Out-of-Band Signaling Scheme for High Speed Wireless LANs

Abstract

In recent years, the physical layer data rate provided by 802.11 Wireless LANs has dramatically increased thanks to significant advances in the modulation and coding techniques employed. However, previous studies show that the 802.11 MAC operation, namely the distributed coordination function (DCF), represents a limiting factor: the throughput efficiency drops as the channel bit rate increases, and a throughput upper limit does indeed exist when the channel bit rate goes to infinite high. These findings indicate that the performance of the DCF protocol will not be efficiently improved by merely increasing the channel bit rate. This paper shows that the DCF performance may significantly benefit from the adoption of two separate physical carriers: one devised to manage the channel access contention, and another devised to deliver information data. We propose a scheme, referred to as out-of-band signaling (OBS), designed to reuse (and remain backward compatible with) the existing 802.11 medium access control (MAC) specification. Performance evaluation of OBS is carried out through analytical techniques validated via extensive simulation, for both saturation and statistical traffic conditions. Numerical results show that OBS improves the throughput/delay performance, and provides better bandwidth usage compared with the in-band signaling technique employed by DCF.

Full text

Page 1

3256 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 6, NO. 9, SEPTEMBER 2007
Out-of-Band Signaling Scheme for
High Speed Wireless LANs
Juki Wirawan Tantra, Member, IEEE, Chuan Heng Foh, Member, IEEE, Ilenia Tinnirello, and Giuseppe Bianchi
Abstract—In recent years, the physical layer data rate pro-
vided by 802.11 Wireless LANs has dramatically increased thanks
to significant advances in the modulation and coding techniques
employed. However, previous studies show that the 802.11 MAC
operation, namely the Distributed Coordination Function (DCF),
represents a limiting factor: the throughput efficiency drops as
the channel bit rate increases, and a throughput upper limit
does indeed exist when the channel bit rate goes to infinite
high. These findings indicate that the performance of the DCF
protocol will not be efficiently improved by merely increasing the
channel bit rate. This paper shows that the DCF performance
may significantly benefit from the adoption of two separate
physical carriers: one devised to manage the channel access
contention, and another devised to deliver information data.
We propose a scheme, referred to as Out-of-Band Signaling
(OBS), designed to reuse (and remain backward compatible with)
the existing 802.11 medium access control (MAC) specification.
Performance evaluation of OBS is carried out through analytical
techniques validated via extensive simulation, for both saturation
and statistical traffic conditions. Numerical results show that
OBS improves the throughput/delay performance, and provides
better bandwidth usage compared with the in-band signaling
technique employed by DCF.
Index Terms—Wireless LAN, IEEE 802.11, computer network
performance.
I. INTRODUCTION
A
area networks (WLANs) [1] have shown the most rapid and
notable increase in the achievable data transmission rate.
Born as a WLAN standard for slow data rate applications
(1 and 2 Mbps), the 802.11 physical layer specification has
been dramatically enhanced well above the most optimistic
initial plans. Building on the impressive market success of
the 11 Mbps IEEE 802.11b [2] physical layer enhancement
standardized in 1999 (and better known with its layman name
Wi-Fi, inherited by the relevant interoperability certification),
today most of the deployed network interface cards and access
points comply with the IEEE 802.11a/b/g specifications, and
hence support bit rates up to 54 Mbps in both the 2.4 GHz
(IEEE 802.11g [3]) and the 5 GHz ISM band (IEEE 802.11a
[4]).
MONG the various wireless technologies appeared in
the last 15-20 years, perhaps 802.11 wireless local
Manuscript received January 20, 2006; revised December 8, 2006; accepted
April 21, 2007. The associate editor coordinating the review of this paper
and approving it for publication was H.-H. Chen. I. Tinnirello’s research was
supported in part by the Italian Research Project PRIN 2005 MIMOSA.
J. W. Tantra and C. H. Foh are with the Centre for Multimedia and Net-
work Technology, School of Computer Engineering, Nanyang Technological
University, Singapore (email: jw.tantra@ieee.org, aschfoh@ntu.edu.sg).
I. Tinnirello is with the Department of Electrical Engineering, Universit` a
di Palermo, Palermo, Italy (email: ilenia.tinnirello@tti.unipa.it).
G. Bianchi is with the Dipartimento di Elettronica, Universit` a degli Studi
di Roma Tor Vergata, Roma, Italy (email: giuseppe.bianchi@uniroma2.it).
Digital Object Identifier 10.1109/TWC.2007.06029.
In the last few years, several proprietary solutions have been
offered by a multiplicity of vendors to support bit rates up to
108 Mbps and beyond. The pressuring demand for higher ca-
pacity, the promising and successful adoption of breakthrough
multiple-input multiple-output (MIMO) technologies, and the
need to develop interoperable products, have fostered IEEE
to form, in January 2004, a new Task Group, 802.11n [5],
chartered to develop a new high data rate amendment to
the 802.11 standard. From the original 100 Mbps goal, the
802.11n targeted maximum data rate has been continuously
increased during the task group n activities, to as much as
the current “beyond 500 Mbps” targets (and with a not so
unrealistic sight towards the Gbps barrier).
However, even impressive advances in terms of physical rate
are insufficient, by themselves, to increase the performance
experienced by the end users. In fact, the 802.11 medium
access control (MAC) protocol introduces severe overheads
which significantly reduce the throughput experienced at the
MAC Service Access Point (SAP) interface. For example, it
is straightforward to show that, with 802.11b, the maximum
throughput achievable at the MAC layer is typically1of the
order of 6-7 Mbps. Indeed, Xiao and Rosdahl [6] have shown
that the performance of the IEEE 802.11 MAC protocol drops
as the channel bit rate grows, and that a somewhat surprisingly
small throughput upper limit (e.g., in the order of 50 Mbps
for 802.11a with 1000 bytes payload sizes) does exist when
the channel bit rate goes to infinity.
These findings demonstrate that the performance of a
WLAN is significantly limited by the mechanisms employed
to control the access to the shared medium, and their re-
lated overhead. However, not only a complete re-design of
the 802.11 MAC protocol is highly unrealistic, as it would
impede backward compatibility with legacy devices, but also
amendments of the MAC operation should be designed with
great care, and should be devised to reuse as much as possible
the MAC primitives already deployed in the existing protocol
stacks.
The goal of this paper is to show that a simple way to
increase the performance of a high speed WLAN is to employ
two distinct carriers, one data channel devised to deliver
the actual information data, and another channel, referred to
1Actually, the precise value depends on the payload size and on some
physical layer parameters. For example, straightforward computation shows
that with 1470 bytes application layer datagrams encapsulated in UDP/IP
packets, and 192 µs long PLCP preamble, the application-layer throughput
perceived by a single user on an ideal 802.11b WLAN employing the DCF
basic access method is only 6.107 Mbps. Almost all the overhead is indeed
due to the 802.11 MAC operation. In fact the throughput experienced at the
MAC sub-layer rather than at application-layer, i.e. considering the UDP/IP
28 header bytes, plus the 8 bytes LLC-SNAP encapsulation as “useful” data,
is only marginally greater (6.257 Mbps).
1536-1276/07$25.00 c ? 2007 IEEE

Page 2

TANTRA et al.: OUT-OF-BAND SIGNALING SCHEME FOR HIGH SPEED WIRELESS LANS3257
as signaling channel, devised to manage the channel access
contention. Our proposed approach,referred to as Out-of-Band
Signaling (OBS), relies on standard-compliant mechanisms.
The signaling channel is in fact managed with a procedure
identical to the RTS/CTS handshake, while the data channel
is governed by the traditional procedures and control frames
developed for the Point Coordination Function (PCF) specified
in the 802.11 standard. Our study indicates that the current in-
band signaling method used in IEEE 802.11 on a high speed
channel actually causes higher bandwidth wastage compared
to our proposed out-of-band signaling method. The bandwidth
gained from the use of two channels in OBS exceeds the ad-
ditional bandwidth required for the signaling channel; hence,
the overall system performance is improved.
The rest of this paper is organized as follows. Section II
revisits related work presented in the literature. Section III
briefly reviews the IEEE 802.11 MAC protocol, and describes
how the protocol is adapted to operate within the proposed
OBS scheme. In Sections IV and V, we analyze the per-
formance of the OBS scheme under saturation condition and
discuss the results, respectively. Section VI provides analysis
and performance results of OBS scheme under statistical
traffic. Conclusions are drawn in Section VII.
II. RELATED WORK
Since the release of the IEEE 802.11 standard, several
efforts have been made to improve its performance. One com-
monly employed strategy is the improvement of the contention
resolution mechanism and the related backoff operation, in
order to achieve a higher protocol efficiency. Some examples
employing this strategy are given as follows. Cali et al. [7]
propose an adaptive backoff mechanism devised to overcome
the performance impairments that the standard “static” 802.11
backoff mechanism encounters in congestion situations, i.e.
when the default contention window settings are shown to
be suboptimal. Wang et al. [8] propose to slowly reduce
the contention window value when frames are successfully
transmitted. This leads to a less aggressive channel access
behavior which in turns results in lower steady-state collision
probabilities. Choi et al. [9] introduce a reservation scheme,
where each station broadcasts its future backoff information
on a successful transmission to avoid others from colliding
with its scheduled transmission in the future.
Another approach to improve protocol performance is the
minimization of the per-packet overheads due to overheads
of protocol headers. Xiao [10] analyzes the efficiency of
burst transmissions2, frame concatenationand packing. Similar
concepts are being discussed in [5], where the MAC acknowl-
edgments are aggregated into a single block ack as in [11].
Finally, in [12] the same TXOP is used for bi-directional
traffic, thus allowing a receiver to send back a frame burst
to its sender before performing the block ack transmission.
In infrastructure networks, the usage of centralized schedul-
ing algorithms deployed on top of extended versions of the
PCF is also a viable approach to improve performance. For
example, Ganz and Phonphoem [13] propose the use of
superpoll instead of single poll to better utilize the channel.
2This is known as Transmission Opportunity TXOP in the standard.
The access point polls multiple stations in one transmission
with indication of the transmission order of polled stations.
In [14], Lo et al. introduce CP-Multipoll where the access
point broadcasts the backoff counters of all polled stations for
efficient transmission scheduling. Lim et al. [15] propose a
polling mechanism where the AP first polls the stations to no-
tify data transmissions occurring on a separate channel. After
receiving the queue information from the stations, the AP uses
superpoll to poll the stations for the frame transmissions.
A rather different strategy to increase the performance of
WLANs is the multichannel approach which allows the MAC
protocol to simultaneously access multiple channels. There are
two commonly considered methods in this strategy, namely,
the aggregation of multiple orthogonal channels into a single
high data rate channel for operation, and management of mul-
tiple orthogonal channels for independent operation. Authors
in [16] employ the former method where two 20 MHz OFDM
channels are aggregated into a single 40 MHz channel with
backward compatibility consideration. However, as studied
in [6], such an aggregation does not overcome the MAC
overhead limits and therefore does not improve the asymptotic
protocol efficiency. The latter method that considers manage-
ment of multiple orthogonal channels has received increasing
attention in the recent literature, especially in the area of multi-
hop WLAN networks. Nasipuri et al. [17] propose a soft
channel reservation scheme where each station listens simulta-
neously to all the channels, and on a per-packet basis, selects
the least congested channel for transmission. This scheme has
been simplified in [18], in which a dedicated common channel
is introduced for broadcasting control messages. Furthermore,
the authors of [19] propose the use of fixed time intervals for
control message exchange within a selected data channel to
reduce control message overheads.
The scheme proposed in this paper falls under the strategy
of multichannel approaches. However, unlike previous work in
this area mainly devised to improve the performance of multi-
hop networks, in this paper we show that the availability of
two separate channels, one for resource reservation (i.e. sig-
naling) and the other for actual data delivery, indeed improves
the performance of a single-hop WLAN. The uniqueness of
our scheme is the combination of asymmetric dual-channel,
random access reservation and a polling mechanism for data
transmission that minimizes protocol overheads. Our early
protocol ideas and a preliminary performance investigation
through analytical modeling were presented in the conference
works [20] and [21], respectively. In this paper, we signif-
icantly revise and extend these earlier works, specifically i)
we provide a deeper understanding of the OBS operation and
effectiveness; ii) we extend the relevant performance evalu-
ation; iii) we provide a refined and more detailed analytical
modeling of the DCF backoff operation, and iv) we extend the
study of the OBS performance to the case of statistical traffic
(i.e. non saturation operation).
III. OUT-OF-BAND SIGNALING SCHEME
Rather than being a new MAC proposal for high speed
Wireless LANs, the OBS scheme is devised to operate,
with minimal and backward-compatible modifications, the
traditional IEEE 802.11 MAC mechanisms (specifically the

Page 3

3258IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 6, NO. 9, SEPTEMBER 2007
Distributed Coordination Function - DCF - and the Point
Coordination Function - PCF) over two separate carriers. One
low bit rate channel is devised to manage the contention
to access the shared medium, while a second high bit rate
contention-less channel is devised to deliver data frames. After
a very brief review of the basic principles of DCF and PCF (the
reader interested in additional details may refer to [1]), this
section follows up by describing the proposed OBS operation.
A. Distributed Coordination Function - DCF
The DCF employs the Carrier Sense Multiple Access with
Collision Avoidance (CSMA/CA) MAC protocol with binary
exponential backoff. The DCF does not use collision detection
function as the stations cannot detect collision by listening to
their own transmission; thus, it employs handshaking method,
which makes use of positive acknowledgment.
The basic access method employs a two-way handshaking
method. When a station has a new frame to transmit, it will
monitor the channel activity. If the channel is detected idle for
a period of time called distributed interframe space (DIFS),
the station can transmit immediately. If the channel is busy,
the station will defer until the end of transmission and a
random backoff interval is selected. The backoff counter is
decremented as long as the channel is sensed idle, stopped
when channel activity is detected, and reactivated when the
channel is sensed idle for more than a DIFS again. The station
transmits its frame when the backoff counter reaches zero.
The DCF uses a slotted binary exponential backoff tech-
nique. The period following an idle DIFS is slotted and the
backoff time counter is measured in terms of slot time. The slot
time is the time needed for any station to detect transmission
from other stations. It accounts for the propagation delay, the
time needed to switch from the receiving to the transmitting
state and the time to notify the MAC layer about the state of
the channel.
The backoff time is uniformly chosen in the range (0,CW-
1), where CW is the current contention window. At the first
transmission attempt, CW is set to the minimum contention
window (CWmin). After each unsuccessful transmission, CW
is doubled until it reaches the maximum contention window
(CWmax).
When the destination station successfully receives a frame,
it will transmit an acknowledgment (ACK) frame after a short
interframe space (SIFS). If the sender does not receive the
ACK frame within a specified ACK Timeout, or it detects the
transmission of a different frame on the channel, it reschedules
the frame transmission according to the specified backoff
rules. Additionally, the standard specifies that the frame will
be dropped after a certain number of retransmissions.
The DCF also defines an optional four-way handshaking
method for frame transmission, which is known as RTS/CTS
method. When a station is ready for a transmission, it performs
the specified backoff technique. The sender transmits a special
RTS frame when its backoff counter reaches zero. When the
receiver receives the RTS frame, it responds with CTS frame
after a SIFS. The sender may then transmit the frame only if
the CTS frame is correctly received.
The RTS and CTS frames carry the information of the
length of the frame to be transmitted, which is used to update
a network allocation vector (NAV) by other stations. The
NAV contains the information about the period of time in
which the channel will remain busy, hence a station can delay
transmission by detecting either RTS or CTS to avoid colli-
sion. The RTS/CTS method improves the system performance
when large frames are transmitted, as it reduces the duration
involved in transmission collisions.
B. Point Coordination Function
The IEEE 802.11 also specifies the optional PCF which is
implemented on top of the DCF. The PCF operation makes use
of polling commands issued by a “Point Coordinator” (PC),
which coincides with the Access Point (AP) in infrastructure
networks; hence, it is contention free. The AP uses point
coordination interframe space (PIFS) when issuing polls. The
PIFS is longer than SIFS but shorter than DIFS, hence the AP
can take control of the channel and stop all the asynchronous
traffic while it issues polls and receives responses. In PCF, the
ACK can be combined with data or poll frame, thus it has less
overhead compared to DCF.
C. Out-of-Band Signaling Scheme
OBS utilizes two different physical channels in a WLAN,
where one channel is operating at a low bit rate for channel
assignment purposes, and another channel is operating at a
high bit rate for the actual data transmissions. We shall refer
to these low and high bit rate channels as the “signaling” and
“data” channels, respectively.
In what follows, we shall describe the OBS operation
assuming that a reservation is performed for each single frame
to be transmitted (as discussed later in this section, the pro-
posed protocol can be significantly enhanced by performing a
reservation for several data frames). When a station is ready
for the transmission of a data frame, it transmits a Request
For Transmission (RFT) frame using the IEEE 802.11 basic
access method on the signaling channel. The RFT frame is
basically an IEEE 802.11 MAC control frame that uses one
of the unallocated frame type in the IEEE 802.11 standard. We
set the length of the RFT frame to be equal to the length of
RTS frame in the RTS/CTS method. When the AP receives the
RFT frame, it acknowledges the request and schedules the data
transmission on the data channel. Such a data transmission
occurs through the standard PCF operation, and Poll+ACK
frame are thus used for consecutive frame transmissions, to
reduce the overhead of polling. Note that the transmission of
the data frame may occur either right after the end of the
RFT/ACK handshake, or after an arbitrary time delay. Fig. 1
illustrates the operation of the OBS scheme.
The basic idea behind the OBS scheme is that, in the
DCF basic access method, transmission collision is the main
overhead of data transmissions. While the collision overhead
might be significantly reduced by the RTS/CTS operation,
due to the shorter length of the actual colliding frames, the
extra overhead introduced by the transmission of the RTS and
CTS control frames is a significant penalty, especially when
the channel data rate increases. By separating the contention
and the actual data transmission onto two different channels,
and using the low bit rate channel for the contention while
dedicating the high bit rate channel to data transmissions, the

Page 4

TANTRA et al.: OUT-OF-BAND SIGNALING SCHEME FOR HIGH SPEED WIRELESS LANS3259
costly transmission collisions, as well as the idle periods, can
be avoided entirely on the high bit rate channel. Additionally,
the use of short RTS/CTS-like frames for contention control
over the signalling channel is effective in reducing collision
overhead without incurring any significant overhead due to the
signalling channel lower rate. Thus, this separation mechanism
will improve the overall performance of WLANs.
The use of the OBS scheme also brings other advantages to
WLANs. The first immediate benefit of OBS is its flexibility
in terms of transmission scheduling. The AP in fact collects
requests from stations on the signaling channel, but is not com-
mitted to follow up with the relevant data frame transmission
in the same order. This allows the deployment of scheduling
approaches devised to provide service differentiation based on
requested quality of service (QoS) and the network setting. A
possible mechanism to achieve such service differentiation is
the introduction of fair queueing algorithm such as Deficit
Round Robin (DRR) [22] to ensure fairness or provide rate
differentiation among stations.
Additionally, any improvement to the PCF protocol is
applicable to OBS. Some examples include the combination
of ACK or Poll frames with data frames, and use of superpoll
[13] which further reduce transmission overheads. Moreover,
as in any reservation-polling scheme, another significant ben-
efit of OBS is the ability to use a single reservation occurring
on the signaling channel for the transmission of multiple data
frames or the set-up of a virtual circuit; thus significantly
reducing the relative overhead of the reservation handshake.
Furthermore, an important characteristic designed on pur-
pose in our OBS proposal is the backward compatibility with
the current IEEE 802.11 standard. In principle, each of the two
OBS channels, namely signaling and data, may be shared with
legacy DCF stations. In fact, the signaling channel relies on the
DCF operation (where the data frame is substituted, for OBS
stations, with the RFT frame), while the data channel relies
on standard PCF, which is meant to be deployed on top of
the DCF operation (the point coordinator reserves the channel
by using a PIFS as inter frame space, as briefly reviewed in
the previous section). More specifically, we deem convenient
to deploy a dedicated low rate signaling channel from free
available spectrum3, while the data channel can be deployed
using a standard IEEE 802.11a/b/g channel, and can be shared
with legacy DCF stations. Note that the data channel sharing
is ultimately controlled by the OBS operation, as the duration
of the PCF super frame is determined by OBS, and may
be dynamically adapted to the amount of traffic offered by
the OBS stations. Clearly, if we dedicate longer PCF super
frame to OBS data stations, we are left with lower amount
of resources to dedicate to legacy DCF stations. Arbitrary
policies may be considered to balance the channel sharing
between legacy and OBS stations: depending on the policy
employed, differences in experienced performance may arise
between these two classes of stations.
3The low bit rate channel has a maximum bit rate that is lower than the
data channel bit rate; Considering the same modulation technique for wireless
transmission for both channels, a 12 Mbps signaling channel requires about
20% of the spectrum allocated for a 54 Mbps channel in IEEE 802.11a.
Similar to the data channel, the bit rate for the signaling channel can be
lowered by link adaptation algorithm because of distance or noise.
SIFSSIFS
Sta−A PC
SIFS
PIFS
idle
slots
idle
slots
RFT ACKRFT
Sta−ASta−B
ACK
PollData
DIFS
Poll+ACK
time
Data Channel
Sta−B PC
Data
time
Signaling Channel
Fig. 1.The OBS scheme illustrated.
IV. SATURATION ANALYSIS OF
OUT-OF-BAND SIGNALING SCHEME
In this section, we describe the analytical model developed
for the study of the OBS scheme. In the proposed analysis, we
assume (worst case) that every frame transmission undergoes
an explicit reservation through the signaling channel. In other
words, we do not account for the possibility of a station to
inform the AP that one or more additional frames are queued.
In general, a station may be found in three possible states.
We define:
• Ready state: the station is currently scheduling its RFT
transmission on the signaling channel;
• Backlogged state: the station is currently waiting for a
poll message from the AP on the data channel;
• Idle state: the station has no frames available for trans-
mission.
A station switches from idle to ready when a frame arrives into
its local buffer. Then the transition from ready to backlogged
occurs when the station successfully exchanges the RFT/ACK
message with the AP, and thus waits for a Poll command
to be received on the data channel. When the data frame is
successfully transmitted, the station either returns to the idle
state, if no more frames are available in the local buffer, or
to the ready state if one or more frames are buffered for
transmission. The assumption of saturation [23] conditions
therefore implies that the station will only alternate between
ready and backlogged state.
The model proposed in what follows builds on the Markov-
ian Framework modeling technique presented in [24]. The key
idea is to model the number of stations in the backlogged state,
i.e. waiting for a Polling command from the AP or engaged
in the transmission of the data frame, with a single server
queue (SSQ). An arrival to the SSQ occurs when a station
switches from the ready to the backlogged state, i.e. when the
station successfully completes an RFT/ACK handshake on the
signaling channel. A departure from the SSQ occurs when a
data frame is successfully transmitted and thus the station re-
enters (owing to the assumption of saturation conditions) the
ready state.
Assume now that the total number of stations in the network
is n. Let m be the (time varying) number of backlogged
stations, i.e. the number of stations waiting in the SSQ. Hence,
the number of stations in the ready state is n−m. This remark
allows to model the arrival process to the SSQ with a state-
dependent arrival rate λn−m, being the rate of a successful

Page 5

3260IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 6, NO. 9, SEPTEMBER 2007
RFT/ACK message exchange conditioned to the fact that
n − m ready stations are competing to access the signaling
channel. In fact, a success of such an exchange generates a
new backlogged station, i.e., a new arrival to the SSQ. Hence,
the arrival process of the SSQ is the service process of the
signaling channel. The SSQ service rate μ is independent
of the SSQ state. In order to complete the model, we need
to i) determine the arrival rate λn−m for all the possible
values m, and ii) model the SSQ queueing system with proper
assumptions on the arrival and departure process statistics (for
which a Poisson approximation is clearly unrealistic). These
two issues are separately tackled in the next two subsections.
A. SSQ arrival rate computation
By model construction, the arrival rate to the SSQ is
equal to the service rate of the signaling channel, i.e. the
rate at which successful RFT/ACK handshakes occur. As it
will be demonstrated by the comparison between analytical
and simulation results, a good approximation is to compute
λn−m as the steady-state service rate of an 802.11 DCF
network with n−m stations. This computation can be easily
accomplished by applying the well known throughput analysis
proposed for the IEEE 802.11 DCF protocol starting from
[23], and adapting the analysis to the DCF basic access method
corresponding to the exchange of the RFT and the relevant
ACK frames (remember that the rest of the handshake, namely
the data frame exchange, occurs on the data channel and thus
is separately modeled).
In more details, in this paper we rely on the more general
modeling framework proposed in [25], which allows to take
into account retransmission limits (first developed in [26])
as well as a large class of more general backoff schemes
(although this latter feature is of no specific interest in this
paper). From [25], we can jointly compute the probability τ
that a station transmits in a randomly chosen slot, and the
probability p that a transmitting station (among n − m ready
stations) experiences a collision, by solving the following
system of two non-linear equations:
τ =
1
?R
1 +
1−p
1−pR+1
i=0piE[bi]
(1)
p = 1 − (1 − τ)(n−m)−1
(2)
where R is the retransmission limit, i.e. the maximum number
of retries after which a frame is dropped, and E[bi] is the
sequence of mean backoff values employed at each backoff
stage i. Since standard DCF is employed, in our first approx-
imation4,
E[bi] =min(2i(CWmin+ 1) − 1,CWmax)
2
, i = 1,2,...,R.
(3)
4In this formula, as well as in the following throughput formula, for
simplicity of presentation, we neglect the issue discussed in [25], [27]
concerning the different access probabilities experienced by the transmitting
and the listening stations for what concerns the slot immediately following a
transmitted frame in Legacy DCF (this problem has been amended in the
802.11e new specification of the backoff counter decrement). Indeed, its
impact in terms of numerical accuracy is marginal if standard 802.11a/b/g
DCF parameters are employed.
From the values p and τ, the service rate λn−mis readily
computed [23], [25] as the IEEE 802.11 DCF channel through-
put measured in frames per second, i.e.
λn−m=
PsPtr
(1 − Ptr)σ + PtrPsTs+ Ptr(1 − Ps)Tc
where σ is the length of a slot time, Ptr= 1−(1 −τ)(n−m)
is the probability that there is at least one transmission in the
considered slot time, Ps = (n − m)τ(1 − τ)n−m−1/Ptr is
the probability that a transmission occurring on the channel
is successful, and Ts and Tc are the average successful
transmission slot time and the average collision slot time,
given by [23]
?
Tc= RFT + DIFS + δ
(4)
Ts= RFT + SIFS + δ + ACK + DIFS + δ
.
(5)
B. Queueing model for the SSQ
Knowing that the signaling channel operates the basic ac-
cess method of the IEEE 802.11 MAC protocol, and according
to [24], the service time distribution of the IEEE 802.11
MAC protocol can be accurately described by an appropriate
Erlang distribution under saturation load condition5, then the
interarrival time of the SSQ can be accurately modeled by an
Erlang distribution.
We adopt the usual assumption in the MAC protocol per-
formance analysis that the frame size is constant. As a result,
the protocol service time distribution on the data channel is
deterministic. In the Markovian Framework, this distribution
can be approximated by an Erlang distribution due to its small
variance characteristics [24]. Consequently, we construct a
state dependent Ej/Ek/1/n (SD-Ej/Ek/1/n) queue to model the
OBS scheme, where Ej and Ek indicate the Erlang distribu-
tions with j and k stages, respectively. The arrival rate of our
SSQ is state dependent because different numbers of ready
stations require different time periods to obtain a successful
RFT/ACK message exchange on the signaling channel6.
The service rate of the SSQ, μ, is given by
μ = 1/Tcycle,
(6)
where the Tcycleis the period of a polling cycle, which can
be expressed as
Tcycle= POLL + SIFS + δ + TDATA+ SIFS + δ. (7)
The balance equation set of the SSQ model is provided
in Appendix I. The Markov Chain state {x,y,z} denotes the
situation that the SSQ has x backlogged stations, and the SSQ
is in the Erlang arrival stage y and Erlang service stage z.
We compute the stationary probabilities of the SD-Ej/Ek/1/n
queue with j = 16 and k = 32. These settings are chosen
based on the study given in [24] and [28].
5While [24] shows the result using simulation study, in [28] we have
performed a comprehensive analytical study on the service characteristics of
the IEEE 802.11 MAC protocol showing that it is possible to find an Erlang
distribution that shares the same statistical characteristics with the service
time distribution of the IEEE 802.11 MAC protocol.
6The duration difference is caused by the employed contention protocol;
generally higher number of stations requires longer time to resolve the
contention as more collisions will occur.

Page 6

TANTRA et al.: OUT-OF-BAND SIGNALING SCHEME FOR HIGH SPEED WIRELESS LANS3261
Let pibe the probability that there are i backlogged stations
in the network, where i = 0,1,...,n. Relating to the SSQ,
pi=?
the SSQ, λ, can be computed by
y,zπi,y,z, where πi,y,z is the steady state probability
of the SSQ being in state {i,y,z}. The mean arrival rate of
λ =
n
?
i=0
(λn−i· pi),
(8)
and the system throughput, γ, can be obtained by
γ = λ · dp,
(9)
where dpis the average size of the payload in a frame.
Let m be the average number of backlogged stations.
Knowing pi, the value m can be computed by
m =
n
?
i=0
(i · pi).
(10)
Using Little’s formula, we calculate the average queuing
delay of a frame on the data channel, Wq, as
Wq= m/λ.
(11)
The average queuing delay corresponds to the time period
between a successful RFT/ACK message exchange and the
frame transmission time. This queuing delay time does not
include the time period of the contention and RFT/ACK
message transmission on the signaling channel.
To compute the transmission delay, we first derive the
average number of ready stations, adjusted for frames dropped
due to retry limit. The mean number of ready stations, E[ns],
is given by
E[ns] =
n
?
i=0
(n − i)[1 − P(pck drop)n−i] · pi,
(12)
where P(pck drop), given in [25], is the probability that a
station will drop its frame due to retry limit; the variable n−i
is the number of ready stations. The probability P(pck drop)
is computed by
P(pck drop) = τ(1 − p)
pR+1
1 − pR+1
R
?
i=0
(1 + E[bi]).
(13)
Having E[ns], the average signaling delay, Wsig, is com-
puted through Little’s formula, which gives
Wsig= E[ns]/λ.
(14)
The mean MAC transmission delay of OBS is simply the sum
of the queueing delay and the signaling delay:
Ws= Wsig+ Wq.
(15)
V. SATURATION PERFORMANCE OF OBS
To illustrate the effectiveness of the OBS scheme, its
performanceis compared with that of the existing IEEE 802.11
MAC protocol. We first report and discuss saturation through-
put and delay results, which are useful to understand the
protocol performance in high load conditions. Then, we follow
up with the optimization of the OBS performance in the same
saturation conditions.
TABLE I
IEEE 802.11A PARAMETERS.
Slot Time
SIFS
DIFS
PIFS
Preamble Length
Propagation Delay (δ)
CWmin
CWmax
Retry Limit (long)
9μs
16μs
34μs
25μs
20μs
1μs
16
1024
7
0
10
20
30
40
50
60
70
80
0 50 100 150 200
Throughput (Mbps)
Channel bit rate (Mbps)
OBS 12 Mbps signaling channel 20 stations
OBS 6 Mbps signaling channel 20 stations
IEEE 802.11 basic access 20 stations
IEEE 802.11 RTS/CTS 20 stations
Analytical results
Fig. 2. Saturation throughput with various data channel bit rate.
Unless otherwise specified, in the rest of this section we use
constant MAC frames each carrying a 1500 bytes payload. As
mentioned in Section III, in the analysis we require OBS to
enforce an RFT/ACK handshake for each single data frame
to be transmitted. We rely on the physical layer parameters
of the IEEE 802.11a [4] (reported for the convenience of the
reader in Table I). As a general graphic notation, the symbols
shown in the figures represent simulation results, while the
solid lines plot analytical results. The analytical results for
the IEEE 802.11 schemes (basic access and RTS/CTS) are
computed according to the model presented in [25].
A. Performance Comparison
Fig. 2 shows the saturation throughput, in Mbps, versus a
varying data channel bit rate employed by the OBS scheme,
ranging from a few Mbps up to more than 200 Mbps. Two
different OBS signaling channel rates, namely 6 and 12 Mbps,
are assessed. In the following, we refer to an OBS scheme
using an x Mbps signaling channel with the synthetic notation
OBSx. For comparison purposes, results for both the DCF
access methods (basic and RTS/CTS) are also reported. The
number of stations in the figure is fixed and set to 20.
From Fig. 2, several important considerations can be drawn.
Firstly, the figure shows an excellent match between our
proposed analytical framework and the simulation results for
the OBS scheme, which confirms the accuracy of our analysis.

Page 7

3262IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 6, NO. 9, SEPTEMBER 2007
0
10
20
30
40
50
60
70
5 10 15 20 25 30 35 40 45 50
Throughput (Mbps)
Number of stations
OBS 12 Mbps signaling channel 1500 bytes
OBS 12 Mbps signaling channel 1000 bytes
IEEE 802.11a basic access 1500 bytes
IEEE 802.11a basic access 1000 bytes
IEEE 802.11a RTS/CTS 1500 bytes
IEEE 802.11a RTS/CTS 1000 bytes
Analytical results
Fig. 3.Saturation throughput with 108 Mbps channel bit rate.
Secondly, Fig. 2 allows the comparison of the OBS per-
formance with that achieved by the 802.11 access methods.
From the figure, we see that OBS12can provide a maximum
saturation throughput of almost 75 Mbps from a 150 Mbps
data channel. For comparison, the IEEE 802.11 basic access
method only achieves 50 Mbps throughput from a 150 Mbps
data channel. Since OBS requires an additional of 12 Mbps
signaling channel, the total usage of the bandwidth is the
bandwidth combination of the two OBS channels, which gives
162 Mbps. If this were the bit rate of the IEEE 802.11 basic
access method, as can be seen from Fig. 2, its throughput
would still be just around 50 Mbps, which remains below the
performance of the OBS scheme. A similar comparison can be
carried out for various data rates reported in the figure, and
shows that OBS performance consistently outperforms both
legacy DCF access methods7. This shows that the use of the
signaling channel in OBS allows a higher utilization in the
data channel. Comparing to the existing IEEE 802.11 schemes,
the throughput gain in the data channel for OBS exceeds the
additional bandwidth required for the signaling channel.
Thirdly, and perhaps more descriptive of the OBS operation,
Fig. 2 shows that, for the same data channel rate, a difference
between the performance of OBS6and OBS12emerges only as
long as the data channel rate increases above a given threshold.
Specifically, when the data channel is below about 100 Mbps,
OBS6 and OBS12 achieve identical throughput performance.
Above this threshold, OBS6 performance flattens out, while
OBS12 performance further increases until the data channel
rate reaches about 150 Mbps. The reason behind this behavior
will be discussed in more details in the next subsection V-B.
In order to assess the OBS performance for a varying
number of stations, and compare it with that achieved by
the legacy 802.11 access methods, Fig. 3 reports the OBS
saturation throughput versus the number of stations, for a data
channel rate set to 108 Mbps (i.e., the double of the current
maximum channel bit rate of IEEE 802.11a and g), and a
signaling channel rate set to 12 Mbps. Although OBS requires
an additional 12 Mbps bandwidth for the signaling channel,
7The only apparent exception is for very low channel data rates, where
clearly a 6 Mbps signaling channel rate reported in the figure is over-
dimensioned with the actual data channel needs.
0
2
4
6
8
10
12
14
5 10 15 20 25 30 35 40 45 50
Delay (ms)
Number of stations
OBS 12 Mbps signaling channel 1500 bytes
OBS 12 Mbps signaling channel 1000 bytes
IEEE 802.11 basic access 1500 bytes
IEEE 802.11 basic access 1000 bytes
IEEE 802.11 RTS/CTS 1500 bytes
IEEE 802.11 RTS/CTS 1000 bytes
Analytical results
Fig. 4. Mean saturation transmission delay with 108 Mbps channel bit rate.
the comparison presented in Fig. 3 nevertheless shows that
OBS throughput advantage exceeds the 12 Mbps employed
for such a signaling channel. Moreover, the performance
advantage is indeed considerable as we recall that the DCF
would not be capable of converting all this extra 12 Mbps
bandwidth available into throughput performance (as clearly
shown and discussed previously in Fig. 2).
Moreover, Fig. 3 shows that OBS, similar to the 802.11
RTS/CTS method and unlike the basic access method, holds
the important property that throughput performance is mar-
ginally dependent on the number of competing stations. This
suggests that, similarly to those demonstrated by literature
work for RTS/CTS [23], OBS is also loosely dependent on
the employed MAC parameter settings, e.g. a too small setting
of the CWmin parameter when a large number of stations
compete does not cause throughput degradation. However,
the figure shows that OBS does not exhibit a decrease in
performance such as what happens for RTS/CTS when the
channel bit rate increases.
Furthermore, Fig. 3 shows the OBS performance for smaller
payloads (1000 bytes). As expected, the performance ad-
vantage decreases with a smaller payload, as the savings
introduced in a reduced collision overhead are lower than the
case of a large frame payload. Nevertheless, the performance
advantage of OBS over legacy DCF remains notable also in
this case.
Finally, we conclude this section by reporting delay per-
formance. Fig. 4 plots the mean MAC transmission delay of
the OBS scheme compared with the IEEE 802.11 schemes
for various numbers of competing stations. With a frame size
of 1000 bytes, OBS maintains a MAC transmission delay
of 8.9 ms even with 50 stations under saturation load, thus
outperformingthe correspondingIEEE 802.11 delays of 15 ms
and 13.4 ms for RTS/CTS and the basic access method,
respectively. The immediate benefit of the lower transmission
delay is a better support for delay-sensitive applications such
as Voice over Internet Protocol (VoIP) or video conferencing8.
8These applications in any case would furthermore dramatically benefit
from the ability to reserve the data channel access once per flow rather than
once per data frame.

Page 8

TANTRA et al.: OUT-OF-BAND SIGNALING SCHEME FOR HIGH SPEED WIRELESS LANS3263
B. OBS Performance Discussion
Fig. 2 suggests that the OBS performance is the result of
a compromise between signaling channel effectiveness and
data channel throughput. In fact, as discussed before, the
performance of the OBS scheme increases steadily at low data
channel bit rate region, it then stays relatively flat at high data
channel bit rate. Most importantly, the throughput value at the
point where OBS performance flattens out heavily depends on
the signaling channel bandwidth.
The intuitive reason behind this behavior can be described
as follows. In the low data channel bit rate region, the
data channel is always fully utilized with data transmission,
and the limit of the throughput is mainly bounded by the
transmission overhead caused by the PCF operation (i.e.,
physical preambles, polls, positive acknowledgment, inter-
frame spaces). Hence with higher bit rate on the data channel,
higher throughput is achieved. At the turning point from the
increasing trend to the flat trend, the successful requests from
the signaling channel are just enough to saturate the data
channel, thus the increment of the data channel bit rate beyond
the turning point does not elevate the throughput performance
further.
This intuitive reason can be more formally supported as
follows. Consider a spectrum that allows a channel to operate
at data rate R. For OBS, this spectrum may be split into
two independent channels operating at Rdataand Rsigwhere
Rdata+ Rsig ≤ R due to the spectral splitting overhead.
Because of the different rates and access rules, the maximum
throughput in frames per second on the two OBS channels
is generally different. Specifically, the signaling channel is
regulated by the DCF operation, where RFT/ACK handshakes
occur instead of the traditional DATA/ACK handshakes.
For the sake of the following discussion, the signaling
channel throughput Ssig can be conveniently approximated,
in frames per second, by (4) computed for a fixed number of
competing stations n (the approximation being the assumption
of a constant n rather than of a variable number of competing
stations which is more correctly done throughout the analysis
presented in Section IV).
Conversely, in the data channel there is a centralized sched-
uler and the maximum throughput Sdata, measured in frames
per second, is equal to
Sdata=
1
HPCF+ F/Rdata,
(16)
where HPCF is the PCF transmission overhead, including the
POLL+ACK frame, the physical overheads and the inter-frame
spaces, and F is the frame size.
The overall OBS data throughput is clearly the minimum
between Ssig and Sdata. In fact, whenever the successful
reservation rate is lower than the data channel throughput,
the OBS throughput is bounded by the limited number of
reserved transmissions. Conversely, when the reservation rate
is greater than the data transmission rate, the latter becomes
the performance limiting factor.
Fig. 5 graphically illustrates the signaling channel through-
put Ssig and the data channel throughput Sdatafor the case
of a fixed total rate R subdivided between data and signaling
channels. Both cases of R = 54 and R = 108 Mbps are
0
1000
2000
3000
4000
5000
6000
7000
0 5 10
OBS signaling channel bit rate Rsig (Mbps)
15 20 25 30 35 40
Throughput (frame/s)
R=108 Mbps
R=54 Mbps
OBS signaling channel Ssig
OBS data channel Sdata
IEEE 802.11 basic access
Fig. 5.
for a given overall bandwidth constraint.
Comparison of the OBS performance versus the DCF performance
considered. Specifically, the figure reports in the x-axis the rate
Rsigassigned to the signaling channel, which results in a rate
Rdata= R − Rsig assigned to the data channel. For sake of
comparison, the figure reports also the throughput, measured
in frames per second, achieved by the DCF basic access
method for both the R = 54 (about 2000 frames/second) and
R = 108 (slightly above 3000 frames/seconds) Mbps channel
rate cases. From the figure, we can draw two important
conclusions. Firstly, the OBS throughput performance reaches
the maximum whenever the data and control channel have the
same throughput. Secondly, for both the R = 54 and R = 108
Mbps channel rate cases, there exists a large region, i.e. a
large amount of signaling channel rate settings, where the OBS
performance outperforms DCF (we remark that, unlike Fig. 2,
this figure has been obtained in the assumption that the same
total rate R is available to both OBS and DCF schemes).
C. Signaling Channel Optimization
The throughput results plotted in Fig. 2 have shown that
there exist a turning point in the OBS throughput performance
such that OBS operates at its maximum throughput given
a suitable pair of bit rates of signaling and data channels.
From the discussion carried out in the previous section as
well as from the related Fig. 5, it is now clear that such
an optimum operation point is achieved when the signaling
channel throughput equals to the data channel rate throughput.
To provide an exact computation (unlike the approximated
treatment carried out in the previous descriptive section) of
such an optimum operational point, it suffices to set the
service rate of the RFT/ACK signaling channel handshake,
λ as computed in (8), to be equal to the service rate μ
of the data channel, provided by (6). More specifically, the
quantity λ depends on the bit rate of the signaling channel
and on the number of stations contending on the signaling
channel, whereas the quantity μ depends on the bit rate of the
data channel and the frame size. To be precise, the signaling
channel bit rate affects RFT in (5) which in turn affects (4)
and λ in (8). Data channel bit rate influences TDATAwhich
in turn influences μ in (6). For a given data channel rate,
the optimal signaling channel rate which maximizes the OBS

Page 9

3264IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 6, NO. 9, SEPTEMBER 2007
0
20
40
60
80
100
120
60 80 100 120 140
Signaling channel bit rate (Mbps)
Data channel bit rate (Mbps)
30 stations 1000 bytes
10 stations 1000 bytes
30 stations 1500 bytes
10 stations 1500 bytes
Fig. 6. Minimum signaling channel bit rate required to achieve full utilization
on the data channel.
throughput is readily computed by imposing the condition
λ = μ.
In Fig. 6, we show the bit rate pairs of signaling and data
channels required to optimize the OBS operation. As can be
seen, for the considered frame sizes, OBS requires relatively
low bit rate signaling channel to fully utilize the data channel
which is below 100 Mbps. Our particular interest here is the
data channel of 108 Mbps bit rate. It is found that the signaling
channel bit rate between 6 Mbps and 20 Mbps appears to be
a good choice for commonly used frame sizes, and hence we
consider 12 Mbps for the signaling channel bit rate for studies
conducted in the previous subsections.
VI. ANALYSIS OF THE OUT-OF-BAND SIGNALING
SCHEME UNDER STATISTICAL TRAFFIC
The performance study of a protocol under saturated traffic
load indicates the protocol performance under the extremely
heavy load condition. However, the protocol often operates
under non-saturated load conditions. It is thus important
to also evaluate the performance of a protocol under non-
saturation traffic, and here we focus on statistical traffic.
There are several approaches in modeling the IEEE 802.11
MAC protocol in the literature [24], [28]–[31]. The approach
we consider for OBS performance study under non-saturation
condition is the Markovian Framework presented in Sec-
tion IV. We extend the Markovian Framework to capture the
characteristics of statistical arrivals, and analyze the through-
put and delay performance of OBS under the considered
traffic. All analytical results are compared and validated by
simulation.
A. Markovian Framework for Out-of-Band Signaling Scheme
under Statistical Traffic
We consider Poisson process with one frame buffer as
arrival process for each station. In other words, each station
holds only a frame in its local MAC buffer. Furthermore, we
assume that the stations are statistically identical and indepen-
dent. We reuse the definitions of ready station, backlogged
station, and idle station, which are defined in Section IV.
Under Poisson arrival, an idle station switches to a ready
station when it generates a frame into its local buffer according
to a Poisson process. A ready station switches to a backlogged
station when it has successfully exchanged the RTS/CTS
message with the AP. Under statistical traffic and one buffer
assumption, a backlogged station switches to an idle station
after its successful frame transmission on the data channel.
Applying Markovian Framework, we characterize the sys-
tem with two queues in tandem. The first queue models the
stage that a station switches from an idle station to a ready
station indicating an arrival to the queue, and from a ready
station to a backlogged station indicating a departure from
the queue. The arrivals to this queue is an aggregation of
arrivals from all idle stations, where the inter-arrival time
has exponential distribution, and the aggregated arrival rate
depends on the number of idle stations. The service process
of the queue is the RFT/ACK frame exchange process, which,
as described in the previous section, can be modeled by an
Erlang distribution.
The departure of the first queue goes into the second queue.
The second queue models the actual data transmission stage,
that is, a backlogged station switching back to a ready station
after its successful frame transmission. The arrival process of
the second queue is the output process of the first queue, which
has Erlang distribution. The service process of the second
queue depends on the data frame size. For the constant frame
size assumption, an Erlang distribution for service time is used
as in the previous section.
Let the total number of stations in the network be n. Define
v to be the number of ready stations, then the number of idle
stations is n−v. The arrival rate λn−vis the state-dependent
Poisson arrival process, which is given by
λn−v= (n − v) ·ˆλ,
whereˆλ is the individual arrival rate of a station.
The service rate of the first queue, μ1, is the rate of a
successful RFT/ACK message exchange when there are v
ready stations, which is given by (4). The service rate of
the second queue, μ2, is the rate of frame transmission with
polling scheme and constant frame size given by (6).
The balance equations set of the system is provided in
Appendix II. The Markov Chain state {v,x,y,z} denotes the
situation that the system has v ready stations and x backlogged
stations, and the system is in the Erlang service stages y and
z of the first and second queues respectively. Similar with
the previous section, we compute the stationary probability of
the queue with j = 16 (the number of Erlang stages for the
first queue), and k = 32 (the number of Erlang stages for the
second queue).
Let pv,x be the probability that there are v ready sta-
tions and x backlogged stations in the network, where v =
0,1,...,n and x = 0,1,...,v. Relating to the system, pv,x=
?
the system, λ, can be computed by
?
(17)
y,zπv,x,y,z, where πv,x,y,zis the steady state probability of
the system being in state {v,x,y,z}. The mean arrival rate of
λ =
n
?
v=0
λn−v·
v
?
x=0
pv,x
?
,
(18)

Page 10

TANTRA et al.: OUT-OF-BAND SIGNALING SCHEME FOR HIGH SPEED WIRELESS LANS3265
and the system throughput, γ, can be obtained by
γ = λ · dp,
(19)
where dpis the average size of the payload in a frame.
Let m be the average number of backlogged stations. The
value m can be computed by
m =
n
?
v=0
v
?
x=0
(x · pv,x).
(20)
The computation of the mean MAC transmission delay is
similar with that in the previous section (Equations (11)–(15)).
Given the number of backlogged stations, m, and the mean
arrival rate, λ, by Little’s formula, we calculate the average
queuing delay of a frame transmission, Wq, as
Wq= m/λ.
(21)
The mean number of ready stations is given by
E[ns] =
n
?
v=0
v
?
x=0
(v − x)[1 − P(pck drop)v−x] · pv,x. (22)
The mean signaling delay can be determined using Little’s
formula as
Wsig= E[ns]/λ.
(23)
The mean MAC transmission delay is simply the sum of
the queuing delay and the signaling delay, which is
Ws= Wsig+ Wq.
(24)
B. Performance Comparison
Fig. 7 shows the mean transmission delay of the OBS
scheme compared with the IEEE 802.11 schemes under sta-
tistical traffic. We use 12 Mbps signaling channel for the
OBS scheme as justified in the previous section. Similarly, we
use 108 Mbps data channel for the comparison. The symbols
shown in the figures represent the simulation results, while the
solid lines show the analytical results. Analytical results for
the IEEE 802.11 schemes are computed using the model in
[24], [28]. The results show that OBS has lower delay under
medium to high load comparedwith the IEEE 802.11 schemes.
Under very light load condition, all schemes perform similarly
well where they offer mean MAC transmission delay of below
0.5 ms.
The throughput versus mean transmission delay of the
OBS scheme is plotted in Fig. 8 and compared with the
IEEE 802.11 schemes. OBS can achieve 57% throughputlevel,
whereas IEEE 802.11 with basic access and RTS/CTS methods
achieve 41% and 34% throughput level respectively. OBS also
maintains delays lower than 1.6 ms, whereas IEEE 802.11 has
delays up to 2.4 ms and 2.9 ms for basic access and RTS/CTS
methods, respectively.
The results reported in the two figures verify the accuracy
of our analytical framework. The figures show close match
between the analysis and the simulation results. This analytical
framework is useful for determining the number of application
streams that can be supported by an access point (see for
example [32]). It is also useful for admission control for
quality of service demanding applications such as VoIP and
video streaming.
0
0.5
1
1.5
2
2.5
3
0 500 1000
Arrival rate (packets/second)
1500 2000 2500 3000
Delay (ms)
OBS 10 stations 1500 bytes
OBS 10 stations 1000 bytes
IEEE 802.11 basic access 10 stations 1500 bytes
IEEE 802.11 basic access 10 stations 1000 bytes
IEEE 802.11 RTS/CTS 10 stations 1500 bytes
IEEE 802.11 RTS/CTS 10 stations 1000 bytes
Analytical results
Fig. 7.Mean transmission delay under statistical traffic.
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
0 0.1 0.2 0.3 0.4 0.5 0.6
Delay (ms)
Throughput
OBS 10 stations 1500 bytes
OBS 10 stations 1000 bytes
IEEE 802.11 basic access 10 stations 1500 bytes
IEEE 802.11 basic access 10 stations 1000 bytes
IEEE 802.11 RTS/CTS 10 stations 1500 bytes
IEEE 802.11 RTS/CTS 10 stations 1000 bytes
Analytical results
Fig. 8. Throughput versus mean transmission delay under statistical traffic.
VII. CONCLUSION
In this paper, we have proposed and analyzed the OBS
scheme for high speed WLANs. OBS uses a low bit rate
channel for signaling and a high bit rate channel for the
actual data transmission. We showed that the use of out-of-
band signaling technique achieves higher overall throughput
despite the need for an additional low bit rate channel for
signaling. Besides, OBS maintains backward compatibility
with the existing users of the IEEE 802.11 standards, where
users of the IEEE 802.11 standards can access OBS WLANs,
however, they will not enjoy the performance benefit.
To illustrate the performance advantage of OBS, we applied
Markovian Framework to study the throughput and delay
performance under the saturation load and the statistical traffic
conditions. The analytical results, validated by simulation,
indicated performance advantages of OBS compared with
the current IEEE 802.11 schemes. We also investigated the
channel bit rate settings for optimal performance in OBS,
where we limit our study to the issue of protocol parameter
design for the justification of OBS parameter selection.

Page 11

3266IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 6, NO. 9, SEPTEMBER 2007
APPENDIX I
BALANCE EQUATIONS OF
OBS UNDER SATURATION CONDITION
For simplicity of formulation, let the invalid states of the
Markov Chain model, i.e. state {x,y,z} outside the range
of 0 ≤ x ≤ n, 0 ≤ y < j, and 0 ≤ z < k, has stationary
probability of zero. We also assume that the arrival and service
rates are zeros for out-of-range parameters i.e. the arrival rate
is non-zero if 0 ≤ x < n and the service rate is non-zero if
0 < x ≤ n. The balance equation set for the SSQ is given by
0 =
−
+
jλn−xPx,y−1,z+ kμ(x)Px,y,z−1
0 =
−
+
jλn−x−1Px−1,j−1,z+ kμ(x)Px,0,z−1
0 =
−
+
jλn−xPx,y−1,0+ kμ(x + 1)Px+1,y,k−1
0 =
−
+
jλn−x−1Px−1,j−1,0
+
kμ(x + 1)Px+1,0,k−1.
?
?
?
?
jλn−x+ kμ(x)
?
?
?
?
Px,y,z
jλn−x+ kμ(x)
Px,0,z
jλn−x+ kμ(x)
Px,y,0
jλn−x+ kμ(x)
Px,0,0
APPENDIX II
BALANCE EQUATIONS OF
OBS UNDER STATISTICAL TRAFFIC
Similar with the previous section, let the invalid states of the
Markov Chain model, i.e. state {v,x,y,z} outside the range
of 0 ≤ v ≤ n, 0 ≤ x ≤ v, 0 ≤ y < j, and 0 ≤ z <
k, has stationary probability of zero. Assume that the arrival
and service rates are zeros for out-of-range parameters i.e. the
arrival rate λ is non-zero if 0 ≤ v < n, the service rate μ1is
non-zero if 0 ≤ x < v, and the service rate μ2is non-zero if
0 < x ≤ v. The balance equation set for the system is given
by
?
+
λv−1Pv−1,x,y,z
+
jμ1(v − x)Pv,x,y−1,z+ kμ2(x)Pv,x,y,z−1
0 =
−
+
λv−1Pv−1,x,0,z
+
jμ1(v − x − 1)Pv,x−1,j−1,z+ kμ2(x)Pv,x,0,z−1
0 =
−
+
λv−1Pv−1,x,y,0
+
jμ1(v − x)Pv,x,y−1,0+ kμ2(x + 1)Pv+1,x+1,y,k−1
0 =
−
+
λv−1Pv−1,x,0,0
+
jμ1(v − x − 1)Pv,x−1,j−1,0
+
kμ2(x + 1)Pv+1,x+1,0,k−1.
0 =
−
λn−v+ jμ1(v − x) + kμ2(x)
?
Pv,x,y,z
?
λn−v+ jμ1(v − x) + kμ2(x)
?
Pv,x,0,z
?
λn−v+ jμ1(v − x) + kμ2(x)
?
Pv,x,y,0
?
λn−v+ jμ1(v − x) + kμ2(x)
?
Pv,x,0,0
REFERENCES
[1] IEEE Std. 802.11, “Wireless LAN medium access control (MAC) and
physical layer specification,” 1997.
[2] IEEE Std. 802.11b, “Part 11: Wireless LAN medium access control
(MAC) and physical layer (PHY) specifications: high-speed physical
layer extension in the 2.4 GHz band,” Sept. 1999.
[3] IEEE Std. 802.11g, “Part 11: wireless LAN medium access control
(MAC) and physical layer (PHY) specifications amendment 4: further
higher data rate extension in the 2.4 GHz band,” June 2003.
[4] IEEE Std. 802.11a, “Part 11: wireless lan medium access control (MAC)
and physical layer (PHY) specifications: high-speed physical layer in the
5 GHz band,” Sept. 1999.
[5] IEEE Draft Std. 802.11n, “Part 11: wireless LAN medium access
control (MAC) and physical layer (PHY) specifications: enhancements
for higher throughput.”
[6] Y. Xiao and J. Rosdahl, “Throughput and delay limits of IEEE 802.11,”
IEEE Commun. Lett., vol. 6, no. 8, pp. 355–357, Aug. 2002.
[7] F. Cali, M. Conti, and E. Gregori, “IEEE 802.11 protocol: design and
performance evaluation of an adaptive backoff mechanism,” IEEE J. Sel.
Areas Commun., vol. 18, no. 9, pp. 1774–1786, Sept. 2000.
[8] C. Wang, B. Li, and L. Li, “A new collision resolution mechanism
to enhance the performance of IEEE 802.11 DCF,” IEEE Trans. Veh.
Technol., vol. 53, no. 4, pp. 1235–1246, July 2004.
[9] J. Choi, J. Yoo, S. Choi, and C. Kim, “EBA: an enhancement of the IEEE
802.11 DCF via distributed reservation,” IEEE Trans. Mobile Comput.,
vol. 4, no. 4, pp. 378–390, July-Aug. 2005.
[10] Y. Xiao, “Packing mechanisms for the IEEE 802.11n wireless LANs,”
in Proc. IEEE Globecom, Nov. 2004.
[11] IEEE Std. 802.11e, “Supplement to part 11: wireless medium access
control (MAC) and physical layer specification: medium access control
(MAC) enhancements for quality of service (QoS),” 2005.
[12] C. Liu and A. Stephens, “An analytic model for infrastructure WLAN
capacity with bidirectional frame aggregation,” in Proc. IEEE WCNC
2005.
[13] A. Ganz, A. Phonphoem, and Z. Ganz, “Robust superpoll with chaining
protocol for IEEE 802.11 wireless LANs in support of multimedia
applications,” Wireless Networks, vol. 7, no. 1, pp. 65–73, Jan. 2001.
[14] S.-C. Lo, G. Lee, and W.-T. Chen, “An efficient multipolling mechanism
for IEEE 802.11 wireless LANs,” IEEE Trans. Comput., vol. 52, no. 6,
pp. 764–778, June 2003.
[15] T. M. Lim, J. W. Tantra, C. H. Foh, and B. S. Lee, “Out-of-band polling
scheme for QoS support in wireless LANs,” International J. Commun.
Syst., special issue on QoS support and service differentiation in wireless
networks, vol. 17, no. 6, pp. 643–661, Aug. 2004.
[16] Y. Utsunomiya, T. Tandai, T. Adachi, and M. Takagi, “A MAC protocol
for coexistence between 20/40 MHz STAs for high throughput WLAN,”
in Proc. IEEE VTC-Spring 2006.
[17] A. Nasipuri, J. Zhuang, and S. R. Das., “A multichannel CSMA MAC
protocol for multihop wireless networks,” in Proc. IEEE WCNC, Sept.
1999.
[18] S.-L. Wu, Y.-C. Tseng, C.-Y. Lin, and J.-P. Sheu, “A multi-channel MAC
protocol with power control for multi-hop mobile ad hoc networks,”
Computer Journal, vol. 45, no. 1, pp. 101–110, 2002.
[19] J. So and N. H. Vaidya, “Multi-channel MAC for ad hoc networks:
handling multi-channel hidden terminals using a single transceiver,” in
Proc. ACM MobiHoc, May 2004.
[20] J. W. Tantra, C. H. Foh, and B. S. Lee, “An efficient scheduling scheme
for high speed IEEE 802.11 WLANs,” in Proc. IEEE VTC, Oct. 2003.
[21] J. W. Tantra, C. H. Foh, G. Bianchi, and I. Tinnirello, “Performance
analysis of the out-of-band signaling scheme for high speed wireless
LANs,” in Proc. IEEE Globecom, Nov. 2004.
[22] M. Shreedhar and G. Varghese, “Efficient fair queuing using deficit
round-robin,” IEEE/ACM Trans. Networking, vol. 4, no. 3, pp. 375–385,
June 1996.
[23] G. Bianchi, “Performance analysis of the IEEE 802.11 distributed
coordination function,” IEEE J. Sel. Areas Commun., vol. 18, no. 3,
pp. 535–547, Mar. 2000.
[24] C. H. Foh and M. Zukerman, “Performance analysis of the IEEE 802.11
MAC protocol,” in Proc. European Wireless, Feb. 2002.
[25] G. Bianchi and I. Tinnirello, “Remarks on IEEE 802.11 DCF perfor-
mance analysis,” IEEE Commun. Lett., vol. 9, no. 8, pp. 765–767, Aug.
2005.
[26] H. Wu, Y. Peng, K. Long, S. Cheng, and J. Ma, “Performance of
reliable transport protocol over IEEE 802.11 wireless LAN: analysis
and enhancement,” in Proc. IEEE INFOCOM, June 2002.
[27] C. H. Foh and J. W. Tantra, “Comments on IEEE 802.11 saturation
throughput analysis with freezing of backoff counters,” IEEE Commun.
Lett., vol. 9, no. 2, pp. 130–132, Feb. 2005.
[28] C. H. Foh, M. Zukerman, and J. W. Tantra, “A Markovian framework
for performance evaluation of IEEE 802.11,” IEEE Trans. Wireless
Commun., vol. 6, no. 4, pp. 1276–1285, Apr. 2007.

Page 12

TANTRA et al.: OUT-OF-BAND SIGNALING SCHEME FOR HIGH SPEED WIRELESS LANS3267
[29] K. Duffy, D. Malone, and D. J. Leith, “Modeling the 802.11 distributed
coordination function in non-saturated conditions,” IEEE Commun. Lett.,
vol. 9, no. 8, pp. 715–717, Aug. 2005.
[30] G.-R. Cantieni, Q. Ni, C. Barakat, and T. Turletti, “Performance analysis
of finite load sources in 802.11b multirate environments,” Computer
Commun., vol. 28, no. 10, pp. 1095–1109, June 2005.
[31] O. Tickoo and B. Sikdar, “A queueing model for finite load IEEE 802.11
random access MAC,” in Proc. IEEE ICC, June 2004.
[32] P. Patil and V. Apte, “Sizing of IEEE 802.11 wireless LANs,” in Proc.
ACM WMASH, Sept. 2005.
Juki Wirawan Tantra (S’02-M’07) received his
BEng (Hons) and Ph.D. degrees from the School
of Computer Engineering, Nanyang Technological
University, Singapore in 2003 and 2007, respec-
tively. His main research interests are wireless net-
works, network QoS, and performance modeling.
Chuan Heng Foh (S’00-M’03) received his B.S. de-
gree in electronic engineering from Fu Jen Catholic
University, Taipei Hsien, Taiwan, R.O.C., in 1992,
the M.S. degree from Monash University, Victoria,
Australia, in 1999, and the Ph.D. degree from the
University of Melbourne, Melbourne, Australia, in
2002.
From July 2002 to December 2002, he was a Lec-
turer at Monash University. He is now an Assistant
Professor with the School of Computer Engineering,
Nanyang Technological University, Singapore. His
research interests include protocol design and performance analysis of mobile
wireless and optical networks.
Ilenia Tinnirello has been Assistant Professor at the
University of Palermo since January 2005. She re-
ceived the Laurea degree in Electronic Engineering
and the Ph.D. on Communications, respectively in
April 2000 and February 2004. She has also been
Visiting Researcher at the Seoul National University,
Korea, in 2004, and at the Nanyang Technological
University of Singapore in 2006. Her research ac-
tivity has been mainly focused on wireless networks
and in particular on: multiple access algorithms with
quality of service provisioning; cross-layer interac-
tions between access solutions and physical layer; mobility management and
load balancing in wireless packet networks.
Giuseppe Bianchi received the ”Laurea” degree in
Electronic Engineering from Politecnico di Milano,
in 1990, and a specialization degree in Informa-
tion Technology from CEFRIEL, Milano, Italy, in
1991. He has been research consultant for CEFRIEL
(1991-1993), Assistant Professor at the Politecnico
di Milano (1993-1998), Associate Professor at the
University of Palermo (1998-2003) and Associate
professor at the University of Roma Tor Vergata
(2004-2006) where, since January 2007, he became
Full Professor. He spent 1992 as visiting researcher
at the Washington University of St. Louis, MO, USA, and 1997 as visiting
professor at the Columbia University of New York, NY, USA. His research
interests include WLAN/WMAN systems, privacy enhancing technologies,
and performance evaluation.