IEEE COMMUNICATIONS LETTERS, VOL. 18, NO.

5, MAY 2014

809

Asym-MAC: A MAC Protocol for Low-Power Duty-Cycled

Wireless Sensor Networks with Asymmetric Links
Myounggyu Won, Member, IEEE, Taejoon Park, Member, IEEE, and Sang H. Son, Fellow, IEEE

AbstractDuty-cycling is a primary technique significantly

improving the energy efficiency of Wireless Sensor Networks
(WSNs). Thus, a large number of MAC protocols have been
developed for duty-cycled WSNs. Especially, recently proposed
receiver-initiated MAC protocols are well suited for extremely
low duty-cycled WSNs. However, as it is shown in this paper, they
perform poorly when highly asymmetric links are present. In this
paper, a hybrid MAC protocol called Asym-MAC is proposed
which takes advantage of both receiver-initiated and senderinitiated MAC protocols to combat performance degradation
due to asymmetric links. Experimental results demonstrate that
Asym-MAC has up to 2.8X higher packet reception ratio and
up to 66.7% smaller packet transmission delay compared with
A-MAC, the state-of-the-art receiver-initiated MAC protocol.

TABLE I
P ERCENTAGE OF ASYMMETRIC LINKS [10]

Tx Power
-19dBm
-14dBm
-5dBm

Symmetric
50%
65%
88%

Asymmetric
43%
22%
6%

Unidirectional
7%
13%
6%

IRELESS Sensor Networks (WSNs) comprise of a

multitude of nodes with wireless communication, computation, and sensing capabilities. WSNs have been deployed
for diverse applications in medical, military, and manufacturing fields revolutionizing the way humans perceive physical
environments. In recent years, WSNs have become a key
component for Cyber-Physical Systems (CPS) [1].
Energy-efficiency is a fundamental issue in WSNs because
nodes are usually battery-powered due to the ubiquitous characteristics of WSNs. One major breakthrough for promoting
energy-efficiency is duty-cycling that allows nodes to periodically cycle between a sleep state and an awake state. In the
sleep state, nodes completely turn off their radio and processing hardware modules, thereby significantly saving energy
the current consumption of a typical Mica2/TelosB node
operated by two AA batteries is 216A/21A, 3.2mA/2.4mA,
7.0mA/21mA, and 8.5mA/23mA for sleep, idle listening,
transmission (TX at 0dBm), and reception (RX) modes, respectively [2][3]. There are mainly two types of duty-cycling
mechanisms: synchronous and asynchronous duty-cycling. In
synchronous duty-cycling, all nodes have synchronized duty
cycles, whereas, in asynchronous duty-cycling, nodes have
their own duty cycles. Between the two, asynchronous duty
cycling has been more popular because synchronous dutycycling requires time synchronization. Various MAC protocols
have been proposed for asynchronously duty-cycled WSNs.
For example, LB-MAC [4] is designed to prolong network
lifetime; DutyCon [5] provides end-to-end delay guarantee.

Recent studies are largely categorized into transmitterinitiated and receiver-initiated mechanisms. Transmitterinitiated MAC protocols like X-MAC [6] and BoX-MACs [7]
rely on transmission of preamble packets, thus suffering
from wasted energy for sending the preamble packets. On
the other hand, in receiver-initiated MAC protocols like RiMAC [8] and A-MAC [9], a receiver broadcasts a probing
packet whenever it wakes up from the sleeping state, while
a sender with data to transmit waits in the listening state
until a probing packet from the indented receiver is received.
Once the probing packet is received, the sender begins its data
transmission. This way receiver-initiated MAC protocols avoid
the overhead of sending preamble packets.
However, as it is shown later, receiver-initiated MAC protocols perform poorly in networks with asymmetric links. A
problem arises when a sender fails to receive a probing packet
from the intended receiver. Especially, if a probing packet is
missed multiple times due to the asymmetric link, the sender
must stay in the listening state until the probing packet arrives,
wasting energy and increasing delay. In the worst case of a
unidirectional link (or a very highly asymmetric link), data
packets are dropped, degrading PRR.
In this paper, Asym-MAC, a MAC protocol designed for
low-power duty-cycled WSNs with asymmetric links is proposed. A key idea is to dynamically switch between two modes
of operation: transmitter-initiated and receiver-initiated modes
depending on the asymmetry of a link which is measured
as a PRR difference between the two directions (up/down
links) a larger PRR difference means higher asymmetry [10].
More specifically, the receiver-initiated mode is used as the
default MAC protocol; when the link is determined to be
asymmetric, Asym-MAC switches its mode to the transmitterinitiated protocol. The hybrid method is shown to achieve
up to 66.7% smaller delay and 2.8X higher packet reception
ratio compared with the state-of-the-art receiver-initiated MAC
protocol A-MAC, in a given experimental setup.

Manuscript received December 3, 2013. The associate editor coordinating

the review of this letter and approving it for publication was S. Vural.
The authors are with the Department of Information and Communication
Engineering, Daegu Gyeongbuk Institute of Science and Technology (DGIST), Daegu, South Korea (e-mail: {mgwon, tjpark, son}@dgist.ac.kr).
T. Park and S. H. Son equally contributed to this work.
Digital Object Identifier 10.1109/LCOMM.2014.032014.132679

Recent studies have shown that a significant portion of

radio links are asymmetric in a typical WSN deployment [10].
Table I shows the percentage of asymmetric links measured
in Kansei testbed [11]. With smaller TX power, the testbed

Index TermsWireless sensor networks, MAC protocols.

I. I NTRODUCTION

II. M OTIVATION

c 2014 IEEE
1089-7798/14$31.00 

810

IEEE COMMUNICATIONS LETTERS, VOL. 18, NO. 5, MAY 2014

Preambles

Preambles
Sender

Listen B

DATA
Time

Case 1: probing pkt

not received
Receiver

B A

DATA

Case 2: data not

received
Time

Asym-MAC
Sender

Asym-MAC
Receiver

Wait for Auto-ACK

Fig. 1. Operation of a receiver-initiated MAC protocol. Fig. 2.

Listen for

DATA
Sender

DATA

Time

Time

C
C
A

C
C
A

T-Mode

Operation of the Asym-MAC protocol.

had more asymmetric links, and a non-negligible portion of

the links were unidirectional.
Figure 1 illustrates how asymmetric links may degrade the
performance of receiver-initiated MAC protocols. The receiver
periodically broadcasts a probing packet denoted by B. If B is
received by the sender, the sender starts its data transmission
denoted by DAT A. However, due to the asymmetry of the
link, two potential failure scenarios may arise. The first case
(Case 1) is when B is failed to be delivered to the sender. In
this case, the sender has to wait for the next probing packet
in the listening state, consuming time and energy. The second
case (Case 2) arises when the sender fails to transmit data
due to the low-quality link from the sender to the receiver
(i.e., the down-link). The focus of this paper is on addressing
the issue of the asymmetry of the link from the receiver to
sender (i.e., the up-link). However, it is worth to note that a
simple yet effective strategy, e.g., increasing transmit power
or increasing the number of retransmissions, can be adopted
to address Case 2, thereby attempting to address both cases.
III. P ROTOCOL D ESIGN
The objective of this work is to design a MAC protocol
that works well under the circumstances of highly asymmetric
links. To achieve this goal, a hybrid approach that leverages the
advantages of both transmitter-initiated and receiver-initiated
MAC protocols is employed. Specifically, Asym-MAC reduces
the overhead of preamble packets by relying on the design of a
receiver-initiated MAC protocol. At the same time, it addresses
the issue of asymmetric links by appropriately integrating the
design principles of a transmitter-initiated MAC protocol into
the receiver-initiated MAC protocol.
Figure 2 illustrates how Asym-MAC works. Asym-MAC
has two modes of operation: R-mode and T-mode. R-mode is
the default mode in which a receiver-initiated MAC is used.
Asym-MAC switches its mode to T-mode when it fails to
receive a probing packet from the intended receiver more
specifically, when a timeout (waiting time exceeding a probing
period) occurs for receiving the probing packet times, where
is a system parameter. In T-mode, the transmitter sends
a series of small preamble packets to notify the receiver of
the pending data transmission. Once the receiver captures
the signal of on-going transmission of preamble packets, it
receives data from the sender.
A challenge is that, while the sender knows when to change
its mode to T-mode (i.e., when a timeout for receiving a
probing packet from the intended receiver occurs more than
times), the receiver has no idea about when to change the
mode. A key idea to address this challenge is basically to
add a short clear channel assessment (CCA) period at the end

C
C
A

DATA

Receiver1

DATA
Time

R-Mode

C
C
A

Time

B
Receiver2
Time

Fig. 3.

A collision scenario.

Algorithm 1 Asym-MAC
1: /* Sender s sending data to Receiver r */
2: if A probing packet received from r then
3:
Send data to r.
4: else if Number of timeout > then
5:
Switch to T-mode and Send data to r.
6: end if
7: Switch back to R-mode.
8: /* r receiving data from s */
9: if Preamble received from s then
10:
Wait for data from s after the preamble.
11: end if
12: Go to sleep mode.

of each probing packet transmission. Adding the CCA period

does not degrade the energy efficiency much given the facts
that: first, the CCA check lasts for only a very short period
of time, and second, the CCA check consumes an order of
magnitude smaller energy than the TX and RX modes [9]. In
each CCA period, the receiver checks for potential preamble
packet transmission from the transmitter who is in T-mode,
so that the receiver receives data regardless of the quality
of the up-link. The pseudo-code of Asym-MAC presented in
Algorithm 1 summarizes the operation of Asym-MAC.
Another challenge is a potential collision between a preamble and a probing packet. Consider a scenario consisting of
three nodes denoted by Sender, Receiver1 and Receiver2 (See
Figure 3). If Receiver2 s probing packet is sent out right
after the transmission of Senders preamble, Receiver1 will
not be able to detect the preamble during the short CCA
period due to the collision of the preamble packet from Sender
and probing packet from Receiver2 . To address this issue,
the CCA-Extension is employed: when a collision occurs, a
receiver performs another CCA check, increasing the chance
of receiving the next preamble.
IV. P ERFORMANCE E VALUATION
The PRR and packet transmission delay of Asym-MAC are
measured and compared against A-MAC [9]. For completeness, performance measurements for a transmitter-initiated
MAC protocol, i.e., BoX-MAC-1 [7], are also provided, which
can be considered as an upper bound performance as a
transmitter-initiated MAC protocol is not affected by the
asymmetry of the link (i.e., from receiver to sender) because
it does not rely on a probing packet. However, it should be
noted that a sender-initiated MAC protocol suffers from low
energy efficiency and causes the hidden terminal problem as

WON et al.: ASYM-MAC: A MAC PROTOCOL FOR LOW-POWER DUTY-CYCLED WIRELESS SENSOR NETWORKS WITH ASYMMETRIC LINKS
30

100

AMAC
AsymMAC
BoXMAC1

3000

20

15

10

90
80

2500

70

PRR (%)

Perpacket delay (msec)

Asymmetry (%)

25

2000

1500

0.3

0.6

0.9

1.2 1.5 1.8

Distance (meter)

2.1

2.4

2.7

Fig. 4.
Asymmetry for different
distances.

AMAC
AsymMAC
BoXMAC1

60
50
40
30

1000

20
10

500

811

10

20

30

40

50

60

70

80

90

10

Fig. 5.
metry.

Per-packet delay per asym-

well as low channel utilization [8]. For this experiment, a pair

of TelosB nodes running TinyOS 2.1.0 is deployed in an office
area. One node transmits a packet of 15 bytes every second
a total of 1,000 packets to the other node. To investigate the
effect of collisions, the number of receivers within one-hop of
the sender is increased up to 10 nodes.
Previous work [12] notes that there are various factors
causing asymmetric links like hardware, transmit power, interference, and distance between nodes. Among the factors,
Sang et al. [10] considered the transmit power and distance as
major factors. Thus, to characterize the asymmetry of the link
between the two nodes, the PRR difference between the up and
down links (i.e., the asymmetry) is measured by varying the
distance between the two nodes with TX power of -25dBm.
Figure 4 illustrates that the PRR difference is small when the
two nodes are close enough to each other, while the PRR
difference significantly increases when the distance becomes
greater than 2.1 meters, showing the asymmetry of the link.
A challenge is that the asymmetry of the link changes over
time due to the changes in the interference level around the
experimental environment (e.g., WiFi signals in the building);
thus, it is difficult to reproduce the asymmetry for the same
distance for different runs of experiments unless a complete
control over the WiFi signals in the building is given. As a
result, the degree of asymmetry is controlled by artificially
dropping packets at the MAC layer while using the maximum
transmit power (0dBm) with the minimum distance (less than
5cm) between the two nodes it is experimentally confirmed
that with this optimal setting, nodes constantly maintained
PRR greater than 97% in both directions of the link. Specifically, the packet drop ratio of the up-link is changed from
10% to 100%, creating controllable asymmetry from 0% to
90%. Now given the control over the asymmetry of the link,
the measurements of PRR and delay of Asym-MAC, A-MAC
and BoX-MAC-1 by varying the asymmetry can be performed.
A. Per-packet Delay
Per-packet transmission delay is defined as a period of time
between loading a data packet in a transmit buffer and successfully sending it to the intended receiver. For this experiment,
an average of per-packet delay for 1,000 packet transmissions
is measured; the probing interval is set to 500msec (i.e., the
receiver sends a probing packet every 500msec); and is set
to 1. Figure 5 depicts per-packet delay as a function of the
asymmetry of the link. It is observed that per-packet delay
increases as the asymmetry increases. This is mainly because

20

30

40

50

60

70

80

90

Asymmetry (%)

Asymmetry (%)

Fig. 6.
(500msec).

PRR per asymmetry

the sender misses more probing packets when the asymmetry

is high; when the transmitter misses a probing packet, it
needs to wait for the next probing packet, thereby increasing
per-packet delay. Asym-MAC has delay no longer than 1sec
because it changes its mode to T-Mode when it consecutively
fails to receive a probing packet twice (i.e., = 1). In
contrast, A-MAC has particularly high per-packet delay when
the asymmetry is very high (i.e., over 50%), because the higher
the asymmetry, more likely to have consecutive failures of
receiving a probing packet. Overall, in the given experimental
setup, Asym-MAC successfully reduces the per-packet delay
of A-MAC by up to 66.7%.
B. Packet Reception Ratio (PRR)
The PRR of Asym-MAC and A-MAC is measured. Figure 6
shows the results (probing interval of 500msec). As shown,
PRR decreases for both Asym-MAC and A-MAC as the
asymmetry increases. Since a data packet is sent per second,
missing a probing packet at least two consecutive times leads
to a packet drop, thereby degrading PRR, assuming no buffering scheme is used. Note that for small asymmetry, both AMAC and Asym-MAC perform relatively well; however, PRR
of A-MAC starts to decrease significantly at high asymmetry.
The reason is that it is more likely for the transmitter to miss
a probing packet two consecutive times when the asymmetry
is very high. In contrast, since Asym-MAC changes its mode
to T-mode and sends the packet directly through the downlink when it detects a missing probing packet more than
times, Asym-MAC has higher PRR than A-MAC. The same
experiments with the probing interval of 1sec are performed,
i.e., even a single miss of a probing packet potentially leads to
decreased PRR. Figure 7 proves this by showing that PRR for
both Asym-MAC and A-MAC degrades even for very small
asymmetry. However, it is worth to note that PRR of AsymMAC is still higher than A-MAC, because although AsymMAC drops a packet for the first miss of a probing packet, it
starts to combat the loss starting from the second loss of the
probing packet because = 1.
C. Energy Consumption
The average current drawn by both receiver and transmitter
is measured for the three protocols. For this experiment, the
probing interval is set to 500msec, asymmetry to 50%, and
tx power to 0dBm. Figure 8 shows that compared with AMAC, the receiver of Asym-MAC consumes more energy,

812

IEEE COMMUNICATIONS LETTERS, VOL. 18, NO. 5, MAY 2014

100

100

12
Receiver
Sender

95

80
75
70

AMAC
AsymMAC
BoXMAC1

65

10

20

85

80

4
75

2
30

40

50

Asymmetry (%)

Fig. 7.

90

PRR (%)

Average current draw (mA)

PRR (%)

85

60

95

10

90

PRR per asymmetry (1sec).

Probing interval=500msec
Probing interval=1000msec
70

AMAC

AsymMAC

BoXMAC1

Fig. 8. Average current consumption.

which is the cost for Asym-MACs CCA checks. Similarly,

the transmitter of Asym-MAC consumes more energy than
A-MAC; however, interestingly, the difference is not large
with an increase of only 2.67%, implying that the transmitter
of A-MAC consumes energy unnecessarily waiting for the
probing packet especially when the link is asymmetric, although Asym-MAC consumes energy by changing the mode
and transmitting the preambles. Compared with the receiverinitiated protocols, the graph shows that BoX-MAC-1 has
higher energy consumption, which coincides with the reported
results [8].
D. Collisions
The different numbers (1 10) of receivers are deployed
within the communication range of the sender to see the
impact of collisions. All nodes periodically broadcast a probing packet, potentially causing a collision at the sender. The
asymmetry of the link is set to 50%, and PRR is measured
by varying the number of receivers. As Figure 9 shows, PRR
decreases when the number of nodes is increased. The results
indicate that although collisions between a probing packet
and a preamble are avoided at each CCA check using the
CCA-Extension scheme, a data packet may still collide with
a probing packet, which is a known drawback of receiverinitiated MAC protocols [9]. However, it is worth to note that
PRR may improve if the probe times are carefully designed
using a protocol like [13].
E. Storage Overhead
One concern for Asym-MAC is the storage overhead compared with A-MAC, because Asym-MAC has more functionalities. Thus, RAM and ROM sizes required for the
implementation of Asym-MAC and A-MAC on a TelosB mote
are measured. A simple RadioCountToLeds application [14]
is implemented on top of the two MAC protocols running on
TinyOS 2.1.0. It is found that Asym-MAC requires 16.3% and
32.9% more space for ROM and RAM respectively. However,
considering the small RAM and ROM sizes of A-MAC, we
believe that the required space for Asym-MAC is acceptable.

10

Number of nodes

Fig. 9.

PRR per number of nodes.

V. C ONCLUSION
Asymmetric links frequently observed in real-world deployment of WSNs degrade the performance of existing receiverinitiated MAC protocols. In this paper, a hybrid MAC protocol
(Asym-MAC) designed for WSNs with highly asymmetric
links is proposed. Future work includes the investigation of
optimal for various parameters like probing interval.
R EFERENCES
[1] R. R. Rajkumar, I. Lee, L. Sha, and J. Stankovic, Cyber-physical
systems: the next computing revolution, in Proc. 2010 DAC.
[2] V. Shnayder, M. Hempstead, B.-R. Chen, G. W. Allen, and M. Welsh,
Simulating the power consumption of large-scale sensor network
applications, in Proc. 2004 Sensys.
[3] Crossbow Inc, Telosb datasheet. Available: http://www.willow.co.uk/
TelosB Datasheet.pdf
[4] Y. Peng, Z. Li, W. Zhang, and D. Qiao, LB-MAC: a lifetime-balanced
MAC protocol for sensor networks, in Proc. 2012 WASA.
[5] X. Wang, X. Wang, G. Xing, and Y. Yao, Dynamic duty cycle control
for end-to-end delay guarantees in wireless sensor networks, in Proc.
2010 IWQoS.
[6] M. Buettner, G. V. Yee, E. Anderson, and R. Han, X-MAC: a short
preamble MAC protocol for duty-cycled wireless sensor networks, in
Proc. 2006 Sensys.
[7] D. Moss and P. Levis, BoX-MACs: exploiting physical and link layer
boundaries in low-power networking, in Technical Report SING-08-00,
Stanford University, 2010.
[8] Y. Sun, O. Gurewitz, and D. B. Johnson, RI-MAC: a receiver-initiated
asynchronous duty cycle MAC protocol for dynamic traffic loads in
wireless sensor networks, in Proc. 2008 Sensys.
[9] P. Dutta, S. Dawson-Haggerty, Y. Chen, C.-J. M. Liang, and A. Terzis,
Design and evaluation of a versatile and efficient receiver-initiated link
layer for low-power wireless, in Proc. 2010 Sensys.
[10] L. Sang, A. Arora, and H. Zhang, On link asymmetry and one-way
estimation in wireless sensor networks, ACM Trans. Sensor Networks,
vol. 6, no. 2, pp. 125, 2010.
[11] E. Ertin, A. Arora, R. Ramnath, M. Nesterenko, V. Naik, S. Bapat,
V. Kulathumani, M. Sridharan, H. Zhang, and H. Cao, Kansei: a testbed
for sensing at scale, in Proc. 2006 IPSN.
[12] G. Wang, D. Turgut, L. Blni, Y. Ji, and D. C. Marinescu, A MAC
layer protocol for wireless networks with asymmetric links, Ad Hoc
Networks, vol. 6, no. 3, pp. 424440, 2008.
[13] J. Degesys, I. Rose, A. Patel, and R. Nagpal, Desync: self-organizing
desynchronization and TDMA on wireless sensor networks, in Proc.
2007 IPSN.
[14] RadioCountToLeds-TinyOS. Available: http://www.tinyos.net/tinyos-2.
x/apps/RadioCountToLeds/