08 01 01

Hybrid Wireless-Optical Broadband Access
Network (WOBAN): Capacity Enhancement for
Wireless Access
Abu (Sayeem) Reaz, Vishwanath Ramamurthi, Suman Sarkar, Dipak Ghosal, and Biswanath Mukherjee
University of California, Davis, USA
Email: {asreaz,rama,sumsarkar,dghosal,bmukherjee}@ucdavis.edu
Abstract— Noting that the optical part of a hybrid wireless- average packet delay in the wireless mesh of WOBAN relative
optical broadband access network (WOBAN) has high capacity, we to previous approaches. An improved algorithm for Capacity
need to enhance the capacity for wireless access using a low-cost and Delay-Aware Routing (CaDAR) [3] exploits the work by
solution. Our prior work developed a solution where each wireless
node was equipped with a single radio. Deploying multiple radios, Fratta et al. [4] and Kleinrock [5] on capacity assignment (CA)
say two, at each node will improve the performance of wireless and flow assignment (FA), which were originally designed for
access, but this will also increase the cost of the solution. However, general packet networks using optimal capacity assignment
deploying multiple radios at only a few nodes, especially those and flow deviation on links to minimize delay.
that are overloaded with traffic, can lead to a less-costly solution, As mentioned earlier, equipping nodes in the wireless part
possibly without sacrificing performance. Hence, we study how
to optimally place a limited number of additional radios at the of a WOBAN with multiple radios enables the WOBAN to
wireless nodes to save the overall network cost. We formulate this carry more traffic. In [6], the authors propose a two-radio
problem as an Integer Linear Program (ILP) and solve it using wireless mesh architecture. They exploit spatial reuse and self-
a standard solver such as CPLEX. As expected, by deploying organize the network for channel allocation through clustering.
multiple radios at bottleneck wireless nodes, we can obtain almost In [7], the authors present an algorithm called “Localized
the same performance as a WOBAN with multi-radios at all nodes.
sElf-reconfiGuration algOrithms” (LEGO) which detects the
Keywords: Wireless-optical hybrid network, wireless mesh failures locally and generates a local network reconfigura-
network (WMN), multiple radios, routing, delay, capacity tion plan and self-heals the multi-radio wireless mesh. In
assignment. [8], the authors compute the minimum number of WMN
nodes that operate as relay stations and their corresponding
I. I NTRODUCTION channel configurations such that a pre-specified subscribers
A hybrid Wireless-Optical Broadband Access Network traffic demand can be satisfied. In [9], optimization models
(WOBAN) is an optimal combination of an optical back-end to minimize cost are proposed for planning a WMN while
(also called optical backhaul) and a wireless front-end for an providing full coverage to wireless mesh clients, and taking
efficient access network [1]. At the back-end of the network, into account traffic routing, interference, rate adaptation, and
Optical Line Terminal (OLT) resides in the Central Office channel assignment. In [10], the authors address the issue of
(CO) and is connected via optical fiber to multiple Optical placing wireless nodes and to find the minimal configuration
Network Units (ONU). At the front-end, a set of wireless to satisfy network coverage, connectivity, non-uniform Internet
nodes (routers) forms a wireless mesh network (WMN). End traffic demand, and the candidate positions for wireless nodes
users, both mobile and stationary, connect to the network are pre-decided. These works, however, did not study the
through these nodes, whose locations are fixed in a WMN. A radio-assignment problem in a WMN.
selected set of these nodes, called gateways, are connected to One of the benefits of using WMN as the front-end of a
the optical part of the network. Usually, gateways are attached WOBAN is its cost effectiveness [1]. So, it is important to a
with one of the ONUs [1]. Figure 1 shows the architecture of design the WMN part that operates with high performance
a WOBAN. and is cost-effective. Deploying multiple radios at nodes
An end user sends packets to a nearby wireless node improves the performance of WMN. But multi-radio nodes
of the WOBAN. These packets travel through the wireless are significantly more expensive than single-radio nodes [11].
mesh, possibly over multiple hops, and reach the OLT via If deploying multiple radios at a few strategic nodes can
the gateways. As the optical part of a WOBAN has higher give almost the same performance as multiple radios at every
capacity compared to the wireless part, capacity enhancement node of the WMN, we can obtain a cost-effective and high-
of the wireless nodes is essential to support higher traffic in performance WMN. In our work, we study the impact of radio
a WOBAN. For capacity enhancement, wireless nodes need to assignment to the nodes and see how only a few nodes with
be equipped with multiple radios (MR) which can enable the multiple radios can achieve a desired performance.
nodes to carry higher traffic from the end users. The rest of the study is organized as follows: Section II
Several routing algorithms have been proposed for WOBAN. discusses the requirement of multiple radios in a WOBAN.
Delay-Aware Routing Algorithm (DARA) [2] reduces the Section III presents the impact of multiple radios. Section IV

978-1-4244-2324-8/08/$25.00 © 2008 IEEE.
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2008 proceedings.

Fig. 1. Architecture of a WOBAN.

presents how multiple radios can be assigned in a WOBAN op-
timally. Illustrative numerical examples are shown in Section ux vx wx
V. Section VI concludes the study.

II. R EQUIREMENT OF M ULTIPLE R ADIOS IN WOBAN uv vw wx
Figure 2 shows some link flows on a WMN, which is a
multi-hop wireless network. Node u sends data to node x u v w x
with traffic intensity γux . Similarly, nodes v and w send data
to node x with traffic intensities γvx and γwx , respectively. Fig. 2. Flows on links in a multi-hop wireless network.
Now, as we can see in Fig. 2, link (u, v) carries traffic
from node u only. On the other hand, link (w, x) carries remain under-utilized. Figure 3 shows the load distribution
traffic from nodes u, v and w. So, we can find the flows among the links for the topology in Fig. 4 for a load of 5.85
on each of the links as λuv = γux , λvw = γux + γvx , and Mbps at each node for CaDAR [3] where each node is equipped
λwx = γux + γvx + γwx . As the flow on a link cannot be with a single radio and has a capacity of 54 Mbps, as in
greater than its capacity, a link with higher flow, e.g., link IEEE 802.11g [12]. We observe that more than half the links
(w, x), becomes more saturated than a link with lower flow, remain idle while about 4% of the links carry more than 7.68
e.g., link (u, v), and becomes a bottleneck for the WMN. Mbps of traffic. This may result in capacity exhaustion of the
Now, if node u wants to send more data to node x, it cannot mesh network for a particular routing algorithm. An interesting
do so, because a link downstream, namely link (w, x), is aspect is to study the impact of multiple-radio placements.
over-utilized. So, it is logical to increase the capacity of the As multi-radio wireless nodes are more expensive and require
links that are over-utilized, instead of all links, because links complex channel assignment and scheduling, we can put
with lower flows will remain under-utilized. The capacity of multiple radios only at those nodes which are overloaded
a wireless link depends on the capacity of the radio at the and observe how much traffic a routing algorithm can carry
wireless nodes which the link is connecting. If the outgoing instead of putting multiple radios at every node. This reduces
links of a node are over-utilized, the node may require higher cost and complexity of deployment while providing similar
capacity, and it is a candidate for multiple radios. In a WMN, performance as a WOBAN in which all wireless nodes have
assigning multiple radios only at the appropriate nodes can multiple radios. (In this study, we restrict each node to have
lead to similar performance as that of fully-deployed multi- at most two radios.) A partially-deployed multi-radio wireless
radio WMN (where all nodes have multiple radios) while mesh is a network where only a subset of wireless nodes have
reducing the overall network cost, thereby leading to better multiple radios.
cost-benefit ratio.
Since traffic flows in a WOBAN are between the wireless III. I MPACT OF M ULTIPLE R ADIOS IN A WOBAN
nodes and the gateways and since these flows may be asym- We define ω(N ) to be the set of N nodes of the wireless
metric, some links may get overloaded while others may mesh of a WOBAN and η(u) to be the neighbor set of node u.

978-1-4244-2324-8/08/$25.00 © 2008 IEEE.

Flow 0 4% the transmission power of a wireless node. Hence, we refer to
Flow (0,2.56]
Flow (2.56,5.12] 9% this as “throughput-delay ratio” or TDR of the system.
Flow (5.12,7.68] Each link in a WMN has a throughput and a delay [2], [3],
Flow > 7.68
so each link can be considered as a system. So, for a wireless
link (u, v), if the flow is λuv , and delay for independent
arrivals is μCuv1 uv , then TDR of link (u, v) is:
−λ
20%
52% puv = λuv (μCuv − λuv ); u ∈ ω(N ); ∀v ∈ η(u) (3)
We want to design the network in such a way that, on average,
the ratio of throughput and delay is balanced on each link. So,
network-wide average TDR of all the links, P , is:
N N
15% 1
P = λuv puv ; u ∈ ω(N ); ∀v ∈ η(u) (4)
γ u=1 v=1
Fig. 3. Distribution (%) of flow per link for CaDAR [3].
If we maximize the average TDR of the network, we obtain,
Node u has a transmission link to each of its neighbors. In on average, delay sensitivity and high throughput for each link
general, the flows on node u’s transmission links, λuv ; ∀v ∈ of the network. If we have R radios to be assigned to the
η(u) could be different. To minimize delay, radio capacity at network (including the original single radio at each node), we
node u, ζu , should be distributed properly among the links need to determine how to assign the additional radios while
(u, v); ∀v ∈ η(u) based on their respective flows. Hence, maximizing P . Each node needs to have at least one radio.
capacities on these links, Cuv ; ∀v ∈ η(u) need to be assigned Hence, we have R − N radios to be distributed. Let us use a
based on λuv ; ∀v ∈ η(u). Thus, capacities on any outgoing binary indicator, xu , to identify if a node u should be equipped
link from u can be derived as [3], [5]: with multiple radios. Then, we can determine a set of the
√ nodes where the R − N additional radios (the excess capacity
λuv (ζu − v λμ ) λuv
uv
of the network) should be deployed. For any given routing on
Cuv = + √ ; u ∈ ω(N ); ∀v ∈ η(u)
μ v λuv the WMN part of WOBAN (λuv ), we develop a mathematical
(1) formulation to maximize the average TDR of the network to
In Eqn. (1), λμ represents the minimum required capacity
uv
find xu . This formulation turns out to be the following Integer
(Cmin ), the second term represents the fraction of excess Linear Program (ILP):
1
capacity (CA ) assigned to the link, and μ is the average packet Maximize:
size. Now, if any link (u, v) becomes overloaded, ζu needs N N
1
to be increased. In the wireless part of a WOBAN, ζu can be P = λ2 (μCuv − λuv ); u ∈ ω(N ); ∀v ∈ η(u) (5)
uv
increased only by increasing the number of radios. So, if the γ u=1 v=1
capacity of a radio is CR and node u is equipped with two with respect to the following constraints:
radios, ζu = 2 × CR . Then, CA is going to be higher and will
enable link (u, v) to carry higher flow. N

If γ is the total traffic generated in the network, the system ζu = Cuv ; u ∈ ω(N ); ∀v ∈ η(u) (6)
delay (i.e., average network-wide packet delay) [5] can be v=1

expressed as: μCuv ≥ λuv ; u ∈ ω(N ); ∀v ∈ η(u) (7)

1
N N
λuv xu ∈ {0, 1}; u ∈ ω(N ) (8)
T = ; u, v ∈ ω(N ); u = v. (2)
γ u=1 v=1
μCuv − λuv ζu = CR (1 + xu ); u ∈ ω(N ) (9)
When a node is equipped with multiple radios, it can support N

higher traffic in the network (γ) while the system delay (T ) xu = R − N ; u ∈ ω(N ) (10)
is reduced. u=1

The objective function, Eqn. (5), maximizes P . Among the
IV. O PTIMUM P LACEMENT OF R ADIOS IN WOBAN constrains, Eqn. (6) indicates that the capacity of any node,
In a network, the desired properties are high throughput u, is distributed among its outgoing links, (u, v), where v is
and low delay. Since these are competing metrics, Kleinrock a neighbor of u. This constraint assigns the capacity on the
et al. have introduced the notion of “power” [13]. Power of links and bounds the link capacity by the capacity of the node.
a system is defined by the ratio: P ower = T hroughput . It
Delay Equation (7) states that the flow on a link cannot be greater
gives a natural measure for representing any system, where the than the link’s capacity. Equation (8) introduces the binary
system needs to operate with high throughput and lower delay. indicator variable for each node, u. If any node u should be
However, in a wireless network, the term “power” represents equipped with multiple radios, the value of xu is 1, otherwise

978-1-4244-2324-8/08/$25.00 © 2008 IEEE.

0.7
the value is 0. Equation (9) gives the capacity of nodes. Each
node should have at least one radio; and depending on the
0.6
value of xu (Eqn. (8)), any node u can have two radios,
and can have higher radio capacity. Equation (10) bounds the 0.5
0 MR nodes
2 MR nodes
number of radios to be distributed by xu by the number of

System delay (msec)
12 MR nodes
additional radios that needs to be distributed. 0.4

V. I LLUSTRATIVE N UMERICAL E XAMPLES
0.3

0.2

0.1

0
0 2 4 6 8 10 12 14
Load at each node (Mbps)

Fig. 5. System delay vs. load with 0, 2, and 12 MR nodes.

4
x 10
8

7

6

TDR of the network (Mb/s2)
5

4
0 MR nodes
Fig. 4. Topology used for analytical study. 3 2 MR nodes
12 MR nodes
We use the 25-node network configuration in Fig. 4 to 2
study the advantage of deploying multiple radios. Each ra-
1
dio has a capacity of 54 Mbps. We analyze the impact of
different number of radios in WOBAN and how it affects the 0
0 2 4 6 8 10 12 14
performance. We use CaDAR [3] to obtain the flows on the Load at each node (Mbps)
network. Performance of CaDAR compared to other schemes
is presented in [3]. For radio distribution, we evaluate the ILP Fig. 6. TDR vs. load with 0, 2, and 12 MR nodes.
described in Section IV and distribute the radios to different
nodes. We solved the ILP using ILOG CPLEX 9.0 on a Ubuntu
Linux operating system on a Intel Core 2 Duo machine with significantly while carrying higher traffic. We observe that,
1 Gigabyte RAM. The ILP was solved in 0.01 second. In the when the number of MR nodes is increased from 2 (two
illustrative examples below, ‘MR’ represents ‘multi radio.’ For gateways) to 12 (almost half the nodes), the TDR of the
MR nodes, we consider two radios and study their impact. This network is increased, but the highest TDR value is achieved
can also be scaled to a higher number of radios per node. for both 2 and 12 MR nodes at the load of 8.3 Mbps. Since
In the WMN part of a WOBAN, due to its multi-hop nature, TDR is the ratio of throughput and delay, the reason for higher
traffic gets aggregated as it moves closer to the gateways (Fig. TDR achieved by higher number of MR nodes is the lower
4). So, some of the nodes, particularly the gateways, become system delay at the same load of 8.3 Mbps (see Fig. 5). This
the bottleneck of the network. Figure 5 shows the impact of indicates that, even if we assign more radios in the network, the
using multiple radios in a WOBAN. We see that the network throughput is not increased because of the bottleneck nodes.
can carry almost twice the traffic than a single-radio WOBAN if Rather, multiple radios at more nodes reduce the average
multiple radios are assigned to the two gateways (Fig. 4) and packet delay and lead to higher average TDR of the network.
the rest of the nodes have one radio each. If we increase the Figure 7 illustrates the effect of the number of MR nodes
number of MR nodes to 12, we see that the network carries on WOBAN. We observe that, as the number of MR nodes
almost the same traffic because the capacity of the bottleneck increases in the WMN of WOBAN from 12 to 25 in the 25-
nodes remain the same. However, it reduces the network-wide node topology, the system delay decreases by about 0.02 msec
average packet delay by about 0.02 msec. for loads higher than 4 Mbps while supporting almost the same
Figure 6 shows that, by assigning multiple radios at the amount of traffic. So, we deduce that, after assigning multiple
bottleneck nodes, the average TDR of the system is increased radios to the bottleneck nodes, we cannot gain much with MR

978-1-4244-2324-8/08/$25.00 © 2008 IEEE.

0.4 4
x 10
2 4

0.35 Delay
TDR

TDR of the max delay path (Mb/s )
Max delay on a single path (msec)

2
0.3 12 MR nodes
16 MR nodes
System delay (msec)

25 MR nodes
0.25

1 2
0.2

0.15

0.1

0.05
0 2 4 6 8 10 12 14 0 0
Load at each node (Mbps) 0 2 4 6 8 10 12 14
Load at each node (Mbps)

Fig. 7. System delay vs. load with 12, 16 and 25 MR nodes. Fig. 9. Delay and TDR of maximum-delay path for 12 MR nodes.

nodes. deployment scheme using an ILP for WOBAN. We observed
4
that, if multiple radios are assigned at only the bottleneck
x 10
9 nodes (e.g., gateways), there is significant performance im-
provement. After that, increasing the number of radios does
8
not improve the performance much. We conclude that we can
7 increase the capacity of the WMN part of a WOBAN in a cost-
TDR of the network (Mb/s2)

6
effective way by assigning multi-radios only at selected nodes.

5
R EFERENCES
4 12 MR nodes [1] S. Sarkar, S. Dixit, and B. Mukherjee, “Hybrid wireless-optical broad-
16 MR nodes band access network (WOBAN): A review of relevant challenges,”
3 25 MR nodes IEEE/OSA Journal of Lightwave Technology, vol. 25, no. 11, Nov. 2007.
[2] S. Sarkar, H. Yen, S. Dixit, and B. Mukherjee, “A novel delay-aware
2 routing algorithm (DARA) for a hybrid wireless-optical broadband
access network (WOBAN),” IEEE Network, vol. 22, no. 3, pp. 20–28,
1 May-Jun. 2008.
[3] A. Reaz, V. Ramamurthi, S. Sarkar, D. Ghosal, S. Dixit, and B. Mukher-
0 jee, “CaDAR: an efficient routing algorithm for wireless-optical broad-
0 2 4 6 8 10 12 14
Load at each node (Mbps) band access network,” in Proc. IEEE ICC, Beijing, China, pp. 5191–
5195, May 2008.
Fig. 8. TDR vs. load with 12, 16, and 25 MR nodes. [4] L. Fratta, M. Gerla, and L. Kleinrock, “The flow deviation method:
An approach to store-and-forward communication network design,”
Networks, vol. 3, no. 2, pp. 97–133, Mar. 2007.
Figure 8 shows that, as the number of MR nodes increases, [5] L. Kleinrock, Queueing Systems, Volume II: Computer Applications,
the TDR is increased. But we observe that the maximum TDR Wiley-Interscience, 1976.
[6] J. Zhu and S. Roy, “802.11 mesh networks with two-radio access points,”
of the network is achieved while carrying almost the same in Proc. IEEE ICC, Seoul, Korea, pp. 3609–3615, May 2005.
traffic. This is consistent with the results in Fig. 7. [7] K. Kim and K. Shin, “Self-healing multi-radio wireless mesh networks,”
Figure 9 shows the delay and the TDR on the maximum- in Proc. ACM MobiCom, Montreal, Canada, pp. 326–329, Sep. 2007.
[8] A. So and B. Liang, “Minimum cost configuration of relay and channel
delay path between a wireless node and a gateway for 12 MR infrastructure in heterogeneous wireless mesh networks,” in Proc. IFIP
nodes. We observe that the TDR is maximized at a load of 8 Networking, Atlanta, GA, pp. 275–286, May 2007.
Mbps, which is also observed for the network-wide average [9] E. Amaldi, A. Capone, M. Cesana, I. Filippini, and F. Malucelli, “Op-
timization models and methods for planning wireless mesh networks,”
packet delay in Figs. 6 and 8. This indicates that the maximum Computer Networks, vol. 52, no. 11, pp. 2159–2171, Aug. 2008.
TDR averaged over all the links of the network is also obtained [10] J. Wang, B. Xie, K. Cai, and D. Agrawal, “Efficient mesh router
for a single path. placement in wireless mesh networks,” in Proc. IEEE International
Conference on Mobile Adhoc and Sensor Systems (MASS), Pisa, Italy,
pp. 1–9, Oct. 2007.
VI. C ONCLUSION [11] Wi-Fi Planet, “Wi-fi products.” http://products.wi-
In a wireless-optical broadband access network, the optical fiplanet.com/wifi/mesh/.
[12] http://standards.ieee.org/getieee802/802.11.html.
backhaul generally has higher capacity than the wireless front- [13] C. Kolias and L. Kleinrock, “The power function as a performance
end. So, we need a cost-effective way of increasing the and comparison measure for ATM switches,” in Proc. IEEE Globecom,
capacity of the wireless access network by assigning multiple Sydney, Australia, pp. 381–386, Nov. 1998.
radios at selected nodes. We presented a optimum multi-radio

978-1-4244-2324-8/08/$25.00 © 2008 IEEE.

08 01 01

More Related Content

What's hot (20)

Viewers also liked (8)

Similar to 08 01 01 (20)

More from Shahab Shahid (14)

08 01 01