Beruflich Dokumente
Kultur Dokumente
Reference
Guide
2.5.0
AT250_ARG_E2
AFP Reference Guide
Contact Information
Forsk (USA Office) 200 South Wacker Drive L sales_us@forsk.com Sales and pricing information
Suite 3100 { support_us@forsk.com Technical support
Chicago, IL 60606 « +1 312 674 4846 General
USA +1 888 GoAtoll (+1 888 462 8655) Technical support
¬ +1 312 674 4847 Fax
Forsk (China Office) Suite 302, 3/F, West Tower, www.forsk.com.cn Web
Jiadu Commercial Building, L enquiries@forsk.com.cn Information and enquiries
No.66 Jianzhong Road, « +86 20 8553 8938 Telephone
Tianhe Hi-Tech Industrial Zone, ¬ +86 20 8553 8285 Fax (Guangzhou)
Guangzhou, 510665, +86 10 6513 4559 Fax (Beijing)
People’s Republic of China
The product or brand names mentioned in this document are trademarks or registered trademarks of their respective reg-
istering parties.
This document begins with a basic user guide containing a short operational introduction to the AFP process in Atoll. Then
it goes on to summarize most aspects of the practical planning process and provides detailed discussions on certain topics.
It also explains the means to evaluate a frequency plan. Furthermore, a chapter is dedicated to advanced topics and trou-
bleshooting in the end.
The appendices deeply detail the technical aspects of the cost function, the BSIC allocation algorithm, the IM calculation,
and the dimensioning process.
Table of Contents
1 Overview ........................................................................................... 3
1.1 Introduction to AFP ..........................................................................................3
1.1.1 Frequency Assignment as a Cost Minimization Problem....................................... 3
1.1.2 Abbreviations ......................................................................................................... 3
1.2 Architecture......................................................................................................5
7 Appendices ..................................................................................... 63
7.1 Appendix 1: Description of the AFP Cost Function........................................63
7.1.1 Notations.............................................................................................................. 63
7.1.2 Cost Function....................................................................................................... 64
7.1.3 Cost Components ................................................................................................ 65
7.1.3.1 Separation Violation Cost Component .........................................................................65
7.1.3.2 Interference Cost Component ......................................................................................67
7.1.4 I_DIV, F_DIV and Other Advanced Cost Parameters .......................................... 68
7.2 Appendix 2: Interferences..............................................................................70
7.2.1 Using Interferences.............................................................................................. 70
7.2.2 Cumulative Density Function of C/I Levels .......................................................... 70
7.2.3 Precise Definition................................................................................................. 70
7.2.4 Precise Interference Distributions Strategy ......................................................... 71
7.2.4.1 Direct Availability of Precise Interference Distributions to the AFP ..............................71
7.2.4.2 Efficient Calculation and Storage of Interference Distributions ....................................71
7.2.4.3 Robustness of the IM ...................................................................................................71
7.2.5 Traffic Load and Interference Information Discrimination .................................... 71
7.3 Appendix 3: BSIC Allocation..........................................................................73
7.3.1 Definitions ............................................................................................................ 73
7.3.2 Hard Criterion ...................................................................................................... 73
7.3.3 Soft Criterion........................................................................................................ 73
7.3.4 Behaviour............................................................................................................. 73
7.4 Appendix 4: Traffic Capture and Dimensioning .............................................74
7.4.1 Introduction .......................................................................................................... 74
7.4.2 Traffic Map Generation ........................................................................................ 74
7.4.3 Traffic Capture Process ....................................................................................... 74
7.4.3.1 Inputs ...........................................................................................................................74
7.4.3.2 The Engine...................................................................................................................75
7.4.3.2.1 Traffic Distribution ................................................................................................................. 75
7.4.3.2.2 Average Timeslot Capacity ................................................................................................... 76
7.4.3.2.3 Integration ............................................................................................................................. 77
7.4.3.3 Outputs.........................................................................................................................77
7.4.4 Network Dimensioning Process ........................................................................... 78
7.4.4.1 Inputs ...........................................................................................................................78
7.4.4.2 Dimensioning ...............................................................................................................78
7.4.4.3 Outputs.........................................................................................................................78
List of Figures
1 Overview
This document describes every aspect of frequency planning in Atoll, from high level description of the frequency planning
process to the practical level detail. Main topics covered in this document include AFP prerequisites, AFP usage, AFP min-
imisation target and some possible problems that may come up during training.
This document begins with a basic user guide, a short operational introduction to the AFP process in Atoll, and goes on to
summarize most aspects of the practical planning process with detailed discussions on certain topics. It also explains the
means to evaluate a frequency plan available in Atoll. A chapter is dedicated to advanced topics and troubleshooting in
the end.
Four appendices contain in-depth information on technical aspects of the cost function, the BSIC allocation algorithm, the
IM calculation and the dimensioning process respectively. All in all, this document is almost self sufficient with respect to
the use of Atoll AFP.
The Atoll AFP considers a large number of constraints and directives; for example, ARFCN separation requirements be-
tween transmitters, interference relations, HSN assignment methods, frequency domain constraints, a certain fractional
load to maintain etc. Hence, the AFP depends on a variety of input data, such as the interference matrix, neighbourhood
relations, traffic information and so on.
This document not only explains how to use the Atoll AFP, by describing the AFP GUI, but also includes detailed descrip-
tions of the various constraints, directives, and data sources. The primary target of this document is to explain the technical
background of the AFP.
The approach of cost minimization is not only the most common approach to the FAP but probably also the easiest to un-
derstand and control. It provides the user with means of guiding the AFP in its task. For example, by setting the cost of
interference violation low, the AFP will concentrate its efforts on resolving the separation violations.
There are AFP tools in which certain types of objectives are presented as “hard constraints”. If a hard constraint is not sat-
isfied, the AFP does not offer any solution or offers a partial solution (with fewer frequencies and satisfying hard con-
straints). The philosophy of hard constraints vs. soft constraints has nothing to do with the quality of an AFP engine, it is
merely a behaviour convention. In Atoll, we prefer always offering a solution to offering partial assignments or violating
domain limitations. This ensures that you will always get a result when you launch the Atoll AFP. This result will very well
depict the difficulty of the FAP. The cost of this solution will clearly indicate if unacceptable violations have occurred or if
this plan has improved the current frequency plan.
The cost function definition permits you to place as much emphasis as required on certain elements of the cost function.
This manipulation will make the AFP behave as if it were guided by hard constraints, from the optimisation viewpoint, while
retaining its property of being a quality monitor and a hardness-of-assignment monitor both.
1.1.2 Abbreviations
Some abbreviations and terminologies used in the document are listed below:
TRX Transceiver
Broadcast Control CHannel. A term usually employed in Atoll to refer to the TRX carrying this
BCCH
channel.
Traffic CHannel. A term usually employed in Atoll to refer to a TRX carrying traffic with usually the
TCH
same coverage area as the BCCH.
Inner Traffic CHannel. A term usually employed in Atoll to refer to a TRX carrying traffic but usually
TCH_INNER
having a coverage area less than that of a TCH.
HR/FR Half Rate/Full Rate
CS Circuit-switched
PS Packet-switched
HCS Hierarchical Cell Structure
Subcell An entity defined by the pair [TX, TRX Type]
HO Handover
Kbps Kilobits per second
GoS Grade of Service
QoS Quality of Service
KPI Key Performance Indicators
TL Traffic Load
P Probability
C Carrier power (Signal strength)
C/I Carrier to Interference ratio
AFP Automatic Frequency Planner/Planning
DTX Discontinuous transmission
GUI Graphical User Interface
FP Frequency Plan
BBH Baseband Hopping
SFH Synthesized Hopping
NH No Hopping
Mobile Allocation List. In the context of SFH, MAL is the group of frequencies used by the
MAL
frequency hopping TRX.
HCS Hierarchical Cell Structure
AMR Adaptive Multi-Rate
HR / FR Half-rate / Full-rate
CC Concentric Cells
Transmitter Atoll synonym for cell or sector in conventional GSM jargon
FER Frame Erasure Rate
FH Frequency Hopping
DL_PC Down Link Power Control
RRM Radio Resource Management
Synchronised
Transmitters that are synchronised and can, therefore, share the same HSN.
transmitters
A project can be saved in a filename.ATL file or as a database. In both cases, most of the project’s
Data Model
information is saved in database tables. We refer to these tables as the data model.
IM, IMco, IMadj Interference Matrix, Co-channel / Adjacent-channel Interference Matrix
HO Handover or Handoff
FN Frame Number
CDF Cumulative Density Function
TSC Training Sequence Code
FAP Frequency Assignment Problem
# Number of
1.2 Architecture
The Atoll Automatic Frequency Planning (AFP) module is an optional module that enables you to generate frequency plans
for GSM and TDMA networks automatically. The Atoll AFP module can allocate the following parameters:
• Frequencies
• Frequency hopping groups (MAL)
• HSN, MAIO
• BSIC (TSC planning)
• TRX rank (can be used to prioritise the use of good frequencies)
• Performance Indicators at Site/Cell/TRX levels
Atoll works with an open AFP interface. Any AFP built using this interface can be able to allocate the following additional
parameters. Future versions of the Atoll AFP module are planned to assign the following parameters as well:
Atoll AFP implements simulated annealing, taboo search, graph heuristics and machine learning. It manages its time re-
sources to match the users time directive. If allowed enough time, the AFP will employ a major part of this time in “learning”
the network. During the learning phase, the AFP tunes up its internal parameters. Towards the end of the user-defined
time, the AFP switches to a randomised combinatorial search phase.
Remark:
The role of this learning phase is extremely important in order to get good results. You should often let the AFP run over
a night or a weekend by specifying corresponding target time. If you never run the AFP specifying a long time period, it
will never be able to calibrate itself and will always perform from 10 to 70 solutions and stop.
Network learning is performed by executing numerous fast and deterministic instances of the AFP. The one that obtains
the best performance is memorized in the document and is, therefore, the most suitable for the specific network. The next
time an AFP is executed it will start where the learning process ended and it will use the parameter profile of the best so-
lution stored in the document.
Note: The following scenario will demonstrate the usefulness of AFP learning capabilities:
- Create a GSM GPRS EGPRS project and import its network elements and maps.
- Create a copy of “Atoll AFP module” and name it “Atoll AFP module 2”.
- If the network has X transmitters, run “Atoll AFP module 2” for X / 10 minutes to obtain a cost Y. (Short exe-
cution)
- Now run “Atoll AFP module 2” for a longer time (for example, X / 5 hours).
- Another cost, Z, is obtained, which is better than Y (i.e. Z < Y). The network dependent information is mem-
orized in the “Atoll AFP module 2” instance whereas the “Atoll AFP module” instance remains unchanged.
- Now if you perform a short execution with “Atoll AFP module 2”, you can get the improved result (Z) right
away. While a short execution of the “Atoll AFP module” instance will give the initial cost (Y).
- If X / 5 hours is too long, you can perform the “learning” on a small (representative) part of the network.
The Atoll AFP is built based on a specified COM interface designed as a part of Atoll’s open platform strategy. The interface
is designed in such a way that puts aside elements that are not inherent to the AFP process. At the same time, through the
modelling capabilities of the planning tool, the AFP can support complete list of features expected from an AFP.
The figure below gives a better view of interaction of the AFP with other elements in Atoll:
• Specify the AFP module you would like to use and set its parameters,
• Choose the network parameters and AFP performance indicators you want the AFP to allocate,
• Specify the network’s default separation requirements,
• Consult the network’s “Exceptional Pairs” and define other separation constraints for them, and
• Indicate whether interferences are to be included in calculations or not.
The last of the wizard dialogs contains some global parameters that often vary from one AFP instance to another:
The most important option here is the one proposing the two sources of the traffic load information. Traffic load can be read
directly from the subcells table, which could have been filled manually, by the dimensioning process or by a KPI calculation.
You can also specify that the traffic load should be read from the default traffic capture (explained later).
Notes:
1. In case the traffic load is taken from the Subcells table, committed after a KPI calculation, you must be
aware of a certain difference: in the KPI calculation, Atoll divides the captured traffic by the timeslot capac-
ity of the existing number of TRXs, while the AFP requires it to be divided by the timeslot capacity of the
required number of TRXs.
2. The traffic load is artificially increased to 0.1, if it is too low (less than 0.1), in order to maintain the AFP
robust against partial data conditions. Hence, the AFP cannot completely ignore the existence of a fre-
quency in a TRX.
Clicking Validate will start the data verification and storage optimisation aimed at providing fast access to data needed by
the AFP. This stage may generate many warnings for real-life networks (for example, values out of range). These are dis-
played in the Event viewer. It is recommended to revise the network data according to these messages and continue once
all the data are clean and coherent. If a certain message is not clear or self evident, you can always contact Forsk’s tech-
nical support. The figure below depicts the Event viewer with some sample messages:
This means that the value entered in the AFP weight column of the Transmitters table for the transmitter 19941 is invalid.
In the database, this field’s name is “COST_FACTOR”. A value of –2 for the cost factor implies that the AFP should gener-
ate the worst assignment possible for the transmitter. It would be interesting to investigate the origin of this erroneous value
as it may avoid possible errors in the future. Atoll automatically resets this value to 1 in order to avoid such calculation
errors.
This message informs that 3678 subcells were loaded successfully. The next section explains the significance of the term
‘effectively selected’ and why 3678 subcells were loaded and only 6 selected for the AFP process.
Both the RING and the SEL parts of the network are loaded. It is important to know which subcells are loaded as the cost
is calculated for all loaded subcells. The RING part is frozen for all assignments (BSIC, HSN, MAL, MAIO and channels).
The SEL part may be assigned some parameters but only the ones specified in the dialog above. For example, if the user
did not select BSIC, it will not be assigned.
In addition to the generic freezing options above, there are some finer freezing options available in the data structure:
1. Individual transmitters can be frozen for channel (and MAL), HSN and/or BSIC assignment.
2. Individual TRX’s can be frozen for channel (and MAL) assignment.
In an Atoll project, it is strongly recommended to avoid TRX’s without channels. For this reason, never create transmitters
automatically if there are no channels to assign to them. Therefore, if the user does not ask for MAL/MAIO assignment, all
SFH subcells are considered frozen and no TRX will be created for them. The same occurs when only a MAL/MAIO as-
signment is requested. In this case, all NH and BBH subcells will be considered frozen and no TRXs will be created.
Note: See Developer Reference Guide for details on the TO_ASSIGN and FROZEN assignment states available
in the AFP API.
Example 1: When partial IM info exists, we can see that 9 transmitters out of 12 do not have any
interferers.
Example 2: When complete IM info exists, observe that the IM topology is more or less normal.
This dialog also lets you define a generator initilialisation number. This number serves as a directive of randomness for
the AFP process being launched. If the generator initialisation is set to 0, the AFP will be fully random. An integer other
than 0 will define a given deterministic sequence for the AFP process. Each generator initialisation number (other than 0)
corresponds to a deterministic sequence. Therefore, each AFP instance launched with the same generator initialisation
number will yeild the same results.
However, you must keep in mind that all the AFP computations are deterministic in the start, independent of the generator
initialisation. The AFP must be allowed to compute during the target time to observe the effects of randomness.
Notes:
1. Since the method chosen by the AFP depends on the target time provided, you might not get the same
results using the same generator initialisation number if the defined target times are different. Therefore, to
actually get the exact same results from the AFP process, you must define a certain target time and a cer-
tain generator initialisation.
2. The AFP may be perfectly deterministic during a portion of the target computation time (5 - 15%). During
this period, the randomness seed will have no effect on the solutions. If you want to see the effect of ran-
domness, let the AFP calculate until the end of the target time, or set a shorter target time.
Important: If only a short time is specified, the full optimisation potential of the AFP will not be utilised.
The window below opens when the AFP is started, and displays information about the AFP process:
The Progress section of this window in the top left displays the target time allocated to the AFP, the time elapsed and the
number of AFP solutions that have been evaluated so far.
The calculation status window in the top right gives some general information about the current solution in real time. This
display depends on the selected AFP module. This section lists the status of the current solution, the initial cost, the cost
of the current best solution, the cost of the previous solution and whether the previous solution was kept or rejected.
The Event viewer has been made accessible through the AFP progress dialog in order to help the user keep track of all
the important warnings and messages generated before and during the AFP process. This also enables you to export
these messages as an AFP log file.
If a solution is kept, a corresponding message appears in the Event viewer. Double-clicking the message in the Event view-
er will open a dialog with the full details of this message, which will look something like the following figure.
The Best Frequency Plan Costs section displays the current values of modifiable and total costs, and their respective sep-
aration components. This section also displays the total weighted Erlangs of the network concerned in the AFP process,
i.e. the total cost of a 100% interfered frequency plan). It gives a general idea of how good the cost of a certain frequency
plan is. The cost of any solution remains between 0 and the Network Weighted Erlangs. The cost is as better as it is closer
to 0.
Apart from this information, this section also contains a table listing the initial frequency plan and all the AFP solutions kept
so far sorted in ascending order of cost. This table can display:
• Modifiable costs
• Total costs
• Frozen costs
• Summed components
• Main components (separation violation cost component, interference component and modified TRX component)
• Additional taxes (corrupted, missing or out of domain TRXs)
For detailed description of modifiable and non-modifiable parts of the total cost, please refer to section 4.4.1.11.
After the AFP is allowed to compute solutions and try to optimise the network for a while, the AFP progress dialog would
look somewhat like this:
Using the buttons available in the Plan comparison section in the bottom right, it is possible to visually compare the initial
frequency plan and the current best solution (with the Best Plan column in the AFP cost details table checked). Clicking
these buttons opens dialogs containing graphs corresponding to ’Cost Distribution on Frequencies’ and ’Usage Distribution
on Frequencies’.
Cost ( f ) = ∑ FL ( i ) × Cost ( i )
i ∈ TRXs using f
Where, FL(i) is the fractional load of frequency f in the MAL of i, and cost(i) is the AFP cost of TRX i in Erlangs.
You can pause or stop the AFP process any time to check the current best solution, and resume optimising the network
after you have checked it. Pausing the AFP process opens the AFP results window with the current best solution results
listed.
This window contains a table listing all the assigned resources.These resources and related items (transmitters, subcells)
are coloured differently to indicate different reasons, such as:
Positioning the cursor over a resource in the table displays the reason for its colour in a tool tip.
The results window displays all the results of the AFP session. It is possible to only display some of the results by checking/
un-checking the relevant choices in the Display options menu. You can choose to display the results related to:
• Cells (BSICs)
• Subcells (HSNs)
• TRXs (Channels/MAL, MAIO) and related separation violations
Selected AFP performance indicators (AFP TRX ranks, and total and separation costs at TRX, subcell, transmitter and site
levels) will also be available in the results window. These AFP performance indicators are also available to export. You
can choose whether to display the AFP indicators in the results as separate columns. The Show AFP Indicators command
in the Display options menu controls the display of AFP TRX ranks, and total costs and separation cost components at
TRX, subcell, transmitter, and site levels.
As the network had been loaded according to both the items to assign and the ones they relate to, it is possible to display
the frequency plan of either:
It is also possible to display detailed information about separation constraint violations, i.e. the co-channel and adjacent
channel collision probabilities for relevant TRXs. You can choose to display these separation constraint violations through
the Display options menu.
The Separation violations column lists each each type of separation constraint violation realted to a given TRX, i.e. excep-
tional pair, co-transmitter, co-site, or neighbour. Another column titled ’With the TRX’ contains a button for each type of
separation constraint violation. This caption of this button shows the TRX with which the separation constraint violation
occurs. Clicking this button takes you to the corresponding TRX row in the table. Right-clicking a row with a separation
constraint violation opens a Separation Constraint Violations context menu, which opens a dialog mentioning the reason
of violation when clicked. For example:
Use the Commit button to assign the allocated resources and AFP performance indicators. The resume button permits re-
suming the AFP optimisation from where it stopped the last time.
Note: At the bottom of the AFP results window, messages related to the last solution are displayed. These may
list problems as well.
The AFP result dialog is a non-blocking dialog. It enables the user to access other Atoll windows while the AFP is still pend-
ing. Thus, it is possible to view other data or warning/error messages in the Event viewer (for example, the history of AFP
solutions). From this stage, it is possible to commit, to resume or to quit the AFP. It is good practice to keep a report through
the export option before resuming the AFP. The user can also partially commit some of the results as explined next.
The dialog examples below depict a case where removing a TRX eliminates a separation constraint violation on neigh-
bours. Once a TRX is manually removed from the resulting plan, separation violations are recalculated (may take a few
seconds). If the TCH TRX of transmitter Site36_3, causing neighbour separation constraint violations, is removed from the
sample frequency plan below, the resulting frequency plan has no neighbour separation constraint violations on the TCH
TRX of transmitter Site36_1.
It is possible to specify the action to be taken with each TRX individually, or globally delete all TRXs with separation viola-
tions. It is also possible to mix the old plan and the new plan. Though this is not recommended, since it can cause interfer-
ences of which the user might be unaware.
The dialog examples below depict how this operation can be carried out.
The Delete the TRX option implies that the resulting frequency plan will not respect the number of required TRXs. In the
above example, note than the neighbour separation constraint violations at transmitter Site36_1 vanished once the TCH
TRX at Site36_31 was deleted.
The aim of this tool is to find the TRXs in the currently proposed frequency plan that cause constraint violations of any of
the four following types:
1. Co-transmitter
2. Co-site
3. Neighbour
4. Exceptional Pair
Once it finds the TRXs that satisfy the criteria, it sets their corresponding values to Delete the TRX in the Channel Assign-
ment column of the AFP results window.
This tool lets you resolve any type of constraint violations for different types of TRXs, control or traffic. You can also define
a threshold of co-channel and adjacent channel collision probabilities. This restriction will only set those TRXs to Delete
the TRX, which have a co-channel or adjacent channel collision probability higher than the threshold you defined.
In this example, frequency 11 is not a good choice since it is used as a neighbour co channel. Frequencies 10 and 12
present similar characteristics.
On the other hand, frequency 14 is a good one and can be possibly allocated. None of the frequencies {13, 14, 15} are
allocated at the selected transmitter of at its neighbours.
Therefore, it is mandatory that the AFP user acquires a minimum level of knowledge regarding Atoll data model. This chap-
ter familiarises the user with the essentials of the data model and depicts their relations with the AFP.
• The unfrozen part of the AFP cost is 0 and the AFP stops due to this fact.
• There appear to be close frequency reuses in the resulting frequency plan.
This means that the problem is too “easy” for the AFP and the user would like to create a more difficult IM in order for the
AFP to have a more difficult problem to solve.
The best method to accomplish this is to increase the cell edge reliability and recalculate the IMs. When the reliability re-
quirement is elevated, a larger part of the standard deviation is reduced from “C” when calculating the C/I for each IM entry.
The user should also verify that the standard deviation is properly defined in all clutter classes and its default value. This
verification is more important in the case of Atoll documents converted from older versions or connected to a database.
When calculating an IM, or when generating an interference study, HCS layers are used in generating service zone maps,
the basis of these calculations. If two transmitters belong to different layers, they can both serve the same pixel even if
received signal from one is much stronger than the other’s. For equal HO margins, more HSC layers mean higher overlap-
ping levels in the network. As the overlapping level increases, the constraint level in the IM and the amount of interference
in an interference study also increase.
Note: Be sure to study the priority mechanism in your network, both in the re-selection process and in the han-
dover process. Define the corresponding HCS layers once you know its working. When using a traffic
model, make sure that there are a few levels of mobility in order to model high speed / low speed mobility
behaviours.
3.1.3 Subcells
Subcells are defined as a group of TRXs in the same transmitter. Two subcells of the same transmitter can request fre-
quencies from different domains, require different C/I qualities, have different downlink power offsets and even have dif-
ferent Radio Resource Managements (RRM). Different RRMs can lead to different service zones under the same cell.
Subcells are crucial for modelling concentric and dual band transmitters. In these cases, the TRXs belonging to the “inner”
subcell serve traffic within a limited zone.
The other important property of concentric cells is the fact that a downlink power offset is associated with each subcell.
The inner subcells can have higher DL PC implying that the frequencies assigned to the inner zones will interfere less with
other transmitters. Concentric cells permit a higher reuse pattern between inner zones, providing up to 40% increase in
capacity.
Atoll can fully exploit this increase in capacity since it calculates interferences between subcells. It uses the power offset
and the C/I threshold that defines the subcell boundaries. Furthermore, it is also possible to define separation constraints
at subcell level.
Atoll provides the possibility to define these thresholds at subcell level resulting in maximum flexibility and option to support
a mixture of old and new equipment. Moreover, the safety margins corresponding to these values can be defined in the
AFP cost definition (see C/I weighting in the next chapter).
The benefit of this method is to facilitate Atoll in exploiting the fact that a larger number of TCH channels can be assigned
with quality requirements lower than the BCCH quality. This results in less constraining interferences and an easier and
faster assignment.
3.1.4 TRXs
Atoll’s TRX table enables the following:
• Support of an external ID space of the TRXs of a transmitter (important for import and export utilities).
• MAL / channel at TRX level.
• MAIO at TRX level.
• Fine freezing: The user can freeze specific TRXs in an unfrozen transmitter.
The TRX table does not contain an “active” field. Therefore, all TRXs in it should contain a valid frequency or MAL and are
all considered to be on air. It is better to remove an entire TRX record than removing only the frequency or MAL from its
channels list.
Note: When freezing channels, keep in mind that the MAIOs are not frozen.
• ARFCNs
ARFCN is the method employed by the GSM/DCS standards to enumerate 200 kHz frequency carriers.
• Frequency Bands
Frequency Bands are subgroups of ARFCNs. Different equipment may be limited to different frequency bands (BTS, MS,
…). In addition, propagation models use the central frequency of the band for calculating propagation.
• Frequency Domains
Domains are used for managing the usage of the Frequency Bands. For example, an operator may use frequencies 1 to
50 while the other uses 52 to 100. Splitting the band on channel usage basis is of great importance as well (BCCH frequen-
cies, TCH frequencies, Hopping layer).
• Domain Groups
Domain groups are used for further managing the use of the frequencies in a domain. For example, f1 and f2 can be as-
signed at the same transmitter if and only if they belong to the same group. Another frequent use for groups is in the MAL
assignment.
In Atoll, a domain is defined as a union of groups. It points to a frequency band and must be included therein. The AFP
respects domain limitations at subcell level.
• The channel list in the Transmitters table is the intersection of all channels appearing in the TRXs of a transmitter.
• The hopping mode of a transmitter is the hopping mode of it’s default traffic carrier (the TCH TRX Type)
• The frequency band of the transmitter (the one used by the propagation model to deduce the central frequency),
is read from the domain of the BCCH subcell of the transmitter.
Atoll considers the low level to be the accurate source of information. For example:
• Atoll will automatically update the TRX table if the channel list of a transmitter in the transmitter table is changed.
• The frequency band of a transmitter cannot be edited.
These redundancies provide some additional features (for example, grouping transmitters according to the frequency
bands).
On the other hand, there is a chance of mistakes and bugs which may damage a redundancy in the ATL file. Therefore, it
is recommended that the audit tool be used from time to time in order to fix these problems (right click on the Transmitters
folder, choose Audit from the Subcells menu).
For example, if two subcells are neighbours and at the same site, their associated separation requirement will be according
to the co-site separation rules. And, if this separation requirement is not fulfilled, their separation violation costs will be
weighted by the co-site weight.
The GSM standard requires this desired behaviour but does not specify any amplification level. It is recommended to be
sure that the physical equipment in the network support this value. The value of this parameter is used in the AFP when
extracting the interference caused by an adjacent channel, and in Atoll in interference and C/I studies.
It might be a good idea to use a safety margin for this parameter and set it to 16 dB, for example.
Notes:
1. Rank = 1 is the best rank.
2. TRX Rank is the corresponding field in the TRX table.
As it is during an AFP process that frequencies and MALs/MAIOs for different TRXs of a subcell are chosen, the AFP tool
stores and manipulates the information about TRXs in good and in bad conditions.
If you choose AFP Rank indicator to be allocated when starting an AFP session, each cost improving solution will go
through a TRX rank assignment. If no improving plan is found, TRX rank will be assigned for the initial plan (like BSIC).
TRX ranking within a subcell is performed on the basis of TRX costs.
In many cases of MAL/MAIO assignment, only one or two of a TRX’s MAIOs violate separation constraints. Therefore, a
higher ranking will be assigned to the MAIO violating the separation constraints.
TRX ranks may be required by the OMC in order to optimise the spectral efficiency. In some networks, a part of the deci-
sion-making process at the OMC may be transferred to the BTSs when this information is available. Even if such a “smart”
system exists, it might be better to know the TRX ranks in advance to improve predictability and consistent behaviour.
Apart from these uses, AFP TRX ranks can be used in post-AFP optimisation. For example, once you perform AFP, you
can freeze all TRXs with ranks less than or equal to X. So that a new AFP instance will concentrate on a smaller subset of
the most interfered TRXs in the most loaded subcells.
A TRX will not be considered frozen for TRX Rank assignment if and only if it is selected for AFP allocation and has not
been frozen at Transmitter level or by the AFP launch Wizard.
In order to be able to compute and display these results, you must add AFP_COST and AFP_SEP_COST fields (of type
SINGLE) to the TRX, Subcells, Transmitters and Sites tables. AFP_COST field and AFP_SEP_COST field correspond to
the total cost and separation cost component respectively. These AFP performance indicators are available in the list of
AFP performance indicators to be computed available when launching the AFP tool.
The AFP cost assignment to the TRXs, subcells, transmitters and sites is carried out at the same time as the TRX rank
assignment. Once a frequency plan is committed, the next instance of the AFP can concentrate more on the problematic
TRX/subcell/transmitter/site to improve results. Another use of this feature can be to automatically limit the modification
scope to the problematic cells/sites. This feature can deliver a significant quality gain.
1. To reduce the required number of TRXs where they are least needed and spectrum not available.
2. To indicate the least loaded TRXs, since they are less important and interfere less with others TRXs.
3. To reduce the constraint level of the IM (for example where interferences are limited to low density surface).
4. To provide an accurate quality estimation of the resulting frequency plan.
The first point is currently implemented through a dimensioning model, which is explained in this chapter. Moreover, it will
be available as an AFP option in one of the future versions (i.e. the AFP will optimise the decisions so as not to respect the
required number of TRXs. In other words, it will perform a spectrum oriented dimensioning). The second role is carried out
by the traffic loads. In order to understand traffic loads better, traffic capture is also described subsequently. The third point
is explained alongwith the description of the IM and the last point is detailed in chapter 5.
The Atoll traffic model is quite advanced. To gain familiarity with the concepts of user profiles, environments, services, mo-
bile types, terminal types etc. a new user should refer to the Atoll User Manual. These traffic model entities can be used to
benefit from all possible capacity gaps in a network. The simplest application here, would be to use a clutter weight oriented
model. The more advanced models and techniques of creating traffic maps, based on traffic-by-transmitter etc., are also
explained in detail in the Atoll User Manual. This chapter provides with the basic know-how on creating the simplest model
using clutter information and clutter weights.
Both traffic maps are stored in the document and can be exported. An exported vector map is smaller than a raster one.
The AFP combines this traffic capture with the number of required TRXs and their timeslot configurations to generate traffic
loads (assuming the AFP will create the required number of TRXs indicated in the subcell table).
The dimensioning process reads the basic information contained in the traffic capture to find out the number of TRXs need-
ed to support a user defined blocking rate, HR ratio etc. (See Appendix 4 for details).
The KPI calculations combine traffic capture with the current number of TRXs in the network and their timeslot configura-
tions to generate current traffic loads.
Important: Keep in mind that the required number of TRXs is the number of TRXs required to carry a given traffic. This
is the number of TRXs (usually) calculated through the dimensioning process. The number of existing
TRXs is the current actual number to TRXs at a transmitter.
See Appendix 3 to understand further why traffic loads and interference information are not combined together in Atoll.
This section details the use of this new feature in various scenarios.
Tip:
Check neighbour allocation before running the AFP. Often a bad neighbour relation definition causes poor frequency
plan performance.
The configuration presented below is recommended in order to use the resulting neighbour importance in the AFP.
AFP can be launched once the results of the neighbour allocation have been generated and committed.
• OMC HO statistics
• Test mobile data measurements (which ignore interferences between non-neighbours)
• Other
As with any other source of information, it is the user’s task to prepare and import this external data. The units of the neigh-
bour importance are probabilities and are expected to reamin between 0 and 1.
The neighbour importance of the original neighbour assignment is probably more reliable than the one calculated using
path loss calculations.
1. Export the current neighbour relation into a file called AllCurrentNei.txt using the generic export feature available
through the context menu of the table,
2. Export all the relations for which there are reliable neighbour importance into a file named
AllCurrentNei_Importance.txt,
3. Import the file AllCurrentNei.txt into the neighbour exceptional pairs so that the existing neighbour allocation is
forced (usual operation for extending an existing allocation),
4. Run automatic neighbour allocation in order to extend your neighbour relations and/or assign importance where it
was not already assigned. To keep important values lower than X%, all Max% values in the importance part of the
dialog should be kept less than X. For example, if X is 50%, the configuration shown below can be used,
As can be observed in the figure above, all new neighbour relations have weak importance values.
The screenshot below shows that the neighbour relations now comprise old neighbours with a higher importance and new
neighbours with a lower importance automatically calculated by Atoll.
This optimisation can currently be carried out with the help of the dimensioning model. In future versions, it may be avail-
able directly through the AFP.
Running the AFP once more can return an improved frequency plan, as the following example shows:
The above experiment proves that the capacity difference between the two networks is very low (first column, around 2
Erlangs). This means that the reduction of 5 TRXs leads to a very minor decrease in capacity. This is due to the fact that
this was done by dimensioning considerations rather than other possible considerations.
The AFP generates a better plan after this decrease in the number of TRXs. The AFP cost units are Erlangs, therefore, we
can compare the 2 Erlangs lost because of capacity to the 10 Erlangs gained because of better AFP cost.
Note: It is possible to set a “Maximum number of TRXs” in the Transmitters table. You can copy and paste the
current demand to this column, thus forcing the dimensioning process to respect the current state of the
network as an upper bound. This possibility is a handy in all possible cases of difficult frequency allocation.
Atoll AFP gives each separation violation a cost equivalent to a certain amount of interference, making it possible to sum
both costs and minimize their sum. For example, the user can define that a separation violation of 1 “costs” the same as
x% of interfered traffic. This is weighted by the type of violation (co-transmitter separation violations have higher impact
than neighbour separation violations). Through this equivalence, It is possible to sum separation violation and interference
costs that share a common unit, i.e. percentage of interfered traffic.
Following this principle, all other cost elements are also calculated in the same manner, the cost of Missing TRXs, the cost
of corrupted TRXs, the cost of a TRX assigned out-of-domain frequencies and the cost of changing a TRX’s assignment.
Example:
• Imagine a network with 6 TRXs, all having a separation constraint of 1 with each other (i.e. 6 nodes, 15 Edges).
• The following 3 cases demonstrate the way the AFP calculates the cost of an allocation.
Two TRXs have good assignments Only one TRX has a good assignment No TRX has a good assignment
• Atoll AFP prefers Case 1 by default. Nevertheless, it can be configured to opt for Case 3.
The parameters that control the capability of Atoll AFP to be more Edge-oriented than Node-oriented are explained next.
But, before this explanation, following are the three main advantages of the Node-oriented approach:
The Node-oriented approach is an important feature of the Atoll AFP and provides a tighter correspondence between the
AFP cost and the network quality.
• If TRX α is corrupted, the tax of being corrupted is added to the cost, and multiplied by T(α).
• If TRX α is missing (the required number of TRXs and the actual number of TRXs being different), the tax of miss-
ing TRX is added to the cost, and multiplied by T(α).
• If TRX α has out-of-domain frequencies assigned to it, the tax of out-of-domain frequency assignment is added to
the cost, and multiplied by T(α).
• Otherwise, the separation cost, the interference cost and the changing load of this TRX are summed up (probabi-
listically) and added to the cost, and multiplied by T(α).
• If this sum is very small, it is discarded (see the Quality Target section)
Here, T(α) is an estimation of the traffic Erlangs using TRX α weighted by the AFP weight for this TRX.
The user can fully control the AFP cost target by determining the value of the cost function parameters. Some of these
parameters belong to the data model, e.g. “Maximum MAL Length” and “Minimum C/I”, while others are present in the
specifc AFP GUI. Appendix 2 explains how to find each of these parameters.
Note: The AFP cost is the cost of the entire loaded network, not only the cost of the selected or non-frozen TRXs.
In many cases, the AFP is authorized to change only a part of the network. Therefore, the part of the cost
corresponding to the non-frozen part of the network and the part of the cost corresponding to the frozen
part of the network are indicated.
Let Sij denote the required separation between two transmitters in a network. If f1 is assigned at i and f2 at j such that
( f 1 – f 2 ) < S ij , this means that the separation constraint is not satisfied. A separation constraint can be violated strongly or
weakly:
For example, the pair of frequencies (1, 2) violates a separation requirement of 3. The pair of frequencies 1 and 3 violate
this requirement as well but is still a better solution than (1, 2) and, therefore, should have a lower cost.
Frequencies that are part of a MAL with a low fractional load and that disobey a separation constraint, should not be weight-
ed the same as in non-hopping separation violation. In fact, the separation component is weighted by the burst collision
probability, which is the multiplication of the victim’s fractional load and the interferer’s fractional load.
Example: Let us consider the following simple case of a network comprising two TRXs in the same cell. The first, TRXi,
has a MAL denoted as MALi. It is interfered by TRXk having MALk. TRXi and TRXk have a separation requirement of 2.
Their MAL lengths are respectively 5 and 4. Unfortunately, one of their frequencies is the same (i.e. separation = 0), while
all other frequencies are correct. For the case of a co-channel violation when the required separation is 2, the cost of the
separation violation is 90%, as shown in the dialog bellow:
Since only one channel of each TRX causes interference, with length(MALi)=5 and length(MALk)=4, the collision probabil-
ity is 1/20. Hence, the cost to consider is divided by 20, i.e. 90/20 = 4.5% for each TRX.
In addition, since the two TRXs have different MAL lengths, they have different interferer diversity gains: a gain of 1.4 for
MAL length of 5 and a gain of 1.2 for MAL length of 4 (see Appendix 2 for details).
( 1.4 ⁄ 10 )
Applying interference diversity gain of 1.4 dB means that the cost will be divided by the value: 10 ≈ 1.38 . For TRXi,
this will give 4.5% / 1.38 = 3.25%.
1 90
For TRXk, the cost to consider will be ------ ------------------------ = 3.41% .
20 10 ( 1.2 ⁄ 10 )
Now, in order to get the exact contribution to the separation cost component, these values are multipilies by the traffic load
(Erlangs / timeslot) and by the number of traffic carrier timeslots in each TRX. Assuming the traffic load to be 1 and that
each TRX has 8 traffic carrier timeslots, we will get (8 x 3.25 + 8 x 3.41), i.e. about 0.5 Erlangs for the two TRXs together.
Note: In this example, the AFP weight was assumed to be 1, the traffic loads were assumed to be 1, no DTX was
involved, no other interference or violation was combined with the above, the global separation cost was
set to be 1, and the co-transmitter separation weight was set to 1 as well.
The solution to this problem is provided by a probabilistic approach: all different costs are considered as bad events and
are combined as if these events were independent.
The probabilities of events in this example are p(Violation) = 0.3 and p(Interference) = 0.4. The cost of the two together is
given by:
Let P1, P2, ….Pn be the violation probability costs of the given TRX (one for each of its n violations).
Let Pn+1, Pn+2, ….Pm be the interference probability costs of the given TRX (one for each of its (m-n) interferences).
n
⎛ ⎞
The separation cost of this TRX will therefore be: ⎜ 1 – ( 1 – P i )⎟
∏
⎜ ⎟
⎝ i=1 ⎠
m+1 n
⎛ ⎞ ⎛ ⎞
⎜ ( 1 – P i )⎟ – ⎜1 – ( 1 – P i )⎟
The additional cost of this TRX will be ⎜ 1 – ∏ ⎟ ⎜ ∏ ⎟
⎝ i=1 ⎠ ⎝ i=1 ⎠
The interference cost uses the “min C/I” value, defined at subcell level, for which it may have precise pair-wise interference
information. It may apply various gains to this C/I quality target due to frequency hopping and/or DTX.
In certain cases, creating more TRXs will not only generate interferences for the newly assigned TRXs but also for other
TRXs that, otherwise, would have correct assignments. This parameter enables the user to tune the AFP in its tradeoffs
between respecting the number of required TRXs and optimising quality.
Note: The Atoll AFP always assigns as many TRXs as the “Required TRXs” field indicates. The user can only
decide regarding the better plan, i.e. the previous plan having a better quality but some missing TRXs or
the new plan having lower quality but having all required TRXs assigned. In future versions, the AFP will be
capable of optimising these decisions as well.
In some cases, fixing the assignment on a group of corrupted TRXs will not only result in an interfered assignment for these
TRXs but also for many other TRXs that, otherwise, would have correct assignments. It is for this reason that this tax is
higher than 100%.
Note: A TRX is corrupted not only if it contains frequencies that are out of domain but also when:
- An NH TRX has a MAL with more than one frequency
- A TRX has no channel at all
- A group constrained SFH TRX is assigned a MAL that is not strictly a group of its domain
- An SFH TRX has no MAIO, or no HSN, or has an out of domain HSN / MAIO value.
In future versions, different corruption reasons will have different tax levels and the assignment of a corrupted TRX will not
always be ignored.
If a TRX is assigned out-of-domain frequencies (channels) but has correct ARFCNs, it will have dual influence on the cost:
Note: If SFH and group constrained subcells have out-of-domain channels, and are frozen, the frozen TRX will be
ignored altogether.
Modifiable and non-modifiable parts of the total netwok cost are linked in concept with the definition of the AFP scope, sec-
tion 2.3. Four groups of transmitters can be defined with respect to AFP:
- If BSIC assignment is required, then all the second order neighbours (neighbours of neighbours) will be includ-
ed in RING.
The total cost of the network corresponds to SEL+RING. It contains the modifiable as well as the non-modifiable parts of
the network costs.
The modifiable part of the total cost of the network corresponds to SEL. However, this cost does not include costs corre-
sponding to frozen entities of transmitters in SEL.
The non-modifiable part of the total cost of the network corresponds to RING. It includes the costs corresponding to frozen
entities of transmitters in SEL.
In each instance of an AFP process, there might be entities frozen by the user. In addition to the generic freezing options,
there are finer freezing options available in the data structure:
1. Individual transmitters can be frozen for channel (and MAL), HSN and/or BSIC assignment.
2. Individual TRX’s can be frozen for channel (and MAL) assignment.
The figure below depicts the Cost tab of the AFP properties dialog:
Test case:
A network with 90 transmitters in total, 15 frozen transmitters and sum of required TRXs = 257.
Only 193 good TRXs were already allocated.
64 TRXs should be created / newly allocated with as little influence as possible on the other 193 TRXs.
For a cost of changing a TRX = 1 AFP changed only 98 TRXS
For a cost of changing a TRX = 0.3 AFP changed only 129 TRXS
For a cost of changing a TRX = 0.1 AFP changed only 139 TRXS
Therefore, the user must specify a margin around which a “slope” is created, as illustrated in the figure below. This figure
corresponds to an interference relation between two TRXs. It describes the distribution of traffic according to C/I conditions.
It depicts the effect of 3 different quality requirements on the interference cost of a co-channel frequency reuse.
It can be observed that, when a low quality (C/I > 8 dB) is required, less traffic is considered as interfered than for a high
quality (C/I > 11 dB).
The option ‘C/I >10 dB + 2dB margin’ has the advantage of not being too strict on one hand, and yet trying to achieve high
quality if possible. It is visible from the above figure how it integrates the different traffic classes into the interference cost.
The figure below shows the Separation Weights tab of the AFP properties dialog.
Other tabs of the AFP module properties dialog are more advanced. Please refer to the Atoll User Manual for more infor-
mation on AFP module properties tabs.
Atoll interference studies are geographic studies that analyse each point on the map. For each point, an interference study
estimates the carrier power and then sums up the interference powers taking burst collision probabilities, DTX, and traffic
load into account.
As the AFP works with interference matrices, it is limited to coarse estimations of interference combination and loses
knowledge of the geographical location of interference events. It is due to this reason that the interference study output is
much more accurate than the AFP cost.
The column “Erlangs (based on traffic load)” is available in the interference study report as seen above. This column uses
a traffic model similar to the one used by the AFP:
• It spreads the traffic of each subcell (#TRX x traffic load x # timeslots) on the service zone of the subcell.
• Then, it sums up the interfered traffic (in Erlangs) of each interfered TRX.
This means that it is a TRX based estimation of interference and is much more accurate than any other tool available in
Atoll. In order to be able to use this option, you must check the “Detailed results” option when specifying conditions for an
interference study.
For each TRX, Atoll retains a map of all the pixels that do not comply with the quality threshold (one map for each TRX).
The threshold can be global or dedicated to each subcell (from the Subcells table). The maps can be visualised by the user
and contribute to the overall statistics. The column mentioned earlier weighs the “bad” surface of each TRX map by the
traffic carried by the TRX as seen by the AFP. Therefore, it is the most appropriate tool for frequency plan interference
evaluation.
In case of large networks, where it might be easier to compare these results using MS-Excel, you might not get a 100%
match between these reuslts and small variations in the order may exist.
The different information seen in this screen shot is part of the report obtained by creating several instances of an interfered
zone study. The interfered zone study whose properties dialog is shown, is the closest to the default configuration. Its in-
terference definition references the subcell quality threshold.
In the two other interfered zones studies, the global threshold for the minimum and maximum TRX C/I to be included in
each TRX’s map were fixed at the reference values. These studies show that there are more “weak interferences” (11.3
Erlangs) than “strong interferences” (6.3 Erlangs). The “weak interferences” being 8 < C/I <= 12 and the strong interference
being C/I <= 7.
The consistency check tool is described in detail in the Atoll User Manual. A quantified summary of separation constraint
violations is available. It makes the evaluation of a frequency plan much easier than before. The section below demon-
strates how this evaluation can be carried out.
The FISFE expression is part of our open architecture AFP strategy. There are more than one AFP integrated with the Atoll
platform and, therefore, it is interesting to know the frequency plan FISFE values. The answer must be independent of the
AFP used to generate the plan.
The value “7.5” is arbitrary. It is used to upscale the FISFE(FP) value to “close to Erlang” units. On the other hand, the
weights specified for each component reflect the idea of their relative importance.
N1 through N7 are now available in the summary of the consistency checking tool.
• N1 is the number of TRXs subject to a co-channel reuse violation where the separation violation concerns two non
SFH TRXs.
• N2 is the additional number of TRXs subject to an adjacent-channel reuse violation where the separation violation
concerns two non SFH TRXs.
Note: If a TRX is counted in N1 it will not be counted in N2.
• N3 is the additional number of TRXs subject to a co-channel reuse violation where the separation violation con-
cerns two TRX where at least one is SFH.
Note: If a TRX is counted in N1 or N2 it will not be counted in N3.
• N4 is the additional number of TRXs subject to an adjacent-channel reuse violation where the separation violation
concerns two TRX where at least one is SFH.
Note: If a TRX is counted in N1, N2 or N3 it will not be counted in N4.
The pair-wise violation counting is less important than TRX counting and therefore has a low weight coefficient. On the
other hand, it cannot be ignored, since in 1/1 SFH plans, all TCH TRXs will have violations with neighbours and yet it would
be required to minimize the volume of these violations. N6 and N7 are defined as following:
• N6 is summed over all pairs subject to a co-channel reuse violation, where at least one TRX in the pair is SFH and
where the probability of a burst having a co-channel violation is summed.
• N7 is summed over all pairs subject to an adjacent-channel reuse violation, where at least one TRX in the pair is
SFH and where the probability of a burst having an adjacent-channel violation is summed.
In the following example, the values of N1 through N7 are {0, 23, 62, 35, 123, 7.695, 2}
The relative load is the same as the number of TRXs if no synthesised hopping is involved. When synthesised hopping
exists, the frequency load is the sum of 1/(MAL size) of all the TRXs using this frequency.
1. When the FAP is easy, the AFP reaches a 0-cost solution and stops immediately. If it was instructed to use the
minimum possible spectrum, the AFP will use the smaller ARFCNs more than the larger ones (and will leave the
largest ARFCNs untouched, for future use). Otherwise, the AFP will try to spread the spectrum usage. By default
this directive is free for AFP tuning. Therefore, the AFP will not create a uniform frequency usage distribution when
the FAP is easy.
2. If the FAP is somewhat difficult, the frequency usage distribution will be somewhat non-uniform. In order for it to
be uniform, the corresponding directive must be set in the Spectrum tab of the AFP properties dialog.
3. If the FAP is difficult, all frequencies will be used and the allocation heuristics will result in a balanced allocation.
Lastly, when the FAP is extremely difficult (many separation violations, for example), an unbalanced allocation
may result because of the reason explained in chapter VII.1, concerning the fact the AFP cost is a TRX (node)
based cost.
4. AFP can also create non-uniform frequency distributions due to the domain range effects described below.
Advice:
Do not expect a balanced distribution of frequencies in every case. Sometimes the best solution requires an unbal-
anced assignment.
In a domain of 1 to 7 in a typical hexagon network with 6 neighbours for each transmitter, one solution can be a 1/4 pattern
using only the frequencies 1, 3, 5 and 7, while 2, 4 and 6 will not be used at all. Another solution for the same can be a 1/
7 pattern. Although, if the separation constraints between neighbours are 2, the 1/4 pattern (based on odd number frequen-
cies only) will give a better result.
The success of the assignment depends on the definition of groups in the domain. Each MAL length, represented by one
or more groups in the domain, is supposed to fulfil the conditions:
• that there must be a number of other groups having the same length, and
• that on the whole they cover the domain as much as possible.
For example, a domain containing groups of lengths 3, 5 and 8 will be a badly planned domain if there are many groups of
length 3, many of length 8 and only one of length 5.
If restricted to such a domain, the AFP will not produce an optimum plan. On the other hand, by adding a few more groups
of length 5, the quality can be much improved. Another solution could be to simply remove the single group of length 5.
When many groups are defined, the quality is almost as good as with free assignment.
Notes:
1. Currently, the AFP always assigns the same MAL to all TRXs within a subcell.
2. The “group constrained” assignment mode is applicable for SFH only. In NH and BBH, the assignment
mode is always free.
1. If two subcells have different domains, they cannot belong to the same atom.
2. If two subcells have different limitations on “Max MAL Length”, they cannot belong to the same atom.
A warning is generated when HSN assignment directives contradict with these restrictions.
An important feature is the possibility to force the AFP to always assign the same MAL among subcells of an atom. Fur-
thermore, improved results can be obtained by post-relaxation of this constraint, performed on a carefully selected and
small subset of transmitters.
Since fractional load cannot always be obtained, this parameter is considered as a guide rather than a constraint. When it
can be obtained, AFP chooses either a MAL length 1/δ times longer than the number of TRXs in the biggest subcell of the
atom or a MAL length 1/δ times longer than the sum of all TRXs in the atom. These are called “the short MAL strategy” and
“the long MAL strategy” respectively. The user can choose between the two in the MAL tab of the properties dialog. The
value of the fractional load parameter can also be edited and, furthermore, it can even be automatically calibrated by the
AFP.
According to the adapted convention on HSNs for BBH TRXs, the AFP allocates different HSNs to the BCCH TRX and
TCH TRXs. The 1st HSN corresponds to timeslots 1 through 7 of the BCCH and TCH TRXs, and the second HSN corre-
sponds to the timeslot 0 of the TCH TRXs only. The second HSN is used in studies.
The user can control the HSN allocation so that it performs one of the following:
The BSIC allocation is compliant with the BSIC domains of transmitters and the strategy indicated in the BSIC tab of the
AFP module property dialog. Either the algorithm selects a minimum number of BSICs in the related BSIC domain (Minimal
option), or it chooses as many BSICs as possible while keeping them evenly distributed in the related BSIC domain (Max-
imal and homogeneous option).
Notes:
1. In the Atoll AFP, the directions of neighbour relations are all considered equivalent, so that the BSIC alloca-
tion is subject to harder constraints. Let us consider the following two examples:
A A
N N N N
B C B C
From the operational point of view, the first case is much more critical in terms of (BSIC, BCCH) collision.
However, both cases are considered as violations of hard constraints by the AFP and error messages are
generated (if the domain limitation provokes violations).
2. The soft criteria add additional constraints due to interference and adjacent reuse between BCCH chan-
nels. Warning messages are generated when soft criteria are not fulfilled.
3. Atoll AFP also considers Training Sequence Code collisions for synchronous networks. The AFP tries to
avoid having the same TSC when interference exists (i.e. TCH interference, not only BCCH reuse).
• The first is available in the GUI and protects Atoll’s tables from NULL and out of range values, non-existing refer-
ences and other inconsistencies. The deficiency of these checks is that these can be partially bypassed by auto-
mated import procedures and external database manipulations.
• The second comprises Network Validation processes as described in the User Manual. The setback of this check
is that the user can ignore it altogether.
• The third check is available when launching the AFP. As the AFP puts together information collected from many
different sources, it is quite possible that some elements be conflicting and/or missing. Before launching the AFP,
a validation process is launched that controls the possibility to launch the AFP.
If possible, missing / incorrect data are changed and default values are used (with warnings for each in the Event viewer).
If any serious problem is detected, an error message appears and access to the AFP is denied. The table below lists the
most common data limitations enforced in order to protect the frequency planner.
1. Neighbour allocation
2. Network dimensioning
3. Prediction studies
4. Interference calculation
To avoid confusion, it is recommended that the user be aware of service zone model being used. In fact, the user should
decide a unique service zone definition to use throughout a project.
Advice:
Before running the AFP, select the definition of the service zone to use for the AFP and all Atoll prediction studies
intended for AFP results analysis.
Moreover, for imported interference matrices, there is no means to check whether the imported files are a complete source
of information. The user could very well calculate interference matrices for a part of the network, export it, and later import
it as interference data for the entire network. This would result in an interfered frequency plan.
Advice:
1. Check that the options in the Atoll interference study are consistent with the AFP interference calculation
settings.
2. Verify that the small IM report does not indicate the existance of many non-interfered transmitters.
3. If AFP interferences have been imported, check that they are consistent with the service zone of Atoll inter-
ference study.
The “All servers” option implies that a transmitter is assumed to serve the region where its signal is stronger than a mini-
mum reception threshold. This means that more than one transmitter can be serving a given point. The service zone is
based on the minimum reception level “C”.
The “Best server” option implies that a transmitter is assumed to serve the region where its signal is the strongest among
all signals stronger than a minimum reception threshold. In other words, where C/N > 0 (C being reception level and N
being the strongest neighbour reception level). If a margin M is defined, then the condition translates to C/N > – M. There
may be more than one serving transmitters at a point but only one best server.
By tuning the minimum reception level and the margin value, any desired server overlapping can be achieved. Large over-
laps will result in a denser interference relation, more neighbours and a rather pessimistic analysis of C/I levels or interfer-
ences. It might be interesting to manipulate the overlapping zones study and the coverage by transmitter study to
determine the correct service zone model.
Coming back to the question of choosing between the two models, since both “C” and “N” values contain errors, the better
model would be the one with the least error. The choice depends on the correlation between the errors in “C” and in “N”. If
these errors are geographically correlated, the “C/N > – M” model should be preferred. Moreover, the error in “C” and the
one in “N” are not evenly distributed: the further you are from the site, the bigger the error is. Therefore, when “N” is much
stronger than “C” it is probably more reliable as well. For these reasons, it is recommended to use the “C/N > – M” model.
As a final argument, it should be kept in mind that the C/N > – M model is bound by the C model.
Advice:
Use a "Best server with margin" coverage study as service zone for both AFP and Atoll studies dedicated to analyse
AFP results.
The fact that “C/N > – M” is bounded by the “C” model might incur misunderstandings. For example, increasing the margin
will not essentially increase the service zone, but change nothing because service zones are limited by coverage and not
by other best servers.
Note: For simplicity, hierarchical layers have been ignored in this discussion.
Advice:
Select "Subcell min C/I threshold" instead of "Specified min threshold" in the Atoll interference study Condition tab.
Since the calculation zone limits the influence area in the interference studies (and IM calculation), all subcell service zones
are, therefore, limited to the calculation zone. Hence, the transmitters at the borders will have very small service zones. All
of the traffic corresponding to these transmitters will be considered as if concentrated in these few pixels and their interfer-
ence matrix entries will be calculated so.
Advice:
Use a focus zone inside the calculation zone in order to focus the study on properly modelled transmitters. Preferably,
use calculation zones that are geographically isolated. See the example below:
• Launch the AFP during X * 30 minutes (half of the time), then stop it and commit the results (if good).
• Freeze all TRXs in the network.
• Find the areas that generate problems. For example, some sites with separation violations.
• Unfreeze the worst 10 sites.
• For each such site, unfreeze 2 – 4 neighbouring transmitters.
• Run the AFP for an additional X * 30 minutes (the remaining half of the time).
AFP is also capable of committing cell level and TRX level quality indicators into the data structure. This makes the selec-
tion of hot spots a much easier task.
7 Appendices
7.1 Appendix 1: Description of the AFP Cost Function
7.1.1 Notations
The notations listed hereafter are used to describe the cost function:
For example:
- When i is NH, A i, g = 1 ⇔ g is a single member group containing one of the frequencies assigned at TRGi.
When we talk about “TRXs of i using g”, and in the case of BBH, then there are |g| such virtual TRXs, each using the entire
group g and having a virtual MAIO [0, |g| - 1].
- When i is SFH, A i, g must be ≤ #TRX i . A i, g = n ⇔ g is the set of frequencies assigned to n TRXs of TRGi.
We assume all the groups assigned to TRGi to have the same length.
• TSi denotes the number of timeslots available for each TRX in TRGi.
• TLi is the traffic load of TRGi. This parameter may either be calculated during dimensioning or be user specified.
• TSUi refers to the downlink timeslot use ratio (due to DTX) at TRGi.
• CFi is the cost factor of TRGi (AFP Weight field).
• QMINi is the minimum required quality (in C/I) at TRGi.
• PMAXi is the percentage permitted to have quality lower than QMINi at TRGi.
• REQi corresponds to the required number of TRXs at TRGi.
A communication uses the group g in TRGi if its mobile allocation is g. The probability to be interfered is denoted by
P i, i', g ( A ) (i’ is the TRX index). Different TRX indexes may have different MAIOs. P i, i', g ( A ) is a function of the whole fre-
quency assignment. The precise definition of the term “to be interfered” is provided afterwards. The probability penalty due
to violating a separation constraint is P i, i', g ( A ) . It is a function of the whole frequency assignment as well.
ATOM ( i ) ≡ ATOM ( k ) i and k are synchronised, have the same HSN, the same MAL length and the same hopping mode.
NH TRGs or BBH TRGs are always in separate atoms. If two TRGs interfere but are not in the same atom, these can be
taken as unsynchronised. The quality of unsynchronised TRGs is a function of all possible frequency combinations. For
synchronised TRGs, pairs of frequencies emitted at the same time are known.
The cost function Ψ is reported to the user during the AFP progress with the help of its 5 components: Ψ mis , Ψ sep ,
Ψ comp , Ψ corr and Ψ dom .
where,
Ψ comp represents the additional cost component (interference, cost of changing a TRX)
Note: Do not let this form of representation mask the fact that the cost function is a TRX based cost function.
Ψ mis = ∑ ( MIS_TRX i × λ ) × TL i × CF i × TS i
i ∈ TRGs
Ψ corr = ∑ ( CORR_TRX i × Ω ) × TL i × CF i × TS i
i ∈ TRGs
Ψ dom = ∑ ( DOM_TRX i × ω ) × TL i × CF i × TS i
i ∈ TRGs
⎛ ⎞
⎜ ⎟
⎜ ⎟
⎜ δ' i, i', g ( A )⎟ × TL i × CF i × TS i
Ψ sep = ∑ ⎜ ∑ ⎟
i ∈ TRGs ⎜ ⎟
⎜ g∈2
ARFCN
⎟
⎝ i' ∈ TRXs of i using g ⎠
⎛ ⎞
⎜ ⎟
⎜ ⎟
⎜ δ'' i, i', g ( A )⎟ × TL i × CF i × TS i
Ψ comp = ∑ ⎜ ∑ ⎟
i ∈ TRGs ⎜ ⎟
⎜ g∈2
ARFCN
⎟
⎝ i' ∈ TRXs of i using g ⎠
• λ is the cost value for a missing TRX. This value can vary between 0 and 10. The default cost value is set to 1 and
can be modified in the AFP module properties dialog.
• CORR_TRX i is the number of corrupted TRXs for the subcell i.
• Ω is the cost value of a corrupted TRX. This value can vary between 0 and 10. The default cost value is set to 10
and can be modified in the AFP module properties dialog.
• DOM_TRX i is the number of TRXs, for the subcell i, having out-of-domain frequencies assigned.
• ω is the cost value of a TRX with out-of-domain frequencies assigned. This value can vary between 0 and 1. The
default cost value is set to 0.5 and can be modified in the AFP module properties dialog.
If i’ is valid, the algorithm evaluates the cost of a valid TRX. This cost has two components, δ' i, i', g ( A ) and δ'' i, i', g ( A ) .
If the option “Take into account the cost of all the TRXs” available in the AFP module properties dialog is selected, then,
Or if the option “Do not include the cost of TRXs having reached their quality target” available in the AFP module properties
dialog is selected, the algorithm compares P' i, i', g ( A ) + P'' i, i', g ( A ) with the quality target specified for i, P MAX :
Then δ' i, i', g ( A ) = P' i, i', g ( A ) and δ'' i, i', g ( A ) = P'' i, i', g ( A ) .
Otherwise,
P' i, i', g ( A ) is the same as δ' i, i', g ( A ) (separation violation probability penalty) and P'' i, i', g ( A ) the same as δ'' i, i', g ( A ) (com-
plementary probability penalty due to interference and the cost of modifying a TRX) in most cases. These are explained in
detail in the next sections.
Let SEP_CONSTRi, k denote the required separation constraint between TRGi and TRGk. Let Cost s, z denote the user
defined separation penalty for a required separation “s” and actual separation “z”. SEP i, k, v is used instead of
Cost SEP_CONSTR as abbreviation.
i, k, z
Note: The AFP module properties dialog takes probability percentages as inputs while this document deals in
probability values.
ξ ii'kgg'k' is considered to be the effect of a separation violation on the i' th TRX of TRGi assigned the group g, caused by
the k' th TRX of TRGk assigned the group g' .
γ denotes the overall weight of the separation violation cost component. This value can be between 0 and 1, set to 1 by
default. It can be modified in the AFP module properties dialog.
γik represents the weight of the specific separation constraint between i and k. This specific weight depends on the type
of separation violation and follows the following priority rule:
• Exceptional pairs
• Co-transmitters
• Co-site
• Neighbours
For example, if a pair of subcells are co-site and neighbours at the same time, they will be considered as co-site because
higher priority. Hence, γik of these subcells will be the weight of co-site relations. If only a neighbour relation exists be-
tween two subcells, then γik will be further weighted by the neighbour relation importance. The value of γik remains be-
tween 0 and 1. The default weights of each type of separation are available in the Separation cost tab.
If ATOM ( i ) ≠ ATOM ( k )
∑ SEP i, k, f – f'
f∈g
f' ∈ g'
Then ξ ii'kgg'k' = γ × γik × --------------------------------------------- , which is same for all values of k.
g × g'
If ATOM ( i ) = ATOM ( k )
∑ SEP i, k, g υ – g' τ
f_n ∈ { 0, 1, ..., F_N – 1 }
Then ξ ii'kgg'k' = γ × γik × -----------------------------------------------------------------------------------
-
F_N
In the above equations, F_N ( g ) is the number of frames in the MAL g. F_N ( g ) = g .
Note: Since F_N ( g ) = F_N ( g' ) , we shortly denote the two as F_N .
th
While υ = ( f_n + MAIO A ) modulo F_N and g υ is the υ frequency in g,
i, g, i'
th
And τ = ( f_n + MAIO A ) modulo F_N and g' τ is the τ frequency in g’.
k, g', k'
In addition, frequencies belonging to a MAL with a low fractional load, and breaking a separation constraint, should not be
weighted equally as in a non-hopping separation breaking case. Therefore, the cost is weighted by an interferer diversity
gain.
ˆ 1
G i, k, g, g' = -------------------------------------------------------------------
( 0.1 × SEP_GAIN ( i, k, g , g' ) )
-
10
The separation gain, denoted by SEP_GAIN ( i, k, g , g' ) is basically a function of the MAL length (and, of course, of the
hopping mode). With frequency hopping, the effects of DTX and traffic load become more significant (due to the consider-
ation of the average case instead of the worst case). For this reason, it is possible to consider these effects in
SEP_GAIN ( i, k, g , g' ) through the relevant option available in the Advanced tab of the AFP module properties dialog.
Without this option, the SEP_GAIN ( i, k, g , g' ) is:
I_DIV ( g ) is the user defined interferer diversity gain (dB) for a given MAL length. It is used in P i, i', g ( A ) definition as well.
On the other hand, if this option is selected, the SEP_GAIN ( i, k, g , g' ) becomes,
( 2 + ASYN_GAIN ( i, k, g' ) )
SEP_GAIN ( i, k, g , g' ) = I_DIV ( g ) +⎛0.5 × TSU_GAIN ( k ) × min ⎛ 10, 4 + ⎛ ( 2 + I_DIV ( g ) ) × ------------------------------------------------------------------------⎞ ⎞ ⎞
⎝ ⎝ ⎝ 4 ⎠⎠⎠
1
Where TSU_GAIN ( k ) = log 10 ⎛ -------------------------------⎞ ,
⎝ TL k × TSU k⎠
ˆ
More than one separation violations may exist for a TRX. Many “small” G i, k, g, g' and ξ' ii'kgg' have to be combined to form
one cost element, the P' i, i', g ( A ) . This is done through iterating over all violating assignments and by summing up an equiv-
alent to the probability of not being violated while considering each separation violation as an independent probability
event. This sum is naturally limited to 100% of the TRX traffic, and is given by,
⎛ ⎞
⎜ ˆ ⎟
⎜1 –
⎜
∏ ( 1 – ξ ii'kgg'k' × G i, k, g, g' ) ⎟
⎟
P' i, i', g ( A ) = ⎜ ⎟
k ∈ TRGs
⎜ ⎟
⎜ g' ∈ 2
ARFCN
⎟
⎜ k' ∈ TRXs of k using g' ⎟
⎝ ⎠
In the above formula, if ( k = i ) , then ( k' ≠ i' ) , so that interference with itself is not taken into account.
Note: Interference histograms are described in User Manual (GSM GPRS EGPRS project management, GSM
GPRS EGPRS network optimisation, GSM GPRS EGPRS generic AFP management). Interference histo-
grams can also be exported to files. For further description, refer to chapter Appendix 2: Interferences of
this document.
• The QMINi C/I quality indicator corresponds to the accumulated interference level of all interferers while the C/I
interference histograms correspond to pair-wise interferences.
• Both QMINi and the histograms correspond to a single frequency. In case of a MAL containing more than one fre-
quencies, interferences on several different frequencies of a MAL must be combined.
This estimation, presented below, is the simplest possible as it solves the first problem by linear summation and truncation
at the value of 1 and it solves the second problem by averaging and adding the two diversity gains:
Hereafter, α denotes the global weight of interference cost component. This value can vary between 0 and 1 and is set to
0.35 by default, which can be modified in the AFP module properties dialog.
Let MAIO A be the j’th MAIO of A k, g' , where j is one of the { 0, 1, ..., A k, g' – 1 } TRXs.
k, g', j
As said earlier, in case of BBH, we consider g' virtual TRXs, the jth TRX has the MAIO j.
Similar to the definition of ξ ii'kgg'k' , ξ' ii'kgg'k' is defined as an interference event. ξ' ii'kgg'k' is the effect interference on the
i' th TRX of TRGi assigned the group g, caused by the k' th TRX of TRGk assigned the group g' .
If ATOM ( i ) ≠ ATOM ( k )
C
Probability ⎛ ----- < Q_UB i, k, f, f'⎞
⎝ I ik ⎠
Then ξ' ii'kgg'k' = ∑ --------------------------------------------------------------------------------
g × g'
f ∈ g, f' ∈ g'
If ATOM ( i ) = ATOM ( k )
Then,
C
⎧ Probability ⎛ ----- < Q_UB i, k, f, f'⎞ ⎫
⎪ ⎝ I ik ⎠⎪
ξ ii'kgg'k' = ∑ ⎨ -------------------------------------------------------------------------------
⎪ F_N
-⎬
⎪
f_n ∈ { 0, 1, ..., F_N – 1 } ⎩ ⎭
Where,
f = gυ ,
f' = g' τ ,
⎧ ⎫
⎪ ⎪
⎪ ⎪
⎪ ⎪
Therefore, we have, P'' i, i', g ( A ) = 1 – ⎨ ( 1 – P' i, i', g ( A ) ) ×
⎪
∏ ( 1 – ξ ii'kgg'k' ) ⎬ – P' i, i', g ( A )
⎪
⎪ ⎪
⎪ k ∈ TRGs ⎪
⎩ g' ∈ 2
ARFCN ⎭
k' ∈ TRXs of k using g'
In the above formula, if ( i = k ) , then ( k' ≠ i' ) , so that interference with itself is not taken into account.
The sum is limited to 100% of the TRX traffic. INTERF_GAIN ( i, k, g , g' ) is quite similar to SEP_GAIN ( i, k, g , g' ) . The
only difference is the frequency diversity gain, F_DIV ( g ) , added to SEP_GAIN ( i, k, g , g' ) .
1. Non-linearity of Frame Error Rate (FER) with respect to average C/I conditions and MAL length.
2. Interference Diversity Gain. This factor represents that the effect of average negative effects over user geographic
location are directly proportional to the MAL length.
3. Frequency Diversity Gain. This factor models the gain due to diversity of multi-path effects and should be applied
to the interference cost component only.
4. The fact that long MALs with synthesized hopping permit discarding the worst case estimation and include a gain
due to DTX and low traffic load at the interferer end.
The Advanced properties tab shown in the figure below facilitates modelling these effects.
The Interference Diversity Gain table lists the values of I_DIV provided as a functions of MAL length. This gain is applied
to the interference cost component and to the separation constraint violation cost component. Therefore, it provides a
means to model the non-linear FER effects and interference diversity both. The default values in this table correspond to
the curve y = 2 × log 10 ( x ) . This equation generates values somewhat lower than empirical best-found values (this is be-
cause we prefer a slightly pessimistic cost function to be on the safe side).
The other table contains the F_DIV values, which are the same as the I_DIV values by default.
In Atoll, both the IMco and IMadj are represented by this Cumulative Density function This implies that each query for the
probability to have C/I conditions worse than X dB requires a single memory access: the co-channel interference probabil-
ity at X dB. In order to deduce the adjacent interference probability value, Atoll looks up the cumulative density function at
the value corresponding to X - Y dB, Y dB being the adjacency suppression value. The following example may be helpful
in further clarifying this concept:
Example: Let [TX1, BCCH] and [TX2, BCCH] be the interfered and interfering subcells respectively. The service areas for
both have been defined by Best Server with 0 dB margin. The interference probability is stated in percentage of interfered
area.
Figure 7.2: The cumulative density of C/I levels between [TX1, BCCH] and [TX2, BCCH]
In this case, we observe that the probability for C/I (BCCH of TX2 effecting the BCCH of TX1) being greater than 0 is 100%
(which is normal because TX1 is the Best Server). The probability of having a C/I value at least equal to 31 dB is 31.1%.
For a required C/I level of 12 dB on the BCCH of TX1, the interference probability is 6.5% (as this requirement is fulfilled
with a probability of 93.5%).
Note: The subcell power offset does not enter the calculation results in the .clc file. It is added later by the AFP
interface. On the other hand, its influence on the subcell service zone is taken into account in the .clc file.
All these factors cannot be pre-calculated since it is the AFP that determines the MAL length and the MAL frequencies.
By not mixing any of the elements above, the interference information keeps its original probability units and is easier to
check and validate. Therefore, the user spends less time on interference recalculations than in the case of a two-entry ma-
trix (where “everything” is included).
1. The mixed option: The interference information contains the traffic information as well. In this way, each IM entry
will contain the quantity of traffic interfered if a co-channel / adjacent channel reuse exists.
2. The separated option: The AFP has separate access to traffic load information and to interference probabilities
(As in Atoll).
Knowing the difference between the two alternative solutions explains why the second strategy has been opted for for Atoll.
However, in detail, this has been done because:
• Option 2 is a superset that contains option 1. But option 1, being a subset, does not contain option 2 (i.e. once the
information are mixed they cannot be separated).
• It does not create any overhead (the size of the additional information is negligible compared to the size of the IM).
• It helps keeping the unit definitions simpler.
A third option also exists. Though, this option is so practically useless due to its inefficiency. It consists in mixing IM and
traffic but still keeping the traffic in its isolated form. This is again a bad idea because of the unit definition and the variety
of IM sources. It involves less benefits than the option chosen in Atoll.
The hard criterion is easier to satisfy but must not be violated since it causes handover failures. It is based on the second
order neighbour relation and BCCH co-channel reuse.
The soft criterion uses interference information as well and expands the BSIC reuse prohibition over adjacent-channel
BCCH use. It aims at inducing a larger (BSIC, BCCH) reuse distance. When unable to satisfy this criterion, Atoll BSIC al-
location algorithm will use the interference and separation relations to choose the “least interfering” BSIC to assign.
7.3.1 Definitions
Here,
• Let Pred_Nei_Of_Nei(X, Y) be true iff there exist Z such that Pred_Nei(X, Z) and Pred_Nei(Y, Z).
• Let Pred_Co_BCCH(X, Y) be true iff X and Y both have the same BCCH.
• Let Pred_Adj_BCCH(X, Y) be true iff X and Y both have adjacent BCCH frequencies.
• Let Pred_Int(X, Y) be true iff X interferes with Y or Y interferes with X.
7.3.4 Behaviour
As obvious, the soft constraint is much stronger and has a higher probability of not being respected.
For any criterion not respected, the AFP issues a warning message counting the number of times the AFP was blocked
due to each of the two criteria. The Consistency checking tool considers the cases where the hard criterion is not respected
as errors. The previous algorithm (version 2.2.1) had a weaker soft criterion and had a tendency of using fewer BSICs.
For a realistic test-bench network (using only one NCC), following BSIC usage results were obtained:
• Traffic Capture
• Network Dimensioning (and Key Performance Indicator calculation)
Traffic Capture in Atoll is defined as the process of reading raw traffic data from the traffic maps and integrating them to
generate traffic demand for each subcell. In order to perform a Traffic Capture, Atoll collects traffic information (in the form
of Services, Mobility types, Terminal types) from traffic maps and network parameters/criteria (HCS priorities, service zone
model, HR/FR activity percentages, other compatibility criteria, etc.) from the traffic model. Then, it distributes this traffic
according to the criteria over the network’s subcells, and generates and assigns the calculated traffic demand to each
subcell. Traffic Demand is defined as the number of Circuit Switched/Packet Switched traffic Erlangs/kbps in a subcell.
The results from this process are ready to be utilized in the next step.
Network Dimensioning is the process that determines the number of TRXs per subcell required to carry the traffic assigned
to it while respecting the QoS/GoS criteria defined. The traffic captured by the network’s subcells in the previous step is
the main input to this process along with the Quality of Service (or Grade of Service for Circuit Switched traffic) criteria
defined by the user. Assigning a large number of TRXs to a subcell may guarantee the QoS required but will make the
generation of a good frequency plan more difficult. This will also increase the infrastructure cost that depends directly upon
the quantity of equipment required. Therefore, network dimensioning is a more sophisticated process than a mere estima-
tion. Additional outputs of this process are the Traffic Load values (used by the Automatic Frequency Planners) and the
Key Performance Indicators (KPIs).
7.4.3.1 Inputs
The following inputs are required for performing a traffic capture:
The traffic model data contain the requirements for a certain class of traffic. Numerous classes of traffic can be created
through combinations of [Services, Mobility types, Terminal types]. For example, a Packet Switched service requiring a
certain kbps data throughput, allowing a certain maximum number of simultaneous timeslots to be used, with a minimum
blocking probability, together define a minimum Quality of Service criterion. When this service is accessed by a user with
certain mobility through a certain type of terminal device, this constitutes a particular class of traffic.
Traffic maps can comprise many such traffic classes with user densities for each geographically distributed over the entire
network or a part of it. Different traffic maps, raster, vector and live data, can be overlaid as shown in the figure below.
During the traffic distribution, Atoll combines or integrates the traffic data read from these multiple layers of traffic maps to
generate the traffic per bin.
The next section describes how these data, traffic information, compatibility criteria, and requirement criteria, are utilized
in distributing and assigning this traffic to the network’s sectors (subcells, to be precise).
Once the Best Server per HCS layer prediction study has been performed in the background, Atoll proceeds to distributing
the traffic over the different HCS layers and between the subcells of each layer.
In Atoll, a user can define any number of HCS layers as required and assign priorities to them. A possible, rather probable,
priority structure could be:
Note: In Atoll, the priority levels increase directly as the integer assigned. This means that priority 3 is higher than
priority 2, which is higher than priority 1, and so on.
Atoll starts by assigning traffic to the HCS layer with the highest priority and then moves down the priority scale. There are
two levels of control for filtering the traffic for each layer:
• Global control
• Local control
• Global Control
A user can define each HCS layer to have a certain priority in order to control its importance at the time of traffic distribution.
The priority parameter can hence be termed as one of the global control parameters that guide Atoll to start with a certain
HCS layer. Each HCS layer is assigned a maximum speed limit. Any user with mobility higher than the maximum allowed
on a certain layer cannot be allocated to it. Therefore, all traffic with speeds higher than that permissible on an HCS layer
will not be assigned to it. Next, each layer can have a certain operating frequency. Although this parameter is managed at
the TX level, a network can very well have, for example, a GSM 900 layer and an 1800 layer. Only the traffic compatible
with the operating frequency band of a layer will be distributed over that layer. A network can also have, for example, a CS
GSM layer and a PS EDGE layer. This technology compatibility criterion is also taken into account when distributing traffic
over layers. This parameter is also at the level of transmitters in Atoll.
• Local Control
The local control parameters enable much finer tuning at the level of subcells. Each subcell in a network can be assigned
a ‘Target Traffic Overflow’. This is the percentage of traffic that is allowed to overflow from this subcell to the lower priority
layer beneath it. The figure below gives an idea of the effect of this parameter over traffic distribution.
In the example above, when the Target Traffic Overflow for the microcell layer is defined to be 0, the entire traffic (compat-
ible to the micro layer, of course) covered by this cell will be carried by it. On the other hand, if there is a percentage of
traffic allowed to overflow from the micro layer to the layer underneath (10% in this example), Atoll will be permitted in this
case to assign this overflowing traffic to the macro layer. This parameter enables fine tuning of the network’s traffic distri-
bution policy. This may be useful in order to avoid over-dimensioning the micro layer, or to avoid reaching the upper limit
on the maximum number of TRXs that a sector can accommodate.
To be a bit more precise on the two compatibility criteria, i.e. the frequency band and technology, this compatibility com-
parison is in fact performed at the sector level. Atoll compares whether the traffic to be assigned is technologically com-
patible with the higher priority sector before handing it over to the lower priority one. For example, a user with a GSM
terminal using the 900MHz band cannot be assigned to a sector that operates on the 1800Mhz band. Similarly, a terminal
accessing a PS service using the EDGE technology cannot be allocated to a sector that can only perform CS functions.
This function enables the distribution of traffic between sectors of the same layer when there are more than one servers
providing coverage to that traffic. As, for example, in the figure below, there is an overlap between the coverage of A and
B that corresponds to a certain HO margin (fixed by the user). The traffic under this overlapping region is covered by both
servers. Therefore, if the traffic complies with the compatibility criteria for both sectors, it will be equally distributed between
the two. Similarly, when there are more than two servers for traffic in a bin, it will be equally distributed among all the po-
tential servers of that bin.
7.4.3.2.3 Integration
Atoll integrates the demand in terms of # TSL from the above results to generate the resulting total demand per sector for
each service. This means that the CS traffic demand in Erlangs and PS traffic demand in kbps are converted into a total
demand for each sector. This traffic demand per sector is later utilized when performing the network dimensioning process.
7.4.3.3 Outputs
The outputs of this entire process are CS and PS traffic demands in Erlangs and kbps respectively, the average CS and
PS traffic demand in terms of # TSL, and the average traffic demand in terms of # TSL. The Traffic Capture process gen-
erates an item called ‘Traffic Capture X’, that is required as input to the dimensioning process.
7.4.4.1 Inputs
The input to the dimensioning process is the output of the traffic distribution process described earlier. Once the traffic has
been distributed over the network’s layers and sectors, the dimensioning process can be launched. Apart from the traffic
demand, the dimensioning process also takes into account the minimum required QoS and GoS criteria defined for each
type of service, the upper and lower limits on the # TSL for each service, and the upper limits on the number of TRX sup-
ported by each sector.
7.4.4.2 Dimensioning
The dimensioning engine in Atoll converts the traffic demand into the required # TSL, and eventually the required # TRX,
following the QoS and GoS criteria. For example, for a CS service, let’s say voice, allowing a minimum GoS of 2%, the
traffic demand in Erlangs can easily be converted into # of TSL required to carry that traffic demand while respecting the
required GoS through the Erlang B and Erlang C formulas. The type of model, either Erlang B or Erlang C, can be selected
by the user.
The computation of # TSL required to carry a certain PS traffic demand is more complicated than the simple conversion
process described above. PS traffic implies a more complex definition of Quality of Service than the simple Grade of Serv-
ice of CS traffic. It implies, apart from a certain blocking probability, a certain data throughput (kbps) to be maintained with
a maximum allowable delay. These parameters that indicate the PS traffic quality requirements are known as Key Perform-
ance Indicators or KPIs.
7.4.4.3 Outputs
To abridge the numerous results generated at the end of a dimensioning process, we can say that the principal result is
always the # TRX required for each subcell, and eventually each sector. Another chief output of the dimensioning process
is the Traffic Load. More specifically, the Atoll Network Dimensioning process provides the # TSL required for CS traffic,
for PS traffic, and the KPIs and the Traffic Load for each subcell. The Traffic Load is defined as the ratio of traffic demand
to the # TSL carrying that traffic. It is perhaps one of the most important outputs of the process as it is further utilized in the
Automatic Frequency Planning process. The consumption for a subcell can be extracted through multiplying the # TRX
required by the Traffic Load and by the TSL multiplexing factor (8 for the GSM standard).
Atoll also incorporates an intelligent allocation methodology for assigning the CS and PS TSL inside a TRX. The user has
the possibility to define the TSL allocation schemes for each type of TRX. This is known as timeslot Configuration.
Release 2.5.0
AT250_ARG_E2
7 rue des briquetiers – 31700 – Blagnac – France
April 2006
Tel: +33 (0)5 62 74 72 10 – Fax: +33 (0)5 62 74 72 11
http://www.forsk.com – email: forsk@forsk.com