Sie sind auf Seite 1von 15

Fading in RF Systems and Its Usage in CelPlanner

There are multiple ways to estimate link availability. The two most common sets of modeling
approaches for terrestrial wireless communications (excludes satellite and spatial communication
links) may be distinguished as:

 Point-to-point: Fixed line-of-sight (LOS) wireless links


 Point-to-Multipoint: Mobile and fixed wireless links that may or may not
have LOS.
The Point-to-Point category includes most microwave links, which typically are designed with
LOS, and the fading effects observed in those cases are mostly caused by multipath effects such
as: (i) interference between direct rays in a varying refractive index gradient atmospheric medium;
(ii) ground-reflected components; and (iii) partial reflections from atmospheric elevated layers. In
addition to multipath, other effects that cause fading in signals in this category may include
antenna decoupling, earth surface intrusion (even in LOS condition), and precipitation (rain) in
the propagation path. Performance models that deal with fading margins in this context include
the ITU-R 530-8 and Barnett-Vigants multipath outage probability models.

The Point-to-Multipoint category consists of mobile and fixed point-to-multipoint


communication links, where the propagation path may vary with time, and the LOS condition
may not be secured at all times. The typical received signal is a combination of multiple rays that
have traveled different paths, most of them indirect, some reflected off buildings and other
surrounding objects, and others diffracted from rounded objects and knife-edges, such as building
corners.

These multiple samples of the same signal arrive at the receiver with different delays and
amplitudes, depending on the path they have traveled, and combine constructively or destructively
at different occasions (multipath). This set of elements that cause such reflection and diffraction
effects varies with time as the mobile moves or suffers micro motion around its position, while
the surrounding environment may also change its geometry. Performance models that deal with
fading margins in this context typically include different ways to statistically model fading;
CelPlanner supports four different statistical models to deal with that:

 Short-term Fading (Rayleigh distribution)


 Short-term Fading (Rician distribution)
 Long-term Fading (Log-Normal distribution)
 Combined short-term and long-term fading (Suzuki distribution)
With respect to the multipath delay spread described above, the impact of multipath fading
depends on the nature of the transmitted signal when compared to the fading characteristics. If
the propagation channel has a flat response over the frequency domain compared to the
bandwidth of the transmitted signal, the spectral characteristics of the transmitted signal are
preserved at reception, and the channel is referred to as a flat fading channel. In the opposite
scenario, when the channel response is narrower (in the frequency domain) than the transmitted
signal bandwidth, besides the variation in amplitude there is also a distortion induced by
intersymbol interference (ISI), and this effect is referred to as frequency selective fading.

Most systems are assumed to suffer flat fading for the purpose of performance analysis.
Techniques such as OFDM are used to allow the transmission over large bandwidths while
mitigating the effect of frequency selective fading, by the use of multiple smaller carriers, which
can be modeled as individually suffering flat fading. The models described here refer to flat fading
channels.

With respect to the coherence time of the fading, a channel is referred to as fast fading when the
received signal varies rapidly compared to the symbol duration (i.e., the coherence time of the
channel is smaller than the symbol period of the transmitted signal). Conversely, a channel is
referred to as slow fading when the channel impulse response changes at a rate much slower than
the transmitted baseband signal, and may be considered static over one or several reciprocal
bandwidth intervals. Most channels at large bandwidths are considered slow fading, because the
user mobility (and the mobility of objects in the channel) present small velocities compared to the
large transmitted baseband signal bandwidth.

Flat fading channels, regardless of suffering slow or fast fading, can be described using the
following main distributions.

Short-term fading:

The expression “short-term” intends to describe the signal behavior with respect to its time
variations (for the same receiver location):

 Gaussian, or ideal channel: when there is only one stationary dominant signal, i.e., the
only noise present at the receiver is the Additive Gaussian White Noise (AWGN)
developed at the receiver. This situation is almost impossible to achieve in the mobile
environment.
 Rayleigh channel: where all components are indirect, i.e., the received signal is a
combination of the indirect rays (this is the worst case scenario). The power
distribution function (p.d.f.) of the received signal envelope, when the multipath
components are independent, follows a Rayleigh distribution.
 Rician channel: This is the intermediary situation between the Gaussian channel and
the Rayleigh channel. The received signal includes a dominant stationary component
(typically the LOS direct path) plus additional indirect paths (multipath components).
Long-term fading (Area fading):

In addition to the signal variations over time at a given location, another way to describe
signal strength distribution is to observe how the received signal varies with location, even for
points that are equidistant from the transmitter. These variations are typically caused by signal
shadowing (obstructions) on the different propagation paths that surround the transmitter.

These shadow-related variations with location in the coverage area of a transmitter are
described by the Log-normal distribution (which means a Gaussian distribution around the
mean received signal value, when described in dBs).
It is important to notice that, even though the shadowing log-normal distribution is usually
referred to, in literature, as the distribution associated with “long-term fading”, it consists, in
fact, of two different effects. While the term “short-term fading” or “long-term fading” refer
to variations over time (for a given location), the log-normal distribution refers to a variation
over a set of locations (which may be defined by a circumference, the whole area within a
circle, or the resolution area of a grid bin).

The term “long-term fading” is used interchangeably with shadowing because this effect is
noticed and measurable only if there is no short-term fading present, or if short-term fading is
filtered during measurement post-processing. Considering that the term “fading” refers
specifically to time variations of the signal, a better term to define this effect would be
“shadowing dispersion”, to avoid the misleading connotation of the term fading.
Nevertheless, to keep consistency with current literature, in this document, the term long-
term fading will be used to refer to this effect.

The standard deviation of the log-normal distribution associated with shadowing depends on
the area of scope over which it has been calculated and has to be consistently applied to that
scope. In other words, if predicting signal strength as a function of distance, statistics must be
processed for multiple distances, to find the average received signal and standard deviation at
each distance.

On the other hand, if running a “per-pixel” type of prediction, which already considers clutter
and shadowing effects at each different propagation path, the correct standard deviation
corresponds to the signal variations that are expected within the analysis resolution (e.g. 30x30
meter pixels). One common mistake found in propagation studies is the application of
distance-based standard deviations (calculated over the entire circumference, and most
commonly referred to in books) on top of a pixel-based prediction, which leads to
exaggerated shadowing fading margins.

Combined short-term and long-term fading:

Short-term and long-term (shadowing dispersion) fading are not exclusive effects. If we filter
(low pass) over time the signal strength variations at each location, the effects of short-term
fading are removed, and the local variations in signal strength may be used to obtain the
standard deviation of the shadowing dispersion in the area.

There are multiple approaches that propose performance (outage) models to deal with both
short-term and long-term fading simultaneously, such as the Suzuki distribution.

For the calculation of fading margin, CelPlanner uses the formula of the fading probability
density distribution (pdf) (Log-normal, Rayleigh, Rician, or Suzuki), which are temporal
distributions of the signal envelope at one given location, and integrate this probability over
multiple locations. Because these distributions refer to point-to-multipoint applications, where
multiple receivers surround a transmitter and vice-versa, two approaches are typically used in
calculating these probabilities: border (or edge) and area.

Considering a variable W representing the power received at a given location, and a threshold W0
(both in dBm) and assuming that both the probability density function p(W) of W, and its mean
signal strength MW are known (this mean signal strength can be determined by the use of
prediction models or measurements); the two most common approaches to obtain the probability
of outage are the following:
Edge:

CelPlanner calculates the probability that the received signal power W is above the threshold
W0. This proportion is referred to as  and is defined as:

 
  prob(W  W0 )   p(W )dW
W0

Or, as a function of the fading margin:

  probW  Mw  Mg 

Where:

W = actual received signal level (dBm)

Mw = mean signal strength of points located at the pixel (dBm). In CelPlanner


predictions, this is the level that is calculated by the propagation models as the mean
value expected at a given location. Even though the median is not the same as the mean,
it is usually said that this is the 50% confidence level value, i.e., roughly 50 % of real
measurements at a location are expected to be below and 50% above the calculated Mw
value.

W0= threshold for the received signal level (dBm). In CelPlanner predictions, this is the
level that is displayed as the “predicted dBm” value at a given location, with the fading
margin already discounted, compared to the Mw value calculated by the propagation
model, i.e., W0 is typically a value smaller than Mw, and the difference between them is
called the “Fading Margin”

Mg = Fading margin, expressed in dB as

m 
Mg (dB)  M W  W0  10 log w 
 w0 

w0= threshold for the received signal level (mW)

mW= signal strength at that location (mW)

In other words, the confidence level  showed above expresses the probability of the actual
signal (W) to be larger than the predicted signal (Wo = Mw - Mg) at any given location, given
that the mean of the signal at the location is Mw. That is, it expresses the percentage of time
where the signal suffers fading effects that have amplitudes smaller than Mg. In that context,
it is usual to describe the confidence level distributions as a function of the margin Mg.

Note: This probability calculation can be solved for any point or set of points for which the
distribution is known. In practice, typical applications of this approach include (i) a set of
points on a circumference of a given radius (in the simplistic assumption of propagation
models based on distance), or (ii) a set of points within an analysis pixel, in more elaborate
analysis that take into account different clutter and diffraction effects that vary per pixel.

Area:

In this approach, CelPlanner obtains the proportion of locations within the circular area
defined by the radius L where the received signal W is above the threshold W0. The
calculation assumes that mobiles are uniformly distributed within the cell area, i.e., the
proportion of locations is the same as the proportion of mobiles. This proportion is referred
here as  and is defined below.
 1
A A
 prob(W  W0 )dA

Where A = L2 and dA is an infinitesimal area (dA = ldld). Therefore, the desired
probability is the average probability of W exceeding W0 over the entire circular area:

1 L 2

L2 
0 0
prob(W  W0 )l dl d

Which gives:

2 L

L2 0
prob(W  W0 )l dl

Note: The area approach was developed at a time when predictions were typically circular,
and signal strength predicted values were associated to a whole circumference contour. In
those circumstances, it was desirable to evaluate the probability of having signal strength
above a required threshold inside the circular area, i.e., all points contained in the circle were
taken into account for the “averaging area calculation”, including those close to the
transmitter at the center of the circle (with very strong signal) as well as those close to the
edge of the circle. Therefore, it is easy to see that fading margins had to be smaller in the area
approach (as probabilities of being above a threshold were higher for the same local mean).

Log-Normal Distribution (for long-term fading)

The Log-normal distribution is used to describe signal variations that are location related, typically
caused by signal shadowing (obstructions).

If the excess path loss is defined as the ratio between the actual received signal and that which
would be received in free-space, this variable will have a Gaussian distribution with parameters
that depend on the type of environment and obstructions (buildings, tunnels, hills, trees, etc)
found in the propagation path. In the logarithmic scale (dBs), this distribution is known as the
lognormal distribution.

Therefore the envelope of the received signal W, measured in dB, has a lognormal probability
density function given by:
 1 W  M 
2

pW  
1
exp   W
 
2 W  2   W  

Where MW and (W)2 are, respectively, the mean and variance of W given in decibels. Typical
values of W vary with the scope of the distribution. For grid analyses of 100 m x 100 m (about 3
sec resolution), typical values of W are found to be between 2.5 and 3 dB, while at 1 arc.sec.
resolution (30 m x 30 m grids), they smaller than 2 dB.

Outage Probability Calculation - Edge approach

The formula below expresses the outage probability of the received signal W exceeding a
given threshold Wo at any point on the border (edge) of a circle around the transmitter, given
that the average signal level at that border is mw , is:

1 W  MW 
  probW  W0   1  erf  0 

2   2 W 

where

erf( ) is error function given by

erf x0   2 exp  x 2 dx


x0 1
0

and

 = confidence level of a given threshold on the edge approach.


W = actual received signal level (dBm)
W = standard deviation of W (dispersion of variations in dB) around the local mean.
Mw = mean signal strength of points located at the pixel (dBm).
Wo= threshold for the received signal level (dBm).
Considering the definition of fading margin in dB as

m 
Mg (dB)  M W  W0  10 log w 
 w0 

The confidence level for the log-normal distribution can be rewritten as

1   Mg 
  probW  Mw  Mg   1  erf  

2   2 W 

Outage Probability Calculation - Area approach

The formula below expresses the outage probability of the received signal w exceeding a given
threshold Wo at any point within the area of a circle around the transmitter, given that the
average signal level at the border of that circle is mw , is:

1
  M  W0   2M W  W 10 log e  2 W2    M W  W 10 log e  2 W2  
  1  erf  W   exp   * 1  erf   
  10 log e 2 W  
 2 W
2 2
2
   100 log e     

where

 = confidence level of the threshold wo


W0= threshold for the received signal level (dBm)
Mw = mean signal strength of points located at the edge (dBm)
 = local attenuation factor (typically 3 to 8) Obs. This parameter gives the standard loss per
decade in the area (e.g. if  = 4, the path loss grows with distance at a rate of 40 dB per
decade)
(a,x) is the Incomplete Gamma Function, that can be numerically approximated with good
accuracy.
The Rayleigh Distribution (for short-term fading)

For Rayleigh channels, it is assumed that there is no dominant signal (direct path) at the receiver
location. All components are indirect, i.e., the received signal is a combination of the indirect rays.
The power distribution function (p.d.f.) of the received signal envelope follows a Rayleigh
distribution such as the following:

1  w 
p( w)  exp   
mw  mw 

Where w is the power received at a location (in watts or mWatts), and mw is the mean signal
strength at that location.
Outage Probability Calculation - Edge approach

The following formula expresses the outage probability of the received signal w exceeding a
given threshold wo at any point on the border of a circle around the transmitter, given that
the average signal level at that border is mw , is:

 w0 
  prob(W  W0 )  exp   
 m w 

where

 = confidence level of a given threshold on the edge approach


wo= threshold for the received signal level (mW)
w = actual received signal level
mw=mean signal strength of points located at that distance (mW)
Fading Margin Calculation - Edge approach

Since the previous equation is invertible, the fading margin may be expressed as an equation
as well, given by:

w0
  ln  
mw

or, if expressed in dBs:

m  w 
M arg in (dB)  M W  W0  10 log w   10 log 0   10 log ln  
 w0   mw 

where

 = confidence level of a given threshold on the edge approach


wo= threshold for the received signal level (mW)
Wo= threshold for the received signal level (dBm)
mw=mean signal strength at that location (mW)
Mw = mean signal strength of points located at the border (dBm)
Outage Probability Calculation - Area approach

The formula below expresses the outage probability of the received signal w exceeding a given
threshold wo at any point within the area of a circle around the transmitter, given that the
average signal level at the border of that circle is mw , is:
2

2 m   2 w 

   w   , 0 
  w0    mw 

where

 = confidence level of the threshold wo


wo = threshold for the received signal level (mW)
mw = mean signal strength at that location (mW)
 = environment standard atennuation (typically varies from 2 to 8) Obs. This parameter
gives the standard loss per decade in the area (e.g. if  = 4, the path loss grows with distance
40 dB per decade)
(a,x) is the Incomplete Gamma Function, given by:

x, y    t x 1 exp  t dt


y

The Rician Distribution (for short-term fading with combined LOS and NLOS)

CelPlanner only calculates outage probability y and fading margin for Rician Distribution using
the Edge approach; if users select the Area approach, the tool automatically switches the
calculation to Rayleigh Distribution.

When using Rician distribution, users have the choice of calculating the K factor based on a
predicted or constant value. When working with predicted K factor, users must setup clutter
factor for each morphological type.

When the option of K factor Constant is selected, the Prediction Margin shown in the
bottom of the dialog corresponds to the sum of the all the attenuation (human, penetration)
and fading (shadow and multipath). When the option From prediction is selected, the
output of Prediction Margin shows only what could be called a minimum prediction
margin, because that value includes all other factors (human, penetration, shadow fading) but
not multipath fading because that is calculated through the K factor prediction and varies per
morphology.
The k–Factor Prediction

The K factor is calculated in two circumstances:

 As a prediction to display as on raster output (available from the Prediction menu,


called “K factor (linear)”;
 As an “auxiliary prediction”, calculated internally for use as an input in the calculation
of the fading margin.
Users must take note that the K factor is service class dependent because it is associated to the
environment configuration. For every pixel, the factor (linear) is calculated as:
Where:

Fs is the morphology factor


Fh is a receive antenna height factor:
Fb is the antenna beamwidth factor
K0 is a regression coefficient (constant = 10)
 is a regression coefficient (constant = - 0.5)
d is the distance between the CPE location (pixel) and the sector that provides the
strongest individual signal, which is calculated as the individual power dBm coming from
each sector (even in the DVB-H systems, it does not look at C/(N+I), or aggregate
power)
For the morphology factor (Fs), most authors suggest a valua of 1 in the summer (tress full of
leaves) and 2.5 in the winter (no leaves). Thus, CelPlan’s suggestion for the different
morphological types is the following:

Open areas: Fs = 2.5 to 3.0


Dense foliage: Fs = 1.0
Intermediate density: Fs between 1 and 2.5
The receive antenna height factor (Fh) is considered by most authors as the following:

Fh = ( hrx / 3 )^0.46 (where hrx = receive antenna height in meters)

CelPlanner uses this same formula, but the terminal (CPE) antenna height (hrx) is automatically
taken from the parameter Antenna Height, in the User Terminal Configuration dialog box,
and is class-dependent. This causes the k factor to be the same both when receiving (downlink)
and transmitting (uplink), for the same path loss prediction and fading margin in both directions
(symmetry).

The antenna beamwidth factor (Fb) is proposed by most authors as:

Fb = ( b / 17 )^-0.62 (where b = antenna beamwidth in degrees)

In CelPlanner, the beamwidth (b) is automatically taken from the parameter Antenna
Horizontal Aperture, in the User Terminal Configuration dialog box, and is class-dependent
(terminal).
The Suzuki Distribution (for combined long and short-term fading)

Most channels experience a local signal variation (short-term fading) that can be assumed, in the
worst case, to have a Rayleigh distribution (typically expected in outdoor scenarios). Considering
that the mean value of the signal W received at a location is given by mw, the probability dense
function of this local mean, over a given area follows a lognormal distribution, which has an area
mean value given by MWA and standard deviation WA. The pdf of W is therefore given by the
density of W conditional to the local mean mw (Rayleigh) averaged over all possible values of mw
(i.e., averaged over the lognormal distribution).

 
w   w 2   1  m  M 2 
p( w)   mw 10 exp    exp   w   dmw
WA

8 W2      
mw
10  4 x 10 10
 
2  W

Traffic simulation with fading

Snapshot Initialization: for each session, the tool creates a vector with the prediction from each
sector in the system whose prediction radius extends to the call (session) location.

To consider Shadow Fading margins during the simulation, the following procedure is performed
for each snapshot:

Draw Penetration Loss and Human Body Attenuation

These values are selected (random draw) once for each session within a snapshot and do
not vary with server selection. Because neither parameter can assume negative numbers
(gains), they are limited to a minimum value of zero.

Draw Shadow Fading

Shadow fading values are generated as a different random variable for each sector.
However, sectors of the same site are correlated, and therefore a correlation factor
between sectors of same site is required. Users can define this parameter in the Intra Cell
Correlation Factor field. The Inter Cell Correlation Factor represents the correlation
between sectors of different sites. These factors vary between 0 and 1, and the inter cell
factor has to be smaller than or equal to the intra cell factor.
To give more insight to the meaning of the correlation factors, it is important to highlight
some interesting points:

 when the Intra Cell Correlation Factor is equal to 1, fading values are
identical among sectors of same site.
 when the Inter Cell Correlation Factor is equal to 1 (which implies the
intra cell factor is also 1), fading values for all sites and sectors.
 when the Intra Cell Correlation Factor is equal to 0 (which implies the
inter cell factor is also 0), all components are independently random.
In the context of one session, one value of Shadow Fading is generated for each sector,
and is applied to all predictions coming from that sector. To generate fading values from
each sector in the system, CelPlanner proceeds as follows, for each session.

 Generate a random number (Fcomm), from 0 to 1, that is the common part of the
random fading value for a session. Therefore, it is applied to all fading values of
that session (independent on which sector/site it comes from). It can be
abstracted as related to the physical shadowing fading causes located close to the
mobile, such as a valley, or trees, i.e., this fading “belongs to the session” and
doesn’t vary with the direction the signal is coming from.
 Generate a random variable (Fsite) for each different site in the list of servers of a
call. This factor is related to the shadow fading experienced on the path from the
site location to the session (mobile location). Therefore, the same value is used
for signals from all sectors that come from the same location (same site). . This
factor has no effect when the Inter Cell Correlation Factor is set to 1, and, in
this case, Fcomm is applied to all elements.
 For the same session, fading coming from different sectors may have some
components that are uncorrelated with each other, e.g. the antennas may not be
exactly in the same position for each sector, or diversity configurations may exist.
This variable is related to the individual sector component on the shadow fading,
and is generated independently as a random variable (Fsector) for each sector
serving the session. This factor has no effect when the Intra Cell Correlation
Factor is set to 1.
The signal coming from each individual sector is then affected by Shadow Fading by
combining these three components (Fcomm, Fsite, Fsector).

Draw Multipath Fading

The multipath fading calculation follows a very similar procedure to the shadow fading
calculation.

Calculate Delta for Path Loss

The final delta to be used for path loss calculation is a combination of the values described in the
previous steps:
 Penetration loss per session
 Body loss per session
 Shadow fading (per sector and session)
 Multipath fading (per sector and session)

Das könnte Ihnen auch gefallen