The aim of this chapter is to study traffic offloading for the SCaaS business model, and gain insight into its techno-economic trade-offs. The scenario is characterized by multiple non-overlapping hotspot areas served by multiple MNOs and where a third party has deployed a number of SC clusters. As the hotspot areas are geographically limited, each MNO covers the hotspot areas with a single eNB1, as shown in Fig. 3.1a. Therefore, the system model is constituted by one eNB per MNO and NC SCO clusters (one in each hotspot area). The set of eNBs is denoted by N = {1, 2, . . . , N } and the MNO owner of eNB i ∈ N is denoted by M N Oi. The set of SCO clusters is denoted by NC = {1, 2, . . . , NC}, where an SC cluster l ∈ NC consists of Nscl small cells, and
is connected to the internet or to the core network through a backhaul network with a capacity CBHl (in Mbps). CBHl is therefore the capacity of the link between the
SC cluster l and the core network/internet, and it is upper bounded by the backhaul technology, such as a wireless millimetre-wave link or an optical fibre connection [84]. We assume a frequency reuse factor 1 for the SC tier. Although in Fig. 3.1a eNBs from different MNOs are not co-sited, the subsequent analysis can be also used for RAN sharing scenarios (i.e. when multiple MNOs share a single eNB), as it only assumes that the average spectral efficiency of the diverse eNBs is the same. The notation used henceforth is summarized in Table3.1.
Each BS is characterized by the spectral efficiency, defined as the transmission rate achievable per bandwidth unit, and the available bandwidth. For a given M N Oi, the available licensed bandwidth is denoted by Bi (in MHz) and the spectral efficiency SEi (in Mbps/Hz) can be approximated by SEi = E[log2(1 + SIN Rik)], where E[·] is the mathematical expectation and SIN Rik is the Signal-to-Interfernece-plus-Noise ratio of a user k served by M N Oi. For an SC cluster l, the spectral efficiency is defined analogously and denoted by SEscl. In turn, the maximum bandwidth supported by each
SC cluster is denoted by Bsc and it is limited either by the technology’s specifications or by the deployed hardware’s capabilities. Based on these definitions, the capacity of M N Oi, defined as the maximum throughput that can be served by M N Oi, is given by Ci = BiSEi. As for the capacity of the SC cluster l, it is given by NsclBscSEscl.
Note that, whereas the traffic served by M N Oi is limited by its capacity Ci, the traffic served by the SC cluster l is limited either by CBHl or by the SC cluster capacity, i.e.
min{CBHl, NsclBscSEscl}. 1
This study is focused on a single macrocell sector, hence all hotspot areas are considered to be located in the geographical area of a macrocell sector. However, the results can be extrapolated to a larger area that consists of multiple macrocell sectors.
Table 3.1: MNO and SCO Notation
Notation Description
N Set of eNBs
NC Set of SC clusters
Nscl Number of small cells in SC cluster l
CBHl Backhaul capacity in SC cluster l
Bi eNB i’s available bandwidth
SEi eNB i’s expected spectral efficiency
Ci eNB i’s maximum capacity without offloading
Bsc Maximum bandwidth supported by all SC clusters
SEscl SC cluster l’s expected spectral efficiency
Li M N Oi’s offered load
Lhi M N Oi’s offered load in all SC clusters
Lhil M N Oi’s offered load in SC cluster l
Lni M N Oi’s offered load outside of SC clusters
T Set of a 24-hour day equal-period timeframes xi M N Oi’s transferred bandwidth vector
xil M N Oi’s transferred bandwidth at SC cluster l Xi M N Oi’s total transferred bandwidth percentage Lmci Load served by eNB i
Lsci M N Oi’s load served by all SC clusters Lscil M N Oi’s load served by SC cluster l LTi M N Oi’s total served load or throughput
D, C Dedicated spectrum, Co-channel deployment superindexes
Θi Throughput difference between co-channel and dedicated spectrum deployment Pi M N Oi’s profit
bi M N Oi’s bid vector
bil M N Oi’s bid at SC cluster l Ri M N Oi’s revenue
CLi M N Oi’s load cost
ai, di M N Oi’s cost shaping parameters Psc SCO’s total profit
Pscl SCO’s profit at SC cluster l
CLscl SCO’s cost at SC cluster l
ascl, dscl SCO’s cost shaping parameters
Pscmin
l SCO’s minimum profit for SC cluster l
z SC capacity pricing factor
bminil M N Oi’s reserve price for SC cluster l
a Forecast value of any parameter a
Let us define the offered load of M N Oi as Li (in Mbps). This offered load can be divided into two components: the offered load generated within each hotspot (Lhil with
Lhi =
P
l∈NCLhil) and the offered load generated elsewhere (Lni), i.e. Li = Lhi+ Lni.
Each of these components can vary in time and space. During time periods where Lhi
is low, the need for capacity provided by the SCO declines. Conversely, high hotspot loads result in a raising interest for the usage of the SCO infrastructure. Hereafter, the day is divided into a set T = {1, 2, . . . , T } of equal timeframes, during which the
load is considered constant. It is nonetheless worth noting that, even though the load in a timeframe t ∈ T is not necessarily the same every day, it follows a daily pattern. Even though the real-time offered load and user SE do not remain constant during a timeframe, their instantaneous variations from their average (i.e. Li, SEi and SEscl)
can be disregarded when considering a timeframe’s offloading decision [29].
As it will be shown in Section 3.4, an M N Oi needs to estimate both Lhi and the
corresponding bandwidth resources transferred to the SC clusters (henceforth denoted as xi), for serving its load as well as its profit maximizing objective. To that end, and since it is not possible to know beforehand the actual spectral efficiency of each user roaming the system, we use SEi and SEscl. However, this assumption introduces some
limitations. Particularly, as a user roams a small cell, her spectral efficiency will be at times either higher or lower than SEscl, leading to lower or higher requirement of xi
respectively. In case of higher xi requirements, it is possible that part of Lhi will not
be served. This in turn can further lead to MNO revenue and profit loss as it will be shown in Section3.3.2.
Regarding the provided SC capacity, it is true that it should reflect the MNOs’ demand. However, increasing the capacity requires the further instalment of SC and backhaul infrastructure. Such deployments and the corresponding SC capacity pricing for the recoup of their investment are always part of a long-term business plan. Hence, since our system model examines SCaaS on a day-to-day basis and short timeframes according to the traffic’s trends (e.g. hours), the study of additional SC infrastructure and its corresponding pricing are out of the scope of this contribution.