**5. Channel assignment algorithms for WMN**

As has been already mentioned, CA in a multi-interface WMN consists of assigning channels to the radio interfaces in order to achieve efficient channel utilization for minimizing interference and to guarantee an adequate level of connectivity. Nowadays, there exist many approaches to solve the channel assignment problem. These approaches can be divided into three main categories (Conti et al., 2007):


In this section several channel assignment approaches are compared by QoS parameters mentioned in the previous section.

Channel Assignment Schemes Optimization for Multi-Interface Wireless Mesh Networks Based on Link Load 95

cause nodes A and B have no common channel, a channel re-assignment is required. Link between nodes A and B needs to be assigned one of the channels from [1, 2, 6, 7]. Based on the channel expected loads, link between nodes A and B is assigned channel 6, and channel 7 assigned already to link between nodes B and D is reassigned to channel 6 (Raniwala et al.,

The *First Random Channel Assignment algorithm* (FRCA) is a dynamic and centralized load aware channel assignment and routing algorithm for multi-interface multi-channel WMN (Pollak, Wieser, 2012). This approach takes into account the network traffic profile. FRCA algorithm assigns radio channels to links considering their expected loads and interference effect of other links, which are in interference range and which are tuned to the same radio

In the first phase, algorithm estimates initial loads on all links based on the initial routes created by routing algorithm. After load estimation, FRCA randomly assigns channels to all

In the second phase, FRCA algorithm uses similar steps as in the first phase, but channel assignment and routing iterations are based on results from the first phase. If some of the link load is higher than link capacity, the algorithm goes back and tries to find better solution. Algorithm's iterations end when no further improvement is possible. In

2004, Yulong Chen et al., 2010).

**Figure 20.** An example of channel revisit in LACA approach

**5.3. First random channel assignment** 

FRCA algorithm consists of two basic phases:

channel.

1. Initial phase

2. Optimization phase

nodes for each radio interface.

#### **5.1. Common channel assignment**

The Common channel assignment (CCA) is a simplest fixed channel assignment approach (Adya et al., 2004). In this CA approach all radio interfaces of each node were tuned to the same set of channels. For example, if every node has two radio interfaces then each node uses the same two channels (Fig. 19). The main benefit of this approach is the network connectivity. The connectivity is the same as that of a single interface approach, while the using of multiple radio interfaces can improve network throughput. However, if the number of non-overlapping channel is much higher than the number of radio interfaces, the gain of the CCA may be limited. CCA scheme presents a simplest channel assignment approach but it fails to account for the various factors affecting CA in a WMN. This solution will decrease the utilization of network resources (Yulong Chen et al., 2010).

**Figure 19.** Example of common channel assignment approach

#### **5.2. Load aware channel assignment**

*Load aware channel assignment* (LACA) represents a dynamic centralized channel assignment and routing algorithm, where traffic is mainly directed toward gateway nodes (Raniwala et al., 2004), assuming that the offered traffic load on each virtual link is known. Algorithm assigns channels by such a way to ensure the network connectivity while takes into account the bandwidth limitation of each link. At the beginning, LACA estimates the total expected load on each virtual link based on the load imposed by each traffic flow. In the next step CA algorithm visits each virtual link in decreasing order of expected traffic load and greedily assigns it a channel. The algorithm starts with an initial estimation of the expected traffic load and iterates over channel assignment and routing until the bandwidth allocated on each virtual link matches its expected load. While this CA approach presents a method for CA that incorporates connectivity and flow patterns, the CA scheme on links may cause a "*ripple effect*", whereby already assigned links have to be revisited, thus increasing the time complexity of the scheme.

An example of node revisiting is illustrated in Fig. 20. In this example each node has two radio interfaces. The channel list of node A is [1, 6] and channel list of node B is [2, 7]. Because nodes A and B have no common channel, a channel re-assignment is required. Link between nodes A and B needs to be assigned one of the channels from [1, 2, 6, 7]. Based on the channel expected loads, link between nodes A and B is assigned channel 6, and channel 7 assigned already to link between nodes B and D is reassigned to channel 6 (Raniwala et al., 2004, Yulong Chen et al., 2010).

**Figure 20.** An example of channel revisit in LACA approach

#### **5.3. First random channel assignment**

The *First Random Channel Assignment algorithm* (FRCA) is a dynamic and centralized load aware channel assignment and routing algorithm for multi-interface multi-channel WMN (Pollak, Wieser, 2012). This approach takes into account the network traffic profile. FRCA algorithm assigns radio channels to links considering their expected loads and interference effect of other links, which are in interference range and which are tuned to the same radio channel.

FRCA algorithm consists of two basic phases:

1. Initial phase

94 Wireless Mesh Networks – Efficient Link Scheduling, Channel Assignment and Network Planning Strategies

The Common channel assignment (CCA) is a simplest fixed channel assignment approach (Adya et al., 2004). In this CA approach all radio interfaces of each node were tuned to the same set of channels. For example, if every node has two radio interfaces then each node uses the same two channels (Fig. 19). The main benefit of this approach is the network connectivity. The connectivity is the same as that of a single interface approach, while the using of multiple radio interfaces can improve network throughput. However, if the number of non-overlapping channel is much higher than the number of radio interfaces, the gain of the CCA may be limited. CCA scheme presents a simplest channel assignment approach but it fails to account for the various factors affecting CA in a WMN. This solution will decrease

*Load aware channel assignment* (LACA) represents a dynamic centralized channel assignment and routing algorithm, where traffic is mainly directed toward gateway nodes (Raniwala et al., 2004), assuming that the offered traffic load on each virtual link is known. Algorithm assigns channels by such a way to ensure the network connectivity while takes into account the bandwidth limitation of each link. At the beginning, LACA estimates the total expected load on each virtual link based on the load imposed by each traffic flow. In the next step CA algorithm visits each virtual link in decreasing order of expected traffic load and greedily assigns it a channel. The algorithm starts with an initial estimation of the expected traffic load and iterates over channel assignment and routing until the bandwidth allocated on each virtual link matches its expected load. While this CA approach presents a method for CA that incorporates connectivity and flow patterns, the CA scheme on links may cause a "*ripple effect*", whereby already assigned links have to be revisited, thus increasing the time

An example of node revisiting is illustrated in Fig. 20. In this example each node has two radio interfaces. The channel list of node A is [1, 6] and channel list of node B is [2, 7]. Be-

**5.1. Common channel assignment** 

the utilization of network resources (Yulong Chen et al., 2010).

**Figure 19.** Example of common channel assignment approach

**5.2. Load aware channel assignment** 

complexity of the scheme.

2. Optimization phase

In the first phase, algorithm estimates initial loads on all links based on the initial routes created by routing algorithm. After load estimation, FRCA randomly assigns channels to all nodes for each radio interface.

In the second phase, FRCA algorithm uses similar steps as in the first phase, but channel assignment and routing iterations are based on results from the first phase. If some of the link load is higher than link capacity, the algorithm goes back and tries to find better solution. Algorithm's iterations end when no further improvement is possible. In optimization phase, FRCA uses greedy load-aware channel assignment algorithm similar to the one used in LACA algorithm (Raniwala et al., 2004). In this algorithm virtual links are visited in decreasing order of the link expected load. To find routes between nodes, FRCA uses shortest path routing based on minimum hop count metric (Kaabi et al., 2010).

#### *5.3.1. Link load estimation*

This approach is based on the concept of load criticality. The method assumes perfect load balancing across all acceptable paths between each communicating pair of nodes. Let *P*(*s, d*) denote the number of acceptable paths between pair of nodes (*s, d*), *Pl* (*s, d*) is the number of acceptable paths between (*s, d*) which pass a link *l*. And finally, let *B(s, d)* be the estimated load between node pair (*s, d*). Then the expected traffic load Φ*l* on link *l* is calculated as (Raniwala et al., 2004):

$$\Phi\_l = \sum\_{s,d} \frac{P\_l(s,d)}{P(s,d)} \cdot \mathbf{B}(s,d) \tag{1}$$

Channel Assignment Schemes Optimization for Multi-Interface Wireless Mesh Networks Based on Link Load 97

Source (s) Destination (d) γ(s, d) (Mbps) a g 0.9 i a 1.2 b j 0.5

(source, destination) (a, g) (i, a) (b, j) Possible paths a-c-g i-e-a b-f-j a-c-d-g i-e-d-a b-f-i-j a-d-g i-d-a b-e-i-j a-d-c-g i-d-c-a b-e-i-f-j a-d-h-g i-d-e-a b-e-d-i-j

P (source, destination) P(a, g) = 8 P(i, a) = 8 P(b, j) = 5

specific link) (Badia et al., 2009, Conti et al., 2007, Raniwala et al., 2004).

In the next step we calculate *P*(*s, d*) for each flow. We need to determine all the possible paths between source and destination. Table 3 shows all possible paths between communication node pairs for the WMN topology in Fig. 21. Values *P*(*s, d*) and the corresponding link traffic load (*Φl*) is calculated using equation (2). Results are shown in table 4. Based on these calculations, we can estimate the load between each neighboring nodes. The result of calculation *Φ<sup>l</sup>* is the expected traffic load of link *l* (i.e. the amount of traffic expected to be carried over a

l Pl(a, g) Pl(i, a) Pl(b, j) Φl (Mbps) a-c 2 3 0 0.675 c-g 2 2 0 0.525 c-d 2 1 0 0.375 d-g 2 1 0 0.375 a-d 4 3 0 0.9 g-h 0 1 0 0.15 d-h 1 1 0 0.2625 a-e 2 2 0 0.525 d-e 1 2 1 0.5125 d-i 1 3 1 0.6625 h-i 2 2 0 0.525 e-i 1 2 2 0.6125 b-e 0 0 3 0.3 b-f 0 0 2 0.2 f-i 0 0 2 0.2 i-j 0 0 20.2 f-j 0 0 20.2

 a-d-i-h-g i-d-g-c-a a-e-d-g i-h-d-a a-e-i-h-g i-h-g-c-a

**Table 3.** Possible data flows between communicating nodes

**Table 4.** The results of calculation Φ*<sup>l</sup>* on specific link *l*

**Table 2.** Traffic profile with three data flows

This equation implies that the initial expected traffic on a link is the sum of the loads from all acceptable paths, across all possible node pairs, which pass through the link. Because of the assumption of uniform multi-path routing, the load that an acceptable path between a pair of nodes is expected to carry is equal to the expected load of the pair of nodes divided by the total number of acceptable paths between them. Let us consider the logical topology as shown in Fig. 21 and assume that we have three data flows reported in table 2.

**Figure 21.** Multi-interface and multi-channel WMN

Because we have three different communications node pairs, we have

$$\Phi\_{l} = \frac{P\_{l}(a,\emptyset)}{P(a,\emptyset)} \cdot \gamma^{(a,\emptyset)} + \frac{P\_{l}(i,a)}{P(i,a)} \cdot \gamma^{(i,a)} + \frac{P\_{l}(b,j)}{P(b,j)} \cdot \gamma^{(b,j)} \tag{2}$$


**Table 2.** Traffic profile with three data flows

uses shortest path routing based on minimum hop count metric (Kaabi et al., 2010).

*5.3.1. Link load estimation* 

(Raniwala et al., 2004):

optimization phase, FRCA uses greedy load-aware channel assignment algorithm similar to the one used in LACA algorithm (Raniwala et al., 2004). In this algorithm virtual links are visited in decreasing order of the link expected load. To find routes between nodes, FRCA

This approach is based on the concept of load criticality. The method assumes perfect load balancing across all acceptable paths between each communicating pair of nodes. Let *P*(*s, d*) denote the number of acceptable paths between pair of nodes (*s, d*), *Pl* (*s, d*) is the number of acceptable paths between (*s, d*) which pass a link *l*. And finally, let *B(s, d)* be the estimated load between node pair (*s, d*). Then the expected traffic load Φ*l* on link *l* is calculated as

*l*

This equation implies that the initial expected traffic on a link is the sum of the loads from all acceptable paths, across all possible node pairs, which pass through the link. Because of the assumption of uniform multi-path routing, the load that an acceptable path between a pair of nodes is expected to carry is equal to the expected load of the pair of nodes divided by the total number of acceptable paths between them. Let us consider the logical topology

Wired Network

i

6

*P(s,d) B(s,d) P(s,d)*

(1)

b

5

1

1

*l ll (a,g) (i,a) (b,j)*

*P(a,g) P(i,a) P(b,j)*

f

5

2

j

*P(a,g) P(i,a) P(b, j)* (2)

*l s,d*

as shown in Fig. 21 and assume that we have three data flows reported in table 2.

a

1 3

c e

2

3

3

3

d

2

4

5

h

Because we have three different communications node pairs, we have

4

g

**Figure 21.** Multi-interface and multi-channel WMN

*l*

6


**Table 3.** Possible data flows between communicating nodes

In the next step we calculate *P*(*s, d*) for each flow. We need to determine all the possible paths between source and destination. Table 3 shows all possible paths between communication node pairs for the WMN topology in Fig. 21. Values *P*(*s, d*) and the corresponding link traffic load (*Φl*) is calculated using equation (2). Results are shown in table 4. Based on these calculations, we can estimate the load between each neighboring nodes. The result of calculation *Φ<sup>l</sup>* is the expected traffic load of link *l* (i.e. the amount of traffic expected to be carried over a specific link) (Badia et al., 2009, Conti et al., 2007, Raniwala et al., 2004).


**Table 4.** The results of calculation Φ*<sup>l</sup>* on specific link *l*

#### *5.3.2. Link capacity estimation*

The link capacity (channel bandwidth available to a virtual link) is determined by the number of all virtual links in its interference range that are also assigned to the same radio channel. So when estimating the usable capacity of the virtual link, we should consider all traffic loads in its interference range. According to the channel assignment rules, the higher load a link is expected to carry, the more bandwidth it should get. On the other side, the higher loads its interfering links are expected to carry, the less bandwidth it could obtain. Thus, the link capacity should be proportional to its traffic load, and be inversely proportional to all other interfering loads. Thus, the capacity *bw(i)* assigned to link *i* can be obtained using the following equation:

$$bw\_{(i)} = \frac{\Phi\_i}{\sum\_{j=\text{left}f(i)} \* \mathbb{C}\_{ch}} \* \mathbb{C}\_{ch} \tag{3}$$

Channel Assignment Schemes Optimization for Multi-Interface Wireless Mesh Networks Based on Link Load 99

and LACA reached the best performance with only 4 radio interfaces. Results show that further increasing of number of radio interfaces didn't increase the network performance, so

**Average End-to-End Delay**

CCA LACA FRCA

> CCA LACA FRCA

**Figure 22.** Average values of end-to-end delay for various radio interfaces and different CA schemes

LACA algorithm.

0

2000

4000

6000

8000

10000

12000

**Throughput (kbps)**

0.00 100.00 200.00 300.00 400.00 500.00 600.00 700.00 800.00 900.00

**End-to-end delay (ms)**

Figure 23 shows the average values of network throughput. The lowest value of average throughput for all radio interfaces was achieved in WMN with CCA scheme. This approach reached the best results for 6 radio interfaces. Others CA algorithms (FRCA and LACA) achieved the best performance with only 4 radio interfaces, with FRCA slightly outperformed

**Average Throughput**

2468 **Number of Radio Interfaces**

**Figure 23.** Average values of network throughput for various radio interfaces and different CA schemes

2468 **Number of Radio Interfaces**

the optimal number of radio interfaces for LACA and FRCA algorithm is 4.

where *Φi* is the expected load on link *i*, Intf(i) is the set of all virtual links in the interference range of link *i* (i.e. links *i* and *j* operates on the same channel). *Cch* is the sustained radio channel capacity (Badia et al., 2009, Conti et al., 2007, Raniwala et al., 2004).

#### **5.4. Simulation results**

In this section, the performance of proposed FRCA concept is evaluated and compared with CCA (Adya et al., 2004), LACA (Raniwala et al., 2004) and a single interface architecture by using NS-2 simulator (ns-2, 2008). Simulation model consisted of 25 static wireless mesh nodes placed in an area of 1000 x 1000 m (Fig.4a). The distance between nodes was set to 200 m. The capacity of all data links was fixed at 11Mbps. All nodes have the same transmission power and the same omni-directional antenna. The transmission range was set to 200 m and interference range was set to 400 m. For traffic generation, 25 CBR (Constant Bit Rate) flows with packet size 1000 bytes were used. Flows were created between randomly chosen node pairs. For simulation evaluation, the same metrics like in section 3.1 was used.

#### *5.4.1. Different number of radio interfaces*

From previous sections the conclusion about optimal number of six radio interfaces was gained. This conclusion was based on simple common channel assignment scheme CCA, which was used in simulations. With using more sophisticated channel assignment scheme it is possible to expect that the same results in QoS parameters may be reached with less number of interfaces. So the performance evaluation of chosen CA schemes was based on changing number of radio interfaces (between 2 to 8 radio interfaces for each node).

Figure 22 shows the average values of end-to-end delay for various number of radio interfaces. From results it is obvious that the highest value of delay (792.64 ms) was reached in WMN with CCA scheme. Lowest value (101.42 ms) reached WMN with FRCA algorithm for 4 radio interfaces. For CCA scheme the optimal number of radio interfaces was 6, but FRCA and LACA reached the best performance with only 4 radio interfaces. Results show that further increasing of number of radio interfaces didn't increase the network performance, so the optimal number of radio interfaces for LACA and FRCA algorithm is 4.

98 Wireless Mesh Networks – Efficient Link Scheduling, Channel Assignment and Network Planning Strategies

The link capacity (channel bandwidth available to a virtual link) is determined by the number of all virtual links in its interference range that are also assigned to the same radio channel. So when estimating the usable capacity of the virtual link, we should consider all traffic loads in its interference range. According to the channel assignment rules, the higher load a link is expected to carry, the more bandwidth it should get. On the other side, the higher loads its interfering links are expected to carry, the less bandwidth it could obtain. Thus, the link capacity should be proportional to its traffic load, and be inversely proportional to all other interfering loads. Thus, the capacity *bw(i)* assigned to link *i* can be

> *i (i) ch j*

(3)

*j Intf (i) bw \* C* 

where *Φi* is the expected load on link *i*, Intf(i) is the set of all virtual links in the interference range of link *i* (i.e. links *i* and *j* operates on the same channel). *Cch* is the sustained radio

In this section, the performance of proposed FRCA concept is evaluated and compared with CCA (Adya et al., 2004), LACA (Raniwala et al., 2004) and a single interface architecture by using NS-2 simulator (ns-2, 2008). Simulation model consisted of 25 static wireless mesh nodes placed in an area of 1000 x 1000 m (Fig.4a). The distance between nodes was set to 200 m. The capacity of all data links was fixed at 11Mbps. All nodes have the same transmission power and the same omni-directional antenna. The transmission range was set to 200 m and interference range was set to 400 m. For traffic generation, 25 CBR (Constant Bit Rate) flows with packet size 1000 bytes were used. Flows were created between randomly chosen node

From previous sections the conclusion about optimal number of six radio interfaces was gained. This conclusion was based on simple common channel assignment scheme CCA, which was used in simulations. With using more sophisticated channel assignment scheme it is possible to expect that the same results in QoS parameters may be reached with less number of interfaces. So the performance evaluation of chosen CA schemes was based on

Figure 22 shows the average values of end-to-end delay for various number of radio interfaces. From results it is obvious that the highest value of delay (792.64 ms) was reached in WMN with CCA scheme. Lowest value (101.42 ms) reached WMN with FRCA algorithm for 4 radio interfaces. For CCA scheme the optimal number of radio interfaces was 6, but FRCA

changing number of radio interfaces (between 2 to 8 radio interfaces for each node).

channel capacity (Badia et al., 2009, Conti et al., 2007, Raniwala et al., 2004).

pairs. For simulation evaluation, the same metrics like in section 3.1 was used.

*5.3.2. Link capacity estimation* 

obtained using the following equation:

*5.4.1. Different number of radio interfaces* 

**5.4. Simulation results**

**Figure 22.** Average values of end-to-end delay for various radio interfaces and different CA schemes

Figure 23 shows the average values of network throughput. The lowest value of average throughput for all radio interfaces was achieved in WMN with CCA scheme. This approach reached the best results for 6 radio interfaces. Others CA algorithms (FRCA and LACA) achieved the best performance with only 4 radio interfaces, with FRCA slightly outperformed LACA algorithm.

**Figure 23.** Average values of network throughput for various radio interfaces and different CA schemes

As we can see from figure 24 the highest value of packet loss for all number of interfaces was reached in WMN with CCA approach, with the best value reached for 6 radio interfaces (63.56 %). The best result (5.86 %) reached FRCA algorithm for 4 radio interfaces, whereas algorithm LACA with the same number of radio interfaces reached value 9.47%.

Channel Assignment Schemes Optimization for Multi-Interface Wireless Mesh Networks Based on Link Load 101

In this chapter, the study of optimal number of radio interfaces and new channel assignment approach was presented (FRCA). The study of optimal number of radio interfaces was created for two different topologies (grid and random), different number of data flows and different number of nodes. The study was based on increasing number of radio interfaces (1 to 8) for each mesh nodes. The results show that by increasing the number of interfaces it is possible to increase network capacity by enhancing of QoS parameters. For all simulations of WMN with common channel assignment method CCA, the number of six radio interfaces appears as an optimum solution, because the further increasing of the number of interfaces improved the capacity of WMN only slightly and using more than seven radio interfaces

For further increasing of network performances more sophisticated channel assignment algorithms were used. The new channel assignment approach called First random channel assignment (FRCA) was compared with existing channel assignment algorithms (CCA, LACA). The results show that by using the suitable CA algorithm it is possible to further increase the network capacity. From all results it can be concluded that the multi interface approach with suitable CA algorithm can dramatically increase the whole network performance. In that case, if it is used the simplest CA approach (CCA), we need to assign for each node up to 6 radio interfaces to maximize network performance, but by using suitable dynamic CA algorithm (e.g. FRCA or LACA), the network performance may be maximized

*Department of Telecommunications and Multimedia, University of Zilina, Slovakia* 

This work was supported by the Slovak Scientific Grant Agency VEGA in the project No.

Adya, A.; Bahl, P.; Padhye, J.; Wolman, A. & Lidong Zhou (2004). A multi-radio unification protocol for IEEE 802.11 wireless networks, *Broadband Networks*, 2004. BroadNets 2004,

Badia, L.; Conti, M.; Das, S. K.; Lenzini, L. & Skalli, H. (2009). Routing, Interface Assignment and Related Cross-layer Issues in Multiradio Wireless Mesh Networks, in *Handbook of Wireless Mesh Networks*, Springer Publishers, ISBN 978-1-84800-908-0, London 2009 Calvo R. A. & Campo J. P. (2007). Adding Multiple Interface Support in NS-2, Jan. 2007. Chi Moon Oh; Hwa Jong Kim; Goo Yeon Lee & Choong Kyo Jeong (2008). A Study on the Optimal Number of Interfaces in Wireless Mesh Network, In *International Journal of Future Generation Communication and Networking*, IJFGCN Vol. 1, No. 1, pp. 59 – 66, Dec. 2008

**6. Conclusion** 

decreased the network performance.

with only 4 radio interfaces.

Stefan Pollak and Vladimir Wieser

pp. 344- 354, 25-29 Oct. 2004

**Author details** 

**Acknowledgement** 

1/0704/12.

**7. References** 

Figure 25 shows average values of average jitter. The best values of average jitter were again reached with FRCA algorithm for 4 radio interfaces (124.8 ms). CCA algorithm reached the best value for 6 radio interfaces (601.25 ms) and LACA approach for 4 radio interfaces (167. 27 ms).

**Figure 24.** Values of packet loss for various radio interfaces and different CA schemes

**Figure 25.** Average values of jitter for various radio interfaces and different CA schemes

Channel Assignment Schemes Optimization for Multi-Interface Wireless Mesh Networks Based on Link Load 101
