802 J. OPT. COMMUN. NETW./VOL. 5, NO.

7/JULY 2013

Monir Hossen and Masanori Hanawa

Dynamic Bandwidth Allocation Algorithm

With Proper Guard Time Management
Over Multi-OLT PON-Based Hybrid
FTTH and Wireless Sensor Networks
Monir Hossen and Masanori Hanawa

AbstractA passive optical network (PON) is a highly

capable access network that effectively converges several
service providers, without suffering from any bandwidth
deficiency. However, a PON consisting of a single optical
line terminal (OLT) for multiple service providers increases
the computational complexity for data packet processing in
the OLT, resulting in a longer time delay and more packet
loss. A multi-OLT PON-based access network is an effective
solution for reducing the computational complexity of data
packet processing in a hybrid network of multiple service
providers. The most important issue concerning the multiOLT PON is the sharing efficiency of upstream channels
among multiple service providers having different packet
lengths and data rates. In this paper, we propose a dynamic
bandwidth allocation algorithm called adaptive limited dynamic bandwidth allocation for multi-OLT PON (ALDBAM).
The proposed scheme is a modified version of adaptive limited dynamic bandwidth allocation (ALDBA) algorithms
that we proposed before, where both the ALDBA1 and
ALDBA2 schemes are combined with proper guard time
management and a modified multipoint control protocol.
The simulation results show that the ALDBAM scheme provides lower packet delay with higher bandwidth utilization, higher upstream efficiency, and higher throughput
than the conventional ALDBA1 and ALDBA2 schemes.
Index TermsDBA algorithm; FTTH; Hybrid network;
Multi-OLT PON; Wireless sensor network.

I. INTRODUCTION

ubiquitous city (u-City) is an autonomous city. A

significant number of service providers, such as
fiber-to-the-home (FTTH), wireless sensor networks
(WSNs), high-definition TV (HDTV) or video on demand
(VoD), and Femto networks (FNs), will comprise modern
u-Cities. Each home and business appliance in a future
u-City will be equipped with several sensor nodes remotely
monitored by owners and service providers. Usually, sensor
nodes will send notifications to the central office (CO)
Manuscript received January 7, 2013; revised April 18, 2013; accepted
May 22, 2013; published July 1, 2013 (Doc. ID 182841).
Monir Hossen is with the Interdisciplinary Graduate School of Medicine
and Engineering, University of Yamanashi, Kofu-shi, Yamanashi 400-8511,
Japan.
Masanori Hanawa (e-mail: hanawa@yamanashi.ac.jp) is with the Interdisciplinary Graduate School of Medicine and Engineering, University of
Yamanashi, Kofu-shi, Yamanashi 400-8511, Japan.
http://dx.doi.org/10.1364/JOCN.5.000802

1943-0620/13/070802-11$15.00/0

concerning any abnormality of any household or commercial device in a sensor network that is expected to be
deployed for all households and commercial systems in
the u-City, e.g., gas systems, temperature and pressure
monitoring systems, electric sparking and smoke detection
systems, automobile systems, and medical sensor nodes in
a hospital. Constructing a closed, specific-use network for
each individual application and accommodating several
users using different access terminals and servers require
an enormous amount of time and expense. To overcome the
enormous expense and deployment of several backbone
networks, passive optical network (PON)-based converged
networks have been proposed to connect the FTTH and
WSNs in a single optical network [1], because PON systems
can effectively share the upstream channel and the CO
equipment over high-speed and high-capacity bandwidth
demands [2].
One of the most critical issues for converging FTTH access networks and several service providers, e.g., WSN,
HDTV/VoD, FNs, etc., in a single optical line terminal
(single-OLT) PON is the requirement of more computational complexity for data packet processing in the OLT.
Because all of these service providers in a u-City have several features, e.g., device capacity diversity, application diversity, mobility, numbering and routing diversity, security,
and privacy, they significantly differ from conventional
access networks. This is why the current access network
architecture is not capable of integrating these service providers efficiently [3]. To mitigate this problem, some polling
algorithms have been proposed to allow additional time in
the OLT for computation and management in addition to
the guard time between every two successive optical network units (ONUs) [4]. More recently, a multi-OLT PONbased hybrid network combining FTTH access networks
and WSNs in a single PON has been proposed to reduce
the computational complexity in the OLT [5]. Even though
the single-OLT PON, as explained in [1], can be used to connect multiple service providers in a u-City like an openaccess network, a multi-OLT PON can play a vital role
in alleviating the ascended problems in a single-OLT
during management and computing of data packets from
multiple service providers with less overhead in the upstream channel. In a multi-OLT PON, each OLT independently handles the control messages and data packets of
2013 Optical Society of America

Monir Hossen and Masanori Hanawa

each service provider in a u-City. However, both an effective

network structure and an efficient dynamic bandwidth
allocation (DBA) scheme suitable for the multi-OLT PON
are essential for obtaining optimum services from the networks. In a time division multiple access PON, the system
performance depends on the sharing efficiency of the upstream channel. DBA plays a vital role in improving the
access network efficiency and bandwidth management of
the upstream channel.
Intensive research has been conducted on DBA algorithms over a PON, and among them, the popular schemes
are the limited service (LS) [4], early DBA (E-DBA) [6], excessive bandwidth reallocation (EBR) [7], and limited sharing with traffic prediction (LSTP) [8] schemes. In the LS
scheme, the granted time slot length depends on the dynamic network traffic, and the maximum length of a time
slot is upper-bounded by the maximum transmission window max. The E-DBA scheme reduces the idle period in the
usual DBA scheme by analyzing the historical traffic management. The EBR scheme redistributes the available
bandwidth from the lightly loaded ONUs to the heavily
loaded ONUs that are also incorporated with the priority
scheduling of network traffic. In the LSTP scheme, data delay is reduced by predicting the traffic that arrived during
the waiting time. The more recently proposed adaptive limited dynamic bandwidth allocation (ALDBA) [1] algorithms
are more suitable for multi-OLT PON-based hybrid networks, because these algorithms consider two different
maximum transmission windows, W max
FTTH for the FTTH terminal and W max
WSN for the cluster head (CH) of a WSN, for
two different service providers to improve the bandwidth
sharing efficiency. However, there is a possibility of utilizing guard time savings for the heavily loaded ONUs, which
can improve the quality of services (QoSs) for a multi-OLT
PON. Moreover, all of the existing algorithms have been
proposed for a single-OLT PON, and without any modification, they are not suitable for a multi-OLT PON. One of the
main reasons is that a single polling table in a multi-OLT
PON will be shared by all the OLTs that require modification in the multipoint control protocol (MPCP) [9] and
control message scheduling algorithm.
In this paper, we propose a new DBA algorithm for a
multi-OLT PON called adaptive limited dynamic bandwidth allocation for multi-OLT PON (ALDBAM) and analyze the performance of the multi-OLT PON using the
ALDBAM algorithm. Our proposed algorithm has the following characteristics. First, the conventional ALDBA1
and ALDBA2 schemes are combined in the ALDBAM
scheme with proper guard time management and the modified MPCP. Second, improved QoS provisioning is achieved
in a multi-OLT PON by utilizing the excess bandwidth savings from the reduced guard time for the heavily loaded
ONUs. Third, we provide a detailed network architecture
for a multi-OLT PON with upstream frame formats.
Fourth, a Gate message scheduling algorithm is modified
appropriately for the multi-OLT PON and ALDBAM algorithm. We have conducted extensive theoretical and
numerical analyses of QoS provisioning in terms of the
packet delay, bandwidth utilization, time jitter, upstream
efficiency, and throughput. The analyses are conducted

VOL. 5, NO. 7/JULY 2013/J. OPT. COMMUN. NETW. 803

for a multi-OLT PON consisting of two OLTs and two different FTTH and WSN service providers. Compared with the
single-OLT PON, the multi-OLT PON provides better
performance.
The rest of this paper is organized as follows. The network architecture with upstream and downstream frame
formats and the modified MPCP for a multi-OLT PON
are investigated in Section II. Guard time management
and the ALDBAM scheme with the Gate message scheduling algorithm in a multi-OLT PON are presented in
Section III. In Section IV, we explain the simulation environment in detail. Section V elucidates the simulation
results. Finally, our conclusions are presented in
Section VI.

II. NETWORK ARCHITECTURE AND THE MPCP

MULTI-OLT PON

FOR A

In this section, we first explain the network structure of

the hybrid multi-OLT PON combining an FTTH access network and WSN with upstream and downstream frame formats. Then we explain the modified version of the MPCP
suitable for the hybrid multi-OLT PON and the proposed
ALDBAM scheme.

A. Network Architecture of a Multi-OLT PON

One of the main aspects of the PON architecture that
helped it become a popular network is its simplicity. The
OLT is the main element of the network and is usually
placed in the CO. ONUs serve as an interface between
the OLT and customers through a splitter/combiner and optical fiber links. A tree-topology-based hybrid multi-OLT
PON consists of multiple OLTs that are connected to several ONUs of the FTTH access network and WSN service
providers. In the multi-OLT PON, a cluster-based WSN is
considered where each cluster consists of a static CH connected to the ONU through an optical fiber [10]. Most PON
systems consist of one OLT and N ONUs connected to the
FTTH terminals with different round-trip-time (RTT) delays. In contrast, the hybrid multi-OLT PON structure consists of multiple OLTs and several ONUs from different
service providers, e.g., ONUs connected to FTTH terminals
and ONUs connected to CHs of a WSN. In the multi-OLT
PON, the number of OLTs depends on the practical scenario, the number of service providers installed, in a u-City.
If a u-City comprises m different service providers, then the
number of OLTs will be m, e.g., OLT1 for FTTH terminals,
OLT2 for WSNs, OLT3 for HDTV/VoD, and OLT m for FNs.
Therefore, the number of OLTs and ONUs may vary; however, for simplicity, only two OLTs and four ONUs for both
services with both upstream and downstream packets are
shown in Fig. 1. Here, the single splitter can be divided into
two and a longer feeder optical fiber can be installed
between them, although it is not shown in Fig. 1.
In the downstream direction of the multi-OLT PON, each
OLT will alternately broadcast data to the network
through a passive splitter. The destination ONU will

804 J. OPT. COMMUN. NETW./VOL. 5, NO. 7/JULY 2013

Monir Hossen and Masanori Hanawa

Fig. 2. (a) Downstream data transmission and (b) upstream data

transmission in a hybrid multi-OLT PON.
Fig. 1. Network structure and data transmission for a hybrid
multi-OLT PON.

selectively extract the broadcasted data from the OLTs.

Figure 2(a) shows downstream data transmission in a
multi-OLT PON using the ALDBAM scheme.
In the upstream direction of the multi-OLT PON, data
packets from any ONU will reach both of the OLTs. However, data packets from an ONU will be accepted only by
the designated OLT. Other OLTs will discard the data packets from that ONU and will wait for data packets from the
next ONU. Figure 2(b) shows the upstream frame format of
the hybrid multi-OLT PON using the ALDBAM scheme.
Here, ONU1 is an ONU connected to an FTTH terminal
communicating with OLT1. In contrast, ONU2 is an
ONU connected to the CH of a WSN communicating with
OLT2. Each data packet of ONU1 contains a user identification (ID) number with the payload of that user and may
be multiplexed with different users if the ONU consists of
multiple users. In contrast, each 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
different sensor nodes of different services. Here, the node
ID is a unique number for each sensor node, and the 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 different service providers) to recognize each sensor node uniquely.

from the OLT to an ONU. On the other hand, the MPCP

uses a DBA algorithm to allocate the transmission window
or timeslots for every ONU and to share the single optical
fiber link with multiple ONUs [9]. Figure 3 illustrates the
MPCP for a hybrid multi-OLT PON. In the downstream
transmission, the MPCP maintains a timestamp with its
local time and broadcasts a Gate message to all the ONUs.
In the upstream transmission, all the ONUs share a
common channel to transmit data to the OLTs. The upstream transmission window of each ONU also contains
a report message at the end of its timeslots to request
the desired transmission window in the next time cycle
T cycle , depending on the ONUs buffer occupancy. Upon receiving the report message at the OLT, the MPCP incorporated with the DBA algorithm determines the allocated
transmission window and recalculates the required overhead and RTT to update the polling table. In a multiOLT PON, the conventional MPCP is modified, where both
OLTs share a common polling table to store the RTT of each
ONU that ensures timing synchronization among all ONUs.

B. MPCP for a Multi-OLT PON

The MPCP provides timing reference to synchronize
ONUs and allocate bandwidth or timeslots to ONUs to allow efficient transmission of data in the upstream direction. In the MPCP, timing synchronization among ONUs
is achieved by calculating the RTT and by maintaining a
polling table. The RTT depends on the physical distance

Fig. 3. MPCP operation in a hybrid multi-OLT PON.

Monir Hossen and Masanori Hanawa

III. GUARD TIME MANAGEMENT

AND

VOL. 5, NO. 7/JULY 2013/J. OPT. COMMUN. NETW. 805

ALDBAM SCHEME

In this section, we first explain the guard time management scheme in a multi-OLT PON, followed by the proposed ALDBAM scheme. The proposed ALDBAM scheme
utilizes the total guard time savings by fair distribution
to the heavily loaded ONUs. Furthermore, the scheduling
algorithm of the Gate message is modified for the
ALDBAM scheme.

A. Guard Time Management in a Multi-OLT PON

Fig. 5. Guard time management in a multi-OLT PON.

Guard time is required to avoid the turn on/off delay of

an optical transceiver and fluctuation of the RTT and to
provide times for clock and data recovery (CDR). A typical
PON system has to cope with these constraints by providing enough space as a guard time between the data packets
of every two consecutive ONUs. Figure 4 illustrates guard
time management in a conventional single-OLT PON
system. The guard time T G between every two consecutive
ONUs in the conventional PON is
T G
T off  T FRTT  T on  T CDR ;

(1)

where T off is the laser off time, T FRTT is the fluctuation of the
RTT, T on is the laser on time, and T CDR is the time for CDR.
In a conventional PON, max is constant for each ONU.
The maximum granted transmission window to each ONU
by the OLT, GOLT , can be calculated as follows:
GOLT

max

T on  T CDR 

T max
D

 T off ;

T FRTT  T CDR :

The total guard time savings T GS

can be calculated as follows:
T GS

NT G T G

M

(3)

in a multi-OLT PON

NT on  T off .

(4)

In a multi-OLT PON, the maximum transmission window for each ONU of an FTTH terminal is W max
FTTH , while
the maximum transmission window for each ONU of a
WSN is W max
WSN . The maximum granted transmission windows to the ONUs of the FTTH terminals and WSN by
OLT1 and OLT2 can be expressed, respectively, by
max
GOLT1
W max
FTTH
T CDR  T DFTTH ;

(5)

max
GOLT2
W max
WSN
T CDR  T DWSN ;

(6)

(2)

where T max
is the length of the maximum granted data
D
packets.
In a multi-OLT PON, two OLTs alternately receive data
from two consecutive ONUs, and the laser on and off times
can be easily avoided. When the data of ONU1 is received
by OLT1, OLT2 is in a sleeping condition at that time and
can wake up early to provide enough time for T on and receive data during T off of OLT1. In contrast, when the data
of ONU2 is received by OLT2, OLT1 is in a sleeping condition at that time and can wake up early to compensate for
T on and receive data during T off of OLT2. Figure 5 shows
the guard time management in a multi-OLT PON system
where only the CDR and RTT fluctuation times are used as
the guard time. T G M is the guard time between every two
consecutive ONUs in a multi-OLT PON in the following:

Fig. 4. Guard time management in a conventional PON.

TG

where GOLT1 is the granted window by OLT1, GOLT2 is the

granted window by OLT2, T max
DFTTH is the length of the
maximum granted data packets to the ONU of the FTTH
terminal, and T max
DWSN is the length of the maximum
granted data packets to the ONU of the WSN.

B. ALDBAM Scheme
In this scheme, we consider a hybrid multi-OLT PONbased access network with two OLTs and N ONUs. Here,
N is divided into two groups and N
N FTTH  N WSN ,
where N FTTH is the number of ONUs connected to the
FTTH terminals and N WSN is the number of ONUs connected to the CHs of the WSN. Usually the packet size
of the WSN is smaller, and the data rate is lower than
the FTTH access network. This is why the usual maximum
transmission window of the WSN will be smaller than the
maximum transmission window of the FTTH terminals,
max
i.e., W max
WSN < W FTTH . Owing to these packet length and data
rate differences, the total available bandwidth savings in
the proposed scheme, W TS , is calculated as in the ALDBA1
scheme [1]:
max
W TS
N WSN W max
FTTH W WSN :

(7)

806 J. OPT. COMMUN. NETW./VOL. 5, NO. 7/JULY 2013

Monir Hossen and Masanori Hanawa

W TS is divided by N to calculate the average available

bandwidth savings for each ONU (i.e., W avg
W TS N), and
this average bandwidth savings is used to provide some
transmission window to the deferred data during the waiting time between the transmission of the Gate and Report
messages. Usually, the waiting time in a PON is equal to
the RTT of each ONU and delay of the Gate starting time
from the OLT. The OLTs predict the amount of deferred
data during the waiting time for each ONU and allocate
the additional bandwidth up to W avg in addition to the
granted window GOLT1 or GOLT2. Prediction of the deferred
data during the waiting time depends on the current queue
occupancy, RTT of each ONU, and Gate starting delay from
the OLTs:
W pred


i; j

WR
i; j RTTi  T GD 
T acq
i; j

(8)

where W pred
i; j is the predicted window size for ONU i at time
cycle j, T acq
i; j is the acquisition time of the present data in the
queue, W R
i; j is the requested window by ONU i at time cycle

j, T GD is the Gate starting delay, and W pred

W avg .
i; j
Owing to the bursty nature of the network traffic [11],
some ONUs might have traffic demand less than W max
FTTH
or W max
WSN, called lightly loaded ONUs, while other ONUs
max
might have traffic demand higher than W max
FTTH or W WSN,
called heavily loaded ONUs. This results in some amount
of excessive bandwidth from the lightly loaded ONUs. The
total excessive bandwidth in the hybrid multi-OLT PON, as
in the ALDBA2 scheme [1], is calculated by
W excess
Total; j

LX
FTTH

m
1

 LX

WSN

R
R
W max
W max
FTTH; j W m; j 
WSN; j W n; j ;
n
1

(9)
where W excess
Total; j is the total excessive bandwidth at time cycle
j; LFTTH and LWSN are the number of lightly loaded ONUs
connected to the FTTH terminals and CHs of the WSN, respectively; and W R
mn; j is the requested window size of a
lightly loaded ONU mn at time cycle j.
In the ALDBAM scheme, this total excess bandwidth
from the lightly loaded ONUs is incorporated with T GS T
in Eq. (4). These two excess bandwidth savings from Eqs. (4)
and (9) can be fairly distributed to the heavily loaded
ONUs, without changing the length of T cycle. The following
equation is used to fairly distribute the total excessive
bandwidth in Eq. (9) and the total guard time savings in
Eq. (4) among the heavily loaded ONUs to solve the congestion problem in the hybrid multi-OLT PON:
W excess


i; j

R
W excess
Total; j  T GS T W i; j
PH
;
R
k
1 W k; j

(10)

is the excessive bandwidth for ONU i at time

where W excess
i; j
cycle j and H is the number of heavily loaded ONUs.
The bandwidth allocation formulas for the ALDBAM
scheme in a multi-OLT PON are as follows:

(
j
Gi;OLT1

pred
WR
i; j W i; j

For lightly loaded ONUs

excess
W max
W pred
For heavily loaded ONUs
i; j
FTTH W i; j

(11)
(
j


Gi;OLT2

pred
WR
i; j W i; j

excess
W max
W pred
i; j
WSN W i; j

For lightly loaded ONUs

For heavily loaded ONUs

(12)
j
where Gi;OLT1
is the granted window to ONU i of the FTTH
j
terminal by OLT1 at time cycle j and Gi;OLT2
is the granted
window for ONU i of the CH of the WSN by OLT2 at time
cycle j.

An illustrative example of bandwidth allocation in the

ALDBAM scheme for heavily loaded ONUs is shown in
Fig. 6. The bandwidth allocation conditions in Fig. 6 follow
Eqs. (11) and (12) for heavily loaded ONUs. Here, T cycle; j is
the length of a polling cycle at time cycle j. The maximum
max
transmission window W max
FTTH or W WSN and excessive bandexcess
excess
excess
widths W 1; j ; W 2; j ; ; W i; j ; ; W excess
with predicted
N; j
pred
pred
pred
windows W pred
1; j ; W 2; j ; ; W i; j ; ; W N; j are alternately allocated by OLT1 or OLT2 to the heavily loaded ONUs
1; 2; ; i; ; N of both service providers at time cycle j.
pred
In contrast, the requested windows W R
are ali; j with W i; j
located by OLT1 or OLT2 to the lightly loaded ONU i at
time cycle j, as shown in Eqs. (11) and (12).

As network complexity has increased with the history of

Internet development due to the inclusion of more diverse
and new inconsistent functions [3], the ALDBAM scheme
also requires more computational complexity than the
LS scheme [4]. Because the ALDBAM scheme needs to calculate predicted traffic, access bandwidth, and lightly
loaded and heavily loaded ONUs, this requires a larger
number of summation and multiplication operations than
the LS scheme. However, these complexities do not heavily
affect the online bandwidth allocation. Moreover, deployment of multiple OLTs can share the overall complexity
to reduce the computing time more than the singleOLT PON.
The main differences between the proposed ALDBAM
scheme and the existing ALDBA1 and ALDBA2 schemes
in [1] are as follows:
1) Consideration of multiple OLTs for multiple service
providers in a single PON.
2) Calculation of the total guard time savings by proper
guard time management in a multi-OLT PON and utilization of this guard time savings for heavily
loaded ONUs.
3) Appropriate modification of the MPCP for a multiOLT PON.
4) Provision of detailed analysis of upstream and downstream frame formats with different maximum transmission windows for different OLTs and service
providers.
5) Consideration of the Gate starting delay to calculate
the predicted traffic in the waiting time.

Monir Hossen and Masanori Hanawa

VOL. 5, NO. 7/JULY 2013/J. OPT. COMMUN. NETW. 807

RTT1 +T GD 1

D a ta fro m ON U 1 (F TTH)
max
G1 = WFTTH
+ W1,excess
+ W1,pred
j
j

W1,pred
j

ON U 1

RTT2+TGD2
ON U 2

D a ta fro m ON U 2 (WS N )
max
G 2 = WWSN
+ W2,excess
+ W2,pred
j
j

W2pred
,j

RTTN +T GD 1

D a ta fro m ON U N (WS N )

ON U N
R TGD1
R =W
1
1, j

max
G 2 = WWSN
+ WN,excess
+ WN,pred
j
j

WNpred
,j
max

max

GI EO TD(FTTH)

OLT1

R
=W
2, j

TGD2

max
GI EO TD(WSN)

OLT2

R
=W
N, j

max

GI EO TD(WSN)

Tim e c y c le j, Cy c le tim e = T cycl e, j

Tim e c y c le j-1

GI EO TD(FTTH)

TGD2

Tim e c y c le j+1

G 1/2 = Ga t e m essa ge fr om OLT1/OLT2,T GD1/2 = Ga t e st a r t in g dela y of OLT1/OLT2, RTT+T GD = Wa it in g t im e, a n d R = Repor t m essa ge

Fig. 6. Illustrative example of the ALDBAM scheme for heavily loaded ONUs.

6) Modification of the Gate message scheduling algorithm

for a multi-OLT PON and the ALDBAM algorithm.

C. Gate Message Scheduling Algorithm in a

Multi-OLT PON
In upstream transmission, a scheduling algorithm for
Gate messages in a multi-OLT PON is very important to
prevent data collision due to multiple ONUs transmitting
at the same time. Figure 7 shows the Gate message scheduling algorithm in a multi-OLT PON for the ALDBAM
scheme. As the scheduling of Gate messages depends on
the RTT and granted window sizes of different ONUs, a
starting Gate message can be sent by any of the OLTs. Gate
messages for different ONUs in a multi-OLT PON are
scheduled using the following formulas:
j
TGi1;

TGi;1 j  RTTi  T FRTT   T CDR  T DFTTH
2

RTTi1  T FRTT ;

in terms of the average packet delay, bandwidth utilization,

jitter, upstream efficiency, and throughput. All these
parameters are evaluated by simulation results. The evaluation was performed using laboratory-made computer
simulation programs. We considered a hybrid PON
architecture with two OLTs and 32 ONUs in a tree topology. In the ALDBAM scheme, we have considered two different maximum transmission windows for two different
service providers for upstream transmission and two different maximum transmission windows for two OLTs for
downstream transmission. We have also incorporated the
impact of guard time savings by fairly distributing to
the heavily loaded ONUs. The downstream and upstream
channel speeds were considered at 1 Gbps. The distance
from an ONU to the OLT is assumed to be random and
in the range of 1020 km. All the data packets were assumed to have the same priority, meaning the service policy was on a first-in first-out (FIFO) basis with an infinite
buffer size for each ONU. A highly bursty self-similar

(13)

j
j
TGi2;

TGi1;
 RTTi1  T FRTT   T CDR
1
2

 T DWSN RTTi2  T FRTT ;

(14)

j
are the time epochs for OLT1 and
where TGi;1 j and TGi1;
2
OLT2 when Gate messages are transmitted to ONU i and
j
ONU i  1, respectively, at time cycle j, and TGi2;
is the
1
time epoch for OLT1 when the Gate message is transmitted
to ONU i  2 at time cycle j.

IV. PERFORMANCE EVALUATION

BY

SIMULATION

In this section, the performance of the proposed

ALDBAM scheme for a hybrid multi-OLT PON is evaluated

Fig. 7. Scheduling diagram for a Gate message in a multi-OLT

PON.

808 J. OPT. COMMUN. NETW./VOL. 5, NO. 7/JULY 2013

Monir Hossen and Masanori Hanawa

network traffic model, as most network traffic can be

characterized by self-similar and long range dependence
[11], was used to generate the data packets for both the
FTTH terminals and sensor nodes of the WSN. This traffic model generated traffic from 0 to multiple packets in
each active ONU in every time cycle, and the total requested window size of an ONU depended on the number
of packets multiplied by the maximum length of a
max
packet, PBmax
FTTH for FTTH terminals and PBWSN for the
CH of the WSN. The maximum packet lengths for the
max
FTTH terminals Bmax
FTTH and WSNs BWSN were 1500 bytes
[12] and 1024 bytes [13], respectively. The processing
time was assumed to be 10 s for the proposed ALDBAM
scheme, as used by Hwang et al. [6]. All analyses were
performed for a nonuniform offered load in the range
of 01.0 with a variable cycle time in the range of
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 I.
The hybrid multi-OLT 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 ALDBAM scheme is the reduction of the end-to-end packet delay by allocating a larger
transmission window to the heavily loaded ONUs from
the lightly loaded ONUs and guard time savings. Reduction of packet delay is also achieved by reducing the data
processing time and guaranteed scheduling of Gate
messages in the multi-OLT PON.
The bandwidth utilization BWU of a multi-OLT PON
using the ALDBAM scheme is expressed by

BWU

N FTTH GOLT1  N WSN GOLT2

;
N FTTH GOLT1  N WSN GOLT2  NT C

(15)

TABLE I
SIMULATION PARAMETERS
Symbol
N
N OLT
D
T on
T off
T FRTT
T CDR
T cycle
T proc
Ru
BR
BE
Bmax
FTIH
Bmax
WSN
P

Quantity

Value

Total number of ONUs

Number of OLTs
Distance between OLTs and ONUs
Laser on time
Laser off time
Fluctuation of RTT
CDR time
Cycle time
Processing time
Transmission speed
Length of report message
Length of Ethernet overhead
Maximum packet length of the
FTTH terminal
Maximum packet length of the WSN
Number of generated packets

32
2
1020 km
1.5 s
1.5 s
1.5 s
0.5 s
0.53 ms
10 ms
1 Gbps
576 bits
304 bits
1500 bytes
1024 bytes
010

where T C is the summation of BR Ru, BE Ru , T FRTT , and

T CDR . The proposed scheme can achieve better bandwidth
utilization by utilizing excessive bandwidth from lightly
loaded ONUs and guard time savings for the heavily loaded
ONUs.
The burst network traffic and DBA in a PON provide
variation in T cycle in every time cycle. Due to this variation
in T cycle , the arrival times of data packets fluctuate in different time cycles. To measure the variation in the data
packet arrival time, the jitter performance of the proposed
ALDBAM scheme was analyzed. The jitter may be calculated by
Jitter

1
n

q
X j
2
T avl T j1
avl  ;

(16)

where n is the total number of time cycles, T javl is the data

packets arrival time at time cycle j, and j
1; 2; 3; ; n.
The ratio between the successful upstream transmission
and the total generated traffic in the network is called the
upstream efficiency. The expression for the upstream
efficiency of a PON system is
UE

N FTTH GOLT1  N WSN GOLT2

;
max
N FTTH PBmax
FTTH  N WSN PBWSN  NT C

(17)

where UE is the upstream efficiency.

V. SIMULATION RESULTS
In this section, the system performance of the proposed
ALDBAM scheme for a multi-OLT PON is compared with
that of the existing ALDBA1 and ALDBA2 schemes for a
single-OLT PON. All the performance parameters are analyzed for nonuniform burst traffic in both the upstream
and the downstream directions. All the results are presented using contour plots over a wider range of offered
loads and cycle times. Lighter colors signify better performance in all of the contour plots in this paper.
Figures 8(a)8(c) show the end-to-end average packet delay of the existing ALDBA1 and ALDBA2 and the proposed
ALDBAM schemes, respectively, for N FTTH N WSN
1616
using contour plots for different offered loads and cycle
times. From these three contour plots, the ALDBAM
scheme clearly provides a wider area of lowest packet delay
in both directions of offered loads and cycle times. On the
other hand, the highest packet delay in the ALDBAM
scheme is 1.8 ms, whereas the highest packet delays in
the ALDBA1 and ALDBA2 schemes are 3 and 2.5 ms, respectively. Figure 8(d) shows a comparison of the average
packet delay among the three schemes for a 2 ms cycle
time. The ALDBAM scheme provides approximately 75%
and 30% less delay than the ALDBA1 and ALDBA2
schemes, respectively, at an offered load of 1.0. However,
the effectiveness of the ALDBAM scheme becomes more
significant at higher data rates and a larger number of
service providers and OLTs.
Figures 9(a) and 9(b) compare the average packet delay
among the ALDBA1, ALDBA2, and ALDBAM schemes for

Monir Hossen and Masanori Hanawa

Fig.
8. Average
packet
N FTTH N WSN
1616.

delay

VOL. 5, NO. 7/JULY 2013/J. OPT. COMMUN. NETW. 809

in

milliseconds

for

a 2 ms cycle time by changing the ratio of the number of

ONUs connected to the FTTH terminals and to the CHs
of the WSN. The delay characteristics of the three schemes
in Figs. 9(a) and 9(b) are also similar to those in Fig. 8(d).
However, the average packet delays for all three schemes
are far less when the number of ONUs connected to the
CHs of the WSN is larger than the number connected to
the FTTH terminals, as shown in Fig. 9(b), where
N FTTH N WSN
824. The first reason is that a larger number of ONUs from the WSN provides less aggregated traffic
in the network as the data rate is lowered, and the packet
size is smaller than those of the FTTH terminals. The
second reason is that a larger number of ONUs from the
WSN provides more bandwidth savings that is utilized
by the deferred data.
The contour plots of Figs. 10(a)10(c) show the bandwidth utilization of the ALDBA1, ALDBA2, and ALDBAM
schemes, respectively. From the analysis of these three contour plots, it is clear that the ALDBAM scheme provides far
superior bandwidth utilization than both the ALDBA1 and
ALDBA2 schemes. The highest bandwidth utilization in
the ALDBAM scheme is 0.95. In contrast, the highest bandwidth utilization in the ALDBA1 and ALDBA2 schemes is
0.8 and 0.9, respectively. Moreover, the ALDBAM scheme
provides higher bandwidth utilization from a much lower

Fig. 9. Comparison of average packet delay for a 2 ms cycle time.

Fig. 10. Bandwidth utilization for N FTTH N WSN
1616.

value of the offered load. From the comparison of the bandwidth utilization at a 2 ms cycle time in Fig. 10(d), the
bandwidth utilization in the ALDBAM scheme exceeds
0.85 at an offered load of 0.12. In contrast, the bandwidth
utilization exceeds 0.85 at an offered load of 0.45 and 0.85
in the ALDBA2 and ALDBA1 schemes, respectively. Similarly, if we draw a horizontal line at a bandwidth utilization of 0.85, as shown in Figs. 11(a) and 11(b), we can then
see that the ALDBAM scheme continually provides similar
performance for different ratios of the number of ONUs
connected to the FTTH and WSNs.
Usually, in a DBA scheme, we cannot avoid jitter because
of bursty network traffic. The contour plots in Fig. 12 show
the jitter performance of the existing ALDBA1, ALDBA2,
and proposed ALDBAM schemes. From the contour plots, it
is clear that the ALDBAM scheme provides less jitter than
the ALDBA1 and ALDBA2 schemes. The comparison of
jitter for a 2 ms cycle time with different numbers of ONUs
from two service providers is shown in Figs. 12(d), 13(a),
and 13(b). Jitter characteristics at a 2 ms cycle time for
every combination of ONUs from the FTTH terminals
and WSN are similar in all the three schemes.
Upstream efficiencies are compared among the three
schemes by the contour plots in Figs. 14(a)14(c). In this
case, the proposed ALDBAM scheme provides better

Fig. 11. Comparison of bandwidth utilization for a 2 ms cycle

time.

810 J. OPT. COMMUN. NETW./VOL. 5, NO. 7/JULY 2013

Fig. 12. Jitter in milliseconds for N FTTH N WSN
1616.

upstream efficiency than the ALDBA1 and ALDBA2

schemes. The highest upstream efficiency area in the
ALDBAM scheme is broadened in the directions of both
maximal cycle times and offered loads. This means that
the ALDBAM scheme can provide better performance from
a lower offered load and cycle time to a higher offered load
and cycle time. From the analysis of Figs. 14(d), 15(a), and
15(b), we can say that the ALDBAM scheme is also consistent for maintaining higher upstream efficiency for every
combination of numbers of ONUs from two different service
providers. To compare the results of the upstream efficiency among the three schemes more efficiently, we draw
a horizontal line at the 80% upstream efficiency level in
Figs. 14(d), 15(a), and 15(b). From these three figures, it is
clear that the ALDBAM scheme provides about a two times
greater offered load than the ALDBA2 scheme for both
N FTTH N WSN
1616 and N FTTH N WSN
824 with an
upstream efficiency higher than 80%. Moreover, the ALDBAM scheme provides four times more offered load than
the ALDBA2 scheme for the N FTTH N WSN
248 case
with an upstream efficiency higher than 80%. However,
the ALDBA1 scheme never provides an upstream efficiency
higher than 80%.

Monir Hossen and Masanori Hanawa

Fig. 14. Upstream efficiency for N FTTH N WSN
1616.

Fig. 15. Comparison of upstream efficiency for a 2 ms cycle time.

ALDBAM schemes, respectively. As expected, ALDBA1

has the lowest throughput, and the maximum throughput achieved in the contour plot of Fig. 16(a) is 0.6. On
the other hand, the maximum throughput achieved
by the ALDBA2 scheme is 0.8, as shown in the contour
plot in Fig. 16(b). Ultimately, the ALDBAM scheme
achieves the highest throughput of 0.9 in Fig. 16(c).

Finally, we compare the improvement in throughput

when the ALDBAM scheme is used for a hybrid multiOLT PON. Figures 16(a)16(c) show the contour plots of
the throughput for different offered loads and cycle times
for the existing ALDBA1, ALDBA2, and proposed

Fig. 13. Comparison of jitter for a 2 ms cycle time.

Fig. 16. Throughput for N FTTH N WSN
1616.

Monir Hossen and Masanori Hanawa

VOL. 5, NO. 7/JULY 2013/J. OPT. COMMUN. NETW. 811

compared with the ALDBA1 and ALDBA2 schemes,

respectively.

ACKNOWLEDGMENT
This work was supported in part by the JSPS-NRF
bilateral joint research project.

Fig. 17. Comparison of throughput for a 2 ms cycle time.

The main reason for the low throughput for the ALDBA1
scheme is less utilization of the upstream channel due
to the lightly loaded ONUs, whereas the ALDBAM
scheme gains the utilization of excess bandwidth from
the lightly loaded ONUs and guard time savings for the
heavily loaded ONUs. From the comparison of throughput
among the three schemes for the 2 ms cycle time, the
ALDBAM scheme achieved more than 15% and 35% higher
throughput than the ALDBA2 and ALDBA1 schemes,
respectively, for every case of the ratio of ONUs from
two different service providers, as shown in Figs. 16(d),
17(a), and 17(b).
Even though the existing ALDBA2 scheme [1] can
achieve better results in a single-OLT PON than the conventional LS scheme [4], it still has its limitations, because
the ALDBA2 scheme does not consider the utilization of
guard time savings in a multi-OLT PON. In a multi-OLT
PON, the computation time for data packet processing in
the OLT is also reduced by dividing the upstream traffic
from the ONUs of different service providers among multiple OLTs. The proposed ALDBAM scheme copes with all
the limitations of the ALDBA1 and ALDBA2 schemes
and provides enhanced performance. Moreover, the effectiveness of the proposed scheme will be more significant
if the analyses are repeated for a larger number of OLTs
and service providers.

VI. CONCLUSION
Our proposed ALDBAM scheme enhances the performance of a multi-OLT PON by the reduction of data
processing time, fair distribution of excess bandwidth from
the lightly loaded ONUs to the heavily loaded ONUs,
proper guard time management, and the perfect scheduling algorithm of Gate messages from multiple OLTs. The
proposed ALDBAM scheme outperformed the ALDBA1
and ALDBA2 schemes in terms of the average packet delay,
bandwidth utilization, jitter, upstream efficiency, and
throughput. The main contribution of the proposed ALDBAM scheme is that it can provide better bandwidth sharing efficiency and utilization due to smaller cycle times and
lower offered loads. The ALDBAM scheme utilizes guard
time savings in a multi-OLT PON and provides better
QoS than the ALDBA1 and ALDBA2 schemes. The ALDBAM scheme provided 75% less delay with 35% higher
throughput and 30% less delay with 15% higher throughput for a 2 ms cycle time at an offered load of 1.0 when

REFERENCES
[1] M. Hossen and M. Hanawa, Adaptive limited dynamic bandwidth allocation scheme to improve bandwidth sharing
efficiency in hybrid PON combining FTTH and wireless
sensor networks, IEICE Trans. Commun., vol. E96-B,
no. 1, pp. 127134, Jan. 2013.
[2] L. G. Kazovsky, W. Shaw, D. Gutierrez, N. Cheng, and S.
Wong, Next-generation optical access networks, J.
Lightwave Technol., vol. 25, no. 11, pp. 34283442,
Nov. 2007.
[3] AKARI Project Group, sponsored by NICT, NEW generation
network architecture AKARI conceptual design, ver. 2, Aug.
2009 [Online]. Available: http://www.nict.go.jp/en/photonic_
nw/archi/akari/conceptdesign_e.html#block_top1.
[4] G. Kramer, B. Mukherjee, and G. Pessavento, IPACT: A
dynamic protocol for an Ethernet PON (EPON), IEEE
Commun. Mag., vol. 40, no. 2, pp. 7480, 2002.
[5] M. Hossen and M. Hanawa, Network architecture and performance analysis of multi-OLT PON for FTTH and wireless
sensor networks, Int. J. Wireless Mobile Netw., vol. 3, no. 6,
pp. 115, Dec. 2011.
[6] I. Hwang, Z. Shyu, L. Ke, and C. Chang, A novel early DBA
mechanism with prediction-based fair excessive bandwidth
allocation scheme in EPON, J. Comput. Commun., vol. 31,
no. 9, pp. 18141823, June 2008.
[7] 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.
[8] Y. Luo and N. Ansari, Limited sharing with traffic prediction
for dynamic bandwidth allocation and QoS provisioning over
Ethernet passive optical networks, J. Opt. Netw., vol. 4, no. 9,
pp. 561572, Sept. 2005.
[9] Carrier Sense Multiple Access with Collision Detection
(CSMA/CD) Access Method and Physical Layer Specifications, Amendment: Media Access Control Parameters, Physical Layers, and Management Parameters for Subscriber
Access Networks, IEEE Standard 802.3ah, 2004.
[10] M. Hossen, K. Kim, and Y. Park, A PON-based large sensor
network and its performance analysis with Sync-LS MAC
protocol, Arabian J. Sci. Eng., to be published [Online]. Available: http://link.springer.com/content/pdf/10.1007%2Fs13369
01305718.pdf#.
[11] W. Willinger, M. S. Taqqu, and A. Erramilli, A bibliographical
guide to self-similar traffic and performance modeling for
modern high-speed networks, in Stochastic Networks: Theory
and Applications (Royal Statistical Society Lecture Notes
Series, vol. 4). Oxford, UK: Oxford University, 1996,
pp. 339366.
[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, in IEEE 4th Int.
Conf. Broadband Communication, Networks and Systems,
Raleigh, NC, Sept. 2007, pp. 486494.

812 J. OPT. COMMUN. NETW./VOL. 5, NO. 7/JULY 2013

Monir Hossen and Masanori Hanawa

[13] V. Rajendran, K. Obraczka, and J. J. Garcia-Luna-Aceves,

Energy-efficient, collision-free medium access control for
wireless sensor networks, Wireless Netw., vol. 12, no. 1,
pp. 6378, Feb. 2006.

Medicine and Engineering at the University of Yamanashi, Japan.

His present research focuses on PON-based hybrid networks combining FTTH and wireless sensor networks and their bandwidth
allocation algorithms.

Monir Hossen received a B.Sc. degree in

electrical and electronic engineering from
Khulna University of Engineering & Technology (KUET), Bangladesh, in 2002. He
joined KUET as a faculty member in the
Electronics and Communication Engineering Department in 2004. He completed
his M.Sc. in electronics engineering at
Kookmin University, South Korea, in 2010.
Currently, he is working toward a Ph.D.
at the Interdisciplinary Graduate School of

Masanori Hanawa received B.E., M.E.,

and Ph.D. degrees from Saitama University,
Japan, in 1990, 1992, and 1995, respectively.
In 1995, he joined the University of
Yamanashi, Japan, as a Research Associate.
Since 2002, he has been an Associate Professor at the university. His research interests
include optical signal processing and optical
fiber communications, including optical code
division multiplexing.