Route Discovery

As in the DSDV routing protocol, our proposed modification requires each mobile node to maintain a routing table which lists all the possible destinations within the network. As shown in Table. 3.1, each entry is tagged with some important routing information such as a sequence number, the metric and the next hop to each destination, and the install time of each entry. The sequence number, as mentioned before, provides a judgment on the freshness of a route. Every time a destination node advertises its routing table, the corresponding sequence number is increased.

Upon receiving the routing information, a route with a more recent sequence number is always preferred. For those candidates with the same sequence number, a route with the smallest metric is selected, and the corresponding node is chosen as the next hop. In a multi-hop environment, the next hop succeeds the data packets and relays forward the destination. The metric represents the crowd level of each route all the

way to the destination. The definition and calculation of the crowd level will be discussed later. The install time field indicates when the entry is installed in the routing table. Examining the install time of each route helps to determine when to delete stale routes. In fact, not all of the information in the routing table is exchanged through the network. As shown in Table. 3.2, update packets contains only information about reachable destinations and the corresponding metrics and sequence numbers.

In addition to the routing table, a neighborhood table is maintained in each mobile node. As shown in Table. 3.3, a neighborhood table lists all the reachable neighbors and the corresponding directions. Each entry in the neighborhood table is also tagged with the install time as in routing tables. Every time a node hears a signal from one of its adjacent nodes, the DOA of this signal is estimated at the PHY layer.

Since the signal can be received by this node, the sender of the signal represents a reachable neighbor of this node. If this neighbor is never heard before, it is added into the neighborhood table along with the DOA. If this neighbor is already recorded in the neighborhood table, the DOA is updated. Therefore, a node can be aware of all its reachable neighbors and the corresponding direction to each. The neighborhood table not only helps to calculate the crowd level of a route to a certain destination, but also guides the antenna beam to the right direction. Recall that at the MAC layer, an RTS packet requires a DOA caching mechanism to be sent directionally toward an intended node. That is, when relaying a data packet, a node must check its routing table for the next hop. After making sure that it has a route to the destination, an RTS packet is sent directionally toward the next hop according to the DOA information recorded in the neighborhood table.

In the following, we present a routing strategy to discover a route along which packet transmission experiences the least interference. An intuitive way to achieve

this goal is to relay packets through a sparsely populated area. Firstly, the crowd level is calculated as follows. When a node receives a signal from one of its neighbors, the DOA of this signal is estimated at the PHY layer, and the neighborhood table is updated as mentioned before. If this received signal is a routing update, the node will refer to its neighborhood table for the calculation of crowd level before updating its routing table. According to the neighborhood table, the number of neighbors within a certain range of the DOA which the signal comes from can be determined. This number is regarded as the crowd level of this single hop in the corresponding direction. Secondly, the routing table is updated. The metric of each entry in the routing update is accumulated with the newly derived single-hop crowd level. As a result, the crowd level of each route in which the routing update source node is used as the next hop can be derived. An entry with a small metric represents a route along which nodes are sparsely located, and thus is preferred. The routing information update process is shown by a flow chart in Fig. 3.6. A latest sparse route can thus be derived.

Fig. 3.7 is an example of sparse route selection. As shown in this figure, by the advertisement of routing update, node D is a reachable destination of node A through two different routes. Upon receiving the routing updates of nodes B and E, node A chooses one of these two nodes as the next hop to node D. According to the neighborhood table in node A, there are five neighbors in the direction of node E while there is only one neighbor in the direction of node B. Therefore, the metric in the routing updates of node E and node B are added by five and one, respectively.

Similarly, as node B receives the routing update from node C, the metric is added by one. With the same sequence number, it is obvious that node A chooses node B as the next hop to reach node D. As a consequence, a sparse route is discovered. Table 3.1 shows the update of node A’s routing table. Table 3.1(a) is the routing table of node A

which is updated based on the routing update sent by node E at time T50. The table lists all reachable destinations and the next hop to each. The corresponding metric and sequence number of each entry are also listed for reference. To reach nodes C and D, node A relays its packets through node E. Table 3.1(b) is the routing update which was advertised by node B at time T52. This table lists all reachable destinations of node B. Each entry is tagged with a metric and a sequence number which is generated by the corresponding destination. The larger the sequence number is, the newer the route is. This routing update informs all the neighbors of node B that these destinations are reachable through node B. Upon receiving this routing update, node A refreshes its routing table which is shown in Table 3.1(c). For a certain destination, a route with a newer sequence number is preferred. Originally, to reach node C, node A has a route via node E with a sequence number of S962. After receiving node B’s routing update, the next hop is replaced with node B, and the new route has a sequence number of S964. To reach node D, node B has a route with a sequence number of S1000 which is the same as that listed in Table 3.1(a). However, the route via node B has a smaller metric and is preferred. This update of routing table completes a sparse route from node A to node D.

From the previous example, we note that using node B as the next hop, node A needs three hops to reach node D while in the case of using node E as the next hop, node D is only two hops away. Since our objective is to find a sparse route, the obtained sparse route is not necessarily the shortest route. However, update information along a shorter path could arrive earlier, and is preferred for its newer sequence number. Not long after the arrival of the update along the sparse route, the former one is replaced. In order to suppress the fluctuations of the routing table, routing updates are kept for a short period of time between the arrival of the first and the arrival of the sparse route. After the arrival of the sparse route, a node can make a

right decision. The delay of update also helps to reduce the number of rebroadcasts of possible route entries that normally arrive with the same sequence number.

3.4.2.2

Route Maintenance

The mobility of nodes causes broken links. The broken link may be detected by the layer-2 protocol, or it may also be inferred if no broadcasts have been received for a while from a former neighbor. Since a broken link could result in serious transmission error, a broadcast routing update containing this information should be arranged immediately. In the routing table, a broken link is described by a metric of

∞ . Once a node detects a broken link, i.e. an unreachable next hop, it immediately assigns an ∞ metric to any route through that next hop and generates an updated sequence number. This is the only situation when the sequence number is generated by any mobile node other than the destination node. Sequence numbers defined by the originating mobile nodes are generated as even numbers, and sequence numbers generated to indicate ∞ metrics are odd numbers. Upon receiving the notification of a broken link, a node updates its routing table and continues to advertise the broken link. When a node receives an ∞ metric, and it has a later sequence number with a finite metric, it triggers a route update broadcast to disseminate the important news about that destination.

In summary, the amount of overhead required for the proposed routing strategy is the extra memory for neighborhood table and the computation of crowd level.

However, both the MAC and routing functionality can benefit from the establishment of neighborhood table. Furthermore, the directional antenna adopted at the PHY layer has no effect on the amount of overhead. In other words, a smaller sector size or beamwidth of a directional antenna does not result in more overhead.

3.5 Summary

From the previous chapter, we know that the throughput reduction results from the medium reservation policy in MAC protocol, and the merits of directional antennas cannot be fully exploited without the modification of upper layer protocols.

In this chapter, we propose an integrated refinement of MAC and routing protocols with the use of directional antennas. A node equipped with directional antennas has the ability to estimate the DOA of an incoming signal and thus can identify the relative directions of its neighbors. Before transmitting data packets, nodes exchange RTS/CTS packets directionally, therefore, only those nodes located within the direction of transmission are blocked. Moreover, if a node receives an RTS or a CTS packet from its neighbors, a DNAV is set according to the DOA of that control packet, and the duration as well. With the help of DNAV, a node is blocked only for those directions where RTS/CTS packets come from, and spatial reuse can thus be achieved. Furthermore, to fully exploit the advantages of directional antennas, we propose a routing strategy to discover a route that experiences the least interference.

The DOA information helps a node to identify the relative directions of its neighbors and the crowd level in every direction. Our proposed routing strategy prefers a next hop which has fewer neighbors. Therefore, packets can be routed to the destination through a sparsely populated area. The joint design of MAC and routing protocols fully exploits the advantages of directional antennas and improves the throughput performance significantly. In the following chapter, we present some computer simulation results to verify improvements over the omni-directional approach.

Figure 3.1: Network capacity is improved via directional antennas. Four sessions is allowed to be held simultaneously without interfering with each other. While in the case of omni-directional communication, most nodes are blocked.

A

B

C

D

E

F G

H J

Central Element

Passive Parasitical Element

Figure 3.2: Antenna gain pattern of an 8-element circular array.

Figure 3.3: A 7-element electronically steerable passive array radiator antenna.

The central element is connected to the main RF radiator. Each passive parasitical element is loaded with a variable reactor.

A

B RTS

STEP 1:

A

B CTS

STEP 2:

A B

DATA STEP 3:

A

B ACK

STEP 4:

Figure 3.4: A modified RTS/CTS exchanging mechanism using directional antennas. The updates of DOA information are achieved by RTS/CTS exchanging. With the latest DOA information, transmission of data packet can be much more reliable.

D

DNAV² (325°) DNAV¹

(115°)

A B

C

E

Ongoing Data Transmission

Figure 3.5: Spatial reuse can be achieved by the adoption of DNAV. Two DNAVs reserve the wireless medium for nodes B and C. The blank area represents available directions for node A’s transmission.

Figure 3.6: Flowchart of routing information update.

Receive routing update packet

For each entry, check newer seq. number

Check new dst.

Add new entry to routing table

Yes

No

Update routing info.

Newer

Check crowd level

Equal

Lower

Unchanged ^Higher

Older

Update neighborhood

table

Figure 3.7: Two different routes are available for node A to reach node D. By the calculation of crowd level, a sparse route can be determined, and packets will not be relayed through a crowded area.

Metric

+1

Metric

+1

Metric

+5

A

B

C

D

E

Table 3.1: Routing table.

Destination Metric Sequence number Next hop Install time

Table 3.2: Update packet.

Destination Metric Sequence number

Table 3.3: Neighborhood table.

Neighbor Angle Install time

Table 3.4: Example of routing table update. With the same sequence number, a route with a smaller metric is preferred. A sparse route from node A to node D is completed after the update of routing table.

(a) Routing table of node A after receiving the update from node E

Destination Metric Sequence number Next hop Install time A 0 S1012 A T46 D 6 S1000 E T50 C 7 S962 E T50 E 5 S850 E T50

(b) Routing update sent by node B at time T52 Destination Metric Sequence number

B 0 S920 A 1 S1012 C 1 S964 D 2 S1000

(c) Routing table of node A after receiving the update from node B Destination Metric Sequence number Next hop Install time

A 0 S1012 A T46 D 3 S1000 B T52 B 1 S920 B T52 C 2 S964 B T52 E 5 S850 E T50

Chapter 4

Computer Simulations

To evaluate the performance of the proposed system architecture, we use the NCTUns 1.0 network simulator [23]. As NCTUns does not support network environment with directional antennas, we implement the directional antenna module based on the system architecture described in Chapter 3. Each node in the network is assumed to have an electrically steerable directional antenna system. For simplicity, as most studies on wireless ad hoc networks using directional antennas, a flat-topped antenna pattern is used in our simulation. A flat-topped antenna pattern has a constant gain for all directions within the specified angle, and has a lower constant gain for all other directions, representing side lobes of the pattern. This beam is steered under the control of the routing protocol as discussed in Section 3.4.2. In our simulations, a directional antenna gain of 8 is assumed within a 60-degree angular region, and a lower gain of 0.2 is assumed for all other regions. The destination of each transmitting node is chosen randomly. The packet length is constant and equals to 1000 bytes. The approximate omni-transmission range is 250 meters. We use the two-ray ground propagation model as the path loss model. Each simulation run is conducted for 200 seconds, and each data point is the average of 3 simulation runs.

The following sections present some simulation results that show how the capacity of wireless ad hoc networks depends on these parameters.

4.1 Blocking Problem

Recall that in Chapter 2, we have shown that the RTS/CTS exchanging mechanism may lead to a congestion situation where most of the nodes in a network are unable to transmit any packets during long periods of time. As a result, the network throughput decreases with increasing traffic load instead of maintaining its peak value. In this section, we present the simulation result of the blocking problem.

We use a rectangular grid arrangement of the nodes. The network consists of 25 static nodes arranged as a 5×5 square. The distance between nodes horizontally and vertically is 200 meters. Fig. 4.1 shows the simulation result of the throughput performance as the traffic load increases. The traffic load is defined as the percentage of source nodes in the network wishing to transmit data. Two different configurations of operating mode are included: directional mode and omni-mode. In the directional mode, as described in Chapter 3, each node adopts the DMAC and the modified DSDV routing protocol for directional transmission. In the omni-mode, each node transmits and receives packets omni-directionally with the standard IEEE 802.11 DCF and unmodified DSDV routing protocol. From Fig. 4.1(a), it is clear that the throughput of the network operating in the omni-mode goes to zero as the traffic load is increased, which implies that the network behaves like a congested network. While in the case of directional mode, the throughput performance is greatly improved and maintains an acceptable value under heavy traffic loading. Another important factor, the average delay of packet delivery is also improved in the operation of directional mode. Fig. 4.1(b) shows the average packet delay as the traffic load increases. The difference between these two operating modes is not significant under low loading situation. However, as the traffic load exceeds 50%, the average delay of standard omni-mode increases drastically, while in the case of directional mode, the average

delay remains stable until the traffic load exceeds 80%. Also, under any load situation, operating in the directional mode yields a lower average delay. This gives an indication of the alleviation of RTS/CTS-induced congestion problem by spatial reusing.

4.2 Effects of Mobility

The main objective of the simulations is to qualitatively analyze the capacity improvement in wireless ad hoc networks when directional antennas are used for communication. In this section, we present the simulation result of the overall system throughput. The network consists of 15 mobile nodes randomly distributed in an 800 meters ×800 meters square area, in which 5 nodes are randomly selected as source nodes wishing to send data packets. In the mobility scenario, the random waypoint model is used as the mobility model in which each node moves straight towards a randomly chosen destination at a constant speed. After the node reaches the destination, it chooses another point and moves toward the new destination without pause. Three different configurations of operating mode are included: directional mode, omni-mode, and direction mode with unmodified DSDV routing protocol. In the third configuration, the PHY and MAC layers are implemented as the directional mode, while the routing layer adopts the unmodified DSDV routing protocol.

Fig. 4.2(a) shows the simulation result of the throughput performance as the moving speed of nodes in the network increases. For all the three cases, the throughput performance goes down as the mobility increases. However, it is clear that the two modes using directional antennas at the PHY layer and the corresponding DMAC have outstanding performance over the omni-mode. This indicates that the potential of spatial reuse has a significant improvement on

throughput performance. Furthermore, our modified DSDV routing protocol has a better performance especially when nodes move slowly. Under high mobility, these two routing protocols have similar performance. Note that these two modes using directional antennas have the same configuration at the PHY and MAC layers.

Therefore the two routing protocols present different routes along which packets are relayed. Our modified DSDV routing protocol discovers a sparse route while the DSDV routing protocol discovers the shortest route. Normally, a route discovered by our modified DSDV has a larger hop count, and hence is less reliable under high mobility. Fig. 4.2(b) shows the packet delivery ratio which is defined as the number of packets received successfully by the destination divided by the number of packets generated by source nodes. The packet delivery ratio indicates how many packets the data source can deliver to the destination over multiple hops without packet drops due to queue overflow or transmission failure. It is clear from this figure that the packet delivery ratio decreases as node mobility increases. Operating in the omni-mode has the lowest packet delivery ratio, and again spatial reuse improves the

Therefore the two routing protocols present different routes along which packets are relayed. Our modified DSDV routing protocol discovers a sparse route while the DSDV routing protocol discovers the shortest route. Normally, a route discovered by our modified DSDV has a larger hop count, and hence is less reliable under high mobility. Fig. 4.2(b) shows the packet delivery ratio which is defined as the number of packets received successfully by the destination divided by the number of packets generated by source nodes. The packet delivery ratio indicates how many packets the data source can deliver to the destination over multiple hops without packet drops due to queue overflow or transmission failure. It is clear from this figure that the packet delivery ratio decreases as node mobility increases. Operating in the omni-mode has the lowest packet delivery ratio, and again spatial reuse improves the