Formalizing Packet Level Incoordination in IEEE 802.11 Ad Hoc

0 downloads 0 Views 421KB Size Report
Marc Torrent-Moreno, Steven Corroy and Hannes Hartenstein ... torrent@tm.uni-karlsruhe.de, [email protected], ..... Fading Channels - 2nd Edition.
Poster Abstract: Formalizing Packet Level Incoordination in IEEE 802.11 Ad Hoc Networks: 1-Hop Broadcast Performance Analysis Marc Torrent-Moreno, Steven Corroy and Hannes Hartenstein Institute of Telematics, University of Karlsruhe, Germany [email protected], [email protected], [email protected]

I. I NTRODUCTION Assume a communication scenario in wireless ad hoc and sensor networks where each node periodically sends out status information as a 1-hop broadcast to inform neighboring nodes. These scenarios are typical for monitoring and safety related applications where each node needs information about its neighborhood, e.g., in vehicular ad hoc networks. In IEEE 802.11 based ad hoc networks, one reason not to receive a packet is due to collisions caused by the well known hidden terminal problem (stressed in the addressed scenario where RTS/CTS mechanism is not applicable). It is also well known that wireless communication protocols may achieve different performance results when assuming different radio propagation models and implementations [1]. While the effect of the hidden terminal problem is well understood when utilizing a deterministic1 radio propagation model, it has a new ‘flavor’ with the utilization of more realistic2 ones, which do not experience a fixed Carrier Sense range (CS). Our contribution in this document is a formalization of a metric able to characterize the effect of different propagation models on IEEE 802.11 networks, and its evaluation (using the ns-2.28 simulator) with respect to packet reception rates. The new metric, which we name Packet Level Incoordination (PLI), is based on the CSMA/CA principle and measures the inability of CSMA/CA of correctly coordinate the nodes on a network from a specific node’s perspective (see Fig. 1). This new metric can assist researchers to understand the details of wireless communications, and thus, in the process to find the best strategies (e.g., power control) when designing optimal protocols. The reminder of the paper is the following: Section II briefly presents radio propagation principles, and the way they and the physical layer are implemented in the used simulator. In Section III, we provide the general definition of PLI and the simulation results. Section IV concludes the paper. 1 Deterministic radio propagation models derive a fix attenuation of the power of the transmitted signal for a given distance traveled from the sender to the receiver. 2 With ‘more realistic’ we refer to probabilistic models, which derive an attenuation following a specific probability function corresponding to different multipath scenarios with amplitude fluctuations during time due to the relative movement of the surrounding.

B C A

CS

Fig. 1. Impact of CSMA/CA’s incoordination. In case nodes B and/or C do not sense an ongoing transmission of node A, they may also transmit creating a collision with node A’s message. The potential impact of node C, which may not sense node A’s message only in the case a probabilistic propagation model is used, can be much worse due to its larger common coverage area with node A.

II. R ADIO P ROPAGATION

AND

S IMULATOR

In order to understand the behavior of wireless networks radio wave propagation has to be considered in detail. In this section, we offer a brief description due to comprehensibility of the results, for more details please refer to [2]. In a real scenario, the attenuation of a transmitted signal is not only caused by free-space loss. Indeed many phenomena such as path-loss (which is the attenuation due to interferences like reflections, diffractions or scattering between the radio waves and the environment) or fading (which is the attenuation due to the movement of the environment relative to the sender) can lead to strong loss in the power of the transmitted signal. Simulation tools use different models to represent these phenomena in various scenarios. In our experiments, we take into account three radio wave propagation models. A deterministic one: Two Ray Ground (TRG) which models the direct path of radio waves between sender and receiver and one reflection to the ground. And two probabilistic: Log-Normal Shadowing (LNS), which models the path-loss relative to a reference distance and the possible variations in the content (density, relief) of the environment; and Nakagami (Nak), which models fast-fading and is shown to match empirical data [3]. Figure 2 shows the probability of reception over the distance for these models with different parameters’ values, which are used to model different scenarios. In the standard ns-2.28 distribution, the reception and

Probability of successful reception

TRG Nak m=3 Nak m=1 LNS β=2 σ=5 LNS β=2 σ=10

1 0.8 0.6 0.4 0.2 0 0

20

40

60

80

100

120

140

160

180

Distance (m)

Fig. 2. Probability of reception for different propagation models with respect to the distance in the absence of interferences from other nodes and a transmission power that provides a communication range of 100m with TRG.

interference models are implemented as follows: A received signal is said to be received successfully if the received power P > RxT h (RxT h the reception threshold) and not sensed if P < CST h (CST h the carrier sense threshold). If two packets, A and B, are on the channel at the same time at a specific location, one can be received (captured) if PA > PB + CPT h (CPT h denotes the capture threshold).

III. PLI D EFINITION

AND

S IMULATION

CSMA/CA is based on the principle that all nodes in a network should be able to detect when another node is transmitting a frame on the medium. In our case, a station with a packet to send must first ‘listen’ to the medium, during DIFS (Distributed Interframe Space), in order to avoid collisions. According to the standard [4], a node intending to send a message while the channel is busy, has to wait at least until DIFS after the finalization of the ongoing transmission. We will account as incoordination all cases not fullfiling this rule. Still, before defining the metric we should remark that: i) it does not make sense to compute PLI from a global perspective, i.e., nodes ‘far enough’ from each other, which transmissions do not spatially overlap, should not be considered as uncoordinated even if their packets overlap in time, ii) a fix radio coverage area (and CS) does not exist with probabilistic models, what makes difficult to identify nodes that are ‘far enough’ from each other, and iii) performance is generally evaluated from a ‘receiver perspective’, i.e., the amount of data successfully received by a node.Therefore, we propose the following definition: Packet Level Incoordination, receiver perspective: In case a node A has sensed two frames from nodes B and C, i.e., the messages from B and C arrive at A’s location with power greater than CST h , node A will account C’s packet as uncoordinated if: tB + tPBC < tC < tB + tPBC + lB + DIF S, where tB and tC are the times when nodes B and C started to transmit their packets respectively; tPBC is the time needed for a radio wave to propagate from B to C; and lB is the length ‘in seconds’ of the message sent by B (Fig. 3). Note that two packets colliding because they had chosen the same backoff timer after contention would not be considered a PLI. Thus, by measuring the probability of incoordination we obtain

Fig. 3. The diagrams show the power of the signals sent by B and C at the correspondent node’s location. In situation (a) node C senses the channel busy when intending to transmit a message; therefore, it defers (coordinates) its transmission. In situation (b) node C senses the channel idle when intending to transmit; thus, the message is sent uncoordinated, what results, in this case, in a collision at node A.

a metric that reflects the ratio of overlapping messages due to CSMA/CA inability to coordinate wireless nodes, i.e., the major cause of packet collisions, sensed by a potential receiver. We have simulated two different circular system areas (Small and Big scenario), where nodes sending broadcast periodic messages have been randomly located (different seeds have been used to obtain statistical significance, a 95% confident interval is presented). This shape has been chosen for spatial uniformity with respect to the node in the center, which has been analyzed and the results presented in the following. In all cases, the physical layer parameters have been configured as in [5] and the packet sizes have been fixed to 500Bytes.

A. Small Scenario The Small scenario consists of a circle of 78m radius with 31 nodes. At first, we configure each node to send 4 packets/s with an intended Communication Range CR = 100m (CS = 156m). This scenario has been selected as toy example, where all nodes reside (when assuming TRG) inside the CS of each other, i.e., ideally no incoordination would occur. Fig. 4(a) presents the ratio of uncoordinated packets sent by a node placed at a distance d = 50m from our reference node (in the center). Note that, even though each node is located inside the intended CS of all the other nodes and the offered load to the medium is relatively low, the only propagation model that results in a totally coordinated scenario is TRG. Observe also the correspondence with Fig. 5(a), i.e., a higher incoordination corresponds to a lower probability of successful reception of packets transmitted by those nodes. In such scenario, a straightforward strategy to achieve a lower PLI is to increase the transmission power. In Figures 4(a) and 5(a) we also represented the results of the same scenario when configuring all nodes with a CR = 200m (CS = 312m). As we can observe the incoordination has decreased significantly and the probability of successful reception has increased. Note though, that due to the topology of the scenario, interferences from/to other nodes and the relative load offered to the network have not increased. Thus, the increase of the values in Fig. 5(a) corresponds in a major part to the improvement of the received power than to the

0.03 0.025 0.02 0.015 0.01 0.005 0

1 0.8

S LN

S LN

ak N

ak N

2 β=

=3

2 β=

m

=1

m

G TR

Probability of incoordination

LNS β=2 σ=10 LNS β=2 σ=5 Nak m=1 Nak m=3 TRG

Carrier Range 100 meters Carrier Range 200 meters Probability of incoordination

Probability of incoordination

0.04 0.035

0.6 0.4 0.2 0

0.8 0.6 0.4 0.2 0

1 σ=

5 σ=

0

0

(a) Small Scenario (at d = 50m)

LNS β=2 σ=10 LNS β=2 σ=5 Nak m=1 Nak m=3 TRG

1

20

40

60 80 100 Distance (m)

120

140

0

(b) Big Scenario (4 pckts/s, 100m CR)

20

40

60 80 100 Distance (m)

120

140

(c) Big Scenario (2 pckts/s, 141m CR)

0.8 0.6 0.4 0.2 0

S LN

S LN

ak N

ak N

2 β=

2 β=

=3

m

=1

m

G TR

0.8 0.6 0.4 0.2 0

0

1 σ=

5 σ=

(a) Small Scenario (at d = 50m)

TRG Nak m=3 Nak m=1 LNS β=2 σ=5 LNS β=2 σ=10

1

Probability of successful reception

Carrier Range 100 meters Carrier Range 200 meters

1

Probability of successful reception

Probability of successful reception

Fig. 4. Probability of packet level incoordination of nodes at different distances from the reference node and for different propagation models. TRG Nak m=3 Nak m=1 LNS β=2 σ=5 LNS β=2 σ=10

1 0.8 0.6 0.4 0.2 0

0

20

40

60 80 100 Distance (m)

120

140

(b) Big Scenario (4 pckts/s, 100m CR)

0

20

40

60 80 100 Distance (m)

120

140

(c) Big Scenario (2 pckts/s, 141m CR)

Fig. 5. Probability of successful reception from nodes at different distances from the reference node and for different propagation models.

improvement of the PLI itself. As we will see in the results of the next subsection, with a bigger scenario (where the CS of the nodes do not cover all the space) the strategy to improve reception rates is not as straightforward.

B. Big Scenario The Big scenario consists of a circle of 600m radius with 1800 nodes (same density of nodes/m2 as the Small scenario). At first, we configure each node to send 4 packets/s with a CR = 100m (CS = 156m). We can observe in Fig. 4(b) the probability that a node located at a distance d from the reference node transmits an uncoordinated packet; and in Fig. 5(b) the probability to receive a packet successfully from those nodes. Observe how locating nodes further than CS and utilizing probabilistic propagation models challenge the proper functionality of CSMA/CA, in comparison with the Small scenario. Also, a higher incoordination corresponds, as expected, to a lower probability of reception for all distances. Second, we repeat our experiment configuring all nodes to transmit 2 packets/s with a CR = 141m (CS = 212m). Note that this configuration results in an offered load to the medium (with respect to CS) equal than the previous configuration. Comparing Figures 4(b) and 4(c) we can observe a slightly higher level of PLI in the second case, what corresponds to slightly lower probabilities of reception for nodes closer than 60 meters to the reference node (Figure 5(c)). For nodes further than 60m the impact of the attenuation caused by radio propagation overcomes the impact of incoordination with respect to other messages, i.e., reception rates are higher when using a higher transmission power.

IV. S UMMARY

AND

O UTLOOK

In this document, we have conceptualized the effect that radio propagation models, both deterministic and probabilistic, have on IEEE 802.11 networks, which we called Packet Level Incoordination (PLI). Further, with the use of ns-2.28 simulator we have presented a first set of results with respect to the relation of probability of successful reception and PLI in two different scenarios and for three radio propagation models. Our future work comprises a deeper and broader analysis of PLI in different scenarios, not only 1-hop broadcast but also considering the case of unicast flows.

R EFERENCES [1] M. Takai, J. Martin, and R. Bagrodia, “Effects of Wireless Physical Layer Modeling in Mobile Ad Hoc Networks,” in Proc. ACM MobiHoc, October 2001. [2] M. K. Simon and M.-S. Alouini, Digital Communication over Fading Channels - 2nd Edition. Wiley, 2005. [3] V. Taliwal, D. Jiang, H. Mangold, C. Chen, and R. Sengupta, “Empirical Determination of Channel Characteristics for DSRC Vehicle-to-vehicle Communication,” in Proc. ACM VANET, October 2004. [4] “IEEE Std. 802.11-1999, Part11: Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) specifications,” IEEE Std. 802.11, 1999 edition. [5] M. Torrent-Moreno, D. Jiang, and H. Hartenstein, “Broadcast Reception Rates and Effects of Priority Access in 802.11-Based Vehicular Ad-Hoc Networks,” in Proc. ACM VANET, October 2004.

Suggest Documents