An Infrastructure Cost Evaluation of Single- and Multi-Access Networks with Heterogeneous Traffic Density Anders Furuskär and Magnus Almgren
Klas Johansson
Wireless Access Networks Ericsson Research Kista, Sweden [anders.furuskar, magnus.almgren]@ericsson.com
Wireless@KTH, The Royal Institute of Technology Electrum 418, S-164 40 Kista, Sweden Email:
[email protected]
Abstract — Traditional performance measures like capacity, cell radius and supported QoS are often insufficient when comparing wireless networks with different network architectures and cost structures. Instead, in this paper, infrastructure cost is used to compare different operator deployed single- and multi-access wireless networks, including 3G, WLAN and proposed 4G radio access technologies. For this purpose a model for the geographical distribution of traffic is introduced. Despite the spatially nonuniform traffic demand, single-access solutions like WCDMA High-Speed Downlink Packet Access (HSDPA) or Long-Term 3G Evolved, with high capacity macro cellular base stations, typically yield the lowest costs per user. In particular this holds for a hypothetical Long-Term 3G Evolved system operating in 450MHz spectrum, which indicates the importance of good coverage. Operator deployed WLAN-only solutions are more expensive even for small fractions of supported users. Multi-access solutions, combining for example WCDMA DCH or HSDPA with WLAN, do not seem to provide better cost efficiency than standard hierarchical cell structures in single-access systems. Instead, multi-access solutions have to be motivated by other factors like peak data rates and spectrum availability. Keywords: Infrastructure cost, Tele-economics, multi-access, WCDMA, 3G, Long-Term 3G Evolution, 4G, WLAN
I.
INTRODUCTION
Mobile network operators are typically interested in maximizing the profit determined by the revenue generated by their systems and their costs. Traditional performance measures used for single-access cellular systems, such as coverage and capacity, are effective measures of the relative improvements for specific systems. Considering also deployment aspects and that different systems typically have different cost structure, technical measures, like spectral efficiency, are however, as discussed in e.g. [1], insufficient to compare different systems. Ideally both costs and revenues should be included in the analysis, as in [2], and availability of spectrum, previous assets, and other strategic issues need to be taken into account. Due to difficulties in, e.g., predicting end users willingness to pay, this is quite complex. A simpler initial step, for relative comparisons only, is to compare the system cost for equal potential revenues (to simplify; number of supported users) and this is also the focus of this paper.
More specifically, the radio access network infrastructure cost is normalized per user and compared between different single- and multi-access system concepts as a function of traffic intensity per user and relative user coverage. The system concepts compared include the cellular systems WCDMA DCH, WCDMA HSDPA, and preliminary Long-Term 3G Evolution and 4G proposals, as well as the WLAN system concepts IEEE 802.11b, a and g. Also multi-access combinations of these, an expected characteristic of future wireless networks [3], are included. To evaluate the benefits of the multi-access networks, a heterogeneous traffic density model is applied. Recently a number of wireless network infrastructure cost analyses for single-access networks have been presented, e.g. [1] and [2]. These conclude that the infrastructure cost, including both capital expenditures (CAPEX) and operational expenditures (OPEX), is largely proportional to the number of access points deployed. Equivalent cost figures per access point, are also presented, which can be used to simply assess the total infrastructure cost for a given deployment. Models of the spatial distribution of mobile users have been presented in e.g. [4] and [5]. This paper combines the above results and extends the scope to cover multi-access networks. In what follows, Section II briefly discusses the impact of the spatial traffic distribution on the cost efficiency of different single- and multi-access system concepts, and presents the deployment principles used in this study. An overview of the radio network, user behavior, and economical models and assumptions is given in Section III. Numerical results are presented in Section IV, followed by conclusions in Section V. II. DEPLOYMENT ASPECTS In scenarios with homogeneous, uniform traffic densities, single-access solutions with only one type of access point manage to maximize cost efficiency. For example, with a high area traffic demand micro cellular base stations may be most cost efficient, whereas macro base stations typically yield the lowest cost in areas with less traffic per area unit. Solutions with a mix of macro, micro and pico cells, as well as multi-access concepts, may be expected to be more cost efficient than single access networks only in scenarios with heterogeneous traffic densities. Yet, this is not a sufficient condition. It is also required that traffic peaks (hotspots) are very few and strong.
1) Varying but low (average) traffic density
2) Varying and high
3) Strongly varying, high, and correlated Micro cell area capacity
Traffic Density
Macro cell area capacity
Traffic Density
Traffic Density
Average
Figure 1. Simple example of traffic density variations over space, and deployment of ‘macro’ (light grey) and ‘micro’ (dark grey) access points.
This is explained by the following example based on macro and micro cellular base stations. Assume that a single-carrier macro cell layer is deployed for fundamental coverage. As depicted in the first example in Figure 1, the capacity of this network is sufficient to serve traffic density patterns for which the average is within the macro cell area capacity. Local areas with traffic demands exceeding the area capacity are supported. Then, if the average traffic density exceeds the total area capacity of the first macro carrier, either a second macro carrier, additional macro base stations or micro cells can be deployed. What solution that brings the lowest cost depends on the statistical distribution of traffic. In the second example, traffic is high and strongly varying with many relatively small peaks. Then adding a carrier to the macro-cellular layer, or deploying macro base stations more densely, is typically most efficient. Deploying a micro cell in each of the many local traffic density peaks would require a large amount of micro sites, and hence be costly, since each micro base station would have excess capacity and be poorly utilized. In situations like the one in example three, however, micro cells are motivated. Here, the peaks in traffic density are very strong and rather few, so that only a few micro cells need to be deployed, and the cost for this is lower than that of extending the macro layer. A heterogeneous traffic density alone is thus not sufficient for motivating micro and pico cell, or multi-access solutions from an infrastructure cost perspective. There are also requirements on (i) a high overall traffic density, (ii) strong variations in traffic density, and (iii) special spatial correlation properties. A. Deployment Principle The deployment principle used in this study is to first deploy macro cells for full area coverage, and then complement with micro or pico cells, or WLAN access points, where it is needed for capacity reasons. In more detail, first, the system area Asys is divided into N different 40x40m elements. The (center) position and the traffic generated in element n are denoted Pn and TEn respectively. For each RAT r, candidate access point sites are positioned on a regular hexagonal grid, with site-to-site distances according to the AP cell radii. The position of AP m is denoted Sm. Based on the site positions, an association between sites and traffic elements is made, so that the elements in the set Srm belong to AP m. The association is done so that traffic elements are associated with the closest site:
S mr = {n arg k min{d ( APk , Pn )} = m}
(1)
where d(APk, Pn) is the distance between AP k and element n. The offered traffic per AP is calculated as:
Tmr =
∑ TE
n
(2)
n∈S mr
In each site m, Nrm cells (transceivers) are then deployed to fulfill the offered traffic, while not exceeding the maximum number of cells per AP, denoted NAPmax: T r N mr = min m , N APmax, m C r
(3)
where Cr is the capacity per access point of RAT r. Note that if Trm = 0, no access point is deployed. If Trm > Cr, all the offered traffic cannot be handled by RAT r. In that case elements are allocated in an increasing order of offered traffic TEn until the maximum capacity per AP CrNAPmax is reached. Remaining traffic that has to be served with other RATs (with a smaller cell radius and higher area capacity) will then belong to elements with the highest traffic TEn. More formally, the traffic elements Srm are sorted in order of offered traffic and indexed n’. The number of elements Mrmax,m that are served by AP m is then determined by: M r M max,m TE n ' ≤ C r N APmax,m = max M : n '=1
∑
(4)
Then, the offered traffic in elements n’=1.. Mrmax m is set to zero to determine the offered traffic for the next RAT to be deployed, i.e., 0 TE n ' = TE n'
r n' ≤ M max m r n' > M max m
(5)
Once the above deployment is completed, which results in 100% coverage if no traffic remains after the last RAT is deployed, the cost for covering smaller fractions of users is calculated. This is done through sorting the access points in order of supported traffic divided by the access point cost. Given the final deployment for full coverage, deployment in this order represents the most cost efficient way to support a given traffic. It should be noted that no attempts have been made to optimize the deployment principle. Instead, the target has been a simple principle that is reasonably good and fair between system concepts. It could e.g. be noted that limiting WLAN access point positions to a regular grid is probably not optimum, but neither is allowing only one cell radius for cellular macro, micro, and pico cells. This tradeoff is discussed further in [7].
Figure 2. A sample of a traffic density map and WCDMA macro and WLAN access point deployment.
III.
MODELS AND ASSUMPTIONS
This section describes the user behavior, system, and radio network models used to evaluate the different system concepts. Macroscopic models are used to enable a conceptual comparison between the concepts for different traffic densities. A. Traffic Density Models In order to capture the effects discussed in Section II, a heterogeneous user behavior is assumed. In short, based on the measurements and model proposed in [4] and statistics from [6], it is assumed that the user density is log-normally distributed around a ‘large-scale’ mean. The ‘small scale’ standard deviation of this distribution is adjusted so that assumed peak values in user density are achieved with reasonable probability. To fit the ‘cell-level’ user density standard deviation to the value 0.4 (log-scale) reported in [4], a spatial correlation is assumed between elements. Reference user densities are created by multiplying typical suburban (su) and city centre (cc) population densities, 500 and 20.000 inhabitants/km2 respectively, with an assumed service penetration of 90% and an operator market share of 30%. Users are further characterized by an average busy hour traffic intensity, measured in data generation per unit time. As a basis for this, the traffic intensity of a private voice user during busy hour is used. This is assumed to be 20mErlang x 10kbps = 0.2kbps. Multiplying this with a factor N then forms traffic intensity reference values. As a reference, assuming that 0.6% of the monthly traffic is generated during each busy hour (typical for voice), a 1GB/month user corresponds to 13kbps, or N = 66. Traffic density maps (10x10km) are created by multiplying the user densities and per-user traffic intensities. The gray scale contour in Figure 2 depicts a realization of traffic density generated by the model. Note that no explicit service is assumed. The evaluation is applicable to all services for which the system models, i.e. access point capacity and coverage, are valid, and that are within the capabilities of the access technology. These capabilities differ significantly between some of the access technologies. For example, 4G concepts should be compared with WCDMA only for services supported by both networks.
TABLE I. ACCESS POINT CHARACTERISTICS. Radius Capacity 1000m [3-9] x 1Mbps WCDMA DCH macro 250m [1-2] x 1Mbps WCDMA DCH micro 100m 1Mbps WCDMA DCH pico 1000m [3-9] x 2.5Mbps WCDMA HS macro 250m [1-2] x 2.5Mbps WCDMA HS micro 100m 2.5Mbps WCDMA HS pico 1000m 3 x 15Mbps S3G macro 250m 15Mbps S3G micro 100m 15Mbps S3G pico 2500m 3 x 15Mbps S3G macro 450 700m 3 x 100Mbps 4G 175m 100Mbps 4G micro 70m 1Gbps 4G pico 1850m 100Mbps 4G relay 40m 6Mbps IEEE 802.11b 40m 22Mbps IEEE 802.11g 20m 22Mbps IEEE 802.11a 20m 100Mbps IEEE 802.11n
Cost Coeff. 1 (55/45%) 0.45 (45/55%) 0.3 (35/65%) 1 (55/45%) 0.45 (45/55%) 0.3 (35/65%) 1 (55/45%) 0.45 (45/55%) 0.3 (35/65%) 1 (55/45%) 1 (55/45%) 0.45 (45/55%) 0.3 (35/65%) 6.4 (65/35%) 0.13 (3/97%) 0.13 (3/97%) 0.13 (3/97%) 0.13 (3/97%)
B. System and Radio Network Models Access points of different access technologies are characterized with different maximum cell radii and capacities; see Table I (herein a cell is defined as a combination of a sector and carrier frequency). All figures are for the downlink and roughly valid for an urban environment without strict requirements for indoor coverage. However, with the simplified modeling used, without explicit radio network models, the models are applicable for arbitrary environment, deployment and service scenarios for which the system models are valid. The WCDMA DCH and HS-DSCH figures, assuming a 15MHz spectrum allocation, are taken as reference values, and Long-Term 3G Evolved [8] (henceforth shortly denoted ‘S3G’) and 4G figures are derived from these. For S3G, a 20MHz spectrum allocation is assumed. Together with a spectrum efficiency assumption of 0.75bps/Hz/cell, this results in a capacity per cell of 15Mbps. The same power density as for WCDMA is also assumed, resulting in the same cell radius. To investigate the impact of coverage, a hypothetical S3G system operating in 450MHz spectrum, is also studied. Its cell radius is simply based on frequency difference and a path-loss exponent of 3.5. For 4G, a 100MHz spectrum is assumed, together with a slightly improved spectrum efficiency of 1bps/Hz/cell. This results in a capacity per cell of 100Mbps. A four times lower power density is assumed for the wider 4G carrier than for WCDMA. Assuming a distance attenuation exponent of 3.5, this results in a 30% reduced cell radius. Micro and pico cell capacities are assumed equal to the macro-cell capacities (per cell). The WLAN figures assume single-cell, non-interfered access points. In coordinated multi-cell scenarios these figures decrease some 20-40% for 802.11b and 802.11g. In noncoordinated multi-operator scenarios, the capacity is shared equally between the operators. A simple 2-hop regenerative relaying concept is also evaluated. It is assumed that the access point is surrounded by a ring of six relay nodes, each with the same cell radius as a regular macro cell access point. This results in an equivalent cell radius of √7 of the original cell radius. The capacity is limited by the access point, and assumed to remain at 100Mbps despite the potentially favorable channel conditions towards the relay
10
-2
0
10
Traffic Density [Mbps/km2]
10
2
3
2
30€/month
10
-2
0
10
Traffic Density [Mbps/km2]
10
cc1000
-1
su1000
10
0
cc10
10
1
su100
10
cc100
10
802.11g WCDMA DCH WCDMA HS DCH & 11g HS & 11g S3G S3G 450 4G 4G relay
cc1
su1000
cc10
su100
-1
cc1
10
0
su10
10
1
10
Fraction of Supported Users 20%
4
su10
30€/month
10
10
su1
2
Infrastructure Cost per Month and Gbyte [€]
3
cc1000
10
802.11g WCDMA DCH WCDMA HS DCH & 11g HS & 11g S3G S3G 450 4G 4G relay
cc100
10
Fraction of Supported Users 90%
4
su1
Infrastructure Cost per Month and Gbyte [€]
10
2
Figure 3. Infrastructure cost per 1GB/month user versus traffic density for 90% supported users.
Figure 4. Infrastructure cost per 1GB/month user versus traffic density for 20% supported users.
nodes. Note that more sophisticated relaying concepts than that evaluated here exist, with potential to further improve coverage and capacity. The access points are also characterized with the cost coefficients given in Table I. These estimate the total infrastructure cost associated with one access point, including CAPEX for radio access network equipment and site build out, as well as OPEX for site rental, transmission, power consumption and O&M over a 10-year period, assuming a 10% discount rate. The figures build on those used in [1], where in turn equipment cost estimates were provided by the Gartner Group and other cost estimates were based on [2]. In this study minor updates for radio network controllers, power consumption and O&M, and addition of WLAN, also based on [2], have been made. The coefficients in Table I are normalized to an estimated value for a cellular macro base station, assumed to be €300k (slightly higher than in [1] due to the above modifications). The components of the cost coefficients are further discussed in [1] and [2], Table I merely includes the fractional CAPEX and OPEX, which in turn are dominated by site and transmission costs respectively. The total infrastructure cost for green-field operators can be calculated as the number of access points of each type multiplied with the corresponding cost coefficients. Note that this model excludes costs for core network nodes as well as costs for spectrum (due to differences in regulation, a generally applicable spectrum cost model is very difficult to define). This makes the costs incremental, i.e. measuring the additional cost for covering a new area, once the core network and spectrum is paid for. Terminal costs, as well administrative costs, e.g. for marketing and billing, are also excluded.
GB and month is marked. This is a rough estimate of what a typical user is willing to spend on mobile communications today. Generally, for all system concepts, the infrastructure cost per GB decreases with traffic density while the systems are coverage limited, and flattens when the system becomes capacity limited.
IV.
NUMERICAL RESULTS
In this chapter numerical results are presented on the form infrastructure cost per transferred data unit (1GB) and month versus traffic density and fractional coverage. In Figure 3 90% of the traffic (users) are supported and in Figure 4 only a small fraction, 20%, of the users are served. Some reference levels are marked on the traffic density axis. These are combinations of suburban (su) or city center (cc) environments as defined above, and traffic intensities per user measured in N times voice (0.2kbps). On the cost axis, a reference level of €30 per
A. Commercially Available Systems Beginning with the 802.11g WLAN and 90% of the traffic served, it is seen that for low traffic densities the cost per GB is very high. In a suburban environment with a voice-like traffic intensity per user (su1), the infrastructure cost reaches €10.000/GB/month. To get down to a reasonable cost per GB (€30), a traffic density of 10 Mbps/km2 is required, approximately corresponding to su500 or cc10 scenarios. For a fraction of supported users of only 20%, as shown in Figure 4, the cost per user for 802.11g decreases significantly (as expected). The reasonable cost of €30/GB/month is now reached at 1Mbps/km2 instead, or roughly a cc1 scenario. This indicates the degree of coverage that can be expected to be profitable for WLAN only operators. WCDMA DCH and HSDPA yield about 50 times lower cost for moderate traffic densities. These systems reach €30 per GB and month already at 0.2Mbps/km2 corresponding to su10 scenarios. WCDMA HSDPA becomes capacity limited at higher traffic densities than WCDMA DCH, and therefore yields lower costs at high traffic densities. The crossover point between WCDMA HSDPA and 802.11g is about 100Mbps/km2, or cc100. With 20% of the users covered WCDMA HSDPA is more expensive than WLAN at 30Mbps/km2 (su1000/cc30), whereas with 90% coverage WLAN only systems gives a lower cost at first around 100 Mbps/km2 (cc100). The multi-access concepts, WCDMA DCH or WCDMA HSDPA combined with 802.11g, are seen to yield the lowest cost of the included subsystems. However, the gain as compared to, e.g., a single access WCDMA HSDPA system (with hierarchical cell structures) is evident only at very high traffic (> 300Mbps/km2). With the models and assumptions used, there is hence no significant multi-access cost reduction. On the other hand there is neither any loss, and there is thus no
cost drawback for a mobile network operator deploying macro and micro cells first, and then adding WLAN only in hotspots, as compared to a pure WLAN operator. It may also be noted that HSDPA alone is a better solution than both WCDMA DCH and 802.11g for traffic loads up to 100Mbps/km2. B. Future Concepts The S3G concept, with similar coverage and cost characteristics as WCDMA DCH and HSDPA, also yield the same cost as these concepts at low and moderate traffic densities (while the systems are coverage limited). S3G however remains coverage limited for higher traffic densities, and yields lower costs for traffic densities exceeding 2Mbps/km2. S3G is also seen to be a better alternative than WCDMA HSDPA combined with 802.11g for the full range of studied traffic densities. The benefit of large coverage is also seen from the hypothetical S3G 450 system. Due to its large cell radius, it yields cost almost 10 times lower cost than the other cellular concepts for traffic densities up to around 2Mbps/km2. Even for high traffic densities it is better than standard S3G despite the same capacity per AP. This indicates that despite a high mean traffic, there are large areas with less traffic where a large coverage per AP is important. The preliminary 4G concept, without relaying, is seen to suffer somewhat from its reduced coverage for traffic loads up to about 10Mbps/km2. Beyond this level it yields the lowest cost per user. The 4G concept with relaying provides slightly lower cost than 4G without relaying for moderate traffic densities, but still higher cost than both WCDMA and S3G. C. Pricing Strategy and Service Offering Consequences The cost per data unit can be mapped to a cost per user (and service) in several ways. This is a quite complex area, which has been subject to many studies. Value based pricing is, however, nowadays most often used in practice and there is typically little relation between price and production cost [9]. Yet, the (incremental) production cost will tell if a service would be profitable or not given the end user pricing possibilities.. Assuming a traffic independent cost the average cost per user is given by the total infrastructure cost divided with the number of supported users. This can also be calculated by multiplying the cost per unit data with the average per user traffic intensity. Alternatively, assuming a linearly traffic dependent cost, the cost per individual user is given by multiplying the cost per unit data with the individual user traffic intensity. Several alternatives in between these ‘extremes’ of course exist. The results presented here are valid for all these alternatives. An interesting observation is that, with the models and assumptions used, the incremental infrastructure cost for 1GB per month per user can be kept below €30 for traffic densities exceeding about 0.2Mbps/km2. Adding margins for excluded costs (marketing, customer care, core network and service platforms, profit, taxes, etc.) about 1Mbps/km2 is probably a more realistic value. This roughly corresponds to a city center area with today’s voice traffic, or a suburban area with 40 times this traffic per user.
V.
CONCLUSIONS
The results of this study indicate that, with the models and assumptions used, single-access solutions with high capacity macro cells yield the lowest costs per user, despite a spatially non-uniform traffic demand. Examples of such systems are WCDMA HSDPA and the preliminary S3G concept. Operator deployed WLAN-only solutions yield high costs even if the requirement on fractions of supported users is small (20%). Except for at very high traffic, a multi-access network composed of WCDMA DCH or HSDPA macro and micro cells combined with IEEE 802.11g access points do not yield lower costs per user than using a single-access WCDMA DCH or HSDPA network consisting of macro, micro and pico cells. This is because several WLAN access points are required in each ‘hotspot’ due to the poor coverage, which results in an excess capacity and poor utilization of each AP. A hotspot concept with better coverage could thus lead to better results also for moderate average traffic densities. For mobile network operators having a 3G license introducing WLAN hotspots hence need to be motivated by other factors; such as access technology capabilities and spectrum. Among the 4G concepts, it is seen that a simple relaying solution with macro-like relays nodes may yield improved cost efficiency in areas with moderate traffic demand, up to some 10s of Mbps/km2. For traffic densities beyond this level, this particular relaying solution is not motivated. The overall lowest cost is enabled by a hypothetical high-capacity cellular system operating in 450MHz spectrum. This indicates the importance of good coverage, which is of course valid also for alternative means to achieve it. Future studies could make use of more refined system and economical models. In particular empirical data on traffic demand for mobile data services would be useful to improve the heterogeneous traffic density model. REFERENCES [1]
[2] [3]
[4] [5]
[6] [7]
[8]
[9]
K. Johansson, et al., “Relation between base station characteristics and cost structure in cellular networks”, in the Proc. of IEEE Personal, Indoor and Mobile Communications (PIMRC), 2004. F. Loizillon et al., “Final results on seamless mobile IP service provision economics”, IST-2000-25172 TONIC Deliverable no. 11, Oct. 2002. N. Niebert et al., “Ambient Networks: An Architecture for Communication Networks Beyond 3G”, in IEEE Wireless Communications, Vol. 11, Issue. 2, April 2004, pp. 14-22. U. Gotzner et al., ”Spatial Traffic Distribution in Cellular Networks”, in Proceedings of IEEE Vehicular Technology Conference, 1998. R. Ganesh and K. Joseph, “Effect of non-uniform traffic distributions on performance of a cellular CDMA system”, Universal Personal Communications Record, October 1997. US Census Bureau, Table GCT-PH1. “Population, Housing Units, Area, and Density”: 2000, available at http://factfinder.census.gov/, K. Johansson and A. Furuskär, “Cost efficient capacity expansion strategies using multi-access networks”, in Proceedings of IEEE Vehicular Technology Conference spring, 2005. Third Generation Partenership Project (3GPP), RP-040461, “Proposed Study Item on Evolved UTRA and UTRAN”, available at www.3g pp.org/ftp/tsg_ran/TSG_RAN/TSGR_26/Docs/PDF/RP-040461.pdf. T. T. Nagle and R. K. Holden, "The Strategy and Tactics of Pricing", Second Edition, Prentice Hall, 1997.