Académique Documents
Professionnel Documents
Culture Documents
Computer Communications
journal homepage: www.elsevier.com/locate/comcom
a r t i c l e i n f o a b s t r a c t
Article history: Mobile prole of WiMAX technology, based on IEEE 802.16e standard, is due to support multi-application
Received 20 July 2010 services, while beneting a wide range of implementation exibilities. The current document presents a
Received in revised form 23 March 2011 simple capacity estimation methodology for WiMAX systems. In this paper, the term capacity concerns
Accepted 4 August 2011
with the maximum number of simultaneous multi-service users that each specic Mobile WiMAX access
Available online 16 August 2011
point can jointly support. A dynamic PHY+MAC overhead removal method is formulated to achieve an
improved system goodput with respect to users distribution and multi-burst construction strategy. A ser-
Keywords:
vice delivery model is presented that investigates the QoS requirements of the services to be supported,
Mobile WiMAX
IEEE 802.16e
along with elaborating an application prole as a consistent input for capacitating and dimensioning
Capacity estimation studies. The minimum resource consumption calculation for the joint-application users is derived at each
Overhead given time. An incremental algorithm compares the optimal available resources with minimum service
Multi-service trafc demand for each number of users to arrive at maximum system capacity. Using the proposed algorithm,
different simulation scenarios are studied based on the most used WiMAX implementation parameters in
practice. The simulation results prove the signicant roll of overhead calculation in performance evalu-
ation studies. Furthermore, these results can be used for network planning and dimensioning purposes,
as well as providing reference measures for scheduling performance analysis and detailed simulations.
2011 Elsevier B.V. All rights reserved.
0140-3664/$ - see front matter 2011 Elsevier B.V. All rights reserved.
doi:10.1016/j.comcom.2011.08.002
110 A.M. Ahmadzadeh et al. / Computer Communications 35 (2012) 109119
(determined by capacity and number of users to be supported). resources over time. Fig. 1 illustrates the TDD-OFDMA frame struc-
Furthermore, by knowledge of the two key features above, the ture. As can be seen, the UL ? DL subframes are separated with a
operators can have a good approximation for investment estima- Receive Transition Gap (RTG) dened as 60 ls for the bandwidths
tion of network deployment. of 5 and 10 MHz. The DL ? UL Transmit Transition Gap is denoted
The rest of the paper is organized as follows: in Section 2 we as TTG. There are 47 OFDM symbols in the entire frame, for the
study the Mobile WiMAX standard with a clear description over aforesaid bandwidths (transition gaps excluded). The duration of
the features related to our methodology. Section 3 presents an each of the subframes is adjustable (to support asymmetric trafc)
overview of the previous works related to capacity estimation for and is reected in DL:UL symbol ratio, dened as 35 symbols in DL
WiMAX systems. We present an overview of the proposed capacity and 12 for UL for the ratio of 3:1, to 26 vs. 21 for ratio of 1:1,
estimation algorithm in Section 4. In Section 5 we calculate the (35/12, 34/13, . . . , (N/M), . . . , 26/21); where N + M = 47. In order to
systems available resources by formulating a PHY+MAC overhead mitigate the Inter Symbol Interference (ISI), a Cyclic Prex (CP) is
removal methodology and with a look over resource allocation introduced in the standard that copies a portion of the ending of
strategies. In Section 6 we present a system delivery model based the useful symbol (Tb) to the beginning of the OFDM signal wave.
on data-rate requirements of different services to be supported and Therefore the symbol duration (Ts) is; (Tb) increased with an order
elaboration of a statistical application prole. Computer simula- of 1/G, where G can have a value of 4, 8, 16 or 32. As long as the CP
tions and numerical results are presented in Section 7 followed duration is larger than the channel delay spread, the ISI is com-
by a conclusion at last section. pletely eliminated.
Fig. 1. OFDMA-TDD frame structure indicating control messages and resource allocation procedure.
Table 2
SOFDMA PHY parameters in PUSC mode.
BW FFT Size No. Data SCs No. Pilot + Null SCs No. Sub-Channels
DL UL DL UL DL UL
5 MHz 512 360 272 60 + 92 136 + 104 15 17
10 MHz 1024 720 560 120 + 184 280 + 184 30 35
2.4. Resource allocation contain a number of data-regions and each subframe may contain
several burst of varying size (with maximum number of 16 bursts
The minimum possible data allocation unit (MAU), assigned to a in DL [3]). As pictured in Fig. 1, when a specic DL-burst is meant
connection by scheduler, is called a slot. In OFDMA-PHY a slot re- for a given MS (Mobile Station), the burst properties along with the
quires both a time and a frequency dimension. Hence, The number unique Connection Identier(s) (CID(s)) of the corresponding con-
of available slots in either of the DL or UL subframes depends on: nection(s), is reected in the related Information Element (IE)
(a) the number of OFDM symbols in each subframe specied by within the DL-MAP message. Note that each MS may simulta-
DL:UL symbol ratio, and (b) The applied subcarrier permutation neously request different applications, hence having different con-
mode. Using PUSC, the 48 data subcarriers of a downlink-subchan- nections and different CIDs. All users track the DL-MAP and each
nel are equally divided over 2 OFDMA symbols (24 + 24). While, in user prepares to receive the downlink data which is addressed to
uplink the subcarriers are spread over 3 symbols following its specic CID(s). The DL-MAP-IE informs the user about the rect-
(12 + 24 + 12) scheme [5]. Each slot may carry different amount angular place of the DL resource allocation within the DL subframe,
of data-bits depending on the applied modulation and FEC coding as well as the necessary information for decoding the message.
level, resulting in different data-block-size per sub-channel. IEEE Similarly, the resource allocation in UL is specied with UL-MAP-
802.16e standard supports QPSK, 16-QAM and 64-QAM modula- IEs. However, the UL data burst is allocated as a horizontal strip
tion with various rates of convolutional codes (CC) or turbo codes per subscriber. The transmission starts at a particular slot and con-
(CTC), while the support for 64-QAM is optional in uplink. Table 3 tinues during the entire UL subframe. The horizontal allocation is
shows the block-size related to each Modulation and Coding used to minimize the number of subcarriers for each MS and hence
Scheme (MCS) in PUSC mode. maximizes the power per subcarrier.
A contagious series of slots can be assigned to a given user,
called user data-region. In other words, a data region, commonly 2.5. Application classes and QoS
depicted as a rectangle, is a two dimensional allocation of a group
of contiguous subchannels (i.e. with contiguous logical numbers) The overall applications to be supported with Mobile WiMAX
in a group of contiguous OFDM symbols. Each data-burst may are classied into 5 major categories by WiMAX Forum [4]. These
classes are; Interactive Gaming, VoIP, Streaming, Web Browsing,
and File Transfer. The major metrics for this classication are the
Table 3 data-rate, the latency and the jitter requirements of each applica-
MCS block-size and distribution. tion. For each application (i.e. data service), these parameters must
Tab Modulation CTC Block size (bit) Min SNR Weight
be known by the system to be able to guarantee a quality user
experience. In other words, based on QoS requirements of different
1 QPSK 1/2 48 2.9 0.20
2 QPSK 3/4 72 6.3 0.17
existing data services, the system decides on their transmission
3 16-QAM 1/2 96 8.6 0.20 ordering and scheduling on the air interface. In WiMAX, this task
4 16-QAM 3/4 144 12.7 0.13 is done by assigning Service Flows to each connection between
5 64-QAM 1/2 144 13.8 0.15 the BS and the MS at MAC level. In fact, a service ow is a unidirec-
6 64-QAM 2/3 192 16.9 0.10
tional ow of packets that is provided with a particular set of QoS
7 64-QAM 3/4 216 18 0.05
parameters. These service ows are: UGS, rtPS, ErtPS, nrtPS and BE
112 A.M. Ahmadzadeh et al. / Computer Communications 35 (2012) 109119
in decreasing serving priority order [4]. The properties of these ser- highly correlated to the number of active users in the system, their
vices ows will be further studied in Section 6.2. service types and the frequency reuse pattern. However, the calcu-
A typical service ow management strategy for WiMAX systems lations in this paper are OFDM based and the vast details of TDD-
is made of following components [14]; (a) an admission control OFDMA congurations are not discussed.
mechanism determines whether a new request for a connection One key element in capacity estimation analyzes is to dene a
can be satised according to the remaining free resources, (b) a service delivery model for the system under study. The service
buffer manager controls the buffer size (c) and the scheduling delivery model needs the knowledge of available system resources,
mechanism determines the sending sequence of the packets of along with the resource consumption, also known as system load.
the admitted connections. The scheduler decides which ones to In Mobile WiMAX networks, the users may have different sets of
send and when to send. Thus, there may be one separate queue trafc with different QoS requirements, be in different receiving
for each service ow and each user. The scheduler matches the conditions, and use different coding and modulation schemes. Con-
channel rates (frequency resources) with respect to users channel sequently, they may consume quite different amounts of the
quality and assigns the transmission slots within the frame (time system resources. Accordingly, the overhead required to accommo-
resources) according to the priority of each service ow. A sched- date the users signaling and control information may vary, resulting
uler design for multiple trafc classes in OFDMA networks is pro- in varying amount of available resources. Hence, characterizing a
posed in [8] that is well suited for WiMAX systems. In service delivery model turns into a complex problem.
mentioned study, the traditional MLWDF (Modied Largest The study done in [6] as a part of the Application Work Group of
Weighted Delay First) scheduling scheme is extended for multiple WiMAX Forum, tries to model a consistent input for application-le-
trafc classes, where the used scheduling metrics are; Head of line vel proles. The authors provide statistical proles that include
packets waiting time for restricted delay trafc and queue length trafc-mix ratio, data-session attempts for applications and diur-
information for BE trafc. In another study [15], the scheduling is nal-application trafc distribution. In our work, we use these data
viewed as a cross-layer scheme for AMC permutation mode. The to dene a simplied application prole for our capacity estima-
resources are allocated with respect to the MAC and PHY joint re- tion algorithm, although implementation of the detailed prole
sponses for different QoS requirements. Since the scheduling algo- will be straightforward. The aforesaid article also studies the sys-
rithms for WiMAX are not standardized, a considerable number of tem performance for VoIP and TCP applications, by presenting
other studies may be found in the literature that study these fea- the maximum number of users that the system can support for
tures. For detailed information, the reader is referred to [16] that each of those applications. However, they do not consider the co-
summarizes various scheduling strategies and provides perfor- existence of different applications. Furthermore, the presented re-
mance evaluation comparisons on different methods. However, sults are for the cases where the entire frame is utilizing a single
in this paper we do not simulate a complex smart scheduler and MCS level, which means the link adoption key feature is not pres-
instead we replace its complexity with a simple statistical trafc ent. The resource overheads are roughly estimated and the impact
model, introduced in Section 6.2 called application prole. of burst construction limitations in resource allocation is not
considered.
The authors in [18] discuss a conditional optimization problem
3. Previous work in order to minimize the required amount of system shared re-
sources, while the QoS requirements of all the scheduled users
The network development life cycle has several phases. From are satised. This problem can be seen as a generic system load
the technical point of view, the most important phases can be clas- model, in which, either the sum of timefrequency resources con-
sied into the following general stages: sumed by all the users sDL or the sum of transmission power val-
ues consumed by all the users qDL , is considered as the restricting
System denition and analysis. factor. In other words, the system will be fully loaded when one of
System design. these two factors reaches its maximum available value. The system
Implementation and testing. load is optimal when resource/power allocation is minimized with
respect to; the set of modulation and coding schemes (q), the set of
The attention of the available literature is mostly drawn on the sec- transmission power values (p) and the set of frequency subchan-
ond phase. However, in more complex networks, such as mobile nels (f) that are assigned to all the users in downlink subframe.
and wireless networks, the general phase of designing the system Eq. (1) depicts this relation, where sDL and qDL correspond to nor-
can be divided into further subclasses. Among them we can men- malized values (i.e. sum value divided by max value). However,
tion: (a) capacitating and dimensioning, (b) operational design, as the authors of this research mention, the solution to this prob-
and (c) performance enhancement. Our study in this paper is mostly lem is an exhaustive search and is inapplicable in practice.
focused on capacitating and dimensioning part and covers its exten-
sive concerns. In a wireless network such as Mobile WiMAX, by uDL minf fmaxsDL ; qDL g: 1
knowing how many users can be supported by each specic hot-
spot, the number of required hotspots to desirably cover the service An iterative solution for the above optimization problem is dis-
area can be calculated. cussed in [19] for WiMAX systems using frequency diversity frame
The previous studies in the literature are mostly concentrated structure (i.e. PUSC). In this reference, the authors present heuristic
on operational and enhancing phases of designing a WiMAX sys- algorithms in order to minimize the sum of allocated resources sDL
tem. As far as we know, none or few of these studies are concerned in a frame, while considering the resource and power constraints
with the objective of this paper. Therefore, this section reviews the and satisfying the QoS requirements of the scheduled service ows.
previous works that, although focusing on operational issues, play Note that due to frequency diversity, the subchannel constraint (f) is
a role in our capacitating and dimensioning study. eliminated. The iteration initiates by applying the highest MCS (i.e.
A simple capacity estimation method for WiMAX systems is highest q set) to all the services, which implies the optimal usage of
presented in [17], where authors assume OFDM-FDD frame struc- the available resources and accordingly the highest transmission
ture. The cell capacity results are presented as a function of the power. If the restricting factors are exceeded, the algorithm selects
number of users, where only two types of trafc services (real-time one service ow and reduces its MCS level, thus updating the q set
and FTP) are present. The authors conclude that the capacity is for the next iteration. The algorithm may iteratively discard one ser-
A.M. Ahmadzadeh et al. / Computer Communications 35 (2012) 109119 113
vice ow, in cases where the examined sets of q do not satisfy the minimum service demand are recalculated and compared. The
constraints. The presented solution imposes a high level of imple- increment continues as long as the remaining resources can satisfy
mentation complexities. Real-time simulation of the presented the additional demand imposed by the new user. When the de-
algorithm entitles massive data correlations that make it inappro- mand exceeds the available system resource, the algorithm intro-
priate for network planning and capacitating studies. The algorithm duces the largest N as the measure for system capacity.
needs to keep track of the time-varying consumed transmission
power of all users that, in practice, are randomly distributed inside 5. Dynamic system resource calculation
the coverage area. Apart from complexity, the presented algorithm
contains a number of impractical assumptions. The number of The system available resource is dened as the timefrequency
scheduled services must be known in advance and discarding the resources (slots) that are available for data transmission in a Mo-
service ows in the iterations, in which the constraints are not sat- bile WiMAX frame. As was explained in Section 2.2, the resource
ised, means increasing the dropping probability. The resource allo- characteristics of a Mobile WiMAX system depend on the BW
cation is performed over the entire downlink subframe and the and applied Permutation Mode. However, as was explained in Sec-
xed and variable resource overheads are not taken into account. tion 2.4, the allocation of these resources and thus the data-rate
The resource allocation in downlink subframe is assumed as con- that they provide depend on applied MCS distribution. In this sec-
tentious horizontal strips, which is against the burst construction tion, rst we propose a dynamic solution for the PHY+MAC over-
strategies dened by the standard. head removal problem with concentration in DL subframe, where
the remaining resources can be assigned for pure data transmis-
sion. In second subsection, the burst construction in DL-subframe
4. Overview of the proposed capacity estimation algorithm and thus the offered data-rate provision ability is investigated by
deriving a modulation and coding distribution assumption.
In this paper, the term capacity is concerned with the maximum
number of simultaneous multi-service users that are jointly sup-
5.1. PHY+MAC overhead removal
ported with each specic BS, and hence the corresponding cell
goodput at full-load. In order to calculate the capacity, we propose
The synchronization, signaling and control messages and the
an incremental solution based on the relation introduced in Eq. (1).
other transmission overheads, described in Section 2, are meant
The problem will be divided into two major processes that are to
to maintain the functionality of the WiMAX system and, obviously,
be solved in parallel. One process calculates the optimal available
do not carry any data. In this section we review the most signi-
resources, and the other the minimum resource consumption,
cant overheads in PHY layer along with presenting the detailed
while considering various service provision concerns. Note that,
overhead removal methodology in a dynamic manner. In order to
in a real eld, the absolute solution to the optimization problem
further improve the obtained PHY throughput, we propose novel
introduced in Eq. (1) is not achievable in a reasonable computa-
methods to extract the MAC layer overheads and estimate other
tional time. Therefore, with a detailed view over the technical fea-
wasted resources in dynamic allocation. Recalling Fig. 1, a DL sub-
tures and with a focus on planning and dimensioning purposes, our
frame starts with a one-symbol preamble, thus introducing a xed
proposed algorithm brings the solution to capacity estimation
overhead in DL. A 24 bit Frame Control Header (FCH) follows the
problem of WiMAX system into practice.
preamble that is sent with QPSK 12 and 4 repetitions, thus occupying
In general, as the number of users rises, proportional to the re-
2 rst subchannels in the rst data-symbol (24 4 bits). The DL-
quested application, the resource consumption increases and the
MAP message starts with 12 byte header and contains at least
available resource decreases. Therefore, In order to serve additional
one DL-MAP-IE for each burst that is to be transmitted in DL sub-
users a dynamic algorithm is necessary to perform an instanta-
frame. As indicated in Fig. 1, each IE contains 44 bits of burst infor-
neous comparison of the cells serving conditions. As illustrated
mation and may contain one or more 16 bit-CID based on the
in Fig. 2, the rst block (conned in the dashed rectangle at left
number of active connections in DL. The overall DL-MAP overhead
hand side) calculates the available cell resource at each given time.
can be written as Eq. (2):
The accurate calculation of useful resource requires the knowledge
of number of scheduled users per frame (NActive) as well as their re- OHDL
MAP 96 N Burst 44 N Activ e 16 bits; 2
source allocation pattern. In each step, the number of useful slots
in the DL-subframe is calculated using the dynamic PHY+MAC where NBurst is the number of data-bursts in DL subframe and NActive
overhead removal methodology proposed in Section 5. The optimal is the number of users that are actively communicating with BS at
allocation of the useful slots is derived from modulation and cod- given time instance.
ing distribution at the coverage area which is studied in Section As in UL each user usually transmits in a single burst, the num-
5.2. We assume that the mobile terminals are uniformly distrib- ber of bursts in UL is equal to the number of active users. Therefore,
uted inside the coverage area and the BS has enough transmission the imposed overhead by UL-MAP that starts with a 7 byte header
power to serve all of the distributed terminals. The offered data- and carries UL-MAP-IEs of 32 bits-length will be as in Eq. (3).
rate of the target cell is formulated by studying the burst construc- OHUL
MAP 56 N Activ e 32 bits: 3
tion strategies. For each number of users, the unknown parameters
to the rst process are calculated in the parallel process in the MAP messages are transmitted with most robust MCS level QPSK 12
second block. Furthermore, this block (right hand side dashed rect- to be received with all users located in the service area. For more
angle) is responsible for calculation of the aggregated minimum re- reliability, a repetition index is also applied over MAPs specied
quired data-rate for multi-service users. We elaborate a statistical in FCH. Assuming 4 repetition, once the MAP messages overhead
methodology to be replaced with the complexities of implementa- OHMAP OHDL UL
MAP OH MAP is calculated, it is enlarged to closet mul-
tion of a fair scheduling procedure. Section 6 introduces the con- tiple of 24 and then is multiplied by 4. This is due to the fact that,
sidered application prole and formulates the aggregated data- repetition operates on slots and not on bits.
rate demand as a function of the number of users and based on dif- The Channel Descriptor messages (i.e. DCD and UCD) are indi-
ferent application QoS requirements. cated with dashed lines in Fig. 1. It implies that these massages
As can be seen in Fig. 2, each time the number of subscribers (N) may not appear in all fames. Basically, Channel Descriptor mes-
increments by one, the actual system resource and subscribers sages are sent periodically or whenever the linking conditions
114 A.M. Ahmadzadeh et al. / Computer Communications 35 (2012) 109119
change and new channel prole must be introduced to MSs. In our method. The rst method, although very simple, needs advance
algorithm it is assumed that DCD/UCD messages appear every 100 knowledge of packet distribution which might not be practical
frames that result in 1.17 kbps of overhead for DCD and 1.71 kbps where the frame resources are exible. The second method imposes
for UCD, assuming 7 burst-proles. In addition to overhead intro- extra overhead in both MAC and PHY layers. A more efcient packet
duced with control messages, sometimes in PHY level the amount to burst mapping method is presented in [12] with respect to a min-
of data to be sent in a burst just spills over a slot. In these cases imal component of a DL-burst, which is dened as x slots in the
nearly an empty slot is sent, introducing additional channel over- frequency domain (x P 1) one slot in the time domain. A number
head. To consider this condition, the algorithm assumes a of minimal components are combined in time axis according to the
50% MAU mismatch error for each data burst. packet size and the units formed in this way may also be aggregated
From MAC layer point of view, as explained earlier in data allo- to form the burst as long as the same MCS mode is applied over
cation procedure, MAC-PDU construction presents a number of them. Therefore, one can expect to have one multi-user DL-burst
overhead data-bits per PDU. There are a 6-byte generic header made of the groups of slots that support each specic MCS level
and a 4-byte CRC per PDU. It is likely to have an additional 3-byte in DL transmission. This strategy is used in this paper and is further
overhead for either the packetization or fragmentation subheaders. studied in MCS distribution analysis in next subsection.
Therefore, each PDU presents 13 bytes overhead in MAC level as Eq. (5a) summarizes the overhead imposed on DL subframe by
shown in Eq. (4), where represents the number of PDUs in ith burst. both PHY and MAC layers. By removing the overhead symbols the
NX
useful resources are obtained that can be assigned for data trans-
Burst
OHMAC NPDU 104 bits: mission in DL. The overhead bits associated with UL subframe
PDU i 4
i1 are fairly simpler to calculate, as MAP messages are not present
in UL. Eq. (5b) denotes the overall UL overhead.
Generally, the number of PDUs per burst is variable and is a func-
tion of transmitted packets size and in-use burst construction OHDL preamble FCH OHMAP OHDCDUCD mismath
method. The standard supports multiple users burst in DL direction
OHMAC
PDU bits; 5a
as long as the users application delay requirements are met. The
limitations regarding to burst construction methods are: rectangu-
lar dimensioning and maintaining a single MCS mode during the OHUL anging OHACK=NACK CQICH mismath OHMAC
PDU bits:
entire burst. There are several proposals available in the literature 5b
for burst construction problem. Fixed burst mapping considers a
predetermined number of burst per frame, while Raster technique Having the overhead symbols removed, the useful data symbols
[11] assigns a burst to each data packet using a slot by slot scanning (Nsym) is calculated as in Eq. (6).
A.M. Ahmadzadeh et al. / Computer Communications 35 (2012) 109119 115
user experience, a key feature of each data service is its data-rate Table 4
requirement. In other words, some data services need the system Mobile WiMAX applications and quality of service [4].
to guarantee that at each given time a specic number of bits out Service ow category Applications QoS data rate
of the entire available resources will be allocated for their transmis- specication
sion. With this respect (i.e. the level of required guarantee for data- Unsolicated Grant Service VoIP Sustained Rate (CBR)
rate access), the services are classied into 3 categories; (a) Con- (UGS)
stant Bit Rate (CBR) with strict data-rate requirement (Minimum Real-time Polling Service Streaming Audio or Minimum Reserved Rate
(rtPS) Video
Reserved), (b) Variable Bit Rate (VBR) that is characterized with a Maximum Sustained
strict part (Minimum Reserved) and a upper bound data-rate (Max- Rate (VBR)
imum Sustained), (c) Best Effort Services (BE) that are served using Extended rtPS (ErtPS) Voice with Activity Minimum Reserved Rate
the remaining bandwidth when other services are supported. Based Detection
Maximum Sustained
on this data-rate requirement classication the total available
Rate (VBR)
bandwidth is divided into guaranteed and non-guaranteed parti- No Real Time Polling File Transfer Minimum Reserved Rate
tions. Fig. 3 indicates this bandwidth partitioning scheme and Table Service (nrtPS) Protocol (FTP)
4 shows the relation of each data-rate specication with its corre- Maximum Sustained
sponding application and service ow category. The guaranteed Rate (VBR,BE)
Best Effort Service (BE) Data Transfer Web Maximum Sustained
bandwidth requires unconditional dedication of the resources that
Surng Rate (BE)
grant the needed data-rate. Table 5 depicts the minimum guaran-
teed data-rate requirement for each application category intro-
duced by WiMAX Forum, when applicable. On the other hand, the
non-guaranteed partition must be fairly shared between requesting Table 5
Service delivery model.
services, based on the scheduling policy. The combination of the
minimum guaranteed and non-guaranteed resources allocated to Tab Application class Guaranteed Weight Ave. contention
the contenting users is calculated based on the subscribers applica- BW (Kbps) ratio [6]
[4]
tion prole, as will be explained in following subsection. In each Residential Business
step, a new user will be admitted if there are enough free resources 1 Multiplayer 50 0.25 0.25 0.5
to satisfy its minimum data-rate requirement. Interactive Gaiming
2 VoIP and Video 32 0.1 0.6 0.75
Conference
3 Streaming Media 64 0.125 0.75 1
6.2. Application prole
4 Web Browsing & Nominal 0.325 0.75 0.75
Instant Message
For all broadband wireless networks that support multi-ser- 5 Media Content BE 0.2 0.4 0.9
vices, including WiMAX, an application usage prole is required Downloading
as a consistent input for the purpose of evaluation of the dimen-
sioning and capacitating studies. The application prole shall cover
subscribers application usage trends, where the two major metrics
tention ratio. In other words, if we assume N is the number of the
to be considered are the applications penetration and contention
subscribers that fall within the connes of the target cell, then
ratio. The penetration parameter represents the distribution of
N W app
j represents the number of subscribers that are likely to
applications that most likely users will use, while the contention ra-
use the jth application, and out of them, only N W app
j CRj sub-
tio is a measure of the simultaneity of the usage of each application.
scribers will be connected to the cell at each time instance, where
The collective application distribution information of broadband
CRj is the contention ratio of the jth application.
mobile/wireless service subscribers are mapped into the WiMAX
In order to give reference values for the contention ratios of dif-
application classication model, where we assigned a weight to
ferent applications, the subscribers are classied based on their
each class representing the average penetration of each application
end-user type. With this respect, the users may subscribe for Res-
among the subscribers. The third column of Table 5 illustrates the
idential or Business class services, each with different service pro-
weight W app
j corresponding to jth application category, when le. Statistical studies prove that the attempts per application per
averaged over different applications grouped into a category. In
day are different for different end users [6]. According to global sta-
addition to distribution weights, the application prole should pro-
tistics of WiMAX deployments in 2007 [10], NR = 58% of the users
vide the information on relative active attempts of each application
were in residential prole, while NB = 42% were subscribed for
in each time instance that in this paper we call it applications con-
business class services (NR + NB = 100%). The detailed contention
ratios for residential and business class subscribers are presented
in [6] for each application per day. These ratios are averaged over
the daily trafc trends and are shown in the corresponding column
of Table 5 for each application category. Recalling the example
in previous paragraph, the number of residential subscribers
that are actively using the ith application is estimated as:
N N R W app
j CRRj , where CRRj is the average contention ratio
for jth application in residential class. Following the same pattern,
a good approximation for the total number of active users (NActive)
is obtained. Eq. (9) summarizes the relation between this key
parameter and the parameters specied by the application prole.
&
5
X 5
X
R B
NActiv e N N NR W app
j CRj N B W app
j CRj ;
j1 j1
9
Fig. 3. Bandwidth partitioning.
A.M. Ahmadzadeh et al. / Computer Communications 35 (2012) 109119 117
where NR and NB are the percentage of the end users with Residen- users is calculated independent of available resources XDL Tot N, that
tial and Business prole respectively, with contention ratios of CRRj was calculated in Eq. (9). For each number of users these two val-
and CRBj correspondingly (j = 1, . . . , 5 represents the application cat- ues are compared and if XDL DL
Tot N P RTot N, N will be incremented
egory). The obtained NActive parameter here is also used in the par- by one. The largest N that satises the above condition is intro-
allel process for overhead estimation calculations. duced as the system capacity in supporting multi-services.
Note that the values presented in Table 5 are for general refer- Recalling the parallel process for overhead calculations, apart
ence and can be easily customized for each trafc scenario. In other from NActive parameter, an estimation for the number of PDUs per
words, the average number of attempts per active user during a burst N PDU was also required as in Eq. (4). Having the load per
i
day can be converted into the number of application sessions dur- frame and the number of active users, a reasonable approximation
ing a specic time of the day with diurnal-application distribution for the average-packet-size (PAve) is calculated for each number of
[6]. For instance, substituting the peak hour values instead of diur- users. Assuming k averaged-size-packets per PDU i:e: ARQ
nal averaged ones may alter the application prole for the worst-
BLOCK SIZE k T f XDLTot =N Activ e , each burst will accommo-
case load prole.
date a number PDUs proportional to its MCS distribution weight.
This relation is shown in Eq. (12).
6.3. Subscribers service demand calculation
&
W MCS
Considering the Service delivery model presented in Table 5, the NPDU
i NActiv e i
: 12
k
data-rate requirement for the instantaneous load imposed on the
system is achievable. We assume that all the active users are
Note that the calculated average packet size is not a practical con-
scheduled for transmission in each DL subframe. The application
cept, since the packets generated by the users considerably vary
classes with specied guaranteed data-rate (the rst 3 application
in size for different applications and also over time. However, this
classes in Table 5) are separated from those with non-guaranteed
concept gives a good approximation for estimation of the number
data-rate. The number of subscribers in the rst group N GActiv e is ob-
of PDUs per data burst by introducing a trade-off between different
tained by solving the Eq. (9) for the rst 3 applications. The aggre-
packet sizes.
gated guaranteed data-rate (RG) is then calculated using Eq. (10).
X
3
RG NGActiv e DRj ; 10 7. Experiments and numerical results
j1
In this section, we demonstrate the simulation results of the
where proposed capacity estimation methodology. In rst experiment,
& we study the detailed behavior of the presented algorithm. The
3
X 3
X
NGActiv e N NR W app
j CRRj NB app B
W j CRj ; two input parameter categories used in Experiment-I are specied
j1 j1 in Table 6. The system parameters are the default values recom-
mended by WiMAX Forum system evaluation methodology and
and DRj is the guaranteed data-rate of the jth application as in Table also are the common values used in practice. The subscription ser-
5 (j = 1, . . . , 3). vice parameters are based on the statistical data, as explained in
At each given time, the remaining system bandwidth can be the presented application prole. The numerical simulation results
allocated to non-guaranteed services with respect to users ser- provided in Table 6 for Experiment-I are concerned with following
vice-rate parameter as determined by the subscription contract. conditions; (a) All the active services, including guaranteed and
Each subscriber is due to have access to a non-guaranteed data- non-guaranteed data-rate services, are scheduled in the frame of
rate up to the subscribed service-rate, as long as the existing higher the sector under study at each given time. (b) Non-guaranteed ser-
priority services for that user (here the guaranteed data-rate ser- vices are assigned a data-rate, equal to their service-rate subscrip-
vices) are already served. The service-rate is an application prole tion (refer to Section 6.3). The rst assumption assures that the
input parameter and may be different for different end-user types. delay requirements of all the services are met and the second
Let us assume that each residential subscriber may contract one provides the highest user satisfaction with the price of impos-
CR kbps of service, while the business class subscribers demand
is CB kbps. Therefore, we can introduce a measure for the maxi-
mum amount of aggregated data-rate for non-guaranteed service Table 6
RMAX Input parameters and numerical results of Experiment-I.
Non;G as conducted in Eq. (11).
" " # " ## System Parameters
X
3 X
3
Bandwidth (MHz) 10
RMAX Non;G
Non;G N Activ e N R C R DRj NB C B DRj ; DL:UL Ratio 2:1
j1 j1
Frequency Reuse Pattern 1, 3, 3
11 MCS Distribution Table 2
Subscriber Service Parameters
where NNon;G
N Activ e
Activ e NGActiv e
is the number of active subscribers % NR 58%
using applications with non-guaranteed data-rate, NR and NB are % NB 42%
the percentage of the end users with Residential and Business pro- CR (kbps) 512
le respectively. CB (kbps) 1000
Load Model and Application Prole Table 4
The total aggregated data-rate at each time instance is simply
calculated by the summation of guaranteed and non-guaranteed Simulation Results
Max No. of supported multi-service users 24
data-rates (i.e. RDL DL
Tot RG RNon;G ). Note that RTot N is a function of NActive 16
the number of users and linearly increases as the number of users Max goodput (Mbps at N = 1) 8.51
increments. Given the application prole, RDL Tot N provides the re- Min No. of overhead slots (at N = 1) 79
source consumption metric for each number of supported users. Sector Throughput (Mbps at N = 24) 5.62
Max No. of overhead slots (at N = 24) 279
Following the generic algorithm proposed in this paper, at each gi-
Average-Packet-Size (bytes) 211
ven time, the resource consumption RDL Tot N for each number of
118 A.M. Ahmadzadeh et al. / Computer Communications 35 (2012) 109119
ing heavy load on the system. Fig. 4 illustrates the dynamic trend of wasted resources are caused by burst construction limitations ex-
the algorithm for this experiment. As can be noticed, the resource plained in the text. A considerable number of realizations show
consumption exceeds the available resource for the 25th user. In that an average of 3 slots remain unallocated in multi-burst map-
other words, two graphs representing the linear minimum ser- ping for mixed user trafc scenarios, while having a fair scheduling
vice-load and the DL optimal goodput, cross before meeting the assumption.
data-rate requirement of 25th user. Thus, the specied sector can In Experiment-II, we compare two different service provision
support 24 multi-service users with the sector throughput of scenarios for non-guaranteed services, while the resource con-
5.62 Mbps. By providing reference measures of magnitude, these sumption in both scenarios are the same. In this experiment, the
results can help network planners to study the dimensioning capa- input parameters for both system and subscription parameters
bilities of the implemented system. are the same as Experiment-I. We model Case-1, where the non-
Fig. 5(a) shows the dynamic PHY+MAC overhead increase from guaranteed services are using 50% of their maximum service-rate,
above 3 symbols with single user up to 11 symbols in full-load. The and compare it with Case-2 when 50% of the non-guaranteed ser-
un-integer values of symbols denote the overhead slots (i.e. multi- vices are not scheduled in target subframe, while the existing ones
ples of 24 bits in DL). The signicant difference in number of over- are using the maximum data-rate. In this way, the sectors load
head symbols in Fig. 5(a), underlines the need to consider a (subframe resource consumption) for both cases will be equal,
dynamic strategy for overhead determination with respect to the but the sectors throughput and, hence, the number of supported
number of scheduled users in Mobile WiMAX capacity estimation users will be different. The simulation results show that the rst
analysis. The number of unallocated slots during the resource allo- scenario will support 32 multi-service users, while in the second
cation procedure in Experiment-I is illustrated in Fig. 5(b). These this value increases up to 38 users.
This is due to the fact that by discarding 50% of the users with
non-guaranteed data-rate, the overhead related to their signaling
and controlling messages are also removed from available down-
link subframe resources. Therefore, when the number of supported
users is 32, the level of available resources of Case-2 is higher than
Case-1. As can be seen in Fig. 6, the freed resources are enough to
accommodate 38 32 = 6 extra multi-service users, with specied
application prole. Note that the extra users also introduce addi-
tional overhead to the system. This can be noticed by continuously
declining slope of the available resources of Case-2. But the simu-
lation results prove that the released available resources in Case-2
can support a greater number of users up to the consumption of
5457 kbps. The specied system can support up to 106 users, when
the entire available resources are assigned to the services with
guaranteed data-rate. By providing reference measures of magni-
tude, these results can help the algorithm developers to evaluate
and compare the service provision capabilities of different schedul-
ing techniques.
The proposed algorithm is highly change sensitive and provides
a simple environment to study the impact of each change (i.e. sys-
tem or load input parameters) on the system performance. In con-
Fig. 4. Simulation results of Experiment-I.
tinuation, we repeat the simulations for different system
parameters, while subscriber service parameters are the same as
Table 6 and MCS distribution is as in Table 3. The results are pre-
sented in Table 7 for two different scheduling case, that are, when
Table 7
Estimated capacity for different system parameters.
BW 5 MHz 10 MHz
FRF 1, 1, 3 1, 3, 3 1, 1, 3 1, 3, 3
DL:UL Scheduling Full Guaranteed Full Guaranteed Full Guaranteed Full Guaranteed
1:1 3 15 8 38 8 45 21 91
2:1 4 19 9 41 9 49 24 106
3:1 5 26 13 57 12 64 30 133
the subframe is loaded with all services as in Experiment-I, and the and mobility overheads on the Mobile WiMAX capacity estimation
other when only guaranteed data-rate services are present and are other topics that will be discussed in our related future works.
non-guaranteed services are not scheduled in the target subframe.
It can be observed that the sector capacity can vary between 3 and Acknowledgement
133 multi-users with different system parameters and subframe
scheduling pattern. This work has been partially supported by the Research Project
of Ministerio de Educacion of Spain TEC2006-07010.
[1] IEEE Standard for Local and metropolitan area networks Part 16: Air Interface
In this document we proposed a dynamic capacity estimation for Fixed Broadband Wireless Access Systems IEEE Std 802.16-2004, Oct.
methodology for Mobile WiMAX systems, while the sector under 2004.
study is simultaneously supporting multi-service users. Our algo- [2] DRAFT Standard for Local and Metropolitan Area Networks, Part 16: Air
Interface for Broadband Wireless Access Systems IEEE P802.16Rev2/D9a, Mar.
rithm contains two major blocks that are solved at each given time 2009.
and in parallel in order to, on one hand, calculate the optimal avail- [3] Mobile System Prole Specication, Release 1.5 Common Part WiMAX
able resource with respect to MCS distribution and multi-user Forum Specication, Aug. 2009.
[4] Mobile WiMAX Part I: A Technical Overview and Performance Evaluation
burst construction and, from the other hand, calculate the mini-
WiMAX Forum, Aug. 2006.
mum required resource consumption with respect to application [5] Y. Xiao, WiMAX/MobileFi: Advanced Research and Technology, Auerbach
prole and the multi-service data-rate requirements. The general Publications, Taylor & Francis Group, 2008.
formulations for overhead estimation are provided for any desired [6] B. Kim, J. Yun, Y. Hur, C. So-In, R. Jain, A. Al Tamimi, Capacity estimation and
TCP performance enhancement over mobile WiMAX networks, IEEE
system parameters. The data-rate requirements of the covered Communications Magazine 47 (6) (2009) 132141.
multi-services are grouped into guaranteed and non-guaranteed [7] J.G. Andrews, A. Ghosh, R. Muhamed, Fundamentals of WiMAX: Understanding
data-rates and an application prole is dened as a consistent in- Broadband Wireless Networking, Prentice Hall, 2007.
[8] W-H. Park, S. Cho, S. Bahk, Scheduler design for multiple trafc classes in
put for capacitating and dimensioning studies evaluation. Simula- OFDMA networks, Computer Communications 31 (1) (2008) 174184.
tions are run with example values for MCS distribution and daily [9] R. Jain, C. So-In, A. Al Tamimi, System-level modeling of IEEE 802.16e Mobile
averaged application prole. The simulation results with regard WiMAX networks: key issues, IEEE Wireless Communications Magazine 15 (1)
(2008) 7379.
to PHY+MAC overhead removal methodology underline the signif- [10] A. Fellah, WiMAX Market Trends and Deployments WiMAXCounts
icant roll of overhead calculations in capacity estimation studies. Newsletter Issue. 1 May 2007. Available at: <http://www.wimaxcounts.com/>.
The elaborated statistical application prole replaces the complex- [11] Y. Ben-Shimol, I. Kitroser, Y. Dinitz, Two-dimensional mapping for wireless
OFDMA systems, IEEE Transactions on Broadcasting 52 (3) (2006) 388396.
ity of simulating a detailed scheduling strategy, thus providing ref- [12] T. Ohseki, M. Morita, T. Inoue, Burst Construction and Packet Mapping
erence measures in order to evaluate the performance of different Scheme for OFDMA Downlinks in IEEE 802.16 systems, in: Proceedings of the
scheduling techniques in real time. Having the number of sup- IEEE Globalcom, 2007, pp. 43074311.
[13] Physical Layer Aspects for Evolved UTRA 3GPP Std 25.814 V1.2.3, May 2006.
ported user with each WiMAX hotspot, with regards to the param-
[14] J. Chen, W. Jiao, H. Wang, A Service Flow Management Strategy for IEEE
eters that characterize the access point and the service area, the 802.16 Broadband Wireless Access Systems in TDD Mode, in: Proceedings of
network planners can optimize the service provision in the cover- the IEEE International Conference on Communications, vol. 5, 2005, pp. 3422
age area. Furthermore, the proposed algorithm is highly change 3426.
[15] T. Ali-Yahiya, A.L. Beylot, G. Pujolle, An adaptive cross-layer design for
sensitive. In other words, the algorithm can be simply congured multiservice scheduling in OFDMA based mobile WiMAX systems, Computer
for each set of input parameters, in order to study the impact of Communications 32 (3) (2009) 531539.
each parameter change on system performance. The performance [16] Y.A. Sekercioglu, M. Ivanovich, A. Yegin, A survey of MAC based QoS
implementations for WiMAX networks, Computer Networks 53 (13) (2009)
is evaluated in a single output result, that is, the number of sup- 25172536.
ported multi-users (or the cell goodput). [17] L. Nuaymi, Z. Noun, Simple Capacity Estimation in WiMAX/802.16 Systems,
Further studies shall be done in order to improve the proposed in: Proceedings of the IEEE 17th International Symposium on Digital Object
Identier, 2006, pp. 15.
algorithm with more implementation details, while keeping the [18] S.N. Moiseev, M.S. Kondakov, Fast system load estimation in the IEEE 802.16
simple structure of the proposed approach for capacity estimation. OFDMA network, Computer Communications 32 (2009) 12281232.
In continuation to this work, we plan to propose solutions in order [19] S.A. Filin, S.N. Moiseev, M.S. Kondakov, A.V. Garmonov, D.H. Yim, J. Lee,
S. Chang, Y.S. Park, QoS-Guaranteed Cross-Layer Adaptive Transmission
to integrate the performance enhancing techniques in proposed Algorithms for the IEEE 802.16 OFDMA Systems in: Proceedings of the IEEE
algorithm and to include the short-scale channel state variations Vehicular Technology Conference, Montreal, Canada, 2006, pp. 15.
factor in resource allocation. Furthermore, the impacts of handover