Mac Protocols for Wireless Sensor Networks

Hans-Christian Halfbrodt
Advisor: Pardeep Kumar

Institute of Computer Science

Freie Universitt Berlin, Germany
[email protected]

January 2010

Contents

Abstrac

Introduction in WSN MAC Protocols

Motivation

WSN Background

MAC Requirements

differences to IEEE802.11 (WLAN)

Protocol Presentation

S-MAC

B-MAC

WiseMAC

IEEE 802.15.4

X-MAC

Comparison

S-MAC and B-MAC

X-MAC and WiseMAC

X-MAC and B-MAC

Conclusion and Outlook

Figures

10

Refencies

11

I. Abstract
This work presents different MAC protocols for wireless sensor networks and compares them to
each other. Therefore it starts with an introduction in wireless sensor networks. It will explain the
important role of MAC protocols for energy saving and why currently common protocols don't fit for
the actual requirements. After presenting a selection of MAC protocols their properties will be
compared to each other.

II. Introduction
Motivation
Today it is possible to build very small hardware devices with wireless communication for
monitoring and measuring miscellaneous values of the environment. There are a lot of application
areas. One is monitoring buildings and their surrounding terrain. A common solution is to install
wired sensors. Wired solutions have a continuous energy supply, but it is very expensive to lay or
replace cables for them. Therefore a wireless solution with a battery which can be completely
exchanged periodically is more eligible, but only if the exchange period is not too small. An other
utilization for wireless sensors is medical observation. Patients, who have for example a
pacemaker or comparable medical devices, want to move as free as possible, but the medical
devices have to exchange information about their health status. A small wireless sensor can realize
this task. It could send an emergency signal in case and with a long battery lifetime it needs less
servicing. That seems more useful than a large or even wired device. Other applications can
report position or traffic flow informations. This may be helpful for congestion prevention and
accidents warnings. Today RFID is a common technology for product identification in warehouses.
But at a container harbour are thousands of big containers to monitor. It is not possible for a worker
to just go around and scan every RFID chip. Wireless sensors could provide information about
container contents and status to a central instance. In agriculture sensor nodes could monitor the
growth of field crops on large areas. They could check if the ground is humid and can even report
statistics about the fall of rain. There are a lot of possible applications for the general concept of
wireless sensor nodes, but they are limited by the technical specifications.
WSN Background
The devices are typically equipped with a processor, memory, sensors and communication devices
with radio systems. All these functions can be put together in one microchip. Wireless devices are

very flexible as long as the power supply, usually a battery, provides enough energy. But charging
or replacing a battery can be very expensive and complicated. That is why energy saving is so
important. Usually the sensor nodes don't need to stay active continuously, they only transmit data
periodically. Especially wireless communication consume a lot of energy by amplifying received
signals or sending data. The radio units have generally three states: Transmitting signals costs
most energy. Receiving signal doesn't need the same amount of energy, but it also needs a lot.
The Idle state needs (nearly) no power. To be reliable for network peers the devices have to
interpret the received radio signal even if there is nothing transmitted. Based on these Informations
the idea is to put the devices to sleep when they are not used and let them periodically wake up.
That is not as easy as it sounds. To realize this idea the communication partners have to be
synchronized at least for each single transmission. The wireless extension is managed by a media
access control protocol. The task of this protocol is to transport local data to a target peer over the
physical layer, the networking hardware.

Requirements for Wireless Network MAC Protocols

A Wireless Medium is a shared medium. This means one instance in the signal range is allowed to
send data at most. This data can be received by one or more attendees. The MAC Protocol has to
transmit given information frames over this shared medium to a network peer. To do so it is
required that every network peer has an identifier. This is usually called a MAC address and has to
be unique for the current network. It is even better if the address is global unique because adding
or exchanging sensors in different networks might be possible. How this MAC address looks like is
part of the MAC protocol definition. There should be a large range of addresses because the
number of participants in the network may increase fast. The network size should be scalable, a
small fixed limit of peers is not useful in general. Autonomy should be guaranteed. It would be too
complicated if a human has to set up a lot of single parameters on the mac layer. It is also
expected that the network is self organizing. Possibly new peers appear, others might disappear or
the topology will be changed. A network has to be reliable. It is not useful if a user has no
warranties for successful participation. These are basic properties which should be fulfilled by a
MAC protocol.

Why WLAN IEE 802.11 is not usable for WSN

There are common MAC protocols for wireless communication. One popular is IEEE 802.11 used
for wireless local area networks. WLAN is used for high speed data exchange. A typical bandwidth
today is 54 Mbit per second and is theoretical possible up to 300 Mbit per second (IEEE 802.11n).

To be reliable here means especially be high available. It is not enough to say that every request
will be handled eventually. Latency is important for reducing waiting periods, too. A user who
requests a web page on his laptop wants to see it on his screen as soon as possible. Laptops and
smart phones are typical examples for the utilization of WLAN. They have large batteries and they
are easy to recharge. Because of these requirements WLAN radios are in listen mode when they
are idle. Put to sleep, wake up and connect with a base station or peer would take to long and
incoming traffic would not be noticed if not expected. Idle listen and receiving messages which
were dropped should be avoided by sensor nodes. This is called overhearing. The devices named
above are used at a limited area. They can be connected peer-to-peer or via a central access
point. Sensor networks may be distributed on a larger area. It could be difficult to to provide a large
number of wired access points. In sensor networks you typically have to transmit measured values.
This means you have a many to one communication. These values are periodically updated. Their
size is small, often a few bytes suffice. That is why the update interval is high in comparison to the
time needed for transmission over the network. Most of the time there is no activity in the network.
So always listening to the media would be to expensive in the view of energy costs. The basic idea
is to shut the radios usually down and let them only wake up from time to time. One have to check
for incoming data though. It may be possible that a node has to forward the data. Mobile phones or
WLAN devices have only one hop. Sensor nodes may have many hops. For these reasons sensor
nodes need specialized MAC protocols.

II. Presentation of Mac Protocols

S-MAC
This Protocol is based on the adaptive listening concept. Nodes are in sleep mode and listen
periodically if there is a data transmission announced. In that way the virtual carrier sense can be
used and the radio can remain turned off until the transmission of the neighbour is done. For
synchronization nodes listen to and send routinely SYNC packets via broadcast. Because of the
distributed ad-hoc structure there are nodes which can reach each other directly and nodes which
can't. The groups of nodes which can reach each other are called virtual clusters. Nodes in the
same virtual cluster synchronize each other. Nodes at borders of two clusters may have two wake
up schedules. The clock drift between the peers is very small in comparison to the wakeup period.
But to ensure a received message is complete nodes wait a short time value before sending. To
avoid overhearing the transmission time is part of the announcement. There is no central access
point, network peers communicate with each other. It follows that no instance configures the

network, so every node has to detect its communication peers. Neighbour detection is expensive. A
node with no neighbours

performs the detection more often than a node with one or many

neighbours. There is is no fairness guarantee. To prevent starvation the carrier sense time is
randomized in a certain time window. It is possible that sources can't reach the destination directly.
Then the data will be transmitted over multiple hops, nodes have to forward data (message
passing). To reduce the latency nodes which overhear a neighbours transmission wake up short
time before the end of the transmission to be prepared for message passing. Because of limited
memory and possible different memory sizes in a heterogeneous network S-MAC support
message fragmentation.

B-MAC
B-MAC is designed for an Ad-Hoc network of nodes with N-sender to 1-receiver transmissions.
The basic idea of B-MAC is to keep the protocol simple. That allows very small implementations,
an important point because of the limited available memory. Like the other protocols B-MAC uses
periodically sleep/wakeup cycles. The mechanism used here is called Low Power Listening. LPL
means in the wakeup time the node listens for incoming data transmissions. If there is no data
received, called a false positive, a timeout interrupts the listen state. Otherwise the node waits for
complete packet transmission. To ensure that the received packet is complete from the beginning
there is a preamble time of 100ms added after the wakeup. Fairness is not guaranteed by LPL.
The sleep periods of the nodes can differ to each other, B-MAC is asynchronous. When there is
data to send a node switches the radio mode and starts to send an announcement. This
announcement must be long enough to make sure that the receiver notices, even if the receiver
starts sleeping at the beginning. Afterwards the sender transmits the target address and starts
sending data. Asynchronous networks don't need complicated and expensive synchronization
methods. There is no data fragmentation used in B-MAC. This would be more complicated to
coordinate and B-MAC expects short messages like the ones commonly used for sensor
informations. An other concept to reduce the amount of needed energy is clear channel
assessment (CCA). This is used for clear channel detection. For energy reduction a better
separation between signals and noise on the channel is useful. Therefore the noise must be
analysed. In case of a false positive a sample was put into a queue. It makes sense to capture and
analyse more than one sample because the noise caused by the environment changes
continuously. An optional feature is using acknowledgements. B-MAC has an application interface
for flexible configuring parameters like this. Other options are for example the check interval. A
good value for this sometimes depends on the use case so this can be adjusted by a higher layer
application.

WiseMAC
WiseMac is an infrastructure protocol. It assumes there is one central unit with unlimited energy
supply and connection to an other high speed network e. g. Ethernet to exchange data packages.
That is why this unit is called access point. Because of the independence to batteries the access
point should manage the network. Therefore it has a table containing the wakeup times for every
known node. That reduces the needed listening time. These times are part of acknowledgement
packets send from the nodes. If the access point has information for a specific node it can be
notified as soon as possible. There must be a little delay before message passing because of the
possible clock drift to the nodes. The messages are expected to be short for sensor nodes. The
base station handles one message after the other. Synchronized nodes would have to wait except
for one node. That is why the wakeup schedules of the nodes are asynchronous. The packet
header contains a frame pending bit. It indicates there is more data waiting. After sending an
acknowledgement the node stays awake if the the bit is set. The sender sends the second packet
ongoing to the received ACK-packet.

IEEE 802.15.4
IEEE 802.15.4 defines another mac protocol for wireless sensor networks. The protocol is based
on CSMA/CA [6]. CSMA stands for carrier sense multiple access. A node which want to start a
transmission first listens for a short amount of time to the media if there is traffic on the channel.
The value of listen duration can be changed. If there is traffic the transmission has to wait.
Otherwise the transmission can be started. The extension CA means collision avoidance. A node
announces a packet transmission to its neighbours. Then the other nodes know that the channel is
used. This is used by the other WSN MAC protocols and IEEE802.11, too. IEEE802.15.4 uses no
RTS/CTS (request to send / clear to send) although it is a common combination. Using RTS/CTS
means that the data source sends a channel request message and the destination answers with a
clear / ready message. The reason for not using RTS/CTS is overhead prevention. The MAC
header has only 127 bytes, so additional flags are avoided. It is possible to use peer-to-peer
connections or a star topology. The default is star topology with a central instance. There are two
types of nodes: Full function devices and reduced function devices. Reduced function nodes are
only able to process their own data. They only have few resources like memory and small energy
capacities. Full functional nodes are better equipped and can relay messages for other devices.

X-MAC
X-MAC is meant for ad-hoc structure, too. It uses asynchronous wake-up schedules like B-MAC.
X-MAC tries to improve B-MAC by sending more intelligent announcements. Therefore the
announcements include the target address. There is not a large announcement for a transmission,
but instead short repeated messages were send. That is how the overhearing is reduced. Nodes
who are not addressed as targets can go to sleep after having received one short message. An
other enhancement is that there are short pauses between these preamble messages. This way
the receiver hasn't to wait for the full time period and can send an Acknowledgement to inform the
sending peer about its attendance and the data transmission can begin with a reduced preamble
time. If a second sender observes an announcement to the same target it waits for the finished
transmission and a randomized short period in addition. Then it sends its data direct without an
own announcement. To ensure this process works receivers usually wait an extra amount of time
for incoming data after a packet was received. This amount has to be at least the maximum
randomized time of the second sender. The randomization is needed for conflict avoidance
reasons. There may be others who want to communicate, too.

III. Comparison
S-MAC and B-MAC
Both, S-MAC and B-MAC, were implemented and tested [3]. For test implementation TinyOS [8] is
used, a very small operation system optimized for integrated devices. It is written in C. As test
hardware is Mica2 Motes used. S-MAC has a lot of features for synchronization optimization, while
B-MAC seems to be more simple. That is why the implementation of S-MAC needs more memory.
The test implementation by Polastre, Hill and Culler of S-MAC needs more than 6200 bytes. On
the contrary B-MAC with LPL, Acknowledgements and CTS/RTS needs less than 4700 bytes. The
energy consumption can be measured in energy per data (e. g. Joule per byte). If the radio device
is never sleeping the energy per byte increases linear to the pause between messages because of
the increased idle time. S-MAC without adaptive listening needs more energy with decreased data
rate too, but compared to the idle listen mode it is only a very small rise. This can be explained by
the overhead caused by synchronization and periodically media checking. Adaptive listen can
reduce the energy waste in high traffic loads, but in cases of lower traffic loads the advantage is
small. An important fact for a multi hop network is the latency. Latency is expected to increase with
every hop. Therefore Ye, Heidemann and Estrin measured latency in comparison to the hop count
(Figure 1). As expected the lowest latency occurs with no sleep. Without adaptive listen latency

rises fast. The exchange of short informations over many hops takes much time. Here adaptive
listen can show a real improvement. Because of the idea of being prepared if a neighbour receives
a message which he has to forward the latency increasing rate is nearly the same as by idle power
on. Only at the first hop we have to wait for the listening period. There latency is the same as
without adaptive listen. This causes a little shift of the overall latency with adaptive listening.
S-MACs energy consumption increases linear to the data rate. B-MACs energy use increases of
course with higher data rates too, but it is below S-MACs. A reason is that B-MAC has no
synchronization between the nodes, this causes a large overhead. In the test case from Polastre,
hill and Cullar were up to 20 nodes used. For different variants of B-MAC depending on the
number of nodes the maximum throughput decreases, while S-MAC throughput seems to be
continuously with increasing node count (Figure 2). Nevertheless B-MAC without ACK and RTSCTS is always faster for the tested range than S-MAC with unicast, while the energy consumption
of B-MAC is still lower. With throughput up to 4.5 times of S-MAC it seems that B.-MAC
outperforms. But this behaviour might changes for very large networks. Assuming most traffic is
unicast traffic and you have a large area for these big networks, so that every node can reach only
a small group of neighbours the behaviour for these small sub networks are probably the same as
for a small network. One very important factor for the performance is the network topology. It might
be that the performance of S-MAC in comparison to B-MAC increases if every node reaches only
two neighbours like a chain.

WiseMAC and X-MAC

WiseMAC is based on the concept of IEEE802.15.4 with infrastructure topology. Because of the
central communication control by the access point overhearing is minimized. Also there is no
message forwarding by the nodes. That increases the lifetime of the nodes. The time spend
waiting for ready peers is minimized too, because the central peer is able to listen always when it
is idle. I assume WiseMAC using Acknowledgements for data transmission increase latency
significantly as shown for other protocols.

X-MAC and B-MAC

X-MAC is based on the idea of B-MAC. To show the performance of X-MAC Buettner, Yee,
Anderson and Han tested their implementation in comparison with a LPL based protocol similar to
X-MAC. There is no direct comparison between X-MAC and B-MAC, but an LPL protocol similar to
X-MAC may be comparable to B-MAC. To demonstrate the advantage of the strobed
announcements Buettner, Yee, Anderson and Han set up a network where each node can reach all

others. There are messages send to one destination periodically. The measured duty cycles of
sending and receiving nodes are close to each other for the same protocol (Figure 3). The LPL
protocol duty cycles increase significantly with the node density. Complementary the duty cycle of
X-MAC increases very slow and is always below the one of LPL protocol. Because the usage time
corresponds to the energy usage the result of power consumption per node in relation to the
network density behave by the same scheme. This is expected from the optimized preamble
communication. Nodes which are not affected in the current transmission can quickly go back to
sleep. By the direct transmission of a third waiting node to the the target after a finished
transmission without a preamble procedure time is reduced.

Unfortunately there are no

performance tests comparing this LPL protocol to X-MAC. With a reduced transmission it is
expected that the latency shrinks while the throughput increases.

Conclusion
At my opinion there are many good ideas for intelligent energy reduction. But there is still
optimization possible. Not only Wireless radios don't need to stay active all of the time, sensors are
measuring for a short time and than they can sleep too. In these situation the whole device is idle
and could go to sleep. To do so there could control mechanism accessible by the Mac protocol
because typically the network is more frequently used than the sensor itself. A global control only
by the operating system might cause overhead and delay. Forwarding packages consumes a lot of
energy especially if there is a bottleneck in the topology. IEEE 802.15.4 proposes two types of
nodes [7]. Only one of them is forwarding data. Maybe this idea can be extended. It seems that XMAC and WiseMAC are currently good solutions, but this is only evaluated for small networks with
20 nodes or less. Tests with large networks are required to consolidate the choice of a MAC
protocol for practical use. Especially for area or building observation and controlling it is possible
that the number increases very fast and exceeds 20 nodes. Efficient protocols like B-MAC reduces
their functionality and refer tasks to upper abstraction levels. The efficiency may decrease if the
data transferred over the network increases because of the task shifting. I think there are still a lot
of possibilities for improvement on MAC protocols for wireless sensor networks.

Figures

Figure 1: Latency of S-MAC with and without adaptive listening, published

in [1]

Figure 2: Throughput of S-MAC and B-MAC dependent on the Number

of nodes, published in [3]

10

Figure 3: %Duty Cycle of X-MAC and compared LPL Protocol in

dependence on Node Density, published in [5]

References
[1] W. Ye, J. Heidemann and D. Estrin, Medium Access Control With Coordinated Adaptive
Sleeping for Wireless Sensor Networks IEEE/ACM Transactions on Networking, vol. 12, no. 3,
June 2004
[2] A. El-Hoiydi and J.-D. Decotingie, WiseMAC: An Ultra Low Power MAC Protocol for the
Downlink of Infrastructure Wireless Sensor Networks, CSEM, Swiss Center of Electronics and
Microtechnology, Inc., 2007
[3] J. Polastre, J. Hill and D. Culler, Versatile Low Power Media Access for Wireless Sensor
Networks, SenSys'04, ACM 1-58113-879-2/04/0011, November 3-5, 2004
(cst berkley, jlh labs) ?
[4] IEEE, Wireless Medium Access Control (MAC) and Physical Layer (PHY) Specifications for
Low-Rate Wireless Personal Area Networks (WPANs), IEEE Std 802.15.4-2003, 2003
[5] M. Buettner, G. Yee, E. Anderson and R. Han,X-MAC: A Short Preamble MAC Protocol For
Duty-Circled Wireless Sensor Networks, Department of Computer Science, University of Colorado
at Boulder, CO [USA], May 2006

11

[6] B. Scheers, W. Mees and B. Lauwens, Developments on an IEEE 802.15.4-based wireless

sensor network, Department CISS, Royal Military Academy, Belgium, published in Journal of
Telecommunications and Information Technology, February 2008
[7] J.-S. Lee, Performance Evaluation of IEEE 802.15.4 for Low-Rate Wireless Personal Area
Networks, IEEE Transactions on Consumer Electronics, Vol. 52, No.3, August 2006
[8] TinyOS Homepage, http://www.tinyos.net/, 01-15-2010

12