Joint TDD Backhaul and Access Optimization in Dense Small Cell Networks

This article has been accepted for publication in a future issue of this journal, but has not been
fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TVT.2014.2379013, IEEE Transactions on Vehicular Technology
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY (VT-2013-01903.R1)
Joint TDD Backhaul and Access Optimization

in Dense Small Cell Networks
Mehrdad Shariat, Emmanouil Pateromichelakis, Atta ul Quddus and Rahim Tafazolli
apacity demand in cellular networks has been growing

exponentially and this trend is expected to continue [1].
The ultimate vision is to provide consistent and reliable
communication to create the perception of infinite capacity
for end users. So far, Long Term Evolution of 3G (LTE) and
LTE-Advanced [2]-[3] are key intermediate steps in smooth
migration towards this vision of future wireless networks. In
particular, small cells and advanced relay cells have emerged

as fundamental elements in these systems [4]-[6] to enhance
the efficiency in both capacity and coverage. On the other
hand, a higher degree of interworking has been envisioned
between small cells and conventional macro cells in the future
via the notion of dual connectivity, which implies that the
user can have simultaneous connections to both macro and
small cell stations [7]-[8]. This feature presents potential
advantages such as separation between control and data
channels where macro layer provides control signaling to
small cells to enhance mobility, overhead and energy
efficiency while small cells focus on information delivery in
data plane [8]. This is the subject of an ongoing work in LTEAdvanced and beyond from Release 12 [7].
To support networks comprising several small cells, wired
fiber backhauling might be cost-prohibitive or difficult to
widely deploy in short to medium terms. Therefore, utilizing
wireless backhauling seems a promising migration path.
Emerging wireless systems have already taken initial steps in
this direction, e.g. in 3GPP LTE-Advanced, layer 3 relaying
strategies require self-backhauling on wireless LTE radio
bearers to backhaul traffic between relay and donor eNodeBs
[5]. Layer 3 relays mimic many of the functionalities in
eNodeBs in smaller scale unlike conventional layer 1 or layer
2 relays.
Wireless backhauling for small cells and relay cells can
introduce several challenges [6] when it comes to radio
resource management (RRM) between the backhaul (BH) and
the access links. Here, coupling between the two can limit the
multiuser diversity gain of radio resource allocation in the
access link. In case of TDD between the transmission links,
each packet would be received after two consecutive
transmissions, i.e. on BH link to small cell station and from
small cell station to the end user (access link). Therefore, the
efficiency of system depends upon the balance of resources
between the two links for each small cell. This requires
efficient resource partitioning on BH or access links and ratebalancing strategies between the two.
Manuscript received December 9, 20013; revised June 03, 2014; accepted

November 04, 2014. This work was supported in part by the European
Commission within the 7th Framework Program through the ICT Project
iJOIN under Grant agreement 317941 and in part by the University of Surrey,
Surrey, U.K., through the 5GIC programme.
The authors are with Institute for Communication Systems, home of the 5G
Innovation Centre, Department of Electronic Engineering, University of
Surrey, Guildford GU2 7XH, U.K (email: m.shariat@surrey.ac.uk)
A. Related Works
There exists a very large literature in the area of RRM and
resource allocation for radio access networks like [9]-[10]
where they fundamentally develop a framework for resource
allocation in the context of orthogonal frequency division
multiple access (OFDMA) systems. Also, there are several
research studies addressing resource allocation strategies in
Abstract This paper addresses the problem of joint backhaul

and access links optimization in dense small cell networks with
special focus on time division duplexing (TDD) mode of operation
in backhaul and access links transmission. Here, we propose a
framework for joint radio resource management where we
systematically decompose the problem in backhaul and access
links. To simplify the analysis, the procedure is tackled in two
stages. At the first stage, the joint optimization problem is
formulated for a point-to-point scenario where each small cell is
simply associated to a single user. It is shown that the
optimization can be decomposed into separate power and subchannel allocation in both backhaul and access links where a set
of rate-balancing parameters in conjunction with duration of
transmission governs the coupling across both links. Moreover, a
novel algorithm is proposed based on grouping the cells to
achieve rate-balancing in different small cells. Next in the second
stage, the problem is generalized for multi access small cells.
Here, each small cell is associated to multiple users to provide the
service. The optimization is similarly decomposed into separate
sub-channel and power allocation by employing auxiliary slicing
variables. It is shown that similar algorithms as previous stage
are applicable by slight change with the aid of slicing variables.
Additionally, for the special case of line-of-sight backhaul links,
simplified expressions for sub-channel and power allocation are
presented. The developed concepts are evaluated by extensive
simulations in different case studies from full orthogonalization
to dynamic clustering and full reuse in the downlink and it is
shown that proposed framework provides significant
improvement over the benchmark cases.
Index Terms backhauling, joint optimization, small cells,
rate-balancing
I. INTRODUCTION
0018-9545 (c) 2013 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE
permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
the context of relay-based as well as small cell networks [11][18]. Here, [11] presents an integrated RRM solution for relaybased networks in the context of code division multiple access
(CDMA) systems. Authors of [12]-[14] offer heuristic
resource allocation strategies for in-band relaying in OFDMA
networks. In particular, [14] focuses on uplink in-band
relaying via co-scheduling and load balancing between macro
and relay cell users. In [15], the authors explore resource
allocation and relay selection strategies for OFDMA systems
when a hybrid of amply-and-forward (AF) and decode-andforward (DF) relays, i.e. layer 1 and layer 2 relays are
employed. In [16], the possibility of concurrent transmission
between relays and macro cell networks is analyzed based on
extreme value theory. Authors of [17] proposes a cooperative
resource allocation strategy between macro and relay stations
in downlink transmission and [18] focuses on cell-association
strategies for small cells to achieve better load balancing
between macro and small cells. A detailed survey on RRM for
multi-carrier cellular networks can be found in [19].
Additionally, [20] provides an overview of radio resource
allocation schemes designed for relay-enhanced systems.
The aforementioned research studies are quite inspiring.
However, some are mainly focused on access side ([9]-[10]
and [18]), in particular for conventional macro deployments
([9]-[10]). Others propose metrics that are not readily
applicable to multi-carrier systems nor apply constraints
imposed by BH limitation ([11], [17]). On the other hand,
some solutions are heuristic in nature without detailed
mathematical insights ([12]-[14]). More importantly, the
majority of solutions assume a fixed duration of transmission
between the BH and access links, not utilizing an important
degree of freedom that we intend to explore in this paper.
B. Contributions
In this paper, we systematically decouple the problem of
wireless backhauling for small cells via decomposition theory
[21]-[22] taking into account the interaction of resource
allocation parameters in different links and layers. Here, we
extend our previous works in [23]-[24]. In [23], we proposed a
low-complexity generic framework for resource allocation
based on time-sharing applicable to different scenarios in
downlink and uplink. On the other hand, in [24], we proposed
an efficient graph-based dynamic clustering framework to
control the level of co-tier interference between small cells.
Combining the two solutions lead us to decouple the problem
of resource allocation for small cells networks across BH and
access links by introducing a set of governing variables as a
priority factor. The decoupled elements facilitate efficient
RRM strategies that are flexible and applicable to a wide set of
cases from full orthogonalization to full spectrum reuse
between small cells. In particular, we present a novel
framework to update the duration of transmissions in
downlink between the BH and access links in TDD mode in
conjunction with resource allocation in both links. Here, the
time coupling element across the links is challenging yet
presents another degree of freedom to jointly optimize the
system performance. The mathematical derivations and the
proposed solutions are novel and have not been explored in

the literature at this level to the best of our knowledge. Such
RRM strategies enable flexible cloud-based radio access
network (Cloud-RAN) that can be fully or partially driven by
the Cloud [25]-[26]. The outcome solutions are evaluated by
extensive simulations in different case studies for downlink
transmission.
II. SYSTEM MODEL
Here, the system consists of a local Cloud-RAN Gateway
(L-GW) and a set of small cell stations , transmitting
information to a set of users over the set of sub-channels .
The users are grouped into disjoint subsets such that
1,2, , . We also define equivalent
to cover the BH transmission phase. We assume 1 covers
the BH transmission while 2 refers to access link.
Figure 1 shows a typical example of our target network
with || 3, | | 5. Here, we assume that each
sub-channel is constrained to be exclusively used by a single
link within small cells per time instance. However, across
different small cells, different case studies are considered from
orthogonal allocation to dynamic clustering and finally the full
reuse of sub-channels. The channel knowledge is considered
to be available at L-GW entities connected to Cloud-RAN
across the network. The Cloud-RAN processes the subchannel and power allocation and adjusts the transmission
duration for BH and access links.
Fig. 1. Dense small cell network
III. PROBLEM FORMULATION

The resource (sub-channel, power, transmission duration)
allocation problem can be mathematically formulated as an
optimization problem with a certain objective function
subjects to involving constraints in the problem domain.
Assuming weighted sum-rate (WSR) maximization as the
optimization objective, a generic representation of the problem
is as follows:
max,, s.t.
(1)

"
,
!, $ %&',( ,
#
0 1

()
)1, 2, , + 2

"
$ 1 , - ,
# ,
(2)
(3)

, )0,1+,
. , - ,

/0, , , !, 1,
(4)
)1, 2, , +.
(5)
Here, represents the outcome average rate of user . over

allocated sub-channels across BH and access links whereas

!, is the instantaneous power that user . transmits over sub
channel - on link . , denotes the fraction of time that

user . is allowed to exclusively use this sub-channel on link
per cell. Obviously, (2) provides the power limitation
constraint per node, i.e. %&',( whereas (3) in conjunction
with (4) imposes the intra-cell orthogonality of resource
allocation, i.e. the exclusivity constraint. Equation (5) shows
that the rate per user . would be equivalent to the rate
allocated to this user out of its serving small cell, i.e. . In

depends on the
particular, it is worth noting that

, , !,
allocated share of resources

to a small cell on
different links in BH 1) and access sides ( 2), the
spectrum reuse strategy across different small cells and finally
the parameter 0 . Here, 0 0,1 is the parameter to adjust the
duration of transmission between BH and access links
assuming TDD. We will detail this dependency later in the
paper.
The grouping of (3) with (4) turns the optimization into a
combinatorial problem that is intractable for large sets of subchannels and users. Furthermore, constraint (5) couples the
resource allocation problem across both links and also
between different small cells based on the spectrum reuse
strategy.
In following sections, the joint resource allocation procedure
is addressed in detail for different scenarios.
A. Point-to-Point Small Cells (PPS)
To simplify the analysis, first, a point-to-point model is
considered where each small cell is exclusively associated to a
single user.
Problem A. (PPS without time-sharing): Maximize the
WSR of users by performing the resource allocation while
satisfying the sub-channel and power allocation constraints.
This problem is very similar to the original problem (1)-(5).
Here, we try to expand the definition of (5) from the capacity
point of view taking into account PPS model:

6
8
/0, , , !, 1 min/05 , 1 05 1

5 9 , log 8 /1 + >, !, 1.

5 is
(6)
(7)

the achievable rate of user n at link (l) whereas >,

represents the SNR or SINR density function of the
constituent link. In particular for the second phase, we have:
8
8
8
8
>, ?, B@ ?/@1, !@ , + A , C .
8
Here, ?, represents the channel gain of user . from

8
8
serving cell whereas ?/@ 1, !@ , models the resulting
interference from reusing the same sub-channel by small
cell in non-orthogonal case studies. A models the noise
power over the target sub-channel. It is important to note that

all the small cells are associated with L-GW of Cloud RAN
for the BH link while each user is coupled to its corresponding
small cell for the access link based on (7). As a result, we
6
6
6
6
have: !, !, and > , >, where is the serving
small cell for user n.
1) Relaxation
The problem A is still combinatorial due to the exclusivity
constraint. Moreover, the minimization function as in (6)
couples strongly the resource allocation across the links. To
tackle these issues, the exclusivity constraint of (4) can be
initially relaxed allowing the time-sharing of resources:

, G0,1H . , - .
(8)
The exclusivity constraint can be re-imposed to the outcome

solution, similar to the method we proposed in [23] with
negligible performance loss. Additionally, the minimization
problem can be interpreted as the following sub-problems:
Sub-problem A.1 (Bottleneck on the BH): In this regime,
the achievable rate of access, from small cells to their
associated users, is greater than the BH. As a result, this subproblem is formulated as follows subject to the constraints (2),
(3) and (8):
6
max,, 05 s.t.
(9)
6
05
8
$ 1 05 , . .
(10)
As it can be seen, constraint (10) is employed to ensure the

operation within the assumed regime.
Sub-problem A.2 (Bottleneck on the access): Here, contrary
to the previous case, the BH is dominant due to the superiority
of its channel condition. Hence, this sub-problem is
formulated as follows subject to similar power and timesharing constraints as A.1:
8
(11)
max,, 1 05 s.t.
1 058 $ 056 , . .
(12)
Similar to sub-problem A.1, constraint (12) monitors the

operation within the boundaries of the assumed regime. It can
be easily shown that any other combination of rates can be
mapped into one of the aforementioned regimes through
rearranging the resources among the users. Hence, by
assuming the operation in either regime, the relaxed problem
can be efficiently solved.
Remark 1: Sub-problems A.1 and A.2 are not generally
convex due to presence of inter-cell interference on the access
links in case of spectrum reuse. However, by applying
dynamic clustering framework proposed in [24], users of
different cells can be effectively grouped into clusters with
low-level of intra-cluster interference. Therefore, we decouple
interference coordination problem from resource allocation by
mapping the problem to this graph-based solution. This leads
to near-convexity of the sub-problems above, enabling us to
effectively decompose them as will be outlined below.
To tackle the relaxed problems, we initially focus on A.1

and we form partial Lagrangian as below subject to constraints
(2), (3) and (8):
6
I, , 0, J 05
K / 056 1 058 1.
!, = argmax {R log 8 /1 + >, !1 TU !},

. , - ,

(14)
where TU are the power prices to satisfy (2) and R are the
balancing parameters to govern the balance between both
links according to either (10) or (12).
Proof: It is straightforward to show (14) by applying
standard dual decomposition with dual variables TU for (2)
and considering K as (13), where:
K = 1

R = V
K = 2

(15)
Following a similar approach for A.2, the same optimality

condition holds unless (15) that should be amended as follows:
R = V

K = 1
K = 2
(16)
Corollary 1: The solution to the power allocation problems

(14) is similar to the standard multi-level water-filling as in
(17) where [W]X = max 0, W.

!,
R
1
=Y
Z .
ln2 TU >
,

sub-problems A.1 or A.2 will be the solutions of the following

set of optimization problems:
6, , , \, = argmax ] R log 8 ^1 +

(17)
Proof: A straightforward result of Lemma 1.
3) Time-share allocation
Similar to power allocation scenario, by fixing the average

powers [, = !, , and , the time-share variables can be
re-tuned to increase the aggregate WSR across the network.
Lemma 2: For fixed allocated average powers and
transmission duration 0, the optimal time-share variables for

>, [, B _` , - ,
(13)
Here, the partial dual problem can be formed as LJ

sup,, I, , 0, J subject to time-share and power
constraints where K (Q5 J are dual prices introducing a set
of balancing parameters between BH and access links. As it
can be seen, the sub-problem A.1 can be easily decoupled into
two weighed sum-rate problems for both links based on (13)
6
8
with weights R K and R K , respectively.
Similar decoupling can be achieved for A.2 by simply
swapping the weights. This partial dual decomposition enables
us to apply power and time-share allocation optimality
conditions as derived in [23] with slight tuning of the weights
for individual links.
2) Power allocation
Lemma 1: For a fixed time-sharing policy and
transmission duration , the optimal power allocation
variables for sub-problems A.1or A.2 will be the solutions of
the following set of optimization problem:

(18)
subject to the constraints (3) and (8).

Proof: By fixing the average allocated powers and the
transmission duration, the optimization will be decomposed to
the above equations per sub-channel as the power constraints
(2) are the only factors that couple the optimization problem
across the sub-channels [23].
Corollary 2: The time-share solutions of equation (18) can

be found by solving the following set of equations:
a /b 1 = R clog 8 /1 + b 1

i ,

e
jU,k
, le
U,k
=>
d e
fg8/6Xd e 1
h=
(19)
where a /b 1 is the marginal utility of user n at link (l) and

i are the time-share prices to satisfy the constraints (3) and
(8).
Proof: This is a direct result of Lemma 2 [23].

4) Adjusting the duration of transmission

Theorem 1: There is an optimum value of where the
solutions of both sub-problems A.1 and A.2 converge that is
the optimal solution of problem A where:
5
8
56
0
, . , 0 0,1.
1 0
(20)
Proof: Considering the sub-problem A.1 and (13), the

tuning parameters can be iteratively updated based on the subgradient method as follows:
[Jn

JnX6
=

6nX6
op 0 q
1
0q8nX6
]rJ ,

(21)
where [ ]rs represents the projection of the dual variable on

the corresponding domain (K > 0. op is the coefficient that
regulates the step-size in the (sub)gradient direction and can
be chosen constant but sufficiently small to guarantee the
convergence [21].
In similar manner for sub-problem A.2, the tuning
parameters can be updated as follows:
[Jn

JnX6
=

+ op 1 0q8nX6
0 q6nX6
]rJ .

(22)
As the dual problem converges to the optimal dual prices

(p : JnX6
Jn

), we can clearly see that, there is a
unique solution for 0 as below for both sub-problems:
8n
5
Limit n{ 6n
= 0,
8n
(23)
5
+ 5
. where 0 0,1.
Considering the PPS model, each user is exclusively

n
attached to a single small cell; therefore we can define 0 as
below for each small cell per iteration:
5
8n
5
+ 5
. , {1, 2, , }.
6n
8n
(24)
Considering 0 as the optimal transmission duration for

small cell at iteration p , the transmission durations might
evolve independently for different cells across different
iterations of the optimization process. However, they will
converge to the optimal 0 asymptotically based on Theorem
1. Defining average 0 as (25), it is straightforward to show
that this average transmission duration value also converges
asymptotically to optimal 0 based on (23) and (24).
n
n
0 n = 0
(25)
This averaging mechanism provides a unique value per

iteration across all cells in order to update dual prices based on
(23) or (24) that it asymptotically reaches optimal solution.
To understand the optimality condition, it is instructive to
examine the effect of adjusting the transmission duration on
the joint capacity of BH and access links:
By definition, the instantaneous capacity region is a
set that consists of all the achievable rate vectors for the
current channel state vector under the constraints of a given
resource allocation policy . For example, in problem A, (2)
provides the constraints on the power allocation whereas (3)
and (4) impose the constraints on the time-sharing policy .
Mathematically, the instantaneous capacity region can be
considered as the union of all achievable rate vectors under the
considered policy:
= where ( satisfies .
(26)
In the case of joint BH and access links, each element of a
typical achievable rate vector is the minimum of achievable
rates (of a target user) on both links according to (6). Based on
this assumption, for a fixed , the total capacity region will be
the intersection of corresponding capacity regions on different
links:
(0, = (0, ) /1 0, 1.
6
8
(27)
Here, increasing the transmission duration expands the

capacity region of the BH while it will shrink the region on the
access link and vice versa. The capacity region for any value
of can be considered as a typical set of achievable rate
vectors. As a result, the union of these typical sets can form
the total capacity region based on (26):
= 0, .
Capacity region of BH (=0.5)

Capacity region of Access (=0.5)
Joint capacity region
(28)
In the joint optimization problem (in either links) for a

given weight vector, the optimal solution lies on the boundary
of the capacity region to guarantee optimality. According to
Theorem 1, there should be an optimal where the solutions
of both sub-problems converge. Intuitively, this unified
user 2 rate: r(l)

2
n
solution should be located on the boundary of the capacity

regions of both links. In other words, the boundaries of both
links will intersect at the optimal . Consequently, the
outcome boundary of the joint capacity region can be achieved
by tracing intersection points for different values of . Figure
2 shows the boundaries of the capacity region across both
links for a two-user scenario. As it can be seen, the
intersection points of the corresponding boundaries (with the
same values of ) form the boundary of the joint capacity
region in thick solid line. Figure 3 illustrates the capacity
regions of the same scenario for a uniform quantized set of
(with 100 levels). As shown, the joint capacity region can be
constructed as the union of capacity regions for different
values of .
user 1 rate: r(l)

1
Fig. 2. The Capacity boundaries in joint BH and access links
Capacity region of BH
Capacity region of Access
Joint capacity region
user 2 rate: r(l)

2
0 =
user 1 rate : r(l)

1
Fig. 3. The Construction of joint capacity region for small cells
5) Sub-gradient method with small cell grouping

As it was shown, the optimality conditions for both subproblems A.1 and A.2 follow a generic optimality conditions
according to (17) and (19) per link where tuning
parameter K in conjunction with duration of transmission
balance the operation across both links. In other words, the
newly emerged tuning parameters act as a priority factor in the
power and time-share allocation similar to weighting
parameters . As a result, there is a direct interaction
between the resource allocation and rate-balancing between
the BH and the access links. Therefore, each allocation policy
introduces a new set of balancing parameters whereby each
Algorithm 1: Optimal Rate-Balancing

1.
2.
3.
4.
5.
6.
Initialization: choose a sufficiently small uniform

and n .
set for Jn

Rate update: given Jn
, calculate
6nX6
8nX6
5
and 5
according to (17), (19) and (7).
Transmission duration update: update the
transmission duration parameter according to (24)
and (25).
Small cell grouping: classify the small cells (users)
according to either (10) or (12).
Balancing update: update the balancing

parameters JnX6 according to small cell grouping
sub-algorithm.
Termination: stop when the total balance is achieved
between all BH and access links otherwise go to 2.
40
Negative rate difference (unsatisfied)

Positive rate difference (oversatisfied)
Relative rate difference, Mb/s
30
20
10
-10
-20
-30
-40
0
10
15
20
25
30
35
40
45
iteration number
Fig. 4. Evolution of relative rate difference between access and BH links

based on algorithm 1
13.5
12.75
WSR, Mb/s
tuning set will directly affect the resource allocation.

Remark 2: In practice, as the resources are discrete (in the
form of sub-channels), the resource allocation is quite
sensitive to the tuning of balancing parameters. Therefore to
come up with efficient and practical resource allocation
strategies, some important observations should be taken into
account:
As the rate-balancing governs the coupling between
the links, it requires a slower time-scale compared to
the power and time-share update to ensure stability
and convergence.
The resource allocation is sensitive to the relative
variations of tuning parameters rather than their
absolute values.
The above remark leads us towards a coherent solution to
update the tuning variables:
Small cell grouping (sub-algorithm): Per iteration of
optimization, small cells can be classified based on satisfying
or violating the balancing constraint that is associated with the
problem. Considering the PPS model, each user is exclusively
attached to a single small cell. Therefore, the constraints are
equivalent to per user constraints. As an example, in subproblem A.1, users that satisfy the inequality (10) per iteration
can be grouped as (over)satisfied users whereas others form
the unsatisfied group. Intuitively, the tuning parameters of
(over)satisfied users (or corresponding cells) can remain
unchanged while the unsatisfied users are jointly updated with
a similar increment. This method can avoid the unnecessary
swapping of resources (sub-channels) inside (over)satisfied or
unsatisfied groups and gradually increases the priority factor
of the unsatisfied ones. As a result, the resources are more
steadily transferred from the (over)satisfied group to
unsatisfied one till the total balance is achieved for all the
users (cells).
Based on this sub-algorithm, we can summarize the
outcome joint resource allocation algorithm for problem A as
below:
12
11.25
10.5
9.75
0
10
15
20
25
30
35
40
45
iteration number
Fig. 5. Evolution of WSR in the network based on algorithm 1
Remark 3: In algorithm 1, the rate vectors of both links are

updated based on the balancing parameters of the previous
iteration. In practice, if the variations of balancing parameters
are small compared to the weighting factors of users , it is
possible to fix the priority factor on the bottleneck link
according to either (15) or (16). This approximation, even
though sub-optimal simplifies the outcome algorithm by just
tuning the rates on one link. This form is termed as suboptimal rate-balancing in the simulation section.
Figure 4 and Figure 5 illustrate a typical evolution of
relative rate difference (imbalance) and WSR according to
algorithm 1 in a two-user (two-cell) case. As shown, by each
iteration, the sub-channels are transferred from the
(over)satisfied user with positive rate difference to the
unsatisfied one with negative rate difference and the
transmission durations are adjusted accordingly. This
procedure continues iteratively till the total balance is reached
in both cells.
B. Multi-access Small Cells (MAS)
Here, the developed model of PPS is further extended for
the MAS where each cell is associated with multiple users.
1) Problem formulation
In the case of MAS, the resource allocation in the BH link is
not exclusive to one user per small cell. In other words, the
achieved rate on the BH can be partitioned among all

associated users of that small cell. To enable effective
partitioning of resources in the BH, we define new auxiliary
6
6
6
variables 5 . Unlike 5 in PPS case, 5 are not directly
defined based on the resource allocation policy as in (7) but
they rather act as slicing variables to partition the resources
allocated to a small cell station among its associated users
[21]. This leads us to a new problem definition for MAS
scenario:
Problem B. (MAS without time-sharing): This problem is
similar to problem A. However, (7) should be defined
separately for the BH and access links as below:

= , Q?8 /1 + >, !, 1,
(29)
= , Q?8 /1 + >, !, 1.
(30)
6
5
8
6
8
6
8 8
Here, represents the rate associated to the small cell

in the BH.
2) Relaxation and decomposition
Similar to the problem A, the exclusivity constraint can be
relaxed through the time-sharing as in (8). Moreover, the
minimization problem is decomposed into the following subproblems:
Sub-problem B.1 (Bottleneck on the BH): Similar to subproblem A.1 where (29)-(30) replace (7).
Sub-problem B.2 (Bottleneck on the access): Similar to
sub-problem A.2 where (29)-(30) replace (7).
6
6
Please note auxiliary variables 5 substitute 5 in both
sub-problems.
In sub-problem B.1, by forming the partial Lagrangian
subject to the power and time-share constraints, we have:
I, , 0, = 05
056
1
6
8
05 .
(31)
As mentioned, 5 are not directly related to the resource

allocation policy. As a result, algorithm 1 cannot be employed
immediately based on sub-problem B.1.
3) Resource slicing
However, the following relationship should be satisfied
6

between
and auxiliary variables 5 as the feasibility
constraint for the slicing variables:
6

" 56
.
(32)
This implies that at the optimal point of operation with full

6

balance, equality is achieved i.e. " 5 =
. Therefore,
a new partial Lagrangian can be formulated taking into
account the above as an additional constraint:
I, , 0, , = 05
6
056 1 058
6

/ " 5 1.
(33)
As it can be seen, two sets of price variables emerge in this
sub-problem; are price parameters to balance the

achievable rate per user on access link to the sliced quota of
6

the BH, determined by 5 whereas
are dual variables in
charge of matching the BH capacity to the aggregate sliced
quota per small cell.
The dual form of sub-problem B.2 results in different price
variables. In particular, this form can be more favorable due to
8
direct relation between the access rates 5 and the allocation
policy according to (30) which is similar to (7) in problem A.

Here, the equality relationship between
and the auxiliary
6
variables 5 at full balance can be utilized to better decouple
the problem. Therefore, we can initially relax constraint (12)
according to (32) to the following set of inequalities:

"1 058 0
, .
(34)
As a result, the problem maps to the following partial

Lagrangian subject to the power and time-share constraints:
I, , 0, = 1 05
8
8

/ "1 05 0 1.
( 35)
As it can be seen, the above closed form decouples the

problem in the BH and access links. Here, Corollaries 1 and 2
can be efficiently utilized for the power and time-share
allocation with the following slight modification:
= 1

R = V
= 2
(36)
where is the tuning set similar to K that monitors the ratebalancing across the links. Unlike K , the new tuning set

recovers the allocated rates to the BH of cells
rather than
users on the first link. Rate-balancing algorithm (algorithm 1)
can still be employed to calculate the optimal solution of
6
8
problem B (via B.2) provided that 56 and 5 are substituted
8

with
and " 5 respectively in steps 3 to 5. Please
note that auxiliary variables can be immediately recovered
from (12) based on 58 and 0 assuming the equality holds.
This would satisfy already relaxed constraint (12) and also
(32).
4) Complexity analysis
Having the generic picture of the proposed algorithm based
on MAS, here, we explore the complexity order of the
algorithm. As a recap of notations, , , stand for the total
number of users, sub-channels and small cells, respectively.
We also assume that iterations will be required till the
algorithm 1 converges. Each iteration of algorithm involves
time share and power allocation updates according to (17) and
(19). Assuming single iteration for time-share update and
water-filing based power allocation, the complexity of these
sub-problems will be and 8 for the BH and and
8 for access links. The duration of transmission update
requires executions. Similarly, small cell grouping requires
executions. There will be one iteration update for joint
tuning of parameters according to the proposed algorithm.

Therefore, summing up different steps, the whole complexity
of the algorithm will be in the order of [ + 8 +
+ + 2 + 1]. We will provide more details on the
convergence of algorithm and the required iterations for
convergence in the simulation studies section.
C. Special Case
As a special case, a high capacity BH can be considered
where a Line-of-Sight (LoS) symmetric connection is present
between the L-GW of Cloud-RAN and its associated small
cells. This assumption can simplify the rate-balancing
procedure as we illustrate below in problem C.
Problem C. (MAS with LoS BH): Maximize the WSR on
the access link subject to resource allocation constraints:
max,, 1 05
8
s.t.
1 058 0&' .
(37)
(38)
Here, &' shows the total achievable rate of the BH link

that should be sliced among different small cells based on
their access requirements.
Considering partial Lagrangian (subject to the resource
allocation constraints), we have:
8
I , , 0, = 1 05
(39)
8

/
1 05 0&' 1.
controls
As it can be seen, here, a single tuning parameter

the balancing of rates among the small cells. Therefore, to
meet the balance the transmission duration parameter can be
directly tuned as follows:
58
(40)
0 =
8
&' + 5
where 5 can be directly calculated based on their respective

weights of ) assuming .
8
IV. SIMULATION STUDIES
users are assumed to be almost stationary so the path loss and

shadowing values are fixed during the simulation duration.
The samples are averaged over 2000 independent snapshots.
Furthermore, n and Jn
are set to 0.5 and 0.001,

respectively. For the symmetric (Sym) cases, the small cells
are located at the perimeter of a circle with 200 m distance
from the L-GW; Users are located at equal distance from their
respective small cells (500 m). In asymmetric (Asym)
scenarios, one small cell is shifted to the distance of 900 m;
however, the associated user keeps the same distance as before
to it.
At first, a two-user case is considered under symmetric and
asymmetric bottleneck conditions. Figure 6 shows the
simulation result for this case. As it can be seen, ratebalancing provides considerable improvement in the
efficiency of the resource allocation compared to the cases
with independent resource allocation across BH and access
links. In the case of symmetric bottleneck condition, the suboptimal algorithm provides near optimal performance whereas
in asymmetric bottleneck condition, the gap to the optimal
solution increases. This is consistent with our theoretical
inspection in previous sections.
Figure 7 shows the simulation result for a three-user case.
Here, similar trend is observable in both bottleneck conditions
Table I: Simulation parameters for case studies in A [27]
Small cell
BH
Access
System bandwidth
5 MHz
Carrier frequency
2.0 GHz
Total TX power
37 dBm
Distance-dependent
path loss
Shadowing
Hotzone Model 1
Hotzone Model 1
(Outdoor Macro)
(Outdoor Pico)
Lognormal, zero mean, 8 dB standard
deviation
Rayleigh block fading
Variable based on scenario
Fast fading channel

UE / small cell dropping
In this section, the developed concepts and algorithms are

evaluated by means of extensive simulations.
Without Rate-balancing (Sym)
Sub-optimal Rate-balancing (Sym)
Optimal Rate-balancing (Sym)
Without Rate-balancing (Asym)
Sub-optimal Rate-balancing (Asym)
Optimal Rate-balancing (Asym)
20
WSR, Mb/s
A. Case studies with orthogonal allocation between small

cells
Here, we consider topologies with orthogonal allocation
between small cells. This enables us to focus on the efficiency
proposed algorithms in different cases. We run Monte Carlo
simulations based on the parameters consistent with 3GPP
specifications [27] to examine the effect of bottleneck links on
the considered algorithms. Furthermore, we assume
normalized weights across the case studies here. Some
common simulation parameters of case studies in A are
presented in Table I.
15
10
Scenarios
1) PPS Scenario
Initially, we assume that PPS model is employed where
each small is associated to a single user in its vicinity. The
Fig. 6. Evaluating algorithmic efficiency of rate-balancing for A-1 (M=2)
30
26
Without Rate-balancing (Sym)

Sub-optimal Rate-balancing (Sym)
Optimal Rate-balancing (Sym)
Without Rate-balancing (Asym)
Sub-optimal Rate-balancing (Asym)
Optimal Rate-balancing (Asym)
25
Without Rate-balancing
Sub-optimal Rate-balancing
Optimal Rate-balancing
24
22
20
20
WSR, Mb/s
WSR, Mb/s
15
18
16
14
10
12
5
10
Scenarios
2) MAS Scenario
In second scenario, the MAS model is utilized where each
small cell is associated with multiple users. Here, the users are
assumed to be mobile. As a result, the shadowing and path
loss values changes across different snapshots. The samples
are averaged over 2000 independent snapshots. n and
are set as PPS case. The small cells are located at the
Jn

perimeter of a circle with 500 m distance from the L-GW and
the users are randomly but uniformly distributed in the vicinity
of each small cell with a similar maximum distance from their
respective small cell station.
Figure 8 shows the simulation result for different numbers
of users in this scenario (M=2). As expected, increasing the
number of users enhances the WSR in all cases due to the
better multiuser diversity. Rate-balancing provides significant
improvement in the efficiency of resource allocation compared
to the benchmark algorithms. This scenario indicates that in a
realistic case where a mixture of symmetric and asymmetric
bottleneck conditions exists, the performance of the suboptimal algorithm is quite close to the optimal solution. As a
result, the sub-optimal algorithm can be considered as an
alternative to the optimal one.
B. Case studies with spectrum reuse between small cells
In these case studies, we evaluate the performance of the
proposed algorithm in a more detailed system-level topology
again consistent with 3GPP specifications [27]. In particular,
we examine the impact of different interference coordination
strategies from dynamic clustering to full reuse. The common
simulation parameters of case studies in B are presented in
Table II. Here, we have an outdoor random cell and user
deployment in a cluster in line with scenario 2 of [28]. In
particular, four small cell stations are randomly dropped in a
cluster area (ring) with minimum distances as outlined in
Table II. Concerning the intra-cell scheduling, proportional
fair (PF) scheduling is used for a multi-channel system per
small cell to provide a fair allocation of resources between
multiple users. Therefore, user weights are tuned based on this
algorithm and are normalized to total weights per cell for a fair
comparison. The samples are averaged over 5000 independent
snapshots. Furthermore, n and Jn
are set as case

studies in A.
Number of Users

Table II. Simulation parameters for case studies in B [27]
Small cell
System bandwidth
Carrier frequency
Total TX power
Distance-dependent path loss
Shadowing
Antenna gain / connector
loss
Fast fading channel
UE dropping
Radius for small cell
dropping in a cluster
Minimum small cell station
to UE distance
Minimum small cell station
to base station distance
BH
Access
10 MHz
2.0 GHz
24 dBm
Macro-to-Relay
Relay-to-UE
(Outdoor)
(Outdoor)
Lognormal, zero mean, 10 dB
standard deviation
5 dBi (omni) / 0 dBi
Rayleigh block fading
8 users per cell (32 in total)
90 m
5m
20 m
1) Dynamic clustering
The proposed solution is initially studied under the dynamic
clustering as interference coordination [24]. Here, we compare
three RRM schemes:
Without rate-balancing, where independent resource
scheduling happens on the BH and access links
Fixed partitioning, where the BH resource are equally
partitioned between the small cells in a static manner,
Rate-balancing, where we employ the proposed scheme
combined with dynamic clustering.
We consider two performance indicators, CDF of
normalized WSR (per cell) and CDF of user spectral
efficiency in (b/s/Hz).
As it can be seen in Figure 9 and Figure 10, the proposed
rate-balancing scheme can provide significant improvement in
both user spectral efficiency and WSR (per cell) in this case
study for both performance indicators. It is also interesting to
note that fixed partitioning on the BH can be generally a better
strategy compared to independent but dynamic RRM in BH
and access links considering the results without ratebalancing. However, fixed partitioning can be an expensive
solution for small cell operation as it yields low BH utilization
when it comes to small cells in low-load scenarios.
0.9
0.9
0.8
0.8
0.7
0.7
0.6
0.6
CDF
CDF
0.5
0.5
0.4
0.4
0.3
0.3
0.2
0.2
Without Rate-Balancing
Fixed Partitioning
Rate-Balancing
0.1
0
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
Fixed Partitioning
Rate-Balancing
0.1
0
0
0.5
0.05
0.1
0.15
Normalized WSR (per cell)
0.25
0.3
0.35
0.4
0.45
0.5
Fig. 11. CDF of normalized WSR for case study B-2

1
0.9
0.8
0.8
0.7
0.7
0.6
0.6
CDF
1
0.9
0.5
0.5
0.4
0.4
0.3
0.3
0.2
0.2
Fixed Partitioning
Rate-Balancing
0.1
0
0
0.2
Normalized WSR (per cell)
Fig. 9. CDF of normalized WSR for case study B-1
CDF
10
0.5
1.5
2.5
Fixed Partitioning
Rate-Balancing
0.1
0
0
3.5
0.5
2.5
3.5
0.6
0.65
Fig. 12. CDF of user spectral efficiency for B-2
Fig. 10. CDF of user spectral efficiency for B-1

450
400
350
300
Frequency
2) Reuse one
At this case study, we apply the rate-balancing mechanism
in reuse one without any active interference coordination. The
results are compared for similar RRM strategies and
performance indicators as B-1.
Figure 11 and Figure 12 show simulations results of this
case study. As it can be seen, similar trend is observable as B1 for the performance indicators in particular for CDF of WSR
(per cell). It is worth noting that in reuse one case, the
optimization problem and related capacity regions on BH and
access links are not generally convex. As a result, the duality
gap of proposed solution will be non-zero. Nevertheless, the
system performance can still benefit from rate-balancing and
the joint optimization as it is evident by the results. However,
the major gain of the algorithm is achieved when ratebalancing strategies are employed in conjunction with
interference coordination algorithms like dynamic clustering.
3) Analysis on evolution of and convergence rate
In this sub-section, we provide more detailed analysis on
evolution of as the adjusting parameter for the duration of
phases in the dynamic clustering case. Furthermore, we
examine the convergence rate of the algorithm across different
snapshots for the same case.
1.5
User Spectral Efficiency (b/s/Hz)
User Spectral Efficiency (b/s/Hz)
250
200
150
100
50
0.3
0.35
0.4
0.45
0.5
0.55
Optimal Value of
Fig. 13. Histogram of optimal value of
Figure 13 shows the histogram of optimal value of across

different snapshots. As it can be seen, the value is dynamically
adjusted based on the instantaneous channel conditions in both
links across different snapshots.
Figure 14 illustrates the histogram of the required number
of iterations (for convergence) across different snapshots. As
shown, the algorithm can converge in all snapshots with fewer
than 250 iterations for . Specifically, the majority of
snapshots will converge in around 100 iterations.
[7]
900
800
[8]
700
600
[9]
Frequency
500
400
[10]
300
200
[11]
100
50
100
150
200
250
No. of Iterations till Convergence
Fig. 14. Histogram of required number of iterations
V. CONCLUSIONS
In this paper, the problem of joint resource allocation
between BH and access links in TDD was addressed for dense
small cell networks. The problem was mathematically
decomposed into per link sub-channel and power allocation
where a set of rate-balancing parameters combined with phase
duration parameter governed the coupling among the links.
Moreover, novel algorithms were derived for rate-balancing
by employing the concepts of small cell grouping and resource
slicing. Finally, the efficiencies of proposed concepts and
algorithms were evaluated by system-level simulations. As
shown, joint optimization with rate-balancing could provide
significant improvement over independent resource allocation
across BH and access links and also fixed BH partitioning
strategies. In particular, the proposed algorithm could deliver
significant performance improvement in conjunction with
efficient interference coordination strategies like dynamic
clustering.
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
VI. ACKNOWLEDGMENT
The authors gratefully acknowledge the contributions of Dr.
Reza Hoshyar in line with the earlier work [23]. The views
and conclusions contained here are those of the authors and
should not be interpreted as necessarily representing the
official policies or endorsements, either expressed or implied,
of the iJOIN project or the European Commission.
REFERENCES
[1]
[2]
[3]
[4]
[5]
[6]
Requirements, evaluation criteria, and submission templates for the

development of IMT-Advanced, ITU-R, Accessed May 2014, [Online]
Available: http://www.itu.int/md/R07-SG05-C-0068/
Evolved Universal Terrestrial Radio Access (E-UTRA) and Evolved
Universal Terrestrial Radio Access Network (E-UTRAN), Overall
Description, 3GPP TS 36.300, Mar. 2014
Further Advancements for E-UTRA Physical Layer Aspects, 3GPP TR
36.814, Mar. 2010
A. Damnjanovic, J. Montojo, Yongbin Wei, Tingfang Ji, Tao Luo, M.
Vajapeyam, Taesang Yoo, Osok Song, and D. Malladi, A survey on
3GPP heterogeneous networks, Wireless Commun., vol.18, no.3, pp.1021, Jun. 2011
H. Raza, A brief survey of radio access network backhaul evolution:
part II, IEEE Commun. Mag., vol.51, no.5, pp.170-177, May 2013
C. Hoymann, Wanshi Chen, J. Montojo, A. Golitschek, C. Koutsimanis,
and Xiaodong Shen, Relaying operation in 3GPP LTE: challenges and
solutions, IEEE Commun. Mag.,vol.50, no.2, pp.156-162, Feb. 2012
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
11
C. Hoymann, D. Larsson, H. Koorapaty and Jung-Fu Cheng, A Lean

Carrier for LTE, IEEE Commun. Mag., vol.51, no.2, pp.74-80, Feb.
2013
H. Ishii, Y. Kishiyama, and H. Takahashi, A novel architecture for
LTE-B :C-plane/U-plane split and Phantom Cell concept, in Proc.
IEEE Globecom Workshops,2012, pp.624-630
G. Song and Ye Li, Cross-layer optimization for OFDM wireless
networks-part I: theoretical framework, IEEE Trans. Wireless
Commun., vol.4, no.2, pp.614-624, Mar. 2005
G. Song and Ye Li, Cross-layer optimization for OFDM wireless
networks-part II: algorithm development, IEEE Trans. Wireless
Commun., vol.4, no.2, pp.625-634, Mar. 2005
Y. Liu, R. Hoshyar, X. Yang, and R. Tafazolli, Integrated Radio
Resource Allocation for Multihop Cellular Networks With Fixed Relay
Stations, IEEE J. Sel. Areas Commun., vol.24, no.11, pp.2137-2146,
Nov. 2006
Lei Huang, Mengtian Rong, Lan Wang, Yisheng Xue, and E. Schulz,
Resource Allocation for OFDMA Based Relay Enhanced Cellular
Networks, in Proc. IEEE VTC, 2007, pp.3160-3164
Zhuyan Zhao, Jian Wang, S. Redana, and B. Raaf, Downlink Resource
Allocation for LTE-Advanced Networks with Type1 Relay Nodes, in
Proc. IEEE VTC, 2012, pp.1-5
K. Dereje Woldemedhin, O. Bulakci, A. B. Saleh, S. Redana, and F.
Granelli, Joint backhaul co-scheduling and relay cell extension in LTEadvanced networks uplink performance evaluation, in Proc. European
Wireless, 2012, pp.1-8
T.C.Y. Ng and Wei Yu, Joint optimization of relay strategies and
resource allocations in cooperative cellular networks, IEEE J. Sel.
Areas Commun., vol.25, no.2, pp.328-339, Feb. 2007
O. Oyman, Opportunistic scheduling and spectrum reuse in relay-based
cellular networks, IEEE Trans. Wireless Commun., vol. 9, no. 3, pp.
10741085, Mar. 2010
Qian Li, R.Q. Hu, Yi Qian, and Geng Wu, Intracell Cooperation and
Resource Allocation in a Heterogeneous Network with Relays, IEEE
Trans. Veh. Technol., vol.62, no.4, pp.1770-1784, May 2013
R. Madan; J. Borran, A. Sampath, N. Bhushan, A. Khandekar, and
Tingfang Ji, Cell Association and Interference Coordination in
Heterogeneous LTE-A Cellular Networks, IEEE J. Sel. Areas
Commun., vol.28, no.9, pp.1479-1489, Dec. 2010
S. Sadr, A. Anpalagan, and K. Raahemifar, Radio Resource Allocation
Algorithms for the Downlink of Multiuser OFDM Communication
Systems, Commun. Surveys Tuts., vol.11, no.3, pp.92-106, 3rd quarter
2009
M. Salem, A. Adinoyi, M. Rahman, H. Yanikomeroglu, D. Falconer, Y.D. Kim, and Y.-C. Cheong, An overview of radio resource
management in relay-enhanced OFDMA-based networks, Commun.
Surveys Tuts., vol.12, no.3, pp.422438, 3rd quarter 2010
D.P. Palomar and Mung Chiang, A tutorial on decomposition methods
for network utility maximization, IEEE J. Sel. Areas Commun., vol.24,
no.8, pp.1439-1451, 2006
D.P. Palomar and Mung Chiang, Alternative Distributed Algorithms for
Network Utility Maximization: Framework and Applications, IEEE
Trans. Autom. Control, vol.52, no.12, pp.2254-2269, Dec. 2007
R. Hoshyar, M. Shariat, and R. Tafazolli, Subcarrier and Power
Allocation with Multiple Power Constraints in OFDMA
Systems, IEEE Commun. Lett., vol.14, no.7, pp. 644-646, Jul. 2010
E. Pateromichelakis, M. Shariat, A. Quddus, M. Dianati, and R.
Tafazolli, Dynamic Clustering Framework for Multi-Cell Scheduling in
Dense Small Cell Networks, IEEE Commun. Lett., vol.17, no.9,
pp.1802-1805, Sep. 2013
EU FP7 project iJOIN, iJOIN | Interworking and JOINt Design of an
Open Access and Backhaul Network Architecture for Small Cells based
on Cloud Networks, Accessed May 2014, [Online] Available:
http://www.ict-ijoin.eu
P. Rost, C. J. Bernados, A. De Domenico, M. Di Girolamo, M. Lalam,
A. Maeder, D. Sabella, and D. Wubben, Cloud Technologies for
Flexible 5G Radio Access Networks, IEEE Commun. Mag., vol.52,
no.5, May 2014
Evolved Universal Terrestrial Radio Access (E-UTRA): Further
advancements for E-UTRA physical layer aspects, 3GPP TR 36.814,
Mar. 2010
Small cell enhancements for E-UTRA and E-UTRAN-Physical layer
aspects, 3GPP TR 36. 872, Dec. 2013
12
Mehrdad Shariat received the BSc

degree
in
Telecommunications
Engineering from Iran University of
Science and Technology, Iran in 2005.
He received the PhD degree in Mobile
Communications from University of
Surrey, UK in 2010. Since 2010, he has
been working as a research fellow in the
Institute for Communication Systems
(ICS), home of the 5G Innovation
Centre, University of Surrey, UK. He has been involved in
several UK and EU co-funded projects including Mobile VCE
(Core 4), BeFEMTO, iJOIN, MiWaveS and collaborative
industry projects on small cell backhauling. His research
interests include cross-layer optimization, radio resource
management and packet scheduling for backhaul and access in
cellular and mesh networks.
Atta ul Quddus received the MSc

degree in Satellite Communications and
PhD degree in Mobile Cellular
Communications, both from University
of Surrey UK in 2000 and 2005,
respectively. He is currently a Lecturer
in Wireless Communications in the
Department of Electronic Engineering,
University of Surrey UK. During his
research career, he has led several
national and international research projects that also
contributed towards 3GPP standardization. In 2004, he also
won the Centre for Communications Systems (CCSR)
Research Excellence Prize sponsored by Vodafone for his
research on Adaptive Filtering algorithms. His current
research interests include Machine Type Communication,
Cloud Radio Access Networks, and Device to Device
Communication.
Emmanouil Pateromichelakis received

his Diploma in Information and
Communication Systems Engineering
from University of the Aegean, Samos,
Greece, in 2008. He received his MSc.
and
PhD.
degree
in
Mobile
Communications from University of
Surrey, UK in 2009 and 2013
respectively, and since then has been working as a research
fellow in the Institute for Communication Systems (ICS),
home of the 5G Innovation Centre, University of Surrey, UK.
His research interests include Radio Resource Management,
Multi-cell Cooperation and Scheduling for OFDMA Cellular/
Multi-hop
Networks.
Rahim Tafazolli received the B.Sc.,

M.Sc., and Ph.D. degrees, all in
electrical engineering. He is currently
the Director of the Institute for
Communication Systems (ICS) and the
5G Innovation Center, Faculty of
Engineering and Physical Sciences,
University of Surrey, Surrey, UK. He is
also the Chair of the European Union
NetWorks Technology Platform Expert Group and a board
member of the UK Future Internet Strategy Group (UKFISG). Furthermore, he is the author of more than 500
research papers in refereed journals and international
conferences and has been an Invited speaker. He is also the
editor of two books on Technologies for Wireless Future
published by Wileys Vol.1in 2004 and Vol.2 in 2006. Prof.
Tafazolli was appointed as a Fellow of the Wireless World
Research Forum in April 2011 in recognition of his personal
contribution to the wireless world and for heading one of the
Europes leading research groups.

Joint TDD Backhaul and Access Optimization in Dense Small Cell Networks

Uploaded by

Copyright:

Available Formats

Joint TDD Backhaul and Access Optimization in Dense Small Cell Networks

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Joint TDD Backhaul and Access Optimization in Dense Small Cell Networks

Uploaded by

Copyright:

Available Formats

This article has been accepted for publication in a future issue of this journal, but has not been

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY (VT-2013-01903.R1)