We are IntechOpen,

the world’s leading publisher of

Open Access books
Built by scientists, for scientists

6,300
Open access books available
171,000
International authors and editors
190M Downloads

Our authors are among the

154
Countries delivered to
TOP 1%
most cited scientists
12.2%
Contributors from top 500 universities

Selection of our books indexed in the Book Citation Index

in Web of Science™ Core Collection (BKCI)

Interested in publishing with us?

Contact book.department@intechopen.com
Numbers displayed above are based on latest data collected.
For more information visit www.intechopen.com
Performance Evaluation Methods to Study
IEEE 802.11 Broadband Wireless Networks under the PCF Mode 509

24
0

Performance Evaluation Methods to Study

IEEE 802.11 Broadband Wireless Networks
under the PCF Mode
Vladimir Vishnevsky and Olga Semenova
ZAO Research & Development Company "Information and Networking Technologies"
Russia

1. Introduction
In the Chapter, we consider design concepts and protocols for metropolitan area wireless net-
works, realization methods for adaptive dynamic polling in these networks and investigation
of their main performance characteristics by means of stochastic polling models.
Polling mechanism is widely used in the wireless metropolitan area networks (WMANs). In
the wireless networks with PCF (Point Coordination Function), a base station polls subscriber
stations accordingly to a polling table describing the order of polling. For IEEE 802.11 Wi-Fi
networks, the polling is an option; for WiMAX networks (IEEE 802.16), it is basic. Using the
PCF in the MANs allows to avoid the problem of hidden stations, efficiently schedule an order
of station access to the wireless channel, flexibly control the radio cell operation and change
its parameters correspondingly to the current situation by adjusting only the base station.
The methods to form and keep up the polling table are not specified in the standard thus the
wireless network developers can freely decide on how to realize it. The specific polling mech-
anism and its parameters are the main factors determining the efficiency of the broadband
wireless MAN with centralized control. In the Section, we give the description of the IEEE
802.11 protocols and the main directions of their development, including the recent versions
IEEE 802.11n and IEEE 802.11 VHT. The much attention is given to development and mod-
elling of the algorithms to poll subscriber stations, the schemes of adaptive dynamic polling
(ADP). The adaptive dynamic polling is proposed to cut down the expenses of polling the
empty subscriber stations and stations that stopped working for some reason. The adaptive
dynamic polling is the prospective direction of developing the IEEE 802.11 broadband wireless
networks.
We present the models of adaptive polling system and the system with threshold polling. With
the adaptive polling order, the order of queue visit is cyclic but a server does not visit queues
that were empty at the instant of polling in the previous cycle. Under the threshold polling,
a queue is served only of its length exceeds the given threshold. Such service discipline is a
possible way to assign a priority to a queue depending on the threshold value, and it allows
server to give more attention to queues with high traffic intensity rather than spend time in
queues with low traffic.

www.intechopen.com
510 Trends in Telecommunications Technologies

2. Performance evaluation of the broadband wireless networks

Polling mechanism is widely used in the metropolitan area wireless transmission networks. In
the wireless networks with PCF (Point Coordination Function), a base station polls subscriber
stations accordingly to a polling table describing the order of polling. For IEEE 802.11 Wi-Fi
networks, the polling is an option; for WiMAX networks (IEEE 802.16), it is basic. In this
Section, we consider design concepts and protocols for metropolitan area wireless networks,
realization methods for adaptive dynamic polling in these networks.

2.1 Development of the broadband wireless networks: state of the art and prospects
In the recent years, the wireless transmission networks become the main direction of the
network industry development. It was provided by both the rapid Internet development
and the adoption of new progressive methods for coding, modulation and wireless data
transmission. Recently, it is obvious that broadband wireless networks are without a rival
with their efficiency of deployment, portability, price and area of potential applications.
Wireless technologies displace the wired one almost in all places where they can provide
high-quality data transmission. The tendency is evidently continuing to the future since the
wireless world is more comfortable. Nowadays, the wireless data transmission technologies
have become ingrained in everyday life of millions of people and enterprises. The modern
wireless networks allows solving variety of problems from the indoor network management
to the distributed wireless networks within a city, a region or a country. Low cost, efficiency of
deployment, wide performance capabilities to transmit data, IP telephony and video streams,
all these make the wireless technologies the most rapidly developing telecommunication area.
Rapid growth of the broadband wireless networks often called the "wireless revolution" in the
field of data transmission networks is explained by a number of their own distinctive features,
such as
– flexibility of a network topology enabling the dynamic change of the topology without
time loss when mobile users connect to the network, move or disconnect;
– high data transmission rate (up to 54 Mbit/sec);
– rapidity of designing and realization which is significant because of the strict technical
conditions to network construction;
– high unauthorized access protection level;
– high-priced laying or rent of the fiber optic or copper cable are not needed.
Recently, the wireless technologies provide effective solution of the following problems:
– mobile access to the Internet;
– organization of the wireless radiocommunication between workstations of a local area
network (organization of the wireless access to a local area network resources);
– unification of local area networks and workstations into a single data transmission
network and providing the remote access to the Internet for local area networks;
– last mile problem solution;
– interconnecting the automatic telephone systems with wireless channels of up to 54
Mbit/sec rate;
– creation of the land cellular radio modem data networks.

www.intechopen.com
Performance Evaluation Methods to Study
IEEE 802.11 Broadband Wireless Networks under the PCF Mode 511

The mentioned features of the wireless technologies are substantially brought about by the fact
that wireless networks operating within the range 2.4-6.4 GHz are based on the technology of
broadband, or noise-type signal. The technology was initially used for the military purposes
and recently it is efficiently used for civil radio networks.
The broadband wireless technologies use two radically different methods of frequency band
utilization, they are Direct Sequence Spread Spectrum (DSSS) and Frequency Hopping Spread
Spectrum (FHSS). Both methods imply the frequency band division into n subchannels. Under
DSSS method, each data bit is coded as a sequence of n bits, and all those n bits are transmitted
simultaneously through all n subchannels, and the coding algorithm is individual for each pair
’transmitter-receiver’ so as to provide the transmission security. Under the FHSS method, a
station transmits only through one of n subchannels at each time moment periodically changing
the subchannel. Those change-overs (hops) happen simultaneously for both a transmitter and
a receiver, and their sequence is pseudo-random and is known only for ’transmitter-receiver’
pair that provide the transmission security as well.
Undoubtedly, each method has its own advantages and disadvantages. The DSSS method
allows reaching the maximal throughput and due to the n-modular redundancy it, first, pro-
vides the narrow-band interference immunity, and, second, gives the opportunity to use the
low power signal so as not to interfere with ordinary radio devises. On the other hand, the
FHSS equipment is considerably simpler and cheaper and has the broadband interference
immunity.
To work with the wireless networks, we need the special MAC (Media Access Control) proto-
cols due to the fundamental differences from the cable medium, namely the lack of complete
connection (the stations can be hidden from each other), the wireless medium is not protected
from the outside signals and its signal propagation properties are asymmetric and variable by
times. In order to provide the effective wireless medium access, the international standards,
protocols and recommendations are developed which specify the physical and MAC layers of
wireless networks: Bluetooth, ETSI Hiperlan and IEEE 802.11 for local area networks (LAN);
IEEE 802.11 using the necessary amplifiers and parabolic antennae for metropolitan area net-
works (MAN), and finally, IEEE 802.16 and callular telephony technologies modified for data
and video images transmission (GPRS, UMTS and CDMA-2000) for MANs (Vishnevsky et al.,
2009).
Among the LAN and MAN developers, the IEEE 802.11 protocol is very popular (referred to as
Radio-Ethernet as well) adopted as an international standard in 1997 and having the following
features:
• it can be used in both LANs and MANs;
• both DSSS and FHSS methods of the broadband wireless network deployment are
regulated;
• the huge number of software and hardware of the large companies (such as CISCO
Aironet, Lucent Technologies, Alvarion, etc.) in the world markets support the standard.
The IEEE 802.11 protocol determines two network development topology, they are topologies
with infrastructure and so called ad-hoc-topology. With infrastructure topology, a wireless
network has a single access point (or base station). An access point provides the synchroniza-
tion and coordination for stations within the range, transmits broadcast packets and, what is
of the great importance, can be a portal into the global network. Such topology is referred to
as the Basic Service Set (BSS). To cover the wide area, it is possible to set up several access
points working in different frequency channels and connected to the joint wired or wireless

www.intechopen.com
512 Trends in Telecommunications Technologies

backbone. Besides, the subscriber stations can be provided with a roaming between the access
points. Such topology is called the Extended Service Set (ESS). The ad-hoc topology called
the Independent Basic Service Set (IBSS) is the network performance scheme under which the
numerous stations are connected directly avoiding connection to the special access point. This
regime is effective when the wireless network infrastructure is not constructed (e.g., conference
hall), or can not be constructed by some reason.
The IEEE 802.11 protocol specifies the data transmission rate equal to 1 or 3 Mbit/sec, with this
the packet header and the service information can be transmitted at 1 Mbit/sec. Note that the
transmission rate did not satisfy the users even when the protocol was adopted and approved.
In order to make the wireless technology popular, cheap and, above all, to satisfy the modern
strict conditions of business applications, the developers had to set up new standards which
were the extensions of IEEE 802.11. Consider them in brief.
IEEE 802.1a. The IEEE 802.11a protocol exploits the radio frequency band of 5 GHz (5150-5250
MHz, 5250-5350 MHz and 5725-5850 MHz). In contrast to IEEE 802.11, the protocol applies
not the spectrum broadening technologies but Orthogonal Frequency Division Multiplexing
(OFDM), also referred to as multiple carrier modulation, which uses several carrier signals of
different frequencies each transmitting a number of bites. This technology allows reaching the
following data transmission rates: 6, 9, 12, 18, 24, 36, 48 and 54 Mbit/sec.
IEEE 802.11b. The IEEE 802.11b protocol involves the changes within the IEEE 802.11 physical
layer. The network operates in 2.4 GHz radio frequency band. But the other signal modulation
technology, Complementary Code Keying (CCK) allows reaching the rates 5.5 and 11 Mbit/sec
and increases the connection stability in interference and multipath signal propagation condi-
tions.
IEEE 802.11g. The IEEE 802.11g protocol as long as IEEE 802.11b operates in 2,4 GHz radio
frequency band but applies the orthogonal frequency division multiplexing (OFDM) allowing
to reach the data transmission rate equivalent to IEEE 802.11a (up to 54 Mbit/sec). Nevertheless,
the protocol enables stations to get back to rates 1, 2, 5.5 and 11 Mbit/sec, i.e to CCK modulation.
Therefore, the devices 802.11b and 802.11g are compatible within a segment of broadband
wireless network.
Recently, the standard IEEE 802.11n describing the networks with data transmission rate
100Mbit/sec on the base of antenna system technology MIMO is going to be finished. The
mobile version of the standard (IEEE 802.11p) and the addition IEEE 802.11e to provide the
guaranteed quality of service (QoS).
In 2007, the generalized standard (see IEEE Std. 802.11-2007) was approved involving all stan-
dards finished before June, 2007. They are IEEE 802.11a/b/g mentioned above and additions
IEEE802.11e/h/i/j.
The standard IEEE 802.11 is constantly being improved and developed to provide new cus-
tomer services and to increase the data transmission rate and its quality. In 2009, it is planned
to release a number of new standards being developed from 2003-2004. First of all, they are
IEEE 802.11n and IEEE 802.11s. Though those standards are being finished, many companies
have started production of devices and provide the wireless network operation based on the
draft versions of those standards.
The other standards to be approved in 2009 are:
• standard IEEE 802.11u describing communications between IEEE 802.11 networks and
outer networks;
• standard IEEE 802.11r regulating procedures of switching between subscriber stations
for delay sensible applications like IP-telephony, etc.;

www.intechopen.com
Performance Evaluation Methods to Study
IEEE 802.11 Broadband Wireless Networks under the PCF Mode 513

• standard IEEE 802.11p for operation in dynamic environment, and for fast moving
wireless devices, in particular;
• standard IEEE 802.11v describing the wireless network control protocols;
• standard IEEE 802.11w regulating the methods to protect supervisory frames in a wire-
less network;
• standard IEEE802.11z describing the protocol for direct data exchanging between sta-
tions without using an access point.
The standard IEEE 802.11k should be also mentioned but it is not included into the generalized
standard IEEE 802.11-2007 since its final version was released at the end of 2007 only. The stan-
dard regulates the mechanisms to exchange information about radio resource, radiochannel
performance and load, noise level, etc.
The persistent growth of the transmission data volume, release of new applications, e.g. high
definition video, impose heavy demands on wireless network throughput.
In spite of high data transmission rate in third generation mobile networks based on LTE
technology and in networks IEEE 802.11n (up to 300 Mbit/sec), work on new technologies
creation in the framework of IEEE 802.11 is continued. From 2007, the standard IEEE 802.11
VNT (Very High Troughput) has been started providing a base for very high throughput local
wireless networks with nominal speed up to 500Mbit/sec within the frequency range 6 GHz.
The standard is planned to be finished in 2012.
High network throughput is obtained by the MIMO technology application with 8 spaced
antennas on both transmitting and receiving sides, by band enhancement up to 80 MHz
through multiplexing of four channels of width 20 MGz, also by using OFDMA to organize
frequency division multiple access as in IEEE 802.16. The developed standard supports
compatibility with devices working under IEEE 802.11a/b/g/n.
In Russian Federation, the new technology and both hardware and software for very high
throughput mesh-networks operating in the frequency range 60GHz (Vishnevsky & Frolov,
2009). As compared to existing mesh-networks, the proposed approach provides transmission
rate up to 1000 Mbit/sec and makes the frequency planning and operating in duplex mode
unnecessary.
In the 802.11 protocol, the fundamental mechanism to access the medium is called Distributed
Coordination Function (DCF). This is a random access scheme, based on the Carrier Sense
Multiple Access with Collision Avoidance (CSMA/CA) protocol. Retransmission of collided
packets for each station is managed according to binary exponential backoff rules (Section 2.2).
The alternative access mechanism as an option specified in IEEE 802.11 is point coordination
function (PCF) under which the coordinator station manages the centralized polling of other
stations (Section 2.3).

2.2 Medium access layer in IEEE 802.11. Distributed Coordination Function (DCF)
The IEEE 802.11 protocol is the part of IEEE 802 protocols for local area networks (LANs)and
metropolitan area networks (MANs) involving the well-known protocols IEEE 802.3 (Ethernet
LAN) and IEEE 802.5 (Token Ring LAN). The majority of IEEE 802 protocols determines the
physical and data link layers of the Open Systems Interconnection (OSI) seven-level reference
model of the International Organization for Standardization (ISO). Furthermore, the data
link layer is represented as two sub-layers, the Logical Link Control (LLC) and Medium
Access Control (MAC). Such division is conditioned by the fact that under the same LLC,
the mechanisms providing MAC can be different. The IEEE 802.11 involves the functional

www.intechopen.com
514 Trends in Telecommunications Technologies

description of both MAC and PHY layers. Both layers possess significant features, e.g. the
high packet loss rate due to noise and collisions, and the fact that wireless data transmission
can suffer from the unauthorized access. Logical link control is not considered in the protocol
since it is the same as in IEEE 802.2.
All questions on regulation of the wireless medium sharing by the network stations are de-
termined on MAC-layer. The necessity of such regulation rules is quite obvious. Imagine
the situation when each station of the wireless network sends data to the medium without
observing any rules. As a result of such signals interference, the destination stations can
not receive the data, and even understand the data were destined for them. Therefore, the
stringent regulating rules are essential to determine the wireless medium multiple access. The
multiple access rules can be compared to the rules of the road which regulate the road sharing
by road users.
As it is mentioned above, there are four types of IEEE 802.11 MAC layer multiple access to
wireless medium, they are Distributed Coordination Function (DCF), its extension Extended
DCF (EDCF), Point Coordination Function (PCF) and Hybrid Coordination Function (HCF).
Below, we consider these mechanisms in details.
The DCF is a method to organize peer-to-peer access to the wireless medium. The function
is based on the Carrier Sense Multiple Access/Collision Avoidance (CSMA/CA). With this
access, each station before sending a packet listens to the medium trying to detect the carrier
signal and starts transmission only when the channel is idle. But in this case the probability
that a packet collides with another one upon its transmission is high enough, when two or
more stations find out the channel is idle and start transmission at the moment when some
station is transmitting. In order to decrease the collision probability, the Collision Avoidance
(CA) mechanism is applied. The mechanism is described as follows. A station detecting the
channel idle waits for the prespecified time interval before it starts transmission. The time
interval is random given by two intervals: the Distributed InterFrame Space (DIFS) and the
random Backoff time. Consequently, each station waits for a random time before starting data
transmission that essentially decreases the collision probability since the probability that at
least two stations have the same backoff is negligible.
MAC-layer of the IEEE 802.11 protocol, its modification described in IEEE 802.11e thereof,
specifies five types of a time interval between consecutive data transmissions, namely Inter-
frame Space (IFS). The shortest one is SIFS (Short Interframe Space) used for special sequence
of data exchange, e.g. ACK transmission to acknowledge a frame successful reception. The
SIFS duration is specified by physical layer to give enough time for the system to switch from
reception to transmission or inversely. The other intervals are given in duration increasing
order: PIFS, used by stations with Point Coordination Function; DIFS, used by stations with
Distributed Coordination Function; AIFS, used in Extended DCF (EDCF), and EIFS (Extended
Interframe Space), used by stations after transmission error.
In order to provide stations with the equal access to the channel, it is necessary to determine
the appropriate algorithm to choose the backoff time which is a number of basic time intervals
called time-slots. To choose the random backoff, each station determines the Contention
Window (CW) which is the range the backoff time is chosen from. The minimal CW is 31 time
slots, and maximal one is 1023 time slots. The backoff is determined as:

Backo f f = Random(CW ) · SlotTime,

www.intechopen.com
Performance Evaluation Methods to Study
IEEE 802.11 Broadband Wireless Networks under the PCF Mode 515

where Random(CW ) is an integer uniformly chosen in the range (0, CW − 1). At each time, the
value CW depends on the number n of the attempts failed to send a packet and is given by

CW = CW0 2n ,

where CW0 is the minimal contention window, 0 ≤ n ≤ m, and m is the maximal number of
attempts allowed to send a packet. If the m + 1-th attempt is failed the packet is discarded.
When a station tries to get an access to the channel, after the DIFS expires the backoff starts to
count down. If the channel is idle during DIFS and the backoff, the station immediately starts
transmission as the backoff counter reaches 0.
After the successful packet transmission, the CW is determined anew. If another station
transmits during the backoff time, the backoff counter is frozen until the channel becomes idle
(transmission is finished and DIFS expired). With this procedure, it is easy so see that the more
times the station freezes its backoff counter the grater the probability that the packet waiting
for transmission will not collide.
The described algorithm to get an access to the wireless medium guarantees the equal access
for all stations in the network. But under such procedure, the probability of collision (that
an arbitrary packet collides upon its transmission) is still non-zero. The collision probability
could be reduced by extension of the maximal CW. But it leads to the grater backoff value
which decreases the channel throughput. Therefore, the DCF uses the following algorithm
to minimize collisions. After each successful frame reception, the destination station sends
the ACK (ACKnowledgement) after SIFS to acknowledge the transmission was successful. If
collision happens upon transmission the sender station does not get an ACK, so it finds out
the transmission failed. The sender waits for ACK during EIFS, and if the ACK is not received
the sender increases its CW. Thus, if the CW is 31 slots for the first transmission attempt for
the second one it is 63, 127 for the third, 255 for the fourth, 511 for the fifth, and 1023 for
other attempts. It can be seen that the CW is dynamically increased with collision number
increasing, which allows reducing both the delay and the collision probability.
Note that the sender station can not receive the ACK frame indicating the transmission was
success due to collision or signal distortion. And both reasons are not distinguishable for the
sender station.
As it is mentioned above, all stations are equal to get an access to the wireless medium due to
the contention mechanism, and no station has a priority to transmit data. This restricts DCF to
provide Quality of Service (QoS). In IEEE 802.11e protocol, QoS is enabled by Enhanced DCF
(EDCF). The EDCF mechanism is similar to DCF, but the difference concerns the CW size and
the backoff counter to provide the priority access for various applications to wireless medium.
For EDCF the traffic is divided into categories (TC, Traffic Categories) which differ from each
other by priority to get an access to the transmission medium. Meanwhile, the medium access
mechanism is the same as for DCF contention based.
Before a station detecting the channel idle starts transmission, it waits for AIFS (Arbitration
InterFrame Space) and then counts down the backoff counter. The backoff counter is uniformly
chosen in the range [1, CW (TC) + 1] where CW (TC) is the contention window for the given
TC. TC priority is provided by using the different values of minimal and maximum CW and
AIFS. Thus, the minimal value of AIFS is always DIFS but it can be increased depending on
the TC.
In case of collisions which happen when two or more backoff counter drop to zero simulta-
neously and AIFS intervals are the same, the CW is increased. For EDCF, the new CW size is

www.intechopen.com
516 Trends in Telecommunications Technologies

determined as newCW (TC) = ((oldCW (TC) + 1)PF) − 1, where PF is the constant CW scaling
factor depending on the TC. Note that for DCF we have PF = 2.
Thus in EDCA, the TC is assigned a priority by variation of the parameters CWmin, CWmax,
AIFS, PF. Each station of the wireless network can have up to 8 TC queues for transmission.
But in this case, the TC backoff counters within the same station can drop to zero simulta-
neously which is called the virtual collision. To avoid virtual collisions, the station uses the
special queue scheduler which provides the higher priority TC with the priority access to the
transmission medium.
The considered mechanism of transmission medium multiple access control have the same
bottleneck, so-called problem of hidden stations. The situation happens when two stations
can not listen to each other directly due to natural barriers. Such stations are called hidden.
To avoid the problem, the DCF and EDCF mechanisms have the optional technique known as
Request-To-Send/Clear-To-Send (RTS/CTS).
Accordingly to RTS/CTS, before transmitting a packet, a station operating in RTS/CTS mode
"reserves" the channel by sending a special Request-To-Send short frame. An RTS frame
involves the information on the forthcoming transmission and destination station and is avail-
able to all stations in the network (except ones hidden from the sender station). It allows the
other stations to postpone transmission for the declared transmission time, and during this
time the channel is considered "virtually busy". The destination station acknowledges the
receipt of an RTS frame by sending back a Clear-To-Send (CTS) frame, after which normal
packet transmission and ACK response occurs. Since collision may occur only on the RTS
frame, and it is detected by the lack of CTS response, the RTS/CTS mechanism allows to in-
crease the system performance by reducing the duration of a collision when long messages
are transmitted.
The DCF and EDCF are simple and reliable mechanisms of multiple access to transmission
medium in IEEE 802.11 broadband wireless networks. But they have two shortcomings: the
lack of both QoS support and solution of the hidden stations problem. Thereby, the IEEE
802.11 protocol was supplemented with the alternative methods to multiple access control.

2.3 Point coordination function

The DCF mechanism described above is basic for IEEE 802.11 protocols and can be used in
both ad-hoc wireless networks and infrastructure networks (having Access Point, AP). But for
the infrastructure networks, there are more appropriate methods to control the transmission
medium multiple access, namely, Point Coordination Function (PCF) and Hybrid Coordination
Function (HCF). Note that PCF and HCF mechanisms are optional and applied for networks
with AP only.
In case of PCF, one of stations (access point) is central and called the Point Coordinator
(PC). The PC controls is imposed a responsibility to control the multiple access on the base
of the specified polling algorithm or information on station priority. Thus, the PC polls all
stations in the network listed in its polling table and organizes data transmission between
network stations. As opposed to DCF where each station decides for itself when to start
transmission, the PCF implies that only PC decides what station can get an access to the
channel. It is significant to note that this approach completely avoids the contention access
to the channel unlike DCF, makes collisions impossible and provide the priority access for
time-sensitive applications. Thus, PCF can be used to organize collision-free priority access to
the transmission medium.

www.intechopen.com
Performance Evaluation Methods to Study
IEEE 802.11 Broadband Wireless Networks under the PCF Mode 517

The Point Coordination Function does not contradict the Distributed Coordination Function
and rather supplements it. In fact, the PCF networks can use both PCF and traditional DCF.
During network operating, the time intervals for PCF and DCF alternate.
In order to provide alternating the PCF and DCF modes, the access point which realizes PCF
has to have a priority access to transmission medium. It is possible if an access to the channel
is contention one (as for DCF) but the time interval for the access point to wait the response is
less then DIFS. In this case, when the access point tries to get an access to the channel it waits
the end of transmission as other stations do and since it has the minimal inferframe space, it
gets an access first. The interframe space for the access point is called PIFS (PCF Interframe
Space), diven that SIFS < PIFS < DIFS.
The DCF and PCF mechanisms are combined in so called superframe which is the sum of PCF
interval of the contention-free access called CFP (Contention-Free Period) and the succeeding
DCF interval of CP (Contention Period). The CP length should be enough to enable the
transmission of at least one frame by using PCF mechanism. It is necessary for the association
procedure but it is out of our topic. The superframe starts with the beacon frame, and
all stations having received it postpone their transmissions for the time determined by the
CFP. The beacons contain the information on the CFP duration and allow synchronizing the
operation of all stations.
Under PCF, the access point sends data packets (DATA) destined for stations (if available)
and asks (polls) all stations about the frames waiting for transmission by sending them the
service frame CF-POLL (invitation to transmit). The access point polls stations accordingly
to its polling list (polling table). The methods to form and keep up the polling table are not
specified in the standard thus the wireless network developers can freely decide on how to
realize it.
A station can transmit packets to the channel only when it receives the CF-POLL. Having
received the CF-POLL, the station sends the short frame containing both data (if available)
and acknowledgement of CF-POLL reception after SIFS. If there are no data to transmit, the
station answers with NULL frame containing the header only. If the access point gets the
answer frame, it waits for SIFS and polls the next station. Otherwise if there is now answer
during PIFS, the access point considers the station unaccessible and polls the next station.
In order to cut expenses, an access point can combine the CF-POLL with data transmission
(the frame DATA+CF-POLL). Similarly, the stations are allowed to combine acknowledgement
frame with data transmission (the frame DATA+CF-ACK). Under the PCF, there are four types
of a frame:
• DATA, data frame;
• CF-ACK, acknowledgement frame;
• CF-POLL, poll frame;
• DATA+CF-ACK, combined frame of data and acknowledgement;
• DATA+CF-POLL, combined frame of data and poll;
• DATA+CF-ACK+CF-POLL, combined frame of data, acknowledgement and poll;
• CF-ACK+CF-POLL, combined frame of acknowledgement and poll.
Along with the extended distributed function EDCF in the IEEE 802.11e protocol, the Hybrid
Coordination Function (HCF) is determined. Similarly as the EDCF is extension of the DCF,
the HCF extends the PCF.

www.intechopen.com
518 Trends in Telecommunications Technologies

Due to the fact that the PCF and HCF realize centralized non-collision priority access to
transmission medium, they completely solve the problem of hidden stations and provide QoS.
But regardless of their advantages, the PCF and HCF are harder to be realized then the DCF.
Besides, the point coordination functions imply relatively large number of service frames (CF-
POLL, etc.) that essentially increases overhead expenses of data transmission in the wireless
medium. To organize the wireless network with PCF or HCF, all stations support these regimes
and one station serves as an access point while the DCF allows organizing the ad-hoc network
without access point. The distributed coordination functions are reasonable to be used in the
simplest wireless networks without hidden stations and delay-sensitive applications.

2.4 Problem of hidden stations in wireless metropolitan area networks

In IEEE 802.11 broadband wireless network, the basic structure unit is a radio-cell having star-
shape structure: the center is a base station with omnidirectional antenna which antennae of
all subscriber stations are focused on presenting the radio bridges between wireless network
and local cable networks.
In typical conditions of the wireless metropolitan area network, the subscriber stations do
not have a radio visibility for each other (they are hidden from each other) and have to
communicate through the retransmitting base station disposed on the high altitude (tower
buildings, TV tower, etc.) and providing an access to the outer network. Thus, the broadband
wireless metropolitan area networks have the mesh structure: a mesh is a radio cell and
subscriber stations are connected with the high-speed backborn network.
Below we consider the operation of a radio cell in the broadband wireless metropolitan area
network.
A radio cell has the following features:
• all subscriber stations are hidden from each other and their antennae are focused on the
base station, i.e. data transmission between subscriber stations is possible through the
base station only;
• the distances between the base station and subscriber ones are large enough (several
kilometers) and different.
Each of the IEEE 802.11 protocols supports its own set of transmission rates. For IEEE 802.11b,
they are the channel rates 1, 2, 5.5 and 11 Mbit/sec, for IEEE 802.11a they are 6, 9, 12, 18, 24, 48
and 54 Mbit/sec. A rate is a type of modulation which modulate the carrier radio signal or its
part in the system with several carriers. The "faster" modulations are less noise-resistant, and
otherwise. For this reason, the IEEE 802.11 based devises use more noise-resistant modulations,
or lower transmission rates in cases of the signal depression and large number of losses when
transmitting frames. It is obvious that a frame of n bits is transmitted longer by rate 1 Mbit/sec
(hence it reserves a channel for a longer time) than by rate 11 Mbit/sec. In case when a
subscriber station is far off or is placed in a nasty place, the radio signal it receives from the
base station is weak, and the subscriber station automatically decrease the transmission rate.
It results in decreasing of the network throughput. In case of PCF, the situation gets worth
as it is impossible to control the subscriber station behavior. Remind, that accordingly to the
DCF and EDCF mechanisms, the subscriber station decides by itself when to transmit data.
Besides, regardless of data transmission directions within a radio cell, they come through the
base station, thus it is logical to impose the base station on the radio cell control responsibility.
It easy to see that due to the above-mentioned features, it is not reasonable to use the distributed
coordination functions in broadband wireless metropolitan area networks. Even though the

www.intechopen.com
Performance Evaluation Methods to Study
IEEE 802.11 Broadband Wireless Networks under the PCF Mode 519

equipment supporting the centralized control is more complicated for development and pro-
duction hence it is more expensive, it much more efficiently uses two most valuable resources
of the broadband wireless network, they are frequency and throughput. The centralized
control in the MANs allows to avoid the problem of hidden stations, efficiently schedule an
order of station access to the wireless channel, flexibly control the radio cell operation and
change its parameters correspondingly to the current situation by adjusting only the base sta-
tion. The adaptive centralized control is a prospective direction of developing the IEEE 802.11
broadband wireless networks.

2.5 Adaptive dynamical polling in wireless networks

Since both the Point Coordination Function and Hybrid Coordination Function are based on
the centralized control (polling), the main attention should be focused on the method of the
centralized control realization. The specific polling mechanism and its parameters are the main
factors determining the efficiency of the broadband wireless MAN with centralized control.
The IEEE 802.11 standards allow the developers freely construct the methods of PCF and HCF
realization. The standard does not specify: the method to poll the subscriber stations, method
to construct and support the polling table, the policy to form queues of frames for transmission,
and recommendations on the relation between DCF and PCF intervals in a superframe. Further,
within the Section the main attention is given to development and modelling of the algorithms
to poll subscriber stations, the schemes of adaptive dynamic polling (ADP).
For each subscriber station, the base station forms a queue of frames. Within a queue, the
frames are ordered accordingly to the priority of traffic they belong to. Then the base station
starts the polling cycle. As the cycle starts, it polls the first station in the polling list (polling
table) sending frames from the corresponding queue to the station. Then the base station polls
it sending the CF-POLL (or QoS CF-POLL) and receives the frames to be transmitted from
the station (if available). As the base station finishes polling the subscriber one, it switches to
polling the next station accordingly to the polling table. Note that PCF leads to the overhead
expenses which can not be avoided during data exchange between the base station and the
subscriber one, they are the time to switch to subscriber stations and service frames for polling
them (CF-POLL, etc.).
The development the algorithm of adaptive dynamic polling lead to the following problems:
• choosing the method to poll the subscriber stations;
• choosing the policy to work with a subscriber station (receiving and transmission);
• development of methods to minimize the overhead expenses and;
• development of methods to optimize the system parameters.
Since the base station is substantially the central device in a radio cell of the broadband wireless
MAN, it is reasonable to provide it with the complete functionality concerning the radio cell
control, service of the queues, determination of the polling order and minimization of the
overhead expenses. For the equipment developers, it means that the base station has to carry
the majority of the network features. A subscriber station plays a passive role just answering
the service frames from the base station.
Efficiency of the broadband wireless PCF network performance considerably depends on the
method of polling mechanism realization. At the same time, the choosing of a cyclic polling
type and a queue service discipline is determined by the method of the broadband wireless
MAN (Metropolitan Area Network) application. Normally, there are two situations during the
broadband wireless MAN operating. In the first one, uplink traffic from the base station to end

www.intechopen.com
520 Trends in Telecommunications Technologies

stations dominates over the downlink one. Such a traffic is typical while using a broadband
wireless MAN cell as a "last mile" for Internet provider. In this case, each end station has its
own segment of LAN (Local Area Network) that gets an access to Internet via the wireless
network. The investigation of the real broadband wireless MAN (Vishnevsky, 2000) working
as the "last mile" shows that traffic from the outer network to the local segment is much higher
than the backward traffic.
In the second case, downlink traffic from an end station to the base one dominates over
the uplink one. Such a traffic is typical while using the broadband wireless network as
backbone network to transmit information from the objects "behind" the end stations to the
outer network (usually, to an information data storage center). The situation takes place when
the wireless network is used for the video monitoring systems, automatic process control
systems, telemetering, systems of collection, storage and processing of data.
Further we say a cell works in the "last mile" mode if the downlink traffic is dominate and it
works in the "data collection" mode if the uplink traffic is dominate.
The fundamental difference between two situations is that the base station (coordinator of the
wireless network cell work) knows the parameters of the queue of frames to be transmitted
to end stations in the case of dominating downlink traffic. Thus, the base station can choose
the policy to work with an end station queue before it polls the station. In the second case,
the base station knows nothing about the queues of frames to be transmitted when the uplink
traffic is dominate. Hence, it needs to poll the end station first and then to decide on the policy
to work with the queue.
When a cell works in the "data collection" mode we can neglect the data traffic from the base
station to the end ones considering just uplink traffic from the end stations to the base one.
In this case, the base station polls an end one first, that is connects to the end station and
starts transmitting. Then, the base station makes an attempt to switch to the next end station
accordingly to the polling table. Moreover, the base station does not know if the end station
will respond to the poll and if it will have the frames for transmission. In order to cut down
expenses of polling the empty end stations and stations that stopped working for some reason
we propose not to poll those stations at the next polling cycle.
Thus the rule to poll the end stations could be as follows. The base station polls a subscriber
one if it polled it in the previous cycle and the end station had frames for transmission or
it was skipped. In the case the end station did not respond to the base one in the previous
cycle or it was empty when being polled, it is skipped (not polled) by the base station in the
current cycle. Since the base station can not estimate the number of frames in the queue to
be transmitted in advance (before polling the end station) it is not reasonable to serve the
queue until it is empty. Thus, we propose the discipline to serve the end station: the base
station transmits only the frames which were present in the queue at the station polling epoch.
The adequate model to investigate the performance characteristics of the broadband wireless
network working in the "data collection" mode described above is the polling system with
adaptive polling mechanism and gated service analyzed in Section 3 where we presented the
analytical and simulation results.
When a cell works in the "last mile" mode, the data traffic from the end station to the base one
can be neglected. So we consider just downlink data traffic from the base station to the end one.
In this case, the base station sends frames from the queue to the corresponding end station.
Then, it waits for the successful data transmission acknowledgment from the subscriber station
(during PIFS) and starts sending frames from the next queue. Time to switch between queues
is random so it is impossible to say in advance if the data are transmitted successfully. If the

www.intechopen.com
Performance Evaluation Methods to Study
IEEE 802.11 Broadband Wireless Networks under the PCF Mode 521

queue of frames is short it is obvious that data transmission costs are high. So, it is better
to serve the queue only if its length exceeds the given value called the threshold in order to
cut down the expenses. To maximize the system throughput, the queues of each end station
should be served until empty. But in the case when one or several queues are long enough,
the frame mean waiting time in the queue is large that is unacceptable for some network
applications. So, the most important system parameters are the mean queue length and the
frame mean waiting time. These parameters can be estimated by means of stochastic model
with exhaustive threshold service considered in Section 4.

3. Adaptive polling system

The polling systems are varieties of the queuing systems with multiple queues and one (or
more) server(s) common to all queues that poll the queues and serve the queued customers.
Classification of polling systems and methods to study them are presented in reviews (Levy
& Sidi, 1990; Vishnevsky & Semenova, 2006) and books (Takagi, 1986; Borst, 1996; Vishnevsky
& Semenova, 2006).
The adaptive polling mechanism assumes that the server polls not all queues in a cycle. The
queues that were empty at the previous polling epoch are skipped (not visited) by the server
and are visit in the next cycle only. Unfortunately, the exact analysis of adaptive polling
systems is cumbersome, and in this Section, we provide an approximate analysis.
We consider the polling system with N queues (having unlimited waiting space) and a single
server which is common for all queues. Each queue, say i, has a Poisson input of customers
with parameter λi . Service times at queue i are independent and identically distributed with
(r) ∞
the distribution function Bi (t) with the moments bi = 0 tr dBi (t), r ≥ 1, and Laplace-Stieltjes
∞
transform (LST) βi (s) = 0 e−st dBi (t), i = 1, N.
The server visits queues in cyclic order from queue 1 to queue N accordingly to an adaptive
scheme. A cycle is referred to as the time the server spends visiting (serving) queues 1 through
N. With adaptive mechanism, the server skips (does not visit) queues which were visited in
the previous cycle and were found empty at their polling moments. After being skipped, a
queue is always visited in the next cycle.
If queue i is polled (or visited) by the server, the switchover time is incurred with distribution
(r)
function Si (t) having the LST S̃i (t) and the moments si , r ≥ 1, r = 1, N. A polling moment
is referred to as a moment when the switchover time is finished and the server is ready to
start working at a queue. The service discipline is gated, i. e. the server serves only those
customers that presented at a queue at its polling instant. The customers arriving during the
queue service time will be served in the next cycle.
If N queues were sequentially found empty at their polling moments, the server takes a
vacation having the distribution function F(t) with LST ϕ(s), and moments ϕ(k) , k ≥ 1. After a
vacation, server starts working at the queue that he stopped polling at and went for a vacation.
The approach we use to investigate the model with several queues is based on the decom-
position of the polling system to separate queues, further analysis of a queue as a queueing
system with server’s vacations, and then, application of the obtained results to the system
with several queues. The analysis aims at deriving the first and second moments of the mean
waiting time in a separate queue and is partially based on the results obtained in (Sumita,
1988) for the queueing system with server’s vacations which are discussed in Section 3.1.

www.intechopen.com
522 Trends in Telecommunications Technologies

3.1 Analysis of a single queue

First, consider a single queue with server’s vacations. Within this subsection, we omit the
lower index denoting the queue number. We assume that the vacation time depends on the
fact whether the queue was empty at the moment the server finishes the previous vacation
or it was not. If not, the vacation time has distribution function H (t) with LST h(s) and the
moments h(k) , k ≥ 1. If the queue is empty when server finishes a vacation, the next vacation
period is distributed with the function H̃ (t) having the LST h̃(s), and the moments h̃(k) , k ≥ 1.
To investigate the queue, consider the Markov chain describing the queue state at the end of
vacations. Let tk be the kth epoch when vacation is finished, k ≥ 1, and itk be the number of
customers at the queue at the epoch tk . It is easy to see that the process itk , k ≥ 1, presents the
discrete-time Markov chain, and its one-step transition probabilities

pi,j = P{itk+1 = j|itk = i}, i, j ≥ 0,

have the form

j
(i)

p0,j = ỹ j , pi,j = al y j−l , i > 0, j ≥ 0,
l=0
where
∞ ∞ ∞
(i) (λt)l −λt (∗i) (λt)l −λt (λt)l −λt
al = e dB (t), yl = e dH (t), ỹl = e dH̃ (t), l ≥ 0,
l! l! l!
0 0 0

and B(∗i) (t) is the i-fold convolution of the distribution function B(t).
It can be shown that under the condition ρ = λb(1) < 1 fulfilled, the stationary state probabili-
ties exist
q( j) = lim P{itk = j}, j ≥ 0.
k→∞
These probabilities satisfy the following set of the balance equations, j ≥ 0,
∞ ∞ ∞ ∞
(λt) j −λt  (λt)l −λt (∗i) (λt) j−l −λt
q( j) = q(0) e dH̃ (t) + e dB (t) e dH (t). (1)
j! l! ( j − l) !
0 i=1 0 0

By multiplying the equations (1) by the corresponding powers of z and summing them up,
it readily follows that the probability generating function (PGF) Q(z) = ∞ ( j) j

j=0 q z of the
stationary probabilities satisfies the functional equation

Q(z) = (Q(β(λ − λz)) − q(0) )h(λ − λz) + q(0) h̃(λ − λz), |z| ≤ 1. (2)

The equations of type (2) have been solved in (Sumita, 1988) on the base of the method to solve
the functional equations like (2) described in (Kuczma, 1968) and the corresponding result is
included into (Takagi, 1991), pp. 223-225.
Below, we briefly describe the solution of (2). Consider the sequence of functions η j (z), j ≥ 0,
which are defined recursively,

η0 (z) = z, η j+1 (z) = β(λ − λη j (z)), j ≥ 0. (3)

www.intechopen.com
Performance Evaluation Methods to Study
IEEE 802.11 Broadband Wireless Networks under the PCF Mode 523

It was proven in (Sumita, 1988) that if ρ = λb(1) < 1, the sequence of functions η j (z), j ≥ 0,
converges uniformly to 1 as j → ∞ for all z, 0 ≤ z ≤ 1. By substitution η j (z) instead of z in (2),
we get

Q(η j (z)) = Q(η j+1 (z))h(λ − λη j (z)) + q(0) (h̃(λ − λη j (z)) − h(λ − λη j (z))).

Continuing the substitution for j = 0, 1, . . . , n − 1, we have

n−1
 n−1
 j−1

(0)
Q(z) = Q(ηn (z)) h(λ − λη j (z)) + q (h̃(λ − λη j (z)) − h(λ − λη j (z))) h(λ − ληk (z)).
j=0 j=0 k =0
(4)
b

Here we assume that ck = 1 if b < a.
k =a
Now, let n → ∞ in (4). Using the normalization condition Q(1) = 1, we obtain
∞
 ∞
 j−1

(0)
Q(z) = h(λ − λη j (z)) + q (h̃(λ − λη j (z)) − h(λ − λη j (z))) h(λ − ληk (z)). (5)
j=0 j=0 k =0

Setting z = 0 at the latter equation, we get the relation for the probability q(0) :
 j−1
−1
∞
  ∞
  
q(0) = h(λ − λz( j) ) 1 − (h̃(λ − λz( j) ) − h(λ − λz( j) )) h(λ − λz(k) ) , (6)
 
 
j=0 j=0 k =0

where the quantities z( j) = η j (0), j ≥ 0, are given by the recursion

z(0) = 0, z( j+1) = β(λ − λz( j) ), j ≥ 0. (7)

Finally, the equations (1), (5)–(7) give the probabilities q( j) , j ≥ 0.

To obtain the performance characteristics of the system with adaptive polling scheme, we
need to derive formulas for the first three derivatives Q′ (1), Q′′ (1), Q′′′ (1) of the PGF Q(z)
at z = 1, which can be found by using (5). The formula (5) involves the LSTs (h̃(λ − λη j (z))
and h(λ − λη j (z)), where the functions η j (z), j ≥ 0, are given by recursion (3). Thus, in order
to obtain the formulas for the derivatives Q′ (1), Q′′ (1), Q′′′ (1) of the PGF Q(z), we have to
find the explicit expressions for the derivatives of the functions η j (z), j ≥ 0, at z = 1 :

1 − ρj
η j (1) = 1, η j ′ (1) = ρ j , j ≥ 0, η j ′′ (1) = λ2 b(2) ρ j−1 , j ≥ 0,
1−ρ

1 − ρ2j (1 − ρ j−1 )(1 − ρ j )

η j ′′′ (1) = λ3 b(3) ρ j−1 + 3λ4 (b(2) )2 ρ j−1
, j ≥ 0.
1 − ρ2 (1 − ρ)(1 − ρ2 )
Using the formulas obtained above, we can calculate the derivatives of functions h(λ −
λη j (z)), j ≥ 0, at z = 1: h(λ − λη j (z))|z=1 = 1,

1 − ρj
(h(λ − λη j (z)))′ |z=1 = λh(1) ρ j , (h(λ − λη j (z)))′′ |z=1 = λ2 h(2) ρ2 j + λ3 h(1) b(2) ρ j−1 ,
1−ρ

www.intechopen.com
524 Trends in Telecommunications Technologies

1 − ρj
(h(λ − λη j (z)))′′′ |z=1 = λ3 h(3) ρ3j + 3λ4 h(2) b(2) ρ j−1 +
1−ρ
 2j j−1 )(1 − ρ j ) 
(1) j−1 3 (3) 1 − ρ 4 (2) 2 (1 − ρ
+λh ρ λ b + 3λ (b ) .
1 − ρ2 (1 − ρ)(1 − ρ2 )
Finally, after cumbersome simplifications, the formulas for the derivatives Q′ (1), Q′′ (1),
Q′′′ (1) of the PGF Q(z) at z = 1 are obtained as

′ λh(1) λ(h̃(1) − h(1) )

Q (1) = + q(0) , (8)
1−ρ 1−ρ
λ3 h(1) b(2) + (λh(1) )2 2ρ
 2 (2)
′′ λ2 h ( 2 ) (0) λ (h̃ − h(2) )
Q (1) = + + q + (9)
1 − ρ2 (1 − ρ)(1 − ρ2 ) 1 − ρ2
3 (2) + 2ρλ2 h(1) ) 
(1) (1) λ b
+(h̃ − h ) ,
(1 − ρ)(1 − ρ2 )
′′′ λ3 h(3) λ4 h(1) b(3) 3ρλ4 h(2) b(2) + 3ρ(1 + 2ρ)λ3 h(1) h(2)
Q (1) = + + + (10)
1 − ρ3 (1 − ρ)(1 − ρ3 ) (1 − ρ2 )(1 − ρ3 )
3ρλ5 h(1) (b(2) )2 + 3λ4 (1 + 2ρ2 )(h(1) )2 b(2) 6ρ3 (λh(1) )3
 3 (3)
(0) λ (h̃ − h(3) )
+ + + q +
(1 − ρ)(1 − ρ2 )(1 − ρ3 ) (1 − ρ)(1 − ρ2 )(1 − ρ3 ) 1 − ρ3

λ4 (h̃(1) − h(1) )b(3) 3ρλ4 (h̃(2) − h(2) )b(2)

+ + +
(1 − ρ)(1 − ρ3 ) (1 − ρ2 )(1 − ρ3 )
3ρ2 λ3 (h̃(2) − h(2) )h(1) 3ρ(1 + ρ)λ3 (h̃(1) − h(1) )h(2)
+ + +
(1 − ρ2 )(1 − ρ3 ) (1 − ρ2 )(1 − ρ3 )
3ρλ5 (h̃(1) − h(1) )(b(2) )2 + 3λ4 (1 + 2ρ2 )((h̃(1) − h(1) )h(1) b(2) 6ρ3 λ3 (h(1) )2 (h̃(1) − h(1) )

+ + ,
(1 − ρ)(1 − ρ2 )(1 − ρ3 ) (1 − ρ)(1 − ρ2 )(1 − ρ3 )
Let ζ be the duration of the queue service (between two successive server’s vacations). Using
the relations obtained above, we can get the moments ψ̂(r) , r = 1, 2, 3 of ζ. The number
of customers served between two successive server’s vacations is a random variable, say τ,
′′ ′ ′′′
which has the first three moments L1 = Q′ (1), L2 = Q (1) + Q (1), L3 = Q (1) + 3L2 − 2L1 ,
and the service time of the kth customer in a service period is a random variable, say ξk , which
τ
has the moments b(r) , r = 1, 2, 3. It is easy to see that ζ =

ξk . The random variables ξk are
k =1
mutually independent and independent of τ. Thus, ζ equals to the sum of a random number
of the random variables. Using the technique of the conditional expectations, it can be shown
that the moments ψ̂(r) , r = 1, 2, 3, of the random variable ζ are

ψ̂(1) = b(1) L1 , ψ̂(2) = b(2) L1 + (b(1) )2 (L2 − L1 ), (11)

ψ̂(3) = L1 b(3) + 3b(1) b(2) (L2 − L1 ) + (b(1) )3 (L3 − 3L2 + 2L1 ). (12)

Denote by ψ(r) , r = 1, 2, 3, the conditional moments of the distribution of ζ, duration of the

queue service between two successive vacations, given that the queue is not empty. It is easy

www.intechopen.com
Performance Evaluation Methods to Study
IEEE 802.11 Broadband Wireless Networks under the PCF Mode 525

to see that the conditional moments ψ(r) , r = 1, 2, 3, are expressed in the terms of the moments
ψ̂(r) , r = 1, 2, 3, as follows:
ψ̂(r)
ψ(r) = , r = 1, 2, 3.
1 − q(0)
Thus, the formulas (5)–(7) give the stationary distribution of q(i) , i ≥ 0, the number of cus-
tomers in the queue at the epoch the server finishes a vacation, and the formulas (8)–(10)
express the first three moments of that distribution. ∞
Let W (x), x ≥ 0, be the distribution of the waiting time in the queue, and w(s) = 0 e−sx dW (w)
be its LST. It is easy to see that

w (s ) = w(0) ( s ) + w(1) ( s ) + w(2) (s ),

where w(0) (s) is the LST of the waiting time of an arbitrary customer arrived to the system
when the server was busy, w(1) (s) is the LST of the waiting time of an arbitrary customer
arrived to the system when the server was on an ordinary vacation (with distribution H (t)),
w(2) (s) is the LST of the waiting time of an arbitrary customer arrived to the system when the
server was on a special vacation (with distribution H̃ (t)).
Using the probabilistic sense of LST, one can verify the following formulas are valid

Q(β(λ(1 − β(s)))) − Q(β(s))

w(0) (s) = τ−1 h(s) ,
s − λ(1 − β(s))

h(λ(1 − β(s))) − h(s) (2)

w(1) (s) = τ−1 (Q(β(λ(1 − β(s)))) − q(0) ) , w (s) =
s − λ(1 − β(s))
h̃(λ(1 − β(s))) − h̃(s)
= τ−1 q(0) .
s − λ(1 − β(s))
These formulas lead to the following result.
Theorem 1. The LST w(s) of the waiting time is given by

τ−1
w(s) = (Q(β(s))(1 − h(s)) + q(0) (h(s) − h̃(s))).
s − λ(1 − β(s))

By differentiating w(s) at s = 0, we obtain the formula for the mean waiting time

v(2) λb(2) + 2ρh(1)

W (1) = + (13)
2v(1) 2(1 − ρ)

and for second moment W (2) of the waiting time

v(3) λb(3) + 3λb(2) h(1) + 3ρh(2) 2ρ2 h(1)

 (2) 
(2) (1) λb
W = + +W + (14)
3v(1) 3(1 − ρ) 1−ρ 1 − ρ2

with v(k) = (1 − q(0) )h(k) + q(0) h̃(k) , k = 1, 2, 3.

www.intechopen.com
526 Trends in Telecommunications Technologies

3.2 Approximate analysis of the system with N queues

As it follows from (6)–(14), the key role in calculation of the performance characteristics of
the system considered is played by the form of the functions h(s) and h̃(s), the LSTs of the
distributions of vacation period after visiting an empty or non-empty queue, respectively. In
order to apply the results obtained for the system with single queue to the polling system
with N queues, we have to estimate those LSTs. It is easy to see that they depend on the time
the server spends visiting other queues, and are unknown. Below, we elaborate a procedure
of successive approximations to calculate the first and second moments of waiting time in
each queue on the base of the analysis given in Section 3.1. Note that the unknown LSTs h(s)
and h̃(s) of vacation duration are improved at each stage of the procedure. And below, we
elaborate such a heuristic procedure.
( j) (0) (r) (r)
In the procedure, we use the quantities zi , j ≥ 0, qi , ψi , r = 1, 2, 3, and Wi , r = 1, 2,
i = 1, N, given by formulas (6)–(14) where all LSTs and the corresponding moments are given
the subscript i, the number of the queue.
Since a queue being polled and found empty will not be visited at the next cycle and will be
polled again two cycles later, it seems reasonable to assume that the time when the server is
away from the queue is equal to the double intervisit time as if the queue was visited each
cycle (after being polled and found non-empty). Besides, we need to take into account the
fact that, if all queues are found empty, they are all visited in the cycle following the server’s
vacation. Thus, we assume that h̃i (w) = χ̃i (w)S̃i (w), where χ̃i (w) = (χi (w))2 + q̄i (ϕ(w +
λi (1 − βi (w)))ri (w) − χi (w)), the functions χi (s) are given by formula (17),
N N
(0) (0) (0)
 
q̄i = qj , ri (w) = (q j + (1 − q j )ψ j (w))S̃ j (w).
j=1, ji j=1,ji

It follows that the moments of the distribution functions H̃ (t) and H (t) of a vacation period in
the system with single queue considered in Section 3.1 are defined as
(1) (1) (2) (2) (2) (1) (3) (3) (3) (2) (1) (2)
h̃i = χ̃i + si , h̃i = χ̃i + si + 2χ̃i si , h̃i = χ̃i + si + 3χ̃i si + 3χ̃i si ,

(l)
where χ̃i , i = 1, 3, are calculated as

(1) (1) (1) (1) (2) (2) (1) (1) (2) (2)
χ̃i = 2χi + q̄i (ϕ̂(1) + ri − χi ), χ̃i = 2χi + 2(χi )2 + q̄i (ϕ̂(2) + 2ϕ̂(1) ri + ri − χi ),

(3) (3) (1) (2) (1) (2) (3) (3)

χ̃i = 2χi + 6χi χi + q̄i (ϕ̂(3) + 3ϕ̂(2) ri + 3ϕ̂(1) ri + ri − χi ) ,
ϕ(r) , r = 1, 3, are given by
(2)
ϕ̂(1) = ϕ(1) (1 + ρi ), ϕ̂(2) = ϕ(2) (1 + ρi )2 + ϕ(1) λi bi ,

(2) (3)
ϕ̂(3) = ϕ(3) (1 + ρi )3 + 3ϕ(2) (1 + ρi )λi bi + ϕ(1) λi bi ,
(l)
and the moments ri are defined as

N
(1) (0) (1) (0) (1)

ri = (q j s j + (1 − q j )a j ),
j=1, ji

www.intechopen.com
Performance Evaluation Methods to Study
IEEE 802.11 Broadband Wireless Networks under the PCF Mode 527

N
(2) (0) (2) (0) (2)

ri = (q j s j + (1 − q j )a j +
j=1, ji
N
(0) (1) (0) (1) (0) (1) (0) (1)

+(q j s j + (1 − q j )a j ) (qk sk + (1 − qk )ak )),
k=1, ki,k j
N N
(3) (0) (3) (0) (3) (0) (2) (0) (2) (0) (1)
 
ri = (q j s j + (1 − q j )a j + 2(q j s j + (1 − q j )a j ) (qk sk +
j=1, ji k=1, ki,k j
N
(0) (1) (0) (1) (0) (1) (0) (2) (0) (2)

+(1 − qk )ak ) + (q j s j + (1 − q j )a j ) ( qk sk + (1 − qk )ak +
k=1, ki,k j
N
(0) (1) (0) (1) (0) (1) (0) (1)

+(qk sk + (1 − qk )ak ) (qm sm + (1 − qm )am ))).
m=1, mi,mj,mk

Further, we discuss the problem of determining the LST h(s).

On the first step of the procedure, assume that each queue is visited (polled) in each cycle and
the server serves only one customer in a queue per a visit. From this assumption it follows
that the initial form of the LST hi (s) of the vacation (intervisit time for queue i) could be
σi
hi (s) =
s + σi
with
N
(1) (1)

−1
( σi ) = (b j + s j ), i = 1, N.
j=1, ji

Then, the moments of the vacation period (intervisit time for queue i) are given by equations
(1) (2) (3)
hi = σ−1
i , hi = 2σ−2
i , hi = 6σ−3
i .

(r) ( j)
Using LSTs hi (s), i = 1, N, and the moments hi , r = 1, 3, we calculate the values of zi , j ≥ 0,
(0) (r) (r)
qi , ψi , r = 1, 2, 3, Wi , r = 1, 2, i = 1, N, by the formulas (6)–(14) for each queue i, i = 1, N.
Note that we can use the Lyapunov inequality to check the correctness of numerical calculation
(r)
of the moments. For instance, the values ψi , r = 1, 2, 3, have to satisfy the inequalities

(2) (1) (3) (1) (3) (2) 3

ψi ≥ (ψi )2 , ψi ≥ (ψi )3 , ψi ≥ (ψi ) 2 .

(r)
Having analyzed the moments ψi , r = 1, 2, 3, of queue i service period, we choose the form
of the LST ψi (s) as follows.
(2)
ψi 1
• If the value cψ = (1) is approximately equal to 1, we set ψi (s) = (1) , i.e. the
( ψi ) 2 1+sψi
vacation (intervisit time) is exponentially distributed.

www.intechopen.com
528 Trends in Telecommunications Technologies

• If cψ ≈ 1k , where k is some positive integer, we set

 k
 
 1 
ψi (s) =  (1)
 ,

 ψ
1 + s ki


i.e. the vacation has Erlang-k distribution.

• If cψ > 1, we suppose that a vacation has hyper-exponential distribution, i.e.

(1) (2)
µi µi
ψi (s) = pi + ( 1 − pi ) . (15)
(1) (2)
µi +s µi +s

(1) (2) (r)

The parameters pi , µi , µi are calculated through values ψi , r = 1, 2, 3, in the
following way. First, calculate the values
(l)
ψi v3 − v1 v2 v1 v 3 − ( v2 ) 2
vl = , l = 1, 2, 3, f1 = , f2 = ,
l! v2 − ( v1 ) 2 v2 − ( v1 ) 2
(1) (2)
(1) 2 (2) 2 µi (µi v1 − 1)
µi = , µi = , pi = .
(2) (1)
 
2
f1 + f1 − 4 f2 2
f1 − f1 − 4 f2 µi − µi

If the inequalities v2 > v21 , v3 > v1 v2 , v1 v3 > v22 and f12 > 4 f2 , are fulfilled then we define
(1) (2)
the function ψi (s) from (15). Otherwise, the values µi , µi and pi are chosen to satisfy
the relations

(1) 2 (1) (2) (1)
2(ψi ) (2)
2ψi (1 − pi ) − 2(1 − pi )pi (ψi − 2(ψi )2 )
pi  , 0 ≤ pi ≤ 1, µi = ,
(2) (1) (2)
ψi 2(ψi )2 − ψi pi
(2)
(1) (2) (1) pi µi
ψi µi − 1 + pi > 0, µi = .
(1) (2)
ψi µi − ( 1 − pi )
 
 6pi 6(1−pi ) (3) 
Note that the value pi should be chosen to minimize  (1) 3 + (2) 3 − ψi  for the better
 (µi ) ( µi ) 
approximation (Kazimirsky, 2002).
• If cψ < 1 but cψ does not equal to 1k approximately, the vacation period distribution can
be approximated by the phase distribution.
• If cψ ≈ 0, we use the following form:
(1)
s
ψi (s) = e−ψi .

www.intechopen.com
Performance Evaluation Methods to Study
IEEE 802.11 Broadband Wireless Networks under the PCF Mode 529

Now, suppose the LSTs ψi (s), i = 1, N, are known. Note that queues may have different forms
of the LST of intervisit times (server’s vacations).
Then, we make a simplifying assumption that the probability that queue i is found empty
at an arbitrary polling epoch is independent of the state of the other queues. Note that this
assumption may not be valid at all (e.g. if the other queues are empty, queue i gets more time
for its customers to be served, hence the probability that the queue is empty is greater then
in the case when all other queues are not empty). But this assumption makes it possible to
investigate the model with adaptive polling analytically. Thus, we can rewrite the form of the
LST hi (s) of vacation duration for queue i (its intervisit time) can be rewritten as

hi (w) = χi (w)S̃i (w), (16)

with
N
(0) (0)

χi ( w ) = (q j + (1 − q j )ψ j (w)S̃ j (w)). (17)
j=1, ji

The derivation of (16)–(17) is based on the probabilistic interpretation of LST as the probability
that a catastrophe from a Poisson input with parameter w does not occur during the period
considered. A catastrophe does not occur during a vacation time for the queueing system with
vacations corresponding to queue i if it does not occur during switchover time for this queue
(with probability S̃i (w)) and during the switchover and service periods for the rest of queues
in the current cycle (with probability χi (w)). The first term in (17) implies that a catastrophe
does not occur during the switchover time to queue j and the following service period with
probability 1, if the queue is not visited by the server, and with probability ψ j (w)S̃ j (w) if it
was.
Formulas (16) and (17) result in the relations for the moments of the intervisit time for queue i:
(1) (1) (1) (2) (2) (2) (1) (1) (3) (3) (3) (2) (1) (1) (2)
hi = χi + si , hi = χi + si + 2χi si , hi = χi + si + 3χi si + 3χi si ,
(18)

where
N N
(1) (0) (1) (2) (0) (2)
 
χi = (1 − q j )a j , χi = (1 − q j )a j + (19)
j=1, ji j=1, ji
N N
(0) (1) (0) (1)
 
+ (1 − q j )a j (1 − qk )ak ,
j=1, ji k=1, ki,kj
N N N
(3) (0) (3) (0) (2) (0) (1)
  
χi = (1 − q j )a j +2 (1 − q j )a j (1 − qk )ak +
j=1, ji j=1, ji k=1, ki,kj
N N N
(0) (1) (0) (1) (0) (1)
  
+ (1 − q j )a j (1 − qk )ak ( 1 − q m ) am , (20)
j=1, ji k=1, ki,k j m=1, mi,m j,mk

where, for m = 1, N,
(1) (1) (1) (2) (2) (2) (1) (1) (3) (3) (3) (2) (1) (1) (2)
am = sm + ψm , am = sm + ψm + 2sm ψm , am = sm + ψm + 3sm ψm + 3sm ψm .

www.intechopen.com
530 Trends in Telecommunications Technologies

Then using formulas (16)–(20) for the LSTs and the moments of the vacation duration (of
(0) (r) (r)
intervisit time for a queue), we recalculate the values qi , ψi , r = 1, 2, 3, Wi , r = 1, 2,
i = 1, N, by formulas (6)–(14) for all i, i = 1, N.
(0) (r)
The iterative procedure described above should be repeated until the values qi , ψi , r =
(r)
1, 2, 3, Wi , r = 1, 2, i = 1, N, calculated at the succeeding steps coincide with the necessary
(r)
accuracy. Thus, we get the moments, Wi , r = 1, 2, i = 1, N, of the waiting time in the queues.

3.3 Numerical results

Here we give numerical examples to illustrate how the algorithm works for polling systems
with various numbers of queues and traffic intensity comparing to the simulation results
obtained using the general-purpose simulation system GPSS World (Schriber, 1974). The
object of modeling was represented by a regional broadband wireless network consisting of
several devices operating with one base station. The rates of packet arrivals to the devices and
the rate of processing them are different. The devices are polled cyclically. Packet servicing
is gated, that is, only those packets are transmitted which were at the queue at the polling
moment. The input flows are assumed to be of the Poisson nature, and the times of packet
servicing and polling initialization are assumed to be exponentially distributed.
We assume for simulation that the system is in the stationary mode when at duplication of
the number of the packets passing through the system none of the comparison parameters
changes more than by 0.5%. In the experiments, more than three million of packets passed
through the system.
Average packet service time, switchover time, etc. in the numerical examples are taken from
the real IEEE 802.11a broadband wireless network under PCF mode with realistic packet sizes
and load levels. Thus, the obtained results correspond to such networks and mean waiting
times satisfy the real systems requirements.
Case N = 2. First, consider the symmetric system of two queues with the mean service times
(1) (1) (1) (1)
b1 = b2 = 0.311, the mean switchover times s1 = s2 = 0.091 and the mean time of
server’s vacation ϕ(1) = 0.005. The mean waiting time calculated by the Algorithm (column
"A"). Simulation results (column "S") and relative error of comparison (column "∆") are shown
in Table 1. The first column describes the customers input intensities and the corresponding
traffic intensities. Two last lines of the Table present the results obtained for different mean
server’s vacation times ϕ(1) given that λ1 = λ2 = 0.5.

A S ∆, %
λ1 = λ2 = 0.321, ρ = 0.2 0.289 0.268 7.8
λ1 = λ2 = 0.5, ρ = 0.311 0.392 0.358 9.5
λ1 = λ2 = 0.803, ρ = 0.5 0.659 0.601 9.7
λ1 = λ2 = 1.28, ρ = 0.8 1.73 1.93 10.4
ϕ(1) = 0.05 0.392 0.358 9.5
ϕ(1) = 0.1 0.417 0.384 8.6
Table 1. System with two queues
(1)
Case N = 3. Now, consider the case of three queues with symmetric service bi = 0.044,
(1) (1)
si = 0.1, i = 1, 3, ϕ(1) = 0.1. The mean waiting times Wi , i = 1, 3 obtained by using the
algorithm and simulation are presented in Table 2 for various customer input intensities. The
last two lines contain results for fully symmetric system (all λi , i = 1, 3 are the same).

www.intechopen.com
Performance Evaluation Methods to Study
IEEE 802.11 Broadband Wireless Networks under the PCF Mode 531

Case N = 5. And, finally, consider the case of five queues with λ1 = 1, λ2 = 2, λ3 = 0.5,
(1) (1)
λ4 = 6, λ5 = 0.5, bi = 0.05, si = 0.05, i = 1, 5, ϕ(1) = 0.05. We vary the input intensity by
multiplying all λi by α which takes values 0.285, 0.714, 1, and 1.143. Thus, the traffic intensity,
ρ, varies from 0.2 to 0.8. The results are given in Table 3. The last four lines present results for
the fully symmetric system with λi = 2 multiplied by the same values of α.

A S ∆, % A S ∆, %
λ1 = 2.5 0.342 0.365 6.3 λ1 = 4.375 0.658 0.698 5.7
λ2 = 6 0.335 0.361 7.2 λ2 = 10.5 0.781 0.834 3.0
λ3 = 0.5 0.410 0.440 6.8 λ3 = 0.875 0.778 0.805 3.4
Symmetric system
λi = 3, i = 1, 3 0.387 0.382 1.3 λi = 5.25, i = 1, 3 0.702 0.771 8.9
Table 2. System with three queues

A S ∆, % A S ∆, %
(1) (1)
W1 0.203 0.220 7.7 W1 0.714 0.672 6.2
(1) (1)
α = 0.285 W2 0.199 0.215 7.4 α=1 W2 0.661 0.618 7.0
(1) (1)
ρ = 0.2 W3 0.205 0.222 7.7 ρ = 0.7 W3 0.776 0.738 5.1
(1) (1)
W4 0.197 0.203 3.0 W4 0.705 0.679 3.8
(1) (1)
W5 0.205 0.224 8.5 W5 0.776 0.745 4.2
(1) (1)
W1 1.036 0.974 6.4 W1 0.374 0.398 6.0
(1) (1)
α = 0.714 W2 0.353 0.370 4.6 α = 1.143 W2 0.967 0.934 3.5
(1) (1)
ρ = 0.5 W3 0.393 0.419 6.2 ρ = 0.8 W3 1.153 1.080 6.8
(1) (1)
W4 0.340 0.355 4.2 W4 1.152 1.120 2.9
(1) (1)
W5 0.393 0.422 6.9 W5 1.153 1.090 5.8
Symmetric system
(1) (1)
ρ = 0.2 Wi 0.207 0.216 4.2 ρ = 0.7 Wi 0.759 0.686 10.6
(1) (1)
ρ = 0.5 Wi 0.455 0.391 16.4 ρ = 0.8 Wi 1.003 1.040 3.6
Table 3. System with five queues
(1) (1) (1) (1)
The results for non-symmetric service (b1 = 0.07, b2 = 0.015, b3 = 0.1, b4 = 0.025,
(1)
b5 = 0.4) are given in Table 4.
Below, we discuss the results in brief. For the symmetric systems with N = 2 and N = 3,
the relative error of comparison grows as the traffic intensity ρ grows. But the situation is
different in case N = 5: the approximate and simulation results coincide up to 5% for low and
high traffic intensity, but for ρ = 0.5 and ρ = 0.7, the error becomes unacceptable (grater than
10%). For the system with five queues, the dependence of the relative error on the total traffic
intensity, or the traffic intensity to a queue, is not well-understood. In case of non-symmetric
system with five queues (Tables 3 and 4), the results can not be well explained. In case of
non-symmetric input of customers (Table 3), the coincidence gets better as ρ increases, but
when we make the service non-symmetric (Table 4), it holds only for queues with relatively
high traffic intensities, namely, queues 4 and 5, as long as the results for the rest of queues
behave the different way (the relative error grows as the values of ρi , i = 1, 3, grow). Note that
in Table 3, the results for queues 3 and 5 have to be the same as the queues are identical but
the simulation results differ up to 1%, which is the simulation error.

www.intechopen.com
532 Trends in Telecommunications Technologies

A S ∆, % A S ∆, %
(1) (1)
W1 0.251 0.250 0.4 W1 0.570 0.506 12.7
(1) (1)
α = 0.4, W2 0.248 0.244 1.6 α = 1, W2 0.535 0.475 12.7
(1) (1)
ρ = 0.2 W3 0.251 0.254 1.2 ρ = 0.5 W3 0.592 0.548 7.9
(1) (1)
W4 0.244 0.228 7.0 W4 0.516 0.455 13.4
(1) (1)
W5 0.223 0.254 12.2 W5 0.538 0.559 3.8
(1) (1)
W1 0.318 0.314 1.3 W1 1.016 0.902 12.6
(1) (1)
α = 0.6, W2 0.311 0.302 3.0 α = 1.4, W2 0.901 0.831 8.4
(1) (1)
ρ = 0.3 W3 0.322 0.325 0.9 ρ = 0.7 W3 1.095 0.994 10.2
(1) (1)
W4 0.305 0.281 8.5 W4 0.938 0.896 4.7
(1) (1)
W5 0.281 0.325 13.5 W5 1.082 1.080 0.2
Table 4. System with five queues and non-symmetric service

To complete the numerical analysis, we compare two polling schemes in the system working
in "data collection" mode described in Section 2.5. The comparison is based on the radio cell
model with parameters N = 4, input intensities λ1 = λ2 = 1500, λ3 = λ4 = λ take values
(1) (1) (1) (1)
from 1 to 200 with step 10, mean service times are b1 = b2 = b4 = 1/4500, b3 = 1/3000
(1) (1) (1) (1)
and the mean switch-over times are s1 = s2 = s3 = s4 = 1/1500. The obtained results
are shown on Fig. 1. The figure shows that adaptive dynamical polling can make significant
profit when some stations have light traffic, since the waiting times are the same for classical
cyclic polling system and one with adaptive polling in case of heavy traffic.

Fig. 1. The dependence of the mean waiting time on λ under the adaptive cyclic polling

4. System with exhaustive threshold service

In this Section, we present the polling system modelling a radio-cell working in the "last mile"
mode as described in Section 2.5. We consider the polling system described in Section 3, but
we assume that the service at a queue is exhaustive (the queue is served until empty).
The server visit a queue only if the queue length exceeds the given threshold (ki for queue i), ki ≥
0, i = 1, N. Before starting service at the queue i, the server needs a random time exponentially

www.intechopen.com
Performance Evaluation Methods to Study
IEEE 802.11 Broadband Wireless Networks under the PCF Mode 533

distributed with the parameter si to prepare for work. This time can be considered the time
to switch to queue i which is incurred only if the queue is served. The queue also has a finite
waiting space hi (hi ≥ ki ).
The service time in queue i is exponentially distributed with parameter µi , i = 1, N. The queue
is served until it becomes empty, then the server moves to the nearest queue which has the
sufficient number of customers (having reached its threshold). If the numbers of customers in
all queues are insufficient to get service (less than ki for queue i), the server stops polling until
the number of customers in any queue reaches the threshold.

4.1 Stationary distribution of the system states and the performance characteristics
A system state at an arbitrary time t in a steady-state is presented by a random process
ξ(t) = (m(t), i(t), n(t)), t ≥ 0,
where m(t) = 0, if at time t the server is idle, m(t) = 1, if at time t the server switches to a
queue, m(t) = 2, if at time t the server serves the customers, t ≥ 0; i(t) is the number of queue
which server is attended at time t, i(t) = 0, if the server is idle; n(t) is a vector
n(t) = (n1 (t), n2 (t), . . . , nN (t)),
n j (t) is the number of customers at queue j at time t, j = 1, N.
The stochastic process ξ(t), t ≥ 0, is Markovian. Introduce the stationary state probabilities of
the process ξ(t), t ≥ 0 for r = (r1 , r2 , . . . , rN ), i = 1, N,

a(r) = lim P{m(t) = 0, i(t) = 0, n(t) = r}, 0 ≤ rm < km , m = 1, N,

t→∞

pi (r) = lim P{m(t) = 1, i(t) = i, n(t) = r}, rm = 0, hm , m = 1, N, m  i, ri = ki , hi ,

t→∞

qi (r) = lim P{m(t) = 2, i(t) = i, n(t) = r}, 0, hm , m = 1, N, m  i, ri 1, hi .

t→∞
The balance equations for the stationary probabilities are
N
 N

λa(r) = λ j a(r − e j )I{r j >0} + µ j q j (r + e j )I{r j =0} , r1 < k1 , . . . , rN < kN , (21)
j=1 j=1
 N  N
   

 λm I{rm <hm } + µi  qi (r) =

 λ j qi (r − e j )I{r j >δij } + µi qi (r + ei )I{ri <hi } + si pi (r)I{ri ≥ki } , (22)
m=1 j=1
 N  N
   

 λm I{rm <hm } + si  pi (r) =

 λ j pi (r − e j )I{r j >ki δij } + µi−1 qi−1 (r + ei−1 )I{ri−1 =0} + (23)
m=1 j=1

+λi a(r1 , . . . , ri−1 , ki − 1, ri+1 , . . . , rN )I{r1 <k1 ,...,ri−1 <ki−1 ,ri =ki ,ri+1 <ki+1 ,...,rN <kN } +
i−2

+ µ j q j (r + e j )I{r j =0,r j+1 <k j+1 ,...,ri−1 <ki−1 } +
j=1
N

+ µ j q j (r + e j )I{r j =0,r j+1 <k j+1 ,...,rN <kN ,r1 <k1 ,...,ri−1 <ki−1 } ,
j=i+1

0 ≤ rm ≤ hm , m = 1, N, m  i, ki ≤ ri ≤ hi , i = 1, N.

www.intechopen.com
534 Trends in Telecommunications Technologies

By replacing one of the equations of system (21)–(23) by normalization condition

 N 
 N 

a(r) + qi ( r ) + pi (r) = 1,
r∈Λ i=1 r∈Πi i=1 r∈χi

with Λ = {(r1 , . . . , rN ) : rm < km , m = 1, N}, Πi = {(r1 , . . . , rN ) : 0 < ri ≤ hi , 0 ≤ rm ≤ hm , m =

1, N, m  i}, χi = {(r1 , . . . , rN ) : ki ≤ ri ≤ hi , 0 ≤ rm ≤ hm , m = 1, N, m  i}, i = 1, N, we get a
system of equations for N
N N
i=1 (2hi − ki + 1) j=1, ji (h j + 1) +

j=1 k j unknowns.
Having calculated the stationary state distribution, we can readily derive the following per-
formance characteristics:
1. Mean length of queue j at time when queue i is served (excluding a customer being
served): Lij = (r j − δij )qi (r), i, j = 1, N;

r∈Πi

2. Mean length of queue j at time when the server switches to queue i: Sij = r j pi ( r ) ,

r∈χi
i, j = 1, N;
3. Mean length of queue j at time when the server is idle: U j = r∈Λ r j a(r), j = 1, N;


4. Mean length of queue j at arbitrary time: L j = N i i

i=1 (L j + S j ) + U j , j = 1, N;


5. The mean fraction of time the server is idle: ā = r∈Λ a(r).



Since the waiting space in the system is limited, some arriving customers can be lost. The
j
probability Plost that an arbitrary customer arriving to queue j is lost equals to the probability
that an arbitrary time all h j waiting places are occupied,

N N 
j
  
Plost = qi (r)I{r j =h j } + pi (r)I{r j =h j } , j = 1, N.
i=1 i j r∈Πi i=1 r∈χi

The mean waiting time in queue j can be obtained by Little’s law W j = Li /λi .
Consider the model of the broadband wireless network radio cell working in the "last mile"
mode with adaptive polling mechanism. The model parameters are the following: N = 4,
λ1 = λ2 = 1500, λ3 = λ4 = λ take the values 30, 60, 100. The service intensities µ1 = µ3 =
µ4 = 4500, µ2 = 3000, the intensities of switching between queues s1 = s2 = s3 = s4 = 1500.
The Fig. 2 shows the dependence of the frame mean waiting time on the threshold value ki = k
(i = 1, 4) assumed to be the same for all the queues. We see that the optimal value k varies
from curve to curve. The less the queues are loaded the grater the k minimizing the mean
waiting time. If the system has queues with low traffic it is reasonable to increase their service
thresholds. As a result, the frame mean waiting time in queues with low traffic increases
but it decreases for queues with high traffic so it can reduce the weighted sum of the mean
waiting times in the system. The optimal threshold choosing can be a troublesome problem in
practise since the threshold value depends on the relation between numbers of stations with
light traffic and heavy traffic, station parameters, etc.

www.intechopen.com
Performance Evaluation Methods to Study
IEEE 802.11 Broadband Wireless Networks under the PCF Mode 535

Fig. 2. The dependence of the mean waiting time on the threshold

5. References

Borst, S.C. (1996). Polling Systems, Stichting Mathematisch Centrum, ISBN 90-6196-467-9,
Amsterdam.
IEEE Std 802.11-2007. Part 11: Wireless LAN Medium Access Control (MAC) and Physical
Layer (PHY) Specifications, June 2007.
Kazimirsky, A. (2002). On approximation of arrival flows with MAPs of order two, Proceedings
of IST’2002, pp. 35-40, Minsk, November 2002. (in Russian)
Kuczma, M. (1968). Functional Equations in a Single Variable, Polish Scientific Publishers, War-
saw.
Levy, H. & Sidi, M. (1990). Polling systems: applications, modeling and optimization. IEEE
Transactions on Communications, 38, 1750-1760. ISSN 0090-6778.
Schriber, T.J. (1974). Simulation Using GPSS, John Wiley & Sons.
Sumita, S. (1988). Performance analysis of interprocessor communications in an electronic
switching system with distributed control. Performance Evaluation, 9, 83-91, ISSN
0166-5316.
Takagi, H. (1986). Analysis of Polling Systems, MIT Press, Cambridge, ISBN 978-0262200578.
Takagi, H. (1991). Queueing Analysis: A Foundation of Performance Evaluation, Vol. 1, North-
Holland, ISBN 978-0444817709.
Vishnevsky, V.M. (2000). Wireless networks for broadband access to the Internet. Elektrosvyaz’,
10, 9-13, ISSN 0013-5771. (in Russian)
Vishnevsky V.M. & Frolov, S.A. (2009). Application for an invention No. 2007136908 of
03.02.2009 "Method of development of the very high throughput broadband wirelss
mesh networks".
Vishnevskii, V.M. & Semenova, O.V. (2006). Mathematical methods to study the polling
systems. Automation and Remote Control, 67, 173-220, ISSN 0005-1179.

www.intechopen.com
536 Trends in Telecommunications Technologies

Vishnevskii, V.M. & Semenova, O.V. (2007). Polling Systems: Theory and Applications for Broad-
band Wireless Networks, Moscow, Technosphera, ISBN 978-5-94836-166-6. (in Russian)
Vishnevsky, V.M.; Portnoy, S.L. & Shakhnovich, I.V. (2009). Encyclopaedia WiMAX. Way to 4G,
Moscow, Technosphera. (in Russian)

www.intechopen.com
 Trends in Telecommunications Technologies
Edited by Christos J Bouras

ISBN 978-953-307-072-8
Hard cover, 768 pages
Publisher InTech
Published online 01, March, 2010
Published in print edition March, 2010

The main focus of the book is the advances in telecommunications modeling, policy, and technology. In
particular, several chapters of the book deal with low-level network layers and present issues in optical
communication technology and optical networks, including the deployment of optical hardware devices and the
design of optical network architecture. Wireless networking is also covered, with a focus on WiFi and WiMAX
technologies. The book also contains chapters that deal with transport issues, and namely protocols and
policies for efficient and guaranteed transmission characteristics while transferring demanding data
applications such as video. Finally, the book includes chapters that focus on the delivery of applications
through common telecommunication channels such as the earth atmosphere. This book is useful for
researchers working in the telecommunications field, in order to read a compact gathering of some of the
latest efforts in related areas. It is also useful for educators that wish to get an up-to-date glimpse of
telecommunications research and present it in an easily understandable and concise way. It is finally suitable
for the engineers and other interested people that would benefit from an overview of ideas, experiments,
algorithms and techniques that are presented throughout the book.

How to reference
In order to correctly reference this scholarly work, feel free to copy and paste the following:

Vladimir Vishnevsky and Olga Semenova (2010). Performance Evaluation Methods to Study IEEE 802.11
Broadband Wireless Networks under the PCF Mode, Trends in Telecommunications Technologies, Christos J
Bouras (Ed.), ISBN: 978-953-307-072-8, InTech, Available from: http://www.intechopen.com/books/trends-in-
telecommunications-technologies/performance-evaluation-methods-to-study-ieee-802-11-broadband-wireless-
networks-under-the-pcf-mode

InTech Europe InTech China

University Campus STeP Ri Unit 405, Office Block, Hotel Equatorial Shanghai
Slavka Krautzeka 83/A No.65, Yan An Road (West), Shanghai, 200040, China
51000 Rijeka, Croatia
Phone: +385 (51) 770 447 Phone: +86-21-62489820
Fax: +385 (51) 686 166 Fax: +86-21-62489821
www.intechopen.com
© 2010 The Author(s). Licensee IntechOpen. This chapter is distributed
under the terms of the Creative Commons Attribution-NonCommercial-
ShareAlike-3.0 License, which permits use, distribution and reproduction for
non-commercial purposes, provided the original is properly cited and
derivative works building on this content are distributed under the same
license.