Вы находитесь на странице: 1из 6

2011 IEEE 10th Malaysia International Conference on Communications (MICC)

2nd – 5th October 2011 | Sutera Harbour Resort, Kota Kinabalu, Sabah, Malaysia

Capacity and Coverage Analysis of Rural Multi-


Radio Multi-hop Network Deployment using
IEEE802.11n Radios
Alvin Ting, David Chieng and Kae Hsiang Kwong
Wireless Communications
MIMOS Berhad
Kuala Lumpur, Malaysia
{kee.ting,ht.chieng,kh.kwong}@mimos.my

Abstract— This paper presents the capacity and coverage the recently approved IEEE 802.11n standard which offers
performance analysis of a multiradio multihop network physical data rates up to 600Mbit/s and higher resiliency
deployment using IEEE802.11n radios for a typical rural area in towards interference via MIMO technology may change the
Malaysia. Insights on the relationships between various key perception to a certain extent. To date there are already a wide
design parameters particularly backhaul link rate, backhaul link range of 11n-based wireless mesh/multihop network products
distance/multihop distance, coverage size per MAR/total coverage and solutions in the market. However, the performance of such
size, number of MIMO spatial multiplex (SM) stream, number of network is not well understood especially from the capacity and
MAR per branch and effective capacity per access point or per range (coverage) viewpoints. In particular this paper aims to
user are acquired. Two optimization objectives namely 1) to
understand the relationships between various key design
maximize coverage size and 2) to maximize backhaul distance are
parameters such as backhaul link rate, backhaul link
introduced.
distance/multihop distance, coverage size per MAR/total
Keywords-component; Wireless Mesh/Multihop Network; Rural coverage size, number of MIMO spatial multiplex (SM) stream,
Deployment; IEEE 802.11n; Backhaul number of MAR per branch and effective capacity per access
point which can later be translated into data rate per user using
I. INTRODUCTION IEEE802.11n radios. The study takes into consideration the
unique characteristics in rural deployments, i.e. distribution of
Accelerating broadband penetration and bridging the digital the users is rather concentrated like a hotspot but distributed.
divide between rural and urban communities have long been Unlike the urban case, the distance between hotspots may range
one of the main agenda of governments worldwide. Over the from hundreds of meter to tens of km (between villages). Also
years, various initiatives have been launched in Malaysia to unlike the urban zones, contiguous coverage is not required.
provide Internet access to the rural communities. More recently Fig. 1 provides a snapshot of a typical rural deployment in
Ministry of Information, Communications and Culture and the Malaysia.
Malaysian Communications and Multimedia Commission
launched the “Kampung WiFi” initiative [1], which aims to
accelerate the national broadband penetration at the rural areas.
According to NBI, the broadband access in rural areas shall
Branch 1
include Basic Telephony Access (via fixed and mobile
networks), broadband connected Community Broadband
Libraries (CBLs) and Community Broadband Centres (CBCs).
Motivated by the lacked of wired infrastructure in these areas,
Wireless Mesh Network (WMN) has become a highly Branch 3
promising means to provide broadband access. As for the radio
technology, IEEE802.11 WLAN or WiFi is naturally preferred
due to cost factor and wide spread market adoption.
Branch 2
Over the years WMN technologies, particularly those based
on WiFi radios, have evolved from single radio systems to
multiradio systems involving heterogeneous radio interfaces
such as IEEE802.11a, b, g and n. The most commonly known
industrial practice adopts the architecture which comprises of Figure 1. Case Study: Kampung (village) Ulu Dusun , Sabah,
IEEE802.11a radios at the backhaul and IEEE802.11g for the Malaysia.*courtesy of Google Maps.
access. This is largely motivated by the fact that IEEE802.11a
has more non overlapping channels and much less congested We first, we developed an analytical model that includes
spectrum band. Although WMN offers a wide range of benefits, physical layer, mac layer and propagation model to evaluate
it continues to suffer capacity limitation due to excessive the capacity and coverage for a multi-radio multihop
sharing of capacity as the number of hop increases. To this end infrastructure network.. In this model, access radio and

978-1-4577-0978-4/11/$26.00 ©2011 IEEE 77


backhaul radio are considered to operate in two different From this basic architecture, different topologies as shown
frequencies and not interfering to each other. Inter relationship in Fig. 3 can be formed.
between coverage and capacity between access and backhaul B. Topology Formation
planes is then derived. Based on the model, we apply the
greedy algorithm to find the optimal coverage or backhaul An area of 2.25 square km (1.5x1.5 km) of a village called
distant for the considered multihop network. Kampung Ulu Dusun as depicted in Fig. 1 is selected for our
study. It can be observed that houses are not uniformly
The reminder of this paper is organized as follows. Next distributed in this village and they tend to group in some
section discussed on the related works. Section 3 and 4 presents specific areas such as near to the main roads. We call a group of
the assumptions and system model respectively. Results are houses as a cluster and there are several clusters in the area.
discussed in section 5 and finally the conclusions and future MARs are deployed in the centre of those clusters to provide
work is drawn in section 6. strongest radio signal to most of houses. In total, only 9 MARs
(including gateway) are required to provide full coverage with
II. RELATED WORKS an average of 25 houses covered by each MAR. All MARs are
assumed to be deployed outdoor and the devices are receiving
Rural WMN deployments both large and small, especially the MAR signal indoor. Also shown in Fig. 1 there are three
those based on WiFi radio have been widely reported main branches. Each branch forms a simple chain or tree
worldwide. [2] offers an insightful study on deploying a rural topology with 1, 2 and 5 MARs respectively.
long distance mesh network in the Scottish Highlands and
Islands using IEEE802.11a backhaul and 11b/g access in the backhaul link distance access coverage
5GHz and2.4GHz bands respectively. In this work, two key cell with multiple
aspects are studied namely effect of water level on long MCS
distance links and the practicality of using energy sources
beyond solar. [3] deployed a WMN in the village of Wray,
England that aimed at providing Internet service to the MAR
community. The study shares their experiences which cover a (a) 1 MAR
wide range of issues from both technical and social aspects. A
similar initiative was undertaken in Macha village, Zambia that GW

focuses entirely on proof of concept and investigation into a


wide range of issues covering social, geographical, economic,
business models and sustainability on top of technical (b) 2 MARs (c) 3 MARs
challenges [4]. From their deployments in Digital Gangetic
Plains and Ashwini, [6] presents their experiences and insights
on network planning, MAC protocol enhancements, network
management, power savings, applications and services. A
deployment of a similar nature as to the above was reported in
[5] but limited to experience gained from commercial single
radio 11b/g nodes. To the best of our knowledge no existing (d) 4 MARs
work provides WMN capacity and coverage analysis using
IEEE802.11n radios with aim to understand the relationships
between various key design parameters listed in the previous
section.
III. ASSUMPTIONS
Possible topology with any additional 1
A. Architecture dimmed MR to make up 5 MRs in total

The architecture consists of many Multiradio Access


Routers (MARs) that are interconnected via backhaul radios to
form a mesh network. Here it assumed that all the traffics flow
in multi-hop fashion between MARs and the Gateway (GW)
node, which has the access to the core network. The basic
architecture is illustrated in Fig. 2 as follows:
5GHz (e) 5 MARs (due to limitation of the space, not all scenarios are shown)
Figure 3. Possible topologies with different number of MAR per GW
GW
The dotted lines indicate the possible alternative links that
MAR
transform a multihop into mesh. From capacity viewpoint, a
2.4GHz 2.4GHz 2.4GHz mesh network is assumed to behave like a multihop tree when
With access Without access the whole network is operating at maximum load. In other
words the alternative paths merely provide resiliency and load
Figure 2. Multiradio Multihop Architecture

78
balancing but not additional capacity [13]. Also since we users in lower data rate zone so that equal capacity can be
assume that all MARs provide access, the terms MAR and distributed to all users to achieve fair capacity sharing. Hence,
access cell shall be used interchangeably in the following the effective capacity per MAR is simply the product of mean
sections. capacity per user and total number of users covered by the
MAR:
C. User Distribution and Traffic Assumption
Even though users are distributed sporadically across the (3)
whole deployment area, users are assumed to be uniformly Assuming MARs can simultaneously transmit (or receive)
distributed within the radio cell. The coverage size per cell is traffic to each of its neighbouring MARs as well as to its own
decided by two factors. Firstly, we need to make sure that the users, the maximum mean capacity per MAP is then limited by
capacity of the MR is sufficient to support all users for a chosen the first hop link capacity, :
coverage size. Otherwise, a smaller coverage size is needed.
Secondly, the coverage is determined by the transmission range max (4)
of the desired MCS that meets the target data rate to be
supported at the cell edge. Target data rate per user represents Where is the number of MAR per branch.
the maximum data rate (uplink and downlink) that can be
enjoyed by each end user. This is representative of the In this study, we assumed that the rural houses were mainly
connection speed (or headline speed) typically advertised by a made of thick wood which incurred a wall penetration loss of
network operator. Since WiFi radio is the focus of this paper, 1.609dB according to [14]. As for the propagation model, the
we therefore only consider a generalized capacity. free space path loss model is adopted and given in equation 4:
The overbooking factor is the ratio of the potential
PLpropagation = 10α (log10 (d ) + 3) + 20 log10 ( f ) − 147 (5)
maximum demand to the actual bandwidth consumed.
Typically the overbooking factor ranges from 50:1 to 10:1 [13] Where α is the propagation coefficient, d is the distance in
where factor of 50 or 40:1 is usually associated with web kilometres and f is the carrier frequency in hertz.
browsing application. In other words, the lower the factor, the
higher the average bandwidth or QoS demand for that service. The IEEE802.11n standard is used to represent WiFi and its
receiver sensitivity is obtained from [15] and the values per
D. Interference MCS are tabulated in TABLE I. The general link budget is
Based on the observations found in [2], [7] and [8], given in eqn 5.
interferences resulting from external WiFi or non WiFi
PLmax = EIRP − Rsen + Grx − FM − Lwall (6)
transmitters are minimal and even non-existence in some cases
in rural environment. This leaves the major contribution of Where PLmax is link budget in dB, EIRP is the effective
interference by own networks in form of co channel. Such isotropic radiated power, Grx is the receiver antenna gain, FM is
interference can be mitigated via careful channel planning and fade margin in dB, Rsen is the receiver sensitivity and Lwall is the
directional antenna orientation, which is can be a pretty wall penetration loss in dB. For backhaul link, wall loss is set to
straightforward task due to wide availability of non-overlapping zero and 28dB fade margin is considered instead to ensure link
channels in the 5GHz backhaul. availability of 99.9% as according to [16] based on rayleigh
fading model.
IV. SYSTEM MODEL
TABLE I. MCS AND RECEIVER SENSITIVITY
A. Capacity Per User and Per MAR 802.11n Receiver
MCS Sensitivity (dBm)[15]
Calculation of average capacity per user within a MAR is
based on equation 1 [11]: Type Index(i) 20MHz 40MHz
BPSK 1/2 8 -95.0 -91.0
1 (1) QPSK 1/2 7 -93.0 -90.0
C user =
N user ,i QPSK 3/4 6 -90.0 -87.0
∑i =1 R
N MCS
16-QAM 1/2 5 -87.0 -84.0
eff , i 16-QAM 3/4 4 -84.0 -82.0
Where Nuser,i is number of users in specific Modulation and 64-QAM 2/3 3 -80.0 -78.0
64-QAM 3/4 2 -79.0 -76.0
Coding Scheme (MCS) zone i, , is the effective data rate 64-QAM 5/6 1 -77.0 -74.0
of the MCS zone and is the total number of MCS
considered. , is defined by:
B. Optimization
First Hop BH Optimisation Algorithm
, , (2) Begin
Where is the link efficiency and Define MCS supported by MAR where i 1,2 … ,
, is the raw physical
Where i=1 is highest order MCS type considered and J is the
data rate given by MCS type i.
lowest order MCS considered (typically at the edge of cell)
For the calculation of mean capacity per user, it is assumed
Repeat
that the time of packet transmission is inversely proportional to
Step 1: Find the coverage are of MCS type i, Ai=1 , Ai=2 , .., A i = J
the access link speed. Therefore, for the packets of the same
size, users in high data rate zone will occupy less air-time than (refer TABLE I)

79
, 1 N MAR ← N MAR + 1
, until total number of MARs is fulfilled
where is coverage radius of MCS type i End
Step 2: Find the total number of users supported in coverage
area
V. RESULTS AND ANALYSIS
N user = ∑i =1 Ai × ρ
J
A. General Parameters
where is the user density Table II summarizes the parameters used used in the study.
Step 3: Find effective capacity per MAR using equation 2 TABLE II. RADIO AND ENVIRONMENTAL PARAMETERS
11n 11n
Step 4: Find aggregated capacity requirement for all MARs at Parameters Units 40MHz 40MHz
first hop: (Backhaul) (Access)
Ceff ,hop1 = Ceff ,MAR × N MAR Receive antenna gain dB 15 3
RF frequency GHz 5.8 2.4
Where: N MAR is the number of mesh nodes per branch, NMAR = Channel bandwidth MHz 40 20
1-5 nodes in this study Transmit EIRP power dBm 30 27
Step 5: Map total MARs capacity requirement to first hop BH Link layer efficiency( ) 0.5 0.5
link and select BH with maximum distance that support MARs’ Propagation Coefficient 2.0 (LOS) 2.6 (rural)
capacity requirement. Maximum BH distance is chosen by Margins/Loses
selecting BH where: Interference dB 0 3
Min{C BH ≥ C eff ,hop1 } Fading dB 28 8
Wall Penetration Loss dB - 1.609
and User Requirements
BH Distance = 0, if CBH < Ceff ,hop1 ; ∀ MCS Types Overbooking Factor - 10:1
Data Rate Mbps - 1, 2, 5, 10
Until meeting lowest MCS, J User Density user/SqKm - 100
End
The EIRP limits used are based on the guideline given by
Greedy BH Distance Optimisation Algorithm (maximizing Malaysian Communications and Multimedia Commission. In
multiple hop distance) this study the effective capacity represents the IP layer
Begin throughput where in most cases is approximately 50% of the
N MAR ← 1 , initial number of MAR or coverage cell. raw PHY data rate. The overbooking factor is set at 10:1 to
d total ← 0 , initial total backhaul distance represent higher bandwidth applications such as video
streaming or FTP type applications. User density is derived
Atotal ← 1 , initial total coverage size
from the selected scenario in Fig. 1.
d max ← max d hop(1
N MAR= 5
)
; //derive max 1st hop BH distance
B. Effective Capacity vs. Distance
A perMAR
← A(d max ) ; //derive corresponding coverage area size
N MAR =5
Fig. 4 shows the ideal effective capacity of a MR’s backhaul
Atotal ← ANperMAR
MAR =5
*5 and access link versus distance in km when 4 spatial streams
repeat are used.
d N MAR ← d ( ANperMAR
MAR
) ; //find corresponding BH distance
350
//correspond to max coverage
d total ← dtotal + d NMAR 300
Effective Capacity (Mbps)

N MAR ← N MAR + 1 250

until total number of MAR is fulfilled 200


End
150

Greedy Coverage Optimisation Algorithm 100


Begin
50
N MAR ← 1 , initial number of MAR
d total ← 0 , initial total backhaul distance
0
0 0.2 0.4 0.6 0.8 1 1.2

Atotal ← 1 , initial total coverage size Distance (km )

perMAR
Amax (
← max ANperMAR
MAR =5
)
; //derive max coverage per MAR, 5 Backhaul, 4 Stream Access, 4 Stream

MARs Figure 4. Effective capacity of IEEE802.11n vs. distance


perMAR
Atotal ← Amax *5
repeat As shown, due to the usage of high gain antenna and
perMAR ; //find corresponding BH distance maximum EIRP limit, the backhaul link can reach almost 1km
d N ← d ( Amax )
MAR
and up to 400m for access coverage radius using the most
//correspond to max coverage robust (lowest order) MCS. As for the access, this is only true if
d total ← dtotal + d NMAR

80
there are four antennas at the user end which is unlikely to be Fig. 7 shows that the date rates of 1, 2, 5 and 10Mbps can
the case in practise. still be supported by branch 3 but with much reduced coverage
size.
C. First HopBackhaul Distance vs. Coverage vs. User Data
Rate with Different Number of MAR per Branch D. Backhaul Distance/Coverage Maximization
Fig. 5 shows that with 1-MAR branch case (branch 1), the In this section we adopted the backhaul distance (or
first hop backhaul link can support up to 0.43sqkm of coverage coverage size) maximization algorithm on the 5-MAR branch
cell size with the maximum distance between MARs equal to (branch 3) and the result is presented in Fig. 8 and 9.
0.92 km when user data rate is 1 Mbps. When user data rate is
increased to 2Mbps, the maximum coverage size 0.43sqkm can
no longer be supported at the same backhaul distance. A
slightly smaller coverage size per MAR (0.36sqkm) has to be
tolerated. When the user data rate is further increased to 5 and
10Mbps, several coverage sizes can no longer be supported. For
the 10Mbps case, backhaul distance is reduced to 0.8 km to
support coverage size of 0.21 sqkm. Smaller coverage implies
no service for some users.

Figure 7. Backhaul distance vs. coverage size vs. data rate for 5-MAR branch
using 1 spatial stream at backhaul link (refer branch 3 in Fig.1)

Fig. 8 shows that the total (aggregated) coverage size


achievable against various data rate using the greedy algorithm.
It can be observed that the 5-MAR branch network can support
more than 2sqkm of aggregated coverage size with 1Mbps per
user. Fig. 9 shows that the maximum multihop backhaul
distance (aggregated) remains constant at ~4km for all data
rates and number of spatial streams considered. In this case, the
Figure 5. Backhaul distance vs. coverage size vs. user data rate for 1-MAR total coverage size is being traded off with backhaul distance. It
branch using 1 spatial stream at backhaul link (refer branch 1 in Fig.1) is also clearly shown that higher number of spatial streams
Fig. 6 shows for the 2-MAR branch case (branch 2), similar enable bigger coverage sizes.
trend as in 1-MAR case is observed but with noticeable
reduction in coverage size and backhaul link distance per user 3
Total Coverage for 5 MARs

date rate. From the result, we can deduce that 10Mbps can be
supported by branch 2 provided the cell size is not more than
0.123sqkm. 2
(sqkm)

0
1 2 5 10
Data Rate (Mbps)
1 Stream 2 Stream 3 Stream 4 Stream
Figure 8. Total coverage size of 5-MAR branch with maximised BH distance
Fig. 10 shows coverage size and backhaul distance with
coverage size being optimized by the greedy algorithm.
Maximum achievable aggregated coverage size for 5-MAR
branch is 2.2 square km. For high capacity 4 spatial-stream
backhaul, the coverage size can be maintained by increasing the
Figure 6. Backhaul distance vs. coverage size vs. user data rate for 2-MAR user data rate with no more than one fifth of loss in coverage
branch using 1 spatial stream at backhaul link. (refer branch 2 in Fig.1) when providing 10Mbps data rate per user. Capacity limited 1
spatial-stream backhaul performs the worst in maintaining
coverage size when user data rate increases. The aggregated

81
coverage size of 10Mbps is around 70% smaller than coverage VI. CONCLUSIONS
size of 1Mbps. As shown in Fig. 11 the total backhaul distance This paper presents a capacity and coverage analysis of a
drops gradually when user data rate increases. It is reduced by multiradio multihop network deployment using IEEE802.11n
almost 30% with 10Mbps as compared to 1Mbps user data rate. radios for a typical rural area. Insights on the relationships
between various key design parameters particularly backhaul
link rate, backhaul link distance/multihop distance, coverage
Multihop Backhaul Distance (km)

4.5
4
size per MAR/total coverage size, number of MIMO spatial
3.5 multiplex (SM) stream, number of MAR per branch and
3 effective capacity per access point or per user have been
2.5 acquired. In general, user data rate ranging from 1 to 10Mbps
2 can be supported with a 5-hop multiradio network using
1.5 IEEE802.11n radios with some tradeoffs in terms of maximum
1 coverage size. We have also introduced two approaches that
0.5 maximize the coverage size or backhaul distance respectively.
0 As expected, the backhaul distance maximization algorithm is
1 2 5 10
Data Rate (Mbps) more suited for rural case and coverage maximization on the
1 Stream 2 Stream 3 Stream 4 Stream other hand is more appropriate for urban or suburban
Figure 9. Total backhaul distance for 5-MAR branch with maximised BH deployments.
distance
REFERENCES
[1] “KAMPUNG ’WIFI’ TO ACCELERATE EFFORTS FOR THE
3
NATIONAL BROADBAND INITIATIVE (NBI)”, Press Release,
Total Coverage for 5 MARs

Mihttp://www.skmm.gov.my/link_file/newsdesk/press/2010/PR_KgWiFi
_Julau_Sarawak_170910.pdf
2
[2] G. Bernardi, P. Buneman, and M. K. Marina,“Tegola Tiered Mesh
(sq km)

NetworkTestbed in Rural Scotland”,in WiNS-DR’08, San Francisco, CA,


Sept. 2008.
1
[3] Ishmael, J., Bury, S., Pezaros, D., Race, N.: Rural community wireless
mesh networks. In IEEE Internet Computing, Vol.12, Issue 4, 2008, pp.
22–29.
0 [4] K. W. Matthee et. al., “Bringing Internet Connectivity to Rural Zambia
1 2 5 10
Data Rate (Mbps) Using a Colloborative Approach”. In Proc. 2nd IEEE/ACM ICTD,
1 Stream 2 Stream 3 Stream 4 Stream Bangalore, India, Dec. 2007.
Figure 10. Total coverage size of 5-MAR branch with maximised access [5] D.L. Johnson, “Evaluation of a single radio mesh network in South
coverage size Africa”, in ICTD’07, Bangalore, India, Dec. 2007.
[6] B. Raman and K. Chebrolu, “Experiences in using WiFi for Rural
Internet in India”, in IEEE Communications Magazine, Vol: 45(1), Jan
5
2007.
Multihop Backhaul Distance

4 [7] A. Sheth et al, “Packet Loss Characterization in WiFi-based Long


Distance Networks”, In Proc. IEEE INFOCOM’07, Alaska, May 2007,
3 [8] D. Gokhale, S. Sen, K. Chebrolu, and B. Raman, “On the Feasibility of
the Link Abstraction in (Rural) Mesh Networks”, In INFOCOM’08,
(km)

2 Phoenix, AZ, USA, Apr. 2008.


[9] Utpal Kumar Paul, Riccardo Crepaldi, Jeongkeun Lee, Sung-Ju Lee,
1 Raul Etkin, “Characterizing WiFi Link Performance in Open Outdoor
Networks”, To appear in IEEE SECON’11, Utah USA, June 2011.
0
1 2 5 10 [10] Matthias Lott, I. Forkel, “A Multi-Wall-and-Floor Model for Indoor
Data Rate (Mbps) Radio Propagation”, in IEEE VTC’01 Spring, Page 464-468 vol.1, May
1 Stream 2 Stream 3 Stream 4 Stream 2001, Phodes, Greece.
Figure 11. Total backhaul distance for 5-MAR branch with maximised access [11] Kaveh Pahlavan and Allen H. Levesque, Wireless Information
coverage size Networks2nd Edition, 2005 Wiley Interscience.
[12] A. Ting and D. Chieng, “Design and Capacity Performance Analysis of
Since the backhaul distance optimization algorithm tends to Wireless Mesh Network”, in Mobility’08, Ilan, Taiwan, Sept. 2008.
stretch for longer distance by trading off coverage size, less [13] D. Chieng, D. V. Hugo, A. Banchs, “A Cost Sensitivity Analysis for
number of users can therefore be supported. This algorithm is Carrier Grade Wireless Mesh Networks with Tabu Optimization”,
more suitable for rural deployment where the number of user is Workshop of Carrier Grade Wireless Mesh Network, in Proceeding of
low and distributed. The additional distance provided on IEEE INFOCOM’10, San Diego, California, March 2010.
backhaul enables a great degree of freedom in terms of MAR [14] Mohammed, Y.E, Abdallah, A.S. and Liu, Y.A, “Characterization of
placing. On the other hand, the coverage maximization Indoor Penetration Loss at ISM Band”, in CEEM’03, 4-7 Nov. 2003
algorithm which ultimately converges towards blanket–type [15] Datasheet: Unex DNMA-92: 802.11n a/b/g wifi 2x2 mini-PCI module,
MB92/AR9220.
coverage is believed to be more suitable for urban or suburban
[16] “Wireless Link Budget Analysis, How to Calculate Link Budget for Your
deployments. Wireless Network”, Tranzeo Wireless Technologies Inc,
http://www.tranzeo.com/allowed/Tranzeo_Link_Budget_Whitepaper.pdf

82

Вам также может понравиться