Signal Processing - November 2016 - 117

iterations. It is observed in (5) that the amount of overhead,
and the number of iterations, decreases with e.

User/BS association in heterogeneous
wireless networks
To meet the increasing demands for reliable wireless services and the need for high data rates, service providers (SPs)
are challenged to improve their next-generation heterogeneous wireless access networks (HetNet) by utilizing small,
low-cost, and low-power APs, known as femtocell APs
(FAPs) [47], in addition to their macrocell APs (MAPs).
FAPs can be overlaid on any existing wireless technology
(e.g., 2G, 3G, LTE, WiMAX, etc.) and are envisioned as a
cost-effective solution for improving the performance of
wireless networks while enabling resource-demanding
applications and personalized wireless services [48].
To reap the benefits of FAP deployment, numerous technical challenges such as interference management, optimal
association of user equipment (UE) to APs, and optimal allocating of the FAPs to the SPs must be addressed [49], [50].
Associating the UE to their serving APs for uplink transmission is a key challenge in FAP networks that remains relatively
unexplored, as outlined in [48] and [51]. Inherently, within an
overlaid HetNet, the problem of assigning the UE to their serving APs faces a number of challenges that are significantly
different from classical CU association problems [52]. Unlike
MAPs, the FAPs are typically randomly deployed in the network without predimensioning or planning and are resource
constrained in nature, which can significantly impact the
UE association problem. Moreover, although some operators
might use dedicated FAPs, in general, the FAPs are likely to
be owned by private entities such as SPs or home users [51].
To improve the QoS for their UE and to ensure that every
UE can receive satisfactory wireless service, the SPs would
like to deploy FAPs in their networks [53]. More formally, the
SPs are willing to deploy the FAPs in their network to reduce
their churn rate. The competition among the SPs motivates
the unsatisfied subscribers to change their SP, which is called
churning. It is important for the SPs to manage the network
resources based on the expectation of the UE; otherwise, the
UE may churn to a different provider.
In [8], a distributed framework based on the matching
theory is developed that associates UE with APs and then
establishes an optimal and stable cooperation between multiple competitive SPs and competitive FAPs. Under this framework, the SPs, FAPs, and UE are assumed to be selfish and
rational entities that merely care about their own interests.
To achieve a higher data rate, the UE compete to connect to
a FAP that provides the highest data rate. The limited FAP
resources and competition among the UE motivate the UE
to offer a certain amount of monetary compensation to the
FAPs in exchange for receiving the wireless service. The proposed algorithm models the competition among the UE and
FAPs and determines both 1) the serving AP and the set of
subchannels allocated to each UE and 2) the exact amount of
money transfer that motivates both the UE FAPs to cooperate

such that the final matching is stable and the total UE sum
rate is maximized.
In summary, the distributed mechanism in [8] jointly performs associating the UE to the APs and allocating the FAPs
to the SPs. The proposed mechanism motivates SPs and FAPs
to cooperate with each other such that the total UE satisfaction level is maximized. Then they proposed a distributed subchannel allocation algorithm for the uplink OFDMA networks,
which is beyond the scope of this tutorial.
Figure 6 represents a general configuration of an OFDMA
uplink wireless communication network, comprising multiple
SPs and multiple FAPs. It is assumed that each SP owns multiple MAPs and multiple subscribed UE denoted. Each UE
is required to connect to either a MAP or a FAP to receive
wireless services. The term AP is used to refer to a MAP or
a FAP. The FAPs utilize an open access control mechanism
that allows arbitrary nearby CUs to use the FAP's wireless
services [51]. Each SP can utilize multiple FAPs to serve its
subscribed UE. For simplicity, it is assumed that each FAP
can be used by only one SP at each transmission frame. Note
that the inter-FAP interference occurs when two FAPs allocate
the same subchannel to two different UE; however, due to the
MAP's fixed configuration to MAPs, the inter-MAP interference is negligible. Each AP has access to multiple subchannels.
To guarantee fairness among the UE, a maximum number of
subchannels allocated to each UE was set. Each AP can serve
a maximum number of UE that will be referred to as the quota
of an AP. Note that the quota of the MAPs is much larger than
the quota of the FAPs.
First each UE will form a descending order preference
list in terms of its utility over all the potentially available APs, that is, the first AP in in its list corresponds to
AP @ = argmax U UE . The MAPs are able to accommodate a
large number of the UE, and it is assumed a bid from UE to
a MAP will always be accepted and the matchings are permanent. However, because a FAP has a limited quota, in case
the number of its received bids is more than its quota, the
FAP only selects the most preferred UE, which results in the
highest FAP's utility and rejects the others. Because the UE
are subscribed to different SPs, each FAP has different lists
of temporarily matched UE, and at the end of the association
algorithm they determine which set of temporarily matched
UE will be served by each FAP.
To deploy the FAPs in their networks, the SPs should define
new tariffs within their existing billing system and charge
the UE that ask for more resources (e.g., requesting for more
subchannels) or ask for FAP access. The exact amount of this
charge is determined by another matching algorithm in [8] that
is a matching with transfer that was explained previously.
Finally, to match the FAPs to each SP an iterative matching algorithm with transfer was developed. The purpose of
the proposed algorithm is to obtain a solution to the optimization problem (7) in a distributed way by utilizing the outcome of the FM. The optimization problem that maximizes
the total satisfaction level of the UE subscribed to each SP
can be written as

IEEE Signal Processing Magazine

November 2016

117

Table of Contents for the Digital Edition of Signal Processing - November 2016