4172 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 11, NO.

11, NOVEMBER 2012

Capacity and Delay of
Probing-Based Two-Hop Relay in MANETs
Jiajia Liu, Student Member, IEEE, Juntao Gao, Student Member, IEEE, Xiaohong Jiang, Senior Member, IEEE,
Hiroki Nishiyama, Member, IEEE, and Nei Kato, Senior Member, IEEE
AbstractDue to their simplicity and efciency, the two-hop
relay algorithm and its variants serve as a class of attractive
routing schemes for mobile ad hoc networks (MANETs). With
the available two-hop relay schemes, a node, whenever getting
an opportunity for transmission, randomly probes only once a
neighbor node for the possible transmission. It is notable that
such single probing strategy, although simple, may result in
a signicant waste of the precious transmission opportunities
in highly dynamic MANETs. To alleviate such limitation for
a more efcient utilization of limited wireless bandwidth, this
paper proposes a more general probing-based two-hop relay
algorithm with limited packet redundancy. In such an algorithm
with probing round limit and packet redundancy limit f, each
transmitter is allowed to conduct up to rounds of probing for
identifying a possible receiver and each packet can be delivered
to at most f distinct relays. A general theoretical framework
is further developed to help us understand that under different
setting of and f, how we can benet from multiple probings
in terms of the per node throughput capacity and the expected
end-to-end packet delay.
Index TermsMobile ad hoc networks, two-hop relay, probing,
packet redundancy, throughput capacity, end-to-end delay.
I. INTRODUCTION
A
S nodes move around randomly in mobile ad hoc net-
works (MANETs), network topology varies dramatically
and there may exist no contemporaneous end-to-end path
at any given time instant [1][5]. As a consequence, the
traditional route-based routing schemes like DSR [6], AODV
[7], etc., fail to function properly since they require the
simultaneous availability of a number of links. The two-hop
relay routing which takes advantage of node mobility and
sequences of node contacts to deliver messages from end to
end, since rst introduced in [8], has become a promising
routing protocol for MANETs [1], [9], [10]. As shown in
[1], [10], [11], the two-hop relay and its variants, simple yet
efcient, are able to provide a exible control of both the
throughput and packet delay for the challenging MANETs.
Under such a routing scheme, a packet reaches its destination
either through a direct transmission from the source or by
two-hop transmissions via an intermediate relay node, which
rst receives the packet from the source and then forwards it
Manuscript received March 12, 2012; revised July 26, 2012; accepted
August 20, 2012. The associate editor coordinating the review of this paper
and approving it for publication was L. Cai.
J. Liu, H. Nishiyama, and N. Kato are with the Graduate School of
Information Sciences, Tohoku University, Aobayama 6-3-09, Sendai, 980-
8579, Japan (e-mail: {liu-jia, bigtree, kato}@it.ecei.tohoku.ac.jp).
J. Gao and X. Jiang are with the School of Systems Information Science,
Future University Hakodate, Kamedanakano 116-2, Hakodate, Hokkaido, 041-
8655, Japan (e-mail: [email protected], [email protected]).
Digital Object Identier 10.1109/TWC.2012.091812.120346
to the destination. Therefore, each packet travels at most two
hops to reach the destination.
The two-hop relay algorithm and its variants have been
intensively studied in literature. The algorithms in [8], [12]
[17] can be regarded as the out-of-order routing without packet
redundancy, where a packet has at most one copy and will
be accepted by its destination as long as it has never been
received before. The two-hop relays in [1], [18][25] also
adopt the out-of-order reception but multiple redundant copies
can be distributed for each packet. Later, some new two-hop
relay algorithms with in-order reception have been proposed
in [26][28] where each packet has a xed number of copies
(i.e., with exact redundancy). The two-hop relay schemes in
[9], [11], [29] also belongs to the line of in-order reception
but each packet is allowed to have a limited number of
copies (i.e., with limited redundancy). More recently, a general
group-based two-hop relay with limited redundancy was also
proposed in [30], where each packet is delivered to a limited
number of relay nodes and can be accepted by its destination
if it is among the group of packets the destination is currently
requesting.
Notice that in the available two-hop relay schemes with
packet redundancy (xed or limited), no matter adopting out-
of-order reception [1], [18][25], in-order reception [9], [11],
[26][29] or group-based reception [30], a node, whenever
getting an opportunity for transmission, randomly probes
only once a neighbor node for possible transmission if its
destination node is not within its transmission range. Such
single probing strategy, although simple, may result in a
signicant waste of the precious transmission opportunities
in highly dynamic MANETs. For example, for the case that
the transmitter regards a randomly probed neighbor node as
a relay and hopes to deliver a redundant packet copy to it, it
may happen that the relay is already carrying such a copy for
that packet; on the other hand, for the case that the transmitter
acts as a relay and hopes to forward a packet to the randomly
probed node, this node may have already received all the
packets carried by the transmitter. Thus, when a wrong node is
selected through such single probing strategy, no transmission
can be conducted successfully in the above two cases and the
transmission opportunity of the transmitter will be wasted.
To alleviate the limitation of single probing for a more
efcient utilization of wireless bandwidth, this paper considers
a general probing-based two-hop relay with limited packet
redundancy. The main contributions of this paper are sum-
marized as follows:
We propose a new two-hop relay algorithm with probing
round limit and packet redundancy limit f (2HR-(, f)
1536-1276/12$31.00 c 2012 IEEE
LIU et al.: CAPACITY AND DELAY OF PROBING-BASED TWO-HOP RELAY IN MANETS 4173
Fig. 1. Illustration of cells in a transmission-group with m = 16 and = 4.
for short), where each transmitter is allowed to conduct
up to rounds of probing for identifying a possible
receiver and each packet can be delivered to at most f
distinct relays. This algorithm covers available two-hop
routing protocols [9], [11], [26], [27] as special cases
( = 1).
We further develop a general theoretical framework
to characterize the complicated packet delivery process
under the 2HR-(, f), where the nite-state absorbing
Markov chain technique is adopted to model the packet
dispatching process at the source and the packet receiv-
ing process at the destination. By setting = 1, our
framework reduces to some available models developed
for two-hop relay [9], [11].
With the help of the theoretical framework, closed-form
expressions are derived for the per node throughput
capacity and the expected end-to-end packet delay. Exten-
sive simulation and theoretical results are also provided
to validate the efciency of the new relay algorithm and
corresponding theoretical framework.
The remainder of this paper is outlined as follows. In
Section II, we introduce the system models, the transmission-
group based scheduling scheme and the 2HR-(, f) routing
algorithm. In Section III we develop Markov chain framework
and derive closed-form expressions for the per node through-
put capacity and the expected end-to-end packet delay for
any feasible trafc input rate. We provide numerical results to
validate our scheme and theoretical framework in Section IV,
introduce the available works regarding throughput capacity
and delay analysis in Section V and conclude the whole paper
in Section VI.
II. 2HR-(, f) ROUTING ALGORITHM
A. System models
Similar to [12], [31], we consider in this paper a torus
network of unit area which is evenly divided into m m
cells. Fig. 1 shows an example of a 16 16 cell-partitioned
network where the cells are further divided into 16 distinct
transmission-groups, and all the shaded cells there belong
to the same transmission-group (as to be introduced later in
Section II-B). Time is slotted, and there are n nodes roaming
around in the torus from cell to cell according to the i.i.d.
mobility model [9], [26]. Each node employs a common
transmission range r, and the protocol model [32] with guard
factor is adopted here to account for interference issues.
A whole time slot is allocated only for data transmissions
in one-hop range, and for any node pair the data bits that
can be successfully transmitted from the transmitter to the
receiver is normalized to one packet here. We consider the
permutation trafc pattern widely adopted in previous studies
[8], [9], [31], where there are in total n distinct trafc ows
(one ow corresponds to one source-destination pair). Under
such trafc pattern, each node is not only the source of its
locally generated trafc ow but also the destination of another
trafc ow originated from some other node. The trafc ow
generated at each node is assumed to have an average input
rate (packets/slot).
B. Scheduling Scheme
We consider a local transmission scenario where a trans-
mitter in some cell can only transmit packets to receivers in
the same cell or its eight neighboring cells. Two cells are
called neighboring cells if they share a common point. Thus,
the transmission range can be determined as r =

8/m.
According to the protocol interference model, two links can
transmit simultaneously if and only if they are sufciently far
away from each other. To avoid collisions among simultaneous
transmissions and schedule as many simultaneous link trans-
missions as possible, similar to [9], [33][35], we adopt the
transmission-group based scheduling scheme.
Transmission-group: A transmission-group is a subset of
cells where any two of them have a vertical and horizontal
distance of some multiple of cells and all the cells there
could transmit simultaneously without interfering with each
other.
It is easy to see that under the transmission-group based
scheduling scheme, all the m
2
cells in the network can be
divided into
2
distinct transmission-groups. If we let each
transmission-group become active (i.e., have link transmis-
sions) alternatively, then each cell will also become active
every
2
time slots.
As illustrated in Fig. 1 for the case = 4, there are in
total 16 transmission groups, and all shaded cells belong to
the same transmission-group. Suppose that the transmission-
group 1 is active, and node S in some active cell is transmitting
to node V . Then the distance between node V and some
other transmitter in another active cell is at least 2 cells.
According to the protocol interference model, in order to
ensure the successful data reception at node V , we should
have ( 2)
1
m
(1 + ) r. Notice that m, then the
parameter can be determined as
= min,(1 + )

8 + 2|, m (1)
Now we proceed to introduce the partition of a time slot.
As shown in Fig. 2, each time slot is divided into four sub-
slots. In subslot W
1
, all nodes in an active cell contend to
become the transmitter in a DCF way, where each node there
randomly selects a back-off counter from (0, W
1
] and the node
whose counter is the rst to become zero broadcasts a message
claiming itself as the transmitter. Subslot W
2
is specied for
destination checking where the destination node of the ow
originated from the transmitter will reply to the transmitter if
4174 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 11, NO. 11, NOVEMBER 2012
Fig. 2. Partition of a time slot.
it is inside the one-hop neighborhood. Otherwise, if no reply is
heard from the destination, in subslot W
3
the transmitter will
conduct at most rounds of probing until an eligible receiver
is selected (in each probing round, a neighboring node is
randomly selected as the receiver). Subslot W
4
is reserved for
data transmission from the transmitter to the selected receiver.
If no eligible receiver is selected in subslot W
3
(and thus no
packet can be transmitted), the transmitter stays idle in subslot
W
4
.
C. 2HR-(, f) Routing Algorithm
Now we are ready to introduce the general probing-based
two-hop relay algorithm 2HR-(, f). Under such an algorithm,
each transmitter will conduct at most rounds of probing to
select an eligible receiver when its destination node is not
inside the one-hop neighborhood, and at most f copies will
be distributed out for each packet.
Notice that under the permutation trafc pattern considered
in this paper, there are in total n distinct ows and each node
can be a potential relay for other n 2 ows (excluding
the two ows originated from and destined for itself). We
assume that each node maintains in its buffer n individual
FIFO queues: one local-queue storing the locally generated
packets, one already-sent-queue storing the packets whose f
copies have been distributed but the reception status are not
conrmed yet, and n 2 relay-queues storing packets from
other n2 ows (one for each ow). For throughput capacity
analysis, we assume all queues have enough buffer space such
that no packet overow will happen.
Without loss of generality, we focus on a tagged ow
and denote by S the source node and denote by D the
destination node. We consider a scenario where the source
S and the destination D use the push-type of service for
data transmission. Specically, S periodically sends locally
generated new packets to D via intermediate relay nodes; D
can only passively receive packets from S (or relay nodes)
and never sends any request to S.
It is noticed that for the designing of relay algorithm with
packet redundancy, one common complication is that remnant
copies of a packet that has already been received at its
destination create excess congestion and must somehow be re-
moved. Together with the push-type data transmission, another
complication is that the transmitter has no idea which packet
should be sent to the receiver during each node meeting, since
it may happen that the transmitter initially plans to send a
packet say P, but the receiver has already received a copy of
packet P from another relay node (or the source). In order
to overcome these two complications, we adopt a sequence
number based mechanism for the 2HR-(, f) algorithm. For
the tagged ow, the source S labels each packet P waiting at
the local-queue with a sequence number SN(P) to represent
its arrival order, and the destination D maintains an indicator
number IN(D) to denote that D has received all packets with
sequence number less than IN(D). Every time node S (resp.
node D) moves ahead its local-queue by one packet (resp.
receives a packet), it increases its sequence number (resp.
indicator number) by one. Therefore, each packet is received
in order at the destination.
Every time S is selected as the transmitter in an active cell,
it executes the following Algorithm 1.
Algorithm 1 2HR-(, f) routing algorithm
1: S checks whether its destination D is in the one-hop
neighborhood;
2: if D is within the one-hop neighborhood of S then
3: S executes Procedure 1;
4: else
5: With probability 1/2, S randomly selects to do source-
to-relay transmission or relay-to-destination transmis-
sion;
6: if S selects source-to-relay transmission then
7: S executes Procedure 2;
8: else
9: S executes Procedure 3;
10: end if
11: end if
Procedure 1 source-to-destination transmission
1: S obtains from D the indicator number IN(D);
2: S directly sends to D the packet P with sequence number
SN(P) = IN(D);
3: S deletes all packets with sequence number less than
IN(D) from both local-queue and already-sent-queue;
4: S moves ahead the remaining packets in local-queue and
already-sent-queue;
Remark 1: It is noticed that in Procedure 1, after obtaining
the indicator number IN(D) from the destination D, the
source S deletes all packets with sequence number less than
IN(D) and tries to send a packet P with sequence number
SN(P) = IN(D) to D. Similarly, in Procedure 3, S conducts
similar buffer update and packet transmission after obtaining
the IN(V
i
) from node V
i
. The above handshake process in
Procedures 1 and 3 happens only before the data transmission
during each node meeting, and a receiver will send its indicator
number to the transmitter only during such handshake process.
III. THROUGHPUT CAPACITY AND EXPECTED
END-TO-END DELAY
In this section, we rst introduce some basic probabilities
and explore the service times at the source S and the desti-
nation D, then proceed to derive the throughput capacity and
LIU et al.: CAPACITY AND DELAY OF PROBING-BASED TWO-HOP RELAY IN MANETS 4175
Procedure 2 source-to-relay transmission
1: i 1;
2: while i do
3: S randomly selects a node (say V
i
) out of the one-hop
neighbors;
4: S checks whether the head-of-line (HoL) packet P
h
at
its local-queue is carried by V
i
;
5: if V
i
doesnt carry P
h
then
6: S delivers to V
i
a copy of P
h
;
7: if All f copies of P
h
have been distributed then
8: S puts P
h
into the end of the already-sent-queue;
9: S moves ahead the remaining packets behind P
h
in the local-queue;
10: end if
11: i + 1;
12: end if
13: i i + 1;
14: end while
Procedure 3 relay-to-destination transmission
1: i 1;
2: while i do
3: S randomly selects a node (say V
i
) out of the one-hop
neighbors;
4: S obtains from V
i
the indicator number IN(V
i
);
5: S checks whether it carries a packet P with SN(P) =
IN(V
i
) in its relay-queue specied for V
i
;
6: if S carries such packet P then
7: S sends P to node V
i
;
8: S deletes all packets with sequence number less than
IN(V
i
) from its relay-queue specied for V
i
;
9: S moves ahead the remaining packets in the relay-
queue specied for V
i
;
10: i + 1;
11: end if
12: i i + 1;
13: end while
the expected end-to-end delay under the general 2HR-(, f)
algorithm.
A. Some Basic Probabilities
Lemma 1: Consider a MANET adopting the 2HR-(, f)
routing algorithm. For a given time slot and the tagged ow,
if we use p
1
to denote the probability that the source S
conducts a source-to-destination transmission and use p
2
to
denote the probability that S conducts a source-to-relay or
relay-to-destination transmission, then we have
p
1
=
1

2
_
9n m
2
n(n 1)

_
m
2
1
m
2
_
n1
8n + 1 m
2
n(n 1)
_
(2)
p
2
=
1

2
_
m
2
9
n 1
_
1
_
m
2
1
m
2
_
n1
_

_
m
2
9
m
2
_
n1
_
(3)
Proof: The derivations of probabilities p
1
and p
2
under
the probing-based 2HR-(, f) routing algorithm, are similar
to that in [9]. Please kindly refer to [9] for details.
According to Procedure 2 of the 2HR-(, f) routing algo-
rithm, we can see that when the source node S decides to
conduct the source-to-relay transmission, it will independently
conduct at most rounds of probing (in each probing round,
a neighboring node is randomly selected as the receiver) to
deliver out a copy for its HoL packet P
h
. Then we have the
following lemma.
Lemma 2: In a MANET with 2HR-(, f) routing algo-
rithm, for a given time slot and the tagged ow, suppose
the source S is delivering copies for the HoL packet P
h
at
its local-queue, and there are already j copies of P
h
in the
network, 1 j f. If we denote by P
d
(j) the probability
that S successfully delivers a new copy of P
h
to some relay
node in the time slot, then we have
P
d
(j) =
(m
2
9)
n1
2
2
m
2n2
_
nj1

s=1
j1

t=0
s+t

k=0
_
n j 1
s
__
j 1
t
_
_
s +t
k
_
8
s+tk
(m
2
9)
s+t

1
k + 1

_
1
_
t
t +s
_

_
_
(4)
Now we proceed to explore the probability that the desti-
nation node D may receive a packet whose sequence number
equals IN(D) in Procedure 3. Consider some relay node R
carrying a packet P with SN(P) = IN(D) in its relay-queue
specied for D. For a time slot, suppose R is selected as the
transmitter and R decides to conduct the relay-to-destination
transmission. It is easy to see that R will deliver to D the
packet P if and only if the following two events happen
simultaneously: D is selected as the receiver in the t
th
round
of probing, 1 t ; for the node V
i
selected in the i
th
round of probing, 1 i < t, V
i
,= D, R does not carry any
packet P

with SN(P

) = IN(V
i
) in its relay-queue for V
i
.
Without loss of generality, we denote by p
nc
the probability
that R does not carry any packet P

with SN(P

) = IN(V
i
),
1 i < t, then we have the following lemma.
Lemma 3: In a MANET with 2HR-(, f) routing algo-
rithm, for a given time slot and the tagged ow, suppose there
are already j copies of packet P with SN(P) = IN(D) in the
network, 1 j f +1. If we denote by P
r
(j) the probability
that D successfully receives P in the time slot, then we have
P
r
(j) = p
1
+
(j 1)(m
2
9)
n2
2
2
(n
2
3n + 2)m
2n2

n3

k=0
_
n 1
k + 2
_
1 (
k
k+1
p
nc
)

1
k
k+1
p
nc

9
k+2
8
k+2
(m
2
9)
k
(5)
The proofs of Lemmas 2 and 3 can be found in Appendix A.
Remark 2: Our 2HR-(, f) routing algorithm covers the
routing scheme in [9] as special cases. By setting = 1,
we have P
d
(j) =
nj1
2(n2)
p
2
and P
r
(j) = p
1
+
j1
2(n2)
p
2
,
which reduce to the results derived in [9].
Actually for a tagged ow under the 2HR-(, f) routing
algorithm, the packet dispatching process at the source S
and the packet receiving process at the destination D can be
modeled by the nite-state absorbing Markov chain technique.
In the following, we take the packet dispatching process at S
as an example to justify why it can be modeled by the nite-
state absorbing Markov chain technique. The argument for the
4176 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 11, NO. 11, NOVEMBER 2012
packet receiving process at D easily follows in a similar way.
Without loss of generality, consider a general packet say P at
S.
First, we show that the dispatching process of P at S is a
nite-state Markov chain. It is easy to see that for the 2HR-
(, f) scheme, the dispatching process of P consists of a set
of states, I = h
1
, h
2
, . . . , h
f
, h
f+1
, where state h
x
denotes
that there are x copies of P in the network, 1 x f,
and state h
f+1
denotes that S either has delivered P to D or
has distributed P to f distinct relay nodes. The dispatching
process starts from state h
1
and it moves successively from
one state to another. Furthermore, if we assume that the
dispatching process is in state H(t) at time slot t, H(t) I,
then we have
Pr
_
H(t + 1) = h
x
[H(1), H(2), . . . , H(t)
_
= Pr
_
H(t + 1) = h
x
[H(t)
_
, x [1, f + 1]
Together with the fact that the set I has a limited number of
states (f +1 in total), therefore, the dispatching process of P
at S can be modeled by a nite-state Markov chain.
Now we proceed to show that the above Markov chain
is actually an absorbing Markov chain. According to the
absorbing Markov chain theory [36], [37], we only need to
show that the above Markov chain satises the following
two conditions: 1) there is at least one absorbing state; 2)
it is possible to go from any state to at least one absorbing
state in a nite number of steps. It is easy to see that the
state h
f+1
is an absorbing state, since once S has delivered
P to D or has distributed P to f distinct relay nodes S
nishes the copy dispatching process for P. Furthermore, it is
possible for the chain to go from any state h
x
to the absorbing
state h
f+1
in a single step, x [1, f], since S may directly
deliver P to D via the source-to-destination transmission with
probability p
1
during each time slot (as proved in Lemma 1).
Therefore, we prove that when operating under the 2HR-
(, f) scheme, the packet dispatching process at S satises
the criteria of mathematical denition of nite-state absorbing
Markov chain.
B. Service Times at the Source S and the Destination D
When operating under the 2HR-(, f) routing algorithm,
for the tagged ow each packet will experience two queuing
processes, i.e., the packet dispatching process at the local-
queue of the source node S and the packet receiving process
at the virtual queue of the destination node D [9], [11]. Since
the virtual queue stores the sequence numbers of those packets
not received yet by D, the head-of-line entry of the virtual
queue always equals the indicator number at D, i.e., IN(D).
Before proceeding to derive the service times at S and D, we
rst formally introduce the following denition.
Denition 1: For a general packet P of the tagged ow, its
service time at the source S is dened as the time elapsed
between the time slot when S starts to deliver copies for P
(i.e., when S moves P into the head-of-line at the local-queue)
and the time slot when S stops distributing copies for P;
the service time of P at the destination D is dened as the
time elapsed between the time slot when the indicator number
IN(D) is updated to SN(P) (i.e., when the entry SN(P) is
(a) Absorbing Markov chain for the packet dispatching process at
the source node S.
(b) Absorbing Markov chain for the packet receiving process at the
destination node D.
Fig. 3. Absorbing Markov chain for a packet P of the tagged ow, given
that there are already k copies of P in the network when the entry SN(P)
is moved into the head-of-line at the virtual queue. For each transient state,
the transition back to itself is not shown for simplicity.
moved into the head-of-line at the virtual queue) and the time
slot when D receives P.
For a time slot and a general packet P of the tagged ow,
suppose that there are already k copies of P (including the
original one at the source node S) in the network when the
entry SN(P) is moved into the head-of-line at the virtual
queue, 1 k f + 1. If we denote by P
s
(k) the probability
of simultaneous source-to-relay transmission (from S to some
node without P) and relay-to-destination transmission (from
some relay carrying P to D), we can see that for the packet
P, the dispatching process at S and the receiving process at D
can be modeled by two nite-state absorbing Markov chains
shown in Fig. 3a and 3b, respectively, where the absorbing
state A denotes the termination of the service process.
Given there are already k copies of P in the network when
the entry SN(P) is moved into the head-of-line at the virtual
queue (i.e., D receives the last packet before P), if we denote
by X
S
(k) the service time at S and denote by X
D
(k) the
service time at D, it is easy to see that X
S
(k) (resp. X
D
(k))
corresponds to the time it takes the Markov chain in Fig. 3a
(resp. in Fig. 3b) to become absorbed given that the chain
starts from state 1 (resp. state k).
Lemma 4: In a MANET with 2HR-(, f) routing algo-
rithm, for a general packet P of the tagged ow, suppose
that there are already k copies of P in the network when
the indicator number IN(D) is updated to SN(P), then the
expected service times EX
S
(k) and EX
D
(k) can be
determined as
EX
S
(k) =

k1
i=1
1
P
d
(i)
+
1
p1+P
d
(k)

_
1 +

fk
j=1

1
(k, j)
_
if 1 k f,

f
i=1
1
P
d
(i)
if k = f + 1.
(6)
LIU et al.: CAPACITY AND DELAY OF PROBING-BASED TWO-HOP RELAY IN MANETS 4177
EX
D
(k) =

1
Pr(k)+P
d
(k)Ps(k)
_
1 +

fk
j=1

2
(k, j)
+
P
d
(f)Ps(f)
Pr(f+1)

2
(k, f k)
_
if 1 k f 1,
1
Pr(f)+P
d
(f)Ps(f)
_
1 +
P
d
(f)Ps(f)
Pr(f+1)
_
if k = f,
1
Pr(f+1)
if k = f + 1.
(7)
where

1
(k, j) =
j

t=1
P
d
(k +t 1)
p
1
+P
d
(k +t)

2
(k, j) =
j

t=1
P
d
(k +t 1) P
s
(k +t 1)
P
r
(k +t) +P
d
(k +t) P
s
(k +t)
Proof: The derivations of expected service times
EX
S
(k) and EX
D
(k) under the probing-based 2HR-
(, f) routing algorithm are similar to that in [9], and please
refer to [9] for details.
Lemma 5: Regarding the expected service times EX
S
(k)
and EX
D
(k), we have
EX
S
(k + 1) > EX
S
(k), 1 k f (8)
EX
D
(k + 1) < EX
D
(k), 1 k f (9)
Proof: As the proof of (8) is similar to that in [9], we
omit it here. Before proceeding to prove (9), we rst employ
the mathematical induction to prove the following inequality
P
r
(k) EX
D
(k + 1) < 1, 1 k f (10)
which will be used for the proof of (9).
Initial step: for k = f, it is easy to see that
P
r
(f) EX
D
(f + 1) =
P
r
(f)
P
r
(f + 1)
< 1 (11)
Inductive step: we assume that (10) holds for some k = t,
1 < t f, i.e., P
r
(t) EX
D
(t + 1) < 1. We need to prove
(10) still holds for k = t 1.
P
r
(t 1) EX
D
(t)
= P
r
(t 1)
1 + (P
d
(t) P
s
(t)) EX
D
(t + 1)
P
r
(t) +P
d
(t) P
s
(t)
<
P
r
(t 1) +P
r
(t)EX
D
(t + 1)(P
d
(t) P
s
(t))
P
r
(t) +P
d
(t) P
s
(t)
(12)
<
P
r
(t 1) +P
d
(t) P
s
(t)
P
r
(t) +P
d
(t) P
s
(t)
< 1 (13)
where (12) follows after P
r
(t 1) < P
r
(t). Combining (11)
and (13), we prove (10).
Now we are ready to prove (9). For the case that k = f,
we have
EX
D
(f) EX
D
(f + 1)
=
1 +
P
d
(f)Ps(f)
Pr(f+1)
P
r
(f) +P
d
(f) P
s
(f)

1
P
r
(f + 1)
=
1
P
r
(f + 1)
_
P
r
(f + 1) +P
d
(f) P
s
(f)
P
r
(f) +P
d
(f) P
s
(f)
1
_
> 0
(14)
For the case that 1 k < f, we have
EX
D
(k) EX
D
(k + 1)
=
1 + (P
d
(k) P
s
(k))EX
D
(k + 1)
P
r
(k) +P
d
(k) P
s
(k)
EX
D
(k + 1)
=
1 P
r
(k)EX
D
(k + 1)
P
r
(k) +P
d
(k) P
s
(k)
> 0 (15)
where (15) follows after substituting (10). After combining
(14) and (15), we have (9) and then complete the proof for
Lemma 5.
Lemma 6: For the tagged ow, if we denote by X
S
the
mean service time averaged over all packets locally generated
at the source S and denote by X
D
the mean service time
averaged over all packets received at the destination D, then
we have
EX
S
(1) X
S
EX
S
(f + 1) (16)
EX
D
(f + 1) X
D
EX
D
(1) (17)
Proof: From the denitions of X
S
and X
D
, it is easy
to see that (16) and (17) follow directly after (8) and (9),
respectively.
C. Throughput Capacity of 2HR-(, f)
Before deriving the throughput capacity of the proposed
2HR-(, f) routing algorithm, we rst introduce the following
lemma.
Lemma 7: For the 2HR-(, f) routing algorithm, 1 f
n 2, 1
0
, we have
EX
S
(f + 1) EX
D
(f + 1) (18)
where
0
is given by

0
=
(n f 1)p
2
2(n 2)p
1
f
p
2
f
2
| (19)
Proof: It is easy to observe from (4) that
P
d
(j)
n j 1
2(n 2)
p
2
(20)
From (6), we can see that
EX
S
(f + 1) =
f

j=1
1
P
d
(j)

f

j=1
1
nj1
2(n2)
p
2
(21)
=
2(n 2)
p
2
f

j=1
1
n j 1
<
2(n 2)
p
2

f
n f 1
(22)
where (21) follows after substituting (20).
Similarly from (5), we have
P
r
(j) p
1
+
j 1
2(n 2)
p
2
(23)
Combining (23) and (7), then we have
EX
D
(f + 1) =
1
P
r
(f + 1)

1
p
1
+f
p2
2(n2)
(24)
4178 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 11, NO. 11, NOVEMBER 2012
From (24) and (22), we can see that (18) holds if
1
p
1
+f
p2
2(n2)

2(n 2)
p
2

f
n f 1
(25)
after some basic algebraic operations, (19) follows directly
after (25). Then we nish the proof for Lemma 7.
Theorem 1: In a cell partitioned MANET where nodes
move according to the i.i.d. mobility model and the 2HR-
(, f) is adopted for packet routing, if we denote by the per-
node throughput capacity, i.e., the network can stably support
any trafc input rate ( < ), then for any given f and ,
1 f n2, 1
0
, the per node throughput capacity
can be determined as
= p
1
+
f (m
2
9)
n2
2
2
(n
2
3n + 2)m
2n2
n3

k=0
_
n 1
k + 2
_

9
k+2
8
k+2
(m
2
9)
k

(k + 1)

(n 2)

(n 2 f)

(n 2)
1
(k + 1)
1
(n 2 +kf)
(26)
Proof: From Lemmas 5, 6, 7 and Theorem 1 in [9], we
can see that for any given f and , 1 f n2, 1
0
,
the per node throughput capacity is determined as
=
1
EX
D
(f + 1)
= P
r
(f + 1) (27)
combining with (5), we can see that in order to derive the
throughput capacity , the only remaining issue is to derive
the probability p
nc
.
Notice that according to Theorem 1 in [11], for a general
packet P of the tagged ow, as the trafc input rate approaches
the throughput capacity, i.e., , the destination D receives
the last packet before P (i.e., the indicator number IN(D) is
updated to SN(P)) only after the source node S has already
distributed out all f copies for P. If we denote by P
req
(j)
the probability that there are already j copies of P when D
receives the last packet before P, 1 j f + 1, then we
have
lim

P
req
(f + 1) = 1 (28)
For a time slot, suppose some node R which carries a packet
P with SN(P) = IN(D) in its relay-queue specied for
D, decides to conduct the relay-to-destination transmission.
Without loss of generality, we assume D is selected as the
receiver in the t
th
round of probing, 1 t , and denote
by V
i
the node selected in the i
th
round of probing, 1 i < t,
V
i
,= D. From (28), it is easy to see that the probability that
R does not carry any packet P

with SN(P

) = IN(V
i
)
(1 i < t) in the relay-queue for V
i
, i.e., the probability p
nc
,
can be given by
p
nc
=
n 2 f
n 2
(29)
together with (27) and (5), it follows (26). Then we complete
the proof for Theorem 1.
Lemma 8: For a MANET with the 2HR-(, f) routing
algorithm (1 f n 2, 1
0
), the maximum
per node throughput capacity

is achieved at =
0
.
Proof: From (5) and (27), it is easy to see that when
[1,
0
], monotonically increases with . Then it follows
Lemma 8.
D. Expected End-to-End Delay of 2HR-(, f)
With the help of above theoretical frameworks, we proceed
to analytically derive the expected end-to-end packet delay in
MANETs with the 2HR-(, f) routing algorithm.
Denition 2: For a general packet P of the tagged ow, its
end-to-end delay is dened as the time elapsed between the
time slot when P is locally generated at the source node S
and the time slot when P is received by the destination node
D. The expected end-to-end packet delay is averaged over all
packets received at the destination D in the long run.
If we denote by T
e
the end-to-end delay of packet P
at the tagged ow, since the end-to-end delay T
e
consists
of two parts, i.e., the queueing delay at the local-queue of
the source node S and the packet delivery delay [12], then
the expected end-to-end delay ET
e
can be given by the
following theorem.
Theorem 2: In a cell partitioned MANET where nodes
move according to the i.i.d. mobility model and the 2HR-(, f)
is adopted for packet routing, 1 f n 2, 1
0
,
if the trafc ow locally generated at each source node is
a Poisson stream with average input rate (packets/slot)
( < ), then the expected end-to-end packet delay ET
e

can be determined as
ET
e
=
EX
D
(f + 1)
1
(30)
where is the system load and = /.
Proof: One can easily observe from Lemma 7 that under
the given settings of f and (1 f n2, 1
0
), we
always have EX
S
(f + 1) EX
D
(f + 1). According to
Theorem 1 in [29], for a general packet P, the queueing delay
and the delivery delay can be determined as 0 and
E{XD(f+1)}
1
,
respectively. Summing up these two parts, it then follows (30).
IV. NUMERICAL RESULTS
In this section, we rst provide simulation results to verify
the theoretical models for the per node throughput capacity
and the expected end-to-end packet delay, then proceed to
explore the maximum per node throughput capacity and cor-
responding setting of .
A. Simulation Settings
A dedicated C++ simulator was developed to simulate the
packet delivery process of the proposed 2HR-(, f) routing al-
gorithm, which is now available on-line at [38]. Similar to [39]
the guard factor was xed as = 1. The trafc ow locally
generated at each source node was assumed to be a Poisson
stream with average input rate (packets/slot). Besides the
i.i.d. mobility model, we also implemented the random walk
and random waypoint mobility models to simulate the node
movement in a MANET [13], [40].
B. Theoretical Model Validation
Extensive simulations were conducted to verify our the-
oretical models. Here we presented the simulation results
of two network scenarios, with parameter settings listed in
LIU et al.: CAPACITY AND DELAY OF PROBING-BASED TWO-HOP RELAY IN MANETS 4179
TABLE I
PARAMETER SETTINGS FOR MODEL VALIDATION
Simulation parameter Scenario 1 Scenario 2
Number of users n 100 300
Cell partition mm 8 8 16 16
Probing round limit 2 3
Packet redundancy limit f 2 5
Throughput capacity 1.21 10
3
5.07 10
4
(a) Network scenario (n = 100, m = 8, = 2, f = 2) with
per node throughput capacity = 1.2110
3
(packets/slot).
(b) Network scenario (n = 300, m = 16, = 3,
f = 5) with per node throughput capacity = 5.07
10
4
(packets/slot).
Fig. 4. Comparisons between simulation results and the theoretical ones
for model validation of the per node throughput capacity and the expected
end-to-end delay.
Table I. The simulation results of other scenarios can also be
obtained by our simulator [38]. For the two scenarios listed in
Table I, the comparisons between the simulation results and
theoretical ones are summarized in Figs. 4 and 5. Notice that
all the simulation results of the expected end-to-end delay are
reported with the 95% condence intervals.
Figs. 4 and 5 indicate clearly that our throughput capacity
model could nicely capture the throughput capacity behavior
of MANETs with the 2HR-(, f) routing algorithm. Speci-
cally, one can easily observe from Figs. 4a and 4b that, the
simulated expected end-to-end delay there gradually increases
as the system load increases, and becomes extremely sensi-
tive to the variations of as approaches 1. Such skyrocketing
behavior of expected end-to-end delay can also serve as an
intuitive validation for the throughput capacity derived by
our theoretical model. Recall that P
req
(f + 1) denotes the
probability that there are already f + 1 copies of a packet
(a) Probability Preq(f +1) under network scenario (n = 100,
m = 8, = 2, f = 2).
(b) Probability Preq(f+1) under network scenario (n = 300,
m = 16, = 3, f = 5).
Fig. 5. Probability Preq(f + 1) vs. system load .
P in the network when its destination node receives the last
packet before it. Figs. 5a and 5b show clearly that as
approaches 1, i.e., , we have P
req
(f + 1) 1, which
veries (28) and in turn validates the throughput capacity
results derived in Theorem 1. Regarding the expected end-to-
end delay, one can also observe from Figs. 4a and 4b that for
both network scenarios there, the theoretical expected end-to-
end delay matches nicely with the simulated ones. Thus, our
theoretical models can be used to efciently characterize the
per node throughput capacity and expected end-to-end delay
under the 2HR-(, f) routing algorithm.
It is interesting to observe from Figs. 4 and 5 that for the
two network scenarios there, the simulated expected end-to-
end delay and P
req
(f +1) of the 2HR-(, f) under the random
walk and random waypoint mobility models exhibit very
similar behaviors with that under the i.i.d. mobility model.
Therefore, our theoretical models, although developed under
the i.i.d. model, can also be used to nicely capture the network
throughput capacity and expected end-to-end delay behaviors
under the random walk and random waypoint mobility models.
C. 2HR-(, f) Throughput Capacity Analysis
Based on the theoretical model for per node throughput
capacity, we rst examine the impact of probing round limit
on the per node throughput capacity under the 2HR-(, f)
routing algorithm. For the network scenario (n = 300, m =
4180 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 11, NO. 11, NOVEMBER 2012
Fig. 6. Per node throughput capacity vs. probing round limit .
(a) The maximum throughput capacity

vs. the number of

users n.
(b) The optimum setting of vs. the number of users n.
Fig. 7. The maximum throughput capacity

and the corresponding

optimum setting of for networks with m = 16 and n varying from 150 to
1000.
16), Fig. 6 shows clearly that for each setting of f there, the
per node throughput capacity can be signicantly improved
by enabling the multiple probing technique. For example, for
the setting f = 4, the throughput capacity of = 5 (resp.
= 10) is 5.44 10
4
(resp. 6.70 10
4
) (packets/slot),
which is almost 1.47 (resp. 1.81) times that of = 1 (3.70
10
4
(packets/slot)).
We now proceed to explore how the maximum throughput
capacity

and the corresponding optimum setting of , i.e.,

0
, vary with the number of users n. With m = 16 and
f = 5, 6, 7, we summarize the corresponding results in
Fig. 8. The maximum throughput capacity

vs. packet redundancy limit

f.
Fig. 7. One can easily observe from Fig. 7a that for each f
setting there, the maximum throughput capacity

vanishes
quickly as the number of users n (the node density n/m
2
)
increases. It is also noticed that in Fig. 7a there exists a
throughput capacity jump between some settings of n. This
can be explained as follows. As shown in Fig. 7b, the optimum
probing round limit
0
monotonically increases as n and is
actually a piecewise function of n, i.e., a specic
0
value
only applies to a small range of n. It is easy to see that
in such a small range the maximum throughput capacity

monotonically decreases as n increases; while as n increases

beyond such range, a bigger
0
value is adopted and thus a
higher

is achieved. Thus, the throughput capacity jump

only happens between the n values that are actually around
the border line of such
0
ranges. A further careful observation
of Fig. 7a shows that such jump behavior is obvious only
when n is relatively small and becomes almost negligible as
n increases up.
Fig. 8 illustrates the relationship between the packet redun-
dancy limit f and the maximum throughput capacity

. We
can see from Fig. 8 that for each n setting there, as f increases
the

always rst increases and then decreases, and there

exists an optimum setting of f such that a maximum

is
achieved. For example, for the case n = 450, 600 and 900, a
maximum

of 5.71 10
4
, 4.72 10
4
and 3.41 10
4
(packets/slot) are achieved at f = 4, respectively.
V. AVAILABLE THROUGHPUT CAPACITY AND DELAY
ANALYSIS
A signicant amount of works has been done to analyze
the throughput capacity and delay performances of the two-
hop relay algorithm and its variants.
A. Throughput Capacity Analysis
Since the seminal work of Grossglauser and Tse [8], a lot
of order sense results have been reported for the throughput
capacity of two-hop relay. It was proved that a (1) per node
throughput can be achieved under different mobility models,
such as the i.i.d. model [8], [18], brownian mobility model
[14], random walk model [13] and the restricted mobility
model [15]. Ciullo et al. in [31] showed that when n nodes
are partitioned into m clusters and each cluster-region has
LIU et al.: CAPACITY AND DELAY OF PROBING-BASED TWO-HOP RELAY IN MANETS 4181
a circular shape of radius R, the (1) throughput is also
achievable in the cluster dense regime, while the throughput
becomes (mR
2
/n) in the cluster sparse regime. Later, Li
et al. in [35] proved that the per node throughput capacity
is upper bounded by O(n
1/2
) when the network area is
evenly divided into n
2
cells and each cell is further evenly
divided into squares of area n
2
. Recently, Perevalov and
Blum examined the delay-limited throughput in [19] where
it was showed that the achievable throughput grows as d
2/3
for moderate delay constraint d and scales as (n
1/3
) for a
xed delay constraint.
Recently, closed-form expressions have also been derived
for the throughput capacity in MANETs. Neely and Modiano
in [26] showed that in a cell partitioned MANET with xed
user density, the per node throughput capacity tends to a xed
value as the number of users scales up. Later, Urgaonkar et al.
in [16] derived the exact network capacity and the minimum
time-average power required to support it in a delay-tolerant
mobile ad hoc network with Markovian mobility. The exact per
node throughput capacity has also been examined in [9], [11]
where a general two-hop relay with limited packet redundancy
and transmission power control was considered.
B. Delay Analysis
The scaling law of packet delay under the two-hop relay
algorithm has been intensively studied under different mobility
models. Gamal et al. in [13] reported that the delay scales
as (nlog n) under the random walk mobility model, which
was later proved to also hold under the restricted mobility
model [15]. Later, Gamal et al. in [12] showed that under
the brownian motion, the delay scales as (n
1/2
/(n)) where
(n) is the velocity of mobile nodes. Lin et al. also considered
the brownian mobility model [14], and showed that the delay
is lower bounded by (log n/
2
n
), where
2
n
is the variance
parameter of the brownian motion. Sharma et al. in [20]
showed that when the network is divided into n

cells,
the two-hop delay is (n) for < 1/2 and (nlog n) for
= 1/2 under a family of mobility models. It was also
proved that the delay scales as O(

n) with exact

n packet
redundancy under the i.i.d. mobility model [26], and scales
as (T
p
(n)

n/k(n)) under the random waypoint mobility

model [27], where T
p
(n) is the packet transmission time and
k(n) is the redundancy limit.
Some closed-form results were also reported for the delay
performance of two-hop relay. Groenevelt et al. showed that
the node inter-meeting times were exponentially distributed in
a sparsely distributed MANET and developed a Markov chain
model for the packet delivery delay [21]. Following the same
line, Hanbali et al. established closed-form expressions for the
delivery delay of two-hop relay with limited packet redun-
dancy [22], [23]. Later, Panagakis et al. in [24] analytically
examined the message delivery delay where approximation
expressions were provided for the cumulative distribution
function of message delivery delay. More recently, Liu et al.
derived closed-form expressions to upper bound the expected
end-to-end delay [9], and characterized the exact expected
end-to-end delay in MANETs with generalized transmission
range and limited packet redundancy [29].
VI. CONCLUSION
This paper proposed a general 2HR-(, f) routing algorithm
for efcient utilization of wireless resources in MANETs. A
Markov chain theoretical framework was further developed to
model the performance of the new relay algorithm, based on
which closed-form expressions were derived for the per-node
throughput capacity and expected end-to-end delay. Extensive
simulation and theoretical studies indicate that the theoretical
framework is very efcient in performance modeling for
the 2HR-(, f) algorithm, and the new relay algorithm can
signicantly improve the per node throughput capacity by
enabling more rounds of receiver probing. It is interesting to
notice that our theoretical models for throughput capacity and
expected end-to-end delay, although were developed under the
i.i.d. mobility model, can also be used to nicely capture the
network behaviors under the random walk and the random
waypoint models as well.
It is noticed that in the proposed 2HR-(, f) algorithm,
we considered a very simple scenario where each node is
assumed to have innite buffer space. Therefore, one of
our future works is to further explore per node throughput
capacity and average delay of the 2HR-(, f) in a more general
scenario where each node has limited buffer size. Since the
theoretical framework and closed-form results developed in
this paper hold only for the two-hop relay routing, another
future research direction is to extend the theoretical framework
in this paper to analyze the throughput and delay performance
for the general k-hop relay routing, k 3.
APPENDIX A
PROOF OF THE LEMMAS 2 AND 3
Proof of Lemma 2: For a time slot and the tagged ow,
given that there are already j copies of packet P
h
, the event
that S will successfully deliver out a new copy for P
h
occurs
if and only if the following mutually exclusive sub-events
happen simultaneously: the source node S is in an active cell;
the destination node D is not in the one-hop neighborhood of
S; s nodes out of the nj1 relays without carrying copies of
P
h
are in the one-hop neighborhood of S, 1 s nj 1;
t nodes out of the j 1 relays carrying copies of P
h
are
in the one-hop neighborhood of S, 0 t j 1; S is
selected as the transmitter; S decides to conduct the source-
to-relay transmission; a relay node without carrying a copy of
P
h
is selected as the receiver before rounds of probing are
conducted. Then we have
P
d
(j)
=
1
2
2
_
1
9
m
2
_
_
j1

t=0
nj1

s=1
t+s

k=0
_
j 1
t
__
n j 1
s
_
_
t +s
k
__
1
m
2
_
k
_
8
m
2
_
t+sk
_
1
9
m
2
_
n2(t+s)

1
k + 1

s
t +s
_
1 +
_
t
t +s
_
+ +
_
t
t +s
_
1
_
_
(31)
After some basic algebraic operations (4) follows and we
nish the proof for Lemma 2.
4182 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 11, NO. 11, NOVEMBER 2012
Proof of Lemma 3: As the destination node D may either
receive P from the source node S or receive P from some
relay node, if we denote by p
rd
the probability that D will
receive P from some specic relay, say R, then we have
P
r
(j) = p
1
+ (j 1) p
rd
(32)
Now we proceed to derive p
rd
. Notice that D will receive
P from R if and only if the following mutually exclusive
sub-events happen simultaneously: R is in an active cell; the
destination node of R is not in the one-hop neighborhood
of R; D is in the same cell with R or in one of its eight
neighboring cells; there are k nodes other than D locating in
one-hop neighborhood of R, 0 k n 3; R is selected as
the transmitter; R decides to conduct the relay-to-destination
transmission; D is selected as the receiver in the t
th
round of
probing, 1 t ; for the node V
i
selected in the i
th
round
of probing, 1 i < t, R does not carry any packet P

with
SN(P

) = IN(V
i
) in its relay-queue specied for V
i
. Then
we have
p
rd
=
1
2
2
_
1
9
m
2
_
_
n3

k=0
k

i=0
_
n 3
k
__
k
i
__
1
m
2
_
i

_
8
m
2
_
ki
_
1
9
m
2
_
n3k
_
1
m
2
_
1
i + 2

1
k + 1
_
1 +
k
k + 1
p
nc
+ +
_
k
k + 1
p
nc
_
1
_
+
n3

k=0
k

i=0
_
n 3
k
__
k
i
__
1
m
2
_
i

_
8
m
2
_
ki
_
1
9
m
2
_
n3k
_
8
m
2
_
1
i + 1

1
k + 1
_
1 +
k
k + 1
p
nc
+ +
_
k
k + 1
p
nc
_
1
_
_
(33)
=
(m
2
9)
n2
2
2
(n
2
3n + 2)m
2n2

n3

k=0
_
n 1
k + 2
_
1 (
k
k+1
p
nc
)

1
k
k+1
p
nc

9
k+2
8
k+2
(m
2
9)
k
(34)
After substituting (34) into (32) it follows (5), and then we
complete the proof for Lemma 3.
REFERENCES
[1] T. Spyropoulos, K. Psounis, and C. S. Raghavendra, Efcient routing
in intermittently connected mobile networks: the multiple-copy case,
IEEE/ACM Trans. Netw., vol. 16, no. 1, pp. 7790, Feb. 2008.
[2] Y. Ma and A. Jamalipour, A cooperative cache-based content delivery
framework for intermittently connected mobile ad hoc networks, IEEE
Trans. Wireless Commun., vol. 9, no. 1, pp. 366373, Jan. 2010.
[3] , Optimized message delivery framework using fuzzy logic for
intermittently connected mobile ad hoc networks, J. Wireless Commun.
Mobile Comput., vol. 9, no. 4, pp. 501512, Apr. 2009.
[4] Z. Yang, L. Cai, Y. Luo, and J. Pan, Topology-aware modulation and
error-correction coding for cooperative networks, IEEE J. Sel. Areas
Commun., vol. 30, no. 2, pp. 379387, Feb. 2012.
[5] M. Cello, G. Gnecco, M. Marchese, and M. Sanguineti, A model of
buffer occupancy for ICNs, IEEE Commun. Lett., vol. 16, no. 6, pp.
862865, June 2012.
[6] D. B. Johnson and D. A. Maltz, Dynamic source routing in ad hoc
wireless networks, in Mobile Comput., 1996.
[7] C. E. Perkins and E. M. Royer, Ad-hoc on-demand distance vector
routing, in 1999 WMCSA.
[8] M. Grossglauser and D. N. Tse, Mobility increases the capacity of ad
hoc wireless networks, in 2001 INFOCOM.
[9] J. Liu, X. Jiang, H. Nishiyama, and N. Kato, Delay and capacity in ad
hoc mobile networks with f -cast relay algorithms, IEEE Trans. Wireless
Commun., vol. 10, no. 8, pp. 27382751, Aug. 2011.
[10] , Generalized two-hop relay for exible delay control in
MANETs, IEEE/ACM Trans. Netw., 2012, to appear.
[11] , Exact throughput capacity under power control in mobile ad hoc
networks, in 2012 INFOCOM.
[12] A. E. Gamal, J. Mammen, B. Prabhakar, and D. Shah, Throughput-
delay trade-off in wireless networks, in 2004 INFOCOM.
[13] , Optimal throughput-delay scaling in wireless networkspart I:
the uid model, IEEE Trans. Inf. Theory, vol. 52, no. 6, pp. 25682592,
June 2006.
[14] X. Lin, G. Sharma, R. R. Mazumdar, and N. B. Shroff, Degenerate
delay-capacity tradeoffs in ad hoc networks with Brownian mobility,
IEEE Trans. Inf. Theory, vol. 52, no. 6, pp. 27772784, June 2006.
[15] J. Mammen and D. Shah, Throughput and delay in random wireless
networks with restricted mobility, IEEE Trans. Inf. Theory, vol. 53,
no. 3, pp. 11081116, Mar. 2007.
[16] R. Urgaonkar and M. J. Neely, Network capacity region and minimum
energy function for a delay-tolerant mobile ad hoc network, IEEE/ACM
Trans. Netw., vol. 19, no. 4, pp. 11371150, Aug. 2011.
[17] X. Wang, W. Huang, S. Wang, J. Zhang, and C. Hu, Delay and capacity
tradeoff analysis for motioncast, IEEE/ACM Trans. Netw., vol. 19, no. 5,
pp. 13541367, Oct. 2011.
[18] R. M. de Moraes, H. R. Sadjadpour, and J. Garcia-Luna-Aceves, Taking
full advantage of multiuser diversity in mobile ad hoc networks, IEEE
Trans. Commun., vol. 55, no. 6, pp. 12021211, June 2007.
[19] E. Perevalov and R. Blum, Delay-limited throughput of ad hoc net-
works, IEEE Trans. Commun., vol. 52, no. 11, pp. 19571968, Nov.
2004.
[20] G. Sharma, R. Mazumdar, and N. B. Shroff, Delay and capacity trade-
offs for mobile ad hoc networks: a global perspective, IEEE/ACM
Trans. Netw., vol. 15, no. 5, pp. 981992, Oct. 2007.
[21] R. Groenevelt, P. Nain, and G. Koole, The message delay in mobile ad
hoc networks, Performance Evaluation, vol. 62, no. 1-4, pp. 210228,
Oct. 2005.
[22] A. A. Hanbali, A. A. Kherani, and P. Nain, Simple models for the
performance evaluation of a class of two-hop relay protocols, in 2007
IFIP Netw.
[23] A. A. Hanbali, P. Nain, and E. Altman, Performance of ad hoc
networks with two hop relay routing and limited packet lifetime, in
2006 Valuetools.
[24] A. Panagakis, A. Vaios, and I. Stavrakakis, Study of two-hop message
spreading in DTNs, in 2007 Wiopt.
[25] T. Small and Z. Hass, Resource and performance tradeoffs in delay-
tolerant wireless networks, in 2005 WDTN.
[26] M. J. Neely and E. Modiano, Capacity and delay tradeoffs for ad-hoc
mobile networks, IEEE Trans. Inf. Theory, vol. 51, no. 6, pp. 1917
1936, June 2005.
[27] G. Sharma and R. Mazumdar, On achievable delay/capacity trade-offs
in mobile ad hoc networks, in 2004 Wiopt.
[28] , Delay and capacity trade-off in wireless ad hoc networks with
random mobility, Dept. Elect. Comput. Eng., Purdue Univ., West
Lafayette, IN 2005. Available: https://ece.uwaterloo.ca/

mazum/adhoc
mobility.pdf
[29] J. Liu, X. Jiang, H. Nishiyama, N. Kato, and X. Shen, End-to-end
delay in mobile ad hoc networks with generalized transmission range
and limited packet redundancy, in 2012 WCNC.
[30] J. Liu, X. Jiang, H. Nishiyama, and N. Kato, Group-based two-hop
relay with redundancy in MANETs, in 2011 HPSR.
[31] D. Ciullo, V. Martina, M. Garetto, and E. Leonardi, Impact of correlated
mobility on delay-throughput performance in mobile ad hoc networks,
IEEE/ACM Trans. Netw., vol. 19, no. 6, pp. 17451758, Dec. 2011.
[32] P. Gupta and P. Kumar, The capacity of wireless networks, IEEE
Trans. Inf. Theory, vol. 46, no. 2, pp. 388404, Mar. 2000.
[33] S. R. Kulkarni and P. Viswanath, A deterministic approach to through-
put scaling in wireless networks, IEEE Trans. Inf. Theory, vol. 50,
no. 6, pp. 10411049, June 2004.
[34] C. Zhang, Y. Fang, and X. Zhu, Throughput-delay tradeoffs in large-
scale manets with network coding, in 2009 INFOCOM, 2009.
[35] P. Li, Y. Fang, J. Li, and X. Huang, Smooth trade-offs between
throughput and delay in mobile ad hoc networks, IEEE Trans. Mobile
Comput., vol. 11, no. 3, pp. 427438, Mar. 2012.
LIU et al.: CAPACITY AND DELAY OF PROBING-BASED TWO-HOP RELAY IN MANETS 4183
[36] C. M. Grinstead and J. L. Snell, Introduction to Probability: Second
Revised Edition. American Mathematical Society, 1997.
[37] J. G. Kemeny and J. L. Snell, Finite Markov Chains. D. Van Nostrand,
1963.
[38] C++ simulator for the 2HR-(,f) MANETs. Available: http://distplat.
blogspot.com.
[39] The network simulator ns-2. Available: http://www.isi.edu/nsnam/ns/.
[40] S. Zhou and L. Ying, On delay constrained multicast capacity of large-
scale mobile ad-hoc networks, in 2010 INFOCOM, 2010.
Jiajia Liu received his B.S. and M.S. Degrees
both in Computer Science from Harbin Institute of
Technology in 2004 and from Xidian University in
2009, respectively. He is currently a PhD candidate
at the Graduate School of Information Sciences at
Tohoku University. His research interests include
performance modeling and evaluation, scaling laws
of wireless networks, stochastic network optimiza-
tion, and optimal control.
Juntao Gao received his B.S. and M.S. degrees
both in Computer Science from Xidian University,
Xian, China, in 2008 and 2010, respectively. He
is currently a PhD candidate at the Graduate School
of Systems Information Science at Future University
Hakodate. His research interests are in the areas of
optimal resource allocation, stochastic optimization
and scheduling in wireless networks.
Xiaohong Jiang received his B.S., M.S. and Ph.D.
degrees from Xidian University, Xian, China, in
1989, 1992 and 1999, respectively. Dr. Jiang is
currently a full professor of Future University Hako-
date, Japan. Before joining Future University Hako-
date, he was an associate professor in Tohoku Uni-
versity, Japan, was an assistant professor and JSPS
Post Doctoral Research Fellow in Japan Advanced
Institute of Science and Technology (JAIST). His
current research interests include wireless networks,
optical networks, network coding, etc. Dr. Jiang
has authored and coauthored more than 170 publications in journals and
international conference proceedings, which include IEEE/ACM TRANSAC-
TIONS ON NETWORKING, IEEE TRANSACTIONS ON COMMUNICATIONS,
and IEEE JOURNAL OF SELECTED AREA ON COMMUNICATIONS. Dr. Jiang
was also the winner of the Best Paper Award of WCNC2008 and the ICC2005-
Optical Networking Symposium. He is a senior member of IEEE. (Email:
[email protected])
Hiroki Nishiyama (M08) received his M.S. and
Ph.D. in Information Science from Tohoku Uni-
versity, Japan, in 2007 and 2008, respectively. He
was a Research Fellow of the prestigious Japan
Society for the Promotion of Science (JSPS) until
the completion of his PhD, following which he went
on to become an Assistant Professor at the Graduate
School of Information Sciences (GSIS) at Tohoku
University. He was promoted to his current position
of an Associate Professor at GSIS in 2012, when
he was just 29 years old. He was acclaimed with
the Best Paper Awards in many international conferences including IEEEs
agship events, namely the IEEE Wireless Communications and Networking
Conference in 2012 (WCNC12) and the IEEE Global Communications
Conference in 2010 (GLOBECOM10). He is a young yet already prominent
researcher in his eld as evident from his valuable contributions in terms of
many quality publications in prestigious IEEE journals and conferences. He
was also a recipient of the IEICE Communications Society Academic Encour-
agement Award 2011 and the 2009 FUNAI Foundations Research Incentive
Award for Information Technology. He received the Best Student Award
and Excellent Research Award from Tohoku University for his phenomenal
performance during the undergraduate and master course study, respectively.
His research covers a wide range of areas including trafc engineering,
congestion control, satellite communications, ad hoc and sensor networks, and
network security. He is a member of the Institute of Electronics, Information
and Communication Engineers (IEICE).
Nei Kato (A03-M04-SM05) received his Bachelor
Degree from Polytechnic University, Japan in 1986,
M.S. and Ph.D. Degrees in information engineering
from Tohoku University, Japan, in 1988 and 1991,
respectively. He joined Computer Center of Tohoku
University at 1991, and has been a full profes-
sor with the Graduate School of Information Sci-
ences since 2003. He has been engaged in research
on satellite communications, computer networking,
wireless mobile communications, smart grid, image
processing and neural networks. He has published
more than 300 papers in peer-reviewed journals and conference proceedings.
He currently serves as the Vice Chair of IEEE Ad Hoc & Sensor Networks
Technical Committee, the Chair of IEEE ComSoc Sendai Chapter, a technical
editor of IEEE Wireless Communications (2006-), and editor of IEEE Net-
work Magazine (2012-), an editor of IEEE TRANSACTIONS ON WIRELESS
COMMUNICATIONS (2008-), an associate editor of IEEE TRANSACTIONS ON
VEHICULAR TECHNOLOGY (2010-), an editor of IEEE TRANSACTIONS ON
PARALLEL AND DISTRIBUTED SYSTEMS. He has served as the Chair of IEEE
Satellite and Space Communications Technical Committee(2010-2012), a co-
guest-editor of several Special Issues of IEEE Wireless Communications Mag-
azine, a symposium co-chair of GLOBECOM07, ICC10, ICC11, ICC12,
Vice Chair of IEEE WCNC10, WCNC11, ChinaCom08, ChinaCom09,
Symposia co-chair of GLOBECOM12, and workshop co-chair of VTC2010.
His awards include Minoru Ishida Foundation Research Encouragement
Prize(2003), Distinguished Contributions to Satellite Communications Award
from the IEEE Communications Society, Satellite and Space Communications
Technical Committee(2005), the FUNAI information Science Award(2007),
the TELCOM System Technology Award from Foundation for Electrical
Communications Diffusion(2008), the IEICE Network System Research
Award(2009), the KDDI Foundation Excellent Research Award(2012), IEEE
GLOBECOM Best Paper Award(twice), IEEE WCNC Best Paper Award,
and IEICE Communications Society Best Paper Award(2012). Besides his
academic activities, he also serves on the expert committee of Telecommu-
nications Council, Ministry of Internal Affairs and Communications, and as
the chairperson of ITU-R SG4 and SG7, Japan. Nei Kato is a Distinguished
Lecturer of IEEE Communications Society(2012-213) and the co-PI of A3
Foresight Program(2011-2014) funded by Japan Society for the Promotion of
Sciences(JSPS), NSFC of China, and NRF of Korea. He is a fellow of IEICE.