INTL JOURNAL OF ELECTRONICS AND TELECOMMUNICATIONS, 2010, VOL. 56, NO. 4, PP.

339–344
Manuscript received September 16, 2010; revised November 2010. DOI: 10.2478/v10177-010-0044-x

Overview of Fading Channel Modeling

Krystyna M. Noga and Beata Pałczyńska

Abstract—VisSim and LabVIEW based approaches are pro- phases combine to give a resultant signal that fluctuates in
posed and implemented to demonstrate fading in the commu- time and space. The mobile radio channel is characterized by
nication systems. The introduction to fading, models for flat two types of fading. The short-term fluctuation in the signal
fading like Rayleigh, Weibull, Nakagami-m and the results of
simulations are presented. envelope, caused by the local multipath, is called a small-scale
fading. Such a situation is observed over distances of about
Keywords—Communication channel, random variables, simu- half of a wavelength. The term small-scale fading denotes
lation, Nakagami-m fading, Rayleigh fading, Weibull fading.
rapid variations of the envelope of a received signal. The
second type of fading is called a large-scale fading. The term
I. I NTRODUCTION large-scale fading denotes the variations of the mean envelope
or mean power of a received signal. It is assumed to be a
T HE CHANNEL CHARACTERIZATION in mobile com-
munication systems is an important and fast growing
part of wireless communication. It plays a main role in the
slow process and is usually modeled as having lognormal
statistics. For example, the slow variation of the mean signal
transmission of information through a propagation medium level for outdoor and indoor environments can be described
from the transmitter to the receiver. The propagation path by a lognormal statistics. This paper will be limited to the flat
between the transmitter and receiver can vary from simple multipath fading, as multi-path in the radio channel creates
line-one sight to one that is severely affected by buildings, small-scale fading effects. Usually the fading envelope of
mountains and trees. Radio channels are extremely random the signal is modeled by a Rice, Rayleigh and Nakagami-
and do not offer easy analysis. Modeling radio channels has m distribution [1]–[5]. The Rice and Rayleigh distributions
been one of the most difficult parts of mobile radio system describe multipath effects and Nakagami-m can replace them
design. It is typically done using statistical analysis. Without both. Analyses based on Rayleigh or Rice fading is sufficient
knowledge of channel models, a wireless system could never to model the nature of many different mobile channels. But a
be developed. This knowledge is very important for designing lot of recent publications, i.e. [3], [4], [6], [7], have suggested
radio communication systems. The mobile radio propagation the use of Nakagami–m, Weibull fading models to provide a
channel introduces fundamental limitations to the performance generalized description of fading in wireless systems.
of any wireless communication systems. In digital communi-
cation theory the frequently assumed model for a transmission III. G ENERATION OF U NCORRELATED FADING
channel is the additive white Gaussian noise (AWGN) channel.
However, for many communication systems the AWGN isn’t The random fluctuation in the received signal due to fading
an adequate model. A more precise and complicated channel can be modeled by treating C[τ (t), t] as a random process in
model should be used. In practice, the fading channel model t, where τ is the propagation delay. The multipath channel can
is frequently applied. Typical examples of such channels are be described by a time-varying, complex, lowpass-equivalent
the Rayleigh, Nakagami-m , Rice channels [1]–[4]. They are impulse response C[τ (t), t]. Statistics of the received signal
combined by different researchers in order to form a model in a mobile radio environment are obtained by using the
for describing a particular channel. scattering model and the assumption of large number of
In this paper, VisSim and LabVIEW based approaches are randomly phased components. Since the components of the
proposed and implemented to demonstrate the concept of multipath signal arise from a large number of reflections or
fading. Models for Rayleigh, Weibull, Nakagami-m fading are scattering from rough or surface, according to the central limit
presented and simulated. theorem, C[τ (t), t] can be modeled as a complex Gaussian
realization. The probability density function of the real and
II. FADING IN A W IRELESS E NVIRONMENT imaginary parts, i. e. the in-phase and quadrature components,
are Gaussian. If C[τ (t), t] has a mean equal to zero, then the
Radio waves propagate from a transmitting antenna, and
envelope R(t) = |C[τ (t), t]|has a Rayleigh probability density
travel through the space, undergoing reflections, absorption,
function (pdf) [3], [4].
refraction, scattering and diffraction. The transmitted signal
arrives at the receiver via several paths with different time r  r2 
delays creating a multi-path situation. At the receiver, these pr (r) = 2 exp − 2 , r ≥ 0 (1)
σ 2σ
multipath waves with randomly distributed envelopes and
where: 2σ 2 = E(R2 ) and E(∗ ) denotes statistical averaging.
K. M. Noga is with the Department of Ship Automation, Gdynia There are a number of methods used to generate different
Maritime University, 81–225 Gdynia, Morska 81–87, Poland (e-mail: (non-uniform) variables. Each method is applicable only to a
jagat@am.gdynia.pl).
B. Pałczyńska is with the Department of Marine Telecommunications, subset of distribution. Also, for a particular distribution, one
Gdynia Maritime University, Poland (e-mail: palbeata@am.gdynia.pl). method may be more efficient than the others [8].
340 K. M. NOGA, B. PAŁCZYŃSKA

a) b)

Fig. 2. The generator of the Weibull envelope.

Fig. 1. The Rayleigh fading process – the envelope (a) and its histogram (b). It should be emphasized that in VisSim and LabView
environments used for simulation, Gaussian signal generators
are available.
The√ generation of Rayleigh fading envelope Figure 1 shows the Rayleigh envelope waveform and its
R = X 2 + Y 2 follows from the basic fact that the histogram. The model of fading is simulated using VisSim
envelope of a Gaussian process with independent quadrature environment, created by the Regents of the University of
Y and in-phase X components, with mean value µ = 0 and California.
variance σ2 = 1, has a Rayleigh distribution. Moreover, the Another distribution used to model fading in a multipath
phase θ = arctg(Y /X) has a uniform pdf. This fact can environment is the Weibull distribution [3], [4], [6], [8]. This
be used to generate two samples of a Gaussian variable by distribution seems to exhibit good fit to the experimental
transforming a pair of Rayleigh and uniform variables. If U1 fading channel measurements, for indoor and outdoor envi-
and U2 are two independent variables uniformly distributed ronments. The probability density function of the envelope R
in the unit interval, i.e. U1 := rnd(1) and U2 := rnd(1) is given by
where rnd(x) returns a uniformly distributed random number
between 0 and x, then αrα−1  r α
pr (r) = α
exp − ; r ≥ 0, α ≥ 0,
a>0
p a a
X = µ + σ −2 ln u1 cos(2πu2 ) (5)
p where a is a scale parameter and α is a shape parameter.
Y = µ + σ −2 ln u1 sin(2πu2 ) (2) To generate Weibull variables, the inverse transformation
technique can be used [8]. We should generate the uniform
are independent Gaussian random variables. This algorithm random variable U [0, 1] and then return a(ln u)1/α as Weibull
is known as the Box-Muller method [1], [8]. There is some variable with parameters a, α.
concern that if this method is used with mean value µ 6= 0
The Weibull envelope can also be expressed as a function
from an Linear-Congruential Generators (LCG), the resulting
of the in-phase and quadrature components, so that
X, Y may be correlated [8].
Another method for generating random variables is the R = (X 2 + Y 2 )1/α (6)
convolution technique. This technique can be used if the
random variable X can be expressed as a sum of n random where X and Y are independent Gaussian processes with
variables Zi mean value equal to zero and E(X 2 ) = E(Y 2 ) = σ 2 . In
Xn
X= Zi (3)
i=1
a) b)
If X is a sum of two random variables, then the pdf
of X can be obtained analytically by a convolution of the
pdf’s of Z1 and Z2 . This is why the technique is called
convolution, although no convolution is required in random-
number generation. Using this method and, taking into account
that the sum of large number of uniform U [0, 1] variables has
a normal distribution, for normal variable with mean value µ
and variance σ 2 it can be written [3]
n
!
X n
ui −
2
X = µ + σ i=1p n (4)
12

This method is also known as the sum of 12 [1]. The

value n = 12 represent some compromise between speed and
accuracy. It should be noted that whereas a Gaussian random
variable has value ranging from −∞ to +∞, formula (4) Fig. 3. The envelope and the histogram of Weibull fading process for σ2 =
produced value of normal variable in the interval [−6, +6]. 0.5 and a) α = 1, b) α = 3.
OVERVIEW OF FADING CHANNEL MODELING 341

a) b)
where Xi and Yi are independent Gaussian processes with the
mean value equal to zero and E(Xi2 ) = E(Yi2 ) = σ 2 . Figure
4 shows the Nakagami-m fading envelope and its histogram.
Another model of Rayleigh fading are Jake’s and modi-
fied Jake’s models [3], [11], that are based on summing of
sinusoids. The Jake’s model produces a signal which has
the Doppler power spectrum. It is a deterministic method,
which simulates time-correlated Rayleigh fading waveforms.
Fig. 4. The Nakagami–m fading envelope (a) and its histogram (b) for
The model assumes that a signal propagates through n equal-
m = 2. strength paths and arrives at a moving receiver with uniformly
distributed arrival angles. Therefore this signal experiences a
Doppler shift.
this case the pdf of the envelope R can be written as The fading channel simulator, created using this model, was
built in LabVIEW environment. Its front panel is shown in
αrα−1  rα 
Figure 5. It contains several functional blocks for specification
pr (r) = exp − ; r ≥ 0, α≥0 (7)
r0 r0 of simulation, determination of fading channel model and
visualization of the results (waveform graphs, output data).
where r0 = E(Rα ) = |E(R2 )/Γ(1 + 2/α)|α/2 and Γ(m)
The presented Virtual Instrument (VI) generates the Rayleigh
is a gamma function [9]. In a special case, when α = 2
and Weibull distributions fading profiles. The user specifies
equation (7) describes the Rayleigh distribution and for α = 1,
input data like fading variance and shape parameter for Weibull
it becomes an exponential distribution. If α = 3.602, then the
fading. The VI applies a user-defined profile to fade the com-
Weibull distribution is close to a normal [8]. Figure 2 shows
plex baseband modulated waveform. The simulator performs
a block which generates the Weibull fading envelope. Figure
QAM modulation. The user has to specify the QAM system
3 shows a waveform and a histogram of the Weibull envelope
parameters, i.e. the M -ary number of modulation, which is the
respectively, using VisSim environment. Another generator of
number of distinct states that represent symbols in the complex
Weibull distributed random variables, in Mathcad environment,
baseband modulated waveform, number of message symbols
has been presented in [10].
and type of pulse shaping filters applied by the QAM. Simula-
One of the most versatile fading models is described by
tions can be carried out for determined values of a symbol rate
the Nakagami-m distribution. It has a greater flexibility and
and the specific Doppler spread. Results of the simulation are
accuracy in matching some experimental data than Rayleigh,
shown in form of waveform graphs. The VIs, contained in the
log-normal, Weibull or Rician distributions. The Nakagami-m
area of graphs, show a fading envelope, histograms of a fading
fading distribution was first used for modeling ionospheric
envelope and a instantaneous phase, In-phase and Quadrature-
and troposheric fast fading channels and has been widely
phase (IQ) profile, waveforms of real and imaginary parts of
adopted for multipath modeling in wireless communications.
faded signal, constellations of transmitter and receiver, power
The Nakagami-m distribution gave the best fit to some urban
spectral densities of transmitter, receiver and faded profile,
multi-path data. This distribution is a purely empirical model,
instantaneous power for both transmitter and receiver.
not phenomenological. The probability density function of
Nakagami-m distributed random variables R is [4]
IV. G ENERATION OF C ORRELATED R ANDOM VARIABLE
2  m m 2m−1  m 
pr (r) = r exp − r2 (8) The time-varying nature of the channel C[τ (t), t] is modeled
Γ(m) Ω Ω
as a wide-sense stationary random process in t with an
where Ω = E(R2 ) and the parameter m denotes the reciprocal autocorrelation function [1]
value of the variance of R2 normalized to Ω2 , i.e.
Rc (τ1 , τ2 , ∆t) = E⌊C ∗ (τ1 , t)C(τ2 , t + ∆t)⌋ (10)
2 2 2
m = Ω /E(R − Ω ≥ 1/2
where C ∗ (τ1 , t) denotes the conjugate value.
2 A correlated Gaussian variable can be generated from an
This is called central x distribution with m degrees of
freedom. The Nakagami-m distribution is often used to model uncorrelated Gaussian variable using a linear transformation or
multipath fading because it can model fading conditions that filtering [1], [12], [13]. The coefficient of the linear transfor-
are either more or less severe than Rayleigh fading. When mation can be obtained from the specified correlation function
m = 1, the Nakagami-m distribution becomes the Rayleigh of the output.
distribution. When m = 0.5, it becomes a one-sided Gaussian The correlated complex Gaussian variables W with a cor-
distribution, and when m = ∞, it becomes a no fading responding normalized covariance matrix given by
channel. For integer m, we can write the envelope of received  
signal as 1 ρg1,2 . . . ρg1,N
v  ρg2,1 1 . . . ρg2,N 
um
Kg =  (11)
 
uX .. 
R=t Xi2 + Yi2 (9)  . ... 
i=1 ρgN,1 ρgN,2 ... 1
342 K. M. NOGA, B. PAŁCZYŃSKA

Fig. 5. The simulator of fading channel applied a QAM signal – the front panel of virtual instrument in LabVIEW.

can be generated from uncorrelated complex Gaussian vari- given by [12], [13]
ables V by employing Cholesky decomposition.  √ 
2 |ρgi,j | π π
The cross-correlation coefficient ρgi,j between the complex (1 + |ρgi,j |)Ei 1+|ρgi,j | , 2 − 2
Gaussian samples vi and vj is given by ρri,j = (14)
2 − π2
u21 + u22 where Ei (η, π2 ) denotes the complete elliptic integral of the
ρ2gi,j = (12)
u2 second kind with modulus η [12].
Equality (14) can be calculated by the use of numerical
where E(Xi2 ) = E(Yi2 ) = E(Xj2 ) = E(Yj2 ) = u, methods, namely polynomial approximation (to evaluate the
E(Xi Xj ) = E(Yi Yj ) = u1 , E(Xi Yj ) = −E(Yi Xj ) = u2 elliptical integrals). In this way, all elements of Kr can be
and Yi , Xi are the quadrature and in-phase components of the mapped to corresponding elements in Kg . The N envelopes
complex Gaussian sample respectively. of the Gaussian samples in W correspond to Rayleigh ran-
The procedure for generating the N correlated Gaussian dom variables (R1′ , R2′ , . . . , RN
′
) with normalized covariance
samples is summarized as follows [12], [13]: matrix Kr and equal variance
• Determine the Cholesky matrix S corresponding to Kg π
1 2
(the colouring matrix S is the lower triangular matrix σr2 = 2 − σ (15)
2 2 g
such that SS T = Kg , where S T represents the transpose
of S). The desired Rayleigh envelopes (R1 , R2 , . . . , RN ) are created
• Generate N uncorrelated complex Gaussian vector from the samples (R1′ , R2′ , . . . , RN′
) by evaluating Ri = Ai Ri′
σri
V = {v1 , v2 , . . . , vN } each with variance σg2 . where Ai = σr . The correlated Rayleigh random variables
• Calculate correlated complex Gaussian vector using generated via the presented method can be used to generate
W = SV . the multivariate Weibull random variables with arbitrary cor-
relation matrix [7]. The desired Weibull correlation coefficient
Using the desired normalized covariance matrix of the matrix is expressed as
Rayleigh envelopes expressed as  
  1 ρw1,2 . . . ρw1,N
1 ρr1,2 . . . ρr1,N  ρw2,1 1 . . . ρw2,N 
 ρr2,1 1 . . . ρr2,N  Kw =  (16)
 
.. 
Kr = 

..

 (13)  . ... 
 . ...  ρwN,1 ρwN,2 . . . 1
ρrN,1 ρrN,2 ... 1
Therefore, the correlation coefficient of any pair of Rayleigh
where ρri,j is the cross-correlation coefficient between the random variables in terms of the desired Weibull coefficient
Rayleigh samples, it is possible to determine the corresponding is [7]
correlation matrix of the complex Gaussian samples. π  h 3 3 i 
(1 − ρwi,j )2 F

The exact analytical relationship between ρri,j and ρgi,j is ρri,j = , , 1, ρwi,j − 1 (17)
4−π 2 2
OVERVIEW OF FADING CHANNEL MODELING 343

Fig. 6. The generator of correlated Weibull fading envelopes – the front panel of virtual instrument in LabVIEW.

where F (a, d, z) is the ordinary Gaussian hypergeometric the correlated Weibull fading envelope and its histogram for
function [9]. the chosen value of fading parameters.
A simpler approximate relationship between ρri,j and ρwi,j
can be obtained using a least-squares polynomial fit to (16), V. C ONCLUSION
expressed as
Radio propagation characterization and modeling are very
ρri,j ∼
= 0.10(ρwi,j )2 + 0.89ρwi,j (18) important for the communications engineers. Without proper
knowledge of channel models, a wireless system would never
The Weibull random variables are obtained from the corre- be developed. The mobile radio propagation channel intro-
lated Rayleigh random variables with desired fading parameter duces fundamental limitations of the performance of any
α using the transformation W = R2/α . wireless communication systems.
The front panel of the virtual fading generator, obtained In this paper an overview of modeling the fading in com-
using above procedures is shown in Fig. 6. This VI gener- munication channel is presented. Especially, a simple and
ates the correlated Rayleigh and Weibull random variables, efficient procedure for the generation of uncorrelated and
applying the Monte Carlo Simulation (MCS). The simulation correlated Rayleigh, Weibull, Nakagami-m fading channels
was carried out for the arbitrary Weibull correlation coefficient are described. VisSim and LabVIEW software are simple and
matrix specified by straightforward tools used for investigation of the concept of

1 0.795 0.604 0.372
 fading. Mathematical models make it easier to understand
 0.795 1 0.795 0.604  the relationships between the different parameters involved in
Kw =   0.604 0.795
 (19) fading process. The presented results can be applied to analyze
1 0.795 
0.372 0.604 0.795 1 the performance of digital transmission over fading channels.

This correlation matrix is quite similar to that obtained from R EFERENCES

empirical data [2]. The fading parameter vector is given as
α = [4, 3, 5, 2, 1.3] and r0 = [4, 3, 2, 1]. The Figure 7 presents [1] M. C. Jeruchim, P. Balaban, and K. S. Shanmugan, Simulation of
communication systems, modeling, methodology and techniques, 2nd ed.
New York: Kluwer Academic/Plenum Publishers, 2000.
[2] W. C. Lee, Mobile Communications: Design Fundamentals, 2nd ed.
a) b) New York: Wiley, 1993.
[3] M. Patzold, Mobile fading channels. John Wiley & Sons Ltd., 2002.
[4] M. K. Simon and M. S. Alouini, Digital communication over fading
channels, a unified approach to performance analysis, 2nd ed. New
York: John Wiley & Sons, 2005.
[5] K. M. Noga and B. Pałczyńska, “Simulation of fading in communication
channel,” Polish Journal of Environmental Studies, vol. 16, no. 4B, pp.
146–150, 2007.
[6] J. Cheng, C. Tellambura, and N. C. Beaulieu, “Performance of digital
linear modulations on weibull slow-fading channels,” IEEE Transactions
on Communications, vol. 52, no. 8, pp. 1265–1268, August 2004.
[7] D. W. Matolak, I. Sen, and W. Xiong, “Generation of multivariate
Weibull random variates,” IEEE Transactions on Communications,
Fig. 7. The Weibull fading envelope (a) and its histogram (b) for α = 4. vol. 2, no. 4, pp. 523–527, 2008.
344 K. M. NOGA, B. PAŁCZYŃSKA

[8] R. Jain, Art of Computer Systems Performance Analysis Techniques For [11] P. Dent, G. E. Bottomley, and T. Croft, “Jakes fading model revisited,”
Experimental Design Measurements Simulation And Modeling. Wiley IEEE Electronics Letters, vol. 29, no. 14, pp. 1162–1163, 1993.
Computer Publishing, John Wiley & Sons, Inc., 1991. [12] R. B. Ertel and J. H. Reed, “Generation of two equal power correlated
[9] S. Gradstein and M. Ryzhik, Tables of Integrals, Sums, Series and rayleigh fading envelopes,” IEEE Communications Letters, vol. 2, no. 10,
Products, 6th ed. San Diego: Academic Press, 2000. pp. 276–278, 1998.
[10] K. M. Noga and G. K. Karagiannidis, “Characterization of Weibull fad- [13] B. Natarajan, C. R. Nassar, and V. Chandrasekhar, “Generation of
ing channels,” Electronics and Telecommunications Quarterly, vol. 53, correlated Rayleigh fading envelopes for spread spectrum applications,”
no. 1, pp. 47–62, 2007. IEEE Communications Letters, vol. 4, no. 1, pp. 9–11, 2000.