Carrier-Sence Multiprle Access (CSMA) Protocols

Leonidas Georgiadis
February 13, 2002

Introduction

Communication of information between two or more parties takes place over a variety of physical media
called channels. Such channels can be simple twisted pair cables, coaxial and optical cables, or the free
space. Sometimes the channel is dedicated to a specific transmitter-receiver pair. This may be the case
when two parties establish a telephone conversation over a dedicated cable. Channels of this type are
called point-to-point. There are situations, however, where multiple users need to have access to the
same channel. The most familiar one is human speech communication. When a number of humans
are located in the same room they all use the same channel, the atmosphere, for their conversation
exchange. Computer Local Area Networks (LANs) is another example: a common approach in this case
is to attach a number of computers to the same cable as in Figure 1. Hence, each computer can listen
to the transmission of every other computer attached to the same cable. For a third example, consider
Satellite communication. As shown in Figure 2 a number of terminals need to communicate between each
other but due to physical obstacles they cannot all listen to each other directly. Instead, each terminal
first sends the information to the satellite through the upstream channel. The satellite listens to the
upstream channel, receives the information and then retransmits to the downstream channel, to which
all terminal can listen. Hence the upstream channel needs to be accessed by all terminals. Channels of
this type are called multiple-access.
If the terminals in a multiple-access channel are left unchecked so that they can transmit information
whenever the need to do so, then the possibility arises that more than one terminals attempt to use the
channel at the same time. In such a situation, the concurrently transmitted messages interfere with each
other and generally cannot be received correctly by the intended receivers. Hence, a fundamental issue in
multiple-access channel communication is how to coordinate the transmitting terminals in order to avoid

Figure 1: A Local Area Network

Figure 2: Satellite Communications

or recover from the interference that may result because of concurrent transmissions. The mechanisms
by which this is achieved come by the name multiple-access protocols.
The simplest way to address the coordination problem in multiple-access communication is to avoid
concurrent transmissions altogether. To be more specific we must make certain assumptions about the
manner in which transmission of information takes place. First, as is very common today, we assume
that all information whether sound, picture or text is transformed to a sequence of bits, 0 or 1, and
that each terminal needs to transmit this sequence of bits to the receiving terminal - the receiver knows
how to recover the original information from the received sequence of bits . Let us assume further that
the sequence of bits is subdivided into groups called packets, and that the transmitter needs to transmit
one packet at a time to the receiver. All packets contain the same number of bits B. If bits can be
transmitted over the channel at a rate of R bits per second (bits/sec or bps), then each packet takes
T = B/R seconds to be transmitted. We refer to T as the "length" of the packet.
We are now ready to describe the protocol by which access to the channel is free of concurrent
transmissions. We divide time into fixed intervals of length T called time slots, see Figure 3. Hence each
slot fits exactly one packet. Let the number of terminals that can potentially use the channel be n. We
group the time slots into frames where each frame contains n consecutive time slots. Terminal i is allowed
to transmit in the ith time slot of each frame. The protocol just described is called Time Division Multiple
Access (TDMA) protocol. Since slots are allocated exclusively to each user, no interference occurs and
packets are transmitted successfully. Note that there is still a possibility that the packet may be received
in error due to noise that naturally exists on the channel, but this is a lower level issue which is addressed
by methods that are beyond the scope of the current discussion.
The TDMA protocol in eect divides the channel into n point-to-point channels. While simple, the
protocol has some serious disadvantages. First, if a terminal does not have packets to transmit, the slots
allocated exclusively to it cannot be used be anybody else, even if other terminals have a large number
of packets to transmit and could use these slots. The second disadvantage is related to packet delays.
Since the time interval between two successive slots during which terminal i can transmit is n time slots,
a packet generated randomly at a terminal will take on the average n/2 time slots to be transmitted,
a delay that can be very large if the number of terminals in the system is large. This will happen no
matter whether the rest of the terminals have packets to transmit or not.
The disadvantages of TDMA are due to the fact that a terminal can transmit only during the slots
allocated to it, even if other terminals are inactive. What if we dispense with the idea of allocating slots
exclusively to transmitters? In fact, what if we take the exact opposite approach and allow a terminal
3

Figure 3: The TDMA protocol

to transmit in any slot when it has packets to transmit? In this case, if no other terminal has packets to
transmit, then the given terminal can transmit a large number of packets with very low delay. However, if
more than one terminals wish to send packets in the same slot then a "collision " will occur and no message
will be received correctly. In this case one must specify how the terminals will react and cooperate in
order to make sure that the packets are eventually delivered to their intended destinations. The simplest
idea is to instruct the terminals to retransmit their collided packets. However, if more than one terminals
pick again and again the same slot for retransmission the packets will keep colliding and will never be
transmitted successfully. There are various methods to avoid this situation. We will concentrate on the
most prevalent method encountered in practice, randomized retransmissions. If collisions occur then the
terminals whose packets collided pick randomly some future time slot for retransmission. Hence, while
collision may again occur, it is hoped that eventually the transmitting terminals will each pick dierent
slots for transmission and thus their packets will be delivered successfully to their intended destination.
The algorithm just described comes by the name ALOHA protocol and will be described in more
detail in Section 2. This algorithm constitutes the basis for the development of Carrier Sense Multiple
Access (CSMA) protocol which takes advantage of certain channel features and transmitter capabilities
in order to provide improved performance. The CSMA protocol will be described in Section 3.

The ALOHA protocol

The ALOHA protocol was designed by Abramson [1] to provide radio communication between several
terminals scattered at various places over the islands of Hawaii. The terminals were sending their data
packets to a central station over a common channel (the upstream channel). The central station was
then retransmitting the packets to another channel (the downstream channel) that could be listened to
by all the terminals. The situation is similar to the one described in Figure 2. Collisions could occur
at the upstream channel if more than one terminals were attempting to transmit their packets. If this
happened, the central computer was informing all the terminals that a collision had occurred.
There are two versions of the ALOHA protocol, slotted and unslotted. Slotted ALOHA requires
time to be divided in time slots and terminals to transmit their packets at the beginning of each slot.
Unslotted ALOHA permits the stations to transmit their packets at any time. The retransmission policy
in case of collision is essentially the same for both protocols. In the next two sections we examine these
two variants of the ALOHA protocol. Unslotted ALOHA was the precursor of Slotted, but it will be
more instructive and simpler to concentrate on the Slotted ALOHA first.

2.1

Slotted ALOHA

Let us provide a model for this protocol. As in Section 1 the channel is divided in time slots. Terminals
are synchronized to transmit their packets at the beginning of a time slot. At the end of each time slot
terminals that transmitted their packets during that slot are informed whether there was a successful
transmission or a collision in the slot. If the packet that a terminal transmitted collides with another
packet, then the terminal attempts a retransmission in the next slot with probability p and defers for the
end of the next slot with probability 1 p. In case of deferral, at the end of the next slot the terminal
attempts again retransmission with probability p and defers with probability 1 p. Figure 4 shows an
example of the operation of the ALOHA protocol. At slot 1 three terminals attempt to transmit their
packets and there is a collision. Hence all three terminals will attempt to retransmit their packets. No
terminal chooses to retransmit at slot 2 which is thus idle. Terminals a and b attempt to retransmit at
slot 3 and hence there is again a collision. Terminal b is the only one attempting retransmission at slot
4 and its transmission is successful. The transmissions from terminals a and c collide again in slot 6,
but they eventually pick dierent slots for retransmission and their packets are transmitted successfully
in slots 8 and 9. Note that other terminals may become active (i.e., they may generate a new packet
for transmission) while the retransmission process takes place. These terminals may cause additional

Figure 4: The operation of the ALOHA protocol

collisions. For example, if terminal d generates a new packet and attempts to transmit it in slot 8,
an additional collision will occur. In the network designed by Abramson, all terminals (not only those
that transmitted their packets) can listen to the downstream channel and hence can be informed about
the status of the transmission at the end of the current slot, i.e., whether there was no transmission, a
successful transmission or a collision during the slot. However, the ALOHA protocol does not make use
of this extra information that a terminal can have.
The protocol just described has the desirable property that packets are not delayed at all if only one
terminal needs to transmit at a given time slot. What happens, however, when more than one terminals
attempt transmission? As the example in Figure 4 shows, in this case there will be collisions that will
be followed by retransmission attempts. This results in two ineciencies: a) slots may be wasted due
to collisions and b) slots may remain idle even though there are terminals that need to transmit their
packets; the latter will happen if all packets that attempt retransmission are deferring in the current
slot and no new packets are generated. It is important therefore to know the useful information that
can come out of the channel. An appropriate measure for this information is the average number of
successfully transmitted packet, S, per slot. We refer to S as the throughput of the channel.
Next we provide a method for evaluating S. We need to make first an assumption regarding the
statistics of new packets generation process: the number of new packets, K, generated for transmission
during a time slot, is a Poisson random variable with rate packets/slot. That is, the probability that
K = k is given by
Pr(K = k) = e

k
.
k!

(1)

This model of packet generation is called infinite population model because it implies that the number
of terminals in the system is potentially infinite (the probability that K is any large number is nonzero)
and that each terminal generates packets infrequently, so that packet queues are not formed at the
terminals. It is used because it is simple, a good approximation when the number of terminals is large
and provides some important insights.
There are two sets of terminals that may attempt transmission at the beginning of a time slot. Those
that generate new packets, and those whose generated packets have collided in some previous slot and
attempt retransmission. In the latter case we say that the packets are backlogged . Assume that
the system can reach steady state and let M be the random number of packets (newly generated and
backlogged) transmitted in a given slot in steady state. Denote G packets/slot the average value of
M . Since M includes both newly generated and backlogged packets, it holds G > . Observe that a
successful transmission occurs only when M = 1. Indeed, if M 6= 1 then either the slot will be idle (if
M = 0) or there will be a collision in the slot (if M 2). Therefore, by the definition of S we have
S = 0 Pr (M 6= 0) + 1 Pr (M = 1) = Pr (M = 1) .
Hence, if we knew the statistics of M,then we would be able to evaluate S. The exact evaluation of the
statistics of M is complicated. To simplify the situation we make the additional assumption that M is
a Poisson random variable. Since the rate of M is G, we have from (1).

S = Pr(M = 1) = eG G.

(2)

In Figure 5 we plot S as a function of G given by (2). It can be shown that the maximum value of S is
1/e 0.368 and is obtained at G = 1. Hence the maximum channel throughput of the slotted ALOHA
protocol is 0.368 packets/slot. A conspicuous feature of the plot in Figure 5 is that a given channel
throughput is achieved for two values of G a small, G1 , and a large G2 . The small value implies that the
number of backlogged packets is small while for the large value this number is large. Clearly we would
prefer to operate the system at the value G1 , but why two values appear and what is their meaning?
There are two flaws with the analysis presented above. First, the existence of steady state is assumed
and second the probability distribution of M is assumed to be Poisson. For the Infinite Poisson model,
both these assumptions turn out to be invalid! However, the derived bound on the achievable throughput
is sill correct. A more detailed analysis of the system for finite number of users, which is beyond the scope

Figure 5: The Throughput of the ALOHA protocol

of this presentation, reveals that indeed the throughput of the system is at most 1/e. Moreover it can be
shown that the system behaves qualitatively as follows. There are long periods of time during which the
number of backlogged packets in the system remains small and the system operates well inducing small
packet delays. However, from time to time a large increase in the number of backlogged packets in the
system will occur and system performance in terms of throughput and delay will degrade. Fortunately,
it can also be shown that the time interval for the transition from the "good " state to the "bad " is
generally very large. Hence this instability phenomenon of transiting from good to bad states is not
usually a severe problem in real systems.

2.2

Unslotted ALOHA

In the previous section we assumed that the terminals are all synchronized to begin transmission of their
packets at the beginning of each slot. If this feature is unavailable, the protocol can be easily modified to
still operate. Indeed, the users can be allowed to transmit their new packets at packet generation time.
If a collision occurs, then the terminal attempts a retransmission at a later randomly chosen time.
Let us evaluate the performance of the Unslotted ALOHA system. We adopt the infinite population
model and the notation of Section 2.1. Taking into account the cautionary statements at the end of
Section 2.1, let us assume the existence of steady state and that M (), the number of terminal that
attempt transmission in any time interval of length T is a Poisson random variable with rate G.

Figure 6: Possibility of colissions in the Unslotted ALOHA protocol

If terminal a begins transmission at time t , see Figure 6, its transmission will be successful if no
other packet is begins transmission in the interval [t T, t + T ]. Since this interval has length 2T, the
probability that no packet (other than terminal as packet) is transmitted in the interval [t T, t + T ] is
Ps = P (M (2) = 0) = e2G . We can interpret Ps as the proportion of attempted packet transmissions
that are successful. Now, the rate (average number of packets per time T ) by which packet transmissions
are attempted is G and a proportion Ps of these transmissions are successful. Hence the rate of successful
transmissions is
S = GPs = Ge2G

(3)

From (3) it can be seen that the maximum throughput is 1/(2e) and is obtained for G = 1/2.
We see that the throughput of the Unslotted ALOHA is half the throughput of the Slotted one.
However, Unslotted ALOHA does not require terminal synchronization. In any case, from the previous
discussion we see that the throughput of both systems is much lower than one. Throughput one could
be achieved if the terminals could be scheduled for transmission so that collisions are avoided. On the
other hand, we have seen that the ALOHA protocol is very simple and distributed in the sense that the
terminals operate independently of each other and require very small amount of feedback information to
make their decisions. Moreover, the protocol induces very small packet delays when the system is lightly
loaded. The question arises whether the throughput of the ALOHA protocol can be improved, while

maintaining its desirable features. These considerations lead to the development of CSMA protocols
which we discuss in the next section.

CSMA protocols

In this section we present the versions of CSMA protocols that have found wide application. One can
think of the CSMA protocol as an evolution of ALOHA where certain terminal capabilities are exploited in
order to attain improved performance. It turns out that in real systems the required terminal capabilities
depend on the transmission media, i.e., whether communication takes place over wires - twisted pair,
coaxial, optical - or through radio waves in the atmosphere - wireless communication. Accordingly we
first discuss the CSMA and CSMA/CD protocols that are appropriate for wired communications and
next examine the CSMA/CA protocol which is designed for wireless communications.

3.1

The CSMA and CSMA/CD protocols

As we saw in the previous sections, the throughput loss of the ALOHA protocol is due to the fact
that slots are wasted due to collisions or remain idle while there are terminals having packets ready for
transmission. Let us see whether we can improve this situation while maintaining the desirable features
of the ALOHA system. The throughput of the system can be improved if
1. The likelihood of a collision is reduced
2. The time wasted transmitting garbled data when a collision occurs is reduced.
Consider the possibility of reducing collisions first. Let us assume that a terminal is able to listen to
the channel and detect possible ongoing transmissions - busy channel. The ALOHA protocol can then be
modified as follows. In case the terminal finds the channel busy, it defers transmission for a random time.
Else it transmits its own packet. The protocol just described is called Carrier Sense Multiple Access
Protocol. The term Carrier Sense signifies the capability of the terminal to listen to the channel and
find out whether it is busy or not.
At first sight it seems that with CSMA we succeed in avoiding collisions altogether. Indeed, if all
terminals transmit their packets only when the channel is not busy and pick a random retransmission
time if they find the channel busy, then it seems that a collision will occur only when two or more
terminals begin transmission simultaneously, an event that is quite unlikely. However, the situation is
not as rosy as it seems, due to the finite time it takes for a signal to propagate from one terminal to
10

Figure 7: Collision occurence in CSMA protocol

another. Consider the example in figure 7. Assume that it takes seconds for a signal to be transferred
from terminal a to b and vice versa. At time t terminal a sense the channel free and starts transmitting
a packet. At time 1 < terminal b sense the channel and finds it also free, although the packet from
terminal a is well on its way on the channel. Terminal b starts transmitting its own packet, and ( 1 ) /2
seconds later the two packets begin to collide.
From the previous discussion we see that collisions will still occur with the CSMA protocol. However,
we expect that the likelihood of a collision will indeed be reduced if the maximum signal propagation
delay between two terminals in the system is small relative to the length of a packet. Indeed this is the
case. It can be shown that the throughput of the CSMA protocol is approximately, for small /T ,

SCSMA

1
1 + 2 ( /T )1/2

(4)

When T the previous formula shows that S approaches one successful packet per packet duration
time, i.e., the maximum possible.

11

Let us examine (4) more closely. If the length of the packet is B bits and the transmission rate at
the channel is C bits/sec, then T = B/C. Therefore, we can rewrite (4) as

SCSMA

1
1/2

1 + 2 ( C/B)

(5)

The channel propagation time, , is constant and independent of C and B. Therefore, if the network is
extended to cover a wider area and as a result increases, then the throughput will be reduced. Assume
next that we upgrade the channel to a higher transmission rate while maintaining the same arrangement
of terminals (i.e. keep the same). What will happen to the channel throughput? We need to be careful
here since throughput has been defined as the average number of successful packet transmissions per
packet length T, and T changes as C varies and B remains constant. An appropriate measure in this
U
case is the average number of successfully transmitted bits per second. This latter measure SCSMA
is

simply related to SCSMA , namely

SB
C 1/2
= SCSMA C
1/2
T
1/C 1/2 + 2 ( /B)

(6)

From (6) we see that the channel throughput in bits per second increases with C, however, the increase
U
is proportional to C 1/2 and not C. In fact, the throughput per channel transmission rate, i.e., SCSM
A /C
U
is equal to SCSMA , which decreases as C increases. Also, as seen from (6), for constant C, SCSMA

increases as the packet length B increases. These considerations should be taken into account when
deploying networks operating with the CSMA protocol.
We now turn our attention to the possibility of reducing the time wasted to collisions. Assume that
a terminal is able to continue listening to the channel while it transmits its own packet. In case it
detects that collision occurred, it interrupts its own transmission and attempts retransmission at a later
time. Hence in general, if a collision occurs, a time interval smaller that the packet duration time will
be wasted. In the example of Figure 7, terminals b and a will detect the collision at times t + and
t + 1 + respectively. The CSMA protocol where nodes are interrupting their transmissions when a
collision is detected comes by the acronym CSMA/CD protocol - CD stands for Collision Detection. The
throughput of the CSMA/CD protocol for /T small is given approximately by

SCSMA/CD

12

1
.
1 + 5 ( /T )

(7)

Figure 8: Comparison of CSMA and CSMA/CD protocols

Figure 8 shows the throughput of the CSMA and CSMA/CD protocols for various values of = /T .
We see that both protocols can achieve much higher throughput than the original ALOHA system when
is small. In fact the throughput can be close to 1. We also see that for the same , CSMA/CD can
achieve significantly better throughput that CSMA. This improvement if due of course to the fact that
less time is wasted to collisions in CSMA/CD systems than in CSMA.
Up to now we have specified that in case a terminal encounters a collision, it attempts a retransmission
at some later random time. What is a good method of selecting such a random time? We discuss here
one method that has found wide application. Intuitively, the random retransmission time, R, should
depend on the number of the backlogged users: the larger the number of backlogged users, the more
spread-out the distribution of R should be so that the likelihood of avoiding new collisions is reduced. Of
course, R should not be too spread out because then terminals will attempt retransmissions rarely and
a large portion of time will be left unused. In fact this intuition is correct and can be shown that if the
number of backlogged terminals is known and the choice of R is based on this number, the instabilities
of the CSMA protocol can be eliminated. However, in real systems the number of backlogged users is
generally not known. As an alternative, a terminal may try to obtain an estimate of the number of the
backlogged users based on its retransmission history. This estimate should increase as the number of
collisions encountered during the attempt to transmit a packet increases. Hence the distribution of R
should become more spread out as the number of such collisions increases.

13

The previous discussion justifies the following retransmission strategy: if a terminal encounters k
collisions during the attempt to transmit a packet, then it attempts a retransmission at time R which
is uniformly distributed in the time interval (0, A2k ), where A is a constant. There are various variants
of this strategy, however, the main characteristic of all of them is that the spreading of R increases
exponentially with k. For this reason, this retransmission strategy is known as exponential backo.
3.1.1

Applications of CSMA/CD Protocol

The foremost application of the CSMA protocol is in the technology that connects computer terminals
located within a company, an institution, university campus etc., using wires. Such a technology is known
as Local Area Network (LAN) technology. Over the past years there appeared several LAN technologies,
but the first and by far the most prevalent one is the Ethernet technology also referred to as the IEEE
802.3 LAN technology.
The Ethernet technology was developed in the mid seventies by Bob Metcalfe and David Boggs. Since
then, although it faced challenges by several alternative LAN technologies (token ring, FDDI, ATM),
it still dominates the marketplace. One of the reasons for this success is that the hardware required
for its deployment became very cheap, which in turn is due to the large production volume and to the
simplicity of the multiple access protocol used for communication, which is the CSMA/CD protocol
with exponential backo. Moreover, the Ethernet technology proved capable of adapting itself to user
demands for increased transmission rates. Currently, Ethernet LANs run at speeds of 10Mbps, 100Mbps,
and even 1Gbps.

3.2

The CSMA/CA Protocol

The distributed nature of the CSMA protocol and the low delays it induces when the number of active
terminals is small, make it a very attractive candidate for wireless communication. However, certain
restrictions in such an environment do not permit the direct implementation of the protocol.
Let us recall that in order to be able to implement the CSMA/CD protocol, each terminal needs to
be able to perform the following functions.
1. The terminal must be able to listen to the channel and hear whether one or more of the rest of the
terminals in the channel is attempting a transmission - carrier sensing capability.
2. The terminal must be able to listen to the channel while transmitting and detect whether its
transmission collided with the transmission of some other terminals - collision detection capability.
14

The collision detection capability implies that a terminal must be able to transmit and receive at
the same time, which in a wireless environment can be expensive and is often avoided. Hence, the
transmitting terminal may not able to even ensure the correct delivery of its packet. Moreover, as we
will see below, even if the collision detection capability exist, it is still possible that a transmitting station
does not detect a collision while it is transmitting a packet, but the transmission collides at the receiver.
This lack of collision detection capability can be remedied by having the receiver inform the transmitter
that the transmitted packet has been correctly received. To do this, the receiving terminal, upon correct
reception of a packet, sends a short acknowledgement packet back to the transmitter. This packet is
referred to as the ACK message.
Regarding the carrier sensing capability of the terminals, while possible, it is not always sucient to
ensure with high probability that the channel is free of transmissions. To understand this problem we
must expand on the special restrictions imposed in a wireless environment. A characteristic of wireless
transmission is that terminal a can deliver reliably information to b only if b is within a specified distance
from a. Consider now the situation in Figure 9 where we assume that transmissions are symmetric
in the sense that if terminal a can deliver information to b, then b can deliver information to a. The
transmission from terminal a can reach b but not c. The transmission from c can reach b but not a. Using
the standard CSMA protocol in this environment, certain collisions can still be avoided by sensing the
channel. For example if b is transmitting to a, c can sense the ongoing transmission. However, assume
that while a transmits to b, c receives a packet for transmission. If c listens to the channel, it will not
hear as transmission and therefore, if the standard CSMA protocol is employed, a collision will occur.
This problem is known as the hidden terminal problem. Note that in this case, even if a is able to detect
collisions, it will not be able to realize that a collision occurred since it cannot hear c0 s transmission - as
we saw the latter problem is remedied by the use of the ACK message. Due to the restransmission policy
of the basic CSMA protocol, the system can still operate in this environment in spite of the increased
number of collisions, however, system throughput may decrease dramatically if packet sizes are large. In
fact, plain carrier sensing is not always desirable in this environment. To clarify this point consider again
the situation in Figure 9. Suppose that b is sending data to a and c wishes to send data to terminal
d. If c senses the channel, it will find it busy and therefore will defer transmission. However, since cs
transmission cannot reach a, c could in fact deliver its packet to d without colliding with b0 s transmission.
As a result, plain carrier sensing in this case results in reduced utilization of the system. This problem
is know as the exposed terminal problem.
We next provide a mechanism to address the above mentioned problems. Two control signals are
15

Figure 9: The Hidden and Exposed terminal problems.

introduced. These control signals are short messages (compared to packet sizes) that are exchanged
between the transmitter and the receiver before the initiation of packet transmission. The first control
signal is sent by the transmitter to the receiver and indicates that the transmitter is Requesting To
Send (RTS)" a packet. The receiver, upon correct reception of the RTS, replies that it is Clear To
Send (CTS) the packet. Both RTS and CTS signals include a field indicating how long the packet
transmission and the accompanied ACK message will last. The terminals now act as follows.
If a terminal listens to a CTS signal, it waits until the end of the ongoing transmission; this is
known since it is included in the CTS signal. It then waits for a random amount of time and
attempts to initiate its own transmission process.
Let us see how this rule resolves the hidden terminal problem. Assume for the moment that the
transmission of the CTS and RTS signals is instantaneous and let us return to the situation in Figure 9,
where a needs to transmit a packet to b. Terminal a sends an RTS to b and b replies with a CTS signal.
Terminal c receives the CTS signal and knows that a transmission has been initiated, so it defers its
own transmission. Hence the hidden terminal problems is alleviated. In eect the exchange of CTS and
RTS messages act as a virtual carrier sensing mechanism.
In fact, the RTS and CTS signals can also be used to also address the exposed terminal problem.
Assume that we add the following rule.
16

 If a terminal listens to an RTS signal but not a CTS then it goes ahead with its own transmission,
if any.
In Figure 9 assume that b sends an RTS to a and a replies with a CTS. Terminal c hears the RTS
from b but not the CTS from a, and so it knows that its own transmission will not interfere with the b
to a transmission. Hence it can start its own transmission at any time. Therefore the exposed terminal
problem is avoided.
We assumed above that CTS and RTS signals are instantaneous. Of course, as described in Section
3.1, in a real system transmissions do not take place instantaneously and therefore one cannot assume
that the RTS and CTS signals will be received correctly always and free of collisions. However, by now
we know that by imposing appropriate retransmission rules the system can deal with occasion loss of
RTS of CTS signals. The RTS and CTS are useful if packet sizes are large. For small packet sizes it
is preferable to go ahead with the packet transmission rather than incurring the overhead of RTS-CTS
message exchange.
The modified CSMA system whose principles of operation were described above, comes by the name
CSMA/CA, where CA stands for Collision Avoidance. The acronym signifies that collisions are sought
to be avoided and not that they are avoided altogether. Due to the retransmission policy of the CSMA
system, collisions that may occur are not detrimental: in case of collision, the ACK message or RTS
CTS messages will not be received and the transmitting terminal will defer its transmission for a later
time. However, if the propagation delays are relatively large and the system is heavily loaded, collisions
may degrade the performance of the system.
3.2.1

Applications of CSMA/CA Protocol

The principles of the CSMA/CA protocol have been applied to the specification of the MAC protocol
for Wireless Local Area Networks (WLAN), known as the IEEE 802.11 standard1 . Originally the transmission rates of IEEE 802.11 were 1 and 2 Mbps. The IEEE 802.11b extension to this standard specified
5.5 and 11 Mbps transmission rates, while there is ongoing work that will increase the rate to 20Mbps.
There is currently a great interest in the development of WLAN technologies that not only support high
data rates, but also multimedia communication such as video, audio, videoconfenence communication
etc. The support for multimedia communication imposes additional requirements to the network, such
as low packet delays, low packet loss etc. Networks that are able to provide such support are said to
provide Quality Of Service (QOS). CSMA networks were not designed originally to provide QOS. There
1 Currently

the standard does not incorporate a mechanism for dealing with the exposed terminal problem.

17

is a large amount of onging works that either attempt to adapt the CSMA protocol to these additional
requirements or investigate the feasibility of other approaches.

To Probe Further

The literature on the ALOHA and the various variants of the CSMA protocols is huge and is still
expanding. We do not attempt to provide a detailed account of all the works that contributed to the
development of these protocols. Instead we provide some key references to which the interested reader
may turn either for a more in depth study, or for a more comprehensive account of related work.
The book by Rom and Sidi [2] provides an in depth analysis of the ALOHA, CSMA, CSMA/CD and
various other multiple-access protocols. A nice and detailed exposition of the subject can also be found
in the book of Bertsekas and Gallager [3]. Very readable accounts of the protocols can be found in the
books of Tannenbaum [4] and Kurose and Ross [5]. Information on the IEEE 802.3 and IEEE 802.11
standards and related activities can be found in the web site [6].

References
[1] N. Abramson, The Aloha system - another alternative for computer communications, Proc. Fall Joint
Comput. Conf. AFIPS Conf., p 37. 1970.
[2] R. Rom and M. Sidi, Multiple Access Protocols Performance and Analysis, Springer-Verlag, 1990.
[3] D. Bertsekas and R. Gallager, Data Networks, Prentice Hall, 1992, second edition.
[4] A. Tanenbaum, Computer Networks, Prentice Hall, 3rd edition, 1996.
[5] J. F. Kurose, K. W. Ross, Computer Networking, A Top-Down Approach Featuring the Internet,
Addison Wesley, 2001.
[6] http://standards.ieee.org/getieee802/

18