IEICE TRANS. COMMUN., VOL.E96B, NO.

1 JANUARY 2013

127

PAPER

Adaptive Limited Dynamic Bandwidth Allocation Scheme to

Improve Bandwidth Sharing Eciency in Hybrid PON
Combining FTTH and Wireless Sensor Networks
Monir HOSSEN , Nonmember and Masanori HANAWAa) , Member

SUMMARY
This paper proposes a dynamic bandwidth allocation algorithm that improves the network performance and bandwidth sharing
eciency in the upstream channels of a hybrid passive optical network
(PON) that combines a fiber-to-the-home (FTTH) access network and wireless sensor networks (WSNs). The algorithm is called the adaptive limited dynamic bandwidth allocation (ALDBA) algorithm. Unlike existing algorithms, the ALDBA algorithm is not limited to controlling just
FTTH access networks, it also supports WSNs. For the proposed algorithm, we investigate the dierence in the lengths of generated data packets between the FTTH terminals and sensor nodes of WSN to eectively
evaluate the end-to-end average packet delay, bandwidth utilization, time
jitter, and upstream eciency. Two variants of the proposed algorithm
and a limited service (LS) scheme, which is an existing well-known algorithm, are compared under non-uniform trac conditions without taking into consideration priority scheduling. We demonstrate the proposed
scheme through simulation by generating a realistic network trac model,
called self-similar network trac. We conducted a detailed simulation using several performance parameters to validate the eectiveness of the proposed scheme. The results of the simulation showed that both ALDBA
variants outperformed the existing LS scheme in terms of average packet
delay, bandwidth utilization, jitter, and upstream eciency for both low
and high trac loads.
key words: WSN, DBA algorithm, bandwidth utilization, jitter, upstream
eciency, delay

1.

Introduction

The Ubiquitous-City (u-City) is an autonomous city where

all the information systems and service providers are virtually linked together. The convenience of a u-City is that
anyone can access any service provider from anywhere in
the city by using only one access terminal. Nowadays, the
number of dierent service providers (i.e., FTTH, WSNs,
local area networks, and ad-hoc networks) in a u-City has
increased dramatically. Although a significant number of
service providers providing both wireless and fiber-based
networks already exist in modern cities, there is still no
suitable interlink between the wireless operators and fiberbased operators. Usually, WSN service providers use wireless links to connect wireless sensor nodes to a central office, i.e., personal area network coordinator (PANC). In contrast, optical fiber-based access networks are used for FTTH
terminals. This optical fiber-based network infrastructure
Manuscript received April 17, 2012.
Manuscript revised August 11, 2012.

The authors are with University of Yamanashi, Kofu-shi, 4000862 Japan.

a) E-mail: hanawa@yamanashi.ac.jp
DOI: 10.1587/transcom.E96.B.127

has enough bandwidth that the bandwidth can be eectively

shared with a WSN without any bandwidth bottlenecks occurring in the access networks. J. Tang et al. proposed using
a cluster-based WSN to divide the entire sensor network of
a u-City and to connect all of the static cluster heads (CHs)
to the PANC through a radio-over-fiber (RoF) link [1]. Our
scheme incorporates the co-operative clustering algorithm
proposed by M. Hossen et al., where each cluster contains
a static CH and is connected with the optical network units
(ONUs) of a PON system through an optical fiber link [2].
As the sensor nodes are deployed in a u-City, any sensor
node in a home or oce can act as a static CH and easily
be incorporated into the electric supply system. That is why
there is no possibility of an inadequate power supply, unlike the CHs of typical sensor networks. The use of optical
fiber links in a cluster-based WSN in a u-City is very useful because it will avoid long distance radio frequency (RF)
transmissions [2] from CHs to the PANC, which reduces the
energy consumption of small battery powered sensor nodes
[3].
Sharing a PON between two dierent service providers
(i.e., CHs of a WSN and FTTH terminals) as a common
optical link can provide a cost-eective and flexible infrastructure that will provide the required bandwidth demands
of future hybrid networks. A PON is an inherently costeective [4] technology because it does not require active
components and remote power facilities. Furthermore, sharing the fiber medium for both the WSNs and FTTH access
networks reduces the cost of physically deploying the fiber.
In this paper, we consider a tree-based hybrid PON architecture that comprises an optical line terminal (OLT), a
splitter, and ONUs, where the ONUs are the CHs of a WSN
and FTTH terminals. The PON provides bi-directional
transmissions between the OLT and ONUs through a single
optical fiber link. Multi-point control protocol (MPCP) [5],
[6] plays a vital role in the avoidance of collisions and the
sharing of the single optical fiber link with multiple ONUs in
tree-based PON topology. In the downstream transmission,
the Gate message and data packets are broadcasted from the
OLT to all the ONUs. Each ONU accepts data from the OLT
according to the destination address. The main information
contained in a downstream packet is the upstream transmission time and the length of the transmission window of each
ONU. In an upstream transmission, a PON is a multipoint
to point network [7] where all the ONUs share a common

c 2013 The Institute of Electronics, Information and Communication Engineers

Copyright 

IEICE TRANS. COMMUN., VOL.E96B, NO.1 JANUARY 2013

128

channel to transmit data to the OLT. According to the time

division multiple access (TDMA) principle, only a single
ONU can transmit data in a specified time-slot to avoid data
collision and packet loss. The transmission window of each
ONU also contains a Report message at the end of a timeslot to request the desired window size of the next time cycle in accordance with the ONUs buer occupancy. That is
why a robust bandwidth sharing algorithm is required to allocate the time-slot among the ONUs in the upstream direction. Dierent dynamic bandwidth allocation (DBA) algorithms have been proposed by several researchers. Among
them, IPACT [8] is the most popular approach. In this approach, an OLT-based polling scheme is used to poll the next
ONU before the data packet from the previous ONU has arrived. With these algorithms, all the trac is assumed to be
single class, i.e., service dierentiation is not considered. G.
Kramer et al. studied the priority scheduling principle and
combined it with the DBA algorithm to provide an eective
service for delay and jitter sensitive applications [9]. More
recently, C. Assi et al. proposed a new DBA algorithm with
quality of service (QoS) support that is combined with priority scheduling and queue management [10]. In this scheme,
the ONUs are divided into two groups, i.e., lightly loaded
ONUs and heavily loaded ONUs, and the bandwidth savings from the lightly loaded ONUs are distributed among the
heavily loaded ONUs. However, there is no suitable DBA
algorithm for a hybrid network combining FTTH access networks and WSNs that can improve the fairness of bandwidth
sharing between the two dierent service providers to reduce end-to-end packet delay and improve QoS.
In this paper, we propose using an adaptive limited
DBA (ALDBA) algorithm for a hybrid network combining FTTH access networks and WSNs in a PON-based access network. We found that the data packet lengths for
FTTH terminals and the sensor nodes of WSN were not
similar; therefore, allocating the same maximum transmission window to both service providers is not an ecient approach in terms of reducing end-to-end packet delay and
improving QoS. The proposed scheme maintains two different maximum transmission windows on the basis of the
generated packet length of sensor nodes and the FTTH terminals. The simulation results showed that using the proposed DBA scheme produced less end-to-end packet delay
and jitter with higher bandwidth utilization and upstream efficiency than the existing LS scheme [8] under both low and
high trac loads.
The remainder of this paper is organized as follows.
Section 2 presents the hybrid network structure, ALDBA algorithms, and upstream frame format. Section 3 describes
the performance evaluation conducted by simulation. In
Sect. 4, we explain the simulation results. Finally, conclusions are presented in Sect. 5.
2.

Hybrid Network Structure and ALDBA Algorithms

In this section, we first explain the proposed network structure of the hybrid PON combining FTTH access networks

and WSNs and the upstream frame format. Next, we explain

the proposed ALDBA scheme, which enhances the system
performances of the hybrid network and improves the bandwidth sharing eciency among the ONUs of both service
providers. The proposed ALDBA scheme has two variants:
ALDBA1 scheme and ALDBA2 scheme. In the ALDBA1
scheme, available bandwidth is calculated from the diermax
for the
ence of two maximum transmission windows, WSN
max
for the FTTH access networks, and the
WSNs and WFTTH
total available bandwidth is fairly distributed among all the
ONUs of the network, regardless of the type of service. With
the ALDBA2 scheme, the available bandwidth savings in
the ALDBA1 scheme are used for the deferred data packets
and this scheme is also incorporated into the DBA scheme
of C. Assi et al. [10] without taking into consideration any
intra ONU priority scheduling. This implies that all the data
classes are transmitted on a first-in first-out (FIFO) basis.
2.1 Hybrid PON Combining FTTH Access Networks and
WSNs
A tree-topology-based hybrid PON consists of one OLT located on the tree side with both the FTTH access network
and WSN service providers connected to several ONUs on
the leaf side of the network. The OLT is connected to several ONUs through optical fiber links using a 1:N optical
splitter/combiner. Most PON systems consist of one OLT
and N ONUs connected to FTTH terminals with dierent
round-trip-time (RTT) delays. In contrast, the proposed hybrid PON structure consists of ONUs from two dierent operators, i.e., ONUs connected to FTTH terminals and ONUs
connected to CHs of a WSN. The number of ONUs connected to FTTH terminals and the number of ONUs connected to CHs of the WSN may vary; however, for simplicity, only four ONUs for both services with dierent RTTs
are shown in Fig. 1. The RTT depends on the physical distance from the OLT to an ONU and the OLT maintains a
polling table to store the RTT of every ONU. In the MPCP
of the PON system, the RTT plays a vital role by scheduling
the control messages: Gate message from the OLT to ONUs
and Report message from the ONUs to the OLT.
Figure 2 shows the upstream frame format of the proposed hybrid PON. Here, ONU1 is an ONU connected to
max
and
a FTTH terminal with maximum window size, WFTTH
ONU2 is an ONU connected to a CH of WSN with maxmax
. Every data packet of ONU1
imum window size, WSN
contains a user identification (ID) number and the payload
of that user and may be multiplexed with dierent users
if the ONU consists of multiple users. In contrast, every
data packet of ONU2 contains a node ID number, a service
code (SC), and the payload of the sensor node, which can
be multiplexed with dierent sensor nodes of dierent services. Here, the node ID is a unique number for each sensor
node, and SC indicates the type of service (i.e., gas, water, electricity, etc.) and a code for a service provider (for
discrimination if the same service is provided by dierent
service providers) so that each sensor node can be uniquely

HOSSEN and HANAWA: ADAPTIVE LIMITED DYNAMIC BANDWIDTH ALLOCATION

129

Fig. 3

Fig. 1 Network architecture of hybrid PON with 2 ONUs connected to

FTTH terminals and 2 ONUs connected to WSNs.

Fig. 2

Upstream frame format of hybrid PON.

recognized.
2.2 ALDBA1 Scheme
We consider a hybrid PON-based access network with N
ONUs (i.e., N = NSN + NFTTH , where NSN is the number of
ONUs connected to CHs of WSN and NFTTH is the number of ONUs connected to FTTH terminals). In every DBA
algorithm, the allocation of transmission windows to the
ONUs depend on the requested window size W R and maximum window size max . In the LS scheme, the maximum
allocated window size max depends on the cycle time (i.e.,
length of a polling cycle) T cycle , as shown in the following
equation:
T cycle N(T E + T R + TG )
(1)
N
Here, T E is the length of the Ethernet overhead, T R is the
length of the Report message, and TG is the guard time. In
every DBA scheme, T cycle is a variable parameter that varies
from a minimum value to an upper bound, depending on
the number of active ONUs and their trac load. Making
T cycle too small will result in more bandwidth being wasted
and not utilized by Ethernet overheads (EOs), guard intervals (GIs), and Report messages, as shown in Fig. 2. This
max =

Illustrated example of ALDBA1 scheme.

is because even if an active ONU does not have any data, it

still has to transmit a Report message that includes an EO
and a GI. The maximum transmission window of an ONU
always does not depend on the packet length but on the total
stored data packets after latest upstream transmission. However, typical transmission windows of the ONUs of WSN
are less than that of FTTH access network because of lower
data rate and smaller packet length of WSN. That is why enclosing these two dierent operators data in a single PON
will make T cycle smaller and the network will suer from
bandwidth utilization problems. To overcome this problem
and to improve the bandwidth sharing eciency and QoS of
the proposed hybrid PON, two dierent max are used, i.e.,
max
max
< max for the WSN and WFTTH
= max for the FTTH
WSN
access network, depending on the maximum length of generated packets by the ONUs of FTTH access network and
max
< max , we calculate the total available
WSN. Since WSN
bandwidth savings WT S using the equation below:


max
WT S = NSN max WSN
(2)
Finally, WT S is divided by the total number of ONUs N to
calculate the average available bandwidth savings for each
ONU, i.e., W avg = WT S /N, and this average bandwidth savmax
max
or WSN
if the
ings is added to the maximum window WFTTH
max
max
or WSN
.
requested window W R is larger than WFTTH
The allocated transmission window for the ALDBA1
scheme is calculated using the following formula:
R
avg

if Wi,Rj Wi,max
Wi, j
j +W
ALDBA1
Wi, j
=
(3)
avg
avg

Wi,max
if Wi,Rj > Wi,max
j +W
j +W
Where Wi,ALDBA1
is the window size allocated to ONU i at
j
time cycle j using the ALDBA1 scheme, and Wi,Rj is the requested window size of ONU i at time cycle j, and Wi,max
j is
the maximum window size for the ONU of FTTH terminal
or CH of WSN at time cycle j.
Figure 3 shows an illustrative example of the ALDBA1
scheme. If ONU1 at time cycle j is a lightly loaded ONU,
avg
, then the granted window size is
i.e., W1,R j < W1,max
j + W
W1,R j . In contrast, if the ONU2 at time cycle j is a heavavg
ily loaded ONU, i.e., W2,R j > W2,max
, then the granted
j +W
max
avg
window size is W2, j + W .

IEICE TRANS. COMMUN., VOL.E96B, NO.1 JANUARY 2013

130

2.3 ALDBA2 Scheme

In this scheme, excessive bandwidth from lightly loaded
ONUs, as explained by C. Assi et al. [10], is calculated without taking into consideration intra ONU priority scheduling.
The ALDBA2 scheme is also incorporated with the average
bandwidth savings in the ALDBA1 scheme to provide some
transmission window to the deferred data during the waiting
time between the transmission of Gate and Report messages.
To calculate the excessive bandwidth, the ONUs are divided
into two groups: heavily loaded ONUs and lightly loaded
ONUs. Equation (4) is used to calculate the total excessive
bandwidth of the proposed hybrid PON.
L
FTTH

WTexcess
otal, j =

LSN 

 

max
R
max
R
W
WFTTH,
+
WSN,
m, j
j
j Wn, j

m=1

R
WTexcess
otal, j Wi, j
H
R
k=1 Wk, j

(4)

n=1

(5)

where Wi,excess
is the excessive bandwidth for ONU i at time
j
cycle j and H is the number of heavily loaded ONUs.
Usually, the waiting time in a PON is determined by
the RTT and the delay of Gate starting time TGd of each
ONU. For simplicity, only RTT time is considered in the
analysis. The OLT predicts the amount of deferred data during the waiting time of each ONU and allocates the extra
bandwidth up to W avg in addition to the requested window
W R . Prediction of the deferred data during the waiting time
depends on the current queue occupancy and RTT of each
ONU, as shown in the equation below:
Wi,pred
j =

Wi,Rj RT T
T i,acquisition
j

Illustrative example of ALDBA2 scheme.

average available bandwidth W avg .

where WTexcess
otal, j is the total excessive bandwidth at time cycle
j, LFTTH and LSN are the number of lightly loaded ONUs
connected to FTTH terminals and CHs of WSN, respecR
tively, and Wm/n,
j is the requested window size of lightly
loaded ONU m/n at time cycle j. The following equation is
used to fairly distribute the total excessive bandwidth calculated in Eq. (4) among the heavily loaded ONUs to solve the
congestion problem of the hybrid PON:
=
Wi,excess
j

Fig. 4

(6)

where T i,acquisition
is the acquisition time of present data in the
j
queue, Wi,pred
j is the predicted window size for ONU i at time
avg
.
cycle j, and Wi,pred
j W
The main dierences between the proposed ALDBA2
scheme and the existing scheme in [10] are as follows:

1. Consideration of two dierent maximum transmission

windows W max for two dierent services.
2. More priority is given to heavily loaded ONUs by allocating excessive bandwidth rather than intra ONU priority scheduling among dierent service classes.
3. Prediction of waiting time trac that is served by the

The bandwidth allocation formula for the ALDBA2 scheme

is as follows:
Wi,ALDBA2
j

pred
R
excess

if Wi,Rj Wi,max
Wi, j +Wi, j
j +Wi, j
=
W max +W excess +W pred if W R > W max +W excess

i, j
i, j
i, j
i, j
i, j
i, j

(7)

where Wi,ALDBA2
is the allocated bandwidth for ONU i at time
j
cycle j using the ALDBA2 scheme.
The illustrative example of bandwidth allocation in the
ALDBA2 scheme is shown in Fig. 4. Here, requested window W1,R j with predicted window W1,pred
j is allocated to the
lightly loaded ONU1 at time cycle j. In contrast, maximum
max
excess
transmission window WN,
j and excessive bandwidth WN, j
pred
with predicted window WN,
j is allocated to the heavily
loaded ONU N at time cycle j.

3.

Performance Evaluation by Simulation

In this section, the performances of the proposed ALDBA

schemes for the hybrid PON are evaluated in terms of average end-to-end packet delay, bandwidth utilization, jitter, and upstream eciency. The evaluation was performed
by laboratory made computer simulation programs. We assumed that the architecture of the hybrid PON had one OLT
and 16 ONUs. Furthermore, we assumed that the transmission speed was 1 Gbps for both upstream and downstream
channels. The distances from the OLT to the ONUs was assumed to be random and in the range 1020 km. A highly
bursty self-similar network trac model [11] was used to
generate the data packets for both the FTTH terminals and
sensor nodes of WSN. The burst trac from 0 to multiple packets was generated by this self-similar trac model,
where the length of each packet was not larger than the
maximum packet length Bmax . The service policy was assumed to be on a first-in first-out (FIFO) basis with an infinite buer size for each ONU. The maximum packet lengths
for the FTTH terminals and WSNs were 1500 bytes [12] and
1024 bytes [13], respectively. The ratio of the two maximum
max
max
and WSN
, reflected the ratio
transmission windows, WFTTH
of these two maximum packet lengths. The computation

HOSSEN and HANAWA: ADAPTIVE LIMITED DYNAMIC BANDWIDTH ALLOCATION

131
Table 1

Simulation parameters.

time was assumed to be 10 s for the proposed DBA algorithms, as used by I. Wang et al. [14]. All the analyses were
done for non-uniform oered loads in the range 01.4 with
a variable cycle time in the range 0.53 ms. Our simulation
took into consideration the queuing delay, transmission delay, congestion delay, and processing delay without taking
into consideration any priority scheduling. The simulation
parameters are summarized in Table 1.
The end-to-end packet delay is one of the most important parameters for every network. The proposed hybrid
PON consists of sensor networks and the data of some sensor nodes (e.g., hospital and fire alarm sensor systems) are
delay sensitive. One of the main objectives of the proposed
ALDBA algorithms is the reduction of end-to-end packet
delay by allocating larger transmission window to heavily
loaded ONUs.
Bandwidth utilization of a PON system is expressed by
the following formula:
BU =

total
Wgrant
total
Wgrant
+ N BC

(8)

total
where Wgrant
is the total granted window for all active
ONUs. Since the term NBC in the denominator is constant,
bandwidth utilization can be improved if we can provide a
total
by taking into considlonger total granted window Wgrant
eration the ONUs trac. Note that the proposed scheme
can achieve better bandwidth sharing eciency and can reduce the unused bandwidth by using two dierent maximum
transmission windows for both types of ONU. This granting
of more transmission windows to the heavily loaded ONUs
is the reason why the proposed ALDBA schemes can improve the bandwidth utilization of a hybrid PON.
The nature of usual network trac models is highly
bursty. Due to this bursty network trac, the data packets
of a particular ONU suer from variations in arrival times
in dierent time cycles. To measure the variation in the data
packet arrival time of an ONU, jitter performance of the proposed ALDBA schemes were analyzed. The upstream eciency of a PON system is defined by Eq. (9):

total
is the total generated window of all active
here, Wgenerated
total
ONUs. The calculation of Wgrant
is similar to that of the
total
in the debandwidth utilization, BWU. However, Wgenerated
nominator of upstream eciency consists of a summation of
the total generated windows of all the ONUs in a time cycle even when the generated window is greater than W max .
Because of that upstream eciency can be lower than the
bandwidth utilization.
The following equations are used to calculate the endto-end packet delay DA1 , bandwidth utilization BU A1 , jitter
JA1 , and upstream eciency UE A1 of the ALDBA1 scheme:
N

(WiR +BC )

i=1

if WiR Wimax + W avg

RU
(10)
DA1 =
N

(W max +W avg +BC )

R
max
avg
i=1 i
+ T cng if Wi > Wi + W
RU
N R

Wi

i=1

if WiR Wimax + W avg

R +B

W
i=1 ( i C )

(11)
BU A1 =
N

(Wimax +W avg )

i=1
R
max
avg

if Wi > Wi + W

N (Wimax +W avg +BC )

i=1

 n
2
1   A1
A1
JA1 =
(12)
T avl, j T avl,
j1
n j=1
N

(WiR +BC )

i=1

if WiR Wimax +W avg

A1
RU
(13)
T avl,
N
j =

(W max +W avg+BC )

R
max
avg
i=1 i
+T cng if Wi > Wi +W
RU
N R

Wi

i=1

if WiR Wimax + W avg

R +B

W
(
)

C
i=1 i
(14)
UE A1 =
N

Wimax +W avg )

i=1
R
max
avg

N (W R +B ) if Wi > Wi + W

C
i
i=1

Similarly, the following equations are used for the ALDBA2

scheme to calculate the end-to-end packet delay DA2 , bandwidth utilization BU A2 , jitter JA2 , and upstream eciency
UE A2 :
DA2 =
N

(WiR +Wipred +BC )

i=1

if WiR Wimax +Wiexcess

RU

pred

(W max +Wiexcess +Wi +BC )

i=1 i
+ T cng if WiR > Wimax +Wiexcess
RU

BU A2

N  R pred 

Wi +Wi

i=1



N

pred
R

i=1 Wi +Wi +BC

=

N 

Wimax +Wiexcess +Wipred

i=1

N  max excess pred

i=1

UE =

total
Wgrant
total
Wgenerated

+ N BC

(9)

Wi

+Wi

+Wi

+BC

(15)
if WiR Wimax +Wiexcess

if Ri > Wimax +Wiexcess

(16)

IEICE TRANS. COMMUN., VOL.E96B, NO.1 JANUARY 2013

132
A2
T avl,
j =
N 


WiR +Wipred +BC

i=1

if WiR Wimax +Wiexcess

RU



W max +Wiexcess +Wipred +BC

i=1 i
+ T cng if WiR > Wimax +Wiexcess
RU

JA2 =

UE A2

(17)
n 


2
1
A2
A2
T avl,
T avl,
j
j1
n j=1
N  R pred 

Wi +Wi

i=1

if WiR Wimax +Wiexcess


N 

i=1 WiR +Wipred +BC

=

N 

Wimax +Wiexcess +Wipred

i=1

if WiR > Wimax +Wiexcess

N  R pred 
i=1

4.

Wi +Wi

(18)

(19)

+BC

Simulation Results

In this section, the simulation results are used to compare

the system performances of the proposed ALDBA1 and
ALDBA2 schemes with the performance of the existing LS
scheme.
Figures 5(a), 5(b), and 5(c) show contour plots of the
average end-to-end packet delay for dierent oered loads
and cycle times of the existing LS scheme, and the proposed ALDBA1 and ALDBA2 schemes for NFTTH :NSN =
8:8. From the analysis and comparison of the these three
contour plots, it is clear that the ALDBA1 and ALDBA2
schemes provide less delay than the LS scheme for any value
of a cycle time and oered load. Comparison of average
packet delay among the three schemes is shown in Fig. 5(d)
for 2-ms cycle time. From this result, it is found that the
ALDBA2 scheme provides about 50% and 25% less delay
than the LS and ALDBA1 schemes, respectively, at an offered load of 1.4. The ALDBA2 scheme provides the best

Fig. 5

Average packet delay in ms for NFTTH :NSN = 8:8.

delay performance in hybrid PON. The proposed schemes

are also more suitable for delay sensitive services, e.g., VoIP
service. Since VoIP uses real-time transport protocol, it requires low packet delay.
Figure 6(a) and 6(b) compare the average packet delay with the oered load among the LS, ALDBA1, and
ALDBA2 schemes for 2-ms cycle time by changing the
number of ONUs connected to FTTH terminals and CHs
of WSN. From the comparison of both cases, we can say
that the ALDBA schemes outperform the LS scheme, the
ALDBA2 scheme being the best. The end-to-end packet delays for the proposed ALDBA schemes are far smaller than
for the LS scheme when the number of ONUs connected to
CHs of WSN is larger than those connected to the FTTH
terminals, as shown in Fig. 6(a), where NFTTH :NSN = 4:12.
The main reason is higher bandwidth savings from the larger
number of ONUs of WSNs.
The contour plots of Fig. 7 show the bandwidth utilization of the existing LS and the proposed ALDBA schemes.
From the analysis, it is found that the ALDBA2 scheme provides a wider area of maximum utilization than the LS and
ALDBA1 schemes. Even though Fig. 7(d) shows similar
results for a 2-ms cycle time, the ALDBA2 scheme could
achieve the same results from 3 to 1.25-ms cycle times with
an oered load of 0.75 while the ALDBA1 and LS schemes
could achieve the same result from 3 to 2.0-ms and 2.5-ms
cycle times, respectively, with an oered load of 0.75, as

Fig. 6

Comparison of average packet delay for 2-ms cycle time.

Fig. 7

Bandwidth utilization for NFTTH :NSN = 8:8.

HOSSEN and HANAWA: ADAPTIVE LIMITED DYNAMIC BANDWIDTH ALLOCATION

133

Fig. 8

Jitter in ms for NFTTH :NSN = 8:8.

sults. However, the highest upstream eciency area of the

ALDBA2 scheme is far larger than both the ALDBA1 and
LS schemes in both cases of maximal cycle time and oered
load. The comparison for a 2-ms cycle time in Fig. 9(d)
shows that the upstream eciency drops drastically for an
oered load larger than 0.5 for the LS scheme and 0.6 for
the ALDBA1 scheme. On the other hand, the upstream efficiency of the ALDBA2 scheme increases until the oered
load is1.0.
The first reason for the performance enhancement in
the proposed schemes is the use of two dierent W max for
the two dierent service providers, i.e., the FTTH access
networks and WSNs, to improve the bandwidth sharing efficiency. The second reason is the calculation of excessive bandwidth and average bandwidth savings from lightly
loaded ONUs. Furthermore, the final reason is the utilization of the excessive bandwidth by heavily loaded ONUs
and the average bandwidth savings for the deferred data.
5.

Fig. 9

Upstream eciency for NFTTH :NSN = 8:8.

shown using the white arrows in the contour plots; therefore, the ALDBA2 scheme is more suitable even though the
cycle time is low.
Jitter performances of the existing LS scheme and the
proposed ALDBA schemes are shown in Figs. 8(a), 8(b),
and 8(c). As shown using the white arrows in the contour
plots, the ALDBA2 scheme provides the lowest jitter until the oered load of 1.35 at a 3-ms cycle time while the
ALDBA1 and LS schemes provide the lowest jitter until the
oered load is 1.0 and 0.85, respectively. From the comparison for the 2-ms cycle time in Fig. 8(d), it is found that
the ALDBA2 scheme performs better than both the LS and
ALDBA1 schemes.
Comparison of the upstream eciencies for the same
ranges of cycle time and oered load is shown by the contour plots in Figs. 9(a), 9(b), and 9(c). In this case, the
proposed ALDBA schemes are also better than the existing
LS scheme; this result is consistent with all the above re-

Conclusions

In this paper, we have proposed two new DBA algorithms,

ALDBA1 and ALDBA2, and applied them to a hybrid PON
consisting of FTTH access networks and WSNs. We investigated the dierences in packet lengths between the FTTH
terminals and sensor nodes of WSN. Since the packet length
varies, the proposed hybrid PON has poor bandwidth sharing eciency when used with existing DBAs. The new algorithms outperformed the existing DBA scheme that we
investigated in terms of average end-to-end packet delay,
bandwidth utilization, jitter, and upstream eciency. When
compared with the LS scheme, the ALDBA1 scheme provided 25% less delay while the ALDBA2 scheme provided
50% less delay for a 2-ms cycle time at an oered load of
1.4. With regard to bandwidth utilization and jitter, the proposed ALDBA schemes provided performance close to but
slightly better than that of the LS scheme at a 2-ms cycle
time. However, the ALDBA2 scheme provided about 15%
and 25% more upstream eciency than the ALDBA1 and
LS schemes, respectively, for a 2-ms cycle time at an oered
load of 1.0. Moreover, the new ALDBA schemes provided
a wider area of better performance than the LS scheme for
both cycle time and oered load when used with the hybrid
PON in each simulation. The proposed schemes provide
better QoS than the LS scheme for delay sensitive services.
In our future works, the optimal number of ONUs, larger
transmission distance, and higher channel speed with priority scheduling scheme will be considered for obtaining the
best system performance. Moreover, ALDBA scheme with
open access network is our next research.
Acknowledgment
This work was supported in part by the JSPS-NRF bilateral
joint research project.

IEICE TRANS. COMMUN., VOL.E96B, NO.1 JANUARY 2013

134

References
[1] J. Tang, X. Jin, Y. Zhang, X. Zhang, and W. Cai, A hybrid radio
over fiber wireless sensor network architecture, Proc. 3rd International Conference on Wireless Communications, Networking and
Mobile Computing, WiCom2007, pp.26752678, Shanghai, China,
Sept. 2007.
[2] M. Hossen, K. Kim, and Y. Park, Ecient clustering algorithm for
delay sensitive PON-based wireless sensor network, Proc. 9th International Conference on Optical Internet (COIN), pp.341343, The
Shilla Jeju, Korea, July 2010.
[3] S. Bandyopadhyay and E.J. Coyle, An energy ecient hierarchical clustering algorithm for wireless sensor networks, Proc. IEEE
INFOCOM 2003, vol.3, pp.17131723, 2003.
[4] M.P. McGarry, M. Reisslein, and M. Maier, Ethernet passive optical network architectures and dynamic bandwidth allocation algorithms, Proc. IEEE Commun. Surveys Tutorials, vol.10, no.3,
pp.4660, 3rd Quarter 2008.
[5] S. Blake, D. Black, M. Carlson, E. Davies, Z. Wang, and W. Weiss,
An architecture for dierentiated services, IETF, RFC 2475, Dec.
1998.
[6] IEEE Draft P802.3ahTM /D1.2, Media access control parameters,
physical layers and management parameters for subscriber access
networks, Dec. 2002, http://grouper.ieee.org/groups/802/3/efm
[7] G. Kramer and B. Mukherjee, Ethernet PON: Design and analysis of an optical access network, IEEE Photonic Netw. Commun.,
vol.3, no.3, pp.307319, July 2001.
[8] G. Kramer and B. Mukherjee, Interleaved polling with adaptive
cycle time (IPACT): A dynamic bandwidth distribution scheme in
an optical access network, IEEE Photonic Netw. Commun., vol.4,
no.1, pp.89107, 2002.
[9] G. Kramer, B. Mukherjee, S. Dixit, Y. Ye, and R. Hirth, On supporting dierentiated classes of service in EPON-based access network,
Proc. OSA J. Optical Networks, vol.1, no.8/9, pp.280298, 2002.
[10] C. Assi, Y. Ye, S. Dixit, and M. Ali, Dynamic bandwidth allocation for quality-of-service over Ethernet PONs, IEEE J. Sel. Areas
Commun., vol.21, no.9, pp.14671477, Nov. 2003.
[11] W. Willinger, M.S. Taqqu, and A. Erramilli, A bibliographical
guide to self-similar trac and performance modeling for modern high-speed networks, stochastic networks: Theory and applications, Royal Statistical Society Lecture Notes Series, vol.4, pp.339
366, Oxford University Press, Oxford, U.K., 1996.
[12] B. Lannoo, L. Verslegers, D. Colle, M. Pikavet, P. Demeester, and
M. Gagnaire, Thorough analysis of the IPACT dynamic bandwidth
allocation algorithm for EPONs, Proc. 4th International Conference
on Broadband Communications, Networks and Systems, pp.486
494, 2007.
[13] V. Rajendran, K. Obraczka, and J.J. Garcia-Luna-Aceves, Energyecient, collision-free medium access control for wireless sensor
networks, J. Wireless Netw., vol.12, no.1, pp.6378, 2006.
[14] I. Hwang, Z. Shyu, L. Ke, and C. Chang, A novel early DBA
mechanism with prediction-based fair excessive bandwidth allocation scheme in EPON, Proc. J. Comput. Commun., vol.31, no.9,
pp.18141823, 2008.

Monir Hossen
received a B.Sc. in EEE
from KUET, Bangladesh in 2002. He joined
KUET as a faculty member in the ECE Department in 2004. He completed his M.Sc. in Electronics Engineering at Kookmin University, Korea in 2010. Currently, he is working toward a
Ph.D. at the Interdisciplinary Graduate School
of Medicine and Engineering in University of
Yamanashi, Japan. His present research focus
on PON-based hybrid networks.

Masanori Hanawa
received a B.E., M.E.,
and Ph.D. from Saitama University, Japan, in
1990, 1992, and 1995, respectively. In 1995, he
joined University of Yamanashi, Japan as a research associate. Since 2002, he has been an associate professor of the university. His research
interests are optical signal processing, optical
fiber communications including optical code division multiplexing.