A Multichannel CSMA MAC Protocol for Multihop Wireless Networks

Asis Nasipuri, Jun Zhuang, and Samir R. Das

Division of Computer Science
The University of Texas at San Antonio
San Antonio, TX 78249-0667
E-mail: nasipuri,jzhuang,samir@cs.utsa.edu

Abstract We describe a new carriersense multiple access

(CSMA) protocol for multihop wireless networks, sometimes
also called ad hoc networks. The CSMA protocol divides the
available bandwidth into several channels and selects an idle
channel randomly for packet transmission. It also employs a
notion of soft channel reservation as it gives preference to
the channel that was used for the last successful transmission. We show via simulations that this multichannel CSMA
protocol provides a higher throughput compared to its single channel counterpart by reducing the packet loss due to
collisions. We also show that the use of channel reservation
provides better performance than multichannel CSMA with
purely random idle channel selection.
1

I NTRODUCTION

We consider multihop wireless networks where there is no

cellular infrastructure (such as base stations). Nodes communicate via multihop wireless links. Nodes can be mobile
and dynamic routing protocols are used to establish routes
between a pair of communicating nodes. In literature, terms
such as packet radio networks or ad hoc networks have also
been used to describe such networks. Such networks are
very useful in military and other tactical applications such
as law enforcement, emergency rescue or exploration missions, where cellular infrastructure is unavailable or unreliable. There is considerable interest in using ad hoc networks
in commercial applications as well where there is a need for
ubiquitous communication services without the presence of
a fixed infrastructure.
A central challenge in the design of Medium Access Control (MAC) protocols for such wireless networks has been
to reduce the impact of the so-called hidden terminal problem [4], where a source cannot hear the transmission from
another, distant source and starts transmission assuming the
medium to be free, while the transmitted packets collide
at the receiver that may hear the transmissions from both
sources. A reservation based mechanism using requestto
send/cleartosend (RTS/CTS) packets [4, 3] is commonly
used to address this problem. This solution is also adopted
in the new IEEE standard 802.11 [5] for wireless LANs.
Several other, more recent, protocols also use some form of
RTS/CTS exchanges. Examples include FAMA [6], GAMA
[13], CARMA [7], etc. However, the RTS and CTS packets
themselves are broadcast packets sent using carrier sensing at
the sender and can collide at the receiver(s) due to the hidden
terminal problem. Even though the RTS and CTS packets are

usually short, the problem can be severe at high loads.

Even if RTS/CTS exchanges were free from ill, all packet
transmissions in a multihop wireless network cannot solely
rely on RTS/CTS exchanges. These include broadcast transmissions frequently used by the dynamic routing protocols.
Such transmissions are intended to reach all neighbors. For
example, broadcast floods are used to update routing tables of
all nodes in a portion of the network (e.g., link-state or distance vector protocols [10]) or to discover new or alternate
routes (e.g., on-demand protocols [9, 15]). Such transmissions cannot use channel reservation and must depend on a
pure CSMA technique [11] for channel access.
Our goal in this paper is to investigate a new multiple
channel CSMA protocol that can reduce the effect of the hidden terminal problem. The protocol selects channels dynamically and employs a soft channel reservation. The idea
is somewhat similar to frequency-division multiple access
(FDMA) schemes used in cellular systems. The major difference is that there is no central infrastructure and thus the
channel assignment is done in a distributed fashion via carrier
sensing much as in a traditional CSMA scheme. Use of carrier sensing to perform channel assignment also distinguishes
it from the traditional broadcast scheduling problem [18, 17]
in a spatially disperse packet radio network, where channel
assignment is performed via a central control or via additional message communication and synchronization. None
of them are viable in an ad hoc setting.
2

M ULTICHANNEL CSMA P ROTOCOL

Multiple channel CSMA protocols are not entirely new.

They were shown to be more efficient than their single channel counterparts in wired LANs [2]. Packet transmission over
multiple random access channels in wireless networks was
also explored in [16], where multi-channel slotted ALOHA
[1] was analyzed. A multichannel MAC protocol was presented in [8] for application in a class of ad hoc networks
termed as Reconfigurable Wireless Networks. Some protocols have been proposed for wireless LANs [14, 19] where
each host is allowed to transmit in a unique frequency channel. However, there has not been any demonstration of improved throughput of wireless networks by breaking up the
available wireless medium into several channels and using a
suitable access protocol for reducing the probability of collisions, which is the goal of this paper.

We extend the concept presented in [2] to the wireless scenario to design a new multi-channel CSMA scheme that provides soft channel reservation. In [2], it is shown that for
a wired LAN with channels, use of carrier sensing to randomly select one of the idle channels for transmission has
a throughput advantage that increases with . This is attributed to the reduction of the normalized propagation delay
per channel, which is defined as the ratio of the propagation
time over the packet transmission time. The bandwidth per
channel decreases with larger , thereby decreasing the normalized propagation delay. Hence, the probability of multiple stations sensing the channel to be idle and choosing to
transmit at mutually overlapping times, decreases with .
This, in turn, reduces the probability of collisions.
However, wireless networks follow a different mechanics
than wired networks. Here, lower channel bandwidth and
faster propagation speed typically result in a much smaller
normalized propagation delay as compared to wired networks. Collisions, however, occur for a very different reason. Signal strength reduces with distance and thus it is possible that some nodes in the network cannot hear each other
sufficiently well, the signal strength being below the carrier
sensing threshold. This causes transmitting nodes to be hidden from other transmitting nodes, but still can cause sufficient interference at the receiver for packets to be lost due
to collisions. Signal strengths at the transmitter and receiver
being different, the transmitter is never in a position to detect collisions. In addition, the combination of even weak
signals from many transmitters can raise the overall interference high enough to cause collisions. One of our goals in
the paper is to demonstrate that in spite of these differences,
multichannel CSMA protocols can reduce collisions significantly in wireless networks, albeit for a different reason than
wired nets. We also design a new mechanism to do channel
reservation, which provides additional benefits.
Our scheme breaks up the total available bandwidth into

nonoverlapping channels, where may be much smaller

than the number of hosts. The channels may be created in

the frequency domain (FDMA) or code domain (CDMA),
though we will use frequency domain in our description in
this paper. Due to the absence of network-wide synchronization in such networks, we assume that TDMA is not used.
The radio transmitter and receiver at every host are assumed
to be able to operate in any one of these channels. Note that
each channel has a bandwidth of , where is the total
available bandwidth for communication. Typically, a transmitter tries to reuse the channel it used in its last successful
transmission. In case this reserved channel is busy (determined by carrier sensing) or the most recently used channel
resulted in an unsuccessful transmission, another free channel is selected at random. A back-off and retry strategy is
employed in case there is no free channel.
The detailed protocol operations are now described. The
protocol is described as a multichannel variation of the basic
CSMA/CA (CSMA with collision avoidance) protocol used
in the IEEE 802.11 standard [5] for wireless LANs.
1. Each node monitors the channels continuously,
whenever it is not transmitting. It detects whether or not

the total received signal strength (TRSS) in the channels

are above or below its sensing threshold (ST). The channels for which the TRSS is below the ST, are marked
as IDLE. The time at which the TRSS dropped below
ST is noted for each channel. These channels are put
on a free channel list. The rest of the channels are
marked as BUSY.
2. At the start of a protocol cycle, i.e., when a packet arrives from the traffic generator:
(a) If the free channel list is empty, the node
waits for the first channel to be IDLE. Then it
waits for a period called the Long Interframe
Space (LongIFS), and it waits further for a random access backoff period before transmitting the
packet. It is required that the channel remains
IDLE during this period.
(b) If the free channel list is not empty, the
node checks if the channel that it used successfully in the most recent past, last channel,
is included in the list. If the last channel
is IDLE, then the node chooses this channel for
data transmission in the current protocol cycle.
Else, the node randomly selects a channel from
the free channel list using a uniform random
number generator.
3. Before actually transmitting the packet the node checks
to see whether the TRSS on the chosen channel has remained below ST for at least a LongIFS period.
(a) If not, the node initiates a backoff delay after the
LongIFS.
(b) If yes, then the node initiates transmission immediately, without further delay.
4. Any backoff is canceled immediately if the TRSS on the
chosen channel goes above the ST at any time during the
backoff period. When TRSS again goes below ST a new
backoff delay is scheduled.
5. After the end of a successful transmission (as indicated by an acknowledgement) last channel is set
to the channel used. Else last channel is undefined and a random channel will be chosen from the
free channel list for the retransmission.
When the number of channels, , is sufficiently large, the
above protocol tends to reserve a channel for data transmission for every node. This channel reservation technique minimizes occasions when two contending transmissions happen
to choose the same channel. On the other hand, the flexible
nature of soft reservation allows for using other free channels. It is expected that hosts will tend to dynamically select
free channels in a mutually exclusive fashion so as to enable parallel, interferencefree transmission. Even in heavy
traffic conditions, when the number of channels may not be
sufficient for conflictfree transmissions, the chance of collisions will be reduced because of persistence of every node to
use a reserved channel for itself.

S IMULATION M ODEL

We developed an event driven simulator particularly

suited for MAC layer performance evaluations for wireless networks. The simulator uses a network of wireless nodes, placed in a square grid. In the data we
present, the nodes are stationary. Traffic is Poisson with a
random source node with a randomly chosen destination located within its radio range. The radio range is determined
using the transmitter power, a propagation or path loss model
and the signal sensing threshold (ST). An indoor propagation
model is used to evaluate the path loss between a given pair
of nodes. It uses a piecewise log-log function in which the
dB path loss is assumed to vary linearly with the log of the
distance between the source and the destination. Multipath
fading is simulated as a separate loss component, generated
randomly for each packet transmission and is assumed to remain constant for the duration of the packet.
Each time a node commences or ends transmition, the
simulator evaluates the received signal strength (RSS) in the
chosen channel at each node location. Two components of
the RSS are computed at each non-transmitting node for every channel: (a) the strength of the desired signal, and (b)
the total signal strength (obtained by summing the RSS contributions from all nodes currently transmitting packets in
that channel). From the two basic RSS components, an interference RSS component is computed as the difference between the total and desired signal strengths. The desired RSS
and the interference RSS yield a signal-to-interference ratio
(SIR). The SIR is recomputed for every receiving node every
time there is a change in the RSS on the corresponding channel. A packet is assumed to be received correctly, whenever
the SIR stays above the specified minimum SIR threshold for
the entire duration of the packet. This essentially implements
a power capture model.
In our experiments, we considered two types of receiver
operations. In the first, the receiver is assumed to be able
to receive only one packet at a time. This receiver, which
we will call the single user (SU) receiver, receives only the
first incoming packet, subject to the usual capture model, and
drops any overlapping packet. In the second type, we assume
a multiuser (MU) receiver, which can receive multiple packets simultaneously as long as they are received on separate
channels and each meet the power capture constraints. Multiuser receivers have been studied extensively for CDMA in
[12].
4

250
N=1
N=5 SU-reservation
N=10 SU-reservation
N=10 MU-reservation
N=20 MU-reservation
N=10 SU-random
N=20 MU-random

200

Throughput (Kbits/sec)

150

100

50

0
0

100

200

300
400
500
600
Total offered load (Kbits/sec)

700

800

Figure 1: Throughput performance of a 225 node network

configuration with parameters as given in Table 1.
vation for =1, 5, 10, and 20 channels. These are shown
in Fig. 1. Also plotted in Fig. 1 are the throughput curves
for a multichannel CSMA protocol without channel reservation, where the source simply picks a channel at random
from the set of idle channels. Note that for the single-user
receiver with channel reservation, the throughput increases
from =1 to =5. The improvement is negligible for more
channels. With the multiuser receiver with channel reservation, the throughput increases through 20 channels. As expected, the throughput for the multiuser receiver is higher
than that for the single-user receiver for the same number of
channels.
The throughput with multichannel CSMA using the random channel selection scheme is also better than that of the
single channel scheme. However the performance is poorer
than the multichannel schemes using soft channel reservation. This shows the benefit of channel reservation.
Fig. 2 gives further insight into the performance of the
multichannel MAC protocols by showing the breakdown of

S IMULATION R ESULTS

We ran simulations to measure the throughput in the network. Separate runs were done for different number of channels, all with the same total bandwidth . The parameters
used for the simulation are shown in Table 1.
With these parameters, the maximum number of neighbors (nodes that are within transmission range) for any node
in the network is found to be 13. Throughput curves were
obtained for both the single user and multiuser receiver models using the multichannel CSMA protocol with soft reser-

Table 1: Parameters used for the simulations in Fig. 1

Parameter
No. of nodes
Grid size
Transmitter power
Signal sense threshold (ST)
Minimum SIR
Packet size
Long IFS
Total bandwidth

Values used
225
200 m
35 dBm
10 dBm
10 dBm
5000 bytes
40 sec
1 Mb/sec

30000
25000

1111Transmitted
0000
due to collisions
0000Unsuccessful
1111
Unsuccessful due to destination busy

400
N=1
N=5, SU-reservation
N=10, SU-reservation
N=10, MU-reservation
N=20, MU-reservation
N=10 SU-random
N=20 MU-random

1
0
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1

15000
10000
5000
0

N=

1
0
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1

1
0
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
d

1
0
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
ed

1
0
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
ed

1
0
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
om

1
0
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
m

ed
e
o
erv
erv
nd
erv eserv
nd
res
res
-ra U-ra
res
r
UUSU
US
M
M
M
0
1
10
10
20
20
N=
N=
N=
N=
N=

300

Normalized delay

Packet counts

350

20000

250

200

150

U-

S
=5

100

Figure 2: Packet counts for the 4th simulation points of the

plots in Figure 1.

50

0
0

all packets transmitted. This data correspond to middle

points from Fig. 1. The total number of unsuccessful transmissions is largest for the case =1 and decreases successively from left to right for the protocols with channel reservation. Packet transmissions are unsuccessful due to primarily two reasons - the destination busy situation, and due
to collisions. A single user receiver is busy whenever it is
transmitting or receiving a packet on any channel. A multiuser receivers is unavailable only when it is currently transmitting or receiving on the same channel on which the new
incoming packet is transmitted. Collisions occur when the
received SIR falls below the required minimum SIR threshold.
Note that with the single user receiver, the number of
packets unsuccessful due to the destination busy condition
increases with . This is attributed to the fact that the
bandwidth-per-channel decreases with increasing , causing
the nodes to take proportionately longer times for transmission and reception. However, it is important to note that collisions decrease with increasing using the proposed soft
reservation protocol. Note that between the cases =1 and
=5 with reservation, the reduction of collisions is greater
than the increase in unsuccessful transmissions due to a busy
destination. This benefit is lost, however, beyond
. On
the other hand, since the multiuser receiver can receive multiple packets concurrently, it loses fewer packets due to the
destination busy condition. With random channel selection,
collision counts are higher than that with channel reservation.
This is due to the situations where hidden transmitters are
choosing the same idle channel in the absence of any reservation. This possibility is reduced with channel reservation.
We next estimate the average packet transmission delay
using the multichannel CSMA protocol. In our simulations,
acknowledgements of successful packet reception and retransmission of unsuccessful packets is not yet implemented.
Thus, we evaluate the average delay in successful packet
transmission by following the ideas presented in [11]. The
average packet delay , normalized to the time for transmission of one packet using the whole bandwidth, is determined

50

100
150
Throughput (Kbits/sec)

200

250

Figure 3: Delays of the MAC protocols considered in Fig. 1,

normalized to the transmission time of a packet over a single
channel. The average normalized retransmission delay is
taken as 50.
from

(1)
is the offered load, is the corresponding through

where
put, is the normalized average delay between successive
retransmissions, and is the normalized propagation time.
depends on the packet transmission time, the round trip
propagation delay between the source and the destination,
the transmission time of the acknowledgement packet (),
and the average retransmission delay ( ). Hence

(2)

For our scenario we assume to be negligibly small and exclude it from our calculations. Also, for simplicity we assume that the acknowledgement packets are much smaller
than the data packets and that they are sent over an ideal
channel with negligible delay. Hence, we use
,
where we consider a fixed average retransmission delay for
all to assess the performance of the multichannel protocols. Using the offered load (G) and corresponding throughputs (S) from in Fig 1, we use Eq. 1 to derive the average
transmission delays for each protocol. These are shown in
Fig. 4. Note that the average delay is higher for the multichannel protocols at low traffic loads. This is due to the
smaller bandwidth per channel with higher . However, the
delay is lower at high traffic loads, as the number of retransmissions is smaller compared to the single channel case.
5

C ONCLUSIONS

We have demonstrated that distinct advantages in throughput can be gained in a CSMA protocol by segregating the

available bandwidth into multiple channels, and using the

carrier sense information for selecting idle channels with a
soft channel reservation. This reservation based scheme
performs better than a multichannel scheme with random selection of idle channels. This performance advantage is in
spite of the lower per channel bandwidth. However, a large
number of channels may cause an unacceptably high packet
transmission time. This in turn affects the throughput for single user receivers and the average delays for both single and
multiuser receivers. Our simulation shows that a handful of
channels works very well. Our future work will focus on
more elaborate performance evaluation with a focus on the
determination of the optimal number of channels.
6

ACKNOWLEDGMENT

This work is partially supported by AFOSR grant no.

F49260-96-1-0472, Texas Advanced Technology Program
grant no. 010115-248b, NSF MII grant no. CDA-9633299
and NSF CAREER award no. ACI-9733836.
R EFERENCES
[1] N. Abramson. The ALOHA system - another alternative for computer communication. In 1970 Fall Joint
Computing Conference Proceedings, pages 281285,
1970.
[2] M. Ajmone-Marsan and D. Roffinella. Multichannel
local area network protocols. IEEE Journal on Selected Areas of Communication, 1:885897, 1983.
[3] F. Cali, M. Conti, and E. Gregori. IEEE 802.11 wireless lan: Capacity analysis and protocol enhancement.
In Proceedings of IEEE INFOCOM98, pages 142
149, 1998.
[4] H. S. Chhaya and S. Gupta. Performance modeling
of asynchronous data transfer methods of IEEE 802.11
MAC protocol. In Proceedings of IEEE Personal Communications Conference, pages 815, October 1996.
[5] IEEE Standards Department. Wireless LAN medium
access control (MAC) and physical layer (PHY) specifications, IEEE standard 802.111997, 1997.
[6] R. Garces and J. J. Garcia-Luna-Aceves. Floor acquisition multiple access with collision resolution. In
Proceedings of the ACM/IEEE Mobile Computing and
Networking Conference, pages 1012, November 1996.
[7] R. Garces and J. J. Garcia-Luna-Aceves. Collision
avoidance and resolution multiple access with transmission queues. ACM Wireless Networks Journal, 1998.
[8] Z. J. Haas. On the performance of a medium access control scheme for the reconfigurable wireless networks. In Proceedings of IEEE MILCOM97, November 1997.

[9] D. Johnson and D. Maltz. Dynamic source routing in ad hoc wireless networks. In T. Imielinski
and H. Korth, editors, Mobile computing. Kluwer Academic, 1996.
[10] S. Keshav. An Engineering Approach to Computer Networking : ATM Networks, the Internet, and the Telephone Network, chapter 11. Addison-Wesley, 1997.
[11] L. Kleinrock and F. A. Tobagi. Packet switching
in radio channels: Part1 - carrier sense multiple
access modes and their throughput-delay characteristics. IEEE Transactions in Communications, COM23(12):14001416, 1975.
[12] R. Lupas and S. Verdu. Near-far resistance of multiuser detectors in asynchronous channels. IEEE
Transactions in Communications, COM-38(4):496
508, 1990.
[13] A. Muir and J. J. Garcia-Luna-Aceves. Supporting
real-time multimedia traffic in a wireless LAN. In Proceedings of the SPIE Multimedia Computing and Networking Conference, February 1997.
[14] A. Muir and J. J. Garcia-Luna-Aceves. A channel access protocol for multihop wireless networks with multiple channels. In Proceedings of the IEEE International Conference on Communications, pages 1617
1621, 1998.
[15] Charles Perkins and Elizabeth Royer. Ad hoc ondemand distance vector routing. In Proceedings of the
2nd IEEE Workshop on Mobile Computing Systems and
Applications, pages 90100, Feb 1999.
[16] R. Prasad, M. G. Jansen, and J. P. van Deursen.
Multiple-frequency slotted aloha in a shadowed and
rician faded radio environment. Electronic Letters,
29(21):18281829, October 1993.
[17] R. Ramanathan and E. L. Lloyd. Scheduling algorithms for multihop radio networks. IEEE/ACM Transactions on Networking, 1:166177, 1993.
[18] R. Ramaswami and K. K. Parhi. Distributed scheduling of broadcasts in radio networks. In Proceedings of
IEEE INFOCOM89, pages 496504, 1989.
[19] Z. Tang and J. J. Garcia-Luna-Aceves.
Hopreservation multiple access (HRMA) for multichannel
packet radio networks. In Proceedings of the IEEE
IC3N98, Seventh International Conference on Computer Communications and Networks, October 1998.