8 3 0.0263 0.0093 0.0042 0.0023 0.0015 0.0011 0 8 4 0.0146 0.0053 0.0024 0.0013 0.0009 0.0006 0 8 10 0.0023 0.0009 0.0004 0.0002 0.0001 0.0001 2 2 Goldsmith 1.7477 1.5033 1.4649 1.4598 2 2 Chiani 0.29 0.2614 0.2531 0.2518 2 2 1 0.3084 0.2156 0.1967 0.1941 2 2 2 0.0694 0.0491 0.0458 0.0454 2 2 3 0.0302 0.0215 0.0201 0.02 2 2 4 0.0169 0.0121 0.0113 0.0112 2 2 10 0.0027 0.0019 0.0018 0.0018 2 4 Goldsmith 1.597 1.245 1.1232 1.0691 1.0426 1.0289 2 4 Chiani 0.3329 0.242 0.197 0.1763 0.1662 0.161 2 4 1 0.3038 0.1427 0.0724 0.041 0.0259 0.0182 2 4 2 0.0645 0.0315 0.0212 0.0172 0.0154 0.0146 2 4 3 0.0279 0.0139 0.0095 0.0078 0.007 0.0067 2 4 4 0.0156 0.0078 0.0054 0.0044 0.004 0.0038 2 4 10 0.0025 0.0012 0.0008 0.0007 0.0006 0.0006 2 8 Goldsmith 1.5156 1.1162 0.9508 0.8602 0.8045 0.768 2 8 Chiani 0.3468 0.2057 0.1334 0.0979 0.0795 0.0692 2 8 1 0.3018 0.089 0.015 0.0647 0.0907 0.1055 2 8 2 0.0605 0.0203 0.0083 0.0043 0.0028 0.0021 2 8 3 0.0262 0.0092 0.0041 0.0022 0.0015 0.0011 2 8 4 0.0145 0.0052 0.0023 0.0013 0.0008 0.0006 2 8 10 0.0023 0.0009 0.0004 0.0002 0.0001 0.0001
62
Table 3.10: Difference in dB between exact and approximate for 64-QAM over Ricean fading channel (K= 6 and 12) with MRC of L branches.
ASER for 64-QAM over Ricean With MRC 1.E-01 1.E-02 1.E-03 1.E-04 1.E-05 1.E-06
K L n Difference (dB) 6 2 Goldsmith 1.6093 1.2991 1.2567 1.3102 1.3829 6 2 Chiani 0.3251 0.244 0.2207 0.2264 0.2386 6 2 1 0.3017 0.1548 0.1194 0.1345 0.1622 6 2 2 0.0643 0.035 0.0307 0.0342 0.0394 6 2 3 0.0279 0.0154 0.0136 0.0152 0.0174 6 2 4 0.0155 0.0087 0.0077 0.0085 0.0097 6 2 10 0.0024 0.0014 0.0012 0.0014 0.0015 6 4 Goldsmith 1.5518 1.1833 1.0556 1.0061 0.9913 0.9914 6 4 Chiani 0.3397 0.2252 0.173 0.153 0.1471 0.1469 6 4 1 0.3016 0.1168 0.0379 0.0079 0.0013 0.0017 6 4 2 0.0622 0.0261 0.016 0.0129 0.0122 0.0123 6 4 3 0.0269 0.0116 0.0073 0.006 0.0056 0.0057 6 4 4 0.015 0.0065 0.0041 0.0034 0.0032 0.0033 6 4 10 0.0024 0.001 0.0006 0.0005 0.0005 0.0005 6 8 Goldsmith 1.5037 1.1001 0.9321 0.8409 0.7859 0.751 6 8 Chiani 0.3478 0.1995 0.1254 0.0907 0.0735 0.0647 6 8 1 0.3012 0.0806 0.0253 0.0741 0.0985 0.1119 6 8 2 0.0599 0.0187 0.0072 0.0036 0.0024 0.0018 6 8 3 0.0259 0.0085 0.0036 0.0019 0.0013 0.0009 6 8 4 0.0144 0.0048 0.0021 0.0011 0.0008 0.0006 6 8 10 0.0023 0.0007 0.0004 0.0001 0.0001 0.0001
12 2 Goldsmith 1.5323 1.157 1.0313 1.0005 1.0311 1.1123 12 2 Chiani 0.3419 0.2163 0.1626 0.1483 0.1559 0.1778 12 2 1 0.3007 0.1044 0.0238 0.0019 0.0124 0.0466 12 2 2 0.0611 0.0238 0.0142 0.0128 0.0149 0.0202 12 2 3 0.0264 0.0106 0.0066 0.0059 0.0068 0.0091 12 2 4 0.0147 0.006 0.0037 0.0034 0.0038 0.0051 12 2 10 0.0023 0.001 0.0006 0.0006 0.0006 0.0008 12 4 Goldsmith 1.5117 1.1178 0.9649 0.8945 0.8665 0.8648 12 4 Chiani 0.3458 0.205 0.1386 0.1112 0.1017 0.1017 12 4 1 0.3009 0.0882 0.0085 0.0483 0.0625 0.0636 12 4 2 0.0602 0.0204 0.0095 0.0063 0.0056 0.0058 12 4 3 0.026 0.0092 0.0046 0.0031 0.0028 0.0028 12 4 4 0.0145 0.0052 0.0027 0.0018 0.0016 0.0017 12 4 10 0.0023 0.0009 0.0005 0.0003 0.0003 0.0003 12 8 Goldsmith 1.4879 1.076 0.901 0.8043 0.7459 0.7095 12 8 Chiani 0.3492 0.19 0.1121 0.0773 0.0616 0.0545 12 8 1 0.3008 0.0683 0.0423 0.0912 0.1146 0.1265 12 8 2 0.059 0.0165 0.0052 0.0024 0.0015 0.0011 12 8 3 0.0255 0.0076 0.0028 0.0013 0.0008 0.0005 12 8 4 0.0142 0.0043 0.0016 0.0008 0.0005 0.0003 12 8 10 0.0022 0.0007 0.0003 0.0001 0.0001 0
63
CHAPTER 4
PERFORMANCE OF M-QAM TRANSMITTED OVER NAKAGAMI-m CHANNEL WITH MRC 4.1 Introduction
In this chapter, the Average Symbol Error Rate (ASER) performance of a coherent square M-
QAM with L-th order diversity in frequency nonselective slowly Nakagami-m fading
environment using MRC reception is presented. Nakagami-m distribution is considered here,
because this distribution is not only the best fit for digital signals received in urban radio
multipath fading channels [5], but also because there is a close agreement between the
Nakagami-m distribution and variety of propagation paths spanning nearly all frequency
bands [2]. Moreover, like Rice distribution, Nakagami-m distribution is also a two parameter
distribution, that provides more flexibility and accuracy in matching the observed signal
statistics. M-QAM considered here is the widely used modulation technique due to its high
spectral efficiency.
We assume that there are L diversity channels carrying the same transmitted signal. Each
channel is modeled as frequency non-selective slowly Nakagami-m fading channel corrupted
by additive white Gaussian noise (AWGN) process. The fading processes among the L
diversity channels are assumed to be mutually statistically independent. The noise processes
in the L diversity channels are assumed to be mutually statistically independent, with
identical distribution (iid). For a M-QAM demodulator [5] which makes its decision based on
the output of a linear filter operating on an undistorted symbol waveform, it is well known
that the probability of symbol error caused by AWGN depends only on the instantaneous
SNR (γ) associated with each symbol.
4.2 Performance of M-QAM over Nakagami Channel with MRC diversity
In fading, γ becomes a random variable and for Nakagami-m fading environment, the
probability density function (pdf) of instantaneous SNR γ at the output of the kth channel is
given as [10]-[12],
64
( ) ( )( )
1 exp /k km m
k k kkk
k k
mmpm
γ γ γγ
γ
− −⎛ ⎞= ⎜ ⎟ Γ⎝ ⎠
where Г(.) is the gamma function, mk is the fading parameter of kth channel and kγ is the
average of γ.
For identical and independent channels (iid), the pdf of γ at the output of the MRC may be
written as [10]-[12],
( ) ( )( )
1 exp /mL mL mLm LpmL
γ γ γγ
γ
− −⎛ ⎞= ⎜ ⎟ Γ⎝ ⎠
(4.1)
where, m is the Nakagami fading parameter, which is assumed to be identical for all channels
(mk= m) and kγ is the average SNR associated with each symbol, which is related to γ as
1
L
kk
γ γ=
=∑
Once the statistics of the instantaneous SNR γ are determined as a function of the Nakagami
parameter m and the average instantaneous SNRγ , the Average Symbol Error Rate of
Nakagami channel (ASERNa) as a function of m and γ of the system can be calculated by
averaging the conditional probability of SER over the pdf ofγ . i.e.,
γγγ dpEPASER S∫∞
=0
)()./( (4.2)
where, )/( γEPS is the conditional probability of symbol error which is the SER over non-
fading channel corrupted by AWGN channel. From chapter 2, the SER of QAM which is
given in the exponential form and can be written as
( )⎭⎬⎫
⎩⎨⎧
+−+−= ∑ ∑−
=
−
=
−−−− 1
1
12//
2/
122
)(n
i
n
ni
SbSbbb
QAMii eeaaee
naESER γγ
γγ
(4.3)
We now derive the expression for average SER of coherent square MQAM with L-th order
diversity in frequency non-selective slowly Nakagami-m fading channel using MRC
reception by substituting equations (4.1) & (4.3) in equation (4.2) to get ASERMRC-Na to get
65
( ) ( )⎭⎬⎫
⎩⎨⎧
⎥⎦
⎤⎢⎣
⎡+−+−⎟⎟
⎠
⎞⎜⎜⎝
⎛ −Γ⎟⎟
⎠
⎞⎜⎜⎝
⎛= ∫ ∑ ∑
∞ −
=
−
=
−−−−
−−
0
1
1
12//
2/1 1
22exp
n
i
n
ni
SbSbbb
mLmL
NaMRCii eeaaeedmL
mLnamLASER γγ
γγ
γγγγ
γ
(4.4)
Using the following relationship given in Appendix II
( ) ( ) kk ukdxuxX −∞
− Γ=−∫0
1 exp
( ) ( ) ( )
⎪⎭
⎪⎬⎫
⎪⎩
⎪⎨⎧
⎟⎟⎠
⎞⎜⎜⎝
⎛++⎟⎟
⎠
⎞⎜⎜⎝
⎛+−+⎟⎟
⎠
⎞⎜⎜⎝
⎛+−⎟⎟
⎠
⎞⎜⎜⎝
⎛+Γ⋅
Γ⎟⎟⎠
⎞⎜⎜⎝
⎛=
−−
=
−
=
−−−
− ∑∑mLn
ni i
n
i
mL
i
mLmLmL
NaMRC SbmL
SbmLabmLabmLmL
mLnamLASER
121
1
1222
1γγγγγ
On further simplification, we get
( )⎪⎭
⎪⎬⎫
⎪⎩
⎪⎨⎧
⎟⎟⎠
⎞⎜⎜⎝
⎛++⎟⎟
⎠
⎞⎜⎜⎝
⎛+−+⎟
⎠⎞
⎜⎝⎛ +−⎟
⎠⎞
⎜⎝⎛ += ∑∑
−
=
−−−
=
−−
−
121
11111
221
21 n
ni
mL
i
mLn
i i
mLmL
NaMRC mLSb
mLSba
mLba
mLb
naASER γγγγ
(4.5)
4.3 Results and Analysis of M-QAM over Nakagami Channel with MRC diversity
Equation (4.5) represents the ASER of coherent square M-QAM over Nakagami-m fading
channels that are iid using MRC reception. Then it is varied in γ (average signal-noise ratio),
L (number of MRC branches), Nakagami parameter m, and M (the modulation index).
For L=1 equation (4.5) represent the ASER of M-QAM over single Nakagami-m channel,
which is given by
( )1 2 1
1
1 1 1 1 1 12 2 2
m mm m n n
Sin Nai i ni i
a b a b b bASER an m m mS mS
γ γ γ γ− −− − − −
−= =
⎧ ⎫⎛ ⎞ ⎛ ⎞⎪ ⎪⎛ ⎞ ⎛ ⎞= + − + + − + + +⎨ ⎬⎜ ⎟ ⎜ ⎟⎜ ⎟ ⎜ ⎟⎝ ⎠ ⎝ ⎠ ⎝ ⎠ ⎝ ⎠⎪ ⎪⎩ ⎭
∑ ∑
(4.6)
If we let m = ∞ in equation (4.6) we will get equation (4.3) which is the SER of M-QAM over
AWGN channel. By substituting m =1 in equation (4.6) which represents Rayleigh fading
channel, we get
66
⎩⎨⎧
+=
21γbn
aASERRay ( )12 +−
γba ( )
⎭⎬⎫
++
+−
+ ∑∑−
=
−
=
121
1
1 n
ni i
in
i i
i
SbS
SbSa
γγ
⎩⎨⎧
+=
21γbn
aASERRay ( )12 +−
γba
⎭⎬⎫
+−
++ ∑∑
−
=
−
=
1
1
12
1
n
i i
in
i i
i
SbaS
SbS
γγ (4.7)
And the performance of Rayleigh fading channel when MRC is used can be obtained by
making m = 1 in equation (4.5) to get
( )1 2 1
1
1 1 1 1 1 12 2 2
L LL L n n
MRC Rayi i ni i
a b a b b bASER an L L LS LS
γ γ γ γ− −− − − −
−= =
⎧ ⎫⎛ ⎞ ⎛ ⎞⎪ ⎪⎛ ⎞ ⎛ ⎞= + − + + − + + +⎨ ⎬⎜ ⎟ ⎜ ⎟⎜ ⎟ ⎜ ⎟⎝ ⎠ ⎝ ⎠ ⎝ ⎠ ⎝ ⎠⎪ ⎪⎩ ⎭
∑ ∑
(4.8)
Equation (4.8) can be simplified more to be identical to equation (3.15) which is
1 22 2 2
L L
MRC Raya L a LASERn b L b Lγ γ−
⎧ ⎛ ⎞ ⎛ ⎞⎪= −⎨ ⎜ ⎟ ⎜ ⎟+ +⎝ ⎠ ⎝ ⎠⎪⎩( )∑
−
=⎟⎟⎠
⎞⎜⎜⎝
⎛+
−+1
11
n
i
L
i
i
LSbLSa
γ
⎪⎭
⎪⎬⎫
⎟⎟⎠
⎞⎜⎜⎝
⎛+
+ ∑−
=
12n
ni
L
i
i
LSbLS
γ (4.9)
Fig.4.1 shows 16-QAM over Nakagami-m fading channel for different values of m that are
0.5, 1, 2, and 5. These values are usually used in the literatures because they are close to the
values observed in the experimental work. From fig.4.1 we can see that small change in m
leads to big change in the system performance. For m = 0.5 that represent the worst case or
more severe than Rayleigh that has m=1. Both figs.4.2 and 4.3 show the performance of
square M-QAM over single Nakagami-m channel. Where figure 4.2 is for both m= 0.5 and 1.
And fig.4.3 for m = 2 and 5.
Fig.4.4 shows the performance of Nakagami-m channel with m = 0.5 and 1 using MRC for
different values of L (1, 2, & 4) for 16-QAM and fig.4.5 shows 16-QAM over Nakagami
channel with m = 2 and 5 with the same order of diversity. Form figs.3.4 & 3.5 we can see
that the improvement in the system performance decreases when the m increase and also
when the order of diversity increase (the number of used branches).
67
0 5 10 15 20 25 30 35 4010-6
10-5
10-4
10-3
10-2
10-1
100
SNR per Symbol (dB)
Pro
babi
lity
of S
ymbo
l erro
r
m=0.5m=1m=2m=5AWGN
Fig.4.1.Exact 16-QAM over single Nakagami with m=0.5, 1, 2 and 5
0 5 10 15 20 25 30 35 4010
-4
10-3
10-2
10-1
100
SNR per Symbol (dB)
Pro
babi
lity
of S
ymbo
l erro
r
m=0.5m=1
o 4QAM+ 16QAM* 64QAM
Fig.4.2.Exact M-QAM over single Nakagami with m=0.5 & 1.
68
0 5 10 15 20 25 30 35 4010-6
10-5
10-4
10-3
10-2
10-1
100
SNR per Symbol (dB)
Pro
babi
lity
of S
ymbo
l erro
r
m=0.5m=1
o 4QAM+ 16QAM* 64QAM
Fig.4.3. Exact M-QAM over single Nakagami with m = 2 and 5
0 5 10 15 20 25 30 35 4010-6
10-5
10-4
10-3
10-2
10-1
100
SNR per Symbol (dB)
Pro
babi
lity
of S
ymbo
l erro
r
+ m = 0.5o m = 1* AWGN
m=0.5,L=1
m=0.5,L=2 &m=1, L=1m=0.5,L=4 &m=1, L=2
m=1,L=4
Fig.4.4. Exact 16-QAM over Nakagami with m=0.5, 1 and L=1, 2 & 4.
2 5
69
12 14 16 18 20 22 24 26 28 3010-6
10-5
10-4
10-3
10-2
10-1
SNR per Symbol (dB)
Pro
babi
lity
of S
ymbo
l erro
r
m=2,L=1m=2,L=2m=2,L=4m=5,L=1m=5,L=2m=5,L=4AWGN
Fig.4.5. Exact 16-QAM over Nakagami with m = 2, 5 and L=1, 2 & 4
4.4 Approximation of M-QAM over Nakagami-m Channel with MRC diversity
In the previous sections we discussed the exact system performance where we get equation
(4.5). This equation is good because it is a simple closed form but we need to make n very
large to get the exact solution. In this section, the error between the exact and different
approximation formulas will be analyzed for square M-QAM over single Nakagami-m
channel and when MRC reception is used. These approximated formulas are based on
equation (4.5), Goldsmith [6], and Chiani [1]. Where Goldsmith, and Chiani were introduced
in chapter 2.
The first approximation we call it " n = 1 " which can be obtained by making n=1 in equation
(4.5) we get
( )1 2 12 2
mL mL
MRC Naa b bASER a
mL mLγ γ− −
−
⎧ ⎫⎪ ⎪⎛ ⎞ ⎛ ⎞≈ + + − +⎨ ⎬⎜ ⎟ ⎜ ⎟⎝ ⎠ ⎝ ⎠⎪ ⎪⎩ ⎭
(4.10)
where, ⎟⎠
⎞⎜⎝
⎛ −=M
a 11 , )1(
3−
=M
b , 0N
Es=γ , γ = mean(γ )
For single Nakagami-m channel we let L=1 in equation (4.10) to have the approximate
system performance as
70
( )1 2 12 2
m m
MRC Naa b bASER a
m mγ γ− −
−
⎧ ⎫⎪ ⎪⎛ ⎞ ⎛ ⎞≈ + + − +⎨ ⎬⎜ ⎟ ⎜ ⎟⎝ ⎠ ⎝ ⎠⎪ ⎪⎩ ⎭
(4.11)
As a special case for single Rayleigh channel where m=1 and L=1 equation (4.10) reduced to
22 2Raya rASER
b r⎛ ⎞≈ ⎜ ⎟+⎝ ⎠
2
2a ra
b r⎛ ⎞⎛ ⎞+ −⎜ ⎟⎜ ⎟+⎝ ⎠⎝ ⎠
(4.12)
And for Rayleigh channel with MRC diversity the system performance can be approximated
when n=1 as
,2
2 2
L
Ray MRCa rASER
b r⎛ ⎞≈ ⎜ ⎟+⎝ ⎠
2
2
La rab r
⎛ ⎞⎛ ⎞+ −⎜ ⎟⎜ ⎟+⎝ ⎠⎝ ⎠ (4.13)
Of course by increasing n we will get better approximation as we will see later.
The second approximation that we consider is Goldsmith approximation. The pdf of γ at the
output of MRC of Nakagami channel is given by equation (4.1)
( ) ( )( )
1 exp /mL mL mLm LpmL
γ γ γγ
γ
− −⎛ ⎞= ⎜ ⎟ Γ⎝ ⎠
The ASER is given by equation (4.2)
γγγ dpEPASERA SNaMRCp ∫∞
− =0
)()./(
From chapter 2 we have symbol error rate of M-QAM over AWGN channel is
AWGN 21.5( / ) SER 0.2exp log
1SP E MM
γγ −⎡ ⎤= ≈ ⎢ ⎥−⎣ ⎦ (4.14)
Therefore,
( )( )
2 1
0
(0.2) log 1.5exp1
mLmL
MRC Na
MmL LmASER dmL M
γ γγ γγ γ
∞−
−
⎛ ⎞ ⎛ ⎞−≈ −⎜ ⎟ ⎜ ⎟Γ −⎝ ⎠ ⎝ ⎠
∫
( )( ) ( )2(0.2) log 1.5
1
mL mLMmL LmmLmL Mγ γ
−⎛ ⎞ ⎛ ⎞
≈ Γ +⎜ ⎟ ⎜ ⎟Γ −⎝ ⎠ ⎝ ⎠
71
( )( )21.5 0.2 log
1
mL mLmL Lm M
Mγ γ
−⎛ ⎞ ⎛ ⎞
≈ + ⋅⎜ ⎟ ⎜ ⎟−⎝ ⎠ ⎝ ⎠
( )21.50.2 1 log( 1)
mL
MRC NaASER MmL M
γ−
−
⎛ ⎞≈ +⎜ ⎟−⎝ ⎠
(4.15)
The last approximate expression is based on Chiani approximation. By going though the
same steps in the previous expressions but with SER over AWGN channel as
/ 2
2 / 3
3 2
b bb
AWGNe aeSER a e
γ γγ
− −−⎛ ⎞
≈ + −⎜ ⎟⎝ ⎠
(4.16)
Then approximate system performance can be expressed as
( )
2 / 3/ 21
0
1 exp3 1 4
bmL bmL b
MRC NamL Lm e e aASER a e d
mL
γγγγγ γ
γ γ
−∞ −− −
−
⎛ ⎞⎛ ⎞ ⎛ ⎞−≈ + −⎜ ⎟⎜ ⎟ ⎜ ⎟⎜ ⎟Γ⎝ ⎠ ⎝ ⎠⎝ ⎠
∫
( )
2 / 3/ 21
0
1 exp3 1 4
bmL bmL bmL Lm e e aa e d
mL
γγγγγ γ
γ γ
−∞ −− −
⎛ ⎞⎛ ⎞ ⎛ ⎞−≈ + −⎜ ⎟⎜ ⎟ ⎜ ⎟⎜ ⎟Γ⎝ ⎠ ⎝ ⎠⎝ ⎠
∫
1 21 1 13 2 3 4
mL mL mL
MRC Nab b a bASER amL mL mLγ γ γ− − −
−
⎡ ⎤⎛ ⎞ ⎛ ⎞ ⎛ ⎞≈ + + + − +⎢ ⎥⎜ ⎟ ⎜ ⎟ ⎜ ⎟⎝ ⎠ ⎝ ⎠ ⎝ ⎠⎢ ⎥⎣ ⎦
(4.17)
4.4.1 Analysis of Approximation
Table 4.1 - 4.6 shows the difference in dB for 4-QAM, 16-QAM and 64-QAM for different
values of m and L respectively. The figs.4.6– 4.15 shows the same information in the tables
but graphically. From the tables and figures we can see that Goldsmith approximation is good
to be used if we accept an error of 1 dB for 16-QAM only since its formula has one term only
but for other modulation indexes like 4 or 64-QAM we have maximum error of 2.8 and 3.2
dB respectively at ASER= 10-1 and single Nakagami-m parameter m = 0.5. Also we can see
72
that the approximation of n =1 is better than Chiani approximation although n =1 consists of
2 terms only. And Chiani is better only at high m and SNR.
4.5 Concluding Remarks
The essence of the above results is summarized below:
1. The system performance can be expressed by a simple finite series of elementary
functions that is the exponential.
2. The error decays rapidly with increasing values of n.
3. At fixed n, the error increases when the modulation index is increased.
4. At fixed n, m and L, the error decreases when SNR is increased.
17.2 17.4 17.6 17.8 18 18.2 18.4
10-1
SNR per Symbol (dB)
Pro
babi
lity
of S
ymbo
l erro
r
Exacy
n=1
Chiani
Goldsmith
m=1L=116QAM
Fig.4.6. Exact and Approx. of 16-QAM over Nakagami with m =1 and L =1
73
14.4 14.6 14.8 15 15.2 15.4 15.6
10-1
SNR per Symbol (dB)
Pro
babi
lity
of S
ymbo
l erro
r
Exacyn=1ChianiGoldsmith
m=2L=116QAM
Fig.4.7. Exact and Approx. of 16-QAM over Nakagami with m = 2 and L=1
13 13.2 13.4 13.6 13.8 14
10-1
SNR per Symbol (dB)
Pro
babi
lity
of S
ymbo
l erro
r
Exacyn=1ChianiGoldsmith
m=5L=116QAM
Fig.4.8. Exact and Approx. of 16-QAM over Nakagami with m = 5 and L =1
74
17.4 17.6 17.8 18 18.2 18.4
10-1
SNR per Symbol (dB)
Pro
babi
lity
of S
ymbo
l erro
r
Exacyn=1ChianiGoldsmith
m=0.5L=216QAM
Fig.4.9. Exact and Approx. of 16-QAM over Nakagami with m = 0.5 and L =2
31.4 31.6 31.8 32 32.2 32.4
10-4
SNR per Symbol (dB)
Pro
babi
lity
of S
ymbo
l erro
r
Exacyn=1ChianiGoldsmith
m=1L=216QAM
Fig.4.10. Exact and Approx. of 16-QAM over Nakagami with m =1 and L =2
75
21.4 21.5 21.6 21.7 21.8 21.9 22 22.1 22.2 22.3 22.4
10-3
SNR per Symbol (dB)
Pro
babi
lity
of S
ymbo
l erro
r
Exacyn=1ChianiGoldsmith
m=2L=216QAM
Fig.4.11. Exact and Approx. of 16-QAM over Nakagami with m = 2 and L = 2.
19.5 19.6 19.7 19.8 19.9 20 20.1 20.2
10-3
SNR per Symbol (dB)
Pro
babi
lity
of S
ymbo
l erro
r
Exacyn=1ChianiGoldsmith
m=2L=416QAM
Fig.4.12. Exact and Approx. of 16-QAM over Nakagami with m = 2 and L =4
76
20.7 20.8 20.9 21 21.1 21.2 21.3 21.4 21.5
10-3
SNR per Symbol (dB)
Pro
babi
lity
of S
ymbo
l erro
r
Exacyn=1ChianiGoldsmith
m=5L=116QAM
Fig.4.13. Exact and Approx. of 16-QAM over Nakagami with m = 5 and L=1
19 19.1 19.2 19.3 19.4 19.5 19.6 19.7
10-3
SNR per Symbol (dB)
Pro
babi
lity
of S
ymbo
l erro
r
Exacyn=1ChianiGoldsmith
m=5L=216QAM
Fig.4.14. Exact and Approx. of 16-QAM over Nakagami with m = 5 and L = 2
77
18.4 18.5 18.6 18.7 18.8 18.9 19
10-3
SNR per Symbol (dB)
Pro
babi
lity
of S
ymbo
l erro
r
Exacyn=1ChianiGoldsmith
m=5L=416QAM
Fig.4.15. Exact and Approx. of 16-QAM over Nakagami with m = 5 and L = 4
78
Table 4.1: Difference in dB for 4-QAM over Nakagami fading channel for m = 0.5, 1 and L = 1, 2 and 4.
ASER for 4-QAM 1.E-01 1.E-02 1.E-03 1.E-04 1.E-05 1.E-06
m L n Difference (dB) 0.5 1 Goldsmith -2.832 -2.6626 0.5 1 Chiani 0.18 0.1912 0.5 1 1 0.4769 0.4634 0.5 1 2 0.1158 0.1127 0.5 1 3 0.0512 0.0498 0.5 1 4 0.0287 0.028 0.5 1 10 0.0046 0.0045 0.5 2 Goldsmith -0.9839 -0.5891 -0.5588 -0.5558 0.5 2 Chiani 0.5375 0.5069 0.5035 0.5 2 1 0.1879 0.1685 0.1665 0.1663 0.5 2 2 0.0449 0.0402 0.0397 0.0397 0.5 2 3 0.0198 0.0177 0.0175 0.0175 0.5 2 4 0.0111 0.0099 0.0098 0.0098 0.5 2 10 0.0018 0.0016 0.0016 0.0016 0.5 4 Goldsmith -0.308 0.1344 0.1997 0.2166 0.2215 0.2231 0.5 4 Chiani 0.5923 0.4469 0.4088 0.3975 0.394 0.3929 0.5 4 1 0.1527 0.1276 0.1127 0.1077 0.106 0.1055 0.5 4 2 0.0436 0.03 0.0265 0.0255 0.0252 0.0251 0.5 4 3 0.0191 0.0132 0.0117 0.0112 0.0111 0.011 0.5 4 4 0.0107 0.0074 0.0065 0.0063 0.0062 0.0062 0.5 4 10 0.0017 0.0012 0.001 0.001 0.001 0.001 1 1 Goldsmith -0.9839 -0.5891 -0.5588 -0.5558 1 1 Chiani 0.5375 0.5069 0.5035 1 1 1 0.1879 0.1685 0.1665 0.1663 1 1 2 0.0449 0.0402 0.0397 0.0397 1 1 3 0.0198 0.0177 0.0175 0.0175 1 1 4 0.0111 0.0099 0.0098 0.0098 1 1 10 0.0018 0.0016 0.0016 0.0016 1 2 Goldsmith -0.308 0.1344 0.1997 0.2166 0.2215 0.2231 1 2 Chiani 0.5923 0.4469 0.4088 0.3975 0.394 0.3929 1 2 1 0.1527 0.1276 0.1127 0.1077 0.106 0.1055 1 2 2 0.0436 0.03 0.0265 0.0255 0.0252 0.0251 1 2 3 0.0191 0.0132 0.0117 0.0112 0.0111 0.011 1 2 4 0.0107 0.0074 0.0065 0.0063 0.0062 0.0062 1 2 10 0.0017 0.0012 0.001 0.001 0.001 0.001 1 4 Goldsmith -0.0431 0.3533 0.4033 0.4151 0.4188 0.4201 1 4 Chiani 0.5858 0.3435 0.258 0.2209 0.2028 0.1935 1 4 1 0.1694 0.1119 0.0553 0.0256 0.0102 0.002 1 4 2 0.0438 0.0207 0.0134 0.0105 0.0092 0.0085 1 4 3 0.0187 0.0092 0.006 0.0048 0.0042 0.0039 1 4 4 0.0105 0.0051 0.0034 0.0027 0.0024 0.0022 1 4 10 0.0017 0.0008 0.0005 0.0004 0.0004 0.0004
79
Table 4.2: Difference in dB for 4-QAM over Nakagami fading channel for m = 2, 5 and L = 1, 2 and 4.
ASER for 4-QAM 1.E-01 1.E-02 1.E-03 1.E-04 1.E-05 1.E-06
m L n Difference (dB) 2 1 Goldsmith -0.308 0.1344 0.1997 0.2166 0.2215 0.2231 2 1 Chiani 0.5923 0.4469 0.4088 0.3975 0.394 0.3929 2 1 1 0.1527 0.1276 0.1127 0.1077 0.106 0.1055 2 1 2 0.0436 0.03 0.0265 0.0255 0.0252 0.0251 2 1 3 0.0191 0.0132 0.0117 0.0112 0.0111 0.011 2 1 4 0.0107 0.0074 0.0065 0.0063 0.0062 0.0062 2 1 10 0.0017 0.0012 0.001 0.001 0.001 0.001 2 2 Goldsmith -0.0431 0.3533 0.4033 0.4151 0.4188 0.4201 2 2 Chiani 0.5858 0.3435 0.258 0.2209 0.2028 0.1935 2 2 1 0.1694 0.1119 0.0553 0.0256 0.0102 0.002 2 2 2 0.0438 0.0207 0.0134 0.0105 0.0092 0.0085 2 2 3 0.0187 0.0092 0.006 0.0048 0.0042 0.0039 2 2 4 0.0105 0.0051 0.0034 0.0027 0.0024 0.0022 2 2 10 0.0017 0.0008 0.0005 0.0004 0.0004 0.0004 2 4 Goldsmith 0.0711 0.4218 0.4429 0.4347 0.424 0.4149 2 4 Chiani 0.5743 0.2758 0.1642 0.1142 0.0892 0.0756 2 4 1 0.1876 0.0922 -0.0065 -0.0591 -0.0873 -0.1036 2 4 2 0.0418 0.0149 0.0059 0.0029 0.0018 0.0013 2 4 3 0.0186 0.0066 0.0028 0.0015 0.001 0.0007 2 4 4 0.0104 0.0037 0.0016 0.0009 0.0006 0.0004 2 4 10 0.0016 0.0006 0.0003 0.0001 0.0001 0.0001 5 1 Goldsmith 0.004 0.3841 0.4244 0.43 0.4297 0.4284 5 1 Chiani 0.5818 0.3175 0.2212 0.1777 0.1557 0.1435 5 1 1 0.1759 0.1056 0.0339 -0.005 -0.0262 -0.0382 5 1 2 0.0432 0.0185 0.0104 0.0072 0.0058 0.005 5 1 3 0.0187 0.0082 0.0047 0.0034 0.0027 0.0024 5 1 4 0.0105 0.0046 0.0027 0.0019 0.0016 0.0014 5 1 10 0.0017 0.0007 0.0004 0.0003 0.0003 0.0002 5 2 Goldsmith 0.0925 0.4324 0.4457 0.4316 0.4162 0.4033 5 2 Chiani 0.5713 0.2613 0.1452 0.0945 0.0705 0.0583 5 2 1 0.1921 0.0864 -0.022 -0.0781 -0.1071 -0.1233 5 2 2 0.0412 0.0137 0.0044 0.0017 0.001 0.0008 5 2 3 0.0187 0.006 0.0022 0.001 0.0006 0.0003 5 2 4 0.0104 0.0034 0.0013 0.0006 0.0003 0.0002 5 2 10 0.0016 0.0005 0.0002 0.0001 0.0001 0 5 4 Goldsmith 0.1338 0.4501 0.4463 0.4187 0.3923 0.3703 5 4 Chiani 0.5648 0.2313 0.1082 0.06 0.0425 0.0379 5 4 1 0.2019 0.0723 -0.0555 -0.1145 -0.1409 -0.1531 5 4 2 0.0399 0.0112 0.0017 0.0004 0.0007 0.0006 5 4 3 0.0187 0.0048 0.0012 0.0003 0.0001 0 5 4 4 0.0103 0.0027 0.0007 0.0002 0.0001 0 5 4 10 0.0016 0.0004 0.0001 0 0 0
80
Table 4.3: Difference in dB for 16-QAM over Nakagami fading channel for
m = 0.5, 1 and L = 1, 2 and 4. ASER for 16-QAM 1.E-01 1.E-02 1.E-03 1.E-04 1.E-05 1.E-06
m L n Difference (dB) 0.5 1 Goldsmith 0.6098 0.5 1 Chiani 0.6204 0.5 1 1 0.4689 0.5 1 2 0.114 0.5 1 3 0.0504 0.5 1 4 0.0283 0.5 1 10 0.0045 0.5 2 Goldsmith 0.9194 0.9171 0.9166 0.5 2 Chiani 0.6403 0.5954 0.5912 0.5 2 1 0.2922 0.2682 0.2658 0.5 2 2 0.0693 0.0634 0.0629 0.5 2 3 0.0305 0.0279 0.0277 0.5 2 4 0.0171 0.0156 0.0155 0.5 2 10 0.0027 0.0025 0.0025 0.5 4 Goldsmith 0.9815 0.92 0.9005 0.8944 0.8925 0.5 4 Chiani 0.576 0.4437 0.4103 0.4004 0.3973 0.5 4 1 0.2626 0.1954 0.1733 0.1664 0.1642 0.5 4 2 0.0626 0.0449 0.0403 0.039 0.0386 0.5 4 3 0.0274 0.0197 0.0177 0.0171 0.017 0.5 4 4 0.0153 0.011 0.0099 0.0096 0.0095 0.5 4 10 0.0024 0.0018 0.0016 0.0015 0.0015 1 1 Goldsmith 0.9194 0.9171 0.9166 1 1 Chiani 0.6403 0.5954 0.5912 1 1 1 0.2922 0.2682 0.2658 1 1 2 0.0693 0.0634 0.0629 1 1 3 0.0305 0.0279 0.0277 1 1 4 0.0171 0.0156 0.0155 1 1 10 0.0027 0.0025 0.0025 1 2 Goldsmith 0.9815 0.92 0.9005 0.8944 0.8925 1 2 Chiani 0.576 0.4437 0.4103 0.4004 0.3973 1 2 1 0.2626 0.1954 0.1733 0.1664 0.1642 1 2 2 0.0626 0.0449 0.0403 0.039 0.0386 1 2 3 0.0274 0.0197 0.0177 0.0171 0.017 1 2 4 0.0153 0.011 0.0099 0.0096 0.0095 1 2 10 0.0024 0.0018 0.0016 0.0015 0.0015 1 4 Goldsmith 0.987 0.8618 0.8027 0.7728 0.757 0.7484 1 4 Chiani 0.5241 0.3187 0.246 0.2137 0.1979 0.1897 1 4 1 0.2658 0.1368 0.0705 0.0386 0.0224 0.0138 1 4 2 0.0582 0.0288 0.0193 0.0154 0.0136 0.0127 1 4 3 0.0252 0.0127 0.0086 0.007 0.0062 0.0058 1 4 4 0.0141 0.0071 0.0049 0.004 0.0035 0.0033 1 4 10 0.0022 0.0011 0.0008 0.0006 0.0006 0.0005
81
Table 4.4: Difference in dB for 16-QAM over Nakagami fading channel for m = 2, 5 and L = 1, 2 and 4.
ASER for 16-QAM 1.E-01 1.E-02 1.E-03 1.E-04 1.E-05 1.E-06
m L n Difference (dB) 2 1 Goldsmith 0.9815 0.92 0.9005 0.8944 0.8925 2 1 Chiani 0.576 0.4437 0.4103 0.4004 0.3973 2 1 1 0.2626 0.1954 0.1733 0.1664 0.1642 2 1 2 0.0626 0.0449 0.0403 0.039 0.0386 2 1 3 0.0274 0.0197 0.0177 0.0171 0.017 2 1 4 0.0153 0.011 0.0099 0.0096 0.0095 2 1 10 0.0024 0.0018 0.0016 0.0015 0.0015 2 2 Goldsmith 0.987 0.8618 0.8027 0.7728 0.757 0.7484 2 2 Chiani 0.5241 0.3187 0.246 0.2137 0.1979 0.1897 2 2 1 0.2658 0.1368 0.0705 0.0386 0.0224 0.0138 2 2 2 0.0582 0.0288 0.0193 0.0154 0.0136 0.0127 2 2 3 0.0252 0.0127 0.0086 0.007 0.0062 0.0058 2 2 4 0.0141 0.0071 0.0049 0.004 0.0035 0.0033 2 2 10 0.0022 0.0011 0.0008 0.0006 0.0006 0.0005 2 4 Goldsmith 0.984 0.8163 0.7234 0.667 0.6305 0.6057 2 4 Chiani 0.4935 0.2452 0.1507 0.1075 0.0855 0.0734 2 4 1 0.2722 0.0907 -0.0117 -0.0622 -0.089 -0.1044 2 4 2 0.0548 0.0188 0.0077 0.004 0.0025 0.0019 2 4 3 0.0239 0.0084 0.0037 0.0021 0.0013 0.001 2 4 4 0.0133 0.0047 0.0022 0.0012 0.0008 0.0006 2 4 10 0.0021 0.0008 0.0004 0.0002 0.0001 0.0001 5 1 Goldsmith 0.9862 0.8449 0.7735 0.7344 0.7118 0.6983 5 1 Chiani 0.5122 0.2901 0.2082 0.1705 0.1511 0.1402 5 1 1 0.268 0.1202 0.0401 0.0001 -0.0212 -0.0332 5 1 2 0.0569 0.025 0.0145 0.0104 0.0084 0.0074 5 1 3 0.0247 0.0111 0.0066 0.0048 0.004 0.0035 5 1 4 0.0138 0.0062 0.0038 0.0028 0.0023 0.002 5 1 10 0.0022 0.001 0.0006 0.0004 0.0004 0.0003 5 2 Goldsmith 0.983 0.8061 0.7051 0.642 0.5998 0.5704 5 2 Chiani 0.487 0.2299 0.1318 0.0882 0.0673 0.0566 5 2 1 0.2739 0.0794 -0.0302 -0.0828 -0.1098 -0.1248 5 2 2 0.054 0.0167 0.0056 0.0024 0.0014 0.001 5 2 3 0.0237 0.0075 0.0029 0.0014 0.0008 0.0005 5 2 4 0.0131 0.0042 0.0017 0.0008 0.0005 0.0003 5 2 10 0.0021 0.0007 0.0003 0.0001 0.0001 0.0001 5 4 Goldsmith 0.9806 0.7841 0.666 0.5886 0.5343 0.4945 5 4 Chiani 0.4735 0.1984 0.0954 0.0552 0.041 0.0378 5 4 1 0.2777 0.0545 -0.0679 -0.1204 -0.1436 -0.1543 5 4 2 0.0526 0.0122 0.002 0.0006 0.0008 0.0006 5 4 3 0.0231 0.0056 0.0014 0.0004 0.0001 0.0001 5 4 4 0.0128 0.0032 0.0009 0.0003 0.0001 0 5 4 10 0.002 0.0005 0.0001 0 0 0
82
Table 4.5: Difference in dB for 64-QAM over Nakagami fading channel for m = 0.5, 1 and L = 1, 2 and 4.
ASER for 64-QAM 1.E-01 1.E-02 1.E-03 1.E-04 1.E-05 1.E-06
m L n Difference (dB) 0.5 1 Goldsmith 3.2381 0.5 1 Chiani 0.8592 0.5 1 1 0.4672 0.5 1 2 0.1136 0.5 1 3 0.0502 0.5 1 4 0.0282 0.5 1 10 0.0045 0.5 2 Goldsmith 2.2647 2.1368 0.5 2 Chiani 0.6923 0.6425 0.5 2 1 0.3499 0.3231 0.5 2 2 0.0825 0.076 0.5 2 3 0.0363 0.0334 0.5 2 4 0.0203 0.0187 0.5 2 10 0.0032 0.003 0.5 4 Goldsmith 1.815 1.5504 1.4865 1.4677 0.5 4 Chiani 0.5718 0.4435 0.4115 0.402 0.5 4 1 0.3152 0.2306 0.2051 0.1973 0.5 4 2 0.0723 0.0525 0.0475 0.046 0.5 4 3 0.0316 0.023 0.0208 0.0202 0.5 4 4 0.0176 0.0129 0.0117 0.0113 0.5 4 10 0.0028 0.002 0.0019 0.0018 1 1 Goldsmith 2.2647 2.1368 1 1 Chiani 0.6923 0.6425 1 1 1 0.3499 0.3231 1 1 2 0.0825 0.076 1 1 3 0.0363 0.0334 1 1 4 0.0203 0.0187 1 1 10 0.0032 0.003 1 2 Goldsmith 1.815 1.5504 1.4865 1.4677 1 2 Chiani 0.5718 0.4435 0.4115 0.402 1 2 1 0.3152 0.2306 0.2051 0.1973 1 2 2 0.0723 0.0525 0.0475 0.046 1 2 3 0.0316 0.023 0.0208 0.0202 1 2 4 0.0176 0.0129 0.0117 0.0113 1 2 10 0.0028 0.002 0.0019 0.0018 1 4 Goldsmith 1.6123 1.2598 1.1348 1.0771 1.0476 1.0319 1 4 Chiani 0.5022 0.3097 0.2413 0.2109 0.1957 0.1879 1 4 1 0.3052 0.1493 0.0786 0.0454 0.0287 0.0199 1 4 2 0.0653 0.0328 0.0222 0.0178 0.0158 0.0148 1 4 3 0.0283 0.0144 0.0099 0.0081 0.0072 0.0068 1 4 4 0.0158 0.0081 0.0056 0.0046 0.0041 0.0039 1 4 10 0.0025 0.0013 0.0009 0.0007 0.0007 0.0006
83
Table 4.6: Difference in dB for 64-QAM over Nakagami fading channel for m = 2, 5 and L = 1, 2 and 4.
ASER for 64-QAM 1.E-01 1.E-02 1.E-03 1.E-04 1.E-05 1.E-06
m L n Difference (dB) 2 1 Goldsmith 1.815 1.5504 1.4865 1.4677 2 1 Chiani 0.5718 0.4435 0.4115 0.402 2 1 1 0.3152 0.2306 0.2051 0.1973 2 1 2 0.0723 0.0525 0.0475 0.046 2 1 3 0.0316 0.023 0.0208 0.0202 2 1 4 0.0176 0.0129 0.0117 0.0113 2 1 10 0.0028 0.002 0.0019 0.0018 2 2 Goldsmith 1.6123 1.2598 1.1348 1.0771 1.0476 1.0319 2 2 Chiani 0.5022 0.3097 0.2413 0.2109 0.1957 0.1879 2 2 1 0.3052 0.1493 0.0786 0.0454 0.0287 0.0199 2 2 2 0.0653 0.0328 0.0222 0.0178 0.0158 0.0148 2 2 3 0.0283 0.0144 0.0099 0.0081 0.0072 0.0068 2 2 4 0.0158 0.0081 0.0056 0.0046 0.0041 0.0039 2 2 10 0.0025 0.0013 0.0009 0.0007 0.0007 0.0006 2 4 Goldsmith 1.5185 1.1197 0.9542 0.8633 0.8071 0.7702 2 4 Chiani 0.4652 0.2345 0.1459 0.105 0.0841 0.0726 2 4 1 0.302 0.0908 -0.0131 -0.0631 -0.0895 -0.1046 2 4 2 0.0608 0.0206 0.0085 0.0045 0.0029 0.0021 2 4 3 0.0263 0.0093 0.0042 0.0023 0.0015 0.0011 2 4 4 0.0146 0.0052 0.0024 0.0014 0.0009 0.0007 2 4 10 0.0023 0.0008 0.0004 0.0002 0.0002 0.0001 5 1 Goldsmith 1.5742 1.2033 1.0628 0.9928 0.954 0.9314 5 1 Chiani 0.4877 0.2803 0.2033 0.1677 0.1491 0.1388 5 1 1 0.3038 0.1277 0.0439 0.0031 -0.0184 -0.0305 5 1 2 0.0636 0.0281 0.0165 0.0119 0.0097 0.0086 5 1 3 0.0275 0.0124 0.0076 0.0056 0.0046 0.0041 5 1 4 0.0153 0.007 0.0043 0.0032 0.0026 0.0023 5 1 10 0.0024 0.0011 0.0007 0.0005 0.0004 0.0004 5 2 Goldsmith 1.5005 1.0923 0.918 0.8195 0.7568 0.7142 5 2 Chiani 0.4576 0.219 0.1271 0.086 0.0661 0.056 5 2 1 0.3015 0.0771 -0.0328 -0.0843 -0.1106 -0.1253 5 2 2 0.0598 0.018 0.0061 0.0027 0.0016 0.0011 5 2 3 0.0258 0.0082 0.0032 0.0015 0.0009 0.0006 5 2 4 0.0144 0.0046 0.0019 0.0009 0.0005 0.0003 5 2 10 0.0023 0.0007 0.0003 0.0002 0.0001 0.0001 5 4 Goldsmith 1.4649 1.0377 0.8456 0.7316 0.6556 0.6013 5 4 Chiani 0.4421 0.1871 0.0911 0.0536 0.0406 0.0377 5 4 1 0.3007 0.0478 -0.0721 -0.1224 -0.1446 -0.1548 5 4 2 0.0577 0.0127 0.0021 0.0008 0.0008 0.0006 5 4 3 0.0249 0.006 0.0015 0.0004 0.0002 0.0001 5 4 4 0.0138 0.0034 0.0009 0.0003 0.0001 0 5 4 10 0.0022 0.0006 0.0002 0.0001 0 0
84
CHAPTER 5
CONCLUSIONS AND FUTURE WORK 5.1 Conclusions
In this thesis, we obtained simple closed form expression to determine the performance of M-
QAM transmitted over slow, flat, identically independently distributed (i.i.d) fading channels
and using space diversity in terms of ASER. Three types of fading channels are considered:
Rayleigh Ricean, and Nakagami. These simple and efficient ASER formulas make it possible
for the first time to study, analyze and discuss the parameters of various constellations of
square M-QAM, diversity order and fading parameters precisely and easily. MRC was used
as combining technique to overcome the fading channel. The main advantage of these
expressions, that they show the relation between the square M-QAM modulation index (M)
and the diversity order (L) and the fading parameter (K for Ricean or m for Nakagami-m)
versus SNR in one simple formula.
We used an expression for the SER of M-QAM modulation over a Gaussian channel
(AWGN). This expression is then used to obtain the ASER of square M-QAM using MRC
over both Ricean and Nakagami fading channels. For Rayleigh, it was obtained as a special
case of both Ricean (K=0) and Nakagami-m (m=1). The obtained expressions are in the form
of sum of exponentials or power series where the number of terms can be determined
according to the required accuracy. Theoretically, to get the exact solution the series must be
infinite. But because the series converge rapidly, 10 terms are enough to get an error less
than 0.005 dB in the worst case which is at low order of modulation index (M), diversity, and
low fading parameter (K for Ricean or m for Nakagami-m). The minimum number of terms
to approximate the ASER is two terms, so single term is also investigated which is available
in the literatures. The error for different number of terms are investigated and tabulated. The
error was limited to one dB for n=1 which is 2 terms for all M, L, and the fading parameter
(K or m) but for single term which we call Goldsmith the difference in dB between the exact
and approximate is increased with M and decreased with L, and the fading parameter.
85
5.2 Future Work
Our work could be extended for different types of combining techniques. Also in this thesis
we have assumed that the fading is i.i.d but this scenario no longer exists in real life if the
diversity is used in the mobile terminals, so the work presented here can be extended to the
case of correlated fading. Another possible extension of our work would be to obtain exact
ASER of M-QAM over slow, flat, fading when the fading channels are non-identical.
86
REFERENCES
[1] M. Chiani, D. Dardari, and M. K. Simon, “New Exponential Bounds and Approximation
for the Computation of Error Probability in Fading Channels,” IEEE Transactions on
Wireless Communications, Vol. 2, No. 4, July 2003.
[2] M. K. Simon and M. S. Alouini, Digital Communications over Fading Channels, John
Wiley & Sons, Inc., New York, 2000.
[3] Swokowski, Olinick, and Pence, Calculus, PWS, Boston, 6th Ed, 1994.
[4] Burden and Faires, Numerical Analysis, Brooks/Cole, 6th Ed, 1997.
[5] J. G. Proakis, Digital Communications, McGraw-Hill, Inc., New York, 2001.
[6] A. Goldsmith and S. G. Chua, “Variable-rate variable-power M-QAM for fading
channels,” IEEE Transactions on Communications, vol. 45, pp. 1218–1230, Oct. 1997.
[7] H. Zhang and T. A. Gulliver, “Error probability for maximum ratio combining
multichannel reception of M-ary coherent systems over flat Ricean fading channels,”
Proc. of IEEE Wireless Communications and Networking Conf., WCNC 2004, vol. 1,
pp. 306 – 310, March 2004.
[8] S. Seo, C. Lee, and S. Kang, “Exact performance analysis of M-ary QAM with MRC
diversity in Ricean fading channels,” Electronics Letters, vol.40, no. 8, pp. 485-486, April
2004.
[9] M. S. Patterh, T. S. Kamal, and B. S. Sohi, “Performance of coherent square MQAM with
Lth order diversity in Ricean fading environment,” Proc. of 54th IEEE Vehicular Tech.
Conf., VTC 2001, vol.1, pp. 141- 143, 2001.
[10] I. A. Falujah and V. K. Prabhu, “Performance analysis of MQAM with MRC over
Nakagami-m fading channels,” Electronics Letters, vol.42, no.4, pp. 231- 233, Feb.
2006.
[11] I. A. Falujah and V. K. Prabhu, “Performance analysis of MQAM with MRC over
Nakagami-m fading channels,” Proc. of Wireless Communications and Networking
Conference, WCNC 2006, vol.3, pp. 1332- 1337, April 2006.
[12] M. S. Patterh and T. S. Kamal, “Performance of coherent square M-QAM with Lth order
diversity in Nakagami-m fading environment,” Proc. of 52nd IEEE Vehicular Tech.
Conf., VTC 2000, vol.6, pp.2849-2853, 2000.
[13] A. Annamalai, C. Tellambura, and V. K. Bhargava, "Exact evaluation of maximal-ratio
87
and equal-gain diversity receivers for M-ary QAM on Nakagami fading channels," IEEE
Transactions on Communications, vol.47, no.9, pp.1335-1344, Sep. 1999
[14] M. Nakagami, "The m-distribution - a general formula for intensity distribution of rapid
fading," Statistical Methods in Radio Wave Propagation, Pergamon Press, Oxford, U.K.,
pp. 3-36, 1960.
[15] Braun and Dcrsch, "A physical mobile radio channel," IEEE Transactions on Vehicular
Technology, vol. 40, pp. 472 - 482, May 1991.
[16] S. Lin, "Statistical Behaviour of a fading signal," Bell System Technical Journal, vol. 50,
pp. 3211-3271, Dec, 1971.
[17] W. C. Jakes, Microwave Mobile Communications, John Wiley & Sons. New York, NY,
1974.
[18] R. Bultitude and G. Bedai, "Propagation characteristics on microcellular urban mobile
radio channels at 910 mhz," IEEE Journal on Selected Areas in Communications, vol. 7,
pp. 31-39, Jan. 1989.
[19] R. Steel. ''The cellular environment of lightweight handheld portables," IEEE
Communications Magazine, pp. 20-29, July 1989.
[20] G. D. Gibson, The Mobile Communications Handbook, CRC and IEEE Press. 1996.
[21] J. D. Parsons, The Mobile Radio Propagation Channel, Pentech Press, London. 1992.
[22] U. Charash, "Reception through Nakagami fading multipath channels with random
delays," IEEE Transactions on Communications, vol. 27, pp. 657-670, April 1979.
[23] H. Suzuki, "A statistical model for urban multipath propagation," IEEE Transactions on
Communications, vol. 25, pp. 673-680, July 1977.
[24] T. Aulin, "Characteristics of a digital mobile radio channel," IEEE Transactions on
Vehicular Technology, vol. 30, pp. 45-53, May 1981.
[25] G. L- Stüber, Principles of Mobile Communications, Kluwer Academic Publishers,
Boston, MA, 1996.
[26] J. Parsons, M. Henze, P. Ratliff, and M. Withers, "Diversity techniques for mobile radio
reception,'' IEEE Transactions on Vehicular Technology, vol. 25, pp. 75-85, Aug. 1976.
[27] W. C. Y. Lee, Mobile Communications Engineering, McGraw-Hill, New York, 1982.
[28] M. Schwartz, W. R. Bennett, and S. Stein, Communication Systems and Techniques,
McGraw-Hill, New York, 1966.
[29] D. Brennan, "Linear diversity combining techniques," Proceedings of IRE, vol. 47, pp.
1075-1102, June 1959.
88
[30] C. W. Helstrom, Statistical Theory of Signal Detection, Pergamon Press, London,
England, 1960.
[31] E. K. Al-Hussaini and A. A. M. Al-Bassiouni, "Performance of MRC diversity systems
for the detection of signals with Nakagami fading," IEEE Transactions on
Communications, vol. 33, pp. 1315-1319, Dec. 1985.
[32] N. C. Beaulieu, "An infinite series for the computation of the complementary probability
distribution function of a sum of independent random variables and its application to the
sum of Rayleigh random variables," IEEE Transactions on Communications, vol. 38, pp.
1463-1474, Sept. 1990.
[33] A. A. Abu-Dayya and N. C. Beaulieu, "Microdiversity on Rician fading channels," IEEE
Transactions on Communications, vol. 42, pp. 2258-2267, June 1994.
[34] N. C. Beaulieu and A. A. Abu-Dayya, "Analysis of equal gain diversity on Nakagami
fading channels," IEEE Transactions on Communications, vol. 39, pp. 225-234, Feb.
1991.
[35] I. S. Gradshteyn and I. M. Ryzhik. Table of Integrals, Series, and Products, Academic
Press, San Diego, CA, 5th edition, 1994.