<<

. 21
( 87 .)



>>




1

3/12/2002
6/12/2002
0.8
Packet Loss




0.6


0.4


0.2


0
60 80 100 120 140 160

Distance (meters)

Figure 3.19. 1 Mbps transmission ranges on different days.
97
3.5 IEEE 802.11b


Table 3.6. Estimates of the Transmission Ranges at Different Data Rates
11 Mbps 5.5 Mbps 2 Mbps 1 Mbps
Data TX_range 30 meters 70 meters 90“100 meters 110“130 meters
Control TX_range 90 meters 120 meters



Again, it is interesting to compare the transmission range used in the most popular
simulation tools, like ns-2 and Glomosim, with the transmission ranges measured in our
experiments. In these simulation tools it is assumed TX_range = 250 m. Since the above
simulation tools only consider a 2 Mbps bit rate, we make reference to the transmission
range estimated with a NIC data rate of 2 Mbps. As is clearly shown, the value used in the
simulation tools (and, hence, in the simulation studies based on them) is two to three times
higher that the values measured in practice. This difference is very important, for exam-
ple, when studying the behavior of routing protocols: the shorter is the TX_range, the
higher is the frequency of route recalculation when the network stations are mobile.

3.5.2.1 Transmission Ranges and the Mobile Devices™ Height. During the ex-
periments we performed to analyze the transmission ranges at various data rates, we ob-
served a dependence of the transmission ranges on the mobile devices™ height from the
ground. Specifically, in some cases we observed that although the devices were not able to
communicate when located on stools, they started to exchange packets when they were
lifted up. In this section, we present the results obtained by a careful investigation of this
phenomenon. Specifically, we studied the dependency of the transmission ranges on the
devices™ height from the ground. To this end we measured the throughput between two sta-
tions9 as a function of their height from the ground. Four different heights were consid-
ered: 0.40 m, 0.80 m, 1.2 m, and 1.6 m. The experiments were performed with the Wi-Fi
card set at two different transmission rates: 2 and 11 Mbps. In each set of experiments, the
distance between the communicating devices was set in such a way as to guarantee that
the receiver was always inside the sender™s transmission range. Specifically, the
sender“receiver distance was equal to 30 and 70 meters when the cards operated at 11 and
2 Mbps, respectively.
As clearly shown in Figure 3.20, the height may have a big impact on the quality of the
communication between the mobile devices. For example, at 11 Mbps, by lifting up the
devices from 0.40 meters to 0.80 meters, the throughput doubles, whereas further increas-
ing the height does not produce significant throughput gains. Similar behavior is observed
with a 2 Mbps transmission rate. However, in this case, the major throughput gain is ob-
tained by lifting up the devices from 0.80 meters to 1.20 meters. A possible explanation
for this different behavior is related to the distance between the communicating devices
that is different in the two cases. This intuition is confirmed by the work presented in [21],
which provides a theoretical framework to explain the height impact on IEEE 802.11
channel quality. Specifically, the channel power loss depends on the contact between the
Fresnel zone and the ground. The Fresnel zone for a radio beam is an elliptical area with
foci located in the sender and the receiver. Objects in the Fresnel zone cause diffraction
and, hence, reduce the signal energy. In particular, most of the radio-wave energy is with-
in the first Fresnel zone, the inner 60% of the Fresnel zone. Hence, if this inner part con-

9
In these experiments, UDP is used as the transport protocol.
98 IEEE 802.11 AD HOC NETWORKS: PROTOCOLS, PERFORMANCE, AND OPEN ISSUES




Figure 3.20. Relationship between throughput and devices™ height.


tacts the ground (or other objects), the energy loss is significant. Figure 3.21 shows the
Fresnel zone (and its inner 60%) for a sender“receiver couple at a distance D. In the fig-
ure, R1 denotes the height of the first Fresnel zone. As shown in [21], R1 is highly depen-
dent on the station™s distance. For example, when the sender and the receiver are at an
height of 1 meter from the ground, the first Fresnel zone has contact with the ground only
if D > 33 meters, whereas at heights of 1.5 and 2 meters, the first Fresnel zone contacts
the ground only if D is greater than 73 and 131 meters, respectively. These theoretical
computations are in line with our experimental results.
99
3.5 IEEE 802.11b




Figure 3.21. The Fresnel zone.



3.5.3 Four-Stations Network Configurations
The results presented in the previous sections show that the IEEE 802.11b behavior is
more complex than the behavior of the IEEE 802.11 standard. Indeed, the availability of
different transmission rates may cause the presence of several transmission ranges inside
the network. In particular, inside the same data transfer session there may be different
transmission ranges for data and control frame (e.g., RTS, CTS, ACK). Hereafter, we
show that the superposition of these different phenomena makes it very difficult to under-
stand the behavior of IEEE 802.11b ad hoc networks. To reduce this complexity, in the ex-
periments presented below the NIC data rate is set to a constant value for the entire dura-
tion of the experiment.10 Hereafter, we present only the results obtained with the NIC data
rate constant and equal to 11 Mbps; more results can be found in [3].
The four-stations configuration presented in Figure 3.22 was used in the experiments.
The results obtained are presented in Figure 3.23.
These results were the superposition of several factors. In detail, dependencies were
observed between the two connections, even though the transmission range was smaller
than the distance between stations S1 and S3. Furthermore, the dependency were ob-
served also when the basic mechanism (i.e., no RTS/CTS) was used.11 To summarize, this
set of experiments showed that interdependencies among the stations extends beyond the
transmission range. To explain this, we hypothesized that all stations were inside the same
physical carrier sensing range, and this produced a correlation between active connections
whose effect is similar to that achieved with the RTS/CTS mechanism (virtual carrier
sensing). The difference in the throughputs achieved by the two sessions when using the
UDP protocol (with or without RTS/CTS) can be explained by considering the asymmet-
ric condition that exists on the channel: station S2 was exposed to transmissions of station
S3 and hence, when station S1 sent a frame to S2, this station was not able to send back
the MAC ACK. Therefore, S1 reacted as in the collision cases (thus rescheduling the
transmission with a larger backoff). It is worth pointing out that also S3 was exposed to S2
transmissions but the S2™s effect on S3 was less marked given the different role of the two
stations. When using the basic access mechanism, the S2™s effect on S3 was limited to
short intervals (i.e., the transmission of ACK frames). When adopting the RTS/CTS

10
It is worth pointing out that we experienced a high variability in the channel conditions, thus making a com-
parison between the results difficult.
11
A similar behavior is observed (but with different values) by adopting the RTS/CTS mechanism.
100 IEEE 802.11 AD HOC NETWORKS: PROTOCOLS, PERFORMANCE, AND OPEN ISSUES



Session 1 Session 2


S1 S2 S3 S4

25 m
25 m 80/85 m

Figure 3.22. Network configuration at 11 Mbps.



UDP
2500



2000
Throughput [Kbps]




1500
12
34
1000



500



0
no RTS/CTS RTS/CTS



TCP
2500



2000
Throughput [Kbps]




1500
12
34

1000



500



0
no RTS/CTS RTS/CTS

Figure 3.23. Throughputs at 11 Mbps.
101
3.5 IEEE 802.11b


mechanism, the S2 CTS forced S3 to defer the transmission of RTS frames (i.e., simply a
delay in the transmission), whereas RTS frames sent by S3 forced S2 to not reply with a
CTS frame to S1™s RTS. In the latter case, S1 increased the backoff and rescheduled the
transmission. Finally, when the TCP protocol was used, the differences between the
throughput achieved by the two connections still existed but were reduced. The analysis of
this case is very complex because we must also take into consideration the impact of the
TCP mechanisms that (i) reduce the transmission rate of the first connection, and (ii) in-
troduce the transmission of TCP-ACK frames (from S2 and S4), thus contributing to mak-
ing the system less asymmetric.

3.5.4 Physical Carrier Sensing Range
Results presented in the previous section seem to indicate that dependencies among the
stations extend far beyond the transmission range. For example, taking as a reference
the scenario presented in Figure 3.22, the distance between the two couples of transmit-
ting stations is about three times the transmission range. The hypothesis is that depen-
dencies are due to a large physical carrier sensing that includes all the stations. To vali-
date this hypothesis and to better understand the system behavior, we designed some
experiments to estimate the physical carrier sensing range. A direct measure of this
quantity seems difficult to achieve because the 802.11b cards we utilized do not provide
to the higher layers information about the channel carrier sensing. Therefore, we defined
an indirect way to perform these measurements. We utilized the scenario shown in
Figure 3.24 with fixed distance between each coupled communicating stations [d(1, 2) =
d(3, 4) = 10 meters], and variable distance between the two couples, that is, d(2, 3) is
variable.
The idea is to investigate the correlation among the two sessions while increasing the
distance d(2, 3). To measure the correlation degree, just before running each experiment
we performed some preliminary measurements. Specifically, we measured the throughput
of each session in isolation, that is, when the other session is not active. Then, we mea-
sured the throughput of each session when both sessions are active. Hereafter, Thi(x) de-
notes the throughput of session i (i = 1, 2) when both sessions are active and d(2, 3) = x.
Obviously, Thi( ) denotes the throughput of session i(i = 1, 2) when d(2, 3) = , and,
hence, the two sessions are independent. By exploiting these measurements, we estimated
the correlation existing between the two sessions by the following index:

Th1(x) + Th2(x)
D1(x) = 1 “
Th1( ) + Th2( )




Figure 3.24. Reference network scenario.
102 IEEE 802.11 AD HOC NETWORKS: PROTOCOLS, PERFORMANCE, AND OPEN ISSUES


The D1(x) index takes the value 0 if the two sessions are independent. Taken a session as a
reference, the presence of the other session may have two possible effects on the perfor-
mance of the reference session: (1) if the two sessions are within the same physical carri-
er sensing range, they share the same physical channel; (2) if they are outside the physical
carrier sensing range, the radiated energy from one session may still affect the quality of
the channel observed by the other session. As the radiated energy may travel over unlimit-
ed distances, we can expect that D1(x) may be equal to zero only for very large distances
among the sessions [8].

<<

. 21
( 87 .)



>>