Quality-of-Service Provisioning System
for Multimedia Transmission in
IEEE 802.11 Wireless LANs
Der-Jiunn Deng and Hsu-Chun Yen
Abstract—IEEE 802.11, the standard of wireless local area networks (WLANs), allows the coexistence of asynchronous and time-bounded traffic using the distributed coordination func-tion (DCF) and point coordinafunc-tion funcfunc-tion (PCF) modes of operations, respectively. In spite of its increasing popularity in real-world applications, the protocol suffers from the lack of any priority and access control policy to cope with various types of multimedia traffic, as well as user mobility. To expand support for applications with quality-of-service (QoS) requirements, the 802.11E task group was formed to enhance the original IEEE 802.11 medium access control (MAC) protocol. However, the problem of choosing the right set of MAC parameters and QoS mechanism to provide predictable QoS in IEEE 802.11 networks remains unsolved. In this paper, we propose a polling with non-preemptive priority-based access control scheme for the IEEE 802.11 protocol. Under such a scheme, modifying the DCF access method in the contention period supports multiple levels of pri-orities such that user handoff calls can be supported in wireless LANs. The proposed transmit-permission policy and adaptive bandwidth allocation scheme derive sufficient conditions such that all the time-bounded traffic sources satisfy their time con-straints to provide various QoS guarantees in the contention free period, while maintaining efficient bandwidth utilization at the same time. In addition, our proposed scheme is provably optimal for voice traffic in that it gives minimum average waiting time for voice packets. In addition to theoretical analysis, simulations are conducted to evaluate the performance of the proposed scheme. As it turns out, our design indeed provides a good performance in the IEEE 802.11 WLAN’s environment, and can be easily incorporated into the hybrid coordination function (HCF) access scheme in the IEEE 802.11e standard.
Index Terms—Carrier sense multiple access/collision avoidance (CSMA/CA), point coordination function (PCF), quality-of-service (QoS), 802.11, wireless local area networks (WLAN).
I. INTRODUCTION
F
LEXIBILITY and mobility have made wireless local area networks (WLANs) a rapidly emerging field of activity in computer networking, attracting significant inter-ests in the communities of academia and industry [1]–[6]. In the meantime, the IEEE standard for WLANs, IEEE 802.11 [1], has gained global acceptance and popularity in wireless computer networking markets and has also been anticipated toManuscript received January 31, 2004; revised December 1, 2004. D.-J. Deng is with the Department of Electrical Engineering, National Taiwan University, Taipei 106, Taiwan and also with the Department of Information Management, Overseas Chinese Institute of Technology, Taichung 407, Taiwan (e-mail: djdeng@cobra.ee.ntu.edu.tw).
H.-C. Yen is with the Department of Electrical Engineering, National Taiwan University, Taipei 106, Taiwan (e-mail: yen@cc.ee.ntu.edu.tw).
Digital Object Identifier 10.1109/JSAC.2005.845632
continue being the preferred standard for supporting WLAN’s applications.
In WLANs, the medium access control (MAC) protocol is the key component that provides the efficiency in sharing the common radio channel, while satisfying the quality-of-service (QoS) requirements of various multimedia traffic. That is, MAC protocols that aim to carry multimedia traffic must be able to meet a variety of requirements for a wide range of traffic classes. However, frames in distributed coordination function (DCF), the basic access method in the IEEE 802.11 MAC layer protocol, do not have priorities, and there is no other mecha-nism to enforce a guaranteed access delay bound. As a result, real-time applications such as voice or live video transmissions may suffer from unacceptable delay with this protocol. The second access mode of the IEEE 802.11 MAC layer protocol, point coordination function (PCF), offers a “packet-switched connection-oriented” service, which is well suited for real-time traffic. However, in order to poll the stations an access point (AP) must maintain a polling list, which is implementation de-pendent. What this means is that end-to-end QoS requirements still cannot be satisfied in this scheme since it does not include any access control policy. Besides, it does not include any pri-ority scheme to support handoff management, nor does it apply any bandwidth allocation strategy for handoff calls. However, packet-switched solutions, taking advantage of silences in a given voice call by multiplexing voice data from other calls, are more bandwidth-efficient than circuit-switched solutions. Furthermore, digitized multimedia traffic can be compressed to prevent tremendous bandwidth consuming comparing to uncompressed audio or video traffic. Since the demand for the use of packet-switched techniques for transferring delay-sensi-tive data in wireless environments is inevitable for multimedia applications, several works [7]–[31] have been investigated and discussed along this line of research. Accordingly, the IEEE 802.11 working group is currently working on a new standard called 802.11e [32] to enhance the original 802.11 MAC sublayer to support applications with QoS requirements.
Although the 802.11e standard is for providing QoS support for WLAN applications, the problem of choosing the right set of MAC parameters and QoS mechanism to provide predictable QoS in 802.11 networks remains unsolved [33]. Besides, the process of creating a definitive standard might be too slow for us waiting for it to be ratified. Hence, we implement a QoS pro-visioning system for multimedia transmission in IEEE 802.11 Wireless LANs, with a view to cutting, to a great extent, the telecommunication costs incurred to international enterprises. 0733-8716/$20.00 © 2005 IEEE
Fig. 1. System architecture.
In this paper, we propose an advanced, pragmatic, and yet more complete polling with nonpreemptive priority-based access con-trol scheme for the IEEE 802.11 protocol. Our primary contri-butions are as follows: Under such a scheme, by modifying the DCF access method in the contention period, the protocol offers multiple levels of priorities such that handoff calls can be sup-ported in wireless LANs. Besides, the proposed transmit-per-mission policy and adaptive bandwidth allocation scheme not only separate admitted inactivated users from newly requesting access users, but also derive sufficient conditions such that all the time-bounded traffic sources satisfy their time constraints to provide various QoS guarantees in the contention free period while maintaining efficient bandwidth utilization at the same time. Furthermore, the proposed scheduling algorithm for voice traffic is provably optimal in that it gives the minimum average waiting time for voice packets.
The proposed scheme is performed at each AP in a distributed manner. Such a scheme can be implemented in a broad class of algorithms with relatively minor modifications. In addition to theoretical analysis, simulations are conducted to evaluate the performance of the proposed scheme for integrated traffic. As it turns out, our design indeed provides good performance improvements over the original IEEE 802.11 protocol or the upcoming IEEE 802.11e standard.
The remainder of this paper is organized as follows. Section II describes the proposed scheme in detail. Simulation and exper-imental results are reported in Section III. Section IV concludes this paper.
II. PROPOSEDSCHEME
In this section, we present the proposed scheme in detail. Fig. 1 shows an overview of the proposed system architecture.
Our method involves three basic components: 1) a priority en-forcement mechanism for request access; 2) a packet transmit-permission policy;, and 3) an adaptive bandwidth management strategy.
A. Priority Enforcement Mechanism for Request Access Since a mobile device travels while a connection is alive, the QoS might degrade because of some physical constraints. The problem will become even more challenging because recent wireless networks have been implemented using architecture based on small-size cells (i.e., microcells or picocells) to obtain higher transmission capacity and to achieve better performance. In most of the solutions, bandwidth is reserved for handoff mo-biles in advance to reduce the dropping probability [35]. Some improvements have also been discussed [36]. However, when reserved and unused, the bandwidth is simply wasted. This is where priority schemes come in. The prioritized medium access of the EDCF in IEEE 802.11e is to provide service differenti-ation by allowing faster access to the channel to traffic classes with higher priority. Faster access can be provided by allocating a smaller contention window (CW) or a smaller interframe space (IFS). However, differentiating the initial CW size is better than differentiating the IFS in terms of total throughput and delay [27]. The reason is that the different initial CW size has both the function of reducing collisions and providing priorities, whereas the arbitration IFS has the function of providing priorities, but can not reduce collisions. In this section, we propose a novel method to modify the DCF access method to get many levels of priorities, capable of giving handoff requests higher priority over new connection requests. Under such a scheme, a high pri-ority station is entitled to a shorter waiting time when accessing the medium. Furthermore, when a collision occurs, a high pri-ority station can also take precedence over its lower pripri-ority counterparts in accessing the medium. Besides, the proposed
adaptive CW mechanism can dynamically expand and contract the CW size according to the current channel status. The method is simple, efficient, flexible, and scalable, based only on carrier sensing which can easily be implemented. It could be used as the random access protocol for contention period within a super-frame in the IEEE 802.11 protocol without requiring any com-plicated computation or additional hardware support.
The core of the underlying collision avoidance mecha-nism in carrier sense multiple access with collision avoidance (CSMA/CA) is built upon a random backoff procedure, which generates a random backoff time (an integer value corre-sponding to the number of time slots). Initially, a station computes a backoff time in the range of 0–7. If a station with a frame to transmit initially senses a busy channel, it waits until the channel becomes idle, and then, the station decrements its backoff timer until either the medium becomes busy again or the timer reaches zero. If the medium becomes busy before the timer reaches zero, the station freezes its timer. When the timer finally decrements to zero, the station transmits its frame. If two or more stations decrement to zero at the same time, a collision occurs, and each station will have to generate a new backoff time in the range of 0–15. For each retransmission attempt, the
backoff time grows in the form of ,
where is the number of consecutive times a station attempts to send a frame, is a uniform variate in (0, 1). (Here, represents the largest integer less than or equal to .) For more about the CSMA/CA and the IEEE 802.11 protocols, the reader is referred to [34].
The basic idea behind our method is that prioritized ac-cess to the wireless medium is controlled through different backoff time periods. To this end, instead of using the one defined in [1], we change the backoff time generation
func-tion to for high priority stations and
for low priority stations. That is, the random backoff time is divided into two parts:
and , which are
used by the high priority stations and the low priority ones, respectively. It is clear that the shorter backoff time a station waits, the higher priority this station will get. For example, initially (i.e., ) the high priority stations generate a backoff time in the range of 0–3, and the low priority stations generate a backoff time in the range of 4–7. The former clearly has the edge on the latter in contending the channel. Such an idea can easily be generalized to support multiple-level priorities. Once again, the backoff time generation function is refined as , where is the level of priority. Within a fixed backoff range, the probability of collisions with respect to the same priority level will increase if the number of contended stations increases. To offer a higher degree of flexibility and expandability, it is desirable that the scheme be able to expand or contract the backoff range arbitrarily. To be more precise, we allow different backoff ranges for different priority levels in our scheme by changing the backoff time
generating function to , where is
the level of priority, and and are the parameters used to decide the number of slots in individual priority levels and the number of slots between each priority levels, respectively. In this paper, the real-time handoff traffic requests have the highest
TABLE I
EXAMPLES OFBACKOFFTIME OFINDIVIDUALTRAFFIC
priority among all other requests, and the second priority class is the admitted inactivated video traffic. The new requests and pure data traffic will reside on the lowest priority level, as illustrated in Table I. Note that we give a wider range to the lowest priority level since this type of traffic is likely to be heavier, in comparison with the other two traffic classes. It is important to note that when a station decrements its backoff timer and the medium becomes busy, the station freezes its timer, meaning that a station will raise its priority automati-cally after several transmission failures. Hence, the proposed mechanism is clearly starvation-free.
The collision avoidance portion of CSMA/CA is performed by a variable time-spreading of the users’ access. However, collisions still occur if two or more stations select the same backoff slot. When this happens, these stations have to reenter the competition with an exponentially increasing CW param-eter value, and the increase of the CW paramparam-eter value after collisions is the mechanism provided by CSMA/CA to make the access adaptive to channel conditions. This strategy avoids long access delays when the load is light because it selects an initial (small) parameter value of CW by assuming a low level of congestion in the system. However, it incurs a high colli-sion probability and channel utilization is degraded in bursty arrival or congested scenarios. In other words, this strategy might allocate initial size of CW, only to find out later that it is not enough when the load increases. The size of CW must be reallocated with a larger size, but each increase of the CW parameter value is obtained paying the cost of a collision (bandwidth wastage). Furthermore, after a successful trans-mission, the size of CW is set again to the minimum value without maintaining any knowledge of the current channel status. Besides, the performance of CSMA/CA access method will be severely degraded not only in congested scenarios but also when the bit-error rate (BER) increases in the wireless channel. One principal problem also comes from the backoff algorithm. In the CSMA/CA access method, immediate positive acknowledgment informs the sender of successful reception of each data frame. This is accomplished by the receiver initiating the transmission of an acknowledgment frame after a small time interval, SIFS, immediately following the reception of the data frame. In case an acknowledgment is not received, as we mentioned above, the sender will presume that the data frame is lost due to collision, not by frame loss. Consequently, when a timer goes off, it exponentially increases backoff pa-rameter value and retransmits the data frame less vigorously.
Unfortunately, wireless transmission links are noisy and highly unreliable. The proper approach to dealing with lost frames is to send them again, and as quickly as possible. Extending the backoff time just makes the matter worse because it brings bandwidth wastage.
As mentioned earlier, our scheme has the ability to expand or contract the backoff range arbitrarily by changing the parame-ters and . Hence, we propose an adaptive CW mechanism for our scheme to dynamically expand and contract the CW size according to the current channel status and achieve the theoret-ical capacity limits. This scheme is based on the results of the capacity analysis model of the IEEE 802.11 protocol originally proposed in [37]–[40], as well as the concept introduced in [41]. However, our scheme is simpler and more efficient and accurate, and it does not suffer from the problem of harmful fluctuation reported in [42].
In order to exploit the early and meaningful information about the actual congestion status of a channel, we start by defining the utilization factor of a CW for real-time handoff traffic to be the number of busy slots observed in the first to
slots divided by the size (number of slots) of the current CW for admitted voice traffic in the latest CW. The uti-lization factor, , for admitted inactivated video traffic is de-fined by the number of busy slots observed in the to slots divided by the size of the current CW for admitted inactivated video traffic in the latest CW. The utilization factor for new request and pure data traffic is de-fined to be the number of busy slots observed in the
to slots divided by the size of the current
CW for new request and pure data traffic in the latest CW. Note that the level of priority, , for voice, video and new request/pure data traffic is 0, 1, and 2, respectively, here.
The arguments of how to adjust the CW size to achieve the theoretical capacity limits for these three classes of traffic are essentially the same. Hence, we assume that only one kind of traffic exists in the following argument. Note, however, that the parameters and of different traffic should be adjusted at the same time while changing any one of these parameters.
In practice, the value of has to be updated in every backoff interval to reflect the actual state of the channel. Assume that there are stations working in asymptotic conditions in the system. This means that the transmission queue of each station is assumed to be always nonempty. The stations transmit frames whose sizes are independent and identically distributed (i.i.d.) sampled from a geometric distribution with parameter . Specif-ically, the size of a frame is an integer multiple of the slot size,
and, hence, the mean frame space is .
Let , and denote the average frame
transmission time, the average temporal distance between two consecutive successful transmission, and the average time required for a successful transmission, respectively. Hence, the
protocol capacity is . Also, from the geometric
backoff assumption, all the processes which define the occu-pancy pattern of the channel are regenerative with respect to the sequence of time instants corresponding to the completion of a successful transmission. Hence, the average time required for a successful transmission is bounded above by , where denotes the maximum propagation delay. Since an idle period is made up
of a number of consecutive slots in which the transmission medium remains idle due to the backoff and the collisions and frame loss might occur between two consecutive successful transmissions, we have
(1)
where , and are the lengths of the -th idle
period, frame loss and collision in a virtual time, respectively,
and and are the number of collisions and the
number of lost frames in a virtual time, respectively.
The assumption that the backoff interval is sampled from a geometric distribution with parameter implies that the future behavior of a station does not depend on the past. Hence, the above equation can be rewritten as
(2)
Closed expressions for , and have been
derived in the literature with and
(3) (4) (5)
(6) Hence, is a function of the system’s parameters, the number of active stations , the parameter which defines the geometric-distribution used in the backoff algorithm, and the parameter that characterizes the frame-size geometric dis-tribution. As mentioned earlier, each station transmits a frame with probability . This yields
(7) where is the probability that a transmitted frame encoun-ters a collision or is received in error, and BER denotes the channel bit error rate. Using the Markov chain, we can obtain an explicit expression for the probability as a function of proba-bility
(8) where is the minimum CW, and is the maximum number of backoff stages, i.e., . From (7), we obtain
Substituting , as expressed by (8), into (9), we obtain
(10) Recall that the probability is defined as the probability that a frame transmitted by the considered station fails. Since in each busy slot an eventual frame transmission would have failed, the probability can be obtained by summing up the number of experienced collisions, frame losses, as well as the number of observed busy slots, and then dividing this sum by the total number of observed slots on which the measurement is taken,
i.e., .
In order to maximize the utilization of every slot in a CW, we still need to engineer the tight upper bound of to help us complete this scheme. We start with defining to be
the value of parameter that minimizes . Since is
closely approximated by the value that guarantees a balance between the collision and frame loss and the idle periods in a virtual transmission time. Suppose there are stations making a transmission attempt in a slot. Then, we have
(11)
As a consequence, is a tight upper bound of in a
system operating with the optimal channel utilization level. Sub-stituting , as expressed by (10), we obtain
(12)
More precisely, the capacity of the 802.11 DCF protocol can be improved to achieve the theoretical throughput limit corre-sponding to the ongoing network environment, channel BER, and traffic configuration by dynamically adjusting its CW whose average size is identified by the optimal value, , that is,
when the average size of CW is .
A natural strategy for expansion and contraction is to allocate a new CW size at the end of each transmission time. However, such a common heuristic would conduct the size of CW to fluc-tuate rapidly between expansion and contraction. To avoid this undesirable behavior, each station runs the algorithm to estimate the optimal CW size, and use the following formula to update its CW:
where is a smoothing factor. Finally, instead of
using the backoff time generation function defined in the IEEE 802.11 standard, we refine the backoff time generation function
as to complete our scheme.
B. Packet Transmit-Permission Policy for Real-Time Traffic Serving for the purpose of deciding whether a network ac-cepts a new connection or not, the design of a packet transmis-sion policy is one of the important challenges of traffic control
Fig. 2. State transition diagram of real-time stations.
in wireless networks. The policy is also used by the AP to deter-mine which station gets permission to transmit a packet, espe-cially in the realm of providing QoS. In this section, we present the methodology to highlight both admission control and reser-vation to meet the QoS requirements. Consequently, end-to-end QoS can be satisfied in WLANs. In what follows, we propose a packet transmit-permission policy for the IEEE 802.11 protocol to support integrated multimedia traffic. Our scheme is an en-hanced version of the transmitting policy originally proposed in [43], and substantially extended from [44]. As we shall see later, our method is simple and efficient and can be implemented in the present IEEE 802.11 PCF or the upcoming IEEE 802.11e HCF standard easily. In addition, we also take handoff traffic into con-sideration. Under such a scheme, all voice traffic satisfies their jitter constraints, all video traffic satisfies their delay constraints, and the remaining bandwidth is shared by data traffic fairly and efficiently. Furthermore, the proposed scheduling algorithm for voice traffic is provably optimal, ensuring the minimum average waiting time for voice packets.
At the IP layer, the maximum transmission unit (MTU) is set to be 1500 bytes, which is the maximum MSDU (the packet delivered to the MAC layer by the higher layer) size for the 100basedT Ethernet. Similarly, the 802.11 standard also pro-vides a fragmentation mechanism, which allows the MAC layer to split an MSDU into more MPDUs (packets delivered by the MAC layer to the PHY layer). Hence, to formalize our problem, we assume that all real-time traffic packets have the same size in this paper. Besides, two types of real-time traffic are considered. The first is voice traffic which is characterized by two parame-ters , where is the rate (number of packets per second) of the source and is the maximum tolerable jitter (packet delay variation) for this stream. Jitter is defined to be the difference be-tween the time of two successive departures and the time of two successive arrivals. The second is video traffic which is charac-terized by three parameters and , where is the average rate of the source, is the maximum burstiness of the source, and is the maximum tolerable delay (packet transfer delay) for this stream.
The channel model considers the real-time stations to be in one of three states: “Empty,” “Request,” and “Wait to Transmit.” Stations with empty buffers are said to be in the Empty state. When a packet (or packets) arrives to the buffer of a station in the Empty state, the station enters the Request state. A station in the Request state sends its request via the request access in
Fig. 3. Proposed packet transmit-permission policy.
contention period (CP) and stays in the Request state until its request is successfully received by the AP. When a station in the Request state successfully sends a request, it switches to the Wait-to-Transmit state. A station in the Wait-to-Transmit state listens to the channel until it is polled by AP, at which point it transmits a packet in the contention free period (CFP) of the next time slot, and also transmits a contention-free request (if necessary) using the PGBK request bit. If the station transmits a nonzero PGBK request bit (indicating the station’s buffer is still nonempty), the station stays in the Wait-to-Transmit state. If the station transmits a zero PGBK request bit (indicating the station’s buffer is empty), the station returns to the Empty state. Fig. 2 shows the state transition diagram of the real-time station. In one basic service set (BSS) of the IEEE 802.11 infrastruc-ture network architecinfrastruc-ture, the AP implements a token buffer for each real-time source. In token buffers for voice sources, the smaller the average rate is, the higher the priority becomes. In token buffers for video sources, the one with the smallest maximum delay constraint has the highest priority among all video sources. We depict the packet transmit-permission policy in Fig. 3. In order to gain control of the medium, the AP per-forms the function of the point coordinator by transmitting a beacon frame at the beginning of the CFP after sensing the medium to be idle for a PIFS period. Once the AP has the con-trol of the medium, it performs the following tasks.
1) The AP first scans the token buffers of voice sources according to the preset priority order. If a token is found, it removes one from this token buffer and polls this voice terminal. On receiving a poll the station transmits its packet after a SIFS interval. Then, the AP generates the next token for this voice source after second if the piggyback was set while transmitting the packet, where is the time to transmit a real-time traffic packet. 2) If no tokens are found in the token buffers of voice sources, the AP continues to scan the token buffers for
video sources according to the preset priority order. If a token is found, it polls this video source. And it will not remove the token if the piggyback was set while this video source transmitting it packet. If the piggyback was not set and it is not the last packet (end-of-file) either, the AP removes the token, and then generates the next token for this video source after seconds if there is no new token generated for this video source within , where will be defined later.
3) If there is no token found in all token buffers, the AP will not know which, if any, of the stations has packets to transmit, then, it can end the CFP by transmitting a CF-end frame, and, for assuring the time constraint of ad-mitted real-time traffic, the AP shall announce the begin-ning of the next CFP interval by observing the token buffer of highest priority on its polling list.
Note that the contention-free multipoll (CF-multipoll) mech-anism defined in the IEEE 802.11e standard can be also used here to reduce the polling overhead from which the PCF suffers. In the following theorems, we provide sufficient conditions for all the voice packets to satisfy their maximum jitter constraints and for all the video packets to satisfy their maximum delay constraints, while optimizing the overall utilization of network bandwidth simultaneously.
Assume there are voice sources (indexed by
), and video sources (indexed by ). We
de-note as the traffic parameters of the th voice source, as the traffic parameters of the th video source, and as the time needed for handoff for source .
Theorem 1: Let and
and be the
time to transmit a packet. If and for
all , then all the packets generated by new-call
voice sources meet their jitter constraints. Furthermore, if
handoffed from other cells, then the packet generated by the th source after handoff meets its jitter constraint. (Note that should be adjusted when the bandwidth for real-time traffic changes. More precisely, equals the size of MPDU divided by the bandwidth of channel I for real-time packets or the size of MPDU divided by the bandwidth of channel I and channel II for handoff real-time packets, where channel I and channel II will be defined later.)
Proof: We first prove the handoff part. Suppose that the first token generated from the th voice source after handoff from other cells has a maximum waiting time . We want to
prove that for . For
, which establishes the induction basis.
Suppose our induction hypothesis holds up to the th
voice sources, i.e., for . Now, we
consider the th voice source. Let the instant of the beginning
of handoff be at time 0. Assume that . Then, it
means that up to time the channel must be serving all the voice sources from 1 to . Since the total amount of
packets that can be served within for these
voice sources is at most . Hence, the
total amount of time to serve these packets is bounded above by ,
and since , we have
This contradicts our assumption that . Hence, .
Based on the principle of induction, the statement of the the-orem follows. Using a similar reasoning, it is reasonably easy to show that all the packets generated by new-call voice sources will meet their jitter constraints.
Theorem 2: Suppose voice sources are scheduled in the given priority order. The average waiting time is minimized for
voice packets if for all .
Proof: The proof is done by contradiction. Assume that there exists a minimum average time schedule containing
with . Recall that the total waiting time within is
Since and , then this cannot be
a minimum total waiting time schedule because we could get a shorter total waiting time schedule by simply exchanging the scheduling order of voice sources and —a contradiction. The theorem is, therefore, proven.
Theorem 3: See the equation at the bottom of the page. If and for all , then the delay constraints are satisfied for all the new-call video sources. Furthermore, if for th source which is handoffed from other cells, then the packet generated by the th source after handoff meets its delay constraint.
Proof: Consider a nonnegative, left limited, and right
continuous stochastic process . Let
. We say that is -upper
con-strained if for all . Similarly, is
-lower constrained if for all .
Since the number of departures in from a -leaky
bucket is bounded above by , the departure
process from a -leaky bucket is -upper
con-strained.
Now, consider the first video source. Let
be the stochastic process that denotes the available bandwidth to the first video source at time . If the channel is available to the first video source at time , then . Otherwise,
.
As mentioned above, the maximum number of packets from the voice sources that can be served in is at most . Hence, the bandwidth that is available to the first video source in is at least
. Thus,
.
That is, is -lower constrained.
Let be the amount of workload of
number of departures in from the first video traffic is -upper constrained. We have
. This
shows is -upper constrained.
Consider an instant after the last packet was sent (but not the EOF packet) by the first video source. Mark the instant as time 0. By letting be the amount of backlogged workload from the first video source in the channel at time , we have . Since the next token for the first video source will be generated
at time at the latest. We have .
Note that the delay for an arrival at time is bounded by the amount of time needed to deplete , and the time to deplete
is bounded by .
Maximizing over , we have the following upper bound for the maximum delay:
for the handoff traffic
Since , we have
. Applying the upper constraint for and the
lower constraint for , we have ,
or . This completes the argument
for the first video source.
The argument for the th video source is essentially the same as that for the first video source. However, the lower constraint for the channel needs to be modified since the th video source only uses the remaining channel after all the voice sources and the first video sources. Since the maximum delay of the
th video source is bounded above by ,
the number of packets from the th source that can be served in
is bounded above by . Hence, the
amount of workload from the th source that can be served in is bounded above by
. Parallel to the argument for the first video source, the maximum delay of the th video source is bounded above by the first equation at the bottom of the page (or the second equation at the bottom of the page for the handoff traffic).
Finally, we still need to engineer to complete this scheme. In order to maximize the bandwidth utilization, one should have as large as possible. The largest can be obtained by solving . However, larger will lead to unsmooth video traffic. Therefore, we give a higher priority to the admitted inactivated
Fig. 4. Proposed bandwidth partition.
video traffic in the contention period in order to compensate this shortcoming.
C. Adaptive Bandwidth Management Strategy
Recall that handoff calls can be supported in our proposed method since a real-time handoff station might be given higher priority over new connection requests. However, the handoff dropping probability might still increase if the desired amount of bandwidth in the neighboring cell, i.e., BSS, cannot be provided. In either case, binding the priority to channel access makes these QoS support mechanism unfair. As the number of stations generating high priority traffic increases, they tend to grab the channel. Hence, from the performance viewpoint, it is equally important to guarantee a minimum bandwidth for data traffic in order to maintain a reasonable bandwidth usage. To achieve this goal, in what follows we propose an adaptive bandwidth management strategy. Our strategy not only tries to maximize the bandwidth utilization and reduce the handoff dropping probability and blocking probability but also guarantees a minimum bandwidth for data traffic. In addition, this strategy is very simple and easy to implement without excessive computations.
As shown in Fig. 4, the total bandwidth is divided into three parts: channels I, II, and channel III. We allocate channel I for real-time traffic and channel II for handoff real-time traffic in contention free period. By allowing the handoff real-time traffic to use bandwidth exclusively with preemptive priority over other traffic in channel II, the handoff real-time traffic might have a larger share of pie in bandwidth utilization to reduce the dropping probability. Likewise, the remaining real-time traffic has precedence over the new request/data traffic for using network resources in channel I. Channel III is only reserved for new requests and data traffic to guarantee a minimum
bandwidth for data traffic in the contention period. However, after bandwidth is allocated, network conditions may change. Therefore, the proposed strategy can also adjust the amount of allocated bandwidth based on the measured dropping prob-ability, blocking probprob-ability, and bandwidth utilization. The algorithm to control the size of the allocated bandwidth is summarized in the following.
Function Adaptive Bandwidth Allocation
IF monitored dropping probability> threshold D THEN IF bandwidth utilization< THEN
size of allocated bandwidth II= min fmax fsize of allocated bandwidth I, size of allocated bandwidth IIg 2 up , total bandwidthg
ELSE
size of allocated bandwidth II= min fmax fsize of allocated bandwidth I, size of allocated bandwidth IIg 2 up , total bandwidth2 threshold channel II max g
ELSE
IF monitored blocking probability> threshold B THEN IF bandwidth utilization< THEN
size of allocated bandwidth I= min fsize of allocated bandwidth I2 up , total bandwidth 2 threshold channel I max g ELSE
size of allocated bandwidth I= min fsize of allocated bandwidth I2 up , total bandwidth 2
threshold channel I mediumg ELSE
IF bandwidth utilization< THEN
size of allocated bandwidth II= max fsize of allocated bandwidth II2 down , total bandwidth 2
threshold channel II ming
size of allocated bandwidth I= max fsize of allocated bandwidth I2 down , total bandwidth 2
threshold channel I ming
As the pseudocode illustrates, the handoff dropping proba-bility is the first measure used to adjust the allocated bandwidth. If the dropping probability over the threshold, threshold D, and the bandwidth utilization is not good enough (less than the threshold value ), it implies that there is not so much data traffic. Hence, we increase the size of channel II by a factor up (by changing the value of , i.e., the total time of transmits a real-time packet in a contention free period, in the inequalities of Theorems 1 and 3) to its maximum (total band-width). Otherwise, we guarantee a minimum bandwidth for data traffic by only increasing the size of channel II to the threshold (total bandwidth threshold up II). Then, we use the blocking probability to adjust the allocated bandwidth of channel I in the same way. However, to lower dropping probability will get higher priority than to lower blocking probability in adjusting bandwidth allocation. Finally, the allocated bandwidth will be stable in a good situation if the bandwidth is over the threshold . That is, both dropping probability and blocking probability are under the threshold, and the bandwidth utilization is above the threshold value .
III. PERFORMANCEEVALUATION
In this section, we evaluate the performance of the proposed scheme.
A. Simulation Environment
Our simulation model is built using the Simscript tool [45]. The model represents a BSS in the IEEE 802.11 WLANs with all stations in the BSS capable of directly communicating with the remaining parties. To focus on the access control issue and to reduce the complexity of the simulation, what follows are the basic assumptions in our simulation environment. First, the “hidden terminal” and the “exposed terminal” problems [46] are not addressed in the simulation model. Second, no stations operate in the “power-saving” mode. Third, no interference is considered from nearby BSSs. Finally, the probability of a frame being transmitted successfully is calculated as:
, where is the number of bits transmitted in the frame and BER denotes the bit error rate.
Three types of traffic are considered in the simulation. 1) Pure data: The arrival of data frames from a station’s
higher layer to MAC sublayer is Possion. Frame length is assumed to be exponentially distributed with mean length 1024 octets.
2) Voice traffic: Voice stream is characterize by two param-eters , where is the rate of the source and is the maximum tolerable jitter (packet delay variation) for this stream. Frames of voice traffic that are not success-fully transmitted within its maximum jitter constraint are assumed to be lost. The voice stream is modeled as a two state Markov on/off process, where stations are either transmitting (on) or listening (off). The amount of time in the off or on state is exponentially distributed, where the mean value of the silence (off) period is 1.5 s, and the mean value of the talk spurt (on) period is 1.35 s. The du-ration of each connection is exponentially distributed with mean time 3 min.
3) Video traffic: Each video stream is characterized by three parameters , where is the average rate of the source, is the maximum burstiness of the source, and is the maximum tolerable delay (packet transfer delay) for this stream. We use a source model in [47]. The bit rate of a single source for the th frame is defined by the
recursive relation: = [bit/pixel],
where , and is a sequence of
independent Gaussian random variables which have mean 0.572 and variance 1. Like voice frames, video frames that are not successfully transmitted within its maximum tolerable delay, , are assumed to be lost.
The default values used in the simulation are listed in Table II. The values for the simulation parameters are chosen carefully in order to closely reflect the realistic scenarios, as well as to make the simulation feasible and reasonable.
B. Numerical and Simulation Results
In the first instance, we measure the maximum jitter and delay for the real-time sources both through the analytical models and
TABLE II
DEFAULTATTRIBUTEVALUESUSED IN THESIMULATION
simulations to verify the correctness and usefulness of the pro-posed scheme. In Fig. 5, we show the differences between the analytical models and the simulated results for voice and video sources. They both show that the restricted bounds of maximum jitter/delay in the proposed scheme are more conservative as they are derived from the worst case analysis. Although both of the figures show that the results are quite close, the simulated results still have about 3% to 7% inaccuracy.
In what follows, the performances of the proposed scheme and the conventional IEEE 802.11 protocol are compared based on simulations. The performances of CF-multipoll mechanism are also discussed. In the CF-multipoll mechanism, the AP can poll more than one station simultaneously using a single CFPoll frame to reduce the polling overhead. In the conventional IEEE 802.11 protocol, CSMA/CA is adopted as the random access protocol for the contention period, and a round-robin discipline is chosen as the scheduling policy for AP in the contention free period. That is, all traffics have the same priority. The admission control scheme in the conventional IEEE 802.11 protocol is very simple and intuitive. Assuming there are totally requests in the
request table, if for the
th voice source or for
Fig. 5. Differences between analytical models and simulated results.
Fig. 6. Dropping probability of real-time handoff connections.
the th video source, the request of a new voice or video source is admitted; otherwise, it is rejected. Understandably, the time needed for handoff will be added for the handoff mobile in ad-mission control. The duration of the contention free period and the length of each superframe are set to be 50 and 75 ms, re-spectively, for the conventional IEEE 802.11 protocol since the video, voice and data traffic are assumed to be mixed in the ratio of 1:1:1 and the maximum delay for video traffic is 50 ms in the simulation. The contention free period ends when either its du-ration has reached the maximum value, i.e., 50 ms, or the AP has no more requests in its request table. However, it is noteworthy that the duration of the contention free period and the length of each superframe are dynamically allocated according to the current channel and traffic status by the AP using the proposed packet transmit-permission policy in the proposed scheme.
Simulation results are shown below in the form of plots. Figs. 6 and 7 show the dropping probability of real-time handoff connections and blocking probability of real-time new connections for the proposed scheme and the conventional IEEE 802.11 protocol. Since the handoff dropping probability is the first measure used to adjust the allocated bandwidth
Fig. 7. Blocking probability of real-time new connections.
Fig. 8. Average access delay of voice traffic, variance = 21:17 (proposed scheme-single poll), 15.49 (proposed scheme-multipoll), and 136.09 (conventional IEEE 802.11).
and we also allow the handoff real-time traffic to use band-width exclusively with preemptive priority over other traffics in the reserved region, channel II, the dropping probability will normally be kept under the threshold, i.e., threshold D. The fact that the proposed scheme provides a slightly higher blocking probability in heavy load is obvious because it shows the tradeoff between the dropping probability and the blocking probability. However, it seems counterintuitive that the pro-posed scheme provides lower blocking probability in light load. This is because that the contention free period can start as soon as the request table just becomes nonempty in the proposed scheme, then the AP will end the current contention period. However, once the contention period starts, the real-time traffic is not allowed to be served until the next contention free period in the conventional IEEE 802.11 protocol.
Figs. 8 and 9 compare the average access delays of voice and video traffic from the proposed scheme and the conven-tional IEEE 802.11 protocol, respectively. We can see that al-though there is not much difference in the values of the per-formance measures when load is light, however, the proposed
Fig. 9. Average access delay of video traffic, variance = 103:27 (proposed scheme-single poll), 82.73 (proposed scheme-multipoll), and 152.97 (conventional IEEE 802.11).
Fig. 10. Average access delay of data traffic, variance= 2577:15 (proposed scheme-single poll), 2116.55 (proposed scheme-multipoll), and 2397.36 (conventional IEEE 802.11).
scheme provides significantly better performance than the con-ventional IEEE 802.11 protocol at heavy load. In other words, the average access delay of the proposed scheme remains low when the offered load is high; in contrast, the conventional IEEE 802.11 protocol shows a sharp rise as the load increases since it lacks any priority/reservation scheme and access control policy. The simulation results suggest that the proposed scheme is ap-propriate for transmitting high priority real-time traffic such as voice and video traffic in real-time applications.
Fig. 10 shows the average access delay of data traffic in a mul-timedia communication environment. As expected, the average access delay of data traffic in the proposed scheme is worse than the conventional IEEE 802.11 protocol since it is of low priority. However, the lower priority traffic can have the bandwidth it needs in light load, so it is not wasted. In fact, a minimum band-width for data traffic to maintain a reasonable bandband-width usage can still be guaranteed by using the proposed adaptive band-width management scheme.
Fig. 11. Average bandwidth utilization.
Fig. 11 presents the average bandwidth utilization as a func-tion of the offered load. Average bandwidth utilizafunc-tion is the percentage of the bandwidth actually being used in the total bandwidth. As illustrated in Fig. 11, the average bandwidth uti-lization is slightly lower for the proposed scheme in a highly loaded system because to maintain the desired QoS, it must be more conservative in admitting new connections. It reveals that there is a clear tradeoff between deterministic (hard) QoS supporting and bandwidth utilization. However, it also reveals that our proposed scheme reduces the handoff dropping prob-ability without sacrificing the bandwidth utilization too much. Besides, the multipoll scheme can surely increase the overall performance over the single-poll scheme, as can easily be seen in the figure.
IV. CONCLUSION
In the era of multimedia communication, the design of priority-sensitive network protocols continues to be an impor-tant issue, and broadband wireless links constitute a subclass in which prioritization is key to optimizing the overall per-formance of the network. In this paper, we have proposed a pragmatic polling with nonpreemptive priority-based access control scheme built on well-known protocols, offering easily implemented and yet flexible criteria for traffic prioritization in a wireless environment. By modifying the DCF access method in the contention period, our designed protocol supports mul-tiple levels of priorities such that user mobility can be supported in wireless LANs. Besides, the proposed transmit-permission policy and adaptive bandwidth allocation scheme not only separate admitted inactivated users from newly requesting access users, but also derive sufficient conditions such that all the time-bounded traffic sources satisfy their time constraints to provide various deterministic QoS guarantees in the contention free period while maintaining efficient bandwidth utilization at the same time. Furthermore, the proposed scheduling algorithm for voice traffic is provably optimal in that it gives minimum average waiting time for voice packets. The proposed scheme is performed at each AP in a distributed manner. Through extensive simulations, we have demonstrated a satisfactory performance of our proposed scheme in a quantitative way.
ACKNOWLEDGMENT
The authors would like to thank the anonymous referees for their helpful comments that greatly improved the quality of this paper.
REFERENCES
[1] IEEE Draft Standard P802.11, Jan. 1996. Wireless Medium Access Control and Physical Layer WG, Wireless LAN, IEEE Std. Dept, D3. [2] ETSI TC-RES, “Radio equipment and systems (RES); High
perfor-mance radio local area network (HIPERLAN); Functional specifica-tion,” ETSI, 06 921 Sophia Antipolis Cedex, France, draft prETS 300 652, Jul. 1995.
[3] M. Chelouche, S. Hethuin, and L. Ramel, “Digital wireless broadband corporate and private network: RENT concepts and applications,” IEEE
Commun. Mag., vol. 35, no. 1, pp. 42–51, Jan. 1997.
[4] K. Y. Eng, M. J. Karol, M. Veeraraghavan, E. Ayanoglu, C. B. Wood-worth, P. Pancha, and R. A. Valenzuela, “BAHAMA: A broadband ad hoc wireless ATM local-area network,” in Proc. ICC, 1995, pp. 1216–1223.
[5] P. Pilat, “HomeRF-SWAP: Optimized for home networking,” Intel
De-veloper Update Mag., pp. 1–5, Jun. 2000.
[6] B. A. Miller and C. Bisdikian, Bluetooth Revealed—The Insider’s Guide
to an Open Specification for Global Wireless Communications. Engle-wood Cliffs, NJ: Prentice-Hall, 2001.
[7] J. Geier, Wireless LANs—Implementing High Performance IEEE 802.11
Network, 2nd ed: Sams Publishing, Jul. 2001, p. 19.
[8] R. O. LaMaire et al., “Wireless LAN’s and mobile networking: Stan-dards and future directions,” IEEE Commun. Mag., vol. 34, no. 8, pp. 86–94, Aug. 1996.
[9] R. A. Dayem, Moble Data and Wireless LAN’s Tech-nology. Englewood Cliffs, NJ: Prentice-Hall, 1997, pp. 190–201.
[10] M. Veeraraghavan, N. Cocker, and T. Moors, “Support of voice ser-vices in IEEE 802.11 wireless LANs,” in Proc. INFOCOM, 2001, pp. 488–497.
[11] Y. Wang and B. Bensaou, “Priority based multiple access for service differentiation in wireless ad hoc networks,” in Proc. MWCN, 2000, pp. 14–30.
[12] P. H. Chuang, H. K. Wu, and M. K. Liao, “Dynamic QoS allocation for multimedia ad hoc wireless networks,” Proc. Comput. Commun. Netw., pp. 480–485, Oct. 1999.
[13] J. L. Sobrinho and A. S. Krishnakumar, “Quality-of-service in ad hoc carrier sense multiple access wireless networks,” IEEE J. Sel. Areas
Commun., vol. 17, no. 8, pp. 1353–1368, Aug. 1999.
[14] , “Distributed multiple access procedures to provide voice commu-nications over IEEE 802.11 wireless networks,” in Proc. GLOBECOM, 1996, pp. 1689–1694.
[15] D. J. Deng and R. S. Chang, “A priority scheme for IEEE 802.11 DCF access method,” IEICE Trans. Commun., vol. E82-B, no. 1, pp. 96–102, Jan. 1999.
[16] C. R. Lin, “Multimedia transport in multihop wireless networks,” Proc.
Inst. Elect. Eng. Commun., vol. 145, no. 5, pp. 342–345, Oct. 1998.
[17] J. Y. Yen and C. H. Chen, “Support of multimedia services with the IEEE 802.11 MAC protocol,” in Proc. ICC , vol. 1, Apr. 2002, pp. 600–604. [18] S. T. Sheu and T. F. Sheu, “A bandwidth allocation/sharing/extension
protocol for multimedia over IEEE 802.11 ad hoc wireless LAN’s,”
IEEE J. Sel. Areas Commun., vol. 19, no. 10, pp. 2065–2080, Oct. 2001.
[19] S. Chakrabarti and A. Mishra, “QoS issues in ad hoc wireless networks,”
IEEE Commun. Mag., vol. 39, pp. 142–148, Feb. 2001.
[20] A. Lindgren, A. Almquist, and O. Schelen, “Evaluation of quality of service schemes for IEEE 802.11 wireless LANs,” in Proc. LCN, Nov. 2001, pp. 348–351.
[21] T. Suzuki and S. Tasaka, “Performance evaluation of priority-based mul-timedia transmission with the PCF in an IEEE 802.11 standard wireless LAN,” in Proc. Int. Symp. Pers., Indoor, Mobile Radio Commun., vol. 2, Oct. 2001, pp. G70–G77.
[22] , “Performance evaluation of integrated video and data transmission with the IEEE 802.11 standard MAC protocol,” in Proc. GLOBECOM, vol. 1b, Nov. 1999, pp. 580–586.
[23] S. Sharma, K. Gopalan, and N. Zhu, “Quality-of-service guarantee on 802.11 networks,” Proc. Hot Interconnects 9, pp. 99–103, Aug. 2001. [24] C. Coutras, S. Gupta, and N. B. Shroff, “Scheduling of real-time traffic in
IEEE 802.11 wireless LANs,” Wireless Netw., vol. 6, no. 6, pp. 457–466, Dec. 2000.
[25] C. Andren, “IEEE 802.11 wireless LAN: Can we use it for multimedia?,”
IEEE Multimedia, vol. 5, no. 2, pp. 84–89, Apr. 1998.
[26] F. Eshghi and A. K. Elhakeem, “Performance analysis of ad hoc wireless LAN’s for real-time traffic,” IEEE J. Sel. Areas Commun., vol. 21, no. 2, pp. 204–215, Feb. 2003.
[27] Y. Xiao, “Enhanced DCF of IEEE 802.11e to support QoS,” in Proc.
WCNC 2003, Mar. 2003, pp. 1291–1296.
[28] S. C. Lo, G. L. Lee, and W. T. Chen, “An efficient multipolling mech-anism for IEEE 802.11 wireless LANs,” IEEE Trans. Comput., vol. 52, no. 6, pp. 764–778, Jun. 2003.
[29] G. Anastasi and L. Lenzini, “QoS provided by the IEEE 802.11 wireless LAN to advanced data applications: A simulation analysis,” Wireless
Netw., vol. 6, no. 2, pp. 99–108, Mar. 2000.
[30] O. Sharon and E. Altman, “An efficient polling MAC for wireless LAN’s,” IEEE/ACM Trans. Netw., vol. 9, no. 4, pp. 439–451, Aug. 2001.
[31] D. Gu and J. Zhang, “QoS enhancement in IEEE 802.11 wireless local area networks,” IEEE Commun. Mag., vol. 41, no. 6, pp. 120–124, Jun. 2003.
[32] Part 11: Wireless Medium Access Control and Physical Layer
Specifi-cations: Medium Access Control Enhancements for Quality of Service,
Nov. 2002. IEEE 802.11e draft/D4.0.
[33] W. P. Atikom, P. Krishnamurthy, and S. B. Jee, “Distributed mechanisms for quality of service in wireless LANs,” IEEE Commun. Mag., vol. 10, no. 3, pp. 26–34, Jun. 2003.
[34] B. P. Crow, I. Widjaja, J. G. Kim, and P. T. Sakai, “IEEE 802.11 wireless local area networks,” IEEE Commun. Mag., vol. 35, no. 9, pp. 116–126, Sep. 1997.
[35] B. Moon, C. Oh, A. Ahmad, and K. Kim, “A study of bandwidth alloca-tion strategies in multimedia wireless networks,” in Proc. Conf. Record
APCC, 1997, pp. 509–513.
[36] C. Oliveira, J. Bae, and T. Suda, “An adaptive bandwidth reservation scheme for high-speed multimedia wireless networks,” IEEE J. Sel.
Areas Commun., vol. 16, no. 6, pp. 858–873, Aug. 1998.
[37] F. Cali, M. Conti, and E. Gregori, “IEEE 802.11 wireless LAN: Ca-pacity analysis and protocol enhancement,” in Proc. INFOCOM, San Francisco, CA, Mar. 1998, pp. 142–149.
[38] L. Bononi, M. Conti, and E. Gregori, “Design and performance evalu-ation of an asymptotically optimal backoff algorithm for IEEE 802.11 wireless LANs,” in Proc. 33rd Hawaii Int. Conf. Syst. Sci., 2000, pp. 1–10.
[39] F. Cali, M. Conti, and E. Gregori, “Dynamic tuning of the IEEE 802.11 protocol to achieve a theoretical throughput limit,” IEEE/ACM Trans.
Netw., vol. 8, no. 6, pp. 785–799, Dec. 2000.
[40] G. Bianchi, “Performance analysis of the IEEE 802.11 distributed co-ordination function,” IEEE J. Sel. Areas Commun., vol. 18, no. 3, pp. 535–547, Mar. 2000.
[41] L. Alcuri, G. Bianchi, and I. Tinnirello, “Occupancy estimation in the IEEE 802.11 distributed coordination function,” presented at the ICS2002, Hualien, Taiwan, 2003.
[42] L. Bononi, M. Conti, and L. Donatiello, “Design and performance eval-uation of a distributed contention control (DCC) mechanism for IEEE 802.11 wireless local area networks,” in Proc. Workshop WOWMOM,
MOBICOM, Dallas, TX, Oct. 1998, pp. 1–10.
[43] D. J. Deng and R. S. Chang, “A nonpreemptive priority based access control scheme for broadband ad hoc wireless ATM local area net-works,” IEEE J. Sel. Areas Commun., vol. 18, no. 9, pp. 1731–1739, Sep. 2000.
[44] C. S. Chang, K. C. Chen, M. Y. You, and J. F. Chang, “Guaranteed quality-of-service wireless access to ATM networks,” IEEE J. Sel. Areas
Commun., vol. 15, no. 1, pp. 106–118, Jan. 1997.
[45] CACI Products Company. (1997, Sep.). Simscript II.5, CA 92 037. [On-line]. Available: http://www.caciasl.com/
[46] A. S. Tanenbaum, Computer Networks, 3rd ed. Englewood Cliffs, NJ: Prentice-Hall, 1996, pp. 263–264.
[47] B. Maglaris et al., “Performance models of statistical multiplexing in packet video communications,” IEEE Trans. Commun., vol. 36, no. 7, pp. 834–844, Jul. 1998.
Der-Jiunn Deng, photograph and biography not available at the time of
publication.
Hsu-Chun Yen, photograph and biography not available at the time of
