Académique Documents
Professionnel Documents
Culture Documents
V 8.1
VERSION HISTORY
Version
Date
Author
Comments
1.0
1.1
1.2
1.3
28/09/2007
18/01/2007
12/10/2011
14/06/2013
A. Panman
A. Lodhi
A. Lawrow
V. Kontogeorgos
First version.
Corrected DL RSS equation.
Version 8.0.
Version 8.1
Added Coverage Probability OK array
Updated Nth DL Loss array description
Added Composite Tech arrays
Added Terminal Info Tech Type arrays
CONTENTS
1
What Is A Snapshot?................................................................................................................... 4
Randomness in a Cellular Network .................................................................................. 4
Time-averaging in Coverage Evaluations......................................................................... 7
Time-averaging in Capacity Evaluations .......................................................................... 7
Why Produce Snapshots? .................................................................................................... 7
Activity Factors..................................................................................................................... 8
Formulae ....................................................................................................................................... 9
2.1
Notation ................................................................................................................................. 9
2.2
List of Principal Symbols .................................................................................................... 9
2.3
Downlink Power Formulae .............................................................................................. 11
Snapshot Overview................................................................................................................... 15
3.1
Random Terminal Distribution ........................................................................................ 15
3.2
Time-Slot Iterations ............................................................................................................ 16
3.3
Convergence Test ............................................................................................................... 16
3.4
Gathering Of Results ......................................................................................................... 16
Connection Evaluation in a Snapshot ................................................................................... 17
4.1
Connection Scenario Prioritisation .................................................................................. 17
4.2
GSM Downlink Evaluation ............................................................................................... 17
4.3
Failure Reasons ................................................................................................................... 18
Output Arrays ............................................................................................................................ 19
5.1
Array dependencies ........................................................................................................... 19
Pathloss Arrays ................................................................................................................................ 20
5.2
Coverage and Data Rate Arrays ....................................................................................... 20
5.3
Composite Tech Arrays ..................................................................................................... 21
5.4
Terminal Info Arrays ......................................................................................................... 21
Coverage Array Calculations .................................................................................................. 23
6.1
Notation ............................................................................................................................... 23
6.2
Fades in the Simulation Snapshots .................................................................................. 23
6.3
Fades in Arrays for Mean Values ..................................................................................... 24
6.4
Fades in Coverage Array Calculations ............................................................................ 24
1.1
1.2
1.3
1.4
1.5
WHAT IS A SNAPSHOT?
1.1
In a simulation of a cellular network there are two main types of randomness that one needs to
consider.
We shall consider the spatial domain to be discrete and consisting of a large number of pixels (bins)
some of which will contain terminals. Each possible pattern of terminal locations has an associated
probability of occurrence. We can label these spatial patterns X1, X2, etc and represent the
corresponding probabilities of occurrence by P(X1), P(X2), etc. An example of two spatial patterns X1
and X2 is shown below.
X1
X2
X
X
X
X
X
Each spatial pattern has many possible configurations of transmitting and non-transmitting terminals.
Two such configurations for the spatial patterns X1 and X2 are shown below, with 1 representing a
transmitting terminal, and 0 a non-transmitting terminal.
(X1,T1)
(X2,T1)
0
1
0
0
0
1
(X1,T2)
(X2,T2)
0
0
1
0
1
We call each of these patterns a spatio-temporal pattern to highlight the fact that we have specified
spatial locations of terminals and also their temporal state (transmitting/non-transmitting). We can
label the spatio-temporal patterns for spatial pattern X1 as follows (X1,T1), (X1,T2), etc and their
probabilites of occurrence P(X1,T1), P(X1,T2), etc. Note that the probability of occurrence of a spatiotemporal pattern (Xi,Tj) is proportional to the probability of occurrence of the spatial pattern Xi:
P(Xi,Tj)= P(Xi) P(Tj | Xi).
One can think of a spatio-temporal pattern as being a picture of a real network at a random instant in
time. This is what most people have in mind when one mentions a simulation snapshot, but a
snapshot in our simulator represents something slightly different, as explained below.
The ideal static simulation would calculate an average quantity (e.g. the average traffic load on a cell)
by performing a weighted sum over the set of all possible spatio-temporal patterns (Xi,Tj), with the
weight for a pattern being its probability of occurrence. So the average of some quantity F would be
given by
F ( X , T ) P( X , T ) .
i
(1a)
X i ,T j
P( X ) F ( X ,T )P(T
i
Xi
| Xi ) .
(1b)
Tj
The summations in (1a) and in (1b) are over every conceivable pattern of terminal locations and
activities, including the unlikely ones, so clearly some simplifications are necessary in any practical
static simulator.
1
N
F ( X ,T )P(T
i
Xi
| Xi ) .
Tj
Spatial randomness is therefore handled explicitly by considering a set of sample spatial patterns that
have been selected in a random and unbiased way. There is still the issue of how to handle the
different temporal states for each sample spatial pattern. There are two main approaches we can use:
(Simplification 2).
(Simplification 3).
1
N
F ( X ,T ) .
i
X i ,T j
If we called a spatio-temporal pattern a snapshot then the above formula simply says that we can
approximate F by performing a simple average over the snapshots. This simple average works
because the sample spatio-temporal patterns are selected in a random and unbiased way. Also note that
this averaging explicitly accounts for spatial randomness and explicitly accounts for temporal
randomness.
There are problems with assigning activity flags to terminals however.
For low activity services, the user can do 100s of snapshots and never set an activity flag, and
therefore certain outputs may not have any results. For example, a simulation report may say
that many users are served on a cell but that there is no throughput on the cell. Forcing the
user to run 1000s of snapshots is unacceptable in a commercial tool, so we either have to
remove the problem outputs or calculate them some other way.
Activity flags are set when the terminals are created. For multi-rate packet-switched (PS)
services, different bearers can have different activity factors. Since we do not know which
bearer a multi-rate terminal will ultimately use, we have to define a set of activity flags, one
flag for every bearer that the terminal may use. This is conceptually horrible, and can lead to
convergence issues during iterations.
For the above reasons, we do not use Simplification 2 and use the following simplification instead.
1.2
All our link budgets (in SI units, not dB) are essentially of the form
E = ( P / L ) / N,
where
E = signal to interference ratio for the link in a period of activity,
P = TX power in a period of activity,
L = linkloss,
N = time-average RX interference on the link.
To check for coverage, we set P to the maximum allowed link power and check that E meets
requirements. In other words we examine the link assuming it is active. The time-averaging affects the
coverage evaluation only because we use the time-average interference N in the link-budget.
1.3
1.4
The main purpose of a snapshot is to provide us with measures of system load. In particular, each
snapshot tells us:
By running many snapshots, we obtain values for these quantities for different spatial distributions of
terminals, and can then proceed to analyse DL coverage for the system.
1.5
ACTIVITY FACTORS
As explained in previous sections, activity factors are used to calculate implicit time-averages for
bearers in a simulation snapshot.
The time-slot activity factor is used to calculate the time-average time-slot consumption for a
packet-switched link.
Time-average Time-Slots = Bearer Activity x Time-Slots when Active.
Circuit-switched (CS) services consume time slots even during periods of inactivity. Therefore CS
connections are assumed to have an activity factor of 100%. A CS connection can use either a full-rate
or half-rate bearer corresponding to one time-slot or half a time-slot respectively. Which bearer is
selected depends on the achieved traffic C/I, whether the bearer supports half-rate connections and the
sub-cell traffic load.
PS Activity Factor Calculation
For a PS service, the tool automatically calculates the power-activity factor from the PS parameters in
the service definition:
Ruser
N bytes
N packets
N calls
Tread
Tinter
BLER
r
1 - BLER
Active time:
Inactive time:
Session time:
Activity factor:
Tactive
Tsession
FORMULAE
2.1
NOTATION
k J
2.2
Unless stated otherwise, all quantities and formulae are in standard SI units, not in dB.
Bandwidth (Hz).
hop
Sk
non-hop
Sk
CINR control
Sk
control
CINR Sk
CINR traff
Sk
CINR traff
Sk
G diversity
GSantenna
Diversity gain.
Boltzmanns constant.
pathloss
Sk
Lantenna
Sk
feeder
S
equip
LS
LSantennacorrection
LScell correction
Lindoor
k
hop
S
Feeder loss.
nSnon-hop
N kthermal
EIRP
S
output
S
PSkreceived
N
TRX hop
S
N STRX non-hop
Temperature in Kelvin.
nSnon-hop ).
slot
S
FR
S
CS
S
TSGPRS
SDTX
adj
Straff
Sfrac
k
10
2.3
PSkreceived PSEIRP
1
L
pathloss antenna
Sk
Sk
(2)
indoor
k
where
PSEIRP
PSoutputGSantenna
feeder equip antennacorrection cell correction
LS LS LS
LS
The product
is read from the prediction file. The received signal strength must meet the
Lpathloss
Lantenna
Sk
Sk
DL Thermal Noise
N kthermal k T B k .
(3)
DL Traffic CINR
CINR traff
Sk
hop
non-hop
N TRX hopESk
N STRX non-hopESk
.
BER 1 S
(4)
Where BER-1 is a function which maps bit error rates to values in decibels via the C/I BER conversion
table. For the purposes of the GSM simulator this is a CINR BER conversion table.
Figure 1
Hence, the CINR BER table is used as a means of averaging values in decibels during the CINR
calculation. However, the BER table has a minimum CINR value of -10dB and a maxiumum of 27dB.
The default BER values are plotted in Figure 1.
11
The highest possible CINR value is achieved with a strong signal and no carrier interference. Assume
the maximum received signal strength is 1W and that there is minimal thermal noise. This gives a
maximum CINR value of:
CINR traff
Sk
1
1
150dB .
23
kTB 1.38 10 .293. 200 103
Therefore an equation is required to calculate new BER default values up to 150dB. An assumption is
made that the line between 25 and 26dB can be extrapolated with the same gradient. This gives a
gradient of -2 and an intercept of 42 which yields an equation for BER:
0
-20
-10
10
20
30
40
50
-10
log(BER)
-20
-30
-40
-50
-60
C/I (dB)
Figure 2
nshop
BERCINR Sk
hop
diversity
1 1
E Sk BER G
BER
nshop
(5)
The summation is restricted to the hopping carriers on sub-cell S. The diversity gain is read from the
frequency hopping diversity gain table, shown in Figure 3.
12
Figure 3
The index into the diversity gain table is the number of carriers, n, and is calculated as:
hop
S
otherwise.
1
ant gain
nsnon hop
BER
CINR
Sk
1
nonhop
E Sk
BER G diversity BER 1
hop
ns
(6)
(CINR ) Sk
received
PSk
DTX frac
. J . Jtraff
J
.PJkreceived .u N kthermal
(7)
J S
u , given by
if ,
1
adj
if is adjacent t o
0
otherwise.
13
traff
S
TSslot
.
CS
TS TSGPRS
(8)
Sfrac
N STRX hop
hop
nS
if nShop 0
if nShop 0
14
SNAPSHOT OVERVIEW
The key purpose of a snapshot is to provide us with measures of system load for a particular
distribution of terminals. To obtain these measures of system load, we must calculate time-slot
consumption for all the links in the system. A snapshot involves the following stages:
3.1
The first stage of a snapshot involves creating a random distribution of terminals representing the
offered traffic in the network. The spatial distribution of terminals must be random, but more
importantly it must be unbiased. In other words, it must be reasonable compared to the terminal
density array provided by the user. To see how this is achieved, we need only consider a single pixel
(bin) in the simulation.
Consider a pixel that has a terminal density of D terminal/km2 and an area of A km2, so that the average
number of terminals in the pixel is DA. We note that:
Terminal occurrences within the pixel are independent of each other and are spatially uniform
within the pixel. In other words, a terminal is just as likely to be located at one point within
the pixel as any other point within the pixel.
The probability that two or more terminals are located at exactly the same point within a pixel
is zero. This is simply because there is an infinite number of locations within the pixel.
These imply that terminal occurrence is a spatial Poisson process within the pixel. Therefore the total
number of terminals in the pixel satisfies the Poisson distribution:
P( k terminals )
( DA) k e DA
.
k!
We choose the number of terminals to assign to the pixel by drawing a number from this Poisson
distribution. Doing this at each pixel ensures our terminal distribution is unbiased. Since the sum of
many Poisson distributions is also a Poisson distribution, the total number of terminals in the snapshot
will also be Poisson distributed.
One may note that if the average number of terminals at a pixel is small (DA << 1), then working to
first order in DA,
P( 0 terminal ) (1 DA) ,
P( 1 terminal ) DA .
So one is effectively making a binary decision about whether a terminal should be placed at the pixel.
After creating the random terminal distribution, the terminals are randomly sequenced.
determines the order in which they will be considered during the iterations.
This
Terminals are randomly assigned as being indoor or outdoor, according to their terminal type and the
clutter type in which they reside. Indoor terminals are all given a speed of zero.
15
3.2
TIME-SLOT ITERATIONS
The main task of a snapshot is to calculate the traffic loading of the network. At the beginning of the
snapshot, the system is placed in a state of an unloaded network. This is achieved by making all timeslots available on all sub-cells, i.e. with no terminals connected. Traffic loading is then calculated
iteratively by cycling repeatedly though the list of randomly spread terminals and attempting to make a
connection. The following logic is applied to each terminal:
-
If the terminal is already connected then disconnect it by freeing the time-slots used at the
sub-cell by the terminal.
Recalculate the control and traffic CINR values to account for the new traffic load.
Try and connect the terminal to the network in the most favourable way possible. The
way in which the terminal connects may be different from the way the terminal connected
previously. This difference arises because the system load may have changed since the
previous iteration.
If a connection is possible then connect the terminal. Update the time-slots used at the
sub-cell to reflect the terminals connection. Recalculate the control and traffic CINR
values to account for the altered traffic load.
Several cycles through the terminal list are needed to achieve a stable system load. On the first cycle,
the first few terminals will experience very little interference because the network is not fully loaded.
The lack of interference means that these terminals can use higher data-rate bearers which have high
CINR requirements. As a consequence, these terminals use fewer time-slots on the sub-cell than if they
connected using lower data-rate bearers. As the system becomes loaded, interference increases. On
subsequent iterations, the terminals may achieve lower traffic CINR values and hence connect using
lower data-rate bearers. Lower data-rate bearers require more time-slots than higher rate bearers and
hence consume more time-slots on the sub-cells. After several cycles, the traffic loading no longer
changes significantly. The iterations have converged to produce a plausible picture of served and failed
terminals in the network.
3.3
CONVERGENCE TEST
A good practical measure of convergence is to examine how the traffic load changes between cycles.
The user need only enter one parameter in the convergence criterion for iterations section of the
wizard for this convergence check. After each cycle through the terminal list, the percentage change in
traffic load is noted. If the change in traffic load falls within the specified limit for fifteen iterations
then the snapshot is considered to have converged.
3.4
GATHERING OF RESULTS
The final stage of a snapshot involves gathering results. The information gathered includes sub-cell
information (e.g. time-slot usage), information about the states of connected terminals, and the
failure reasons of terminals which failed to be served.
16
4.1
The connection scenario describes how a terminal connects to the network. It consists of the
following parameters:
Typically, several connection scenarios are available to each terminal. Our snapshot attempts to
connect the randomly spread terminals to the network in the most favourable way possible. Logic is
required for ranking the different scenarios that each terminal may use.
The rules for ranking scenarios during connection evaluation are (in order of decreasing importance):
The connection scenarios for each terminal are evaluated in turn (from most to least favoured) until one
that permits a network connection is found. The scenario employed by a terminal may change each
time it is evaluated in the loading iterations, and this flexibility provides us with link adaptation.
A downlink evaluation is carried out to examine the connection scenario and determine whether a
connection is possible.
4.2
17
4.3
FAILURE REASONS
A connection scenario can fail for one or more of the following reasons:
Low Signal.
The received signal does not meet the receiver sensitivity requirement specified on the
terminal type.
No Cell Time-Slots.
The sub-cell has too few time-slots to serve the bearer.
No Terminal Time-Slots
The terminal has too few time-slots available for the connection.
High Pathloss.
There is no sub-cell on the pixel with a pathloss value sufficient to be considered as a serving
sub-cells.
If all of the connection scenarios available to a terminal fail to produce a connection, then the terminal
is classed as a failure. Each scenario in the list can fail for multiple reasons. Also, different scenarios
in the list can fail for different sets of reasons. All of this makes failure reporting problematic. For
example, consider a terminal with the following scenario list containing only 2 scenarios:
Cell Name
Cell_X
Cell_Y
Cell Layer
CellLayer_A
CellLayer_B
CINR Ctrl
4 dB
20 dB
DL Bearer
12kbps
12kbps
Problems
Low ctrl CINR & traffic CINR
No Cell Time-Slots
In this example, CellLayer_A has a higher priority than CellLayer_B. This makes Cell_X the top
scenario even though it has a worse CINR than Cell_Y. What is the correct reason for failure for this
terminal? There is no correct way of assigning this. The tool records the failure reasons of the top
scenario only. In most cases the top scenario provides the most useful information as to why a terminal
fails. In the above example, the terminal is deemed to fail because of problems with Low Control
Channel CINR and Low Traffic CINR on Cell_X. The terminal contributes to the failure statistics for
Cell_X only, not Cell_Y. Hence the terminal does not affect the failure statistics for No Cell TimeSlots.
18
5
5.1
OUTPUT ARRAYS
ARRAY DEPENDENCIES
All arrays are produced on a per cell-layer basis. Many arrays depend on whether the terminal is taken
to be indoor or outdoor. Indoor arrays use the in-building parameters for the clutter type at each pixel
(i.e. indoor loss and indoor shadow-fading standard deviation).
Coverage arrays can be drawn even if no snapshots have been run, but the user should note that the
arrays then refer to coverage in an unloaded system. To obtain coverage arrays for a loaded system the
user must run some snapshots; the key purpose of running snapshots is to provide measures of traffic
load. The arrays change little after a relatively small number of snapshots have been performed (10s of
snapshots in most cases). This is because only a small number of snapshots are needed to get an idea
of the average loading on each sub-cell.
The following table lists the types of array that are available in the Simulator, and shows some of their
dependencies. Most terms (e.g. Indoor) are self-explanatory.
Fading means the array depends on the standard deviation of shadow fading for the clutter type.
Snapshots done means the accuracy of the array has a strong dependence on the number of snapshots
done, so the array will require 1000s of snapshots to give accurate results instead of 10s of snapshots.
Other Tech means that the presence of the array is dependent on the inclusion of other technology
types in the Simulation. Other technology types can be UMTS and Wi-Fi.
CE=Cell Layer
I=Indoor
T=Terminal
F=Fading
S=Service
R=Reliability
O=Other Tech
n=Snapshots done
CE
X
X
X
X
X
X
Pathloss Arrays
DL Loss
Nth DL Loss
Best Server by RSS
Nth Best Server by RSS
RSS
Nth RSS
All Servers
Coverage Probability
Coverage Probability OK
CINR (Control)
CINR (Traffic + Control)
Achievable Bitrate
X
X
Coverage and Data Rate Arrays
X
X
X
X
X
X
X
X
X
X
Composite Tech Arrays
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
T
X
X
n
CE
X
X
19
PATHLOSS ARRAYS
DL Loss & Nth DL Loss
Dependencies: Terminal, Cell layer, Indoor
These are the downlink losses of the Best Server by RSS and Nth Best Server by RSS. They represent
average values and are therefore calculated with fades of 0 dB.
5.2
These arrays all provide information on coverage levels and coverage probabilities.
Best Server by RSS & Nth Best Server by RSS
Dependencies: Cell Layer
This is the sub-cell that provides the highest (and Nth highest) RSS for the terminal.
Coverage Probability
Dependencies: Terminal, Cell Layer, Indoor, Fading
This is the probability that the Best DL Cell (by RSS) satisfies the RSS requirement specified on the
terminal type. This probability depends on the standard deviation of shadow fading for the clutter type
at the pixel. If this standard deviation has been set to zero, then there are only three possible coverage
probabilities: 0% if the requirement is not satisfied, 50% if the requirement is satisfied exactly, and
100% if the requirement is exceeded.
Coverage Probability OK
Dependencies: Terminal, Cell Layer, Indoor, Fading, Reliability
This is a thresholded version of the Coverage Probability array and has just two values (Yes/No). It
has the advantage of being quicker to calculate than the Coverage Probability array. A value of Yes
means that the downlink coverage probability meets the coverage reliability level specified in Array
Settings Sim Display Thresholds.
CINR (Control)
Dependencies: Terminal, Cell Layer, Indoor
These are the CINR(Control) values corresponding to the best serving sub-cells, i.e. not necessarily the
highest CINR(Control) values.
20
Achievable Bitrate
Dependencies: Terminal, Cell Layer, Service, Indoor
This is the highest bitrate that can be achieved by the terminal based on CINR regardless of system
loading.
5.3
Composite Tech Arrays account for GSM, UMTS and Wi-Fi cells collectively.
Composite: Best Server
Dependencies: Terminal, Indoor, Service, Fading, Reliability, Other Tech
This is the serving cell identity. Wi-Fi is the top priority technology. UMTS and GSM relevant
priorities are as specified in the Service definition. Cells of a specific technology type are ranked by
Signal Strength (Wi-Fi: DL-RSS, UMTS: RSCP, GSM: RSS). The terminals requirements must be
met for the respective technology (Wi-Fi: Required Signal Strength, UMTS: Required RSCP, Ec/Io and
Pilot SIR, GSM: Receiver RSS Sensitivity). The display thresholds are set for each technology type
individually in Array Settings Sim Display Thresholds.
5.4
21
A terminal can fail for multiple reasons. When this occurs, only a single reason is reported when
writing a value at the pixel. This is achieved by sensibly ranking the failure reasons for the terminal
and using the most dominant one. For example, there is no point in indicating a capacity failure for a
terminal if it does not have coverage. In other words, coverage failures rank more highly than capacity
failures.
The failure reason categories are ranked as follows (most dominant failure reasons first). For clarity,
the categories in the legend are shown in the same order.
Success
No covering cells
No valid scenarios
Low Signal
No Cell TS
No Terminal TS
22
The term coverage is used in the classical sense. It looks at the problem of a received signal being of
sufficient strength/quality given that it has been transmitted through a lossy channel with shadow
fading. Coverage probabilities are calculated analytically. This means that coverage plots can be
obtained after running a very small number of snapshots (typically 10s) and plots converge very
quickly. If no snapshots have been run, then coverage plots are still available but they give the
coverage probabilities in an unloaded system.
We will examine the different ways in which shadow fading affects our calculations. There are
essentially three cases to consider.
6.1
NOTATION
We introduce the following notation to represent the probability density function for a normally
distributed random variable with mean and standard deviation .
N (x ; , )
6.2
(x )2
exp
2
2 2
2
Shadow fading is modelled in a snapshot by randomising the pathlosses experienced by the randomly
scattered terminals. Shadow fades are log-normally distributed, and the user specifies the shadow
fading standard deviation for indoor and outdoor terminals in each clutter type. In reality, the fades
between a terminal and the cells that cover it will exhibit a degree of correlation. In particular, a
terminal is likely to have similar fades to cells that are located on the same site. To account for this, the
user specifies two parameters in the Monte Carlo Wizard:
The normalised inter-site correlation coefficient ( c inter ). This is the correlation between
fades to cells on different sites.
The normalised intra-site correlation coefficient ( c intra ). This is the correlation between
fades to cells on the same site.
23
These two parameters must satisfy the constraints 0 c inter c intra 1 . For each randomly scattered
terminal in a snapshot, a set of correlated fades to the covering cells is generated using the following
procedure. All the random numbers mentioned below are independent and normally distributed with
zero mean and unit variance, and is the standard deviation of the shadow fading at the pixel in dB.
The above procedure is performed for each of the randomly scattered terminals at the beginning of a
snapshot. Fades for different terminals are uncorrelated even if they are located in the same pixel.
6.3
Arrays showing the mean level of a quantity (e.g. DL linkloss, RSS, CINR, etc) are calculated with all
fades set to 0 dB.
6.4
RJ .
If sub-cell J has a fade of FJ dB, then the RSS is given by
RJ (10 FJ /10 ) .
We can calculate a coverage probability for the RSS as follows:
Find the fade FJ that causes the RSS to satisfy exactly the RSCP requirement specified on the
terminal type. Call this fade F * . Note that F * may be positive or negative. Any fade bigger
than F * will give an inadequate RSS.
Since FJ is normally-distributed with a mean of 0 dB and standard deviation of dB, the
probability that FJ F * is given by
P( FJ F )
*
N (F
; 0, )dFJ .
F*
This is the probability that the RSS does not meet the requirement.
24