2556 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 59, NO.

5, JUNE 2010
Cross-Layer Optimized MAC to Support Multihop
QoS Routing for Wireless Sensor Networks
Heping Wang, Xiaobo Zhang, Farid Nat-Abdesselam, Member, IEEE, and Ashfaq Khokhar, Fellow, IEEE
AbstractThis paper presents an efcient hybrid medium-
access control (HMAC) protocol with an embedded cross-layer
optimization solution to provide routing-layer coarse-grained end-
to-end quality-of-service (QoS) support for latency-sensitive trafc
ows. A novel channel-reservation technique is proposed to signif-
icantly reduce the end-to-end delay for delay-sensitive trafc ows
by allowing packets to go through multiple hops within a single
medium-access control (MAC) frame and by also giving them
higher priority channel access to reduce possible queuing delay.
Our proposed protocol (HMAC) combines energy-efcient fea-
tures of the existing contention-based and time-division multiple-
access (TDMA)-based MAC protocols and adopts a short frame
structure to expedite packet delivery. Simulation results in ns-2
show that HMAC achieves signicant performance improvements
in energy consumption, latency, and throughput over existing
MAC protocols.
Index TermsEnd-to-end latency, energy efciency, hybrid
medium-access control (HMAC), latency-sensitive trafc, quality
of service (QoS).
I. INTRODUCTION
N
ETWORKING low-cost smart sensors through a wire-
less medium to form wireless sensor networks (WSNs)
has broad applications, such as surveillance, event detection,
etc. However, due to battery-constrained computationally and
communication-wise less powerful nodes in WSNs, traditional
protocols are not suitable for WSNs in terms of energy ef-
ciency, scalability, and design complexity. Current research
efforts for WSNs mainly focus on exploring energy-efcient
network operations in WSNs.
At the network layer, a multihop routing paradigm based on
short-range communication is popularly adopted by existing
network layer protocols to improve energy efciency [20].
However, multihop routing in wireless environments incurs
increased end-to-end delivery latency and exacerbates potential
hidden/exposed terminal problems [16].
Manuscript received January 26, 2009; revised May 27, 2009 and
September 23, 2009; accepted November 9, 2009. Date of publication
February 17, 2010; date of current version June 16, 2010. This work was
supported in part by the U.S. National Science Foundation under Grant CNS-
0910988. The review of this paper was coordinated by Prof. H. Hassanein.
H. Wang is with Lemko Corporation, Schaumburg, IL 60173 USA (e-mail:
hwang10@uic.edu).
X. Zhang is with CISCO Systems, San Jose, CA 95134 USA (e-mail:
xzhang20@uic.edu).
F. Nat-Abdesselam is with the University of Sciences and Technologies of
Lille, 59655 Villeneuve dAscq, France (e-mail: farid.nait-abdesselam@li.fr).
A. Khokhar is with the Department of Electrical and Computer Engi-
neering, University of Illinois at Chicago, Chicago, IL 60607 USA (e-mail:
ashfaq@uic.edu).
Color versions of one or more of the gures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identier 10.1109/TVT.2010.2042185
Existing medium access control (MAC) layer solutions [1]
[6] for WSNs generally aim at reducing energy waste due
to channel idle listening and overhearing. In most of these
solutions, per-hop latency is compromised in favor of energy
efciency, thus deteriorating end-to-end communication delay.
Without considering the performance requirements at upper
layers, an energy-efcient MAC solution for WSN may not
be optimal from the holistic networking and applications per-
spective, where end-to-end latency is sometimes an important
consideration [7].
In this paper, we present a new MAC protocol, which is re-
ferred to as hybrid MAC (HMAC), which is suitable for WSNs
in terms of energy efciency, latency, and design complexity.
HMAC combines channel-allocation schemes from existing
contention-based and time-division multiple-access (TDMA)-
based MAC protocols to allow the realization of tradeoffs
between different performance metrics. It uses a short slotted
frame structure and a novel wakeup scheme to achieve high-
energy performance, low delivery latency, and improved chan-
nel utilization. Compared with existing TDMA-based MAC
protocols [5], [6], HMAC is simple and scalable since each
node does not have to maintain neighborhood information.
In addition, HMAC provides routing layer coarse-grained
quality-of-service (QoS) support at the MAC layer. To the
best of our knowledge, very few existing MAC layer works
handle such QoS issues in WSNs. Quality of service-aware
medium access control [17] assigns each ow a channel-access
priority to reduce the queuing delay for high-priority ows,
but it still suffers from a long end-to-end delay. The MAC
protocols presented in [12][15] reduce the end-to-end delivery
latency while increasing control overhead without considering
different performance demands between ows. The HMAC
design presents an extremely low-cost solution compared with
the designs proposed in [12][15]. Our simulation results on
ns-2 [9] show that HMAC outperforms sensor-MAC (S-MAC)
and routing-enhanced MAC (RMAC) in terms of per-hop la-
tency and delivery ratio while still maintaining superior energy
performance.
II. RELATED WORK
Different WSN applications have their own performance
demands for the underlying MAC protocols. A signicant body
of the existing research in WSNs have mainly explored energy-
efcient MAC techniques, such as channel duty cycling [1], [2],
low power listening [4], and distributed TDMA-based MAC
solutions [5], [6].
0018-9545/$26.00 2010 IEEE
WANG et al.: CROSS-LAYER OPTIMIZED MAC TO SUPPORT MULTIHOP QOS ROUTING FOR WIRELESS SENSOR NETWORKS 2557
A. Contention-Based MAC Protocols
Contention-based MAC protocols [1][3] basically access
the radio channel based on a duty cycling scheme (also called
periodical listen and sleep) to reduce the energy waste in idle
listening in low trafc networks. S-MAC [2] introduces adap-
tive listening into its original design [1] to cut its end-to-end
latency by half. Timeout-MAC (T-MAC) [3] extends S-MAC
by adapting the channel duty cycle based on the trafc load to
improve the MAC performance. T-MAC also proposed a tech-
nique to partially reduce the early sleep-problem in S-MAC.
To further solve the sleep-delay issue in S-MAC and T-MAC,
data gathering MAC [12] and the work in [14] have proposed
pipelined forwarding to reduce the end-to-end latency. RMAC
[13] is the most recent work based on S-MAC and optimizes the
end-to-end delay. RMAC extends the listen period in S-MAC
so that multiple Pioneer frame packets can be accommodated
to reserve the channel and set up the corresponding data trans-
mission schedules in the sleep period within a single MAC
frame time. Different from S-MAC, Berkeley MAC (B-MAC)
[4] is an asynchronous contention-based MAC protocol based
on preamble sampling. Senders put a preamble in front of
each outgoing packet; a nonsender node periodically wakes
up for a short duration to detect such a preamble for possible
packet transmission. Both the energy efciency and the latency
performance of B-MAC depend on the preamble length.
B. Contention-Free MAC Protocols
In contention-free MAC protocols [5], [6], nodes maintain
TDMA-based contention-free schedules for channel access.
These MAC protocols are naturally energy efcient since
each node is only active within a dedicated portion of time
for contention-free channel access. TRAMA [5] is a trafc-
adaptive TDMA-based MAC protocol. Each node must col-
lect consistent two-hop neighbor information to determine
contention-free schedules. TRAMA exhibits higher energy ef-
ciency and channel utilization than S-MAC. In TDMA-W [6],
each node is assigned two slots: 1) a wake-up slot and 2) a
unique send slot within its two-hop neighborhood. A wakeup
mechanism is proposed to reduce the energy waste due to idle
listening. TDMA-W is highly energy efcient and guarantees
deterministic upper bounded end-to-end delays. TDMA-based
MAC solutions are suitable for high-trafc conditions under
which all nodes share a similar trafc pattern. However, they
will incur high design complexity to maintain a contention-free
schedule, long latency in a large-scale network, and lowchannel
utilization in low-trafc situations.
C. Hybrid MAC Protocols
Hybrid MAC protocols [8], [15] combine different medium-
access techniques to improve network performance. Zebra
MAC (Z-MAC) [8] is built on B-MAC [4] and aims to im-
prove channel utilization and latency under different network
situations to overcome the shortcomings of both carrier sense
multiple access (CSMA) and TDMA schemes. Based on the
trafc load in the network, Z-MAC can act as CSMA in a
low-trafc situation or like TDMA in a high-trafc situation.
Funneling-MAC [15] specically handles the funnel effect
issue in areas with more intensive trafc load and eliminates
collisions that frequently happen in such areas. Funneling-
MAC adopts a TDMA scheduling on nodes in the trafc-
intensive region while allowing nodes far away from the sink to
compete channel access in a CSMA fashion. Similar to TDMA-
based MACprotocols, Hybrid MACsolutions incur high design
complexity.
In conclusion, existing body of research work on MAC
protocols for WSNs is generally focused on energy-efcient
designs at the cost of compromising other network performance
metrics, such as delivery latency. Contention-based solutions
[1][3] have lower design complexity and are preferred for
light-trafc applications compared with TDMA-based solu-
tions [5], [6] and hybrid designs [8], [15]. Some optimiza-
tion techniques have been introduced to improve the latency
performance of existing MAC protocols [12][14], but these
techniques have introduced signicant control overhead. Fur-
thermore, providing QoS support at the MAClayer for multihop
routings has also been neglected in most of the existing works.
III. HYBRID MEDIUM-ACCESS CONTROL
PROTOCOL DESIGN
The fact that sensor networks generally exhibit low trafc
loads most of the time motivates us to develop an HMAC
protocol that exploits the energy-efcient characteristics of low-
duty-cycled contention-based protocols and efcient channel-
utilization characteristics of TDMA-based protocols. HMAC
introduces the concept of a short slotted frame structure in
which slots are dynamically shared. A slotted frame structure
helps to improve energy efciency; a short frame length can
intuitively decrease the queuing delays of packets and thus
reduce the latency and collision when the trafc load increases.
Another motivation comes from the observation that there may
exist some ows that have lower end-to-end delay requirement
than others in some WSN applications. Providing end-to-end
QoS support in an energy-efcient way at the MAC layer is
desirable in such applications.
A. Assumptions
We assume that the nodes are randomly distributed over a
given area, and each node has a unique identier. Each node
uses a short-range radio to communicate with neighbors to
save energy. Thus, nodes separated by long distances induce
a multihop network.
In HMAC, the slotted frame structure requires time synchro-
nization within each nodes two-hop neighborhood. We can
provide some guard time at both ends of each slot to deal
with the short-term CPU clock drift. The provided guard time,
together with the existing synchronization mechanisms [10],
[18] or a cheap GPS-based solution, can meet the synchroniza-
tion requirement in HMAC. We can also adopt the lightweight
and energy-efcient time synchronization technique in [19] or
the global schedule algorithm in [14] to synchronize nodes in
HMAC by using the rst wakeup slot in each HMAC frame
to exchange SYNC messages between nodes. In this paper, we
assume that the nodes in HMAC are synchronized.
2558 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 59, NO. 5, JUNE 2010
Fig. 1. HMAC frame structure: wake-up and data slots.
B. Details of Implementation
Time is organized into nonoverlapping frames, and each
frame contains multiple very short wakeup slots (W_SLOT)
and multiple data slots (D_SLOT), as shown in Fig. 1. The
number of W_SLOTs can be set according to the density of
the network, and the number of D_SLOTs can be set based on
the potential trafc load in the network and end-to-end delay
requirements. Each W_SLOT contains a carrier sensing (CS)
and wakeup message (WAKEUP) period, and each D_SLOT
includes CS and periods for RTS/CTS/DATA/ACK messages.
The reason why we provide the CS periods at the beginning
of both wakeup slots and data slots is that possible concurrent
transmissions could be initiated by multiple senders within a
neighborhood. Under such a scenario, channel access conicts
can happen during both receivers wakeup stage and data-
transmission stage when some receivers share the same wakeup
slot or some senders happen to choose the same data slot.
To reduce such channel access conicts, all senders should
perform CS to check the channel availability after backing off
by some duration. The backoff time is randomly distributed
within a predetermined contention window. In Section III-C, we
further discuss how to provide QoS support at the MAC layer
by properly dividing the CS period of each data slot into two
sections, which provide two classes of channel-access priorities
for two different types of trafc ows.
Each node has its own wakeup slot to listen to the channel
for the possible incoming WAKEUP message. It sleeps in all
the other wakeup slots. Since WSNs generally have light trafc
most of the time, we do not assign each node a data slot a priori.
Whenever a node has data to transmit, it just randomly picks
up a data slot and noties the receiver(s) of the corresponding
slot number in the receivers wakeup slot via a WAKEUP
message. The receiver only wakes up in the corresponding data
slot to receive data. One exception is that when a node receives
more than one WAKEUP message during its wakeup slot (we
call it wakeup collision), the receiver wakes up in all the data
slots for possible data transmission because the receiver cannot
determine the data slot the sender will use or how many senders
will send data. This will mostly likely happen in bottleneck
areas. The well-known RTS/CTS mechanism is used to reserve
channel and avoid the hidden terminal problem if more than one
sender exists within a receivers one-hop neighborhood and all
of them happen to choose the same D_SLOT. Before initiating
an RTS message, each sender needs to perform carrier sense
to avoid potential collision of channel contention with other
senders. The RTS/CTS/DATA/ACK message sequence is fol-
lowed only for unicast data packet. For a broadcast packet, once
a clear channel is determined during CS period at the beginning
of a D_SLOT, a sender sends out its packet immediately.
Each node simply determines its W_SLOT number (i.e.,
the position index of the W_SLOT within an HMAC frame)
as n mod M, where n is the node identier, and M is the
total number of W_SLOTs in a MAC frame. This way, each
node does not need to collect the W_SLOT information of
its one-hop neighbors and, hence, reduces the corresponding
communication overhead, which can be very high in a dense or
dynamic network.
The regular unicast data exchange between nodes can be
performed as follows:
1) Each node turns on its radio during its own wakeup slot
and sleeps during all the other wakeup slots.
2) Each sender randomly picks up a data slot and announces
the data slot number along with the receivers node iden-
tier via a WAKEUP message in the receivers wake-
up slot.
3) Upon reception of a WAKEUP message, a node checks
the embedded node identier in the WAKEUP mes-
sage. If it is the intended receiver, then the node turns
on its radio for the incoming data packet in the specied
data slot; otherwise, it just sleeps. If a broadcast address
is included in the WAKEUP message, then all nodes
receiving this message should wake up in the specied
data slot simultaneously.
4) If any collision occurs in a nodes wakeup slot, then the
node turns on its radio for a duration long enough to
receive an RTS packet at the beginning of each data slot
for a possible incoming data packet. If the node learns that
it is the intended receiver from the received RTS message,
then it keeps the radio on to receive the data packet;
otherwise, it returns to sleep in the remaining period of
the data slot. This way, a node can minimize the extra
energy cost under such a situation.
5) In each data slot, unicast data transmission must fol-
low the well-known RTS/CTS/DATA/ACK scheme in
IEEE802.11 [11] to avoid the hidden terminal problem,
since two senders may choose the same data slot to
send data to their receivers at the same time, and the
transmissions happen to be in a common interference
range.
HMAC also provides support for one-hop broadcast op-
eration. When a node has data to broadcast, it sends out a
WAKEUP message containing a broadcast address and a data
slot in each wakeup slot. After receiving such WAKEUP
messages, all neighbors will wake up in the same data slot to
receive the broadcast message. In addition, different from uni-
cast operation, when the sender nds an idle channel during the
CS period in the chosen data slot, it will immediately send out
the broadcast message without following the aforementioned
RTS/CTS/DATA/ACK scheme.
C. QoS Support at MAC Layer
As previously mentioned, there may exist trafc ows with
different performance requirements in some sensor applica-
tions. Some of them are more latency sensitive (referred to
as TYPE I ows in this paper) than others (referred to as
TYPE II ows in this paper). Through cross-layer optimization,
a MAC layer design can provide coarse-grained QoS support to
routing layer by minimizing the end-to-end latency for TYPE I
WANG et al.: CROSS-LAYER OPTIMIZED MAC TO SUPPORT MULTIHOP QOS ROUTING FOR WIRELESS SENSOR NETWORKS 2559
Fig. 2. Channel reservation mechanism across multiple hops.
ows at the MAC layer. In this section, we propose a channel
reservation scheme combined with a prioritized channel-access
technique to support expedited transmission of TYPE I ows.
In the following, N represents the total number of data slots
within an HMAC frame.
1) Channel Reservation: Let us start with an example to
see how our channel reservation scheme works. Considering
a TYPE I ow path shown in Fig. 2, node S is the source,
and node D is the destination. Within a specic HMAC frame,
for any data packet p, if we can deliver it from source S to
destination D in a pipelined fashion as follows: S sends p to
A in D_SLOT 1, A forwards p to B in D_SLOT 2, and then
B sends p to node D in D_SLOT 3; then, the event packet p
can be delivered from S to D within a single HMAC frame.
However, without channel reservation, HMAC will take three
frames time to deliver p. The channel reservation is performed
on a per-ow basis between any sourcedestination pair.
To support ow-based channel reservation, the source node
either issues a special reservation request before sending its
time-critical packets or piggybacks the reservation request in
its rst event packet. This can easily be done at the upper
layer in the source node by setting a special bit (we refer to it
as REV-bit) in each outgoing packet, which indicates whether
the subsequent packets from the specic sources should be
transmitted in a expedited or regular way. By regular
way, we mean that each packet will be delivered only one
hop closer to the destination within each HMAC frame. The
REV-bit helps to reserve the wireless channel along paths for
any duration needed by the expedited ows. During the
transmission of the rst packet with a REV-bit set to 1, the
channel is reserved along the path from the source toward
the destination as follows: If a node receives a packet with
a set REV-bit in the ith D_SLOT, then it will reserve the
((i + 1) mod N)th D_SLOT and forward the subsequent data
packets from the same source node to its next hop in the ((i +
1) mod N)th D_SLOT, provided the receiving node itself is not
the destination. This process continues until the packets arrive
at the destination. Therefore, for each expedited ow, all re-
lated intermediate nodes locally reserve two adjacent D_SLOTs
during the transmission of fast ows: one for receiving data
(we call it R_D_SLOT) from its previous hop and the other for
sending data (we call it S_D_SLOT) to its immediate next hop.
The reserved channel will be released when a node receives a
packet with a clear REV-bit or a predetermined timer expires,
and then, all related nodes release the reserved slots.
During the reservation period, for each packet belonging to
a particular TYPE I trafc ow, receivers at each hop will not
be woken up by WAKEUP messages issued by their previous
hop for the same ow; each participating node will listen to
the channel in its reserved receiving data slot (R_D_SLOT) for
a duration approximately equal to the lengths of CS and RTS
periods; if no RTS is received, then the node will sleep in the
Fig. 3. Prioritized channel access.
remaining time of R_D_SLOT and S_D_SLOT, provided it is
not woken up for data transmission by any other node.
For a TYPE II packet with a REV-bit set to 0, nodes
forward the packet at each hop using the regular HMAC
channel-reservation procedure described in Section III-B.
2) Prioritized Channel Access: To give TYPE I ows higher
channel access priority over TYPE II ows, the random backoff
time for each type of ows is differently determined during the
CS period at the beginning of each D_SLOT. We divide the
CS period into two periods: 1) T
I
and 2) T
II
. TYPE I ows
compete channel access only within the period T
I
, whereas
TYPE II ows compete channel access only within the period
T
II
(refer to Fig. 3). This way, TYPE I ows can be assigned
higher priority for channel access. If a node nds a free channel
within a reserved D_SLOT during T
I
and it happens to have
data to send, then it can compete with other potential senders
for reusing this D_SLOT for TYPE II ows; otherwise, this
D_SLOT is occupied by TYPE I ows.
IV. AVERAGE END-TO-END LATENCY ANALYSIS
In this section, we analyze the average per-hop latency and
multihop end-to-end latency of HMAC for different trafc
ows. We rst consider the scenario without channel reserva-
tion, which is for TYPE II ows or the reservation packets
for the TYPE I ows. To simplify the analysis of the average
delivery delay a message may experience at each hop, we
assume that no collision occurs. We also assume that the trafc
load of the network is light, and there is no buffering of packets
at the intermediate nodes. For the purpose of analysis, we dene
the following notations:
N number of data slots in an HMAC frame;
M number of wakeup slots in an HMAC frame;
i index of data slot in which the sender chooses to send
data packet;
j index of wakeup slot of the receiver (next hop);
t event/packet arrival instance in an HMAC frame;
T
f
length of an HMAC frame;
T
d
length of a data slot;
T
w
length of a wakeup slot;
p(x) probability density function (pdf) of random variable x;
E[x] expectation of random variable x.
Therefore, we have relations 0 i N 1, 0 j M
1, 0 t T
f
, and T
w
= (T
f
N T
d
)/M (see Fig. 4). We
also dene t
ws
= j T
w
.
We assume uniform distributions for random variables i,
j, and t, and the corresponding pdfs are 1/N, 1/M, 1/T
f
,
respectively.
Case 1) If t < t
ws
, or the packet arrives at the sender before
the left end of the wakeup slot j of the next-hop
node, then it is obvious that the sender can send the
2560 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 59, NO. 5, JUNE 2010
Fig. 4. Average delivery latency analysis.
packet to the receiver in the current frame; therefore,
after the sender chooses a data slot i, the packet can
be delivered in the current frame (not including time
for CS/RTS/CTS) with the following delay:
t
i,j
delay
= (T
f
NT
d
t) +iT
d
. (1)
Case 2) If t > t
ws
, or the packet arrives at a sender node
after the right end of the wakeup slot j of the next-
hop node (in fact the time in wakeup slot j when
the sender could not wake up the receiver), then
the sender will have to send the packet in the next
frame; therefore, after a sender chooses a data slot
i, the packet can be delivered in the following frame
with the following delay:
t
i,j
delay
= (T
f
NT
d
+iT
d
) + (T
f
t). (2)
Combining the foregoing two cases for the given slots i and
j, the delivery latency of a packet can be calculated as follows:
t
i,j
delay
=
t
ws

0
[(T
f
NT
d
t) +iT
d
] p(t)dt
+
T
f

t
ws
[(T
f
NT
d
+iT
d
) + (T
f
t)] p(t)dt
=1.5T
f
NT
d
+iT
d
jT
w
. (3)
Since i and j are randomly selected, the average delivery
delay of a packet can nally be determined as
t
i,j
delay
=E

t
i,j
delay

=1.5T
f
NT
d
+E[i]T
d
E[j]T
w
=1.5T
f
NT
d
+
N 1
2
T
d

M 1
2
T
w
=T
f
0.5T
d
+ 0.5T
w
. (4)
Note that (4) only formulates the average per hop latency
that a packet experiences. Therefore, for an n-hop path between
a pair of source S and destination D, the average end-to-end
latency can be determined as
D
n
(S, D) =n t
i,j
delay
=n(T
f
0.5T
d
+ 0.5T
w
). (5)
Based on the foregoing notations, we now determine the
average end-to-end latency for TYPE I trafc ows. For any
n-hop TYPE I ow between a specic pair of source S and
destination D, ow packets will be delivered in the reserved
slots of all nodes along the path from S to D in a pipelined
fashion. It is obvious that a packet can be N hops closer toward
the destination within a single HMAC frame, where N is the
number of data slots in an HMAC frame. The average end-to-
end latency can simply be divided into two parts:
1) Waiting delay or queuing delay t
q
at the source node S:
At source S, a TYPE I data packet may be generated
at any instance t within an HMAC frame and can only
be delivered to the next hop in the reserved data slot
of S determined by the channel-reservation operation.
Therefore, some queuing delay will be incurred on each
data packet at source S. If t is before the left end of the
reserved data slot of S, and we further assume it to be the
rst data slot, then the average waiting delay is
t
1
q
=
M T
w
2
. (6)
However, if the packet is generated after the reserved data
slot of S, then the average waiting time is
t
2
q
= M T
w
+
N T
d
2
. (7)
If we assume a uniform distribution for the random
variable t, then combining (6) and (7), we have the
expectation for t
q
as
t
q
= t
1
q
M T
w
T
f
+t
2
q
N T
d
T
f
=
T
f
2
. (8)
2) Propagation time t
p
a packet travels from source node S
to destination node D: We also assume that source
node S reserves the rst data slot for its TYPE I packets
each time; therefore, it will take at most n/N full
HMAC frames and n mod N data slot time for a TYPE I
packet to travel from node S to node D. The propagation
time t
p
is
t
p
= (n mod M) T
d
+N T
d

n
N

+ M T
w

n 1
N

. (9)
Therefore, under channel reservation, the average end-to-end
latency for TYPE I packet ows can approximately be deter-
mined as
D
n
(S, D)
R
=t
p
+t
q
=(n mod M) T
d
+N T
d

n
N

+M T
w

n 1
N

+
T
f
2
. (10)
V. PERFORMANCE EVALUATION
We have implemented HMACin ns-2 network simulator with
Carnegie Mellon Universitys wireless model extension. In our
simulations, each node is assumed to have one omnidirectional
antenna and uses two ray ground reection radio propagation
models. The transmission and sensing ranges are set to 250 and
550 m, respectively.
WANG et al.: CROSS-LAYER OPTIMIZED MAC TO SUPPORT MULTIHOP QOS ROUTING FOR WIRELESS SENSOR NETWORKS 2561
TABLE I
MAJOR IMPLEMENTATION PARAMETERS
Fig. 5. Multiple-hop chain topology.
We mainly choose the following three metrics to investigate
the performance of HMAC and compare it with S-MAC [1] and
RMAC [13]:
1) average energy consumption: total energy consumed dur-
ing the simulation divided by the total number of nodes
in the network;
2) average end-to-end delay: time difference between the
arrival times of a packet at each pair of source and desti-
nation averaged over all packets successfully received by
destinations;
3) delivery ratio: total number of packets successfully ar-
rived at receivers divided by the total number of packets
sent by all sources
In our experiments, the data packet length is xed as 50 B
for all three MAC protocols under evaluation. The average
power consumption in transmission, receiving, idle listening,
and sleeping states are modeled as 1.83 : 1 : 1 : 0.001. The other
key simulation parameters are summarized in Table I.
In our simulation, we assume that each data packet is for-
warded along the shortest paths toward the destination. In the
following sections, we describe the network topologies used in
our simulations and present the corresponding results.
A. Simulation Topologies
In our simulation, we use two types of popular topologies
adopted in both S-MAC [1], [2] and RMAC [13] to study the
performance: 1) multihop chain network (see Fig. 5) and 2) a
multihop cross network (see Fig. 6).
B. Energy Consumption Comparison
Each source generates constant bitrate (CBR) trafc ows
to its corresponding destination. The trafc arrival interval
of CBR ows is varied from 5 to 50 s with 5-s steps.
From Figs. 79, we can see that HMAC outperforms S-MAC
and RMAC in terms of energy consumption, since HMAC
is scheduling based and uses a novel wake up scheme. Note
that the energy consumed in the synchronization operation in
HMAC is included in the results reported here. Figs. 7 and 8
show that the nodes in HMAC consume less energy than both
S-MAC and RMAC on average. While Fig. 9 demonstrates
Fig. 6. Multiple-hop cross topology.
Fig. 7. Energy consumption in a ten-hop chain topology.
Fig. 8. Energy consumption in a ten-hop cross topology.
Fig. 9. Standard deviation of energy consumption in a ten-hop cross topology.
that the nodes remaining energies in HMAC are more evenly
distributed than in both S-MAC and RMAC, and hence, HMAC
will have a longer lifetime on average. In Figs. 7 and 8, we
should mention that the nodes in S-MAC consume less energy
in the scenario of = 5 than that of = 10. This is because
more packets get dropped at nodes due to collision under the
increased trafc loads, and we assume that the MAC layer does
not cache more than one packet.
C. End-to-End Latency Comparison
We vary the length of paths between each pair of source
and destination nodes. The experimental results are shown in
Figs. 10 and 11 for the two network topologies used in our
2562 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 59, NO. 5, JUNE 2010
Fig. 10. End-to-end latency in a ten-hop chain topology.
Fig. 11. End-to-end latency in a ten-hop cross topology.
experiments. To evaluate how the MAC layer can enhance
coarse-grained QoS support for a particular class of trafc,
rst, only TYPE I ows are considered and generated in source
nodes. The simulation results demonstrate that for TYPE I
ows, HMAC outperforms both RMAC and S-MAC. This
is partly due to the fact that HMAC adopts a shorter frame
structure; at each hop, packets can be transmitted and forwarded
faster in HMAC than in S-MAC and RMAC. In addition, our
proposed channel reservation scheme signicantly reduces the
end-to-end delivery latency for TYPE I ows. In the mean time,
HMAC allows multiple data transmission within a single MAC
frame duration, which can further expedite the data delivery
when more than one sender exists in a small area. This can
be justied from the point view of queuing theory, since a
short frame length means a short service time, which reduces
collision and queuing delay.
Next, we discuss the latency performance of HMAC when
different classes of trafc ows coexist in the network and
validate our performance analysis conducted in Section IV. To
do this, we only use cross topology and perform the experiment
as follows: We let one source generate TYPE I ows and the
other source generate TYPE II ows. Similar to the experi-
ments reported for RMAC [13], we force the two sources to
simultaneously generate their CBRtrafc to intentionally create
channel congestion in the intersection area of the two chains.
The experiment and analysis results are shown in Fig. 12. From
the gure, we can see that under channel reservation, TYPE I
ows have signicantly lower end-to-end latency than TYPE II
ows. This is due to two reasons: 1) HMAC provides higher
priority channel access to TYPE I ows over TYPE II ows,
and 2) the proposed channel-reservation scheme at the MAC
layer allows TYPE I ows to travel across multiple hops within
a MAC frame. The gure also shows that our analysis results
match well with the simulation results.
D. Delivery Ratio Comparison
Next, we evaluate the performance in terms of delivery ratio.
Figs. 13 and 14 show the delivery ratio under different trafc
Fig. 12. HMAC latency. Simulation versus analytical results in a ten-hop
cross topology.
Fig. 13. Delivery ratio in a ten-hop chain topology.
Fig. 14. Delivery ratio in a ten-hop cross topology.
loads in chain and cross topologies, respectively. We change
the CBR trafc rate from 1 packet/s to 1 packet/40 s to evaluate
the performance of these three MAC protocols. For HMAC, we
only test TYPE I trafc ows in both topologies. Different from
the previous tests, to increase the delivery ratio of S-MAC and
RMAC, we adopt a 10% duty cycle for both protocols instead
of the 5% listed in Table I. In our simulation, we assume that
the MAC layer does not cache more than one packet. If a new
packet arrives at the MAC layer for delivery, then the old packet
will be dropped if it is still not delivered. Same as in RMAC,
we intentionally let the two sources generate CBR trafc ows
at exactly same time to create the packet-collision situation
at the middle node in cross topology. Both Figs. 13 and 14
demonstrate that the HMAC network provides higher delivery
ratio than S-MAC and RMAC when the network experiences
high trafc load. This is quite reasonable because HMAC has
a short frame size, which helps nodes to quickly move packets
to their next hops and, hence, signicantly reduces the packet
drop rate due to buffer overow or channel contention. This
fast packet delivery is particularly important when the network
experiences heavy trafc load. For example, when an event is
detected within an area, the nodes in this area may simultane-
ously report the events data, which potentially creates much
more trafc. Fig 14 shows that HMAC signicantly reduces
packet collisions under heavy trafc load.
WANG et al.: CROSS-LAYER OPTIMIZED MAC TO SUPPORT MULTIHOP QOS ROUTING FOR WIRELESS SENSOR NETWORKS 2563
VI. CONCLUSION
Providing coarse-grained end-to-end QoS support for multi-
hop routing in WSNs is important for several delay-sensitive
applications. This paper has investigated QoS-based rout-
ing at the MAC layer. It presents a cross-layer optimized
MAC protocol referred to as HMAC, which is particularly
suitable for WSNs in terms of energy efciency, latency,
and packet-delivery ratio. HMAC combines energy-efcient
schemes from existing contention-based and TDMA-based
MAC protocols for WSN to improve network performance.
A channel-reservation technique is proposed to drastically im-
prove the end-to-end latency for TYPE I trafc ows. Through
extensive simulation, we prove that HMAC provides superior
performance in end-to-end latency, throughput, and per-node
fairness while maintaining high energy efciency of the net-
work compared with existing MAC protocols for WSNs.
REFERENCES
[1] W. Ye, J. Heidemann, and D. Estrin, An energy-efcient MAC protocol
for wireless sensor networks, in Proc. IEEE INFOCOM, New York,
Jun. 2002, pp. 15671576.
[2] W. Ye, J. Heidemann, and D. Estrin, Medium access control with co-
ordinated adaptive listening for wireless sensor networks, IEEE/ACM
Trans. Netw., vol. 12, no. 3, pp. 493506, Jun. 2004.
[3] T. V. Dam and K. Langendoen, An adaptive energy-efcient MAC
protocol for wireless sensor networks, in Proc. 1st ACM Conf.
Embedded Netw. SenSys, Los Angeles, CA, Nov. 2003, pp. 171180.
[4] J. Polastre, J. Hill, and D. Culler, Versatile low power media access
for wireless sensor networks, in Proc. 2nd ACM Conf. Embedded Netw.
SenSys, Baltimore, MD, Nov. 2004, pp. 95107.
[5] V. Rajendran, K. Obraczka, and J. J. Garcia-Luna-Aceves, Energy-
efcient, collision-free medium access control for wireless sensor
networks, Wireless Netw., vol. 12, no. 1, pp. 6378, Feb. 2006.
[6] Z. Chen and A. Khokhar, Self organization and energy efcient TDMA
MAC protocol by wake up for wireless sensor networks, in Proc. IEEE
SECON, Oct. 2004, pp. 335341.
[7] M. Younis, K. Akayya, M. Eltowiessy, and A. Wadaa, On handling QoS
trafc in wireless sensor networks, in Proc. 37th Int. Conf. HICSS, Big
Island, HI, Jan. 2004, p. 90 292.1.
[8] I. Rhee, A. Warrier, M. Aia, and J. Min, Z-MAC: A hybrid MAC for
wireless sensor networks, in Proc. 3rd ACM Conf. Embedded Netw.
Sensys, Nov. 2005, pp. 511524.
[9] The Network Simulator. [Online]. Available: www.isi.edu/nsnam/ns
[10] J. Elson, L. Girod, and D. Estrin, Fine-grained network time synchro-
nization using reference broadcasts, in Proc. ACM SIGOPS, New York,
2002, pp. 147163.
[11] Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY)
Specication, IEEE Std. 802.11-1999 ed., 1999.
[12] G. Lu, B. Krishnamachari, and C. S. Raghavendra, An adaptive energy
efcient and low-latency MAC for data gathering in wireless sensor net-
works, in Proc. 18th Int. Parallel Distrib. Process. Symp., Apr. 2004,
pp. 224231.
[13] S. Du, A. K. Saha, and D. B. Johnson, RMAC: A routing-enhanced duty-
cycle MAC protocol for wireless sensor networks, in Proc. INFOCOM,
May 2007, pp. 14781486.
[14] Y. Li, W. Ye, and J. Heidemann, Energy and latency control in low
duty cycle MAC protocols, in Proc. IEEE WCNC, Mar. 2005, vol. 2,
pp. 676682.
[15] G. Ahn, E. Miluzzo, A. T. Campbell, S. G. Hong, and F. Cuomo,
Funneling-MAC: A localized, sink-oriented MAC for boosting delity
in sensor networks, in Proc. 4th ACM Conf. Embedded Netw. Sensys,
Boulder, CO, Nov. 2006, pp. 293306.
[16] V. Bharghavan, A. Demers, S. Shenker, and L. Zhang, MACAW: A
media access protocol for wireless LANs, in Proc. ACM SIGCOMM,
London, U.K., Sep. 1994, pp. 212225.
[17] Y. Liu, I. Elhanany, and H. Qi, An energy-efcient QoS-aware media
access protocol for wireless sensor networks, in Proc. Int. Conf. Mobile
Adhoc Sens. Netw., Nov. 2005, pp. 189191.
[18] S. Ganeriwal, R. Kumar, and M. Srivastava, Timing-sync protocol for
sensor networks, in Proc. 1st ACM Conf. Embedded Netw. SenSys,
Los Angeles, CA, Nov. 2003, pp. 138149.
[19] M. Xu, M. Zhao, and S. Li, Lightweight and energy efcient time syn-
chronization for sensor network, in Proc. Int. Conf. WCNM, Sep. 2005,
pp. 947950.
[20] S. C. Ergen and P. Varaiya, On multi-hop routing for energy efciency,
IEEE Commun. Lett., vol. 9, no. 10, pp. 880881, Oct. 2005.
Heping Wang received the B.S. degree in computer
science from the University of Electronic Science
and Technology of China, Chengdu, China, the M.E.
degree in computer engineering from Beijing Uni-
versity of Posts and Telecommunications, Beijing,
China, and the Ph.D. degree in electrical and com-
puter engineering from the University of Illinois at
Chicago in 2009.
He has seven years of industry R&D experience
in embedded system design and network protocol
design. He is currently a Lead Software Engineer
with Lemko Corporation, Schaumburg, IL. His current research focuses on
scalable and energy-efcient medium-access control and routing layer protocol
designs and cross-layer performance optimization for wireless sensor networks.
Xiaobo Zhang received the B.S. degree from the
University of Science and Technology of China,
Hefei, China, in 2003 and the M.S. degree and the
Ph.D. degree in electrical and computer engineering
from the University of Illinois at Chicago in 2006
and 2009, respectively.
He is currently a Senior Engineer with CISCO
Systems, San Jose, CA. His research interests in-
clude data gathering and distortion analysis in wire-
less sensor networks.
Farid Nat-Abdesselam (M06) received the en-
gineer degree in computer science from the
University of Sciences and Technologies Houari
Boumediene, Algeria, in 1993 and the Ph.D. degree
in computer science from the University of Versailles
Saint Quentin, Versailles, France, in 2000.
He was previously with INRIA Lille Nord
Europe. He is currently an Associate Professor with
the University of Sciences and Technologies of Lille,
Villeneuve dAscq, France. His research interests lie
in the eld of computer and communication net-
works with emphasis on architectures and protocols for quality of service and
security in internet protocol-based networks, mobile ad-hoc, sensor, vehicular,
and mesh networks, and overlay networks.
Dr. Nat-Abdesselam is a member of the IEEE Communications and Com-
puter Societies.
Ashfaq Khokhar (F09) received the B.Sc. degree
in electrical engineering from the University of En-
gineering and Technology, Lahore, Pakistan, in 1985
and the Ph.D. degree in computer engineering from
the University of Southern California, Los Angeles,
in 1993.
He is currently a Professor with the Depart-
ment of Electrical and Computer Engineering,
University of Illinois at Chicago. His research in-
terests include wireless and sensor networks, multi-
media systems, data mining, and high-performance
computing.
Dr. Khokhar was the recipient of the National Science Foundation CAREER
award in 1998. His paper entitled Scalable S-to-P Broadcasting in Message
Passing MPPs won the Outstanding Paper Award at the International Confer-
ence on Parallel Processing in 1996. He is an IEEE Fellow for his contributions
to multimedia computing and database systems.