Statistical Description of the Wireless Channel

Dr. Ahmed Roumane

April 20, 2025

1 Introduction
In many practical scenarios, it is neither feasible nor necessary to explicitly model all individual propaga-
tion phenomena such as reflections, diffractions, and scatterings that give rise to the various Multi-Path
Components (MPCs). The complexity and computational cost of deterministic modeling make it im-
practical, especially in dynamic or large-scale environments.
Instead, it is often more effective to adopt a statistical approach, in which the propagation channel
is characterized by the probability distributions of its parameters. These parameters can include delay,
angle of arrival, power, or Doppler shift. Rather than assigning specific values to specific locations,
statistical models aim to capture the likelihood of a channel characteristic assuming certain values,
averaged over space, time, or an ensemble of environments.
This abstraction enables the design and analysis of communication systems without requiring precise
knowledge of the environment, and is particularly useful for performance evaluation, protocol develop-
ment, and capacity analysis.

1.1 Scales of Fading and Path Loss

The variations of the received power with distance occur across multiple spatial scales:

• Small-scale fading: On the scale of a wavelength, received power exhibits rapid fluctuations due
to interference between multiple propagation paths (MPCs). These variations are statistical in
nature and are typically described using a local mean power and the distribution of fluctuations
around that mean.
• Large-scale fading (shadowing): When power is averaged over about 10 wavelengths, the
resulting local mean shows variations over much larger distances (on the order of hundreds of
wavelengths). These fluctuations are caused by shadowing from large obstacles and can also be
described statistically.

Figure 1: Received power as a function of distance from the TX.

1
 Figure 2: Types of received power variations.

• Path loss: The overall trend of the received power decreases monotonically with the distance
from the transmitter. This deterministic behavior is described by path loss models that reflect the
average attenuation due to propagation in space.

1.2 Time-Invariant Two-Path Model

To introduce the concept of multipath propagation and small-scale fading, we consider the simplest case
of two time-invariant multipath components (MPCs) received at the same location.

1.2.1 Single Path Case

Let the transmitted signal be a sinusoidal waveform:
ETX (t) = cos(2πfc t)
If the propagation distance is d, then the received signal is:
E(t) = E0 cos(2πfc t − k0 d)
2π
where k0 = λ . Using complex baseband representation:
E = E0 e−jk0 d

1.2.2 Two Path Case

Assume two waves arrive at the RX position r with path lengths d1 and d2 , resulting in delays τ1 = dc01
and τ2 = dc02 . The receiver is in the far field of the reflecting objects (IOs), and the received field is
modeled as the sum of two homogeneous plane waves:
⃗ ⃗
E(⃗r) = E1 e−j k1 ·⃗r + E2 e−j k2 ·⃗r
where ⃗k1 and ⃗k2 are wave vectors with magnitude k0 , indicating direction of propagation.

1.2.3 Interference and Fading

This superposition results in spatially varying interference:
• Constructive interference: occurs when the two waves are in phase, leading to an amplitude of
E1 + E2 .
• Destructive interference: occurs when the phase difference is 180◦ , reducing the amplitude to
|E1 − E2 |. If E1 = E2 , complete cancellation occurs.
The pattern of interference causes small-scale fading, with fading dips appearing roughly every
wavelength (λ). This phenomenon is evident in received power plots and is characteristic of interference
between MPCs in mobile or indoor channels.

2
 Figure 3: Geometry of the time-invariant two-path model.

Figure 4: Interference of two planewaves with E1 = E2 = 1 and arg (k1, k2) = 30°.

1.3 Time-Variant Two-Path Model

In realistic environments, the lengths of propagation paths change over time due to motion of the
Transmitter (TX), Receiver (RX), or interacting objects (IOs). For simplicity, we consider only RX
motion, typical in a cellular downlink scenario.

1.3.1 Time-Varying Fading

As the RX moves, it experiences a time-varying interference pattern caused by multipath components
(MPCs). The spatially varying field strength pattern becomes time-dependent, leading to small-scale
fading over time.

2 Small-Scale Fading Without a Dominant Component

We now investigate a more general case of multipath propagation involving a moving RX and multiple
interaction objects (IOs). Due to the large number of IOs, we transition from a deterministic to a
stochastic description, which is essential for wireless communication systems.

2.1 A Computer Experiment

Consider the following simple computer experiment: The RX moves over a small area, receiving signals
from several IOs distributed uniformly around the RX. The waves are assumed to be homogeneous plane
waves, and the movement of the RX within the area does not affect their amplitudes. Each wave has a
random phase and amplitude. The complex field strengths of the constituting waves are described as:

3
 Figure 5: Projection of velocity vector |⃗v | onto the direction of propagation k.

Figure 6: Instantaneous value of the field strength at time t = 0 – i.e., Re{E}. Superposition of the
eight constituting waves in the area 0 < x < 5λ, 0 < y < 5λ.

E1 (x, y) = 1.0 exp[−jk0 (x cos(169◦ ) + y sin(169◦ ))] exp(j311◦ )

E2 (x, y) = 0.8 exp[−jk0 (x cos(213◦ ) + y sin(213◦ ))] exp(j32◦ )
E3 (x, y) = 1.1 exp[−jk0 (x cos(87◦ ) + y sin(87◦ ))] exp(j161◦ )
E4 (x, y) = 1.3 exp[−jk0 (x cos(256◦ ) + y sin(256◦ ))] exp(j356◦ )
E5 (x, y) = 0.9 exp[−jk0 (x cos(17◦ ) + y sin(17◦ ))] exp(j191◦ )
E6 (x, y) = 0.5 exp[−jk0 (x cos(126◦ ) + y sin(126◦ ))] exp(j56◦ )
E7 (x, y) = 0.7 exp[−jk0 (x cos(343◦ ) + y sin(343◦ ))] exp(j268◦ )
E8 (x, y) = 0.9 exp[−jk0 (x cos(297◦ ) + y sin(297◦ ))] exp(j131◦ )
The total field strength E is obtained by superimposing these constituting waves, which can be
interpreted as adding up complex random phasors 6.

2.2 Statistics of the Field Strength

The field strength statistics are analyzed by plotting the instantaneous value of the total field strength
at a given time. The histogram of the real part of the complex field strength, Re(E), follows a zero-mean
Gaussian distribution, which is a consequence of the central limit theorem (i.e. the distribution of the

4
 Figure 7: Histogram of the field strength of Figure 6 A Gaussian pdf is shown for comparison

Figure 8: Pdf of the received amplitude.

sum (or average) of a large number of i.i.d random variables approaches a normal distribution). The
conditions for the central limit theorem are met, as the eight constituting waves have random angles of
incidence and phases, and no amplitude is dominant. Figure 7 shows that the imaginary part of the
field strength, Im(E), is also normally distributed.

2.3 Envelope Distribution

The behavior of most receivers (RXs) is determined by the magnitude (absolute amplitude) of the
received signal. Figure 8 shows the statistics of the amplitude over that area (a Rayleigh pdf). A
Rayleigh distribution describes the magnitude of a complex stochastic variable whose real and imaginary
parts are independent and normally distributed.

3 Mathematical Derivation of the Statistics of Amplitude and

Phase
After these pseudo-experimental considerations, we now turn to a more detailed, and more mathemati-
cally sound, derivation of Rayleigh distribution. Consider a scenario where N homogeneous plane waves
(MPCs) have been created by reflection/scattering from different incident directions. The incident di-

Figure 9: Pdf of the received phase.

5
rections (IOs) and the transmitter (TX) do not move, and the receiver (RX) moves with a velocity v.
The real part of the received field strength due to the ℓth MPC is thus aℓ cos(ϕℓ ), and the imaginary
part is aℓ sin(ϕℓ ). With absolute amplitudes aℓ and phase ϕℓ .
As before, we assume that the absolute amplitudes of the MPCs do not change over the region of
observation. The sum of the squared amplitudes is thus:
N
X
|aℓ |2 = CP (5.9)
ℓ=1

where CP is a constant. However, the phases φℓ vary strongly and are thus approximated as random
variables uniformly distributed in the range [0, 2π].
The phases at different times (or equivalently, locations) are interpreted as different realizations
of uniformly distributed random variables. This can be justified by considering the Doppler shift for
computation of the total field strength E(t). For an unmodulated carrier, we get the expression in real
passband notation:
N
X
E(t) = |aℓ | cos (2πfc t − 2πνmax cos(γℓ )t + φ0,ℓ ) (5.10)
ℓ=1

where φ0,ℓ is the phase of the ℓth MPC at time 0. Rewriting this in terms of in-phase and quadrature-
phase components in real passband notation, we obtain:

EBP (t) = I(t) cos (2πfc t) − Q(t) sin (2πfc t)

with
N
X
I(t) = |aℓ | cos (−2πνmax cos(γℓ )t + φ0,ℓ ) (5.12)
ℓ=1
N
X
Q(t) = |aℓ | sin (−2πνmax cos(γℓ )t + φ0,ℓ ) (5.13)
ℓ=1

The arguments of the cos and sin functions can thus be interpreted as random variables that take
on different values at different times. For sufficiently large time intervals, these realizations can be
considered independent.
Based on the above assumptions, both the in-phase and quadrature-phase components are sums of
many random variables, none of which dominate. It follows from the central limit theorem that the
probability density function (pdf) of the amplitude distribution is Rayleigh, regardless of the exact pdf
of the constituent amplitudes. A zero-mean Gaussian random variable has the pdf:

x2
 
1
pdf(x) = √ exp − 2 (5.14)
2πσ 2 2σ

where σ 2 denotes the variance.

The pdf of the received phase ψ = arg{E} is uniform:
1
pdf(ψ) = for − π ≤ ψ ≤ π (5.15)
2π
The pdf of the amplitude r follows a Rayleigh distribution:

r2
 
r
pdf(r) = 2 exp − 2 for r ≥ 0 (5.16)
σ 2σ

The squared amplitude (signal power) follows an exponential distribution:

 
1 P
pdfP (P ) = exp − , P ≥0
P̄ P̄

where P̄ is the mean power.

Properties of the Rayleigh Distribution

6
 Figure 10: Pdf of a Rayleigh distribution

• Mean value: E[r] = σ

pπ
2

• Mean square value: E[r2 ] = 2σ 2

• Variance:
π
Var(r) = E[r2 ] − (E[r])2 = 2σ 2 − σ 2 = 0.429σ 2
2
√
• Median value: r50 = σ 2 ln 2 ≈ 1.18σ
• Location of maximum of pdf: Occurs at r = σ
The cumulative distribution function is defined as the probability that the realization of the random
variable has a value smaller than or equal to x. The CDF is given by:
Z r
cdfr (r) = pdfr (u) du
−∞

r2
 
CDF(r) = 1 − exp − 2 (5.19)
2σ
For small values of r, this can be approximated as:

r2
CDF(r) ≈ (5.20)
2σ 2
The Rayleigh distribution is widely used in wireless communications for its accuracy, simplicity, and
ability to model worst-case fading. It relies on one parameter (mean power) and offers mathematical
convenience but may not apply in LOS or ultra-wideband scenarios. Verification uses Weibull paper,
where exponential distributions appear as straight lines.

3.1 Fading Margin for Rayleigh-Fading

In noise-limited systems, the received power determines the system performance. Due to fading effects,
such as Rayleigh fading, the received power exceeds a given minimum value only in a certain percentage
of cases.
The key design question becomes: Given a minimum required receive power Pmin for successful com-
munication, how large must the mean power P̄ be to ensure that the outage probability does not exceed
x%? In other words, what fading margin is necessary?
The cumulative distribution function (CDF) provides the probability that the received power does
not exceed a given threshold. For small Pmin , the CDF of an exponentially distributed received power
(as in Rayleigh fading) can be approximated as:
Pmin
x = cdf(Pmin ) ≈
P̄

7
This implies that to meet an outage probability of x, the required average power must satisfy:
Pmin
P̄ =
x
Example 5.1
For a signal with Rayleigh-distributed amplitude, what is the probability that the received signal power
is at least 20, 6, or 3 dB below the mean power? Compare the exact result and the result from the
approximate formulation of Eq. (5.20).
A power level 20 dB below the mean power corresponds to:
Pmin 1
=
P̄ 100
Then the probability that the received power is below this threshold is:
 
1
P(P < Pmin ) = 1 − exp − ≈ 9.95 × 10−3 (5.23)
100

Similarly, the exact results for power levels 6 dB and 3 dB below the mean are:
(
0.221 for 6 dB
P(P < Pmin ) =
0.393 for 3 dB

The approximate formulation of Eq. (5.20) gives:


Pmin 0.01
 for 20 dB
= 0.25 for 6 dB
P̄ 
0.5 for 3 dB


Thus, the approximation is reasonably accurate for power levels 6 dB below the mean, but breaks
down for higher values of Pmin .

3.2 Fading Margin in Interference-Limited Systems

In interference-limited systems, both the desired signal and the interfering signals are subject to Rayleigh
fading. This adds complexity to the system performance analysis, as both signal and interference fluctuate
randomly.
To analyze such systems, we consider the amplitude ratio r of the desired signal to the interference.
If both signal and interference follow Rayleigh distributions, the probability density function (pdf) of r
is given by:
2θ̃r
pdf(r) =  2 , r ≥ 0
θ̃ + r2
σ2
where θ̃ = σ12 is the ratio of the mean signal power to the mean interference power (SIR).
2
The corresponding cumulative distribution function (cdf) is:

θ̃
cdf(r) = 1 − , r≥0
θ̃ + r2
We can also derive the pdf and cdf of the signal-to-interference power ratio θ as:

θ̃
pdf(θ) = , θ≥0
(θ̃ + θ)2

θ̃
cdf(θ) = 1 − , θ≥0
θ̃ + θ

8
 Figure 11: Magnitude of the electric field strength, |E|, in the presence of a dominant MPC.

Figure 12: Histogram of the amplitudes in the presence of a dominant MPC.

4 Small-Scale Fading with a Dominant Component

4.1 A Computer Experiment
Fading statistics change when a dominant multipath component (MPC)—e.g., a line-of-sight (LOS) or
strong specular reflection—is present. To study this, we repeat the computer experiment of Section 2.1
, but include an additional wave of amplitude 5:

E9 (x, y) = 5.0 · exp [−jk0 (x cos(0◦ ) + y sin(0◦ ))] · exp(j0)

The presence of this dominant component is visible in the field’s magnitude. Compared to the
Rayleigh case, the probability of deep fades is significantly reduced (See figure 11).

4.2 Derivation of the Amplitude and Phase Distributions

Assuming the LOS component has zero phase, the real part of the field is Gaussian with nonzero mean,
and the imaginary part remains zero-mean Gaussian.
The joint PDF of amplitude r and phase ψ is:
 2
r + A2 − 2rA cos ψ

r
pdfr,ψ (r, ψ) = exp − , r ≥ 0, −π < ψ ≤ π
2πσ 2 2σ 2
Integrating over ψ, we obtain the Rice distribution (See figure 12):
 2
r + A2
  
r rA
pdfr (r) = 2 exp − I0 , r≥0
σ 2σ 2 σ2
where I0 (x) is the modified Bessel function of the first kind, order zero.
The mean square value is:

E[r2 ] = 2σ 2 + A2

9
Figure 13: Rice distribution for three different values of Kr – i.e., the ratio between the power of the
LOS

The Rice factor is defined as:

A2
Kr =
2σ 2
For Kr = 0, the Rice distribution reduces to the Rayleigh distribution. For large Kr , it approximates
a Gaussian with mean A, (See figure 13).
This formulation is particularly useful for analyzing and computing reuse distances in cellular systems.

4.3 Phase and Power Distributions in the Presence of a Dominant Compo-

nent
The presence of a dominant multipath component (MPC) changes the phase distribution. Intuitively, as
the dominant component becomes stronger, the phase of the total signal approaches that of the dominant
component. Hence, the phase distribution converges to a Dirac delta function at ψ = 0.
The general phase probability density function (PDF) is:
√ 2  √ 
1 + πKr eKr cos ψ cos(ψ) 1 + erf( Kr cos(ψ))
pdfψ (ψ) = , −π < ψ ≤ π (5.33)
2πeKr
where the error function is:
Z x
2 2
erf(x) = √ e−t dt
π 0
The PDF of the signal power P is given by:

  r !
1 (1 + Kr )P Kr (1 + Kr )P
pdfP (P ) = (1 + Kr ) exp −Kr − I0 2 , P ≥0 (5.34)
P P P

From a mathematical point of view, this is equivalent to the problem of a deterministic phasor
(resulting in a non-zero mean) added to a zero-mean complex Gaussian distribution.

4.4 Nakagami Distribution

The Nakagami-m distribution is widely used to model the statistical behavior of signal amplitude in
wireless communication systems. Its probability density function (PDF) for the amplitude r is given by:
2 m  m  1
pdfr (r) = ( )m r2m−1 exp − r2 , r ≥ 0, m ≥ (5.35)
Γ(m) P̄ P 2
where: - Γ(m) is Euler’s Gamma function,
Z ∞
Γ(x) = tx−1 e−t dt,
0

10
Figure 14: Probability density function (PDF) of the phase of a non-zero-mean complex Gaussian dis-
tribution with standard deviation σ = 1 and mean amplitude A = 0, 1, 3, 10.

which reduces to Γ(x) = (x − 1)! for integer x, - P̄ = E[r2 ] is the mean square value of the amplitude,
and - m is the fading parameter, defined as:
P̄ 2
m= (5.36)
E[(r2 − P̄ )2 ]

If the amplitude follows the Nakagami distribution, the corresponding power P = r2 follows a Gamma
distribution:
 
m mP m−1 mP
pdfP (P ) = ( ) exp − , P ≥0 (5.37)
P Γ(m) P P
The Nakagami and Rice distributions are similar in shape, and one can approximate the other. For
m > 1, the Nakagami fading parameter m can be related to the Rice factor Kr using:

(Kr + 1)2
m= (5.38)
2Kr + 1
and conversely: p
(m2 − m)
Kr = √ (5.39)
m − m2 + m

4.4.1 Remarks
• While both distributions show similar behavior for general values, they differ significantly near
r = 0, which influences the achievable diversity order.
• The Rice distribution gives the exact amplitude PDF for a non-zero-mean complex Gaussian pro-
cess—corresponding to a dominant LOS component and many weak scatterers.
• The Nakagami distribution provides an approximate amplitude distribution for scenarios where the
central limit theorem may not apply, such as ultrawideband (UWB) channels.

5 Temporal Variations for Moving UE

5.1 Doppler Shift vs. Doppler Spread
To distinguish between a Doppler shift and the Doppler spread, consider the following:
A pure frequency shift of the signal, if experienced uniformly across all multipath components (MPCs),
has negligible impact on system performance. This is because the receiver (RX) can estimate the carrier
frequency from the received signal using a technique called carrier recovery, effectively compensating for
the Doppler-shifted carrier.

11
 Figure 15: Superposition of two carriers with different frequencies (beating).

However, this situation only occurs if there is a single MPC. In reality, different MPCs typically arrive
from different directions and experience different Doppler shifts. This leads to Doppler spreading, or
frequency dispersion, which has a much more significant influence on system performance.

5.1.1 Doppler Shift (Single Path)

Consider a sinusoidal wave received by a moving RX. If the RX moves away from the TX with speed v,
the received signal is:
  v 
E(t) = E0 cos (2πfc t − k0 (d0 + vt)) = E0 cos 2πt fc − − k0 d0
λ
Thus, the Doppler shift is:
v v
ν=− = −fc
λ c0
If the direction of motion is at an angle γ to the wave direction, the Doppler shift becomes:
v
ν = −fc cos γ = −νmax cos γ
c0

5.1.2 Doppler Spread (Multiple Paths)

When multiple MPCs are present, each arriving from a different direction, they experience different
Doppler shifts. This leads to a Doppler spread—a range of frequency shifts—causing frequency dispersion.
As a result, the received signal exhibits beating: a periodic amplitude fluctuation at a rate equal to
the difference in Doppler shifts. This creates time-domain fading dips and corresponds to small-scale
fading (See figure 15).

5.1.3 Implications of Doppler Spread

• Fading Rate: Can be found by:
1. Counting dips per second as RX moves through spatial interference pattern.
2. The beat frequency is the difference of the Doppler shifts of the two Doppler-shifted waves.
• Significance of Doppler Spread: Although the Doppler spread is small in magnitude, it is an
important parameter of the wireless channel:

– Doppler spread measures the rate of change of the channel over time.
– The superposition of multiple slightly Doppler-shifted signals results in phase shifts of the
total received signal. These shifts can impair the reception of angle-modulated signals, leading
to random frequency modulation (FM) of the received signal. This effect is particularly
important for signals with low bit rates.
– In Orthogonal Frequency Division Multiplexing (OFDM) systems, which use many
closely spaced subcarriers, Doppler spread can lead to inter-carrier interference. This
effect becomes especially pronounced at high carrier frequencies such as in millimeter-wave
(mmWave) systems.

When a user equipment (UE) moves, the multipath components (MPCs) arriving from different
directions experience different Doppler shifts, leading to a broadening of the received signal spectrum.
This section derives the Doppler spectrum under the narrowband assumption.

12
 Figure 16: Jakes Doppler spectrum.

Let γ be the angle between the UE’s velocity vector ⃗v and the wave direction at the UE. The received
frequency f is shifted from the carrier frequency fc by:
 
v
f = fc 1 − cos γ = fc + ν (1)
c0
fc v
where v = |⃗v |, c0 is the speed of light, and ν is the Doppler shift. The maximum shift is νmax = c0 ,
giving a frequency range [fc − νmax , fc + νmax ].
To analyze statistically, let pdfγ (γ) be the angular power density function of incident waves.
For a uniform angular distribution:
1
pdfγ (γ) = , −π < γ ≤ π
2π
and for an ideal vertical dipole antenna with G(γ) = 1.5, the Doppler spectrum becomes:

1.5 Ω̄
SD (ν) = p , |ν| ≤ νmax (2)
π 2
νmax − ν2
This is known as the Jakes spectrum, with singularities at ν = ±νmax corresponding to for-
ward/backward movement (See figure 16).

Temporal Correlation: The autocorrelation function of fading describes how similar the fading states
at two times separated by ∆t are. The autocorrelation of the in-phase component is:

E[I(t)I(t + ∆t)]
= J0 (2πνmax ∆t) (3)
E[I(t)2 ]
and the normalized envelope covariance is approximately:

E[(r(t)r(t + ∆t) − E[r(t)]2 )]

≈ J02 (2πνmax ∆t) (4)
E[r(t)2 ] − E[r(t)]2
where J0 (·) is the Bessel function of the first kind, zero order.

Remarks:
• The Doppler spectrum describes frequency dispersion and temporal channel variation.
• It is central in modeling fading for both narrowband and wideband systems.
• Alternative models include the Aulin spectrum, Gaussian, and uniform Doppler spectra.

13
 Figure 17: Amplitude correlation as a function of displacement

Example 5.3 Assume that a UE is located in a fading dip. On average, what minimum distance
should the UE move so that it is no longer influenced by this fading dip?
The temporal correlation function when traveling at a speed v through a static environment is equiv-
alent to the spatial autocorrelation function with spatial displacement v∆t. As a first step, we consider
the envelope correlation function, which is plotted in Figure 17.
If we define “no longer influenced” as having an envelope correlation coefficient equal to 0.5, then (on
average) the receiver must move a distance of approximately 0.18 λ. For complete decorrelation from the
fading dip, a movement of approximately 0.38 λ is required.

5.2 Temporal Variations in Fixed and Nomadic Wireless Systems

In a fixed wireless system, as well as in nomadic systems where the UE does not move over the time
period of interest, temporal variations arise solely due to the movement of the interacting objects (IOs).
This leads to a different perspective on modeling temporal channel changes.
In a mobile link, the UE moves through the spatial fading “landscape”, and thus spatial and temporal
fading are directly related via the UE’s speed v. However, in a fixed wireless link, the observed variation
is caused by the motion of scattering objects—typical examples include vehicles or vegetation swaying
in the wind.
Empirical observations have shown that the amplitude statistics of fading in such scenarios often
follow a Rician distribution. However, the Rice factor in this context quantifies the ratio of power in
time-invariant MPCs to that in time-variant MPCs. Thus, the temporal Rice factor is unrelated to
the presence of a line-of-sight (LOS) path or dominant components. For example, in a purely NLOS
environment with many equally strong, time-invariant MPCs, the temporal Rice factor can become very
large—even approaching infinity.
The Doppler spectrum in such systems contains a delta function at ν = 0, representing the static com-
ponents, along with a continuous part that is shaped by the motion and distribution of the moving IOs.
Numerous measurement campaigns have reported that the diffuse part of this spectrum is approximately
Gaussian.

5.3 Generation of Fading

The prevalent approach for simulating Rayleigh fading with a Jakes Doppler spectrum is based on a
discrete sum-of-sinusoids model, given by:
N
X
h(t) = Ci exp (j2πtνmax cos γi + ϕi )
i=1

where typically the parameters are set as:

1 2πi + θi
Ci = √ , γi =
N N

14
 Figure 18: LCR for a Rayleigh-fading amplitude and Jakes spectrum.

In Jakes’ original model, (which models multipath fading due to relative motion), the phases ϕi were
all set to zero, i.e., ϕi = 0. However, this approach leads to a non-stationary fading process. To achieve
statistical stationarity, it is necessary to use random phases. More specifically, both ϕi and θi should be
independently and uniformly distributed random variables.
An alternative approach involves placing interacting objects (IOs) in a geometric environment and
simulating the resulting accumulated signal via simple ray tracing techniques.

6 Temporal Fading Characterization

6.1 Level Crossing Rate
The Doppler spectrum provides a complete characterization of the temporal statistics of fading. However,
for practical system design, more intuitive metrics are often needed. One such metric is the Level
Crossing Rate (LCR), which quantifies how frequently the received signal crosses a specified level in
the positive direction.
Since deeper fades occur less frequently, the LCR depends on the chosen threshold level. Mathemat-
ically, the LCR is defined as the expected rate at which the received signal crosses a threshold r in the
positive direction. See figure 18 The Level Crossing Rate (LCR) is highest near the average level (i.e.,
r
when rrms ≈ 1), because the signal crosses this level most often.
At very low or high levels, the signal crosses less frequently, so the LCR drops.
Knowing the LCR helps in setting adaptive thresholds, error correction, or diversity schemes.

6.2 Average Duration of Fades

Another useful metric is the Average Duration of Fades (ADF). This is defined as the ratio between
the probability that the signal is below a threshold (i.e., the CDF) and the LCR:

cdf(r)
ADF(r) = , for r ≥ 0 (5)
NR (r)
This represents the average time the signal remains below a certain level (i.e., a fade), scaled by the
maximum Doppler frequency fD .
Knowing the AFD is useful for designing interleaving and error correction schemes, selecting channel
estimation intervals, and evaluating how long fades may disrupt communication.

6.3 Random Frequency Modulation

In time-varying channels, phase shifts are random and vary with time—effectively resulting in a random
frequency modulation (FM).

15
 Instantaneous frequency spans the full real line—unbounded by the maximum Doppler shift. Impor-
tantly, strong frequency modulation is more probable during fading dips, where signal levels
are low.
Fading dips cause increased noise susceptibility and stronger FM, leading to possible phase errors in
modulation schemes and contributing to intersymbol interference.

7 Large-Scale Fading
7.1 Large-Scale Fading Distribution
Large-scale fading refers to the slow variations in received signal strength caused by the shadowing of
multipath components (MPCs) by large objects such as buildings. Unlike small-scale fading, which varies
over a few wavelengths, large-scale fading occurs over much longer distances.
To isolate large-scale fading, the received signal is first averaged over a small area (e.g., 10 by 10
wavelengths), yielding the Small Scale Averaged (SSA) field strength. Experimental studies show that
the SSA field strength F , when expressed in dB, follows a Gaussian distribution. Consequently, the field
strength in linear scale follows a lognormal distribution, with the following probability density function
(pdf):
(20 log10 F − µF,dB )2
 
20/ ln(10)
pdfF (F ) = √ exp − , F ≥0
F σF 2π 2σF2
where σF is the standard deviation and µF,dB is the mean value in dB.
Similarly, the received power is also lognormally distributed. For Rayleigh small-scale fading, the
median power in dB is related to the median field strength as:
 
4
µP,dB = µF,dB + 10 log10
π

The corresponding power pdf is:

(10 log10 P − µP,dB )2

 
10/ ln(10)
pdfP (P ) = √ exp − , P ≥0
P σP 2π 2σP2

with σP = σF , typically ranging from 4 to 10 dB.

The lognormal nature of fading arises from cumulative shadowing effects, as each interaction (e.g.,
diffraction, scattering) attenuates the signal. On a logarithmic scale, these attenuations add, leading to
a sum of random variables and thus a lognormal distribution.
In practical scenarios such as interference modeling or channel modeling with multiple MPC clusters,
one often needs to sum several lognormal random variables. The sum is approximately lognormal. A
commonly used approximation is the Fenton–Wilkinson method, under the assumption of equal variances
σc2 . Finally, shadowing shows spatial correlation. That is, the values of shadowing at nearby locations
are not independent. The most commonly used model assumes exponential correlation:
 
∆x
E[F (x)F (x + ∆x)] = exp −
x̄c

where xc is the decorrelation distance, typically between 5 and 50 meters depending on the environment.

16