1

Energy Consumption Optimization in Mobile

Communication Networks
Florian Bahlke and Marius Pesavento
arXiv:1807.02651v1 [eess.SP] 7 Jul 2018

Abstract

This work addresses the challenge of minimizing the energy consumption of a wireless communi-
cation network by joint optimization of the base station transmit power and the cell activity. A mixed-
integer nonlinear optimization problem is formulated, for which a computationally tractable linear inner
approximation algorithm is provided. The proposed method offers great flexibility in optimizing the
network operation by considering multiple system parameters jointly, which mitigates a major drawback
of existing state-of-the-art schemes that are mostly based on heuristics. Simulation results show that
the proposed method exhibits high performance in decreasing the energy consumption, and provides
implicit load balancing in difficult high demand scenarios.

I. I NTRODUCTION

In the evolution of wireless communication networks over the recent years, new technolo-
gies have been proposed to fulfill the increasing performance requirements for the upcoming
fifth generation (5G) and subsequent generations of mobile communication networks [1]–[3].
In addition to serving users with ever increasing data rates, novel applications require very
low latency connections and extreme reliability. Among the most promising technologies for
5G are Massive-MIMO systems, Millimeter-Wave communications and heterogeneous network
structures with dense cell deployments. Between these three options, the latter one arguably
poses the lowest technological risks and level of commitment from network providers. For such
dense and heterogeneous networks, the existing cell architecture is supplemented with additional
cells containing base stations of variable size, both in transmit power and coverage area. This
densification of the network has been identified as a promising and scalable approach for the next
decades of wireless communications [4]–[6]. Due to the increase in intercell interferences limiting
the achievable data throughput, novel control schemes for such networks need to be devised that
supersede the established strategy of deploying additional cells without increasing the amount
 2

of coordination between them [7]–[9]. The wireless communication networks of the future are
envisioned to have a significantly higher energy efficiency in terms of energy consumption per
transmitted bit of data. In the 5G standard, this will be achieved trough intelligent switching of
each cell’s operation between active phases and sleep modes - abandoning the always-on and
always-connected concept of contemporary base stations - a dynamic scaling of the transmit
power, and an energy-focused design of multi-antenna systems [8], [10]–[12].
We propose a method for minimizing the energy consumption of the wireless communication
network, subject to cell load constraints that prevent cells from being unable to serve the demand
of associated users with their available time-frequency resources. This approach is suitable for
the planning the network parameters ahead of operation, and complements energy efficient
transceiver techniques commonly applied in-operation for example to maximize instantaneous
data rates. In previous research, extensive effort has been invested into the analysis and optimiza-
tion of cell loads for heterogeneous mobile communication networks [13]–[18]. The cell load
has been used in various schemes to optimize the transmit powers [19], [20], in the design of
energy-efficient beamformers for multi-antenna systems [21], [22], and to optimize the cell on-off
status to enable scheduling for sleep mode and activity periods [23], [24]. These methods share
one fundamental disadvantage, which is that they cannot jointly optimize the transmit power and
the cell activity status. Switching cells off is just considered implicitly, as the transmit power
being scaled down to zero [19], [25]. The transmit power in a practical system however might
be lower-bounded by a nonzero level, for example due to transmit power independent losses and
nonlinearities in the power amplifiers [26]–[28]. Heuristic approaches also heavily rely on the
cell load being a strictly decreasing continuous function of its transmit power, which requires
multiple simplifications in the way how the network is modeled, particularly regarding the used
adaptive modulation and coding schemes. For example, the load a user adds to a cell needs to
be a strictly decreasing function of the user’s signal-to-interference-and-noise ratio (SINR), the
assignment of users to cells needs to be constant and the operable transmit power range needs
to be lower-bound by zero, which all do not necessarily apply in practical systems.
In this work, we propose an approach for minimizing the network’s energy consumption based
on Mixed-Integer Linear Programming (MILP), which expands upon state of the art solutions
in the following aspects:
• The transmit power and the activity status of the cell (on or off) are jointly optimized. This
 3

leads to a mixed-integer problem as, e.g. the transmit power is optimized on a continuous
scale and the cell activity indicator is binary.
• While the original optimization problem is nonlinear and computationally intractable to
solve, we propose a linear inner approximation. The solution of this approximate problem
is always feasible for the original problem.
• The proposed method easily incorporates additional convex constraints such as minimum
transmit power and minimum SINR threshold constraints as well as upper bounds on the
user rates due to finite modulation and coding schemes.
• The assignment of users to cells is one of the design parameters, and changes dynamically
according to which cell provides the strongest signal. The proposed scheme also allows the
incorporation of other user allocation rules.
The remainder of the paper is structured as follows: In Section II we introduce the system model
for the wireless communication network. A mixed-integer nonlinear programming (MINLP)
approach to minimize the network’s energy consumption is introduced in Section III, for which
we provide an inner linear approximation (MILP). Simulation results for different energy con-
sumption models and a comparative analysis between the proposed and alternative methods are
provided in Section IV. Finally, we summarize the results and provide an outlook onto future
work in Section V.

II. S YSTEM M ODEL

We consider a downlink wireless communication network with K cells, each equipped with
a single antenna base station (BS). The transmit power of the BS in cell k = 1, . . . , K is
denoted as pk , and the vector of all transmit powers as p = [p1 , p2 , . . . , pK ]T . In the following
we assume that each cell is defined as the coverage area of the BS, and therefore we use the
terms interchangeably. In practical networks the transmit power pk is generally confined to lie
in a the interval
0 < PkMIN ≤ pk ≤ PkMAX , (1)

where, due to physical hardware limitations, such as linearity constraints in the power amplifiers
and radiation efficiency requirements of the antenna the thresholds PkMIN and PkMAX are positive
(excluding PkMIN = 0) and finite [26]–[28]. The network contains M demand points (DP) with
DP m = 1, . . . , M having the data demand Dm . A DP may represent a single mobile node,
 4

or in case of a cluster of closely spaced mobile nodes with similar channel characteristics to
the connected BS, the accumulation of multiple nodes. The gain of BS k is denoted by g̃kBS ,
DP
correspondingly we use g̃m as the antenna gain of the DP m. The large-scale path attenuation
PATH
factor of signals transmitted from the BS in cell k to DP m are denoted as g̃mk . Small scale
fading parameters will be neglected because the proposed method is assumed to be applied on
a network planning timescale, and on averaged rather than instantaneous channel information.
Combining the aforementioned factors, we denote as gkm = g̃kBS g̃mk
PATH DP
g̃m the attenuation factor
for transmissions between BS k and DP m.
The SINR of cell k serving DP m can be modeled as
pk gkm
γkm = P 2
, (2)
j=1,...,K\{k} pj gjm + σ

where σ 2 represents the variance of additive white Gaussian noise (AWGN). This corresponds
to an orthogonal frequency division multiple access (OFDMA) system with full frequency reuse
BW
between cells. The bandwidth efficiency of BS k serving DP m is denoted as ηkm [29], and the
total available system bandwidth is W . Gains in data rate achievable in multi-antenna systems
BW
can also be accounted for through the bandwidth efficiency parameter ηkm . The achieved radio
downlink bandwidth [30] of cell k in DP m can, e.g., be determined as
BW
Bkm = ηkm W log2 (1 + γkm ) (3)

where the interferences are treated as noise. To satisfy the data demand Dm in DP m, cell k
needs to allocate the fraction Dm /Bkm of its resources. In order to model the allocation of DPs
to BSs, we use the binary parameter

1 if DP m is allocated to cell k

Akm = , (4)
0 otherwise


and we denote as A ∈ {0, 1}K×M the combined allocation matrix. We assume that each DP is
allocated to a single BS, so that K
P
k=1 Akm = 1. To determine the fraction of available resources

required by a cell in order to satisfy the data demand Dm of all its allocated DPs m, as specified
by Akm , we compute the total load factor ρk of cell k [29], [31] as
M M
X Dm X Dm 1
ρk = Akm = Akm BW
. (5)
m=1
Bkm m=1 W ηkm log2 (1 + γkm )

We further define the vector of load factors ρ = [ρ1 , ρ2 , . . . , ρK ]T . It is observable that ρk > 0
holds and that cell k is not overloaded if ρk ≤ 1. An overloaded cell, with ρk > 1, cannot serve
 5

the minimum data demands Dm of its allocated DPs under the present SINR conditions. Under
these circumstances, new connections cannot be established, and existing connections have to
be dropped. Let
1
f (γ) = (6)
log2 (1 + γ)
denote, for later reference, the inverse rate, measured in time per transmitted bit, corresponding
to a link with SINR γ.
The interference term j=1,...,K\{k} pj gjm + σ 2 in Eq. (2) can be weighted with the load factors
P

ρj of interfering cells, to account for the fact that lightly loaded cells do not need to fully
use their available time-frequency resources and therefore generate on average lower levels of
interference than heavily loaded ones [14], [29], [31]. Because even lightly loaded cells might
fully interfere with each other if there are no coordination mechanisms employed, we will use,
without loss of generality, the "worst-case" assumption of full interference between active cells
in our simulations.
To indicate the on-off activity status of cells, we introduce the binary model parameter

1 if cell k is active

xk = . (7)
0 otherwise


and the vector x = [x1 , x2 , . . . , xK ]T representing the activity status of all cells in the network.
We define the energy consumption of cell k as

Ek = Γ (xk , pk , ρk ) (8)

where Γ (xk , pk , ρk ) is an arbitrary linearly increasing function of the cell’s on-off status xk ,
transmit power pk and load ρk . For example, the energy consumption function used in Eq. (8)
can be defined as
 
p̃k
Γ (xk , p̃k , ρk ) = T0 PkMAX κ1 xk + κ2 MAX + κ3 ρ̃k (9)
Pk
where the parameters κ1 , κ2 and κ3 are weighting factors for the cell’s energy consumption based
on the on-off status, transmit power, and load factor, respectively, and T0 is a time constant. The
load factor of a cell can impact its power consumption because it reflects the amount of its
utilization [32]. Recent network models therefore have established that, especially for small
cells, the power consumption is best modeled as a function of the cell load in addition to the
transmit power [20], [25]. Note that the terms xk , p̃k /PkMAX and ρ̃k cannot exceed the value 1, for
 6

each cell k. For more sophisticated models for the power consumption of mobile communication
BSs, which incorporate energy consumption of wired backhaul, and individual factors for all
components of the BS we refer to [27], [33]–[35]. Since our model can use any combination
of the three factors in Eq. (8), we obtain a highly flexible approach for energy minimization,
which is shown in the following Section III.

III. E NERGY C ONSUMPTION O PTIMIZATION

In this section, we formulate an optimization problem to minimize the energy consumption of

the wireless network as defined in Eq. (8), subject to DP to BS allocation-, minimum SINR- and
cell load constraints. Physical layer requirements of the wireless communications standard, such
as the used modulation and coding scheme in LTE-A, impose a minimum SINR level γ MIN > 0
for the transmission link to provide a non-zero rate, and define τ MAX = f (γ MIN ). The load term
BW
(Dm τ MAX )/(W ηkm ) therefore is the highest load DP m contributes to the overall load (5) of
cell k.
The SINR threshold γ MAX denotes the SINR-level where, for γ ≥ γ MAX , the highest available
modulation- and coding scheme is used such that the maximum rate is achieved, and a further
increase in SINR is not associated with an additional increase in rate. The inverse of the log-term
in Eq. (5) for SINR-levels γ ≥ γ MAX is denoted as τ MIN = f γ MAX . We further define

fτ+MIN (γ) = max f (γ), τ MIN .


(10)

For the allocation of DPs to cells, we assume that cell range expansion is being utilized [14], [36],
with θk denoting the bias value of cell k. DP m is allocated to the cell k that provides the highest
product of received signal power pk gk m and bias value θk . Using as optimization parameters
the binary cell activity indicator x ∈ {0, 1}K×1 and allocation indicator A ∈ {0, 1}M ×K , the
continuous transmit power parameter p ∈ RK×1
0+ and the cell load ρ ∈ RK×1
0+ , the energy
 7

minimization problem can be formulated as following:

K
X
minimize Γ (xk , pk , ρk ) (11a)
x,p,A,ρ
k=1

subject to PkMIN ≤ pk ≤ PkMAX ∀k (11b)

K
X
Akm = 1 ∀m (11c)
k=1
K
X
Akm ≤ xk ∀k, m (11d)
k=1
K
X
Akm θk pk gkm ≥ xj θj pj gjm ∀j, m (11e)
k=1
K
!
X X
Akm pk gkm − γ MIN xj (1 − Ajm )pj gjm + σ 2 ≥0 ∀m (11f)
k=1 j
M
!
X Dm + pk gkm
ρk = Akm f
BW τ MIN PK ∀k (11g)
m=1
W ηkm j=1 xj (1 − Ajm )pj gjm + σ 2

ρk ≤ 1 ∀k (11h)

xk , Akm ∈ {0, 1} ∀k, m (11i)

pk ∈ R0+ ∀k (11j)

In problem (11), the objective (11a) aims to minimize the overall systems’ energy consumption,
which is the sum of the energy consumption of individual cells as defined in (8) and (9). The
constraint (11b) defines the feasible transmit power range of cell k restricted according to (1).
Each DP m is served by exactly one cell k, and only active cells {k|xk = 1} can serve any
DP, as specified by (11c) and (11d), respectively. Constraint (11e) enforces that, each DP m is
allocated to the cell k that provides highest product of received signal power and bias value 1 .
The load constraint that cell k has to satisfy, as defined in (5), is specified in (11h).
Problem (11) is a combinatorial and nonconvex MINLP, and thus generally very difficult to
solve. While significant advancements have been made for convex MINLPs [37], [38], it is
universally agreed upon that nonconvex MINLPs pose a significant computational challenge

1
Typically the DP is allocated to the cell providing the highest received signal power, but this leads to an underutilization of
the low-power small cells. If so-called "range expansion" is utilized, the signal power from small cells is weighted with a bias
value, which corresponds to an increased coverage area [36].
 8

where the chances of finding an optimal solution to any given problem highly depend on the
problem size and structure [39], [40]. To maintain robustness and scalability for schemes based on
network optimization problems, it is therefore advisable to find an MILP that represents a linear
inner approximation or a linear reformulation of the original MINLP. The objective function
(11a) and constraints (11e), (11f) and (11h) contain the bilinear term xk pk . We introduce a new
variable p̃k , pk xk and reformulate (11) as the following equivalent problem:
K
X
minimize Γ (xk , p̃k , ρk ) (12a)
x,p̃,A,ρ
k=1

subject to xk PkMIN ≤ p̃k ≤ xk PkMAX ∀k (12b)

(11c) − (11d)
K
X
Akm θk p̃k gkm ≥ θj p̃j gjm ∀j, m (12c)
k=1
K
!
X X
MIN 2
Akm p̃k gkm − γ (1 − Ajm )p̃j gjm + σ ≥0 ∀m (12d)
k=1 j
M
!
X Dm + p̃k gkm
ρk = Akm f
BW τ MIN PK ∀k (12e)
m=1
W ηkm j=1 (1 − Ajm )p̃j gjm + σ 2

ρk ≤ 1 ∀k (12f)

xk , Akm ∈ {0, 1} ∀k, m (12g)

p̃k ∈ R0+ ∀k (12h)

Using a lifting strategy, we will in the following introduce auxiliary parameters to represent
bilinear products of optimization variables, which a more tractable, linear problem structure at
the cost of increased problem dimensionality. Towards this aim, the bilinear products of binary
allocation parameters Akm and cell transmit powers p̃k in Eqs. (12c), (12d) and (12e) have to
be linearized. We define the set

L := {(r, r, b, a) ∈ R0+ × R+ × {0, 1} × R0+ : a ≥ r − (1 − b)r, a ≤ r, a ≤ br} (13)

with the binary parameter b and the real parameter r with 0 ≤ r ≤ r. The inequalities defining
L in (13) are affine in r, b and a, and (r, r, b, a) ∈ L enforces a = rb, which will be used in
the following reformulations to linearize bilinear products of binary and continuous optimization
parameters [41].
 9

K×M
We introduce an new variable Ωkm and the corresponding matrix Ω ∈ R0+ . For the proposed
lifting approach, we install p̃k , PkMAX , Akm , Ωkm ∈ L ∀k, m in problem (12), which enforces

that Ωkm = p̃k Akm , such that we can reformulate (11) as:
K
X
minimize Γ (xk , p̃k , ρk ) (14a)
x,p,A,ρ,Ω
k=1

subject to (11c) − (11d), (12b), (13)

K
X
Ωkm θk gkm ≥ xj θj p̃j gjm ∀j, m (14b)
k=1
K
!
X X
Ωkm gkm − γ MIN (1 − Ωjm )gjm + σ 2 ≥0 ∀m (14c)
k=1 j
M
!
X Dm + p̃k gkm
ρk = Akm f
BW τ MIN
P 2
∀k (14d)
m=1
W ηkm j=1,...,K (p̃j − Ωjm )gjm + σ

ρk ≤ 1 ∀k (14e)

p̃k , PkMAX , Akm , Ωkm ∈ L

∀k, m (14f)

xk , Akm ∈ {0, 1} ∀k, m (14g)

p̃k , Ωkm ∈ R0+ ∀k (14h)

From (11) to (14), the auxiliary parameter Ω has been used in constraints (14b), (14c) and (14d)
to replace Ωkm = p̃k Akm , whereas the remaining optimization parameters remain unchanged.
The solution of problem (14) can therefore be used to easily obtain the corresponding solutions
for problem (11) and vice-versa. Thus, both formulations can be considered equivalent.
Problem (14) is an integer linear program except for constraint (14d), which is nonlinear due
to the log-term in the function fτ+MIN (γ) as defined in (10), the fractional SINR-term and the
allocation factor Akm . In the following, we propose an affine inner approximation of (14d)-
(14e). We define a set of I linear functions

ui (γ) = αi γ + βi , i = 1, . . . , I, (15)

which satisfy the upper bound property

max ui (γ) ≥ fτ+MIN (γ) ∀ γ ≥ γ MIN , (16)

i

as illustrated in Fig. 1. Since f (γ) in (6) is strictly decreasing, we conclude that all ui (γ) can
be designed such that αi ≤ 0 ∀ i. To approximate the load for γ ≥ γ MAX , as depicted in Fig. 1,
 10

f (γ), ui (γ)

b1

τ MAX

u1

u2
u... uI
bI = τ MIN γ
γ MIN γ MAX

Fig. 1. Illustration of the piecewise linear over-approximation of the cell load function f (γ) with the linear functions ui (γ) in
the SINR interval γ MIN ≤ γ ≤ γ MAX .

a constant function can be used with uI (γ) = βI = τ MIN . The issue of designing a suitable
set of ui that keep the maximum absolute approximation error below a selectable threshold ǫ is
discussed in Appendix A.
We introduce the optimization parameter µkm designed to be an upper bound of the load term
in Eq. (14d), such that
µkm ≥ ui (γ) ∀ i, γ ≥ γ MIN (17)

and the corresponding matrix µ ∈ RK×M

0+ . For the interval γ MIN ≤ γ ≤ γ MAX , we reformulate
the log-term contained in the function fτ+MIN (γ) in the constraint (14d) as
M
X Dm
ρk = Akm µ
BW km
(18)
m=1
W ηkm
where for (15)-(17)
p̃k gkm
µkm ≥ αi P 2
+ βi ∀ i, k, m (19)
j=1,...,K (1 − Ωjm )gjm + σ

We further denote the product of µkm and allocation parameter Akm as Λkm = µkm Akm and the
corresponding matrix as Λ ∈ RK×M
0+ . This bilinear product formulation for Λ is replaced by a
linear reformulation using (13) by adding the constraint that (µkm , β1 , Akm , Λkm ) ∈ L.
In order to approximate the interference levels in the denominator of the SINR term Eq. (2), we
 11

introduce the scalar interference levels Ψnkm with interference scenario index n = 1, . . . , N, and
N ×K×M
the corresponding three-dimensional scalar tensor Ψ ∈ R0+ . We also introduce a binary
interference scenario selection parameter φnkm and the corresponding three-dimensional binary
tensor φ ∈ {0, 1}N ×K×M . To ensure that the solution of the approximate problem is always
feasible for the original, we add the constraint that the selected discrete interference level is
always an over-approximation of the actual interference:
N
X X
φnkm Ψnkm ≥ (1 − Ωjm )gjm + σ 2 ∀k, m (20)
n=1 j=1,...,K

When implementing the selection parameter φ in Eq. (19), we replace the bilinear product
p̃k gkmφnkm with an auxiliary parameter, for which we introduce the lifting variable Φnkm =
N ×K×M
p̃k gkmφnkm with the corresponding tensor variable Φ ∈ R0+ . Again, the product com-
putation of Φ will be replaced by an auxiliary parameter using (13) by adding the constraint
p̃k gkm , PkMAX gkm, φnkm , Φnkm ∈ L ∀n, k, m.

The proposed linear inner approximation of (14) is the following:

K
X
minimize Γ (xk , p̃k , ρ̃k ) (21a)
x,p̃,A,ρ̃,Ω,µ,Λ,φ,Φ
k=1

subject to (11c) − (11d), (12b), (13), (14b) − (14c), (14f), (20)

M  
X dm
ρ̃k = Λ
BW km
∀k (21b)
m=1
W η

ρ˜k ≤ 1 ∀k (21c)
N
X
φnkm = 1 ∀k, m (21d)
n=1
N
X Φnkm
µkm ≥ αi + βi ∀i, k, m (21e)
n=1
Ψnkm

(µkm , β1 , Akm , Λkm) ∈ L ∀k, m (21f)

p̃k gkm , PkMAX gkm, φnkm , Φnkm ∈ L

∀n, k, m (21g)

xk , Akm , φnkm ∈ {0, 1} ∀n, k, m (21h)

p̃k , Ωkm , µkm , Λkm, Φnkm ∈ R0+ ∀n, k, m (21i)

Proposition 1. Problem (21) is an inner approximation of problem (14), i.e. for every point
{x, p, A} solving (21) a feasible point of (14) can be constructed.
 12

Proof. The transmit power constraints (12b), the allocation constraints (11c)-(11d) and the signal
power constraints (14b)-(14c) are identical in problem (14) and (21).
The proposition therefore holds if the load in (21b) is an inner approximation of that in (14d),
specifically if
M   M
X Dm X Dm 1
Λkm ≥ Akm BW
  ∀ k. (22)
m=1
W η BW m=1
W ηkm log2 1 + p̃k gkm
2
P
j=1,...,K (1−Ωjm )gjm +σ

Due to (21f), we have Λkm = µkmAkm , therefore (22) is satisfied if

1
µkm ≥   ∀ k, m, (23)
p̃k gkm
log2 1 + P 2
j=1,...,K (1−Ωjm )gjm +σ

from which, with (21e) and (16) applied to the left- and right-hand side of Eq. (23), respectively,
we obtain
N N
X Φnkm X p̃k gkm
αi + βi ≥ αi P 2
+ βi ∀ i, k, m. (24)
n=1
Ψnkm n=1 j=1,...,K (1 − Ωjm )gjm + σ

Due to the constraints (21g), which implement the bilinear constraint Φnkm = p̃k gkm φnkm , and
due to φnkm ∈ {0, 1} ∀ n, k, m, we have
N
X Φnkm p̃k gkm
= PN ∀ n, k, m. (25)
n=1
Ψnkm n=1 φnkm Ψnkm

Substituting (25) in the left-hand side of (24), we obtain the inequality

N N
X p̃k gkm X p̃k gkm
αi PN + βi ≥ αi P 2
+ βi ∀ i, k, m, (26)
n=1 n=1 Ψnkm n=1 j=1,...,K (1 − Ωjm )gjm + σ

which holds due to the constraint (20) for αi ≤ 0 ∀ i, thus proving the proposition.

The tightness of the approximating problem (21) with regards to problem (14) depends on
two factors. The first factor is related to how closely the linear functions ui approximate the
load function as in Eq. (16). The second factor is how closely the discrete interference levels
Ψnkm approximate the actual interference level j=1,...,K (1 − Ωjm )gjm + σ 2 . Proposition 1 holds
P

irrespectively the choice of the discrete interference levels Ψnkm . Certain changes in interference
levels, specifically the removal of strongest interferers, cause large differences in the load caused
by a DP. The levels Ψnkm can be chosen in such a way that these changes can be reflected by
the selection of a different interference scenario. The accuracy of the inner approximation can
be improved by using a larger number of interference levels, at the cost of increased problem
complexity.
 13

TABLE I
W EIGHTING FACTORS FOR COMPUTATION OF INTERFERENCE SCENARIOS Ψnkm , USED FOR AN OVER - APPROXIMATION OF

THE ACTUAL INTERFERENCE LEVEL .

n= 1 2 3 4 5 6 7

lnP 1 0.75 0.5 0.25 0 0 0

lnS 1 1 1 1 1 0 0

lnR 1 1 1 1 1 1 0

We propose to construct, for each pair (m, k) of DP m allocated to cell k, interference levels
Ψnkm that mainly reflect transmit power changes of the first- and second-strongest interferers
[42]–[44]. With
v = arg max(pj gjm ) (27)
j\{k}

and
w = arg max(pj gjm ) (28)
j\{k,v}}

we compute our interference levels as

X
Ψnkm = lnP pv gvm + lnS pw gwm + lnR pj gjm + σ 2 , (29)
j\{k,v,w}

where the parameters lnP ,lnS and lnR denote the weighting factors for primary-, secondary- and
remaining interferers, respectively. Keeping in mind that we focus on transmit power changes
for the first- and second strongest interferers, a suitable set of weighting factors to compute the
interference levels Ψnkm is shown, for example, in Table I.

IV. S IMULATION R ESULTS

To evaluate the performance of the proposed method, we simulate a heterogeneous wireless

communication network containing 4 macro- and 4 pico cells as illustrated in Fig. 2. The selected
system parameters are summarized in Table II. The selectable transmit power range and antenna
gains are chosen as 36dBm − 46dBm with 15dB antenna gain for macro cells and 26dBm −
36dBm with 5dB antenna gain for small cells. A bias value of θk = 2dB is used for small
cells to slightly increase their coverage area. The proposed method using Problem (21) was
solved using CVX for MATLAB [46], [47] and Gurobi as a MILP solver [48]. For the energy
 14

1,000 macro BS
pico BS
demand point
800

crossrange (m)
600

400

200

0
0 200 400 600 800 1,000
range (m)

Fig. 2. Illustration of the network scenario with 4 macro- and 4 small cells and an example distribution of 20 DPs. The network
area is 1000m times 1000m and path loss between cells and DPs is modeled according to 3GPP TS 36.814 specification.

consumption modeling of cells we use Eq. (9) with κ1 = 0.5, κ2 = 0.5 and κ3 = 0. This implies
that the power consumption of cell k depends on its on-off status indicator xk and its transmit
power pk . The power consumption is modeled this way in order to allow comparability of the
proposed MILP with an established heuristic method proposed in [19] that focuses on transmit
power minimization. As a performance benchmark for our energy minimization algorithm we
use the power scaling method introduced in [19], which we extended in the following ways
to make it applicable to our problem: power scaling is used for all possible configurations of
all cells’ on-off status x. Resulting transmit powers obtained by the algorithm of [19] that lie
below or above the bounds specified in Table II are projected to the lower- and upper bound
respectively. Then, the best configuration that does not violate load constraints is selected as the
solution. This algorithm therefore combines an exhaustive search over all configurations for x
with power scaling being used in each configuration. It is in the following in all figures denoted
as "power scaling + exh. search". The second approach we use for comparison is an exhaustive
search over all combinations of cells being switched on or off, with the transmit powers being
 15

TABLE II
S IMULATION PARAMETERS OF A DOWNLINK LTE NETWORK . T HE TRANSMIT POWER OF THE CELLS IS OPTIMIZED INSIDE A

10dB INTERVAL . R ESULTS ARE AVERAGED OVER 5000 SIMULATIONS WITH FIXED BASE STATION POSITIONS AND

RANDOMLY DISTRIBUTED DP S .

Area size 1000 × 1000 m

Noise power -145 dBm/Hz

System bandwidth W 20 MHz

Position of macro BS MBS1 at [200m, 200m]

MBS2 at [150m, 850m]
MBS3 at [800m, 230m]
MBS4 at [780m, 820m]

MBS transmit power range P MIN . . . P MAX 36dBm . . . 46dBm

BS
MBS antenna gain g̃ 15dB

MBS bias value θk 0dB

Position of pico BS PBS1 at [500m, 700m]

PBS2 at [520m, 310m]
PBS3 at [320m, 500m]
PBS4 at [690m, 490m]

PBS transmit power range P MIN . . . P MAX 26dBm . . . 36dBm

PBS antenna gain g̃ BS 5dB

PBS bias value θk 3dB

DP antenna gain g̃ DP 0dB

PATH
Propagation loss g̃ 3GPP TS 36.814 [45]

Bandwidth efficiency η BW 0.8

MIN
SINR requirement γ −10dB

SINR threshold γ MAX 20dB

fixed to P MAX , which we in the following indicate as "max power cell switching". The solution
of the original MINLP in (11) is unsuitable as a lower bound solution even for small problem
sizes, because even for fixed binary optimization parameters the resulting continuous problem
is still nonconvex.
Deploying M = 20 DPs randomly in the network area illustrated in Fig. 2, 5000 network
scenarios are generated and each DPs data demand in each scenario is scaled between dm =
0.25Mbit/s and dm = 7.5Mbit/s. The proposed energy-minimized solution obtained from solving
 16

probability of obtaining a feasible solution

0.8

0.6

max. power cell switching

0.4
pow. scaling + exh. search
proposed MILP
0.2

0
0 1 2 3 4 5 6 7
user demand in Mbit/s

Fig. 3. Probability of obtaining a feasible solution over increasing user demand, evaluated over 5000 simulations of M = 20
randomly distributed demand points. The proposed MILP-based scheme achieves the highest solution percentage.

problem (21) is compared to the solutions of the aforementioned max. power cell switching
and combined power scaling and exhaustive search methods [19]. The probability of obtaining
a feasible solution with no overloaded cells is illustrated in Fig.3. The proposed MILP based
method is much more likely to find a feasible and power-minimized solution even in high
demand scenarios. In the following we discuss the performance indicators: energy consumption,
cell load, and number of active cells. To ensure a fair comparison, the respective averages were
computed only from those scenarios that were solved by all methods. Fig. 4 shows the average
power consumption achieved by each of the three considered energy minimization schemes.
The proposed MILP-based approach achieves lower power consumption levels than both the
cell switching and the heuristic approach. The cell switching method noticeably achieves good
performance up until about 3Mbit/s, with the performance significantly deteriorating for higher
demands.
In Fig. 5, the average number of active cells is shown. For very low demands, it can be
observed that the number of cells is not increasing continuously with the demand, as the proposed
 17

200

energy consumption in W/T0

150 full power
max. power cell switching
pow. scaling + exh. search
proposed MILP
100

50

0
0 1 2 3 4 5 6 7
user demand in Mbit/s

Fig. 4. Energy consumption for energy minimization schemes over increasing user demand, averaged over 5000 simulations of
M = 20 randomly distributed demand points. The proposed scheme achieves the lowest average energy consumption levels of
the evaluated schemes.

algorithm for some scenarios serves all users exclusively with pico cells, instead of using a single
macro cell. In practice this does not pose a problem since for these low load levels offloading
is not required. On average however less than 4 cells are being used, showing that small cells
are only used sporadically or for low demand levels. For very high demand levels, the proposed
method utilizes the lowest number of cells.
The average load factor of active cells is shown in Fig. 6. It is observable that the cell load
does not converge to 1 even for high loads. It was shown in [19] that for minimum energy
consumption, the load would be equal to 1. This however only holds if the transmit power can
be increased or decreased without bounds (i.e. for P MIN = 0 and P MIN = ∞), and if the cell
load is a strictly decreasing function of the transmit power. With the upper- and lower bounds
on the transmit power, the discontinuities we introduced in the load computation, and the user
allocation changing dynamically with the transmit powers, we observe from Fig. 6 that this
property no longer holds.
 18

4
max. power cell switching
pow. scaling + exh. search
average number of active cells
proposed MILP

1
0 1 2 3 4 5 6 7
user demand in Mbit/s

Fig. 5. Number of active cells for energy minimization schemes over increasing user demand, averaged over 5000 simulations
of M = 20 randomly distributed demand points. For high demand, the proposed scheme on average utilizes the lowest number
of cells.

V. C ONCLUSION

In this paper we proposed a novel method for minimizing the energy consumption of a wireless
communication network, subject to cell load constraints. The transmit power and the cell activity
are jointly optimized in a mixed integer linear problem. Multiple simplifications used in other
state of the art methods to allow the application of heuristic schemes are not required in the
proposed method.
The simulation results show that the proposed approach achieved a further decrease in energy
consumption relative to both an optimization of the cell activity and a comparable heuristic
method. Additionally, it achieves a higher success rate in finding an operable solution for high-
demand network scenarios.
Even though the proposed method the proposed method consists in linear approximations of the
originally mixed integer nonlinear program with bilinear and nonconvex constraints, it still yields
very high complexity, making it impractical for the optimization of large networks. Further work
 19

average load level of active cells

0.8

0.6

0.4 max. power cell switching

pow. scaling + exh. search
proposed MILP

0.2

0 1 2 3 4 5 6 7
user demand in Mbit/s

Fig. 6. Load of active cells for energy minimization schemes over increasing user demand, averaged over 5000 simulations of
M = 20 randomly distributed demand points.

could be dedicated to combining existing heuristic methods with an utilization of the proposed
approach to optimize smaller clusters of the network, to allow for better scalability.

A PPENDIX

As illustrated in Fig. 1, we aim to find linear functions

ui (γ) = αi γ + βi , (30)

indicated with i = 1, . . . , I, which satisfy the condition

1
max ui (γ) ≥ ∀ γ ≥ γ MIN . (31)
i log2 (1 + γ)
with αi ≤ 0 ∀ i. The problem of finding suitable linear functions ui (γ) is equivalent to finding
a set of breakpoints on f (γ) where the linear functions ui are the lines connecting each two
respectively neighboring break points. As discussed in [49], [50], a good breakpoint selection
strategy is to start with the function values of the interval endpoints as the first two breakpoints.
Assuming the line connecting these two points to be the linearization solution, we compute
 20

the position γ where the maximum approximation error occurs. If that error is larger than a
predefined threshold ǫ, we add a breakpoint at that position, and we again determine the linear
functions between neighboring breakpoints. The procedure is then continued until in each interval
between two breakpoints the maximum approximation error is lower than ǫ.
Assuming u(γ) ≥ f (γ), we define the approximation error function

ξ(γ) = u(γ) − f (γ) (32)

1
= αγ + β − (33)
log2 (γ + 1)
and the derivative
dξ(γ) log(2)
=α+ . (34)
dγ (γ + 1) log2 (γ + 1)
dξ(γ)
There is dγ
= 0 for
 q 
2W 1
2
− log(2)
α
γ = δ(αi ) = e ∀ αi < 0 (35)

where W is the Lambert W-Function defined as

y = f −1 (yey ) = W (yey ) . (36)

Algorithm 1 Breakpoint selection algorithm

1: procedure BPS(γ1 , γ2 )
f (γ2 )−f (γ1 )
2: α← γ2 −γ1

3: if |ξ(δ(α))| ≤ ǫ then
4: return {}
5: else
6: return {BPS(γ1 , δ(α)), f (δ(α)), BPS(δ(α), γ2)}
7: end if
8: end procedure

To determine the set of breakpoints we define the procedure BPS (Algorithm 1) which returns
a set of breakpoints necessary between given interval endpoints (γ1 , γ2 ). This way a set of
γ-positions of the breakpoints on f (γ) can be obtained, and by connecting each respectively
neighboring pair of points in this set, the linear functions ui (γ) can be determined and used in
Problem (21).
 21

R EFERENCES

[1] J. G. Andrews, S. Buzzi, W. Choi, S. V. Hanly, A. Lozano, A. C. K. Soong, and J. C. Zhang, “What Will 5G Be?” IEEE
Journal on Selected Areas in Communications, vol. 32, no. 6, pp. 1065–1082, June 2014.
[2] F. Boccardi, R. W. Heath, A. Lozano, T. L. Marzetta, and P. Popovski, “Five Disruptive Technology Directions for 5G,”
IEEE Communications Magazine, vol. 52, no. 2, pp. 74–80, February 2014.
[3] M. Iwamura, “NGMN View on 5G Architecture,” in 2015 IEEE 81st Vehicular Technology Conference (VTC Spring), May
2015, pp. 1–5.
[4] N. Bhushan, J. Li, D. Malladi, R. Gilmore, D. Brenner, A. Damnjanovic, R. T. Sukhavasi, C. Patel, and S. Geirhofer,
“Network Densification: The Dominant Theme for Wireless Evolution Into 5G,” IEEE Communications Magazine, vol. 52,
no. 2, pp. 82–89, February 2014.
[5] X. Ge, S. Tu, G. Mao, C. X. Wang, and T. Han, “5G Ultra-Dense Cellular Networks,” IEEE Wireless Communications,
vol. 23, no. 1, pp. 72–79, February 2016.
[6] L. Wang, K. K. Wong, R. W. Heath, J. Yuan, and J. Yuan, “Wireless Powered Dense Cellular Networks: How Many Small
Cells Do We Need?” IEEE Journal on Selected Areas in Communications, vol. PP, no. 99, pp. 1–1, 2017.
[7] J. G. Andrews, X. Zhang, G. D. Durgin, and A. K. Gupta, “Are we approaching the fundamental limits of wireless network
densification?” IEEE Communications Magazine, vol. 54, no. 10, pp. 184–190, October 2016.
[8] R. Cavalcante, S. Stanczak, M. Schubert, A. Eisenblaetter, and U. Tuerke, “Toward Energy-Efficient 5G Wireless
Communications Technologies: Tools for Decoupling the Scaling of Networks from the Growth of Operating Power,”
Signal Processing Magazine, IEEE, vol. 31, no. 6, pp. 24–34, Nov 2014.
[9] C. L. I, C. Rowell, S. Han, Z. Xu, G. Li, and Z. Pan, “Toward Green and Soft: A 5G Perspective,” IEEE Communications
Magazine, vol. 52, no. 2, pp. 66–73, February 2014.
[10] C. Lange, D. Kosiankowski, A. Betker, H. Simon, N. Bayer, D. von Hugo, H. Lehmann, and A. Gladisch, “Energy
Efficiency of Load-Adaptively Operated Telecommunication Networks,” Journal of Lightwave Technology, vol. 32, no. 4,
pp. 571–590, Feb 2014.
[11] L. M. Correia, D. Zeller, O. Blume, D. Ferling, Y. Jading, I. Gódor, G. Auer, and L. V. D. Perre, “Challenges and Enabling
Technologies for Energy Aware Mobile Radio Networks,” IEEE Communications Magazine, vol. 48, no. 11, pp. 66–72,
November 2010.
[12] W. Vereecken, W. V. Heddeghem, M. Deruyck, B. Puype, B. Lannoo, W. Joseph, D. Colle, L. Martens, and P. Demeester,
“Power Consumption in Telecommunication Networks: Overview and Reduction Strategies,” IEEE Communications
Magazine, vol. 49, no. 6, pp. 62–69, June 2011.
[13] L. You and D. Yuan, “Load Optimization With User Association in Cooperative and Load-Coupled LTE Networks,” IEEE
Transactions on Wireless Communications, vol. 16, no. 5, pp. 3218–3231, May 2017.
[14] I. Siomina and D. Yuan, “Load Balancing in Heterogeneous LTE: Range Optimization Via Cell Offset and Load-coupling
Characterization,” in 2012 IEEE International Conference on Communications (ICC), June 2012, pp. 1357–1361.
[15] ——, “Optimization Approaches for Planning Small Cell Locations in Load-Coupled Heterogeneous LTE Networks,” in
Personal Indoor and Mobile Radio Communications (PIMRC), 2013 IEEE 24th International Symposium on, Sept 2013,
pp. 2904–2908.
[16] F. Bahlke, O. D. Ramos-Cantor, and M. Pesavento, “Budget Constrained Small Cell Deployment Planning for Het-
erogeneous LTE Networks,” in 2015 IEEE 16th International Workshop on Signal Processing Advances in Wireless
Communications (SPAWC), June 2015, pp. 1–5.
 22

[17] F. Bahlke, O. D. Ramos-Cantor, S. Henneberger, and M. Pesavento, “Optimized Cell Planning for Network Slicing in
Heterogeneous Wireless Communication Networks,” IEEE Communications Letters, pp. 1–1, 2018.
[18] Z. H. Yang, Y. J. Pan, M. Chen, H. Xu, and J. F. Shi, “Cell Load Coupling With Power Control for LTE Network Planning,”
in 2015 International Conference on Wireless Communications Signal Processing (WCSP), Oct 2015, pp. 1–5.
[19] C. K. Ho, D. Yuan, L. Lei, and S. Sun, “Power and Load Coupling in Cellular Networks for Energy Optimization,” IEEE
Transactions on Wireless Communications, vol. 14, no. 1, pp. 509–519, Jan 2015.
[20] L. You, L. Lei, and D. Yuan, “Optimizing Power and User Association for Energy Saving In Load-Coupled Cooperative
LTE,” in 2016 IEEE International Conference on Communications (ICC), May 2016, pp. 1–6.
[21] J. Chen, H. Zhuang, and Z. Luo, “Energy Optimization in Dense OFDM Networks,” IEEE Communications Letters, vol. 20,
no. 1, pp. 189–192, Jan 2016.
[22] G. Miao, N. Himayat, G. Y. Li, and S. Talwar, “Distributed Interference-Aware Energy-Efficient Power Optimization,”
IEEE Transactions on Wireless Communications, vol. 10, no. 4, pp. 1323–1333, April 2011.
[23] H. Celebi and . Güvenç, “Load Analysis and Sleep Mode Optimization for Energy-Efficient 5G Small Cell Networks,” in
2017 IEEE International Conference on Communications Workshops (ICC Workshops), May 2017, pp. 1159–1164.
[24] L. Lei, D. Yuan, C. K. Ho, and S. Sun, “Optimal Cell Clustering and Activation for Energy Saving in Load-Coupled
Wireless Networks,” IEEE Transactions on Wireless Communications, vol. 14, no. 11, pp. 6150–6163, Nov 2015.
[25] Q. Kuang and W. Utschick, “Energy Management in Heterogeneous Networks With Cell Activation, User Association, and
Interference Coordination,” IEEE Transactions on Wireless Communications, vol. 15, no. 6, pp. 3868–3879, June 2016.
[26] S. Kandukuri and S. Boyd, “Optimal Power Control in Interference-Limited Fading Wireless Channels with Outage-
Probability Specifications,” IEEE Transactions on Wireless Communications, vol. 1, no. 1, pp. 46–55, Jan 2002.
[27] C. Desset, B. Debaillie, V. Giannini, A. Fehske, G. Auer, H. Holtkamp, W. Wajda, D. Sabella, F. Richter, M. J. Gonzalez,
H. Klessig, I. Gódor, M. Olsson, M. A. Imran, A. Ambrosy, and O. Blume, “Flexible Power Modeling of LTE Base
Stations,” in 2012 IEEE Wireless Communications and Networking Conference (WCNC), April 2012, pp. 2858–2862.
[28] O. Arnold, F. Richter, G. Fettweis, and O. Blume, “Power Consumption Modeling of Different Base Station Types in
Heterogeneous Cellular Networks,” in 2010 Future Network Mobile Summit, June 2010, pp. 1–8.
[29] K. Majewski and M. Koonert, “Conservative Cell Load Approximation for Radio Networks with Shannon Channels and
its Application to LTE Network Planning,” in Telecommunications (AICT), 2010 Sixth Advanced International Conference
on, May 2010, pp. 219–225.
[30] P. Mogensen, W. Na, I. Kovacs, F. Frederiksen, A. Pokhariyal, K. Pedersen, T. Kolding, K. Hugl, and M. Kuusela, “LTE
Capacity Compared to the Shannon Bound,” in Vehicular Technology Conference, 2007. VTC2007-Spring. IEEE 65th, April
2007, pp. 1234–1238.
[31] I. Siomina and D. Yuan, “Analysis of Cell Load Coupling for LTE Network Planning and Optimization,” Wireless
Communications, IEEE Transactions on, vol. 11, no. 6, pp. 2287–2297, June 2012.
[32] K. Son, H. Kim, Y. Yi, and B. Krishnamachari, “Base Station Operation and User Association Mechanisms for Energy-
Delay Tradeoffs in Green Cellular Networks,” IEEE Journal on Selected Areas in Communications, vol. 29, no. 8, pp.
1525–1536, September 2011.
[33] Z. Hasan, H. Boostanimehr, and V. K. Bhargava, “Green Cellular Networks: A Survey, Some Research Issues and
Challenges,” IEEE Communications Surveys Tutorials, vol. 13, no. 4, pp. 524–540, 4 2011.
[34] H. Bogucka and A. Conti, “Degrees of Freedom for Energy Savings in Practical Adaptive Wireless Systems,” IEEE
Communications Magazine, vol. 49, no. 6, pp. 38–45, June 2011.
[35] M. Deruyck, W. Joseph, and L. Martens, “Power Consumption Model for Macrocell and Microcell Base Stations,”
 23

Transactions on Emerging Telecommunications Technologies, vol. 25, no. 3, pp. 320–333, 2014. [Online]. Available:
http://dx.doi.org/10.1002/ett.2565
[36] Q. Ye, B. Rong, Y. Chen, M. Al-Shalash, C. Caramanis, and J. G. Andrews, “User Association for Load Balancing in
Heterogeneous Cellular Networks,” IEEE Transactions on Wireless Communications, vol. 12, no. 6, pp. 2706–2716, June
2013.
[37] H. Hijazi, P. Bonami, G. Cornuéjols, and A. Ouorou, “Mixed-Integer Nonlinear Programs Featuring On/Off Constraints,”
Computational Optimization and Applications, vol. 52, no. 2, pp. 537–558, 2012.
[38] P. Belotti, C. Kirches, S. Leyffer, J. Linderoth, J. Luedtke, and A. Mahajan, “Mixed-integer Nonlinear Optimization,” Acta
Numerica, vol. 22, p. 1Ű131, 2013.
[39] C. A. Floudas, A. Aggarwal, and A. R. Ciric, “Global Optimum Search for Nonconvex NLP and MINLP
Problems,” Computers & Chemical Engineering, vol. 13, no. 10, pp. 1117 – 1132, 1989. [Online]. Available:
http://www.sciencedirect.com/science/article/pii/0098135489870164
[40] F. Trespalacios and I. E. Grossmann, “Review of Mixed-Integer Nonlinear and Generalized Disjunctive Programming
Methods,” Chemie Ingenieur Technik, vol. 86, no. 7, pp. 991–1012, 2014.
[41] L. Liberti and C. C. Pantelides, “An Exact Reformulation Algorithm for Large Nonconvex NLPs Involving
Bilinear Terms,” Journal of Global Optimization, vol. 36, no. 2, pp. 161–189, Oct 2006. [Online]. Available:
https://doi.org/10.1007/s10898-006-9005-4
[42] H.-L. Määttänen, K. Hämäläinen, J. Venäläinen, K. Schober, M. Enescu, and M. Valkama, “System-Level Performance
of LTE-Advanced with Joint Transmission and Dynamic Point Selection Schemes,” EURASIP Journal on Advances in
Signal Processing, vol. 2012, no. 1, p. 247, Nov 2012. [Online]. Available: http://dx.doi.org/10.1186/1687-6180-2012-247
[43] O. D. Ramos-Cantor, J. Belschner, G. Hegde, and M. Pesavento, “Centralized Coordinated Scheduling in LTE-Advanced
Networks,” ArXiv e-prints, Feb. 2017.
[44] S. Gulati, S. Kalyanasundaram, P. Nashine, B. Natarajan, R. Agrawal, and A. Bedekar, “Performance Analysis of Distributed
Multi-Cell Coordinated Scheduler,” in 2015 IEEE 82nd Vehicular Technology Conference (VTC2015-Fall), Sept 2015, pp.
1–5.
[45] “3GPP TS 36.814 Evolved Universal Terrestrial Radio Access (E-UTRA); Further advancements for E-UTRA physical
layer aspects v9.0.0.”
[46] M. Grant and S. Boyd, “CVX: Matlab Software for Disciplined Convex Programming, version 2.1,” http://cvxr.com/cvx,
Mar. 2014.
[47] ——, “Graph Implementations for Nonsmooth Convex Programs,” in Recent Advances in Learning and Control, ser.
Lecture Notes in Control and Information Sciences, V. Blondel, S. Boyd, and H. Kimura, Eds. Springer-Verlag Limited,
2008, pp. 95–110.
[48] “Gurobi Optimizer 6.0,” www.gurobi.com.
[49] M.-H. Lin, J. Carlsson, D. Ge, J. Shi, and J.-F. Tsai, “A Review of Piecewise Linearization Methods,” Mathematical
Problems in Engineering, 2013.
[50] M.-H. Lin and J.-F. Tsai, “Comparisons of Break Points Selection Strategies for Piecewise Linear Approximation,”
International Journal of Mechanical Engineering and Robotics Research, vol. 4, no. 3, pp. 247–250, July 2015.