<<

. 22
( 87 .)



>>

When the D1(x) value is greater than zero, the index does not indicate how strong the
correlation is. To measure this second aspect we introduce the D2(x) index:

Th1(0) + Th2(0)
D2(x) =
Th1(x) + Th2(x)

D2(x) compares the throughput of the two sessions when they are active at the same time
and d(2, 3) = x, with respect to the two-session throughput when all the stations are inside
the same transmission range, that is, d(2, 3) = 0. A D2(x) value equal to 1 indicates the
maximum correlation that exists when all stations are in the same transmission range.
By varying the distance d(2, 3) we performed several experiments to estimate the
above indexes. The results were obtained with the card™s transmission rates set to 2 and 11
Mbps, and are summarized in Table 3.7 and Table 3.8, respectively. As is clearly shown in
the tables, the correlation among sessions is still marked when d(2, 3) is less than or equal
to 250 meters, noticeably decreases around 300 meters, and further decreases (but does
not disappear) when the intersession distance is about 350 meters.
From the above results, we assume that 250 m is approximately the size of the physical
carrier sensing range. After this distance, the correlation among the two sessions is due to
the mutual impact on the channel quality. A set of measurements is currently being made
to further verify the exact size of the physical carrier sensing range.
It is worth noting that the physical carrier sensing range is almost the same for the two
different transmission rates. Indeed, the physical carrier sensing mainly depends on two
parameters: the stations™ transmitting power and the distance between transmitting sta-
tions. The rate at which data are transmitted has no significant effect on these parameters.
The results obtained confirm the hypotheses we made in the previous section to jus-
tify the apparent dependencies existing among the two couples of transmitting stations


Table 3.7. Throughput Values (Card Rate = 11 Mbps, Payload Size = 512 bytes)
Throughput of Session 1 Throughput of Session 2
Access Th1( ) Th1(x) Th2( ) Th2(x)
Mechanism Distance Kbps Kbps Kbps Kbps D1(x) D2(x)
x=0 2780 1849 2981 1768 0.37 1.00
x = 150 1950 1500 2950 2250 0.23 0.96
No x = 180 2920 2210 3040 1580 0.36 0.95
RTS/CTS x = 200 2290 1930 3160 2660 0.16 0.78
x = 250 2820 1700 3170 2760 0.25 0.81
x = 300 2980 2800 3060 2750 0.08 0.65
x = 350 2730 2590 3250 3230 0.03 0.62
103
3.5 IEEE 802.11b


Table 3.8. Throughput Values (Card Rate =2 Mbps, Payload Size = 512 bytes)
Throughput Session 1 Throughput Session 2
Access
Mechanism Distance Th1( ) Th1(x) Th2( ) Th2(x) D1(x) D2(x)
x=0 1279 577 1253 561 0.55 1.00
x = 150 1310 880 1310 780 0.37 0.69
No x = 180 1310 930 1310 820 0.33 0.65
RTS/CTS x = 200 1270 1030 1330 1130 0.17 0.53
x = 250 1300 960 1330 960 0.27 0.59
x = 300 1370 1360 1380 1050 0.12 0.47
x = 350 1360 1110 1400 1390 0.09 0.45



even if the distance between them is about three times greater than the transmission
range.
It is worth noting that the ideal value for D1(0) is 0.5, that is, each session gets half of
the throughput of the session in isolation. This is not true for CSMA MAC protocol as
Th1(0) [Th2(0)] is greater than Th1( )/2 [Th2( )/2]. This result is caused by the smaller
overhead of the backoff algorithm in the experiments with d(2, 3) = 0.


3.5.5 Channel Model for an IEEE 802.11 Ad Hoc Network
The results presented in this paper indicate that for correctly understanding the behavior of
an 802.11 network operating in ad hoc mode, several different ranges must be considered.
Specifically, as shown in Figure 3.25, given a transmitting station S, the stations
around it will be affected by the station S transmissions in a different way depending on
the distance from S and the rate used by S for its transmissions. Assuming that S is trans-
mitting with a rate x (x {1, 2, 5.5, 11}), stations around it can be partitioned into three
classes depending on their distance, d, from S:

1. Stations at a distance d < TX_Range(x) are able to correctly receive data from S, if S
is transmitting at a rate lower or equal to x.
2. Stations at a distance d, where TX_Range(x) < d < PCS_Range, are not able to re-
ceive data correctly from station S. However, as they are in the S physical carrier
sensing range, when S is transmitting they observe the channel busy and, thus, they
defer their transmissions.
3. Stations at a distance d > PCS_Range do not measure any significant energy on the
channel when S is transmitting, therefore they can start transmitting contemporarily
to S; however, the quality of the channel they observe may be affected by the energy
radiated by S. In addition, if d < PCS_Range + TX_Range(x), some interference
phenomena may occur (see below). This interference depends on the IF_Range val-
ue. This value is difficult to model and evaluate as it depends on several factors
(mainly the power at the receiving site) but as explained before, TX_Range(1) <
IF_Range < PCS_Range.

Several interesting observations can be derived by taking into consideration points 1 to 3
above. Firstly, the hidden-station phenomenon, as it is usually defined in the literature (see
104 IEEE 802.11 AD HOC NETWORKS: PROTOCOLS, PERFORMANCE, AND OPEN ISSUES




Tx(1)



S
Tx(2)


Tx(5.5)




Figure 3.25. Channel model for an 802.11 ad hoc network.



Section 3.2.2), is almost impossible with the ranges measured in our experiments. Indeed,
the PCS_Range is more than twice TX_Range(1) (the larger transmission range). Further-
more, two stations, say S1 and S2, that can start transmitting toward the same receiver, R,
must be at a distance 2 · TX_Range(1), and thus they are inside the physical carrier
sensing range of each other. Hence, if S1 has an ongoing transmission with R, S2 will ob-
serve a busy channel and, thus, will defer its own transmission. This means that in this
scenario, virtual carrier sensing is not necessary and the RTS/CTS mechanism only intro-
duces additional overhead.
Although the hidden-station phenomenon, as defined in the literature, does not seem
relevant for this environment, point 3 above highlights that packets cannot be correctly re-
ceived because of interference caused by a station that is “hidden” to the sending station.
An example of this type of hidden-station phenomenon is presented in Figure 3.26. In this
figure, we have two transmitting stations, S and S1, that are outside their respective
PCS_Range and, hence, are hidden from each other. In addition, we assume that the re-
ceiver of station S (denoted by R in the figure) is inside the interference range (IF_Range)
of station S1. In this scenario, S and S1 can be simultaneously transmitting and, if this oc-
curs, station R cannot receive data from S correctly. Also in this case, the RTS/CTS mech-
anism does not provide any help and new coordination mechanisms need to be designed
to extend the coordination in the channel access beyond the PCS_Range.
105
3.5 IEEE 802.11b




Figure 3.26. Interference-based hidden-station phenomenon.



It is worth noting that, in our channel model, the exposed-station definition (see Figure
3.6) must be modified, too. In this scenario, exposed stations are those station at a dis-
tance PCS_Range- TX_Range(1) < d < PCS_Range. Indeed, these stations are exposed to
station S transmissions while they are in the transmission range of stations with d >
PCS_Range. The following example outlines problems that may occur in this case. Let us
denote by S1 a station at a distance d from S: PCS_Range < d < PCS_Range +
TX_Range(x). Station S1 can start transmitting, with a rate x, toward a station E that is in-
side the physical carrier sensing area of S; station E cannot reply because it observes a
busy channel due to the ongoing station S transmissions, that is, E is exposed to station S.
Since station S1 does not receive any reply (802.11 ACK) from E, it assumes an error con-
dition (collision or CRC error condition); hence, it back offs and then tries again. If this
situation repeats several times (up to 7), S1 assumes that E is no longer in its transmission
range, gives up the transmission attempt, and (wrongly) signals to the higher layer a link
breakage condition, thus forcing higher layers to attempt a recovery action (e.g., new
route discovery, etc.”see Section 3.3).
To summarize, results obtained in the configuration we analyzed indicate that the hid-
den-station and exposed-station definitions must be extended. These new hidden-station
and exposed-station phenomena may produce undesirable effects that may degrade the
performance of an ad hoc network, mainly if the TCP protocol is used. Extending the co-
ordination in the channel access beyond the PCS_Range seems to be the correct direction
for solving the above problems.

3.5.6 The Communication Gray Zones Problem
An important problem related to the different transmission ranges of control and data
frames is the so-called communication gray zones problem [18]. This problem was re-
106 IEEE 802.11 AD HOC NETWORKS: PROTOCOLS, PERFORMANCE, AND OPEN ISSUES


vealed by a group of researchers at Uppsala University. While measuring the performance
of their own implementation of the AODV routing protocol [22] in an IEEE 802.11b ad
hoc network, they observed an unexpected large amount of packet losses, especially dur-
ing route changes. They found that the increase in packet loss occurred in some specific
geographic areas that they called “communication gray zones.” In such zones, the packet
loss experienced by a station may be extremely high”up to 100%”thus severely affect-
ing the performance of those applications characterized by a continuous packet flow (e.g.,
file transfers and multimedia streaming). They also found that the ultimate reason for this
phenomenon is that a station inside a gray zone is considered as reachable by a neighbor-
ing station, based on its routing information, but data communication between the stations
is not possible. The same problem was found to affect other routing protocols like OLSR
[5] and LUNAR [25].
To better understand why communication gray zones arise it is worthwhile to briefly
recall how the AODV routing protocol works. AODV is a reactive protocol that discovers
and maintains routes on demand. When a route to a target station is needed, the AODV
protocol broadcasts a route-request message that is then disseminated throughout the net-
work. When the target station (or a valid route to the target station) is found, a route-reply
message is sent back to the requesting station by means of a unicast message. While this
message travels towards the requesting station, routes are set up inside routing tables of
the traversed stations.
In addition to the request“reply mechanism, the AODV protocol uses a sensing mecha-
nism to discover neighboring stations and, based on this, to update, add, or remove routes
in the routing table. Periodically, each station broadcasts HELLO beacons. Upon recep-
tion of a HELLO message from a neighbor, a station becomes aware that the neighboring
station is reachable and can, thus, be used to relay data transmissions. Routing tables are,
thus, updated accordingly.
Several elements contribute to the occurrence of communication gray zones. In partic-
ular, the different properties of HELLO messages with respect to data messages play an
important role. These properties, and their effects, are summarized below.

1. Transmission rate. Since HELLO beacons are broadcast messages, they are trans-
mitted at the basic rate (2 Mbps). On the other hand, data packets (which are uni-
cast) may be transmitted at 11 or 5.5 Mbps. Therefore, HELLO messages have a
transmission range larger than data messages.
2. No Acknowledgment. In 802.11b, broadcast messages are transmitted without ac-
knowledgement. Therefore, a station that receives a HELLO message from a neigh-
boring has no indication whether transmission is possible even in the opposite di-
rection, that is, there is no indication that the link is bidirectional.
3. Packet size. In general, HELLO messages are much smaller in size than data pack-
ets. As is well known, small packets have a lower probability of being affected by
transmission errors, and minor chances of colliding with other packets. Therefore, it
is more likely for a HELLO message to reach a receiver than a data packet, espe-
cially when the link quality is poor.

In addition to the above elements, the effects of fluctuating links need to be taken into
account as well. At the border of the transmission range, the communication quality tends
to be fluctuating. Under such conditions, it may happen that a station sporadically re-
107
3.5 IEEE 802.11b


ceives a HELLO message from a neighbor, but this does not imply that consistent com-
munication between the stations is actually possible. Since the AODV protocol updates
routing tables based on the neighboring sensing mechanism (i.e., based on the reception
of a HELLO message) it may occur that stable and longer routes are replaced by shorter
but unreliable ones.
Figure 3.27 depicts a scenario pointed out by the researchers at Uppsala University, in
which communication gray zones can be experienced by the mobile station MS (see [18]).
In this scenario, stations labeled as GW, FS1, and FS2 are static, while station MS moves
forward and back as indicated in the figure. There is an active communication between the
gateway Station GW and the mobile station MS. Depending on the physical position of
the mobile station, the traffic from MS to GW (and vice versa) is routed trough one, two,
or three hops via intermediate stations FS1 and FS2. Theoretically, the MS always has a
route toward GW. However, while moving from the initial position to the rightmost posi-
tion, MS will pass through two gray zones. Similarly, two gray zones will be traversed in
the reverse path. In [18] it is shown that traversal of the gray zones is associated with time
intervals during which MS experiences a packet loss of up to 100%. The duration of this
time interval, as well as the peak value in the packet loss experienced by the mobile sta-
tion, depends on the specific routing protocol.
Before proceeding, it is important to highlight that the communication gray zone prob-
lem cannot be revealed by using the current simulation tools (e.g., ns-2). Indeed, in the
IEEE 802.11 model implemented by simulation tools, both unicast and broadcast trans-
missions are performed at 2 Mbps and, hence, they have the same transmission range.
Furthermore, connectivity is modeled as on/off, that is, the communication becomes im-
possible as soon as the distance exceeds the transmission range.
In [18] the authors also propose some possible solutions for alleviating the communi-
cation gray zone problem, namely: (i) the exchange of the neighboring set (i.e., stations
include their neighboring set in HELLO messages); (ii) the transmission of N-consecutive
HELLO messages; and (iii) the introduction of a SNR threshold to discard weak control
messages. They have also assessed, by means of experimental analysis, that the SNR-
threshold approach is the most effective and it eliminates the effects of communication
gray zones almost completely.




Figure 3.27. A scenario in which communication gray zones can be experienced (taken from [18]).
108 IEEE 802.11 AD HOC NETWORKS: PROTOCOLS, PERFORMANCE, AND OPEN ISSUES


3.6 EVOLUTION OF IEEE 802.11b FOR AD HOC NETWORKS

In ad hoc networks, each station logically operates similarly to a router. However, from the
physical standpoint, there is a significant difference between a router and a station in an
ad hoc network. Typically, a router has multiple network interfaces, and a packet received
from one interface is retransmitted through a different interface (see the left side of Figure
3.28). On the other hand, in a multihop ad hoc network a station has a single wireless in-
terface and packets are received from and transmitted through the same interface (see the
right side of Figure 3.28).

<<

. 22
( 87 .)



>>