Update of the FIFAC Model

Update of the FIFAC Model
A REPORT PREPARED FOR FICORA
January 2013
© Frontier Economics Ltd, London.
January 2013 | Frontier Economics
i
Update of the FIFAC Model
Executive Summary
1
1
Context
3
1.1
Original FIFAC model................................................................. 3
1.2
Changes in mobile networks and services ................................. 3
2
Fixed cost recovery
2.1
Source of fixed costs in mobile networks ................................... 8
2.2
Recovery of fixed costs .............................................................. 9
3
Required changes in the model
3.1
Scope of changes .................................................................... 11
3.2
Areas where change is necessary ........................................... 11
4
Revised cost allocation for mobile RAN
4.1
Existing approach .................................................................... 12
4.2
Recommended approach ......................................................... 12
4.3
Attributing RAN costs by technology ........................................ 13
4.4
Conversion factors by technology ............................................ 16
8
11
12
Annexe 1: Applying Cost causality
27
Annexe 2: Accounting for QoS in allocation factors
29
Adjusting for quality of service ............................................................ 29
Contents
ii
Frontier Economics | January 2013
Update of the FIFAC Model
Figure 1. Mobile broadband traffic in Finland
6
Figure 2. Voice traffic
7
Figure 3. Example of calculation of conversion factor for SMS
16
Figure 4. Example of conversion factors for packet data
19
Figure 5. Total traffic across network
20
Figure 6. Traffic weighted for QoS
20
Figure 7. Example traffic by network and technology
21
Figure 8. Example of normalised traffic by technology
21
Figure 9. Allocation of RAN costs by class of base station
21
Figure 10. Example of allocation of RAN costs by technology
22
Figure 11. Example of allocation of RAN costs by type of traffic
22
Figure 12. Example of calculation of final conversion factor
23
Figure 13. Capacity vs traffic for real time services
31
Figure 14. Relationship between incremental traffic and capacity
32
No table of figures entries found.
Tables & Figures
January 2013 | Frontier Economics
1
Executive Summary
Frontier has been engaged to update the FIFAC model which Frontier initially
developed in 2005. Since the model was first specified there have been a number
of changes in the mobile sector:

High growth in packet data services as smartphones have been
increasingly used;

Introduction of improved radio access network (RAN) technology
providing more efficient conveyance of packet data from 3G
technologies;

Introduction of ‘’4G”(LTE) technology providing still more efficient
carriage of packet data;

Introduction of software defined radio in the RAN.
While the basic framework of the model is equally valid as it was when first
developed, these developments mean that some of the detailed assumptions need
to be revised to ensure the model is still fit for purpose. The update is designed
to bring the parameters used in the model up to date while the overall costing
framework and structure of the model remains unchanged.
The only change required in the model is an update in the cost allocation of the
RAN to services. This is to ensure that the model fully recognises the multiple
technologies used in the RAN and allocates costs between voice and data
services to reflect the characteristics of these technologies. This is necessary
because the initial model adopted a simplified approach which is no longer
robust given: the rapid increase in data traffic in recent years; and the increase
usage of 3G and 4G technologies.
This version of the report amends the first draft of the report to take account of
comments received by the operators in two areas1:

Taking account of the greater quality of service (QoS) for voice traffic
compared to data traffic;

Taking account of variations in the availability and capacity of different
technologies throughout the country.
These two changes have the effect of increasing the proportion of costs allocated
to voice services.
1
An additional chapter explaining the approach taken to allocating fixed costs has also been added to
the report as background.
Executive Summary
January 2013 | Frontier Economics
1
Context
1.1
Original FIFAC model
3
In 2005, the Finnish Communications Regulatory Authority (FICORA) has
commissioned Frontier Economics Ltd to conduct a study on developing the
evaluation process for determining the cost oriented mobile termination prices in
Finland.
For the purposes of estimating the cost of mobile termination in Finland,
Frontier proposed and developed a top-down element-based Fully Allocated
Cost model (the FIFAC model). The model allocates the operators’ actual
network cost to the services that the operators provide on a current cost
accounting (CCA) basis. This was a two stage process first allocating the
operators costs into general elements of mobile telecommunications network and
then estimating the usage of the network elements by each of the services.
In developing cost allocation models there is always a trade-off between the
precision of the cost allocations on the one hand and the complexity of the
model on the other hand. At the time the FIFAC model was developed 3G
technology was being introduced into the network but the penetration of 3G and
the usage of packet data services were still relative low. Reflecting this a relatively
simple treatment of the costs of the RAN was adopted
1.2
Changes in mobile networks and services
1.2.1
RAN
Since 2005 there have been a number of significant developments in the RAN
including the addition of Long Term Evolution (LTE) technology.
In this section we review the technological developments by technology (2G, 3G
and 4G/LTE) and for the RAN in general.
2G
As a legacy technology there has been limited development in GSM since the
original FIFAC model was developed. The major trends have been the
‘refarming’ of some of the spectrum previously used for GSM services for 3G
and 4G services (900 MHz being used for 3G and 1800 for 4G). Other changes
may be some increased penetration of technologies such as EDGE as legacy
equipment has been replaced.
It is likely that 2G services will remain in service for some years to come to
support legacy terminals and the model should continue to reflect 2G
technology.
Context
4
Frontier Economics | January 2013
3G
At the time that the original report was published, the coverage, penetration and
usage of 3G services was limited.
Since 2005, 3G handsets have become predominant and coverage of 3G
networks has become near universals with the result that a high proportion of
both voice and data traffic are carried over 3G networks.
In addition there have been significant advances in the efficiency of the air
interface for packet data traffic – high speed packet access (HSPA). HSPA allows
higher maximum download speeds for end users and greater throughput per cell.
As there have not been similar advances in the technology for voice calls, the
relative efficiency of packet data and voice calls will have changed significantly.
Continued evolution of HSPA will ensure that 3G services remain competitive
with 4G in the medium term.
LTE
LTE technology is a next generation mobile technology. Existing 3G technology
has its origins in the 1990s and the design decisions reflect a combination of
expectations of user requirements at that point and the available technology at
that point. LTE provides benefits over existing 3G technologies both in terms
of increased capabilities for end users and lower cost for providers through:

A radio interface providing higher peak bandwidth for individual users
throughout the cell and for both static and mobile users, along with an
increased number of simultaneous users and total data throughput;

A radio interface which makes more efficient use of spectrum and is
flexible in terms of bands standardised for use and the quantity of
spectrum required;

A simplified all-IP core network architecture.
LTE technology is being introduced in mobile networks globally. LTE is being
extensively rolled out in a number of EU member states, including Finland.
Software defined radio
Traditionally cellular radio networks have been built using hardware dedicated to
one specific technology. Adding capacity to an existing base station generally
required a ‘truck roll’ for engineers to install additional hardware (transceivers)
using different frequencies
The introduction of single RAN technology based on software defined radio
(SDR) leverages the growth in computing power to move certain functions into
software running on generic hardware, rather than implementing the functions
Context
January 2013 | Frontier Economics
5
through dedicated capacity. The use of single RAN provides a large number of
benefits:





By using generic hardware which benefits from economies of scale the
overall cost of the equipment is much lower than traditional base stations;
Simpler, smaller, more power efficient hardware reduces the operational
expenditure required to house, power and maintain base station equipment;
As the air interface is defined within software the same hardware can be
used to deliver multiple air interfaces;
Where the existing equipment has the required capability, new air interface
technologies can be largely implemented as software upgrades rather than
requiring additional hardware to be purchased and installed; and
Additional capacity (carriers) can be added remotely via software upgrades
rather than requiring site visits to install additional hardware.
Given all of these advantages, single RAN is the preferred technology for the
installation of new base stations or when upgrading existing base stations to add
additional technologies such as 3G and LTE. However the installation of single
RAN technology requires significant one off investment which outweigh any
operational cost savings and thus will not be justified on existing base stations
unless it results in in incremental revenues, for example because of increased
coverage of 3G.
The introduction of SDR based equipment in the network eliminated many
hardware costs specific to one or other technology, which has implications for
service costing.
1.2.2
Core network switching
When packet data services were first launch, packet data capability was added to
existing circuit oriented voice network through an overlay network. This led to
operators operating two ‘domains’: a circuit oriented domain used to deliver
voice calls and a packet switched domain used to provide packet data services
While equipment in the circuit oriented domain used for voice calls has
increasingly used underlying packet switched technology such as IP, there has not
yet been convergence of the two domains. This has resulted in a high degree of
separation of assets used for voice and packet data.
In the longer term, with LTE, the core network is expected to be unified, with a
single converged packet switch network also carry voice traffic. This will provide
two advantages:
Context
6
Frontier Economics | January 2013

By running a single network rather than two parallel networks, costs can
be minimised; and

Moving voice traffic onto a packet switched network allows operators
to take account of the greater flexibility and lower costs offered by
packet switched networks.
However in the medium term, until voice over LTE becomes widespread, we
would expect the current core network topology to remain.
1.2.3
Backhaul and transmission
Increasingly transmission migrated from TDM technologies (PDH and SDH) to
packet switched based technologies (Ethernet) which provide advantages in
terms of lower unit costs and increased capacity given the growth in data traffic.
1.2.4
Service evolution
Since the original model was designed there has been rapid growth in packet data
traffic driven by increased penetration of smart phones and the used of USB
‘dongles’ for mobile broadband access. This trend is expected to continues as
can be seen in Figure 1 below.
Figure 1. Mobile broadband traffic in Finland
Source: FICORA Communications market in Finland Market review 2011
While there has been continuing growth in mobile telephony traffic, this has been
relatively limited and in Finland appears to be largely due to substitution of fixed
calls by mobile calls.
Context
January 2013 | Frontier Economics
Figure 2. Voice traffic
Source: FICORA Communications market in Finland Market review 2011
7
8
Frontier Economics | January 2013
2
Fixed cost recovery
2.1
Source of fixed costs in mobile networks
2.1.1
RAN
A large proportion of costs in mobile access networks are fixed with respect to
the level of traffic, but variable with respect to the area covered. For example
Ofcom have estimated that less than half of the total costs of a mobile network
are incremental with respect to traffic2. Increasing coverage requires additional
base stations to be deployed, as the area covered by each base station is largely
fixed by the propagation characteristics of the spectrum and the technical
requirements of the technology used. In marginal areas with relatively low
population density, the minimum base station configuration required to provide
services will be sufficient to serve all traffic generated within the associated
coverage area3. In these areas costs will be invariant for small increases in traffic
from current levels.
The RAN cost components fixed with respect to traffic, but variable with respect
to coverage include: the base station sites; related infrastructure; antenna; and the
base station electronics4. The costs which are variable with respect to traffic
include software licences related to the capacity of the air interface5.
2.1.2
Backhaul and core network
As well as fixed costs in the RAN, there will be fixed costs in the backhaul
network and the core network.
For example some backhaul links to base stations in low traffic ‘coverage’ areas
are likely to be dimensioned with a minimum amount of capacity, that will not be
dependent on the actual level of traffic on the network.
Some elements of the core network may also have a fixed minimum capacity
which will lead to fixed costs which do not vary with demand.
2
Estimated incremental costs of call termination of 0.69 pence per minute compared to an a cost of
1.61 pence per minute include a mark-up for fixed cost. Wholesale mobile voice call termination
statement 15 March 2011, Table 9.1.
3
As mobile pricing is by its nature independent of location, end user pricing will be set such that
demand in areas of high traffic density matches available capacity in those areas. As a result traffic
in relatively low density areas will be significantly below available capacity.
4
Under single RAN technology, the electronic equipment required does not very with traffic.
5
Under the forward looking technology based on software defined radio, capacity constraints are
related to software licencing rather than physical constraints of the underlying electronics.
Fixed cost recovery
January 2013 | Frontier Economics
2.1.3
9
Networks with decreasing demand
Due to the existence of sunk costs, networks with demand that is decreasing over
time, e.g. 2G networks, may have a greater capacity than required to deliver the
current level of traffic. In this case it is reasonable to take a longer term view,
taking into account the need to dimension the network taking into account traffic
in the past when traffic was growing.
2.2
Recovery of fixed costs
The EC recommendation on Cost Accounting6 states:
“It is recommended that the allocation of costs, capital employed and revenue be
undertaken in accordance with the principle of cost causation […].”
By definition, fixed and common costs are not directly causally related to any
service or product and as such another method must be found. This may
include:

Allocating fixed and common costs in proportion to incremental costs;

Taking account of a broader view of causality;

Taking account of broader efficiency issues in regulated prices.
We discuss each of these possible approaches below:
2.2.1
Allocation in proportion to incremental costs
Recovering common costs in proportional to incremental costs – LRIC plus
equi-proportionate mark up (LRIC+EPMU) – is considered to be a reasonable
neutral recovery of common costs that cannot be allocated on the basis of direct
causality.
The direct estimation of LRIC+EPMU is difficult. In most cases FAC or similar
approaches are used as a proxy for LRIC+EPMU. This is the approach adopted
by the FIFAC model.
2.2.2
Broader view of causality
While a narrow view of cost causality may not provide allocation of fixed and
common costs to individual services a broader view of cost causality may be
taken into account, for example the reasons why certain fixed expenditures were
initially incurred or an analysis of who benefits from the expenditure having
being made.
6
COMMISSION RECOMMENDATION of 19 September 2005 on accounting separation and cost
accounting systems under the regulatory framework for electronic communications (2005/698/EC)
Fixed cost recovery
10
Frontier Economics | January 2013
For example it could be argued that as mobile networks were initially developed
for voice traffic, voice services should recover the majority of fixed and common
costs. However it is not clear whether such an approach would be justified
efficient for a number of reasons:
2.2.3

Investments made in the current lifetime of most assets (e.g. in the last
decade) will have been incurred to deliver both voice and data traffic;

Disproportionate recovery of fixed and common costs from voice
discourages use of voice traffic compared to data traffic, which may lead
to inefficient patterns of demand, for example use of VoIP over data
bearers in place of voice traffic;

It is unclear that an allocation of fixed and common costs related to
coverage from those customers who may high usage of voice services
compared to those who make high use of data services
Broader efficiency considerations
Regulators may distinguish between efficient regulated prices and costs, for
example taking into account broader efficiency considerations such as potential
externalities or competition effects when setting prices. For example the EC in its
recommendation on termination rates, recommends a “pure” LRIC approach
when determining termination rates, with no fixed and common costs recovered
through call termination payments.
We understand that the current legislation in Finland does not provide for a pure
LRIC approach to be implemented.
2.2.4
Conclusion
Given the terms of reference of this project we see no reason to depart from the
current methodology, which uses a FAC approach as a proxy for LRIC+EPMU.
Fixed cost recovery
January 2013 | Frontier Economics
3
Required changes in the model
3.1
Scope of changes
11
There several levels of updates that could be applied to the model. In ascending
order of work required, these are:

Level 1: an update to the existing model parameters (e.g. routing tables,
conversion factors, cost-to-component allocation keys);

Level 2: an update to the components to which costs are allocated,
together with the relevant updates to routing tables and conversion
factors, as well as updates to the Excel model.

Level 3: an update or modification to the underlying core principles of
the cost allocation. This could involve, for example, introducing or
removing stages in the current cost allocation process.

Level 4: a change in the modelling approach, for example moving from
top-down cost allocation model to bottom-up long-run incremental cost
models.
FICORA’s requirement was that Frontier undertake an update covering level 1
and level 2, that is updating parameters and cost categories while leaving the
overall framework unchanged in terms of the basis of the model and the cost
allocation process.
3.2
Areas where change is necessary
Changes to the model need to take account of changes in the cost structure of
mobile networks since 2005, along with the changes in market demand, in
particular the rapid growth of packet data. The principal change is the use of
three generations of technology in the RAN with a significant proportion of
traffic on all three generations. This requires making changes to the simplifying
assumptions used in the existing model for the allocation of RAN costs.
A review of the other elements of the model suggests that the model does not
require material changes.
Required changes in the model
12
Frontier Economics | January 2013
4
Revised cost allocation for mobile RAN
4.1
Existing approach
At the time the FIFAC model was developed most traffic was carried on the 2G
network and packet data traffic was a relatively small proportion of total traffic.
In view of this a simplified approach to the allocation of RAN related costs was
taken:

All costs were allocated to a single network element “BSS”;

The costs of this network element was allocated based on a
combination of routing factors and conversion factors
The conversion factors between SMS, voice and packet data traffic were
calculated according to the characteristics of GSM, SMS voice and GPRS packet
data.
4.1.1
Deficiencies of existing approach
Since the model was developed, a high proportion of traffic has migrated to the
3G network. Recent technology has also increased the efficiency with which
traffic can be carried, in particular the efficiency with which packet data traffic
can be carried on across all networks: 2G (EDGE); 3G (HSPA); and 4G (LTE).
This has enabled the rapid growth of data traffic within the existing spectrum
holdings.
As GPRS is relatively inefficient, the continued use of packet data conversion
factors based on GPRS would allocate more costs to packet data than would be
justified by packet data’s use of underlying resources, such as spectrum and base
stations. This will lead to under-allocation of costs to voice calls, in particular
mobile termination.
4.2
Recommended approach
Using an allocation solely based on the current dominant technology, 3G with
HSPA, would produce a result which is more accurate than the existing
approach. However it would not take account the use of both legacy (2G)
technology and forward looking (4G) technology in the network. In particular,
while 2G technology may be relatively inefficient compared to 3G and 4G, the
continued use of 2G is efficient in a broader sense, in regards to the need to
maintain backwards compatibility for users with legacy handsets. Given that 2G
is likely to be disproportionately used for voice traffic, not taking into account
the characteristics of 2G networks is likely to understate the true opportunity
cost of voice traffic.
Revised cost allocation for mobile RAN
January 2013 | Frontier Economics
13
To produce a more accurate result we propose to apply new conversion factors
which take account of:

The proportion of RAN costs associated with the different technologies
(2G, 3G and 4G);

The efficiency with which each of these technologies use spectrum to
deliver SMS voice and packet date traffic; and

The distribution of traffic across the technologies
We consider each of these factors in turn below.
4.3
Attributing RAN costs by technology
4.3.1
Potential approaches
An approach which takes account of usage of different RAN technologies needs
to allocate RAN costs across these technologies. Some costs, such as technology
specific active equipment, can be directly allocated to one or other technology.
However another method needs to be found to allocate joint and common costs,
such as masts shared between technologies, between the technologies that share
the costs. As increasingly different technologies will share a common grid of sites
and even the underlying active equipment (using SDR) the need to appropriately
allocate joint and common costs will become more important.
There are a range of methods that could be used to allocate RAN costs:

Allocating RAN costs including direct costs and indirect costs such as
mast and towers, in proportion to direct costs, such as 2G and 3G
specific equipment;

Allocating indirect costs such as mast and towers, in proportion to an
underlying physical driver, such as the number of antenna on the tower
serving each technology;

Allocating costs on the basis of the usage of the underlying scarce
resource (spectrum).
We assess each of these potential approaches in turn.
Allocating costs in proportion to direct costs
The first approach effectively recovers indirect costs as a mark-up on direct
costs, but relies on being able to determine direct costs for each technology with
precision.
This would require input of data on the proportion of direct equipment costs for
each technology. Whilst in the past it may have been relatively straightforward to
Revised cost allocation for mobile RAN
14
Frontier Economics | January 2013
distinguish assets and hence costs between different technologies, this may be
increasingly difficult in the future for two reasons:

The introduction of single RAN technology means that a single piece of
hardware may be used to deliver 2G, 3G and 4G technologies; and

Contract with vendors for the acquisition and/or maintenance of the
network may not fully distinguish between costs for different
technologies.
Even where the network currently uses discrete equipment to deliver different
technologies, for example separate 2G BTS and 3G node-B, the modern
equivalent asset (MEA) will be a single-RAN solution. Or this reason it would be
inappropriate to allocate CCA costs, which should reflect MEA principles based
on the historical investment in discrete equipment.
In view of these issues we do not believe that a mark-up based approach is likely
to produce robust results.
Allocation based on a physical driver
The second approach, based on the load placed on infrastructure by different
technologies, would require collection, calculation or estimation of additional
information, for example counts of antennae by site. This could be a significant
resource burden for the operators.
Over time the approach would become less grounded in actual information as
single RAN technology, where active RAV equipment and antennae would be
shared across different technologies, became the norm.
In view of these practical and theoretical difficulties we do not recommend this
approach.
Allocation based on spectrum
Network costs may not be directly related to frequency utilised. However an
allocation based upon the usage of spectrum is theoretically justified if we assume
the resources available for mobile networks, in terms of spectrum and cell sites,
are broadly fixed in the short run. In this case the economic costs of a given
technology, for example 2G, is less driven by the incremental costs of the 2G
network and more by the opportunity cost of using spectrum and cell sites for
2G rather than for any other technology. For example the use of resources to
terminate mobile termination traffic using 2G technology prevent these resources
being deployed by the operator to offer packet services on LTE7. This
opportunity cost perspective is described in more detail at Annex 1.
7
Arguably there are some high frequency bands, such as 2.6 GHz, where spectrum is not scarce, in
that currently the available capacity is not fully utilised. However operators may have rolled out LTE
Revised cost allocation for mobile RAN
January 2013 | Frontier Economics
15
This approach requires less specific data than the other potential approach. There
may be issues where spectrum usage differs significantly in different geographical
areas, for example between urban and rural areas. In this case the allocation of
costs related to base stations in different “geotypes” could reflect the spectrum
used in that geotype for example:

In areas where base stations only provide 2G coverage, all of the costs
associated with base stations would be allocated to 2G services;

In areas where only a minimum coverage network is installed for 3G
services, for example a single carrier, while multiple 2G carriers are
enable to serve voice demand, proportionately more costs would be
allocated to 2G services; and

In dense urban areas, where the total amount of spectrum is used, the
allocation would reflect the usage of the scarce resource.
Ideally such an approach would be adopted on a base station by base station
basis, first identifying the costs associated with each individual base station and
then allocating based on the spectrum used by that base station. However such
an approach would require a large amount of information that is not likely to be
readily available within the operators.
Instead we propose to apply an assumption that the cost for each base station is
broadly constant. The cost of the RAN can then be allocated in a two stage
process, between different “classes” of base station based on the number of base
stations in each class, and between the technologies in each class according to the
spectrum used by each technology in that class.
The simplifying assumption that all base stations have broadly the same cost on a
forward looking basis may be reasonable:

With the use of single RAN equipment, costs will be less dependent on
the capacity deployed at each base station;

Increases in some costs associated with urban sites, for example
increased site rental, will be offset by reduced costs, for example the
cost of infrastructure and transmission, which will typically be higher
for rural sites.
The approach, being based on the usage of the underlying scarce resource rather
than equipment costs, would not be affected by the transition from technology
specific hardware to single RAN technology.
networks with the capability to use large blocks of spectrum in this band in order to allow high
speeds to be offered rather than capacity. This could distort the cost allocation by allocating
disproportionate costs to the LTE network. This could be allowed for by applying a factor to allow
for the much lower utilisation of spectrum by LTE in these bands.
Revised cost allocation for mobile RAN
16
4.3.2
Frontier Economics | January 2013
Conclusion
We propose to allocate the costs of the RAN according to a combination of the
number of base stations and the use of spectrum by classes of base station, as a
robust and practical approach which does not require significant additional data
collection by the operators.
4.4
Conversion factors by technology
The current FIFAC model uses conversion factors to convert SMS and packet
data traffic to voice minute equivalents. In order to update the model to take
account of different we need to review the approach taken.
4.4.1
SMS conversion
SMS traffic continues to grow in Finland, although due to the small size of the
payload per message, the amount of network capacity consumed by SMS will be
relatively small.
The conversion factor from SMS to voice minutes equivalents can be calculated
by comparing the payload for an SMS (140 octets equivalent to 140 bytes) to the
payload for a voice minute (the codec bit rate multiplied by 60 divided by 8).
An example is shown in Figure 3 below:
Figure 3. Example of calculation of conversion factor for SMS
Bytes
Kbps
160
20
Bytes
1 Minute of voice
Conversion SMS to voice minute equiv.
150,000
0.001
SMS size
Voice codec rate
Source: Frontier Economics
As SMS continues to be generally carried by the circuit switched network rather
than the packet switched network, the conversion factor can be similar for both
2G and 3G networks.
4.4.2
Packet data conversion factors by technology
For each technology, costs will be allocated between voice and packet data traffic
based upon the relative amount of voice and packet data traffic that can be
carried by each technology respectively. This will be carried out by setting
conversion factors for each technology.
This requires a three-step process:

Comparison of the peak throughput for voice and data services for each
technology;
Revised cost allocation for mobile RAN
January 2013 | Frontier Economics
17

A conversion factor between voice and data traffic volumes and
measures of peak traffic throughput;

An additional factor to take account of the higher priority given to voice
traffic compared to packet data traffic, which is delivered on a best
effort basis.
Traffic throughput can be most easily thought of in terms of number of channels
(for voice traffic) and Mbps (for packet data traffic) per MHz per sector. This
compares the opportunity cost of voice versus packet data traffic, i.e. if given
spectrum is not used for one service how much throughput for the other service
can be delivered.
As traffic volumes are input into the model in terms of millions of minutes and
Gigabytes rather throughput, a second conversion factor is needed to convert
between measures such as annual number of minutes of Gigabytes of data to
measures of traffic intensity such as Erlangs or Mbps.
2G voice and data
The throughput for GSM voice per MHz can be calculated using the number of
voice channels by carrier (7 channels assuming one time slot is used for
signalling), the width of the carrier (200 kHz) and the frequency reuse factor
(which may vary between networks).
Assuming a reuse factor of 16, the number of voice channels per MHz per cell is
7 / (16*0.2) = 2.19 channels/MHz/cell.
For data traffic the throughput per time slot can vary between 8 kbps (GPRS)
and 59 kbps.
If we assume 2G cells can offer approximately 20 kbps per timeslot and
assuming a reuse factor of 16, the Mbps per MHz per call is: 7 * 0.020/(16*0.2)
= 0.044 Mbps/MHz/cell.
3G voice and data
Much of the efficiency gains brought by 3G compared to 2G is due to the 1
reuse factor, which applies to both voice and data traffic.
To a greater extent than 2G services, the capacity of 3G cells can vary, depending
on a range of factors including the number, usage and location of users within
the cell. In addition packet data throughput has increased over time due to
successive releases of UMTS (HSPA) standards.
If we assume a typical throughput for a 5MHz carrier of 30 voice channels and 5
Mbps this provides throughput of 6 channels/MHz/cell and 1 Mbps/MHz/cell
Revised cost allocation for mobile RAN
18
Frontier Economics | January 2013
LTE data
While there is limited operating experience of LTE currently, it is expected that
spectral efficiency will be significantly greater than currently deployed 3G
technology, perhaps of the order of 2 Mbps/MHz/cell.
Carrier grade voice over LTE is not yet mature and as such voice traffic
generally is not be yet carried on LTE networks (except where end users use
VoIP applications ‘over the top’ of the packet data service). However the
implementation used will be based on VoIP, and a typical codec rate can be used
to the number of voice channels that could be carried in a given.
Conversion from channels/bandwidth to traffic intensity
Traffic volumes are typically provided in terms of millions of minutes for voice
traffic and Gigabytes for data traffic. When dimensioning networks the driver of
costs is not the total volume of traffic but the busy hour traffic intensity.
The conversion factors described above also set out are measures of
instantaneous capacity. In order to compare the two types of traffic we need to
convert these measures of traffic intensity into measures of traffic volumes. This
is straightforward:

1 Mbps bandwidth can carry to 608/89 = 7.5 Mbytes traffic in a minute

1 voice channel can carry 1 minute of voice conversation in a minute.
This simple conversion does not take into account other factors that can affect
the volume of traffic that can be carried over a given network capacity in actual
networks such as the distribution of traffic over the year and within days or any
non-billed capacity used, for example call set up times. However in the context
of a top down FAC model this level of accuracy is not needed. Only if it is
believed that there is a significant and measurable difference between voice and
data traffic is it necessary to make this distinction.
Resulting conversion factors by technology
Figure 4 shows example conversion factors for different technologies. The key
result is that for 3G networks, the transmission of packet data is much more
efficient relative to voice, due to the large improvements in spectral efficiency
due to HSPA technologies.
8
Seconds in a minute
9
bits in a byte
Revised cost allocation for mobile RAN
January 2013 | Frontier Economics
19
Figure 4. Example of conversion factors for packet data
2G
Voice channels per carrier
Packet data bandwidth per carrier
Carrier bandwidth
Frequency reuse factor
Voice Channels per MHz
Mbps per MHz
Mbyte per minute per MHz
Conversion Mbyte to voice minute
Mbps
MHz
/Mhz
Mbps/MHz
Mbyte/MHz
3G
7.00
0.14
0.20
16.00
2.19
0.04
0.33
6.67
30.00
5.00
5.00
1.00
6.00
1.00
7.50
0.80
4G
1,000.00
20.00
10.00
1.00
100.00
2.00
15.00
6.67
Source: Frontier Economics
4.4.3
Allowance for QoS
A final factor takes into account the greater quality of service offered for voice
compared to data when using the same technology. Where there is contention
between voice and packet data services, operators will manage networks such
that voice as a real time and high value service is prioritised over packet data
services. Similarly, additional capacity is more likely to be added when voice
traffic exceeds the available capacity as this will lead to calls being dropped,
compared to packet data where if offered traffic exceeds capacity, data can be
queued or packets dropped to downgrade quality of service.
While there is general agreement that voice traffic requires a higher quality of
service and hence should be prioritised and hence places a greater load on
network, there appears to be no consensus on the magnitude of the difference.
A conservative approach would be to increase the weight of voice traffic by 20%
compared to packet data and voice to take account of QoS issues. The derivation
of this factor is shown in Annex 2.
4.4.4
Calculation of updated usage factors
In order to calculate usage factors per technology from weighted average usage
factors for each technologies we need to:

Estimate the total volume of traffic by service and technology, applying
a weighting to voice to take account of QoS;

Calculate the distribution of the costs of the network across service and
technology, based on the distribution of cell sites by “class”, use of
spectrum by technology in each class and distribution of traffic within
technology;

Compare the costs allocated to each service with the total volume of
each service to calculate the weighted average conversion factor.
Revised cost allocation for mobile RAN
20
Frontier Economics | January 2013
These steps are outlined in more detail below.
Estimate volume of services by technology and service
The first step is to estimate the total traffic across the RAN. This should be
available within the model. The example below in Figure 5 takes the total traffic
across all networks in Finland.
Figure 5. Total traffic across network
2011
SMS
mn
messages
4,565
Voice traffic
mn minutes
16,105
Data traffic
Terabytes
62,000
Source: Communications market in Finland Market review 2011
Traffic then needs to be weighted to take account of the greater quality of service
offered for voice traffic,
Figure 6. Traffic weighted for QoS
QoS weighting
SMS
1.00
Voice traffic
1.25
Data traffic
1.00
Total QoS weighted traffic
2011
SMS
mn
messages
4,565
Voice traffic
mn minutes
19,326
Data traffic
Terabytes
62,000
Source: Frontier Economics
The total traffic then needs to be distributed across technologies. We would
expect the proportions to differ by services, for example a relatively greater
volume of data traffic to be carried over 3G and 4G technology than voice
traffic.
Revised cost allocation for mobile RAN
January 2013 | Frontier Economics
21
Figure 7. Example traffic by network and technology
Distribution of traffic
2G
%
%
%
SMS
Voice traffic
Data traffic
3G
60%
60%
5%
4G
40%
40%
90%
5%
Traffic by network
2G
3G
4G
Total
SMS
mn messages
2,739
1,826
-
4,565
Voice traffic
mn minutes
11,596
7,730
-
19,326
Data traffic
Terabytes
3,100
55,800
3,100
62,000
Source: Frontier Economics
Traffic then needs to be ‘normalised’ by applying the conversion factors
calculated above.
Figure 8. Example of normalised traffic by technology
2G
3G
4G
Total
SMS
million minutes equiv.
2.43
2.43
-
5
Voice traffic
million minutes equiv.
11,596
7,730
-
19,326
Data traffic
million minutes equiv.
20,667
44,640
20,667
85,973
32,265
52,372
20,667
105,304
Total
Source: Frontier Economics
RAN Cost allocation
The allocation of RAN costs to services first allocated costs to different classes
of base station (in the example below geotypes) based on the number of sites.
Figure 9. Allocation of RAN costs by class of base station
Number of base stations
Dense urban
Urban
Suburban
Rural
Total
number
number
number
number
2011
500
1,000
2,000
4,000
7,500
Percentage
6.7%
13.3%
26.7%
53.3%
100.0%
Revised cost allocation for mobile RAN
22
Frontier Economics | January 2013
Source: Frontier Economics
The next step allocates the proportion of costs for each class of base station
between technologies according to the frequency used by each technology.
Figure 10. Example of allocation of RAN costs by technology
Spectrum used by base station type
2G
Dense urban
Urban
Suburban
Rural
MHz (paired)
MHz (paired)
MHz (paired)
MHz (paired)
3G
4G
20
20
10
10
Allocation between technologies(%age of total cost)
2G
Dense urban
3.0%
Urban
6.7%
Suburban
17.8%
Rural
53.3%
RAN cost allocation
81%
Total
15
10
5
0
3G
10
10
0
0
45
40
15
10
1.5%
3.3%
0.0%
0.0%
5%
Total
6.7%
13.3%
26.7%
53.3%
100%
4G
2.2%
3.3%
8.9%
0.0%
14%
Source: Frontier Economics
The proportion of costs for each service within a technology is then allocated on
the basis of the distribution of normalised traffic volumes. This illustrated in
Error! Reference source not found. below. Note that the allocation can be
arried out in terms of percentage of costs, i.e. the absolute level of costs for the
RAN is not needed for this calculation.
Figure 11. Example of allocation of RAN costs by type of traffic
Allocation within
technologies
2G
SMS
Voice traffic
Data traffic
Total
%
%
%
3G
0%
37%
63%
100%
4G
0%
15%
85%
100%
0%
0%
100%
100%
Final allocation by technology and service
2G
SMS
Voice traffic
Data traffic
Total
%
%
%
3G
0%
29%
52%
81%
Revised cost allocation for mobile RAN
4G
0%
2%
12%
14%
Total
0%
0%
5%
5%
0%
31%
69%
100%
January 2013 | Frontier Economics
23
Source: Frontier Economics
Calculation of the final usage factor
The final stage of the calculation is to calculate the ratio between the proportion
of costs allocated to each service and the traffic for each service. The final
conversion factor is than calculated
As can be seen from Figure 12 below, the conversion factor for SMS remains
unchanged (as it is the same across all technologies), while the conversion factor
for data traffic is within the range for the different technologies as would be
expected.
Figure 12. Example of calculation of final conversion factor
Conversion factor calculation
SMS
Voice traffic
Data traffic
Allocation
0.0%
31.1%
68.8%
Traffic
4,565
16,105
62,000
unit
allocation
per traffic
1.72E-08
1.93E-05
1.11E-05
Conversion factor
0.001
1.000
0.574
Source: Frontier Economics
Revised cost allocation for mobile RAN
January 2013 | Frontier Economics
25
Revised cost allocation for mobile RAN
January 2013 | Frontier Economics
27
Annexe 1: Applying Cost causality
When considering the costs of different services and technologies which share a
common scarce resource there are two ways of considering costs:

An incremental cost approach based on estimating the costs of
additional capacity for a given service or the avoidable costs

An opportunity cost approach, based on an understanding of the
resources used, principally spectrum, could be used for if it were not
used for the given service.
An incremental approach considers the additional costs of delivering a increment
of additional demand (or the avoidable cost of not delivering an increment of
demand). This incremental cost is largely due to the costs of adding additional
capacity on existing sites, e.g. carriers, or additional sites, in areas where sites have
reached capacity limits. This reflects the decisions operators need to make when
expanding capacity based on the given technology.
Directly estimating incremental costs requires constructing detailed bottom up
costs models. Even these costs models are likely to be relatively inaccurate, as the
relationship between demand and network dimension in mobile networks is
complex.
An incremental cost approach may give misleading results in the case that
demand for one technology is falling. For example the incremental cost of 2G
services in the short term is likely to be close to zero, as with falling demand
there is likely to be excess (sunk) capacity. Thus additional traffic could be
carried with close to zero cost. However there is an opportunity cost in
continuing to operate a 2G network rather than re-farming the spectrum for use
with new technologies. This opportunity cost should be recognised.
An opportunity cost approach considers what level of demand for services using
other technologies could be delivered if an increment of demand was not
delivered under the service in question, e.g. what demand could be delivered by
3G services if spectrum was freed up by reduced demand for 2G services. This
reflects the decision operators have whether to migrate traffic from less efficient
to more efficient technologies in order to enable re-farming of traffic.
The relative opportunity cost of different services will be inversely related to
spectral efficiency, that is the volume of traffic that can be delivered using a given
amount of spectrum. A technology which makes inefficient use of spectrum will
have a relatively high opportunity costs as using this technology will prevent a
greater volume of traffic being delivered using a more efficient technology.
In equilibrium the two perspectives should produce broadly similar results in
terms of the proportion of costs attributed to each service as incremental costs
should broadly reflect spectral efficiency, that is a technology that uses spectrum
Annexe 1: Applying Cost causality
28
Frontier Economics | January 2013
more efficient should have a lower incremental cost than an inefficient
technology. For example there will be trade-offs between delivering additional
traffic by building additional sites using the existing technology and spectrum
allocations or migrating traffic from less to more efficient technologies.
Annexe 1: Applying Cost causality
January 2013 | Frontier Economics
29
Annexe 2: Accounting for QoS in allocation
factors
With the growth of data services, a growing proportion of the costs of the
network should be recovered from data services.
As there are a large proportion of fixed costs in the network, an increase in data
traffic with stable voice traffic means that the unit cost per voice minute will fall
over time, all other things being equal, This makes the allocation of costs
between voice services and data services critical.
The existing FIFAC model approach is to allocate costs between voice and other
services, by estimating the payload of information transmitted.
One criticism of such an approach is that it does not take account of the greater
QoS offered on voice, with a constant bit rate being required and voice being
delivered in real time.
While data volumes have been relatively low, such aspects have not unduly
distorted the costs allocated between voice and data services. As data volumes
increase significantly, it may be appropriate to take account of such factors. As
such the revised FIFAC model includes a factor which allocated proportionately
more costs to real time, voice services than data.
This annex explains the derivation of this factor
Adjusting for quality of service
Voice services require a minimum quality of service for a call to be successful. A
given amount of bandwidth must be constantly available and the level of delay
(latency) and the variability of this delay (jitter) must be controlled. In contrast
data services are typically best effort and are relatively delay tolerant.
The higher quality of service required of voice limits the volume of traffic that
can be carried in a given capacity. Erlang formulae can be used to determine
what intensity of traffic can be carried at a given quality of service (defined by the
blocking rate) within a given capacity. This can be used to estimate a multiplier
to apply to voice traffic, to take account of the need to allow for lower utilisation.
In contrast, delay tolerant best effort traffic can effectively be carried with very
high utilisation, with traffic being delayed when there is congestion.
Incremental cost approach
Ideally, the allocation of costs to each service should be proportionate to the
incremental cost of that service. The incremental cost is the additional cost (of
the network) if that service is delivered, in addition to other demand.
Annexe 2: Accounting for QoS in allocation
factors
30
Frontier Economics | January 2013
Equivalently the incremental cost is the cost avoided if a given service is not
delivered, but other demand is.
If the cost allocation is proportional to incremental costs, the resulting fully
allocated cost allocation should reflect incremental costs with an equiproportionate mark up.
Ideally the incremental cost for each service would be estimated by modelling the
required cost of q network with full demand, and then a network with demand
less the given service. Such an approach would implicitly take account of QoS
issues. However, modelling the impact of changing demand in such a way is
complex. In particular, if there is a mix of services with different QoS
requirements, the impact of removing one service is difficult to model.
QoS factor
Rather than complex modelling of the whole network, a simple factor is applied
with the factor being higher for those services with a higher QoS requirement.
There are two methods of estimating the factor:
Using a judgemental approach of the appropriate relativities, which may also
take account of demand side willingness to pay as well as supply side
capacity issues; or
Estimating the increased capacity needed to provide services with a given
QoS compared to services with a lower QoS.
We propose to adopt the second approach, as this reflects actual costs, rather
than judgements about unobservable demand side factors.
Estimating the differential between best effort traffic and real time
(voice) traffic
For best effort traffic, which can be delayed or discarded where offer traffic
exceeds capacity on the network, we can assume that at peak times traffic
intensity equals capacity.
For real time traffic, in order to meet a given level of quality of service, the
available capacity must exceed the average offered traffic, as traffic during peak
periods will vary around this average. If additional capacity is not maintained than
calls will be blocked (or existing calls dropped).
Estimating the increase capacity of services with a high QoS should take account
of the existing demand on the network. For example the incremental capacity of
delivering an additional 10 erlangs of voice traffic on top of existing (data) traffic
is less than the capacity required to deliver 10 erlangs of traffic on a standalone
basis.
The reason for this difference is two-fold:
Annexe 2: Accounting for QoS in allocation
factors
January 2013 | Frontier Economics
31

when there is existing traffic on the network, peaks in demand for
higher quality services can be met by delaying other traffic; and

as overall traffic volumes increase, the relative variability of the traffic
decreases, meaning that utilisation increases as volume increases.
This second effect is illustrated in the following chart, which looks at the
relationship between traffic intensity (in erlangs) and required capacity (in
channels) required to meet a given quality of service (blocking rate).
Figure 13. Capacity vs traffic for real time services
Source: Frontier Economics analysis
Another way of looking at this to look at the incremental capacity required for an
incremental increase in demand.
The table below shows that the incremental cost of the first increment of
demand is significantly higher than the incremental cost for a similar amount of
traffic, when there is already traffic on the network.
Annexe 2: Accounting for QoS in allocation
factors
32
Frontier Economics | January 2013
Figure 14. Relationship between incremental traffic and capacity
Relationship between incremental traffic and capacity
(1% blocking rate)
Offered
traffic
(erlangs)
0
5
10
15
20
25
30
35
40
45
50
Channels
required
0
11
18
24
30
36
42
47
53
58
64
Utilisation
45%
56%
63%
67%
69%
71%
74%
75%
78%
78%
Incremental
channels
11
7
6
6
6
6
5
6
5
6
Ratio
incremental
channels/
traffic
2.2
1.4
1.2
1.2
1.2
1.2
1.0
1.2
1.0
1.2
Source: Frontier Economics analysis
This table indicates that at relatively low levels of traffic, after the first increment
of traffic, an increment of 1 erlang of real time traffic requires approximately
additional 1.2 channels to be enabled to maintain the quality of service. This gives
a ratio of 1:1.2 from traffic to capacity
For best effort service, where the traffic can be considered to broadly equal
capacity the relationship is 1:1.
To reflect this difference an uplift factor of 1.2 can be applied to best effort
traffic to reflect the greater incremental capacity for an increment of demand
.required to deliver the required quality of service.
Frontier Economics Limited in Europe is a member of the Frontier Economics network, which
consists of separate companies based in Europe (Brussels, Cologne, London & Madrid) and Australia
(Brisbane, Melbourne & Sydney). The companies are independently owned, and legal commitments
entered into by any one company do not impose any obligations on other companies in the network.
All views expressed in this document are the views of Frontier Economics Limited.
FRONTIER ECONOMICS EUROPE
BRUSSELS | COLOGNE | LONDON | MADRID
Frontier Economics Ltd
Tel. +44 (0)20 7031 7000
71 High Holborn
Fax. +44 (0)20 7031 7001
London
WC1V 6DA
www.frontier-economics.com