Distributed Coordination Function

The fundamental medium access mechanism of the IEEE 802.11 is the DCF, which is based on Carrier Sense Multiple Access with Collision Avoidance (CSMA/CA) and can provide the fair chance of accessing the medium. The CSMA sometimes is called “listen-before-talk” scheme, the STA senses the medium before data delivery. Before the whole control function and the mechanism flow of the DCF, there are three issues should be addressed first: the reliable data delivery, the

inter-frame space and the binary exponential backoff.

The WLAN technology operates on the unlicensed industrial, scientific and medical (ISM) band (802.11b/g on 2.4 GHz and 802.11a on 5GHz). To provide the reliable data delivery on the noisy and interference ISM band, the acknowledgement frame (ACK) is necessary mostly. The source STA expects the ACK frame after the transmission to the destination, and the STA will try to retransmit if it does not get the ACK frame.

The inter-frame space (IFS) is the time interval between frames. There are four kinds of IFS defined to provide multiple priorities for medium access. The four IFSs from the shortest to the longest are short IFS (SIFS), PCF IFS (PIFS), DCF IFS (DIFS) and extended IFS (EIFS).

The SIFS is the minimum time interval between any two frames. Using the smallest gap within the frame exchange can prevent other STAs from attempting to use the medium, thus giving the priority to the completion of the frame exchange in progress. The PIFS is used by the AP to gain the access over the medium. The value of the PIFS is one SIFS plus one slot time¹. The DIFS is used for a STA wishes to start a transmission, which is the SIFS plus two slot times. The EIFS is much larger than other kinds of IFS and is the interval required between a STA’s attempts on the retransmission of a failed packet.

A collision occurs when multiple STAs simultaneously transmit on the shared medium. The DCF protocol mandates the STA attempts to transmit must execute a backoff procedure to reduce the probability of collision and provide fair access opportunities for other aspiring STAs. The binary exponential backoff procedure works as follows: A STA wishes to transmit sets the backoff timer as Equation 1. The

1 The value of the xIFS and a slot time (aslottime) depend on the physical configuration (802.11a, b, g).

value of the timer is a uniformly distributed random number ranged from zero to the Contention Window (CW). The timer decreases when the medium is free and is frozen when the medium becomes busy. After the backoff timer expires, the STA has the opportunity to transmit. The CW will double for each successive attempt to transmit the same packet as shown in Figure 2-3. Once the CW reaches the maximum value (CWmax), it shall stay at the value until it is reset. On the other hand, for a successful transmission, the CW will be reset to CWmin.

max

Figure 2-3, Binary Exponential Increase of CW

Figure 2-4 is the flow char of the STA under DCF access operation. The rule of the DCF is that a STA desiring to transmit senses the medium, if the medium is idle

Figure 2-4, Flow Char of DCF 2.2.2 Point Coordination Function

Apart from the DCF, the Point Coordination Function (PCF) provides contention-free access mechanism to let STAs have priority access to the medium and to implement the time-bounded services. When the PCF is implemented, the time is always divided into two alternated periods (contention-free period and contention period, CFP and CP) and forms a periodic superframe structure. Under this access mode, the medium control is belonged to an access point (AP), which is referred to a point coordinator. The AP polls the STA having the contention-free traffic and then routes the data to the destination.

During the CFP, the AP maintains a polling list which records the eligible CF-Pollable STAs and polls these STAs one by one. To become the CF-Pollable STA,

a STA should request to AP early. Upon receiving a poll frame, the STA transmits its data after a SIFS period. If the AP receives no response from the polled STA, the AP will poll the next STA after a PIFS. In short, there is no idle period larger than a PIFS value. If the AP has completed the polling of all STA recorded on the polling list, it can end the CFP by the CF-End frame and release the access to other contented STA.

To make the bandwidth more efficient, the piggyback scheme is utilized. The concept of the piggyback is that the control frame (e.g. ACK) and the data frame (e.g. Data) are combined whenever applicable, such as Data+CF-ACK, CF-Poll+CF-ACK and Data+CF-Poll+CF-ACK.

A superframe starts from the beacon frame broadcasted by the AP, and then the PCF and DCF operate on the CFP and the CP respectively. Figure 2-5 depicts the superframe structure and the mentioned PCF operation on the CFP. The beacon frame is the management frame that includes the synchronization and other protocol related parameters. One important parameter is the target beacon transition time (TBTT) which announces when the next beacon frame will arrive, so the AP can ideally² generate the periodic superframe.

Beacon D1+Poll

Figure 2-5, PCF Operation on CFP of Superframe

2 In practice, the superframe is not absolute periodic due to the contention. Section 2.3.1 will detail this issue.

2.3 802.11e MAC

There are some limitations for the traditional 802.11 MAC to maintain the QoS requirements, such as the required bandwidth and bounded delay. To support the QoS, the enhanced control functions which include the Enhanced Distributed Coordination Function (EDCF) and the Hybrid Coordination Function (HCF) are promoted by the IEEE 802.11 Task Group E (TGe). The EDCF is a contention-based channel access extended from the DCF, and the HCF combines both the contention-based channel access and the polling-based channel access, that is, EDCF could be viewed as a part of the HCF. The limitations of the original 802.11 MAC and both the enhanced control functions will be described in this section.

2.3.1 Enhanced Distributed Coordination Function

The DCF does not support any QoS assurance. Basically, the DCF provides the fair channel access probability without the concept of service differentiation to all STAs, that is, all STAs contend the bandwidth with the same priority. It is not a desirable feature since the time-bounded services, such as the VoIP and videoconference, have the specified bandwidth and low delay requirements. Several studies have proved the poor performance of the DCF for supporting the real-time services [20, 21].

The EDCF is an extension of the DCF and it utilizes the similar CSMA/CA operation concept. To provide the prioritized QoS, the Access Categories (ACs) and multiple independent backoff entities are introduced. Eight different priority traffics are mapped into four Access Categories (ACs), each AC (queue) performs backoff individually and has its own parameter set, which includes the Arbitration IFS (AIFS[AC]), CWmin[AC], CWmax[AC], and Transmission Opportunity (TXOP[AC])

where the AIFS[AC], CWmin[AC] and CWmax[AC] replace the DIFS, CWmin, and CWmax of the DCF respectively and the TXOP is the maximum duration that a station can transmit. Table 2-1 and Table 2-2 are the traffic priority mapping and the typical EDCF parameters respectively. In Table 2-2, the AIFSn is used to calculate the AIFS (AIFS[AC] = SIFS + AIFSn×aSlotTime). A low-priority AC has larger values of AIFSn, CWmin, and CWmax than a high-priority AC. Hence, the high-priority traffic is likely to access the medium easily than the low- priority traffic.

Table 2-1, Priority to Access Category Mapping

Priority Access Destination

Figure 2-6 is the reference model of the 802.11e STA, the four individual backoff entities perform virtual collision within a STA. As the collision occurs on the medium, it is possible that the backoff timers of the different queues are expired at the same time. The highest priority traffic frame are transmitted, and the other traffic queues perform the collision operation as they collide with other STA when more than one timer expiration within the STA. It should be noted that the highest priority traffic may also collide with other STA on the medium. By the virtual collision mechanism and the AC parameter set, the applications of each STA are served differentially.

8 traffics map into 4 ACs

Figure 2-6, Reference Model of 802.11e STA

2.3.2 Hybrid Coordination Function

The PCF is designed to support the time-bounded service, while it has three main limitations. First, the unpredictable beacon frame delay due to the alternation of the CFP and CP. The STAs are allowed to start their transmission even if the current packet can not finish before the upcoming TBTT. The delay beacon frame leads to extra delay for the time-bounded services. Second, it is no assurance of the transmission time for each polled STA. This makes the AP hard to provide guaranteed performance during the CFP. Third, the communication within an If-BSS should go through the AP and results in bandwidth inefficiency.

The HCF is based on the polling mechanism as defined in PCF and a few improved and extended control mechanisms. First, a STA is not allowed to deliver if the transmission can not be completed before the upcoming TBBT, which solves the beacon frame delay problem. Second, a TXOP is used to limit the transmission time for the polled STA. Third, the direct link protocol (DLP) defines how to communicate within an If-BSS. These improvements can help the AP make the bandwidth allocation more precise and reduce the waste of the bandwidth.

Furthermore, the auxiliary of negotiated traffic specification (TSPEC) and the non-existing boundary of the CFP and CP promote the capability of QoS supporting.

To setup a specific traffic service, a STA must send a request frame containing the TSPEC to the AP. The TSPEC describes the required QoS parameters such as mean data rate, delay bound, and etc. Based on the parameterized requirement, the AP grants or rejects the service. If the service is admitted, the AP is responsible for maintaining the QoS demands. Besides, different from the strictly separate relation of the 802.11 MAC (PCF and DCF), the HCF combines both advantages to adaptively manager the bandwidth. When HCF is implemented, a Hybrid Coordinator (HC)

resides in the AP takes the responsibility of managing the wireless medium. The most differences are that in HCF, the AP has the highest priority so as to poll STAs to maintain the QoS demands even during the CP. The period that the AP polls stations in the CP is called Controlled Access Period (CAP), as shown in Figure 2-7. The pre-information of the auxiliary TSPEC and the flexible drive of the polling-based channel access can make the bandwidth assignment more precise and efficient and further guarantee the required QoS.

Figure 2-7, Superframe Structure, 802.11/802.11e Without/With CAP

Chapter 3 Scheduling Algorithms in WLAN

Chapter 2 has introduced the enhanced medium access control function, and furthermore this chapter describes the scheduling control algorithm applied to the polling-based channel access of the HCF. The simple scheduling control algorithm provided by the vendors of TGe will be addressed first and followed by the proposed timer-based scheduling control algorithm.

3.1 Simple Scheduling Algorithm

The simple scheduling algorithm is based on fixed order polling mechanism. The AP calculates a fixed Service Interval (SI) and TXOP for each station based on Mean Data Rate (ρ), Nominal MAC Service Data Unit (MSDU) Size (L), and Maximum Service Interval (MSI) or Delay Bound (DB) of the TSPEC. Figure 3-1 shows how the simple algorithm to allocate the bandwidth, the unmarked (blank) portion in the SI is reserved (utilized) for the contention-based channel access. When a new stream participates, the AP determines the SI, and then the AP reserves (allocates) the TXOP for each service. The SI and TXOP are calculated as follows:

Figure 3-1, Bandwidth Allocation of Simple Algorithm

For the SI, the AP first calculates the minimum of all MSIs (or DB) as a value m.

Then, the AP determines the SI that is the sub-interval of the beacon interval value and is lower than the value x as shown in Equation 2, where BI represents the beacon interval and the m is an integer. The SI will be re-calculated when a new traffic stream is admitted and the MSI (or DB) of the incoming service is less than the current SI.

For TXOP, as shown in Equation 3, the AP first calculates the number of the MSDUs (N) that arrives at the Mean Data Rate during the SI. After that, from Equation 4, the AP calculates the TXOP as the maximum of the time to transmit frames calculated in Equation 3 and the time required to transmit the maximum size of MSDU (M, 2304 bytes) at transmission rate of user i, Ri, and then plus the overheads (O). The overheads include the PHY and MAC header and inter-frame space.

)}

Next, consider the bandwidth reallocation after the dropped stream. If a stream is dropped, the AP may resume the available time and leave it for contention or move the following TXOPs to prevent the intermittent alternation of the polling-based and contention-based channel accesses. An example depicted the two schemes is shown in Figure 3-2, when the traffic j is removed.

Figure 3-2, Reallocation of TXOPs

3.2 Proposed Timer-Based Scheduling Algorithm

The proposed timer-based scheduling algorithm is based on the concept of the Earliest Deadline First (EDF) and utilizes the mean inter-arrival time of the consecutive frames and the delay bound to effectively control the required transmission rate and delay budget for achieving QoS requirements. In the proposal, the AP sets two timers for each direction of a bi-directional service, and then determines which STA has the lowest timer. The setups of the timers are as follows:

The downlink timer is used to calculate the deadline of a downlink frame. As calculated in Equation 5, the deadline of the frame from the network is the delay bound (DB) minus both the frame age (Age, the time that the frame has stayed in the MAC layer) and the required time to finish the transmission of the frame (Tt).

t

d DB Age T

T = − − (5)

The uplink timer is used to estimate the uplink deadline of the next frame. Figure 3-3 shows how to set the uplink timer. A STA that intends to join the If-BSS will send requests to the AP. If the AP accepts the station, the uplink deadline is set based on Equation 6, which is the estimated time of the first frame plus the delay bound and minus the time to exchange frame. After polling the STA, the new uplink timer will be updated based on Equation 7, which is the remaining value of the timer (Told) plus the mean frame inter-arrival time (Tint).

o e

u T DB T

T = + − (6)

where Te represents the estimated time of the first frame, and To represents the time to exchange frames.

Figure 3-4 is the flow of the timer-based scheduling control algorithm. To facilitate the understanding, the margin timer (Tm, calculated as Equation 8) and the threshold (Tthr) are introduced. The margin timer represents the margin of the most urgent packet and the threshold is used to separate the contention-based and polling-based channel accesses. The AP compares the margin timer with the threshold to determine whether the polling-based access function should be commenced. If the margin timer is less than the threshold, the AP will start the polling-based channel

Figure 3-4, Flow Char of Proposed Algorithm

The threshold has a big impact on the performances. The larger threshold biases against the EDCF since the resource is always occupied by the polling-based channel access, which results in the lower throughput of the non real-time service. On the contrary, the smaller threshold may cause the serious delay of the real-time services.

Hence, there exists a tradeoff between the delay of the real-time services and the throughput of non real-time services.

Based on the flow char of the algorithm, the packet will be transmitted after it becomes the most urgent packet and the timer is less than the threshold. The relation of the delay of the packet (D(n)) and the margin timer is shown in Figure 3-5 and expressed as Equation 9, where n represents the frame sequence. Equation 9 could be written as Equation 10 from the statistics concept, the D in Equation 10 represents the average delay. To provide service differentiation and relative delay fairness, Equation 10 is modified as Equation 11. In this case, compared to the most urgent service, the other services will always tolerate a delay gap and the value of this gap is the difference of delay bound. For a given objective (expected) delay (Dobj), Equation 11 is written as Equation 12. The meaning of Equation 12 is that to achieve the objective delay, the threshold should be at least larger than a specific value. Generally, the delay increases with the loading of the real-time service. It is impracticable to desire an extreme delay for heavy loading. This causes great degradation in the throughput without improving the delay. To simplify, a threshold mapping table is suggested, the detailed setup is described in the next chapter.

thr

Figure 3-5, Relation of Delay and Minimum Timer

3.3 Analysis of Algorithms’ Concept

The last two sub-sections have described how to implement the simple and the proposed timer-based scheduling control algorithms. Before immediate simulation and comparison, this sub-section first analyzes the designed concept by intuition.

The simple algorithm serves each STA once per service interval. The AP regards each stream as constant bit rate (CBR) service and treats them equally without considering the different QoS requirements, that is, the AP reserves the bandwidth for the service if it was admitted. It is inefficient in some scenarios. First, lots of real-time applications are variable bit rate (VBR) service, so the characteristic of the burst data should be emphasized. For example, if the generated data is small and the AP still polls the STA, this will cause the bandwidth inefficiency since the station has no data to transmit. Second, the fixed SI for each service results in the same delay for each service, while this is impracticable since some services have looser requirements

especially under the mixed traffic scenario. In short, the performance degrades under the insidiously simple scheduler design.

Compared to the simple algorithm, the timer-based algorithm is more flexible.

The standard endowed the AP with the highest priority to start the CAP, but the powerful feature is seldom considered in the simple algorithm. On the other hand, the CAP is utilized in the timer-based algorithm as needed. The urgent traffic is served earliest and the service differentiation is also considered. The timer-based algorithm does not emphasize the absolute fairness in delay since the delay requirement is different. By utilizing the influential CAP properly and sacrificing the looser QoS requirement service without violating the demands can make the bandwidth utilization more efficient and adaptable.

Chapter 4 Simulation Results

In this chapter, the comparisons of the mentioned access mechanisms which include the contention-based EDCF, the simple and the timer-based scheduling control algorithms are explored. The simulation model which includes the

In this chapter, the comparisons of the mentioned access mechanisms which include the contention-based EDCF, the simple and the timer-based scheduling control algorithms are explored. The simulation model which includes the