Traffic Engineering with Distributed Dynamic Channel Allocation in

BFWA Mesh Networks at Millimeter Wave Band

Junaid A. Khan and Hussein M. Alnuweiri
Department of Electrical & Computer Engineering, 2356 Main Mall,
University of British Columbia, Vancouver, B.C. Canada, V6T 1Z4
Phone: +1 (604) 822-9185
Emails: {junaidk,hussein}@ece.ubc.ca
physical WMN and its network architecture is presented in
Fig.1. Each node in this network is a wireless router capable
of routing traffic from a source node to a destination node.
Some of these nodes are connected to access networks, some
are points of interface to the core network (internet) or some
other networks and some are just relay nodes. Each node is
connected to other nodes through point-to-point wireless
links using directional antennas. The whole network is a
multipoint-to-multipoint network, where a traffic flow may
use multiple hops to reach destination node. As shown in
Fig.1 (a), with WMN it is possible to walk around the tress
that may not be possible for a PMP network.
Each link in the wireless mesh network under consideration
consists of several radio channels. It is possible by using
adaptive modulation techniques that a wireless link will be up
most of the time at the expense of capacity. It is also possible
to control capacity of each link by allocating or removing
wireless channels from the link. Two adjacent nodes can
communicate to allocate or remove a channel from a link.
Only constraints is that no two links connected to the same
node can share a channel to avoid interference. In the
presence of directional antennas the interference is reduced
considerably but at the same node transmitting signal on one
link can interfere with the receiving signal on another link
connected to the node. Also each node is aware of channels
allocated to the links connected to it, and consequently of the
capacity of the links. The node also knows the reserved
bandwidth (or bandwidth utilization) on such links all the
time. It is appropriate to assume that MPLS (Multi Protocol
Label Switching) tunnels are operational in the network by
reserving the required bandwidth on each route. This
assumption provides a better approximation of bandwidth
utilization (or reserved bandwidth) on each link. Another
assumption is the use of time division duplexing (TDD) on
each link. TDD makes it possible to use same channel on a
link for traffic on both directions. TDD avoids the unused
channel problem present in Frequency Division Duplexing
(FDD) and hence provides better link utilization. In FDD it is
possible that for a short period of time there is no traffic in
one direction and thus the frequency channel allocated to that
direction was not in use, furthermore it is not appropriate to
reallocate channels for such a short period of time. Whereas
TDD shares the channels on both directions, thus if there is
no traffic in one direction for a short time then all the
channels can use most of their time in one direction with

AbstractInherent difficulties in millimeter-wave radio

operations, such as higher atmospheric attenuation,
especially during rainy times, motivated the use of mesh
architecture in millimeter-wave band for broadband fixed
wireless access (BFWA) networks. When used with highly
directional antennas, these mesh networks also provide
better frequency reuse. In a recent proposed architecture
for such networks, a link can have multiple radio
channels. However, to provide traffic engineering with
scalability, it is needed to develop a distributed dynamic
channel allocation algorithm to allocate channels to these
links. This paper proposes a distributed dynamic channel
allocation algorithm that is scalable and able to provide
traffic engineering if invoked periodically. The proposed
solution provides traffic engineering by optimizing link
capacities by adding or removing channels from a link
while maintaining interference constraints, based on
current network conditions. Simulation results suggested
that proposed algorithm performs better than a solution
based on fixed channel allocation.
Index TermsMesh Networks, Wireless Broadband,
Dynamic Channel Allocation, Traffic Engineering,
Millimeter Wave Band.
I INTRODUCTION
nherent operational difficulties at licensed millimeter-wave
radio bands, such as higher atmospheric attenuation,
especially during rainy times, limits the range of
millimeter-wave links to a few kilometers (1-2Km).
Furthermore, Line of Site (LOS) is necessary for operation at
these frequencies, however multipath effect is not that
significant. These requirements make it harder to use these
frequencies in traditional Point to Multi Point (PMP) network
architecture because it needs expensive base station every
few kilometers. Authors in [1] have proposed a mesh
architecture to overcome these difficulties. The work in this
paper is based on this particular wireless mesh network
(WMN) architecture.
There are two main categories of WMNs, logical mesh,
and physical or directed mesh [2][3]. Logical mesh typically
uses omni-directional antennas to minimize complexity. On
the other hand, physical mesh requires the use of strongly
directional antennas to reduce interference. An example

II PROBLEM DESCRIPTION AND FORMULATION

The objective of TE with dynamic channel allocation is to
assign channels in the available spectrum, using a distributed
dynamic algorithm, to each link in the network, such that
a. No two links connected to the same node share the
same channel. This is to avoid interference.
b. Each link has at least one channel.
c. Most of the available channels in the spectrum are
used/reused to achieve better resource utilization. In
other words, summation of link capacities is
maximized.
d. The distribution of channels is balanced. For an
unloaded network, it means that channels are
distributed as evenly as possible. For a network with
some bandwidth reservation, it means that channels are
allocated such that the utilization of links remains
balanced.
e. Maximum link utilization is minimized.
Constraint (a) is to avoid interference between adjacent
links (links connected to same node). Although the use of
directional antennas reduces this possibility, the presence of
side lobs in antenna pattern does not eliminate interference
completely. As mentioned earlier, transmitted signal on a link
can interfere with receiving signal on another link which is
sharing the same node and same channel. It is, therefore,
needed to assign different channels to adjacent links.
However, non-adjacent links have the freedom to choose
same channels, which may not be always possible if omni
directional antennas are used. Constraint (b) is needed to
provide a link with minimum capacity. Objective (c) is used
to avoid waste of resources and to provide more capacity to
links. It is to be noted here that poor channel allocation may
result in many unused channels in a node and hence reducing
the sum of link capacities. Objective (d) provides a balanced
capacity planning. Objective (e) is byproduct of objective (d)
because a dynamically reconfigure able load balanced
network is also optimized with these two objectives. One can
argue that, channel allocation to links must be coupled with
already estimated traffic demand. However, in real life it is
not possible to perfectly predict future traffic demand [4]. In
such a case, a distributed and dynamic channel allocation
helps to provide better traffic engineering, where channels
may be re-allocated, periodically, based on current network
load conditions.

B u il d i n g 1
C o n d os

T re e

T re e
Tr e e

O u td o o r m a l l

T re e

House

T r ee

T re e
S ch o o l
T re e

B u il d i n g 1
T re e
T re e
T re e

T r ee
Condos

H o s p i ta l
S k y s c ra p e r

(a)

(b)

Fig. 1 Wireless Mesh Network Architecture.

TDD bursts, providing better link utilization at that time. The
last assumption is the presence of a control radio channel in
the system. Although the same control radio channel is used
on each link, using a contention avoidance mechanism such
as CSMA/CA (Carrier Sense Multiple Access with Collision
Avoidance) can eliminate interference problem between two
adjacent links. The use of CSMA/CA may introduce some
overhead, but for the purpose of control signals, it may not be
very significant. However, once a channel is assigned to a
link, then the control radio channel is no longer needed for
that link. Control channel is needed only at the initial network
configuration or when a node joins a network. Also, each
node is capable of detecting the presence of a link, and if a
link is down removes all channels from this link.
In spite of all the advantages, WMNs are still in their early
stage of development and several real problems remain to be
solved before an efficient deployment of WMNs is achieved.
This work proposes to solve one such issue, i.e. to provide
Traffic Engineering (TE) solution in physical WMN. TE is a
proven technique, which is used with considerable success in
wired networks. TE's main objective is to increase network
utilization and maximize the number of admitted connection
through load balancing. There are two ways to provide TE in
a network; (a) by using a well-designed TE based routing
algorithm to avoid hot spots or (b) by reconfiguring the
network to eliminate hot spots. It is obvious that hot spot
elimination is a better solution and can be achieved by
increasing the link capacities of congested links. However, in
wired networks it is not possible to change link capacities and
thus they rely on solution (a) in general. Whereas, it is
possible in WMNs to change link capacities by assigning or
removing channels from each link. The proposed solution has
adopted the approach in (b). For this purpose, a distributed
solution to the problem of channel allocation to links in a
physical WMN is developed; when each node invokes this
algorithm periodically it optimizes the network link
capacities dynamically to provide TE.
The rest of the paper is organized as follows. Section 2
covers problem formulation. Section 3 covers some previous
work. Proposed solution is presented in Section 4. Section 5
covers algorithm analysis. The algorithm performance is
illustrated with simulation results in Section 6, and the paper
is concluded in Section 7.

Mathematical Model
A WMN is represented by a graph G(N,L), where N is the
set of nodes and L is the set of links. Let n, and l, be the
wireless node and wireless link IDs. Other parameters are
defined as follows
En : Set of links connected to node n.
Cs : Set of radio channels in the available spectrum.
Un : Set of unused radio channels in node n.
Rl : Set of radio channels allocated to link l.
cl : Single channel capacity on link l.
rl : Reserved bandwidth on link l.

l : Bandwidth utilization of link l.

dynamic solution to minimize the number of links which

interfere with each other while using the same frequency.
Their decision criteria, however, was not given in a formal
algorithm to solve the problem. Also the objective of their
paper is to minimize interference rather than avoiding
interference completely. Almost the same group of authors
presented a solution for simultaneous routing and channel
allocation in [6]. Their objective was to provide traffic
engineering in WMN by reconfiguring all the routes and
channel allocation every few minutes. However, with the
presence of large number of routes, it is not feasible to
recalculate routes and channel allocation and to restore new
paths every few minutes. Furthermore, interference is not
considered in this work. The entire solution is based on a
centralized server, thus limiting the scalability of WMN. The
centralized server introduces an extra bottleneck, because
failure of the server may stop the whole process.

SDn : Standard deviation of bandwidth utilization of En.

SDRn : Standard deviation of the number of radio channels
allocated to links in En.
The objective of the distributed dynamic channel allocation
problem is to assign channels in Cs to every set Rl (l L)
dynamically to optimize certain objectives while fulfilling
certain constraints. Formally,
Minimize
SDn ; if network has some
n N

(1)
reserved bandwidth

SDRn ; otherwise

n N
Minimize U n
(2)
n N

Minimize MAX (l)

Constraints

Ri I R j = ... i, j same E n ;
n N

Rl > 0 ..... l L

(3)

IV PROPOSED SOLUTION

Distributed Dynamic Channel Allocation

The proposed solution to the problem employs message
passing. Each node is capable of invoking the channel
allocation process based on certain conditions. Whenever, a
node joins a network, or periodically, the node checks if there
is a need to invoke local channel allocation/reconfiguration
process. A node n invokes channel allocation/reconfiguration
process when at least one of the following conditions is true.
Condition 1: There are unused channels available in node n
i.e. |Un| > 0.
rA
rB
Condition 2: If
for any two
+ <
c A ( R A 1)
c B RB
links A and B connected to node n, and A has at
least two channels available, > 0 is a
constant.
The first condition is self-explanatory and is used to
achieve objective (2). Second condition states that removing
a channel from link A and allocating it to link B will provide
a better load balancing thus optimizing other objectives. A
positive threshold of (e.g. = 0.1) is added to avoid
instability or rapid changes in WMN system.
If a node n needs to allocate or reconfigure the channel
allocation to the links connected to it, node n sends a channel
allocation request (CR) message to the node on a selected
link indicating that n needs an extra channel on the link. The
link with highest bandwidth utilization is selected by node n
to send CR message. A tie (specially if there is no bandwidth
reservation) is broken by selecting the link with smallest |Rl|.
This optimizes objective in (1) for unloaded network.
Selecting a link at random breaks further ties.
A flow of CR and its responses is shown in Figure 2. In
this figure, node x has decided that it needs an extra channel
on link connecting x and y. Node x sends a CR message with
list of available channels for link xy to node y. List of
available channels in a node n for link B are the channels in

(4)
(5)

The objective functions in (1) represent two scenarios. In

the first scenario, the network has reserved some bandwidth
based on connection requests seen before. In this case, the
objective is to assign channels to links, such that links
connected to the same node will have nearly the same amount
of bandwidth utilization. The second scenario represents the
moment when the network is initialized with no reserved
bandwidth. Recall that a link is a grouping of channels. In
such case, the objective is to assign nearly the same number
of channels to all links connected to the same node. This
objective ensures that if a link connected to a node has high
bandwidth utilization then it should get a channel from one of
its neighboring links to reduce utilization. The objective
function in (2) increases the efficiency of channel usage by
minimizing the number of unused channels (maximizing the
channel reuse) in every node in the network. Note that if all
the channels for n are in use, then |Un| = 0. The objective in
(3) is a byproduct of objective in (1) as every node tries to
reduce the maximum utilization of links connected to it by
reducing the standard deviation of the links utilizations.
There are three constraints in channel allocation.
Constraint (4) states that no two links connected to the same
node can share a channel. This constraint avoids channel
interference on adjacent links. Constraint (5) makes sure that
each link has at least one radio channel assigned to it to avoid
zero capacity links.
III PREVIOUS WORKS
Despite the need for an efficient distributed dynamic
solution for this problem, surprisingly not many have been
reported in the literature. In [5], the authors presented a

1. CR

1. CR

2. +ve ack

2. -ve ack.
with no
available
channels

3. +ve/-ve
confirm.

(a)

3. RR

1. CR

4. -ve ack.
for RR

2. -ve ack.
with some
available
channels

(b)

(c)

3. RR

1. CR

4. -ve ack.
for RR

2. -ve ack.
with some
available
channels

5. -ve/+ve
confirm.
for RR

6. +ve
confirm. for
-ve ack.

(d)

Fig. 2 Flow of channel request (CR) and its responses. (a),(b),(c) and (d) represent different scenarios.
Ux and all the channels on links connected to x (link A in
condition 2) which satisfy condition 2.
Fig. 2 shows four different response scenarios. In scenario
2(a), node y sends a positive acknowledgement that it has an
available channel for xy that matches with one of the
requested channels in CR. Node x then sends a positive
confirmation to y to confirm allocation of the channel on the
link. In this case node x and y also removes the channels from
other links if it is assigned to some other links (not shown in
figure for simplicity). However, it is possible that node x has
received a CR from a node other than y and has already sent a
positive acknowledgment for the same channel. In this case,
node x sends a negative confirmation to y. Scenario 2(b)
shows a case with negative acknowledgment of CR. It is also
possible that node y does not have an available channel that
matches a channel in CR, but it has some other available
channels for xy. In such case, node y sends a negative
acknowledgment with the list of these channels as shown in
2(c).
When node x receives this kind of negative
acknowledgment then it sends a replacement request (RR) to
its neighboring nodes. These are the nodes, which are
connected by links having at least one channel common with
those listed in the negative acknowledgment. The RR
message contains the list of channels in the negative
acknowledgment, and list of channels in CR. The purpose of
RR is to replace a channel in a link xz that matches the
available channel in y with an available channel in x. If z does
not accept RR then it sends a negative acknowledgment for
RR. However, if z accepts the RR based on channels in Uz, or
on condition 2 for link xy as shown in 2(d), then it sends a
positive acknowledgment. After receiving positive
acknowledgment for RR, node x checks whether it has
already received another positive acknowledgement from
some other node and replied to it. If not, x checks for its
available channels again before sending a positive
confirmation to z. After sending a positive confirmation to z,
node x also sends a positive confirmation to y to allocate the
channel that is replaced from link xz to link xy. After
receiving positive confirmations from x both y and z removes
the corresponding channels from other links, if needed. If x
finds that it has already accepted another positive
acknowledgment then it sends a negative confirmation to z. In
such a case, x does not send any message to y.

It is possible that node x is unable to allocate a channel to

link xy in the above algorithm. For further optimization, node
x again checks for need to invoke local channel (re)allocation
process. However, this time node x does not include links to
which it has just failed to allocate a channel. A node x tracks
failed attempts by using a flag for each link connected to it.
This flag for a link is set after sending a CR message on this
link. Once node x achieved a successful channel allocation
then it removes the flag from all links connected to it. The
flags are also removed from all the links connected to node x,
if node x has unsuccessful attempts on all links connected to
it. However, node x does not sends a CR message in such a
case. The flags are removed in this situation because channel
allocation/reconfiguration is not possible at this time but it
may be possible in future. After a successful channel
allocation, a node x again checks for a need to allocate /
reconfigure channel allocation. This is needed because it is
possible that there are some unused channels in Ux which can
be assigned to some links connected to x.
Another scenario that is not mentioned in the abovementioned algorithm is when a node n receives a CR or RR
message from a node m while n itself is waiting for a
response for a CR or RR message that it has sent. In such a
case, n will compare the two links for which it has sent and
received CR or RR messages and chose the one with higher
utilization link. Further ties are broken using the number of
channels assigned to those links followed, by random
selection. If node n decides not to serve an incoming CR or
RR message then it asks node m to check for need for channel
allocation/reconfiguration again and sends the CR or RR
message again, if needed.
V ALGORITHM ANALYSIS

Message Complexity
The message complexity of the proposed algorithm is
defined as the number of messages sent to allocate single
channel. This can be calculated with the help of Fig.2. From
Fig. 2(a) it is clear that best-case message complexity is 3
messages. The worst case scenario can be explained with the
help of Fig. 2 (c) and (d). If node x has |Ux| > 0; to allocate
channels in Ux, x will send CR message (CRM) on every link
in Ex, one by one. In the worst-case scenario, all CRMs,
except the last one, will be unsuccessful and will result in

2,3,9

2,3

8,9

0,7,6

0,6,7,8

1,6,7

1,3

10

1,3,9

13

0,7,6

2,4,8

6
2,4,5

1,2,4

earlier, for single unsuccessful CRM, all RRMs are sent in

parallel, hence it is sufficient to include time for one RRM
and its responses for each unsuccessful CRM. Based on this
worst-case latency for a single unsuccessful attempt is 4T.
For the last successful attempt as shown in Fig.2 (d) latency
is 5T. It is based on the fact that last 2 confirmation messages
are sent in parallel. Therefore for a worst case scenario with
|Ex|-1 unsuccessful CRM and one successful CRM the latency
is
Latencyworst = 4T (|Ex|-1) + 5T = (4 |Ex| + 1)T
(10)

12

2.4.5

0,1,4

5,9

2,7,6
3,5,8,9

1,5,6,7

0,3,8,9

8
4,8,0

11

14
1,3,8,9

0,1,8,4
5,7,6

0,5,7,6

Algorithm Scalability
The proposed algorithm is highly scalable because all the
decisions are made locally and to allocate a single channel to
a link connected to a node n, messages are passed at most to
the extended neighborhood of n. Extended neighborhood of n
are the nodes adjacent to the nodes directly connected to node
n. These nodes will receive the remove channel messages.
Also no node needs network wide information to make a
decision and only depends on local information; hence the
algorithm is highly scalable.

15

Fig. 3 Results for network without any reserved

bandwidth.
LEGEND
(reserved bandwidth, number of channels, utilization)

7
(200,11,0.757)

10

(100,9,0.463)

(100,9,0.463)

11
(200,11,0.757)

13

14

15

(200,13,0.641)

12

VI SIMULATION RESULTS
In order to check our algorithm, we have created a discrete
time simulator. It simulates the message passing between
nodes to simulate the distributed nature of the algorithm. We
have tested our algorithm for a 16 node, lattice type network,
assuming that the number of channels in the spectrum is 10.
In the beginning the network was setup without any reserved
bandwidth. In this scenario the objective of the algorithm is

(200,12,0.694) (200,11,0.757)

(80,8,0.417)
6

(100,8,0.521)

(100,9,0.463)

(80,8,0.417)

(80,8,0.417)

(100,8,0.521)

(200,12,0.694)
8

(100,8,0.521)

(100,8,0.521)

(200,12,0.694)
4

(80,8,0.417)

(100,9,0.463)

(200,12,0.694)

(200,11,0.757) (200,12,0.694)

(200,12,0.694)
0

(200,12,0.694)

to

Fig. 4 Results for static configuration with reserved

bandwidth.
messages as shown in Fig. 2 (c). The last CRM will be
successful and result in messages as shown in Fig. 2(d). The
message complexity for this scenario is calculated as follows.
After receiving ve ack (Ack-CR), with available channels, for
each unsuccessful CRM, x will send at most |Ex|-1 RR
messages (RRM) and will receive |Ex|-1 ve acks for RR
(Ack-RR). Therefore total number of messages for |Ex|-1
unsuccessful CRM will be
(|Ex|-1) (CRM + Ack-CR + (|Ex|-1) (RRM + Ack-RR) ) (6)
In terms of number of messages, it will be
(|Ex|-1) (2 +2 (|Ex|-1)) = 2 |Ex| (|Ex|-1)
(7)
In the same manner for last successful CRM, number of
messages can be calculated as
3 + 3 ( |Ex|-1 ) = 3 |Ex|
(8)
Adding (7) and (8) and after some simplification, message
complexity for worst case scenario is calculated as
2 |Ex|2 +|Ex|
(9)

minimize

SDR
n N

and

. The solution

n N

obtained after running the algorithm is shown in Figure 9. It

is clear from this solution that the difference between the
number of channels on two adjacent links is never more than
one channel. This indicates that the algorithm has
successfully balanced the number of channels on each link.
Also, only the 4 corner nodes (nodes 0,3,12, and 15 in Figure
3) have some unused channels. The reason for these unused
channels is that the corner nodes are connected to only two
links and assigning these unused channels to these links may
cause removal of channels from some other links. It can also
be seen in this solution that there is no interference as no two
adjacent links share the same channel, and every link has at
least one channel assigned to it.
To check the performance of our algorithm when network
has reserved some bandwidth, we have used same lattice time
network. However, this time we have used 33 channels in
LMDS band with 6MHz analog bandwidth per channel. We
consider 16QAM encoding providing 24Mbps digital
bandwidth per channel. We conducted two experiments. In
the first experiment, we use our algorithm initially and load
the network as shown in Figure 4. Initial channel assignment,
reserved bandwidth on links and bandwidth utilization for
this experiment is also shown in this figure. We call this static
configuration, because there is no dynamic channel allocation

Latency to Allocation A Single Channel

The latency to allocate a channel is defined as the time to
allocate a single channel to a link. The unit of time T is
considered as the time to relay one message from one node to
another node. Again best-case latency is calculated to be 3T
based on Fig. 2(a). For worst-case scenario as discussed

LEGEND
(reserved bandwidth, number of channels, utilization)

(80,6,0.55)
6

(200,13,0.641)

(100,8,0.521)

13

10

(100,8,0.521)

14

11
(200,12,0.694)

15

(200,13,0.641)

(100,8,0.521)

(100,7,0.59)
9
(80,9,0.37)

(80,8,0.417)

12

AVG(
l)
MAX(
l)
STD(
l)

Fig. 5 Results for dynamic configuration with

reserved bandwidth.
after the initial assignment. In the second experiment we
consider an instance of dynamic channel assignment. We ran
our algorithm again after loading the network as shown in
Figure 4. The results are shown in Figure 5. It is clear that
after running our distributed algorithm again, the network
becomes more optimized. This is verified after comparing

SD

0.589
0.757
0.131

0.583
0.694
0.098

Table 2: Simulation results for network with reserved

bandwidth.
Network Size
Average Number of
Messages Sent by a
Node
33
10
Channels
Channels
78
25
16 nodes 24 links
89
31
100 nodes 180 links

(200,13,0.641)

Figure 4 and Figure 5 in terms of

n N

(200,13,0.641) (200,12,0.694)

(100,8,0.521)

(200,12,0.694)
8

(100,8,0.521)

(100,8,0.521)

(200,13,0.641)
4

(80,8,0.417)

(100,8,0.521)

(200,12,0.694)

(200,13,0.641) (200,12,0.694)

(200,12,0.694)
0

Table 1: Simulation results for network with reserved

bandwidth.
Static
Dynamic
Configuration Configuration
1.352
1.047
SD

VII CONCLUSION

, average

In this work we have presented a distributed and dynamic

solution to channel assignment problem in BFWA wireless
mesh networks. The proposed algorithm provides traffc
engineering when invoked periodically. The performance of
the proposed solution was supported by simulation and found
to be better than fixed channel allocation. The algorithm is
also highly scalable.
REFERENCE

n N

utilization over all links (AVG(l) ), maximum utilization

over all links (MAX(l) ), and standard deviation of
utilization over all links (STD(l) ). All these results are
shown in Table 1. It is clear from these results that network
performed substantially better after re-running the distributed
algorithm. The lower numbers under the Dynamic
Configuration column indicate better load balancing on all
links. It is also worth to note here that the algorithm is
dynamic in the sense that it performs fine-tuning of previous
channel assignment instead of rerunning the whole
optimization from scratch.

[1]

[2]

Scalability Test
One way to check the scalability of the proposed solution
is to show that average number of messages sent by a node
for initial channel assignment is independent of network size.
For this purpose we have calculated the average number of
messages sent by a node for initial channel allocation. It has
been done for 33 available channels in the spectrum and 10
available channels in the spectrum. The results for this test
are shown in Table 2. It is clear from this table that algorithm
is scalable and average number of messages per node does
not depend upon the size of the network. However, these
depend upon the number of channels available in the
spectrum that is logical that it needs more messages to
allocate more channels. Both networks are of lattice type
networks.

[3]

[4]

[5]

[6]

Yoji Kishi, Satoshi Konishi, Shinobu Nanba, and Shinichi Nomoto, A

proposal of millimeter-wave multi-hop mesh wireless network
architecture with adaptive network control features for broadband
wireless access, In proceedings of IEEE Radio And Wireless
Conference, RAWCON, pp 17-20, August 2001.
Dave Beyer, Wireless mesh networks for residential broadband, In
proceedings of National Wireless Engineering Conference, November
2002.
Paul Piggin, Barry Lewis, and Phil Whitehead, Mesh networks in
fixed broadband wireless access, multipoint enhancements for the
802.16 standard, http://www.ieee802.org/16/docs/ , last accessed May
16, 2005.
Junaid A. Khan and Hussein M. Alnuweiri, A fuzzy constraint-based
routing algorithm for traffic engineering, IEEE GLOBCOM04, vol.
3, pp. 1366-1372, November 2004.
Satoshi Konishi, Shinobu Nanba, Yoji Kishi, and Shinichi Nomoto,
Dynamic and autonomous frequency assignment method for MP-MP
BFWA systems, In proceedings of IEEE Personal, Indoor and Mobile
Radio Communications, vol. 1, pp 359-363, September 2002.
Kishi, Y., Tabata, K., Konishi, S., and Nomoto, S., A proposal of an
adaptive channel allocation and traffic engineering algorithm in multihop mesh networks for broadband fixed wireless access, IEEE
Wireless Communications and Networking, WCNC 2003, vol. 2, pp
1043-1048, March 2003.