This article has been accepted for inclusion in a future issue of this journal.
Content is final as presented, with the exception of pagination.
IEEE/ACM TRANSACTIONS ON NETWORKING
Optimal Partial Relaying for Energy-Harvesting
Wireless Networks
Mohamed Kashef, Member, IEEE, and Anthony Ephremides, Life Fellow, IEEE
Abstract—In this paper, we asses the benefits of using partial re-
laying in energy-harvesting networks. We consider a system com-
posed of a source, a relay, and a destination. Each of the source
and the relay has energy-harvesting capability and generates its
own traffic. The source is helped by the relay through a partial
relaying network-level cooperation protocol. The relay regulates
the arrivals from the source by accepting only a proportion of the
successfully received packets at the relay. The relaying parameter,
which determines the proportion of packets to be accepted, is se-
lected based on the parameters of the network to ensure the sta-
bility of the source and the relay data queues. In this work, we
provide an exact characterization of the stability region of the net-
work. We derive the optimal value of the relaying parameter to
maximize the stable throughput of the source for a given data ar-
rival rate to the relay. Also, we compare the stability region of the
proposed strategy with partial relaying to the stability regions of
simple transmission strategies. Finally, we consider the problem of
network utility optimization in which we optimize over the value
of the relaying parameter for a given pair of data arrival rates for
the source and the relay.
Index Terms—Cooperative communications, energy harvesting,
partial relaying, stability analysis.
I. INTRODUCTION
E NERGY harvesting enables wireless nodes to be
recharged by the surrounding environment. Recent
advances in energy-harvesting materials and ultra-low-power
communications will soon enable the realization of energy-har-
vesting networks [1], [2]. Nodes can harvest energy from
nature through various different sources, such as solar cells,
vibration absorption devices, water mills, thermoelectric gen-
erators, and microbial fuel cells. Examples of the techniques
of energy harvesting from nature can be found in [3]–[5]. The
energy-harvesting nodes are used in different types of networks
such as rechargeable sensor networks [6] and energy harvesting
active networked tags (EnHANTs) [7]. Such networks have
applications in various areas, which motivates studying dif-
ferent aspects related to energy-harvesting networks.
In the systems where nodes harvest energy from nature,
energy can be modeled as an exogenous recharge process.
Manuscript received September 18, 2013; revised August 05, 2014; ac-
cepted September 07, 2014; approved by IEEE/ACM TRANSACTIONS ON
NETWORKING Editor A. Walid. This work was supported in part by the MURI
under Grant W911NF-08-1-0238, the NSF under Grants CCF0905204 and
CNS1147730, and the ONR under Grant N000141110127.
The authors are with the Department of Electrical and Computer Engi-
neering, University of Maryland, College Park, MD 20742 USA (e-mail:
mkashef@umd.edu; etony@umd.edu).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TNET.2014.2361683
1063-6692 © 2014 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
1
Therefore, unlike traditional battery-powered systems, energy
is not a deterministic quantity in these systems, but is a random
process that varies stochastically in time. In our work, we
deal with the harvested energy as a stochastic process without
considering the energy-harvesting technique. When dealing
with nodes powered by nonrechargeable batteries, the common
objectives are short-term such as maximizing the lifetime of
the network [8], [9]. The harvesting capability enables us to
consider different performance measures such as the throughput
and the stability of the network [10].
On the other hand, cooperative diversity enables single an-
tenna users to benefit from the spatial diversity by delivering
data with the help of relay nodes. Cooperative diversity in en-
ergy-harvesting networks can be exploited to enhance the ca-
pacity and the coverage of the networks.
In this paper, we characterize the stability region of a system
that contains a source, a relay, and a destination. The source
and the relay have energy-harvesting capability. Each of the
source and the relay has stochastic data arrivals. The relay reg-
ulates the relaying process by accepting only a proportion of
the source successfully received packets. The relay transmits
over the common medium only when the source is idle. We
start by evaluating the stability conditions for the source and the
relay data queues. Then, we combine the conditions to charac-
terize the stability region as a function of the relaying parameter.
Then, we solve the optimization problem of obtaining the op-
timal relaying parameter that maximizes the stable throughput
rate of the source for a given relay data arrival rate while main-
taining the stability of the relay data queue. Thus, we charac-
terize the stability region of the system over the whole range of
the relaying parameter. Then, we evaluate the stability region
for simple transmission strategies such as no relaying strategy
and fixed resource allocation strategy. We consider time divi-
sion multiple access (TDMA) as an example for fixed resource
allocation strategies. Finally, we study the problem of a general
network utility optimization for a given pair of the source and
the relay data arrival rates. We characterize the relaying param-
eter feasibility range for the optimization problem.
The study of a simple model consisting of only a source, a
relay, and a destination is both instructive and necessary. The
importance of this work is to reveal insights at the conceptual
level about the effects of cooperative relaying and exploiting
channel characteristics on the stability of energy-harvesting sys-
tems. More work needs to be done to exploit the results of this
work in more realistic systems. Also, energy-harvesting capa-
bility can much affect the dynamic behavior of the proposed
system, but it is out of this paper scope.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
2 IEEE/ACM TRANSACTIONS ON NETWORKING
This work also can be considered as a building block in con-
sidering the problem of data transmission with the help of mul-
tiple relays. Employing multiple relays needs further investiga-
tion on the scheduling of these relays transmissions. Also, dif-
ferent levels of information passing in the network have to be
studied including the statistics of the channels between different
nodes. Dealing with multiple copies of the packets at different
relays is another challenge that faces the operation of the system
with multiple relays.
Our contributions in this work can be summarized as follows.
• We investigate the interaction between energy harvesting,
cooperative relaying, and stability in wireless networks.
• We characterize the stability region of the optimal partial
relaying strategy for the source and the relay in a simple
relaying network with a single relay that generates its own
traffic.
• We obtain the optimal value of the relaying parameter to
maximize the source stable throughput for a given relay
throughput. We obtain this optimal value in closed form as
a function of the system parameters.
• We obtain the stability regions for simple relaying strate-
gies, which are the no relaying strategy and the TDMA
strategy.
• We quantify the improvement in the performance due to
the use of optimal partial relaying.
II. RELATED WORK
There has been recent research effort in understanding data
transmission in energy-harvesting networks [11]–[21]. In [11],
an optimal admission control policy is obtained for data trans-
mission with energy-harvesting sensors. In [12], energy man-
agement policies that stabilize the data queue are proposed for
single-user communication under a linearity assumption for the
power-rate relation. In [13], the problem of throughput-optimal
energy allocation is studied for energy-harvesting systems in
a time-constrained slotted setting. In [14] and [15], minimiza-
tion of the transmission completion time is considered in an en-
ergy-harvesting single-user system. In [16], the problem of min-
imization of the transmission completion time for energy-har-
vesting transmitters with batteries of finite energy storage is
considered. In [17] and [18], optimal transmission policies are
obtained for a single-user energy-harvesting transmitter oper-
ating over a time-varying channel. In [19]–[21], optimal trans-
mission policies are developed for broadcast channel with an
energy-harvesting transmitter.
On the other hand, numerous works have been done to an-
alyze cooperative diversity at the physical layer based on in-
formation-theoretic considerations [22], [23]. It has also been
shown that cooperation can be applied at the network layer.
In [24], a network-level cooperation protocol has been used to
increase the stable throughput region for the uplink of a wire-
less network. Also in [25], a network-level cooperation pro-
tocol has been exploited to enhance the performance in a mul-
ticasting scenario. A network-level partial relaying is consid-
ered before in [26], where the stability region of a system with
a source, a relay, and a destination is characterized. The nodes
are non-energy-harvesting, and they access the channel through
Fig. 1. System model.
a random access technique. In [26], the effect of relaying con-
trol on the system performance was investigated. Our work is
different than [26] in that we consider energy-harvesting nodes
with random energy arrivals.
Also, cooperative diversity in energy-harvesting networks at
the physical layer has been considered before in a number of
works as in [27] and [28]. The problem of transmission con-
trol for energy-harvesting networks with network-level coop-
eration has been discussed in [29]. The authors have derived
the maximum stable throughput rate for the source node in a
network consisting of a source, a relay, and a destination. The
relay does not generate its own traffic. The relaying strategy is
TDMA. In this strategy, the odd time-slots are assigned to the
source transmissions, and the even time-slots are assigned to the
relay transmissions. This strategy has low channel utilization
because of the fixed assignment of the time-slots. As a result, it
has been shown in [29] that the direct transmission has higher
stable throughput than this relaying scheme for high energy ar-
rival rates. In our work, we propose a relaying scheme that has
better channel utilization than the relaying scheme in [29]. Also,
we consider the case when the relay has its own traffic, and thus
we consider the stability region of the network.
III. SYSTEM MODEL
A. Network Model
We consider a network that consists of a source node, a relay
node, and a destination node as shown in Fig. 1. Each of the
source and the relay has an infinite data queue for storing fixed-
length packets. These queues are denoted by and , re-
spectively. We assume that the source generates its own traffic,
while the relay both generates its own traffic and relays the
source traffic. The data arrival processes to the source and the
relay data queues are modeled as Bernoulli processes. Also,
each of the source and the relay has an infinite energy queue.
These queues are denoted by and , respectively. The
usage of infinite queues is a reasonable approximation when
the data queues are large enough compared to the packet size
and the energy queues are large enough compared to the en-
ergy unit [30]. Each of the source and the relay can acquire
a single unit of energy at each time-slot with probabilities
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
KASHEF AND EPHREMIDES: OPTIMAL PARTIAL RELAYING FOR ENERGY-HARVESTING WIRELESS NETWORKS
and , respectively, that the energy arrival processes are mod-
eled by Bernoulli processes. All nodes are half-duplex, and thus
they cannot transmit and receive simultaneously. Time is as-
sumed to be slotted such that each packet transmission takes
one time-slot. Transmission of a data packet from a node re-
quires using a single unit of energy from the corresponding en-
ergy queue. For simplicity, we assume that the energy consump-
tion in a node is due to transmission only, and therefore the pro-
cessing and the reception energy are considered to be negligible.
B. Channel Model
All the channels, which are denoted by , and , following two probabilities are defined to be used in the stability
are modeled as independent erasure channels. The channels are
independent of the packet arrival processes and the energy-har-
vesting processes at the source and the relay. The quality of a
channel is represented by the success probability of a packet.
The packet success probabilities are denoted by , and
. These success probabilities are determined by the system
physical parameters such as transmission power, modulation
scheme, coding scheme, and targeted bit error rate.
C. Transmission Strategy
At a time-slot, the source is able to transmit if both its en-
ergy queue and its data queue are not empty. If the packet is ac-
cepted by the destination or by the relay, it is released from the
source data queue; otherwise it is kept in the source data queue
for retransmission. At the beginning of every time-slot, the relay
senses the channel. We assume perfect sensing by the relay for
the source transmissions. If the source is not transmitting, the
relay uses this idle time-slot to transmit a packet from its data
queue to the destination when its energy queue is not empty.
Hence, no explicit channel resources are assigned to the relay. A
packet is released from the relay data queue if it is successfully
received by the destination; otherwise it is kept for retransmis-
sion. We assume that the destination and the relay broadcast an
acknowledge (ACK) signal when accepting a packet over a per-
fect feedback channel. The source keeps a packet that has been
transmitted if no ACK signal is received from either the relay
or the destination, and the relay keeps its transmitted packet if
it is not successfully received by the destination.
We exploit partial relaying cooperation that the relay accepts
only a certain proportion of the successfully received packets.
We need here to distinguish between accepting the packets and
successfully receiving the packets at the relay. The packet is ac-
cepted if it is kept at the relay data queue and hence released
from the source data queue. The packet is successfully received
if it succeeded to go through the channel. The successfully re-
ceived packets may be accepted or not at the relay based on
the harvested energy limitations. The proportion of accepted
packets should match the ability of the relay to forward the
packets. This proportion is determined by the relaying param-
eter , which is the probability of accepting a packet at the relay
data queue given that this packet has been successfully received.
In [31], Loynes’ theorem states that if the arrival and service
processes at a queue are jointly stationary, then the queue is
stable if the average arrival rate is less than the average service
rate. Throughout the paper, we denote the average arrival rate
at the source data queue by . The average arrival rate to the
3
relay data queue due to the source transmissions is denoted by
and the average arrival rate to the relay data queue for the
packets generated by the relay is denoted by . The average
service rate of the source data queue is denoted by . Also, the
average service rate of the relay data queue is denoted by .
IV. STABILITY ANALYSIS
In this section, we derive the stability region of the proposed
transmission protocol. The system is stable if both the source
data queue and the relay data queue are stable. In the following,
we derive the stability conditions for each queue separately. The
analysis. First, the probability that a packet transmitted by the
source is accepted by the relay is denoted by , and its value
is calculated as follows:
(1)
Also, we denote the probability that a packet transmitted by the
source is accepted by either the relay or the destination by ,
and we calculate its value as follows:
(2)
1) Source Data Queue: In order to calculate the maximum
stable throughput rate for the source data queue, we have to
start by considering the service rate for the source energy
queue, which is the rate that the source node can transmit when
its data queue is saturated [32]. Each transmission attempt uses
a single unit energy. The energy departure process depends
on the channel success probability, which we assume to be
stationary. Hence, the service time follows a geometric distri-
bution independent on other network parameters including the
packet arrival process. Therefore, both the arrival and departure
processes are Bernoulli, and hence both are memoryless. As a
result, the source energy queue forms a discrete-time M/M/1
system. It follows from [33] for a discrete-time M/M/1 system
that the probability of energy queue to be not empty is the ratio
between the energy arrival rate and the energy queue service
rate. The energy queue service rate equals 1 as we consider
the case in which the data queue is saturated [32], and hence it
always attempts to transmit. The arrival rate of energy to the
source node is . As a result, the probability of the energy
queue to be not empty is
(3)
The maximum data arrival rate that maintains the stability of
the source data queue is limited by its service rate. A packet at
the source is served if it is successfully delivered to the relay
or the destination. The source data queue service rate is the
product of the success probability given that the source is able
to transmit by the probability that its energy queue is not empty.
The service rate of the source data queue is
(4)
The stability condition for the source data queue is .
2) Relay Data Queue: We start by calculating the probability
that the channel is occupied by the source transmissions, and this
probability is denoted by . As the source data queue forms a
discrete-time M/M/1 system and assuming that the source data
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

4 IEEE/ACM TRANSACTIONS ON NETWORKING

queue is stable, it follows from [33] that the probability is of . The value of the optimal as a function in is substi-
calculated as follows: tuted in (9) to get the bound of the stability region. The optimal
value of is the solution of the problem
(5)
(10)
We have calculated the value of under the source data queue
stability assumption as it is going to be used in evaluating the where the values of , and are obtained from
stability conditions for the relay data queue. If the source data (9) as follows:
queue is not stable, the overall system is not stable. The stability
(11)
conditions of the relay data queue needs to be calculated only
when the source data queue is stable.
The arrival rate for the relay data queue from the source trans-
missions is the probability that a packet is accepted by the relay (12)
at any given time-slot. The relay has an arrival at a time-slot if (13)
the source is transmitting and the channel is not in outage
while the channel is in outage. The arrival rate to the relay
is calculated as follows: The function is a linear function of with a nonnegative
slope. Thus, is nondecreasing function of .
(6) The function is continuous over the interval . We
calculate the first derivative of the function to be
Then, the service rate of the relay data queue equals
(14)
(7)
The complete derivation of the expression of the service rate Thus, the first derivative is always nonpositive, and the function
of the relay data queue is found in Appendix A. For the relay is a nonincreasing function of .
data queue to be stable, the summation of the relay own traffic The function is continuous over the interval . The
arrival rate and the arrival rate of the packets due to the source function can be either nonincreasing or nondecreasing
transmissions has to be less than the relay data queue service based on the system parameters. The value of the first derivative
rate. Thus, the stability condition for the relay data queue is of is calculated as follows:
.
3) Stability Conditions: To ensure that the system is stable, (15)
both the source and the relay data queues have to be stable. As a
result, both the conditions and should
where is the first derivative of with respect to . The
be satisfied. By substituting using (6) in the second condition, it
value of is
is written as
(16)
(8)
Also, the difference between and is
Note that the right-hand side of the inequality is still a func-
tion of . By combining the conditions and moving to one side (17)
of the inequality, we get the general expression for the maximum
stable throughput of the source as follows: Thus, the sign of the term determines the mono-
tonicity of the function . If it is nonnegative, the function
is nondecreasing with respect to , and if the term
is nonpositive, the function is nonincreasing with respect to .
(9) In Sections V-A and V-B, we will consider the optimal value of
in both cases.
The complete derivation for the overall system stability condi-
tion can be found in Appendix B. A.
The same analysis is still valid when the energy arrival pro-
We consider the case in which the channel from relay to the
cesses and the data arrival process are modeled by Poisson pro-
destination has better quality than the channel from the source to
cesses. In this case, the energy queues and the data queues form
the destination. In this case, the function is nondecreasing
M/G/1 systems.
function of . We calculate the intersection points of with
each of and . We denote these points as and ,
V. PARTIAL RELAYING OPTIMIZATION respectively. The value of is evaluated and found to be
In this section, we evaluate the value of the parameter to
(18)
maximize the stable throughput of the source for a given value
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

KASHEF AND EPHREMIDES: OPTIMAL PARTIAL RELAYING FOR ENERGY-HARVESTING WIRELESS NETWORKS 5

Then, the value of the second intersection point is calculated In the case when the quality of the channel is better than
as follows: the quality of the channel , it is preferred to let the source
transmit with the maximum transmission attempt rate that is
(19) equal to . The denominator represents the
rate with which the source data is relayed. The nominator is the
The values of and are always nonnegative when the relay rate with which the relay can forward the source transmissions.
data queue is stable as the value of is positive when The term is the rate with which the relay ac-
the queue is stable. cesses the channel, and it is multiplied by the probability that
The optimal value of in this case is calculated as follows: a packet transmitted by the relay is received by the destination.
Then, the relay data arrival rate is subtracted from the previous
(20) quantity to get the rate with which the relay can forward the
The optimal value of is 1 if the maximum of the intersection source transmissions.
points is larger than 1.
By substitution using the values of and and simpli- VI. SPECIAL CASES
fying the resulted equation, we get In this section, we consider the special cases when a node has
a continuous source of energy for transmission.

A. Case of
(21)
This is the case in which the relay has a continuous source of
In the case when the quality of the channel is better than energy. Thus, the service rate of the relay is limited only by the
the quality of the channel , it is preferred to let the relay channel occupation due to the source transmissions. Then, the
transmit with the maximum transmission attempt rate that is expression of the relay service rate can be rewritten as follows:
equal to . The numerator represents the rate
with which the relay could forward the source transmissions. It (25)
is the difference between the relay service rate and the relay own As a result, the general expression for the maximum stable
traffic arrival rate. The denominator is the rate of the source data throughput of the source is stated as follows:
to be relayed. The term is the rate with which
the source accesses the channel, and it is multiplied by the prob- (26)
ability that a packet transmitted by the source is received by the
The expression of the maximum stable throughput of the source
relay only.
can be written as
B. (27)
In this section, we discuss the optimal relaying parameter As a result, the optimal value of can be obtained following
when the channel from the source to the destination has better the same steps of the general case. When , both
quality than the channel from the relay to the destination. In and are nondecreasing functions in .
this case, the function is a nonincreasing function of . Then, the optimal value of is
We calculate the intersection points of with each of
(28)
and . We denote these points as and , respectively.
The value of is calculated in the Section V-A. The value of On the other hand when , the optimal value of is
is (29)

(22)

The optimal value of in this case is calculated as follows: (30)

It is intuitive that when and , the op-
(23) timal value of is 1. In this case, the channel from the relay to
The optimal value of is 1 when both and are larger than the destination is better than the channel from the source to the
1. On the other hand, the optimal value of is 0 if is non- destination and there is no energy limitation at the relay. Then,
positive considering that the value of is always nonnegative there is no reason for the relay to reject a successfully received
when the system is stable. packet from the source. This explanation is true also when the
By substitution, using the values of and and simpli- value of is large enough to forward all the successfully re-
fying the resulted equation, we get ceived packets from the source. To get the condition on for
the optimal to be 1 when , we get the value at
which . Then, the condition is

(24) (31)
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

6 IEEE/ACM TRANSACTIONS ON NETWORKING

B. Case of . Then, the service rate of the relay data queue

This is the case in which the source has a continuous source equals
of energy. Thus, the service rate of the source is limited only by (40)
the channels success probabilities. Then, the expression of the
source service rate can be rewritten as follows: Thus, the stability condition for the relay data queue is
. To ensure that the system is stable, both the conditions
(32) and should be satisfied. By combining
As a result, the general expression for the maximum stable the conditions on , we get the general expression for the sta-
throughput of the source is stated as follows: bility region as follows:

if (41)

(33)
B. Fixed Resource Allocation
The optimal value of can be obtained following the same
steps of the general case. When , the value of In this section, we evaluate the stability region in the case
is greater than or equal to the value of of TDMA scheduling for the source and the relay. The same
because , that is transmission strategy as described in Section II is exploited ex-
cept for the time allocation. The odd time-slots are assigned for
(34) the source, and the even time-slots are assigned for the relay,
which is the same technique used in [29]. The value of the
Then, the optimal can be defined to be
source service rate that is denoted by is calculated
(35) using the same steps as the calculation of the relay service rate in
Appendix A. The probability with which the source can access
the channel is 1/2. Then, the probability of the source energy
queue to be not empty is calculated as follows:
(36)
When , we found that (42)

(37)
This quantity is a nonpositive quantity from the definition of . ability that the channel is occupied by source transmissions is
Thus, the optimal value of is
(38)
It is intuitive that when and , the optimal
value of is 0. In this case, the channel from the source to the
destination is better than the channel from the relay to the des-
tination and there is no energy limitation at the source. Then,
there is no reason for the source to be helped by the relay. This
explanation is true also when the value of is large enough to
forward all the packets from the source. To get the condition on
for the optimal to be 0 when , we get the value
at which . Then, the condition is
(39)
VII. STABILITY REGIONS OF SIMPLE STRATEGIES
A. No Relaying
We consider the case in which the source and the relay do
not cooperate. The source has higher priority than the relay
that the relay transmits only when the source is idle. The re-
sults for this case are obtained by setting the parameter to
0. The importance of this case is that it represents the case
of the resource allocation with no cooperation and no interfer-
ence between energy-harvesting nodes. The value of the source
service rate with “No Relaying” that is denoted by is cal-
culated as . The value of the probability that
the channel is occupied by source transmissions is calculated as
Then, we obtain the expression for the source data queue ser-
vice rate as . The value of the prob-
calculated as . The calculation of the service
rate of the relay data queue follows the steps. Also, the proba-
bility with which the source can access the channel is 1/2. Then,
the probability of the source energy queue to be not empty is cal-
culated as follows:
(43)
Then, the service rate of the relay data queue equals
(44)
Thus, the stability condition for the relay data queue is
. To ensure that the system is stable, both the
conditions and should be
satisfied. By combining the conditions on , we get the general
expression for the stability region as follows:
(45)
The maximum source stable throughput rate is the minimum
of two functions. The first function is a nondecreasing function
of as shown in Section IV. The second function is a nonin-
creasing function of . The optimal value of is the intersection
point of the two functions if this intersection point is less than
1; otherwise, the optimal value of is 1. Thus, the expression
for the optimal value of is written as follows:
(46)
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

KASHEF AND EPHREMIDES: OPTIMAL PARTIAL RELAYING FOR ENERGY-HARVESTING WIRELESS NETWORKS 7

VIII. UTILITY OPTIMIZATION limits of the range by directly dealing with stability conditions
In this section, we study a general utility optimization of the network.
problem for a partial relaying cooperative network. Let us We start by the first stability condition, which is
denote the network utility by , which is a network quality (48)
of service measure. Optimization objectives examples include
fairness utility maximization, energy consumption mini- We divide both sides by and subtract them from 1. Then, we
mization, and average delay minimization. By appropriately divide both sides by , which is nonnegative. Thus, we
choosing the utility objective function subject to a set of data get
rate constraints, we optimize the network utility for both the
source and the relay. In the optimization problem, we get the (49)
optimal value of with respect to the utility function for a pair
of data rates and that are in the stability region of the Finally, by subtracting both sides from 1 and dividing by ,
network. we get
We consider the optimization of the utility function over the
feasible domain of subject to the constraint of the source and (50)
the relay data arrival rates to be in the stability region of the
The value of this limit can be either positive or negative based
network. By obtaining the feasible range of , we can find the
on the system parameters.
optimal value of the utility function numerically by searching
Then, we deal with the second stability condition. The second
for the optimal value over the feasible range of . We write
stability condition is stated as follows:
the general optimization problem as a maximization problem
as follows:
(51)

By multiplying both sides of the inequality by

subject to
and subtracting the results at both sides
from 1, we get an inequality in which we divide both sides by
. We get the following inequality:

(52)
(47)
The range for to maintain the stability conditions will de-
The constrains are set to make sure that the data arrival rates
pend on the sign of the term multiplied by . If
to the source and the relay are in the stability region of the net-
, then
work. We now get the feasibility region for the utility optimiza-
tion problem. (53)
Feasibility Region
1) If , then
If , then is larger than some negative quan-
tity. This condition is already satisfied by maintaining .
Then, we deal with the third stability condition. The third
2) if , then stability condition is stated as follows:

(54)
3) if , then
By multiplying both sides of the inequality by
and subtracting the results at both sides from
1, we get an inequality for which we divide both sides by
where . Then, we get the following inequality:

(55)

The range for to maintain the stability conditions will depend

on the sign of the term multiplied by . If , then

Proof: We obtain the feasibility range corresponding to a (56)

pair of data arrival rates of the source and the relay. We get the
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

8 IEEE/ACM TRANSACTIONS ON NETWORKING

Fig. 3. Optimal against different system parameters.

Fig. 2. Stability region.

If , then

(57)

By combining the results obtained from the stability condi-

tions, we get the stated feasibility region.

IX. NUMERICAL ANALYSIS

In this section, we present numerical results to illustrate the
previous theoretical development. We illustrate the effects of
system parameters on the performance of the system and the
optimal value of relaying parameter. In the following results, we
fix the channel success probabilities to be
, and . We denote the system with optimal relaying
parameter with the relay senses the channel by “Optimal” and
the system with TDMA channel access techniques by “TDMA.”
In Fig. 2, we show the stability regions of different relaying
schemes. We set the energy arrival rates for the source and the
relay to and . We compare the optimal re-
laying to the cases of full relaying, no relaying, and TDMA. We
show that the stability region of optimal relaying contains the
stability region of other relaying schemes. For the selected pa-
rameters, it is optimal to use full relaying for . When
is larger, the relay does not have enough energy to forward
all the successfully received packets by the source. Also, it is
optimal not to relay source data when . In this case, the
source energy is enough to forward the source data through the
channel , and the relay uses its energy to forward its own
data. On the other hand, the TDMA scheme is optimal at one
point only when . relaying system for different values of . We set the energy
In Fig. 3, we show the value of the optimal relaying param-
eter against , and . The horizontal axis is , and we
show results for and with the values of 0.1, 0.3, and 0.5.
The figure shows that full relaying is optimal for wider range
with the increase of relay energy arrival rate. Also, partial re-
laying is more important for enhancing the performance with
Fig. 4. Stability region against different values of .
the decrease of the source energy arrival rate. Finally, the value
of the optimal relaying parameter decreases with the increase of
the data arrival rate for the relay.
In Fig. 4, we show the stability regions of the optimal partial
relaying system for different values of . We set the energy
arrival rate for the source to . We show the increase in
the stability region with increase of . This increase is much
higher for low values of , while it has lower effect for large
values of . When is large that the network cannot exploit
all the harvested energy, the enhancement in the stability region
happens only when is small such that the source is not able
to use all its harvested energy. The vertical lines in the figure
represent the case when such that the increase in does
not affect the allowable in the system.
In Fig. 5, we show the stability regions of the optimal partial
arrival rate for the relay to . We show the increase in the
stability region with increase of . This increase is much higher
for low values of , while it has lower effect for large values
of . The horizontal lines in the figure represent the case when
such that the increase in exploits unused harvested
energy at the relay without affecting the source performance.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

KASHEF AND EPHREMIDES: OPTIMAL PARTIAL RELAYING FOR ENERGY-HARVESTING WIRELESS NETWORKS 9

shown that the probability of the source to be busy is . Then,

the number of successive time-slots, in which the source is busy,
follows a geometric distribution that .
The expected value of the number of successive time-slots, in
which the source is busy, is . Let be
the number of time-slots needed for the relay to get served in-
cluding those in which the source will be transmitting, then
. This expression results from the fact that
the th transmission of the relay transmissions is followed by
busy period of length . Then, the expected value of the number
of time-slots needed for the relay to get served, including those
in which the source will be transmitting, is calculated as follows:

(58)

Thus, the service rate of the relay data queue is

.
Fig. 5. Stability region against different values of .

APPENDIX B
X. CONCLUSION DERIVATION OF THE MAXIMUM STABLE THROUGHPUT
OF THE SOURCE
In this paper, we have introduced the notion of partial
network-level cooperation for energy-harvesting networks. We combine the source data queue stability condition
The flow from the source through the relay is controlled. We and the relay data queue stability condition in (8) to evaluate the
provide an exact characterization of the stability region for the overall system stability condition. Equation (8) can be divided
discussed system. We have shown that the performance of the to the following two conditions by substituting the func-
system with partial relaying is always better than or equals the tion by its two allowable values. By substituting using for the
performance of simple relaying schemes. We characterize the function, we obtain
feasibility region for a general utility optimization problem for
the discussed system. Also, we have shown that it is optimal to (59)
use full relaying for small data arrival rate at the relay, while it
is optimal to use no relaying when the source has a small data Then, we substitute using where . The
arrival rate. condition can be written as follows:

(60)
APPENDIX A
DERIVATION OF THE SERVICE RATE FOR THE RELAY QUEUE
By moving to the left-hand side of the inequality, we get
FOR TRANSMISSION PROTOCOL WITH RELAYING

We are going to calculate the service rate of the relay data (61)
queue. Let be the probability that a packet is received by
the destination due to a relay transmission. The packet is to be
(62)
decoded successfully when the relay is able to transmit and the
channel is not in outage. The value of is calculated
as . The relay energy queue forms Then, by combining the condition that and the condi-
a discrete-time M/M/1 system. The probability of the energy tions in (59) and (62), we obtain the stability condition for the
queue to be not empty is system that has to be less than the smallest term of these three
. terms as follows:
Let be the number of time-slots needed for the relay to
serve a packet in the relay data queue assuming that the relay
continuously transmits. Then, has a geometric probability (63)
distribution that . Then,
the expected value of the number of time-slots needed until the
packet is decoded correctly by the destination, assuming that the REFERENCES
relay continuously transmits, is . [1] J. Paradiso and T. Starner, “Energy scavenging for mobile and
Let , be a sequence of random variables. The random wireless electronics,” IEEE Pervasive Comput., vol. 4, no. 1, pp.
variable represents the number of successive time-slots in 18–27, Jan.–Mar. 2005.
[2] D. Wentzloff et al., “Energy efficient pulsed-UWB CMOS circuits
which the source is going to be busy before the th relay retrans- and systems,” in Proc. IEEE Int. Conf. Ultra-Wideband, 2007, pp.
mission. This sequence represents an i.i.d. sequence. We have 282–287.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
10
[3] V. Raghunathan, A. Kansal, J. Hsu, J. Friedman, and M. B. Srivastava,
“Design considerations for solar energy harvesting wireless embedded
systems,” in Proc. IEEE IPSN, Apr. 2005, pp. 457–462.
[4] S. Meninger, J. O. Mur-Miranda, R. Amirtharajah, A. Chandrakasan,
and J. H. Lang, “Vibration-to-electric energy conversion,” IEEE Trans.
VLSI, vol. 9, no. 1, pp. 64–76, Feb. 2001.
[5] M. Gorlatova et al., “Challenge: Ultra-low-power energy-harvesting
active networked tags (EnHANTs),” in Proc. MobiCom, Sep. 2009, pp.
253–260.
[6] X. Jiang, J. Polastre, and D. Culler, “Perpetual environmentally pow-
ered sensor networks,” in Proc. Int. Symp. Inf. Process. Sensor Netw.,
Apr. 2005, Art. no. 65.
[7] M. Gorlatova et al., “Energy-harvesting active networked tags (En-
HANTs) for ubiquitous object networking,” IEEE Wireless Commun.,
vol. 17, no. 6, pp. 18–25, Dec. 2010.
[8] G. S. Kasbekar, Y. Bejerano, and S. Sarkar, “Lifetime and coverage
guarantees through distributed coordinate-free sensor activation,”
IEEE/ACM Trans. Netw., vol. 19, no. 2, pp. 470–483, Apr. 2011.
[9] T. Himsoon, W. Siriwongpairat, Z. Han, and K. J. R. Liu, “Lifetime
maximization framework by cooperative nodes and relay deployment
in wireless networks,” IEEE J. Sel. Areas Commun., vol. 25, no. 2, pp.
306–317, Feb. 2007.
[10] S. Sudevalayam and P. Kulkarni, “Energy harvesting sensor nodes:
Survey and implications,” IEEE Commun. Surveys Tuts., vol. 13, no.
3, pp. 443–461, 3rd Quart., 2011.
[11] J. Lei, R. Yates, and L. Greenstein, “A generic model for optimizing
single-hop transmission policy of replenishable sensors,” IEEE Trans.
Wireless Commun., vol. 8, no. 2, pp. 547–551, Feb. 2009.
[12] V. Sharma, U. Mukherji, V. Joseph, and S. Gupta, “Optimal energy
management policies for energy harvesting sensor nodes,” IEEE Trans.
Wireless Commun., vol. 9, no. 4, pp. 1326–1336, Apr. 2010.
[13] C. Ho and R. Zhang, “Optimal energy allocation for wireless communi-
cations powered by energy harvesters,” in Proc. IEEE ISIT, Jun. 2010,
pp. 2368–2372.
[14] J. Yang and S. Ulukus, “Transmission completion time minimization
in an energy harvesting system,” in Proc. CISS, Mar. 2010, pp. 1–6.
[15] J. Yang and S. Ulukus, “Optimal packet scheduling in an energy har-
vesting communication system,” IEEE Trans. Commun., vol. 60, no. 1,
pp. 220–230, Jan. 2012.
[16] K. Tutuncuoglu and A. Yener, “Optimum transmission policies for bat-
tery limited energy harvesting nodes,” IEEE Trans. Wireless Commun.,
vol. 11, no. 3, pp. 1180–1189, Mar. 2012.
[17] O. Ozel, K. Tutuncuoglu, J. Yang, S. Ulukus, and A. Yener, “Transmis-
sion with energy harvesting nodes in fading wireless channels: Optimal
policies,” IEEE J. Sel. Areas Commun., vol. 29, no. 8, pp. 1732–1743,
Sep. 2011.
[18] M. Kashef and A. Ephremides, “Optimal packet scheduling for energy
harvesting sources on time varying wireless channels,” J. Commun.
Netw., vol. 14, no. 2, pp. 121–129, 2012.
[19] J. Yang, O. Ozel, and S. Ulukus, “Broadcasting with an energy har-
vesting rechargeable transmitter,” IEEE Trans. Wireless Commun., vol.
11, no. 2, pp. 571–583, Feb. 2012.
[20] M. A. Antepli, E. Uysal-Biyikoglu, and H. Erkal, “Optimal packet
scheduling on an energy harvesting broadcast link,” IEEE J. Sel. Areas
Commun., vol. 29, no. 8, pp. 1721–1731, Sep. 2011.
[21] O. Ozel, J. Yang, and S. Ulukus, “Broadcasting with a battery lim-
ited energy harvesting rechargeable transmitter,” in Proc. WiOpt, May
2011, pp. 205–212.
[22] A. Sendonaris, E. Erkip, and B. Aazhang, “User cooperation diver-
sity—Part I: System description,” IEEE Trans. Commun., vol. 51, no.
11, pp. 1927–1938, Nov. 2003.
[23] J. N. Laneman, D. N. C. Tse, and G. W. Wornell, “Cooperative diversity
in wireless networks: Efficient” protocols and outage behavior,” IEEE
Trans. Inf. Theory, vol. 50, no. 12, pp. 3062–3080, Dec. 2004.
IEEE/ACM TRANSACTIONS ON NETWORKING
[24] A. K. Sadek, K. J. R. Liu, and A. Ephremides, “Cooperative multiple
access protocols for wireless networks: Protocol design and stability
analysis,” in Proc. CISS, Princeton, NJ, USA, 2006, pp. 1224–1229.
[25] A. Fanous and A. Ephremides, “Transmission control of two-user
slotted ALOHA over Gilbert-Elliott channel: Stability and delay
analysis,” in Proc. IEEE ISIT, Jul. 31–Aug. 5 2011, pp. 2532–2536.
[26] N. Pappas, J. Jeon, A. Ephremides, and A. Traganitis, “Wireless net-
work-level partial relay cooperation,” in Proc. IEEE ISIT, Jul. 1–6,
2012, pp. 1122–1126.
[27] B. Medepally and N. B. Mehta, “Voluntary energy harvesting relays
and selection in cooperative wireless networks,” IEEE Trans. Wireless
Commun., vol. 9, no. 11, pp. 3543–3553, Nov. 2010.
[28] H. Li, N. Jaggi, and B. Sikdar, “Cooperative relay scheduling under
partial state information in energy harvesting sensor networks,” in
Proc. IEEE Global Commun. Conf., Miami, FL, USA, Dec. 2010, pp.
1–5.
[29] I. Krikidis, T. Charalambous, and J. S. Thompson, “Stability analysis
and power optimization for energy harvesting cooperative networks,”
IEEE Signal Process. Lett., vol. 19, no. 1, pp. 20–23, Jan. 2012.
[30] V. Sharma, U. Mukherji, V. Joseph, and S. Gupta, “Optimal energy
management policies for energy harvesting sensor nodes,” IEEE Trans.
Wireless Commun., vol. 9, no. 4, pp. 1326–1336, Apr. 2010.
[31] R. Loynes, “The stability of a queue with non-interdependent inter ar-
rival and service times,” Proc. Cambridge Philos. Soc., vol. 58, pp.
497–520, 1962.
[32] J. Jeon and A. Ephremides, “The stability region of random multiple
access under stochastic energy harvesting,” in Proc. IEEE ISIT, Saint-
Petersburg, Russia, Aug. 2011, pp. 1796–1800.
[33] L. Kleinrock, Queueing Theory, Volume I: Theory. New York, NY,
USA: Wiley, 1975.
Mohamed Kashef (M’14) received the B.Sc. and
M.S. degrees with honors in electronics and electrical
communications engineering from Cairo University,
Cairo, Egypt, in 2006 and 2009, respectively, and
the Ph.D. degree in electrical engineering from the
University of Maryland, College Park, MD, USA, in
2013.
His research interests generally include wireless
communication systems and networks.
Anthony Ephremides (S’68–M’71–SM’77–F’84–
LF’09) received the Ph.D. degree in electrical
engineering from Princeton University, Princeton,
NJ, USA, in 1971.
He holds the Cynthia Kim Professorship of
Information Technology with the Electrical and
Computer Engineering Department, University of
Maryland, College Park, MD, USA, where he is a
Distinguished University Professor and has a joint
appointment with the Institute for Systems Research,
of which he was among the founding members in
1986. He has been with the University of Maryland since 1971. He is the
author of several hundred papers, conference presentations, and patents, has
received numerous awards, and has served the profession in numerous ways.
His research interests lie in the areas of communication systems and networks
and all related disciplines, such as information theory, control and optimization,
satellite systems, queueing models, signal processing, etc. He is especially
interested in wireless networks and energy efficient systems.