Joint Tx/Rx Energy-Efficient Scheduling in

Multi-Radio Networks: A Divide-and-Conquer

Approach
Qingqing Wu, Meixia Tao, and Wen Chen

arXiv:1502.00052v1 [cs.IT] 31 Jan 2015

Department of Electronic Engineering, Shanghai Jiao Tong University, Shanghai, China.

Emails: {wu.qq,mxtao,wenchen}@sjtu.edu.cn.
AbstractMost of the existing works on energy-efficient wireless communication systems only consider the transmitter (Tx) or
the receiver (Rx) side power consumption but not both. Moreover,
they often assume the static circuit power consumption. To
be more practical, this paper considers the joint Tx and Rx
power consumption in multiple-access radio networks, where
the power model takes both the transmission power and the
dynamic circuit power into account. We formulate the joint
Tx and Rx energy efficiency (EE) maximization problem which
is a combinatorial-type one due to the indicator function for
scheduling users and activating radio links. The link EE and the
user EE are then introduced which have the similar structure as
the system EE. Their hierarchical relationships are exploited to
tackle the problem using a divide-and-conquer approach, which
is only of linear complexity. We further reveal that the static
receiving power plays a critical role in the user scheduling.
Finally, comprehensive numerical results are provided to validate
the theoretical findings and demonstrate the effectiveness of the
proposed algorithm for improving the system EE.

I. I NTRODUCTION
The increasing number of new wireless access devices and
various services lead to a significant increase in the demand
for higher user data rate. While the higher energy consumption
is a great concern as well for future wireless communication
systems. Recently, there has been an upsurge of interest in
the energy efficiency (EE) optimization field. Basic concepts
of energy-efficient communications are introduced in [1] and
several advanced physical layer techniques for EE are studied
in [2][6].
However, all the above works only consider one side
power consumption, i.e., either the transmitter (Tx) or the
receiver (Rx) side. In fact, the expectation of limiting electric
expenditure and reducing carbon emissions requires the base
station to perform in an energy-efficient manner [1], while
minimizing the user side energy consumption also deserves
more efforts due to capacity limited batteries and user experience requirements [7], [8]. Moreover, according to [9],
the techniques adopted to improve the EE of one end of the
communication system may adversely affect the EE of the
This work is supported by the National 973 Project #2012CB316106,
by NSF China #61322102, #61161130529, and #61328101, by the STCSM
Science and Technology Innovation Program #13510711200, by the SEU
National Key Lab on Mobile Communications #2013D11. Wen Chen is also
with the School of Electronic Engineering and Automation, Guilin University
of Electronic Technology.

other end. Therefore, it is necessary to consider the joint Tx

and Rx EE optimization, which shall provide more flexibility
for the energy saving at the side interested or both.
For EE oriented research, one of the most important tasks
is to quantify the power consumption of the communication
system [9]. Most of the existing works only consider a constant
circuit power so as to simplify the system analysis and make
the problem more tractable. However, it has been reported in
[1], [10] that a rough modeling for the power consumption
can not reflect the true behaviour of wireless devices and thus
might provide misleading conclusions. Therefore, the power
consumption modeling should not only capture the key system
components but also characterize the reality [1].
The main contributions of this paper are summarized as
follows: 1) We formulate the joint Tx and Rx EE maximization
problem in which the link dependent signal processing power,
the static circuit power as well as the transmission power
are considered based on a comprehensive study [9], while
the power model of existing works [2][5], [11], [12] are
basically special cases. 2) We explore the fractional structure
of the system EE and introduce the concept of the individual
EE, i.e., the link EE and the user EE. Based on these, an
optimal approach of linear complexity is proposed to solve
the non-convex EE maximization problem. Moreover, this
approach can also be used to optimally solve the problem in
[3] where only a quadratic complexity method is proposed.
3) We reveal that the static receiving power has an implicit
interpretation of the optimal number of scheduled users. In
the extreme case when the static receiving power is negligible,
time division multiplexing access (TDMA) is optimal for the
energy-efficient transmission.
II. S YSTEM D ESCRIPTION

AND

P ROBLEM F ORMULATION

A. System Model
Consider a multi-user multi-radio network, where K users
are communicating with one access point (AP) over M orthogonal radio links simultaneously. It is assumed that each
user k, for k = 1, ..., K, is assigned prior with a fixed subset
of radio links, denoted as Mk , and that the radio links of
different users do not T
overlap with each other so as to void
interference, i.e., Mk Mm = . The multiple radio links
can be formed by orthogonal multiplexing techniques, such
as frequency division multiplexing. The channel between the

AP and each user is assumed to be quasi-static fading and

they are all equipped with one antenna. It is assumed that
the perfect and global channel state information (CSI) of all
users is available for the AP, which allows us to do energyefficient scheduling. The channel gain of user k over link i and
the corresponding power allocation of this link are denoted as
gk,i and pk,i , respectively. The receiver noise is modelled as a
circularly symmetric complex Gaussian random variable with
zero mean and variance 2 for all links. Then the data rate of
user k over link i, denoted as rk,i , can be expressed as

pk,i gk,i 
rk,i = B log2 1 +
,
(1)
2
where B is the bandwidth of each radio link and characterizes the gap between actual achievable rate and channel
capacity due to a practical modulation and coding design [3].
Consequently, the overall system data rate can be expressed
as
K
K
X
X
X
Rtot =
k Rk =
k
rk,i ,
(2)
k=1

k=1

iMk

where Rk is the data rate of user k and k which is provided

by upper layers, represents the priority of user k.

B. Joint Tx/Rx Power Consumption Model

In this work, we adopt the power consumption model from
[9] published by Energy Aware Radio and neTwork tecHnologies (EARTH) project, which provides a comprehensive
characterization of the power consumption for each component
involved in the communication.
At the user side, the power dissipation consists of two parts,
i.e., the transmission power and the circuit power. Denote PT k
as the overall transmission power of user k and it is given by
P
iMk pk,i
,
(3)
PT k =

where (0, 1] is a constant which accounts for the efficiency

of the power amplifier. Denote PCk as the circuit power of user
k. According to [9], the circuit power of each device contains
a dynamic part for the signal processing which linearly scales
with the number of active links, and a static part independent
of links for other circuit blocks, i.e.,
PCk (nok ) = nok Pdyn,k + I(nok )Psta,k ,

(4)

At the AP side, the receiving circuit power consumption also

consists of two similar parts as the user device [7], [9]. Denote
Pdyn,0 and Psta,0 as the dynamic and static receiving circuit
power, respectively. Then the overall power consumption at
the AP side can be expressed as
P0 =

K
X

nok Pdyn,0 + Psta,0 .

(7)

k=1

Finally, the overall power consumption of the system can

be expressed as
Ptot =

K
X

Pk + P0 .

(8)

k=1

C. Problem Formulation
Energy efficiency is commonly defined by the ratio of
the overall system rate Rtot over the overall system power
consumption Ptot [2], [3], [5]. Our goal is to jointly optimize
the user scheduling, the link activation and the power control
to maximize the EE of the considered system. Mathematically,
we can formulate the EE optimization problem as (P1)
P
PK
pk,i gk,i

iMk B log2 1 + 2
k=1 k
max P
P

p
K
iMk pk,i
o ) + no P
+ Psta,0
+
P
(n
Ck
dyn,0
k=1
k
k

X
s.t. nok =
I(pk,i ), 1 k K, i Mk ,
iMk

k,i
0 pk,i Pmax
,

1 k K, i Mk ,

(9)

where p , {pk,i |k = 1, 2, ...K; i Mk }. For practical

consideration, we assume that each radio link i of user k has a
k,i
maximum allowed transmit power Pmax
. Note that the authors
in [3] consider a similar problem formulation to (9) but without
power constraint which is thereby a special case of this paper.
The existence of the two
functions, i.e.,


P layered indicator
I(pk,i ) and I(nok ) = I
I(p
)
makes
the objeck,i
iMk
tive function discontinuous and hence non-differentiable. The
global optimal solution of (9) is generally difficult to be
obtained with an efficient complexity. In the following section,
we explore the particular structure of system EE and show that
the global optimal solution can actually be obtained using a
divide-and-conquer approach with low complexity.

where nokPis the number of active links and can be expressed

as nok = iMk I(pk,i ). Here, the indicator function I(x) is
defined as

1, if x > 0,
I(x) =
(5)
0, otherwise.

III. E NERGY-E FFICIENT S CHEDULING

In this section, we solve the system EE maximization
problem directly from a fractional-form perspective. This idea
results from the connection of the EE from three levels,
namely, the link Energy Efficiency, the user Energy Efficiency,
and the system Energy Efficiency.

Pk = PT k + PCk (nok ).

A. Link Energy Efficiency and User Energy Efficiency

Definition 1 (Link Energy Efficiency): The EE of link i of
user k, for i Mk , k = 1, .., K, is defined as the ratio of
the weighted achievable rate of the user on this link over the
consumed power associated with this link, i.e.,
p gk,i

k B log2 1 + k,i
2
,
(10)
eek,i = pk,i
+
P
+
P
dyn,k
dyn,0

Specifically, if pk,i > 0, then I(pk,i ) = 1 means that link

i is active, and if nok > 0, then I(nok ) = 1 means that user
k is scheduled. In (4), Pdyn,k and Psta,k are dynamic and
static components of the circuit power for user k, respectively.
Considering different types of terminals in practical systems,
Pdyn,k and Psta,k can be different for different user k. Now,
the overall power consumption of user k, denoted as Pk , is
(6)

where the link-level power consumption counts the transmission power of the user over the link, per-link dynamic circuit
power of the user and the AP, respectively.
It is easy to prove that this fractional type function have
the stationary point which is also the optimal point [13]. By
setting the derivative of eek,i with respect to pk,i to zero, we
obtain that the optimal power value pk,i and the optimal link
EE under peak power constraint satisfies
k,i
#Pmax
"
2
Bk

, k, i Mk ,
(11)

pk,i =
eek,i ln 2
gk,i
0

2
k
,
where [x]ab , min {max{x, b}, a}. Note that eeB
ln 2 > g
k,i
k,i
i.e., pk,i > 0 always holds for eek,i , since otherwise eek,i

would be zero. Based on (10) and (11), the numerical values of

eek,i and pk,i can be easily obtained by the bisection method.
Definition 2 (User Energy Efficiency): The EE of user k,
for k = 1, ..., K, is defined as the ratio of the weighted total
achievable rate of the user on all its preassigned radio links
over the total power consumption associated with this user,
i.e.,
P
p gk,i

k iMk B log2 1 + k,i
2
EEk = P
, (12)
iMk pk,i
+ nok (Pdyn,k + Pdyn,0 ) + Psta,k

where the user-level power consumption counts the total

transmission power of the user, the overall circuit power of
the user and the dynamic processing power of the AP related
to this user.
Now, we find the optimal power control to maximize the
user EE. The problem is formulated as
max

{pk,i }

s.t.

EEk
X
nok =
I(pk,i ), 1 k K, i Mk ,
iMk
k,i
pk,i Pmax
,

pk,i 0,

1 k K, i Mk ,

1 k K, i Mk .

(13)

Define k as the set of active links for user k and then nok is
the cardinality of k . Given any k , it is easy to prove that
EEk is strictly quasiconcave in pk,i . Thus, similar to the link
EE, the optimal power allocation under set k satisfies

P k,i
Bk
2 max
pk,i =

, k, i k .
(14)
EEk ln 2
gk,i 0
Note that if pk,i = 0, it suggests that this link should
not be active in the optimal solution, but its corresponding
circuit power Pdyn,k + Pdyn,0 has already been accounted in
calculating the total power consumption in (13). Therefore, we
have to obtain the set k in which all radio links are allocated
with strictly positive powers in maximizing EEk .

Let EE
denote the optimal intermediate user EE of user
k
k when its current set of active links is k , and then the

value of EE
can be obtained by (13) and (14). The next
k
theorem provides a general condition for determining whether
an arbitrary link should be scheduled.

Theorem 1: For any link i

/ k , if EE
eek,i , then
k

there must be EEk EEk {(k,i)} eek,i , and the link i

should be activated and added to k ; else if EE

> eek,i ,
k

then there must be EEk > EEk {(k,i)} > eek,i , and the
link i should not be activated and added to k .
Proof: Please see Appendix A.
The interpretation is also obvious: the new link i should
have a better utilization of the power than its user. In what
follows, we introduce how to obtain the optimal user EE
based on the link EE, and the details of this procedure are
summarized in line 1-14 of Algorithm 1.
Sort all radio links of user k according to their link EE eek,i
in descending order, i.e., eek,1 eek,2 ... eek,nk , and
set the initial k = . Then we successively take one link
from the order and judge whether it should be added to k .
Until some link is determined not to be activated or all links
are activated, then based on the current k , we can obtain the
optimal user EE.
Remark 1: The optimality of the proposed procedure for
maximizing the user EE is ensured by the ordering of the link
EE as well as the conclusion of Theorem 1. This idea opens
up a new way to address the fractional-form EE maximization
problem.

B. User Scheduling and Link Adaptation

In this subsection, we show how to solve the original
problem (9) based on the link EE and the user EE. For
the explanation convenience, we first introduce two auxiliary
sets. Denote as the set of active links of all users, i.e.,
= {(k, i) | pk,i > 0, i, k}, with its optimum denoted as .
Denote U as the set of scheduled users which have at least one
active link belonging to set , i.e., U = { k | (k, i) , k, i}.
Apparently, U can be sufficiently determined by .
Given the set of overall active links , and accordingly the
set of scheduled users U , then nok can be readily calculated
and problem (9) is simplified into the following problem
P
P
pk,i gk,i

i B log2 1 + 2
kU k


P
max P
p
i pk,i
o ) + no P
+
P
(n
Ck
kU
k
k dyn,0 + Psta,0

k,i
s.t. 0 < pk,i Pmax
,

1 k K, i .

(15)

Obviously, problem (15) can be verified as a standard quasiconcave optimization problem and thereby can be readily
solved as (13). Then our task is transformed to find the
scheduled users and its corresponding active links. Recall that
in obtaining the optimal user EE, some links may not be
activated and for all inactive links, use (k , i ) to denote them.
Then we define each inactive link, say link i of user k as
a virtual user just like the real users in the system, and let
{(k , i )} = . Therefore, the EE of this virtual user is

exactly the EE of link i of user k, i.e., EE

= eek,i . In

the rest, unless specified otherwise, term user refers to both

real users and virtual users. The difference between the real
user and the virtual user is that each real user may contain
several links and its circuit power includes the static user
scheduling power Psta,k as well as the link-dependent power

Algorithm 1 Energy-Efficient Scheduling Algorithm

1: for k = 1 : K
2:
Compute eek,i for all i Mk , by (10) and (11);
3:
Sort all links of user k in descending order of eek,i ,xxxx
xxi.e., eek,1 eek,2 ... eek,nk ;

4:
Set k = and EE
= 0;
k
5:
for i = 1 : nk

6:
if EE
6 ee
k
S k,i do
7:
k = k {(k, i)} ;

8:
Compute pk,i and EE
by (13) and (14);
k

9:
else EEk > eek,i
10:
k = k ;

11:
EE
return
= EE ;
k
k
12:
end
13:
end
14: end
15: Sort all users (include both real users and virtual users ) in

descending order of EE
, i.e., EE EE , ...,
1
2
k

EE ;
L

16: Set = , U = , and EE

= 0;
17: for k = 1 : L

18:
if EE
6 EE
do
S k
S
19:
= k and U = U {k} ;

20:
Obtain pk,i and EE
by solving problem (15);

21:
else EE > EE
k
22:
= ;

23:
EE
return
= EE ;
24:
end
25: end

Pdyn,k + Pdyn,0 , while each virtual user only contain one link
and its circuit power thereby is given by Pdyn,k + Pdyn,0 .
We first sort all users in descending order according to the

user EE EE
, i.e., EE EE , ..., EE , where
1
2
L
k
L is the overall number of real users and virtual users. Then,
we have the following lemma to characterize a property of the
order.
Lemma 1: Assume that the virtual user is derived from
the link i of the real user k. Following the descending order
of the user EE, the order index of this virtual user must be
larger than that of its associated real user k.

Proof: According to the user EE, we have EE

> ee
k,i ,
k

i.e, EE > EE . Therefore, when they are mixed together

k
to specify the order, the virtual user (inactive link) must be
ranked after its corresponding real user k.
This lemma guarantees that those virtual users (inactive
links) of real user k must be less likely to be active in the
system EE compared with the user k (the active links in the
user EE), otherwise it may lead to the case that some link is
scheduled finally in the system, but its associated real user is
not scheduled, which contradicts the reality.
In the following, we explore the special structures of the
system EE, the user EE and the link EE, and show how to
obtain the optimal set . In each round, we add one user to
the set U following the order and add all its active links in

k to , respectively. Then based on , the optimal system

EE EE
can be calculated as (15). By the following theorem,
we obtain the maximum system EE of problem (9).

Theorem 2: 1) For any k

/ , If EE
EE
, then
k

there must be EE EE EE ; else if EE

>
k
k

EE , then there must be EE > EE > EE ; 2) If

k
k
k
any user k is scheduled, then all active links in terms of the
optimal user EE will also be activated in maximizing system
EE.
Proof: Please see Appendix B.
The first statement suggests that in each round, the compar

ison result of EE
and EE
is necessary and sufficient to
k
determine whether the kth user can be scheduled to improve
the system EE. While the second statement guarantees the optimality of the active links in maximizing the system EE. This
theorem guarantees the optimality of the proposed method
which exhibits the concept of divide-and-conquer following
the EE of three levels. The process of method is summarized
in Algorithm 1 and it is easy to show that the complexity of the
divide-and conquer approach overall has a linear complexity
of the power control.
C. Impact of Static Receiving Power on User Scheduling
The next theorem reveals the relationship between the user
scheduling and the static receiving power.
Theorem 3: 1) The optimal number of scheduled users in
maximizing the system EE is nondecreasing with the static
receiving power Psta,0 ; 2) When Psta,0 is negligible, i.e.,
Psta,0 0, TDMA is optimal for energy-efficient transmission; 3) When Psta,0 is sufficiently large, all users will be
scheduled for energy-efficient transmission.
Proof: Due to the space limitation, we only provide a
sketch of the proof here. It is easy to show that the system EE
is decreasing with the static receiving power Psta,0 . Then, from
Theorem 2, we can show that less users would be scheduled
for a higher system EE. A more detailed proof will be given
in the journal version of this paper.
The intuition is that when Psta,0 is larger, the additional
power consumption brought from scheduling users is less
dominant, which makes it more effective to achieve higher
EE. If there is no additional power consumption for operating
systems, i.e., Psta,0 = 0, the optimal energy-efficient strategy
is only to schedule the best user where the best is in terms
of the user EE. It has the similar interpretation as that of the
throughput maximization problem in TDMA systems: only
the user with the best channel gain will be scheduled. From
Theorem 3, it is also interesting to note that the number of
users scheduled can not be guaranteed although the weights
have been imposed on users, especially for the case with low
static receiving power.
IV. N UMERICAL R ESULTS
In this section, we provide simulation results to validate our
theoretical findings and demonstrate the effectiveness proposed
methods. There are eight equally weighted users in the system
and each user is configured with twenty radio links. Without

TABLE I
SYSTEM PARAMETERS
Description
2 GHz
15 kHz
25 dBm
5000 mW
45 mW
100 mW
5 30 mW
174 dBm/Hz
0.38
1000 m
Okumura-Hata
20 dB
8 dB
Rayleigh flat fading

The number of users scheduled for EE transmission

Parameter
Carrier frequency
Bandwidth of each radio link, B
k,i
Maximal allowed transmit power, Pmax
Static circuit power of the AP, Psta,0
Link dependent power of the AP, Pdyn,0
Static circuit power of user k, Psta,k
Link dependent power of user k, Pdyn,k
Power density of thermal noise variance
Power amplifier efficiency,
Cell radius, r
Path loss model
Penetration loss
Lognormal shadowing
Fading

8
EE Optimal
7

2
3
4
5
6
The static receiving power of the AP, Psta,0 (W)

x 10

Energy efficiency of the system (bits/Joule)

18

Fig. 3.

Dinkelbach method
EE Optimal
EE Receiver
Throughput Optimal
EE Transmitter

16
14
12
10
8
6
4
2
0

10

15

20

25

Transmit power Pk,i

(dBm)
max

Fig. 1.

The system EE versus the transmit power.

x 10

Dinkelbach method
EE Optimal
EE Receiver
Throughput Optimal
EE Transmitter

Throughput of the system (bits/Joule)

8
7
6
5
4
3
2
1
0

10

15

20

25

Transmit power Pk,i

(dBm)
max

Fig. 2.

The system throughput versus the transmit power.

k,i
loss of generality, Pmax
is assumed the same for all users and
is assumed as 1. Other system parameters are listed in Table
I according to [11], [14] unless specified otherwise.

The number of scheduled users versus the static receiving power.

In Fig. 1, we compare the EE of the following methods: 1)

Dinkelbach method: the existing optimal method [4]; 2) EE
Optimal: joint Tx and Rx optimization; 3) EE Transmitter:
based on the Tx side optimization [15]; 4) EE Receiver: based
on the Rx side optimization [15]; 5) Throughput Optimal:
based on the throughput maximization. In Fig. 1, we can
first observe that our proposed method performs the same
as the Dinkelbach method, which demonstrates its optimality.
Moreover, as the transmit power increases, the performance of
the EE Optimal scheme first increases and then approaches a
constant because of its energy-efficient nature, while those of
the Throughput Optimal scheme and the EE Receiver scheme
first increase and then decrease due to their greedy use of
power. It is also interesting to note that the EE Receiver
scheme approaches the EE Optimal scheme in the low transmit
power regime while it is more close to the Throughput
Optimal scheme in the high transmit power regime. A similar
phenomenon can also be found in Fig. 2 in terms of the system
throughput. Moreover, the EE Transmitter scheme results in
both low EE and spectral efficiency due to the fact that only
one user is scheduled, which has been theoretically shown in
Section III-C.
Fig. 3 further demonstrates our theoretical findings in Theorem 3 which characterizes the monotonicity of the number
of users scheduled with Psta,0 . We observe that when Psta,0 is
negligible, the optimal energy-efficient strategy is to schedule
only one user. As Psta,0 increases, more users are scheduled to
improve the system EE through boosting the system throughput.
V. C ONCLUSIONS
This paper investigated the joint transmitter and receiver
EE maximization problem in multi-radio networks. A holistic
and dynamic power consumption model was established for
the considered system. Then, the EE maximization problem
is directly addressed from the fractional perspective, which
results in an linear divide-and-conquer approach. Moreover,

we pointed out that the static receiving power has an implicit

interpretation for the optimal number of scheduled users. In
the extreme case when the static receiving power is negligible,
TDMA is the optimal scheduling strategy. In order to meet
the QoS in practice, we then extended the propose method
to solve the problem with minimal user data rate constraints,
which exhibits good performance with linear complexity.
A PPENDIX A
P ROOF OF T HEOREM 1
Denote pk,i as the optimal power corresponding to eek,i
by (10) and (11). Also, denote pk,i and pk,i as the optimal
S

powers corresponding to EE
{(k,i)} and EEk by (15),
k
k,i
respectively. Let Sk , {pk,i |0 pk,i Pmax
, i Mk , k =
t pk,i
1, ..., K} and Pk,i (pk,i ) = + t Pdyn,k . Then, we have
the following

=
=

S
EE
{(k,i)}
k
Pi
k rk, (pk, )
max Pi =1
pk Sk
=1 Pk,i (pk, ) + Psta,k
Pi1
k rk, (
pk, ) + k rk,i (
pk,i )
Pi1=1
P
(
p
)
+
P
+
P
pk,i )
sta,k
k,i (
=1 k, k,
Pi1

pk, ) + k rk,i (pk,i )

=1 k rk, (
Pi1
pk, ) + Psta,k + Pk,i (pk,i )
=1 Pk, (
)
( P
i1

r
(p
)

r
(
p
)
k
k,i
k
k,
k,
k,i
,
min Pi1 =1

pk, ) + Psta,k Pk,i (pk,i )

=1 Pk, (


min EE
, eek,i .
(16)
k

On the other hand,

S
EE
{(k,i)}
k
Pi1
k rk, (
pk, ) + k rk,i (
pk,i )
Pi1=1
Pk, (
pk, ) + Psta,k + Pk,i (
pk,i )
=1
)
( P
i1
pk, )
k rk,i (
pk,i )
=1 k rk, (
max Pi1
,
pk,i )
pk, ) + Psta,k Pk,i (
=1 Pk, (
)
( P
i1

r
(p
)

r
(
p
)
k
k,i
k
k,
k,
k,i
max Pi1 =1
,

pk, ) + Psta,k Pk,i (pk,i )

=1 Pk, (



(17)
max EE
, eek,i .
k

Based on (16) and (17), we have





min EE
, eek,i 6 EE
{(k,i)} 6 max EEk , eek,i .
k
k
(18)
By (18), Theorem 1 can be easily proved.
A PPENDIX B
P ROOF OF T HEOREM 2
The statement 1) in Theorem 2 can be similarly proved by
an extension of Theorem 1, thus we omit them for brevity.
We now prove 2) by contradiction. Assume that user k is
scheduled, but the link i, for i k , is not activated in
maximizing the system EE, i.e., (k, i)
/ . In Theorem 1,

we have shown that the sufficient and necessary condition of

any i k is that EE
eek,i . Thus, it follows that
k

EE
eek,no eek,i ,
i k ,
k
k

(19)

where nok also denotes the last link activated according to

the link EE order since user k overall has nok links activated.
On the other hand, since user k is scheduled, we must have

EE
EEk by Theorem 2. Combining with (19), it follows
that

EE
i k , (20)
EE eek,no eek,i = EE ,

k
k

where the virtual user expression is adopted, i.e.,{(k, i)} =

. According to 1) in Theorem 2, there must be

S
S
EE
EE
= EE {(k,i)} .

(21)

Therefore, from (21), we can conclude that scheduling the link

i of user k should be scheduled in maximizing the system EE,
which contradicts the assumption that (k, i)
/ .
R EFERENCES
[1] J. Wu, S. Rangan, and H. Zhang, Green Communications: Theoretical
Fundamentals, Algorithms and Applications. CRC Press, 2012.
[2] X. Xiao, X. Tao, and J. Lu, QoS-aware energy-efficient radio resource
scheduling in multi-user OFDMA systems, IEEE Commun. Lett.,
vol. 17, no. 1, pp. 7578, 2013.
[3] G. Miao, Energy-efficient uplink multi-user MIMO, IEEE Trans.
Wireless Commun., vol. 12, no. 5, pp. 23022313, 2013.
[4] D. W. K. Ng, E. S. Lo, and R. Schober, Energy-efficient resource allocation in multi-cell OFDMA systems with limited backhaul capacity,
IEEE Trans. Wireless Commun., vol. 11, no. 10, pp. 36183631, 2012.
[5] K. Cheung, S. Yang, and L. Hanzo, Achieving maximum energyefficiency in multi-relay OFDMA cellular networks: A fractional programming approach, IEEE Trans. Commun., vol. 61, no. 7, pp. 2746
2757, 2013.
[6] H. Zhang, Z. Zhang, X. Chen, and R. Yin, Energy efficient joint source
and channel sensing in cognitive radio sensor networks, in Proc. IEEE
ICC, 2011, pp. 16.
[7] H. Kim and G. De Veciana, Leveraging dynamic spare capacity in
wireless systems to conserve mobile terminals energy, IEEE/ACM
Trans. on Networking, vol. 18, no. 3, pp. 802815, 2010.
[8] S. Luo, R. Zhang, and T. J. Lim, Joint transmitter and receiver
energy minimization in multiuser OFDM systems, arXiv preprint
arXiv:1312.6743, 2013.
[9] G. Auer, O. Blume, V. Giannini, I. Godor, M. Imran, Y. Jading, E. Katranaras, M. Olsson, D. Sabella, P. Skillermark et al., D2. 3: Energy
efficiency analysis of the reference systems, areas of improvements and
target breakdown, EARTH, 2010.
[10] M. Gruber, O. Blume, D. Ferling, D. Zeller, M. A. Imran, and E. C.
Strinati, EARTH: energy aware radio and network technologies, in
Proc. IEEE PIMRC, 2009, pp. 15.
[11] Z. Xu, C. Yang, G. Y. Li, S. Zhang, Y. Chen, and S. Xu, Energyefficient configuration of spatial and frequency resources in MIMOOFDMA systems, IEEE Trans. Commun., vol. 61, no. 2, pp. 564575,
2013.
[12] J. Mao, G. Xie, J. Gao, and Y. Liu, Energy efficiency optimization
for OFDM-based cognitive radio systems: A water-filling factor aided
search method, IEEE Trans. Wireless Commun., vol. 12, no. 5, pp.
23662375, 2013.
[13] S. Boyd and L. Vandenberghe, Convex Optimization.
Cambridge
University Press, 2004.
[14] S. Zhang, Y. Chen, and S. Xu, Joint bandwidth-power allocation for
energy efficient transmission in multi-user systems, in Proc. IEEE
GLOBECOM, 2010, pp. 14001405.
[15] Q. Wu, W. Chen, M. Tao, J. Li, H. Tang, and J. Wu, Resource allocation
for joint transmitter and receiver energy efficiency maximization in
downlink OFDMA systems, IEEE Trans. Commun., 2014, to appear.