QoS-Based Adaptive Contention/Reservation
Medium Access Control Protocols for
Wireless Local Area Networks
Jia-Shi Lin, Student Member, IEEE, and Kai-Ten Feng, Member, IEEE
Abstract—In the conventional IEEE 802.11 medium access control protocol, the distributed coordination function is designed for the wireless stations (WSs) to perform channel contention within the wireless local area networks (WLANs). Research work has been conducted to modify the random backoff mechanism in order to alleviate the packet collision problem while the WSs are contending for channel access. However, most of the existing work can only provide limited throughput enhancement under specific number of WSs within the network. In this paper, an adaptive reservation-assisted collision resolution (ARCR) protocol is proposed to both improve packet collision and reduce the backoff delays from the random access scheme. With its adaptable reservation period, the contention-based channel access can be adaptively transformed into a reservation-contention-based system if there are pending packets required to be transmitted between the WSs and the access point. Moreover, in order to support quality-of-service requirements, the enhanced-ARCR (E-enhanced-ARCR) protocol is further proposed to provide adaptation for multiple prioritized traffic in the WLAN. Analytical models are derived for both proposed schemes to evaluate their throughput performance. It can be observed from both analytical and simulation results that the proposed protocols outperform existing schemes with enhanced channel utilization and network throughput. Index Terms—Wireless local area network (WLAN), IEEE 802.11 standards, medium access control, random backoff mechanism, reservation-based algorithm.
Ç
1
I
NTRODUCTIONI
N recent years, the techniques for wireless local areanetworks (WLANs) have been extensively utilized for both indoor and mobile communications. The applications for WLANs include wireless home gateways, hotspots for commercial usages, and ad hoc networking for intervehi-cular communications. Among different techniques, IEEE 802.11 standard is considered the well-adopted suite due to its remarkable success in both design and deployment. Various amendments are contained in the IEEE 802.11 standard suite, mainly including IEEE 802.11a/b/g [1], [2], [3] and IEEE 802.11e [4] for quality-of-service (QoS) support. The medium access control (MAC) protocol within the IEEE 802.11 standard supports the distributed coordi-nation function (DCF) to regulate the random and complex medium accessing behaviors among the wireless stations (WSs) within the same WLAN. How to alleviate the probability of packet collision is considered a crucial issue to enhance the network throughput for this type of random access schemes. Furthermore, the point coordination func-tion (PCF) initiated by the access point (AP) provides centralized polling-based schemes to support time-con-strained traffic for the WSs.
There are trade-offs between the centralized-based and contention-based schemes under different network envir-onments. It will be beneficial to provide a channel access
mechanism that can adaptively switch between these two types of schemes. Therefore, an adaptive reservation-assisted collision resolution (ARCR) protocol is proposed in this paper in order to alleviate the packet collisions and reduce the backoff delays within the random access scheme. The main feature of the proposed ARCR scheme is that the original contention-based channel access will be adaptively transformed into a reservation-based system in the case that there are pending requests for packet transmission from the WSs. With the adaptable reservation period by exploiting the ARCR algorithm, packet collision resulting from channel contention can be effectively reduced which consequently leads to enhanced network throughput. Furthermore, with the consideration of four prioritized access categories (ACs) within a WS, the enhanced-ARCR (E-ARCR) protocol is further proposed in order to fulfill the QoS requirements. Analytical models for throughput analysis are developed in this paper to provide feasible observations on the behaviors of the proposed ARCR and E-ARCR protocols. Numerical results are conducted via simulations both to provide validation on the analytical models and to evaluate the effectiveness of the proposed schemes. Compared with other existing protocols, the network throughput can be enhanced by adopting the ARCR algorithm, e.g., around 50 percent performance gain with 10 WSs under error-free channel scenario. Moreover, QoS requirements can also be fulfilled with the exploitation of the E-ARCR scheme.
The remainder of this paper is organized as follows: Section 2 provides the related work, and Section 3 briefly summarizes the IEEE 802.11 MAC protocol and the gentle DCF (GDCF) scheme [5], [6]. The proposed ARCR scheme is described in Section 4 associated with its throughput analysis presented in Section 5. The proposed E-ARCR . The authors are with the Department of Electrical Engineering, National
Chiao Tung University, Hsinchu, Taiwan, R.O.C.
E-mail: [email protected], [email protected]. Manuscript received 22 Oct. 2009; revised 24 July 2010; accepted 1 Oct. 2010; published online 16 Dec. 2010.
For information on obtaining reprints of this article, please send e-mail to: [email protected], and reference IEEECS Log Number TMC-2009-10-0449. Digital Object Identifier no. 10.1109/TMC.2010.235.
protocol and its performance analysis are explained in Sections 6 and 7. Section 8 presents the performance validation and evaluation for both the proposed ARCR and E-ARCR protocols; while conclusions are drawn in Section 9.
2
R
ELATEDW
ORKDifferent types of schemes have been proposed in order to resolve the packet collision problem within the WLAN. The adjustment of contention window (CW) size has been considered an effective scheme in most of the existing research work [1], [5], [6], [7], [8], [9], [10]. The binary exponential backoff scheme [1] as described in the IEEE 802.11 MAC protocol controls the waiting time duration for channel contention. The CW size will be increased or decremented with failed or successful transmission, respec-tively. In general, the probability of packet collision can be decreased with augmented value of the CW size, especially with a larger number of WSs in the network. However, enlarged CW size can incur excessive idle time which will consequently degrade the channel utilization. In order to enhance the throughput performance for the conventional IEEE 802.11 protocol, the algorithm proposed in [7] increases the transition rate between the backoff stages associated with decreased value of the minimum CW and incremented value of the maximum CW size. The hybrid algorithm proposed in [8] combines both the exponential and the linear backoff for the purpose of decreasing packet collision, while the slow CW decrease (SD) scheme in [9] either doubles or halves the CW size according to the success of packet transmission. The early backoff announcement (EBA) pro-tocol [10] proposed a WS to record its next backoff number into the MAC head while transmitting data packets. All the other WSs will select their corresponding backoff numbers excluding this value in order to avoid potential packet collisions. The GDCF protocol as proposed in [5], [6] maintains a larger value of the CW size compared to the conventional backoff scheme in order to decrease the probability of packet collision.
Furthermore, in order to provide reliable services for multimedia applications, IEEE 802.11e standard [4] has been proposed to fulfill QoS requirements. For achieving prior-itized channel access, the enhanced distributed channel access (EDCA) mechanism defines four ACs in a WS associated with their distinct arbitration interframe spaces (AIFSs) and CW sizes. In order to provide higher through-put performance comparing with the conventional EDCA scheme, research work has been proposed in [11], [12] by providing adjustment on the four CW sizes for their corresponding ACs in a WS. Adaptation of AIFS has been studied in [13] for achieving stable capacity ratios between the ACs; while random AIFS algorithm was proposed in [14] to both decrease packet collisions and increase throughput performance. With the adjustment of CW size and rando-mized AIFS values, the work proposed in [15] improves channel utilization and fairness by preventing starvation on lower priority classes under higher traffic loads. The piggyback method [16] is utilized by inserting additional fields in order to further enhance network throughput. Nevertheless, all the existing contention-based protocols suffer from the trade-off between packet collision and transmission delay. Moreover, the throughput performance
by adopting these algorithms is greatly influenced by the total number of WSs within the WLAN.
Compared to the DCF-based random access schemes, there are also polling-based algorithms proposed for WLAN in order to provide feasible performance to fulfill time-constrained requirements. Various centralized polling protocols and scheduling algorithms (e.g., [17]) have been proposed to increase the channel utilization for the IEEE 802.11 PCF [1] and the IEEE 802.11e HCF controlled channel access (HCCA) [4]. The operation time period for each WS is divided into cycles of contention period (CP) and contention-free period (CFP), where CFP is utilized by either PCF or HCCA for real-time packet delivery. The work in [18], [19] proposed piggyback schemes for HCCA by adjusting the transmission rate of WS for throughput enhancement. However, the requirement to specifically assign the designated CFP for the implementation of polling-based algorithms will lead to excessive overhead if the WSs have no packet to be delivered to the AP. Moreover, it is considered difficult to determine the ratio of CFP to CP in order to both fulfill the QoS requirement for the WSs and enhance system throughput.
3
P
RELIMINARIESIn this section, both the IEEE 802.11 MAC protocols and the GDCF scheme are summarized which will be utilized for performance comparison with the proposed schemes in Section 8. The IEEE 802.11 MAC protocols, which include both the contention-based and the reservation-based me-chanisms, are utilized as the baseline schemes for perfor-mance comparison. As described in the previous section, most of the existing research [5], [6], [7], [8], [9], [10], [11], [12], [13], [14], [15] considers the adjustment of the CW size within their backoff algorithm in order to alleviate packet collision in the network. The delay coming from the backoff process has not been explicitly considered and reduced in the existing schemes. On the other hand, one of the major design objectives of the proposed ARCR scheme is to reduce the backoff delay by introducing the adaptive reservation table in the AP. Therefore, it is intuitively feasible to consider that the proposed ARCR scheme can outperform the existing schemes with higher system throughput.
In order to quantitatively evaluate the proposed ARCR scheme, the GDCF algorithm [5], [6] is selected from these existing schemes as an enhanced version of the IEEE 802.11 MAC protocol. The reason for selecting the GDCF algorithm as a comparison scheme is as follows: According to the design concept of the proposed ARCR scheme, its benefit will be revealed under larger number of WSs in the network owing to its adaptive scheme for table reservation. The GDCF scheme possesses higher probability of staying at the stages with larger CW sizes compared to the other existing schemes. With this design, the GDCF is capable of allowing a larger number of WSs within the network to contend for the channel access. Therefore, the GDCF protocol is selected for performance comparison with the proposed ARCR scheme.
3.1 IEEE 802.11 MAC Protocol
The DCF is utilized as the basic access mechanism in the IEEE 802.11 MAC protocol. It is based on the carrier sensing multiple access with collision avoidance (CSMA/CA) scheme to ensure that each WS can acquire a fair chance
to access the wireless medium. A WS that intends to transmit data will first sense the channel to verify if it is at the idle state. As the channel is idle for the time interval of the DCF interframe space (DIFS), the random backoff process will be started which is executed in each WS for the purpose of decreasing the probability of data collision.
The random number kdcf at the backoff stage i is chosen
within the range of a uniform distribution U½a; b, i.e., kdcf ¼
U½0; 2iW 1 where W denotes the minimum backoff
window size. It is noted that the backoff stage i corresponds to the number of transmission retries. Moreover, both the request-to-send (RTS) and clear-to-send (CTS) packets exchanged before the data transmission is exploited to resolve the potential hidden terminal problem. In order to avoid packet collision during data transmission, the virtual carrier sensing mechanism carried out by the network allocation vector (NAV) is utilized to record the duration of ongoing data transmission. It is noted that the NAV information adopted within each WS will be delivered to its neighbor nodes. A nonzero NAV value recorded in a WS will consequently prohibit the surrounding neighbor nodes to initiate a new data transmission.
Unlike the contention-based DCF scheme, the PCF supported by the IEEE 802.11 standard is designed to be a centralized polling protocol. Periodic occurrences of CP and CFP are designed for each WS, where CP is operated by DCF and CFP is executed by polling mechanism. The AP will broadcast a beacon message to inform all the WSs regarding the start of CFP. Based on a polling list of WSs recorded within the AP, the AP will sequentially transmit the CF-Poll control frame to the WS within the list by adopting the round-robin scheduling algorithm. If a WS that receives the CF-Poll frame has data to be delivered, the WS will transmit data packets to the AP after waiting for a short interframe space (SIFS). The AP correctly receiving data packets will send a CF-ACK frame in response to the WS after waiting for the SIFS time interval. On the other hand, in the case that the AP did not receive any data packet within the time interval of the PCF interframe space (PIFS), it will continue to poll the next WS in its corresponding polling list. After all the WSs in the list have been consecutively polled, the AP will broadcast the CF-End frame as the indication for the end of CFP. Afterwards, all the WSs in the network will enter into the CP mode with the adoption of the contention-based DCF scheme.
In order to support QoS requirements, the contention-based EDCA and centralized-contention-based HCCA protocols are proposed in the IEEE 802.11e standard. The EDCA protocol inherits the conventional DCF’s CSMA/CA scheme with the enhanced RTS/CTS handshaking process. Furthermore, four prioritized ACs are defined in EDCA in order to support different types of network traffic. The QoS requirements for each AC is defined by selecting feasible values of the CW size and AIFS length. It is intuitive to observe that higher priority AC should possess smaller values of CW and AIFS sizes. Each AC will wait for its AIFS length and independently select its own backoff number. Until the backoff number for a specific AC has been decremented to zero, the corresponding AC can initiate a RTS frame for channel contention. Each AC within a WS is considered as a stand-alone entity to contend with the ACs both in the same WS and the other WSs for channel access in the network. Furthermore, HCCA is designed to be a
modified version of PCF which provides prioritized ACs to conduct centralized polling-based channel access.
3.2 Gentle DCF Protocol
The GDCF algorithm in [5], [6] modifies the conventional backoff scheme within the IEEE 802.11 protocol for the enhancement of network throughput. The major parameter in the GDCF scheme is the design of a successful counter for recording the number of consecutive successful transmis-sions. The counter will be reset to zero every time a failed transmission occurs. Similar to the conventional DCF scheme, the CW size will be doubled if the packet for the WS is failed in transmission. On the other hand, in the case of successful packet transmission, the CW size by adopting the GDCF protocol will not be reset back to the minimal CW size as the DCF scheme. The CW size will be maintained until there exist c successful transmissions of data packets, and the size will be halved only after the c consecutive transmissions have been achieved. Consequently, the packet collision owing to the channel contention can be alleviated with the adoption of the GDCF scheme. How-ever, the network throughput can only be enhanced with the reduction of RTS packet collisions while there exists a large number of WSs within the network. In the case that there is a comparably smaller number of WSs in the considered network, the design of an enlarged CW size will degrade the network throughput, which consequently results in elongated transmission delay.
4
P
ROPOSEDARCR P
ROTOCOLThe design concept of the proposed ARCR algorithm is to adaptively provide reservation periods for specific WSs within the contention-based channel access networks. In order to promote the network throughput without incur-ring excessive control overhead, the piggyback mechanism [16] is utilized to append the control messages after either the data or the acknowledge (ACK) packets. The piggy-backed fields introduced by the ARCR protocol are applied in order to alleviate the RTS/CTS/ACK overheads, to regulate the backoff processes, and to schedule the transmission orders, which ultimately can achieve higher network throughput. With the enhanced channel utilization by adopting the proposed ARCR scheme, it will be illustrated in the numerical evaluation that the overheads from the piggybacked control fields are observed to be insignificant. The detailed functionalities of the proposed ARCR scheme is described in Section 4.1. The examples of both ideal and realistic network scenarios for the proposed scheme are addressed in Sections 4.2 and 4.3.
4.1 Functional Description
As a node intends to transmit data packets within an IEEE 802.11 AP-based network, a RTS/CTS exchange process will be initiated before the transmission of data packets. In the case that there are additional data packets to be delivered, a control field called table-adding request (TAR) will be appended after the data packet to perform piggyback, i.e., denoted as DATAþTAR. On the other hand, the conventional DCF scheme will be adopted if there is no further data packet to be dispatched. The TAR control field is defined as follows:
Definition 1 (TAR). TAR is defined as a control field used to inform the AP that a WS is intending to join the AP’s reservation table.
After receiving the DATAþTAR packet from the WS, the AP will record the MAC address of the corresponding WS within its reservation table T ¼ fTrðSÞ; 8r; Sg that consists a
list of prioritized numbering for each WS, e.g., T0ðAÞ
indicates that WS A is recorded in the first entry (i.e., r¼ 0) of the reservation table T. Consequently, the AP will respond with an ACK packet associated with a piggybacked field called next transmission order NTOðrÞ, which is defined as follows:
Definition 2 (NTOðrÞ).Next Transmission Order (NTOðrÞ) is defined as a control field adopted by the AP to inform a WS that its order for the next transmission is r.
For example, r ¼ 0 indicates that the WS is recorded at the top of the reservation list T, which will be the next WS to conduct packet transmission. Therefore, each WS that are recorded in the reservation table will be informed by the AP with the ACKþNTOðrÞ packet. By adopting the ARCR scheme, the random backoff number karcrfor the WS will be
selected based on the corresponding index r as
karcr¼ U½0; 20W 1; r¼ 0; U½2r1W ; 2rW 1; 1 r M; U½‘ 2M1W ; u 2M1W 1; r > M; 8 < : ð1Þ
where ‘ ¼ r M þ 1, u ¼ r M þ 2, and the parameter M denotes the maximum number of backoff stage. According to the transmission order r, it can be observed from (1) that each specific WS S within the table entry TrðSÞ will possess a
distinct range of values for its corresponding random backoff number karcr. This design will assure that small
value of r will result in smaller random backoff number karcr.
Consequently, based on the reservation system of the ARCR scheme, the WS with the smallest value of r (i.e., at the top of the reservation table) will be ensured to acquire the channel access comparing with the other WSs within the table T. It is also noticed that the backoff scheme is transformed from exponential to linear increase for the purpose of limiting the range of random number karcrafter r > M.
Definition 3 (RTS-R).The RTS-R packet signifies the initiation of the reservation period, which is delivered by the WS after acquiring the ACKþNTOðrÞ packet from the AP.
After the WS is informed by the AP that it will be the next station to conduct packet transmission, the WS is ready to transmit its RTS-R packet in order to initiate the reservation period. The transmission of RTS-R packet will be delivered from the WS after it has succeeded in contending the channel access by adopting its random
backoff number karcr as in (1). After the RTS-R/CTS
handshake has been completed, either the DATAþTAR packet or the DATA packet will be transmitted from the WS to the AP. Once the data transmission has been accom-plished, the table entry TrðSÞ will remain in or be removed
from the reservation table if the DATAþTAR packet or the DATA packet is transmitted, respectively. Furthermore, in the case that there are remaining table entries within T, the
AP will transmit its ACK packet appended with a request for data (RFD) field toward the WS that is recorded within the next table entry. The RFD field is defined as follows: Definition 4 (RFDðrÞ). RFDðrÞ is defined as a control field
utilized by the AP to inform the rth WS in the reservation table that it can conduct packet transmission after waiting for a SIFS duration.
The ACKþRFDðrÞ packet is employed to serve as the indication message from the AP to the WS for requesting the next data transmission, which is delivered within the reservation period. Without conducting the backoff process, the corresponding WS can immediately transmit its
DATAþTAR (or DATA) packet to the AP after a SIFS
interval. The procedures for transmitting the ACK þ RFDðrÞ packet will be continuously conducted until all the table entries within the reservation table T have been processed. The ARCR algorithm will be switched from the reservation-based system back to the contention-based DCF scheme. It is especially noticed that there is only one RTS-R packet required for channel contention within the entire reservation period. With the exploration of adaptive reservation period, the proposed ARCR scheme can reduce packet collision from the RTS packets, which effectively increases the channel utilization.
Furthermore, the fairness for packet transmission be-tween the WSs is also considered within the reservation period of the proposed ARCR scheme. All WSs within the reservation table will be scheduled by the AP based on the round-robin fashion in order to maintain the fairness for packet transmission. Considering that all the WSs con-tinuously have data packets to be delivered, i.e., the
DATAþTAR packets are always transmitted by the WSs,
the WS that is informed by the AP with the order r (i.e., NTOðrÞ) will be assigned with the order of r 1 for its next transmission with NTOðr 1Þ. It is noted that the WS with the order of r ¼ 0 will therefore be assigned with the maximum value of r for its next transmission order.
Fig. 1 shows the flowchart for each WS by adopting the proposed ARCR protocol. The transitions between the conventional DCF scheme and the ARCR algorithm is also illustrated. Either the WS failed in packet transmission or it has no further data to be delivered, the ARCR scheme will be switched back to the DCF protocol with the implementa-tion of random backoff scheme for packet retransmission. Different types of transmission scenarios will be exempli-fied in the following two sections.
4.2 Ideal Network Scenarios
Fig. 2a shows an example for an ideal network scenario by exploiting the proposed ARCR algorithm. In this case, it is assumed that the channel is error-free without the occurrence of packet collision. Three WSs A, B, and C within the network are intending to continuously transmit data packets to the AP. At the beginning time instant t1, no entry is recorded within
the AP’s reservation table T; while the three WSs are contending for channel access by adopting the IEEE 802.11 DCF mechanism. It is assumed that WS A acquires the channel access after the contention, the conventional RTS/ CTS exchange will be conducted between WS A and the AP. The DATAþTAR packet will be delivered from WS A to the AP, where the TAR field indicates the request from node A
that it still possesses remaining data packet to be transmitted. After the table entry T0ðAÞ has been added to the reservation
table T, the AP will transmit the ACKþNTOð0Þ packet to WS Aindicating that it will be the first WS to conduct packet transmission in the next reservation period. It is noted that the NAV vector is utilized to suspend potential channel sensing and packet transmissions from both WSs B and C during the interaction time interval between WS A and the AP.
After WS A completes its first transmission with the AP, the three WSs will continue to compete for the channel
access at time t2. Since WS A has received the NTOð0Þ
packet from the AP, it will employ the random backoff scheme in (1) by adopting the ARCR scheme; while the conventional backoff scheme from the DCF mechanism will be applied to both WSs B and C. Considering that WS B has obtained the channel access, similar procedures between WS B and the AP will be taken place, i.e., the transmission of RTS, CTS, DATAþTAR, and ACKþNTOð1Þ packets between WS B and the AP. The table entry T1ðBÞ will also
be included in the AP’s reservation table T. Due to the reason that both WSs A and B have received the NTOðrÞ packets, the random backoff scheme from (1) is exploited for both nodes at time instant t3; while the conventional
DCF backoff mechanism will be adopted by WS C. Owing to the special design of the random backoff algorithm as in (1), the WS with the smallest r value (i.e., WS A in this case) will be ensured to have the highest opportunity to acquire the channel access among the WSs recorded in the table. Therefore, there will only be either WS A or C that will finally win the channel access after the time instant t3.
Assuming that WS A acquires the channel access after t3,
the RTS-R packet will be initiated by WS A to start the reservation period for both WSs A and B, i.e., tR;1as shown
in Fig. 2. After the reception of the DATAþTAR packet from WS A, the AP will respond with the ACK þ RFDð1Þ packet where the ACK packet is intended for WS A and the RFDð1Þ field is targeting for WS B. Based on the received RFDð1Þ message from the AP, WS B will terminate its backoff process and conduct the transmission of DATAþTAR packet to the AP after a SIFS time interval. It is noted that the cancelation of the backoff process for WS B can reduce the channel idle time, and consequently promotes the network throughput. After the completion of the DATAþTAR packet from WS B, Fig. 1. The flowchart for the behavior of WS by adopting the proposed
ARCR protocol.
the AP will respond with an ACKþNTOð0Þ þ NTOð1Þ packet where the ACKþNTOð0Þ packet is delivered to WS B and NTO(1) packet is intended for WS A. It is noticed that the transmission order within the reservation system has been swapped for the consideration of fairness, i.e., T¼ fT0ðBÞ; T1ðAÞg.
Assuming that WS C finally acquires the channel access at t4, the table-adding procedures will be conducted for WS
C after the completion of its data transmission, i.e.,
T¼ fT0ðBÞ; T1ðAÞ; T2ðCÞg. Consequently, at t5, the
reserva-tion period tR;2 will be utilized to conduct packet
transmission for all the three WSs that are recorded within the reservation table T. Afterwards, the transmission order will be rotated for the purpose to ensure the transmission fairness, i.e., T ¼ fT0ðAÞ; T1ðCÞ; T2ðBÞg. In the case that
there exists a new WS (e.g., WS D) that joins the network at the time instant t6, channel contention will occur between
WSs A and D. Otherwise, a new reservation period tR;3
will be initiated to continuously transmit the packets from WSs A, B, and C.
4.3 Realistic Network Scenarios
Fig. 2b shows the examples for the proposed ARCR scheme to alleviate the packet collision under a realistic network scenario. In this case, it is assumed that the channel is error-prone with the occurrence of RTS/RTS-R packet collision. First of all, the adaptive adjustment of the ARCR scheme owing to the RTS-R packet collision is considered. Assum-ing that the AP’s reservation table is recorded as T ¼ fT0ðAÞ; T1ðCÞ; T2ðBÞg before the time instant t1. Since WS A
is situated at the top of table T, it will possess the smallest backoff number karcr according to (1) which results in the
acquisition of channel access.
WS A will initiate the RTS-R packet to the AP, and it is assumed to be unsuccessfully transmitted due to packet collision with WS D. Without receiving the CTS packet from the AP, WS A will change its channel access mechanism from the ARCR algorithm back to the conventional DCF scheme. As shown in the flowchart from Fig. 1, the random number kdcf will be selected via the original DCF scheme
with backoff stage i ¼ 0, i.e., within the range of
U½0; 20W 1. On the other hand, since WS C did not
obtain the RFDð1Þ field from the AP, it will continue its random backoff process. Therefore, both WSs A and C will be involved in contending the channel access at time t2.
Considering that WS C is successful in acquiring the channel, it will start the reservation period by sending the RTS-R packet to the AP. With the reception of the RTS-R packet, the AP will notice that its first table entry T0ðAÞ is
not available for data transmission. Consequently, the entry
T0ðAÞ is removed such that the reservation table will
become T ¼ fT0ðBÞ; T1ðCÞg.
The transmission priorities that are recorded within the reservation table will be changed after the packet transmis-sions for both WSs B and C, i.e., T ¼ fT0ðBÞ; T1ðCÞg. For the
next reservation period starting from t3, after WS B
accomplishes its packet transmission with the AP, WS C will receive the RFDð1Þ message from the AP and start to dispatch its DATAþTAR packet. Considering that the DATAþTAR packet failed in transmission due to the occurrence of packet error, the AP will wait for a period required for successful packet transmission, i.e., the AP
time-out period, to recognize this situation and consequently remove WS C from its reservation table as T ¼ fT0ðBÞg. It is
noted that if there are still other table entries recorded behind the removed table entry, the AP will continue to initiate the RFD message to the remaining WSs for packet transmissions. On the other hand, without any further acknowledgment from the AP, WS C will change its channel access mechanism from the ARCR algorithm back to the DCF scheme. At time t4, all the three WSs will be in the
process to contend for channel access, and similar proce-dures are implemented to conduct packet transmission.
Similar processes can be examined as above in the case that either the ACK+NTO or the ACK+RFD packet failed in its transmission from the AP to the corresponding WS. The AP will remove the table entry for the WS after waiting for the AP time-out period; while the WS will be adaptively switched back to its original DCF mode for channel contention.
5
T
HROUGHPUTA
NALYSIS OF THEP
ROPOSEDARCR P
ROTOCOLAnalytical study is performed in order to explore the benefits of the proposed ARCR protocol. The backoff process of the DCF scheme is first modeled by the Markov chain model in Section 5.1. The probability for a WS to join the reservation table is derived in Section 5.2. As a consequence, the analytical model of throughput perfor-mance for the proposed ARCR protocol will be obtained in Section 5.3.
5.1 Backoff Process of the DCF Scheme
There are existing research [20], [21], [22], [23], [24], [25] establishing the analytical models for the backoff process of the DCF scheme under different considerations, e.g., fading channel [21], backoff suspension [24], or retry limit [25]. The two-dimensional Markov chain model utilized in [21] is adopted as the baseline model to analyze the random backoff process in the proposed ARCR protocol. As shown in Fig. 3, the parameter p represents the probability of failed transmis-sion due to packet collitransmis-sions or channel noise. Wi¼ 2iW is
defined as the backoff window size at the stage i for
0 i M, where W denotes the minimum backoff window
size. sðtÞ and bðtÞ are defined as the stochastic processes representing the backoff stage and the backoff time counter of a WS at time t, respectively. It is noted that discrete and integral timescale for the decrements of backoff time counter is adopted in the analysis. The backoff time counter will decrease by one in a slot time as the channel is sensed idle. Let bi;k denotes the stationary distribution of the
two-dimensional stochastic process fsðtÞ; bðtÞg as a WS lies at the ith backoff stage with its counter equal to k. As shown in Fig. 3, a WS in backoff stage i will randomly select a number within ½0; Wi 1 and start to count down if the channel is
sensed to be idle. The WS will successfully transmit with probability 1 p after the counter k decreases to zero. It will consequently be reset to the minimum window size, i.e., i¼ 0, for the next channel contention. On the other hand, the WS will be at the ði þ 1Þth backoff stage if collision happens for packet transmission. In the case that the current backoff stage is M and the transmission fails, the next
backoff stage will still remain at the stage M. The relation-ship between each state is derived as follows:
Pðbi;kjbi;kþ1Þ ¼ 1; 0 k Wi 2; 0 i M; Pðb0;kjbi;0Þ ¼ ð1 pÞ=W0; 0 k W0 1; 0 i M; Pðbi;kjbi1;0Þ ¼ p=Wi; 0 k Wi 1; 1 i M; PðbM;kjbM;0Þ ¼ p=WM; 0 k WM 1: 8 > > < > > : ð2Þ
It is noticed that each steady-state probability bi;k can be
expressed as a function of b0;0after transformation based on
the equations in (2). Since the sum of all the states will be equal to 1, namelyPMi¼0PWi1
k¼0 bi;k¼ 1, b0;0can be obtained as
b0;0¼
2ð1 pÞð1 2pÞ
ð1 2pÞðW þ 1Þ þ pW ½1 ð2pÞM: ð3Þ
Let be defined as the probability that a WS transmits a RTS packet in a randomly selected time slot. Based on the model in Fig. 3, a WS can transmits its RTS packets only if the backoff counter k reaches zero. Therefore, the parameter can be acquired as
¼X M i¼0 bi;0¼ b0;0 1 p¼ 2ð1 2pÞ ð1 2pÞðW þ 1Þ þ pW ½1 ð2pÞM: ð4Þ In order to solve and p in (4), another relationship
between these two parameters should be obtained. Let Pf
be denoted as the packet error rate due to the existence of channel noises and Pc be the probability that the packet
issued by one WS collides with those from other WSs. Noted that the packet error rate Pf can be computed from
the bit error rate which is derived from signal-to-noise ratio (SNR) of the channel states. Assuming that there are N WSs in the wireless network, Pc can be interpreted as the event
that at least one WS transmits packets among the remaining N 1 WSs, i.e., Pc¼ 1 ð1 ÞN1. Therefore, the
prob-ability of failed transmission p can be obtained as
p¼ Pcþ Pf PcPf¼ 1 ð1 ÞN1þ Pfð1 ÞN1: ð5Þ
By iteratively solving the nonlinear functions (4) and (5), the two parameters and p can therefore be obtained. In the next section, the behavior that whether a WS will become an entry in the reservation table will be depicted.
5.2 Derivation of Reservation Probability Pt
In this section, the major task is to derive the parameter Pt
which represents the transition probability that a WS either is in or will join the reservation table, named as reservation probability. As described in Section 4.3, the WSs will be added into or removed from the AP’s reservation table T according to the proposed ARCR scheme. Therefore, the total number of effective WSs will vary with the transmis-sion events that happen in the network. In the proposed ARCR protocol, the effective WSs are defined as the set which consists of 1) the WSs that adopt the DCF scheme for channel contention and 2) the WS in the first entry of the reservation table T. Consider that the AP has recorded several WSs in its reservation table T. If a WS successfully completes its transmission by applying the DCF scheme, it will be added as the last entry in T and the number of effective WSs will be decreased by one. On the other hand, the number of the effective WSs will be increased by one if any of the WSs recorded in T is forced to be removed from
the table under certain network scenarios. Let ne;r be
referred as the number of the effective WSs in the network on the condition that there are r WSs in the reservation table T. The relationship between the number of WSs r in the reservation table T and the number of effective WSs ne;r
in the network is represented as
ne;r¼ N; r¼ 0;
N r þ 1; 1 r N:
ð6Þ According to (6), if the reservation table T is empty (i.e., r¼ 0), ne;0 will be equal to N and all the effective WSs will
compete the channel by using the DCF scheme. In the case that there is one WS in T, the parameter ne;1will still be equal
to N since the WS in T will need to contend for channel access with the other N 1 WSs that are not in the table. Considering that there are ne;r-effective WSs in the network,
the numbers of WSs reside inside and outside the reservation table T will be N ne;rþ 1 and ne;r 1, respectively.
A WS which joins in or departs from the reservation table Twill affect the degree of channel contention in the wireless network. If a WS joins in the reservation table T, the number of effective WSs will decrease and the occurrence of packet collisions will be reduced. On the other hand, the transmitted packets will potentially suffer from more collisions when the number of effective WSs is increased owing to the departure of WSs from the reservation table T. To simplify the interactions among the WSs, it is assumed that whether a WS will join in or depart from the reservation table is independent to the strategies adopted by the other WSs. Fig. 4 shows the transitions between the steady states Fig. 3. The two-dimensional Markov chain model for the backoff process
of the DCF scheme.
Fig. 4. The Markov model of reservation probability Ptfor the proposed
according to whether a WS will be recorded in the reservation table T. The parameter t is defined as the steady-state
probability that a WS will reside in the reservation table T, which can be obtained as
t¼ tPtþ ð1 tÞPt¼ Pt: ð7Þ
Note that transition probabilities from both states toward t
are assumed equal to simplify calculation complexity. In
order to solve the reservation probability Pt, another
relationship between tand Ptwill be required. Given that
there are r WSs in the reservation table which corresponds to ne;r-effective WSs in the network, the parameters Pc;rand r
are, respectively, denoted as the probabilities of collisions and the events that a WS transmits its RTS packet in a random slot time. Based on the iterative computation between (4) and (5), the set of parameters Pc;r and rcan be
solved from r ¼ 0 to r ¼ N. Moreover, the probability for a WS to be in the reservation table can be contributed to either one of the following two factors: 1) a WS is added into the reservation table T after successfully transmitting packets via channel contention or 2) a WS that exists in table T has conducted successful packet transmission. Therefore, the
parameter Pt can also be regarded as the probability of
successful transmission considering the situations that a WS is either inside or outside of the reservation table. Based on the value of Pc;ras described above, the probability Ptin the
steady state can consequently be derived as Pt¼ XN r¼0 CrNrtð1 tÞNr ne;r N ð1 PfÞð1 Pc;rÞ þ N ne;r N ð1 PfÞ : ð8Þ
It is noted thatne;r
N in (8) is denoted as the probability that a
WS is required to contend with the other WSs in the network. On the other hand, Nne;r
N represents the
prob-ability that the WS resides within the reservation table to be scheduled for packet transmission. Therefore, only the packet error rate Pfis required to be addressed without the
consideration of collision probability Pc;r. By substituting (7)
into (8), the parameters t and Pt can consequently be
obtained by solving the corresponding nonlinear function.
5.3 Throughput Performance of the Proposed
ARCR Protocol
Compared to conventional analytical models for the DCF scheme, the analysis for throughput performance of the proposed ARCR protocol is to further investigate the effect from the reservation table to the channel contention. Let
Ptr;r be the probability that there is at least one WS
transmitting in a slot time while r WSs are recorded in the reservation table T, i.e.,
Ptr;r¼ 1 ð1 rÞne;r: ð9Þ
Moreover, the probability Ps;r is denoted as the event that
exactly one WS occupies the channel without any transmis-sion from the other WSs given that there are r WSs in the reservation table. The probability Ps;rcan be derived as
Ps;r¼
ne;rrð1 rÞne;r1
Ptr;r
: ð10Þ
To obtain the system throughput with r WSs recorded in T, the average payload delivered in successful transmissions
will be considered. The parameter E½Pr represents the
average payload size for one transmission given that there are rðr 6¼ 0Þ WSs in the reservation table T, which can be obtained as E½Pr ¼ ne;r 1 ne;r E½P þ 1 ne;r ðN ne;rþ1ÞE½P ¼ N ne;r E½P ; ð11Þ where E½P denotes the average intended transmitted payload size for each WS. It is noted that ne;r1
ne;r represents
the probability that the transmitters do not reside in the reservation table T, and each of them has payload E½P to be delivered. On the other hand, the fraction 1
ne;r stands for
the transmission probability of the WS that possesses the first transmission priority among all the WSs in the reservation table T. The total payload issued at this case by the entire r WSs in T becomes ðN ne;rþ 1ÞE½P . In the
case that r ¼ 0, all the WSs will adopt the conventional DCF scheme which results in E½Pr¼0 ¼ E½P that can also be
verified by substituting r ¼ 0 in (11).
In order to evaluate the total required time Tav;rfor packet
transmission given that there are r WSs in the reservation table, the time durations owing to packet collisions Tc,
successful transmissions Ts;r, and noise corruptions Tf;rwill
be taken into account. With the consideration of the three events mentioned before, the average required time Tav;rcan
be derived as
Tav;r¼ ð1 Ptr;rÞ þ Ptr;rð1 Ps;rÞTc
þ Ptr;rPs;rð1 PfÞTs;rþ Ptr;rPs;rPfTf;r;
ð12Þ where represents the slot time. The probabilities Ptr;rand
Ps;r can be obtained from (9) and (10), respectively. The
parameter Tc denotes the time for a WS to sense the
occurrence of packet collisions which can be expressed as Tc¼ TRT SRþ þ TCT Sþ þ TSIF Sþ TDIF S; ð13Þ
where is the propagation delay, and the remaining parameters in (13) are indicated by their corresponding subscripts. Noted that TRT S-R represents the required time
for either the RTS or the RTS-R packet since no additional control field is required by adopting the designed RTS-R packet. On the other hand, the required time for successful transmissions can be acquired as
Ts;r¼
ne;r 1
ne;r
½TRT SRþ TCT Sþ TP HY þ TMACþ TE½P
þ TACKþNT Oþ 3TSIF Sþ 4 þ TDIF S
þ 1
ne;r
½TRT Sþ TCT Sþ TSIF Sþ 2 þ TDIF S
þ ðN ne;rþ 1ÞðTP HY þ TMACþ TE½P
þ TACKþ 2TSIF Sþ 2Þ;
ð14Þ
where TE½P , TACKþNT O, TP HY, and TMACare defined as the
required time intervals for transmitting payload, ACKþ
NTOframe, PHY header, and MAC header. Noted that the
time interval for transmitting the designed RFD field is considered within the MAC header. Similar to the concept in (11), the first term in (14) that associated with probability
ne;r1
that adopts the DCF scheme. The second term associated
with probability 1
ne;r indicates the required time for a
successful transmission while the WS resides in the reserva-tion table, which exploits the ARCR protocol to compete the channel access. Furthermore, a transmitter will need to perceive whether its transmission has completed or not according to the reception of ACK packet. Therefore, the required time owing to noise corruption will be equal to that for successful transmissions, i.e., Tf;r¼ Ts;r. Based on (11)
and (12), the average system throughput S can consequently be derived as S¼ PN r¼0CrNrtð1 tÞNrPtr;rPs;rð1 PfÞE½Pr PN r¼0CrNrtð1 tÞNrTav;r ; ð15Þ
where tcan be obtained by solving (7) and (8). It is noted
that the term Ptr;rPs;rð1 PfÞE½Pr in (15) denotes the
expected payload to be transmitted with r WSs in the reservation table. The validation of throughput perfor-mance S in (15) for the proposed ARCR protocol will be conducted in Section 8.1.1.
6
P
ROPOSEDE-ARCR P
ROTOCOLAs is not considered in the DCF mechanism, the IEEE 802.11e EDCA scheme [4] supports different traffic types and fulfills their corresponding QoS requirements. In order to achieve the advancement from the DCF method to the EDCA scheme, the E-ARCR protocol is proposed as the enhanced version of the ARCR scheme in order to fulfill the QoS requirements as specified in the EDCA scheme. The E-ARCR protocol will support four ACs in a WS in order to serve various traffic types which possess different priorities for the competition of channel access. As specified in the standard, the access categories are denoted as AC[Z] with Z ¼ 3; 2; 1, and 0, where AC[3] represents the highest priority and AC[0] has the lowest priority. The special control functions described in Section 6.1 are designed to facilitate the implementation of the E-ARCR protocol. The operations of the proposed E-ARCR scheme is explained with an arbitrary network scenario in Section 6.2.
6.1 Functional Description
In order to provide prioritized ACs for different traffic, four queues in the same WS are utilized as four virtual stations to contend for channel access. Therefore, instead of adopting a
single reservation table as in the ARCR scheme, the proposed E-ARCR protocol exploits four reservation tables in order to record different types of traffic from all the WSs in the network. Each of the four reservation tables will be labeled as TAC½Z which matches with the AC[Z] traffic, where
Z¼ 3; 2; 1, and 0. The control fields similar to Definitions 1 to 4 are utilized in the E-ARCR protocol associated with different AC[Z]s, including TARðZÞ, NTOðZ; rÞ, RTS-RðZÞ, and RFDðZ; rÞ. For example, TARðZÞ is defined as a control field to inform the AP that AC[Z] of a WS is intending to join the reservation table TAC½Z.
Considering different priorities among the AC[Z]s, the initial window size WAC½Zas well as the maximum backoff
stage MAC½Z will be different between the four AC[Z]s.
Based on the information acquired from the control field NTOðZ; rÞ, the random backoff number kearcr;AC½Z for the
specific AC[Z] within a WS will be selected as kearcr;AC½Z
¼
U½0; 20W
AC½Z 1; r¼ 0;
U½2r1W
AC½Z; 2rWAC½Z 1; 1 r MAC½Z;
U½‘ 2MAC½Z1W AC½Z; u 2MAC½Z1W AC½Z 1; r > MAC½Z; 8 > > > > > < > > > > > : ð16Þ
where ‘ ¼ r MAC½Zþ 1 and u ¼ r MAC½Zþ 2. Moreover,
the AIFS value in the EDCA scheme for each AC[Z] is denoted as AIF SAC½Zin order to govern different waiting
time intervals to start the backoff process. Therefore, the parameters WAC½Z, MAC½Z, and AIF SAC½Z for each of the
four AC[Z] can be manipulated to affect different priorities among the ACs.
6.2 Network Scenarios
The operations of the proposed E-ARCR protocol without packet collisions and channel noise are depicted in Fig. 5. To clearly visualize the network behaviors of the proposed E-ARCR scheme, each of the two WSs is associated with two ACs including AC[1] for high priority and AC[0] for low-priority transmission. Therefore, there will be two reserva-tion tables TAC½1and TAC½0exploited within the AP. At the
beginning, all the four ACs contend for channel by adopting the EDCA scheme and there is no entry recorded in AP’s reservation tables, i.e., TAC½1¼ TAC½0¼ fg. As AC[1] of
node A successfully acquires the channel at time t1, WS A
will be added into the reservation table TAC½1 as the first
entry, i.e., TAC½1¼ fT0ðAÞg. After data packets have been
successfully delivered from AC[1] of WS A to the AP, the AP will transmit the ACKþNTOð1; 0Þ packet to WS A which indicates that AC[1] of WS A has the transmission order of 0. At the time instant t2, all the four ACs will continue to
contend for the channel access. As WS A has received the NTO(1,0) packet from the AP, WS A will adopt the E-ARCR scheme with random backoff mechanism as defined in (16); while the other three ACs will employ the conventional backoff scheme from the EDCA algorithm. Considering that AC[0] of WS B wins the channel contention and hence it will be recorded as a new entry in the reservation table as TAC½0¼ fT0ðBÞg. Similarly, in the case that AC[1] of WS B
acquires the channel access at time t3, it will join in the
reservation table TAC½1 as the second entry, i.e., TAC½1¼
fT0ðAÞ; T1ðBÞg, and finally receives the ACKþNTOð1; 1Þ
packet from the AP after packet transmission.
Assuming that AC[1] of WS A obtains the channel access at time t4, the RTS-R(1) packet will be delivered by WS A to
initiate the reservation period for AC[1] in both WSs A and B. After receiving the DATAþTAR(1) packet from WS A, the AP will respond with the ACK+RFD(1,1) packet where the ACK packet is targeting for AC[1] of WS A and the RFD(1,1) packet is for AC[1] of WS B. According to the received RFD(1,1) message from the AP, AC[1] of WS B can deliver DATAþTAR(1) packet without the requirement for channel contention. At the end of the reservation period, the AP will change the entry order within the reservation table TAC½1 in a round-robin manner, i.e., TAC½1 ¼ fT0ðBÞ;
T1ðAÞg. It is noted that similar reservation period will be
implemented for AC[0] of both WSs A and B. Moreover, the proposed E-ARCR scheme can also be implemented in a more realistic network scenarios with the existence of packet collisions and channel noises, which can be extended from the descriptions as addressed in Section 4.3 for the ARCR protocol.
It is noticed that packet collision will not happen in the original ARCR scheme if all the WSs reside within the reservation table under error-free network environments. In the E-ARCR scheme, however, collisions may still exist even though all ACs in WSs are recorded within their corresponding reservation tables in the AP. The reason is contributed to the usage of more than one reservation table in the network. The first entries in those reservation tables will still contend with each other which results in the occurrence of packet collisions. This is considered the trade-offs by adopting the E-ARCR protocol as the QoS require-ment is specified to be fulfilled. The performance of the proposed E-ARCR scheme will be evaluated and compared in Section 8.
7
T
HROUGHPUTA
NALYSIS OF THEP
ROPOSEDE-ARCR P
ROTOCOLThe throughput analysis of the proposed E-ARCR protocol can be regarded as an extension of that for the ARCR scheme addressed in Section 5. It is noted that certain portion of the WSs will still adopt the conventional EDCA scheme for
channel contention; while others utilize the E-ARCR proto-col. Therefore, the backoff process of the EDCA scheme will first be described in Section 7.1. The reservation probability for the E-ARCR scheme and the corresponding network throughput will be derived in Sections 7.2 and 7.3, respectively.
7.1 Backoff Process of the EDCA Scheme
Existing research work [26], [27], [28], [29], [30], [31] has been conducted to analyze the backoff process of the EDCA protocol. An analytical approach for throughput and delay performance of the IEEE 802.11e EDCA scheme has been proposed in [26], [27] in order to observe the effect of different CWs and retry limits for each AC. Three-dimensional Markov Chain has been utilized in [28], [29] to model the EDCA mechanism. On the other hand, two-dimensional Markov Chain is adopted in [30], [31] by dividing the backoff interval into different time zones, which will be employed as the baseline model for analyzing the performance of the proposed E-ARCR scheme with additional consideration of error-prone channel effects. As was specified in previous work, without loss of generality, each of the N WSs is considered to possess two ACs in the analysis, including AC[1] and AC[0].
Apart from considering the WS as a whole in the DCF and ARCR schemes, each AC in a WS is viewed individually in the backoff process by adopting both the EDCA and E-ARCR protocols. The Markov chain model as shown in Fig. 3 can still be applied to the EDCA scheme except that individual AC is considered instated of the
entire WS. Let AC½Z denote the probability that AC[Z]
transmits the RTS packet in a randomly selected time slot, and pAC½Z is defined as the average probability that AC[Z] fails in transmission due to packet collision or frame errors. Similar to (4), the relationship between AC½Zand pAC½Zcan be
acquired as
AC½Z¼ 2ð1 2pAC½ZÞ ½ð1 2pAC½ZÞðWAC½Zþ 1Þ
þ pAC½ZWAC½Zð1 ð2pAC½ZÞ
MAC½Z1: ð17Þ
It is noted that the averaged value pAC½Z is considered in (17) since the fail transmission probabilities are calculated in two different time zones for each AC[Z], which will be explained as follows: In order to distinguish different QoS requirements among distinct ACs, the ACs which belong to higher priorities will start their backoff processes after a shorter AIFS duration. A smaller number will be obtained by the ACs with higher priorities for backoff countdown, and therefore suffer from fewer channel contentions comparing with the ACs of lower priorities. As illustrated in Fig. 6, resulting from the various values of AIFSs, the backoff period can be divided into two different time regions including Fig. 6. The EDCA backoff process after the occurrence of busy medium.
Zones A and B. Let LAand LBbe referred as the numbers of
time slots in Zones A and B, respectively. The entire time duration LAþ LB can be obtained as the maximal backoff
window size MAC½1 of the highest priority AC[1], i.e.,
LAþLB¼ minf2MAC½Z WAC½ZjZ ¼ 1; 0g ¼ 2MAC½1WAC½1. Within
the duration of Zone A, only the high-priority AC[1]s can decrement their backoff numbers and have the chance to transmit their RTS packets. On other other hand, all ACs including both AC[1]s and AC[0]s will contend for channel access in Zone B if there does not exist AC[1] that intends to transmit in Zone A. As can be expected, the states of channel contention for Zones A and B, respectively, will be different. To evaluate the stationary probability of each zone, it is assumed that every AC is independent to each other and the Morkov chain model for state transition between backoff slots is shown in Fig. 7. Suppose that there are nAC½1 AC[1]s
and nAC½0 AC[0]s in the network, and let qX denote the
probability that there does not exist any AC transmitting in Zone X. Therefore, qAand qBcan be calculated as
qA¼ ð1 AC½1Þ nAC½1 ; qB¼ ð1 AC½1Þ nAC½1 ð1 AC½0ÞnAC½0: ( ð18Þ Let zk be referred as the stationary probability that time
slot k locates in the contention zones, which can conse-quently be acquired as zk¼ z1 Qk i¼2qA ; 1 < k LAþ 1; zk¼ z1 QLAþ1 i¼2 qA Qk i¼LAþ2qB ; LAþ 1 < k LAþ LB: ( ð19Þ By associating (19) with the relationshipPLAþLB
k¼1 zk¼ 1, the
probability z1can be derived as
z1¼ 1 qALAþ1 1 qA þ qALAqB 1 qBLB1 1 qB 1 : ð20Þ
Furthermore, let A and B be the stationary probabilities
for a random time slot lies in Zones A and B, respectively. Both parameters can be calculated by incorporating the results from (19) and (20) as
A¼PLk¼1A zk;
B¼PLk¼1B zLAþk:
ð21Þ Since two different zones are considered in the analytical model of the EDCA scheme, additional derivations are required in order to depict the situations of packet collision. Let Pc;AC½Z;Xbe defined as the collision probability of AC[Z]
given that packet collisions occur within Zone X. The collision probabilities for the two types of ACs in contention zones A and B are, respectively, obtained as
Pc;AC½1;A¼ 1 ð1 AC½1Þ nAC½11 ; Pc;AC½1;B¼ 1 ð1 AC½1Þ nAC½11 ð1 AC½0ÞnAC½01; Pc;AC½0;A¼ 0; Pc;AC½0;B¼ 1 ð1 AC½1Þ nAC½1 ð1 AC½0Þ nAC½01 : 8 > > > < > > > : ð22Þ
Noted that the reason for Pc;AC½0;A¼ 0 in (22) is that AC[0] will
only conduct packet transmission within Zone B. Moreover,
let Pc;AC½Z with Z ¼ 1 and 0 be defined as the average
collision probability of AC[Z] in these two contention zones. Since Aþ B¼ 1, the average collision probability Pc;AC½1
and Pc;AC½0can be obtained by averaging (22) as
Pc;AC½1 ¼ Pc;AC½1;AAþPc;AC½1;BB AþB ¼ Pc;AC½1;A Aþ Pc;AC½1;B B; Pc;AC½0 ¼ Pc;AC½0;B: 8 > < > : ð23Þ
The average probability pAC½Z that AC[Z] fails in transmis-sion due to packet collitransmis-sions or channel noises can be acquired from (23) as
pAC½Z ¼ Pc;AC½Zþ Pf ðPc;AC½Z PfÞ ð24Þ
for Z ¼ 1; 0, and Pfdenotes the packet error rate. From (18)
to (24), it is observed that both pAC½1 and pAC½0 are functions
of AC½1 and AC½0; while (17) provides another relationship
between pAC½Zand AC½Zfor Z ¼ 1 and 0. Consequently, the
unknown parameters pAC½1, pAC½0, AC½1, and AC½0 can be
iteratively solved.
7.2 Derivation of Reservation Probability Pt;AC½Z
The reservation probability Pt;AC½Z will be derived in this
section. It is noted that Pt;AC½Z represents the transition
probability that an AC[Z] of a WS either is in or will join in the reservation table TAC½Zfor Z ¼ 1; 0. As shown in Fig. 4,
the derivation of reservation probability for the ARCR scheme can be extended to the E-ARCR protocol by considering the Markov model for each reservation table TAC½Zwith Z ¼ 1; 0. Let t;AC½Zbe defined as the stationary
probability that an AC[Z] of a WS stays in the reservation table TAC½Z. Similar to (7), the relationship between Pt;AC½Z
and t;AC½Zcan be obtained as
t;AC½Z¼ Pt;AC½Z; ð25Þ
for Z ¼ 1; 0. Another relationship between Pt;AC½Z and
t;AC½Z is required for solving the reservation probability
Pt;AC½Z. Given that there are i AC[1]s and j AC[0]s in the
reservation tables TAC½1 and TAC½0, respectively, the
effective numbers of AC[1]s and AC[0]s that actually contend for channel access become ne;i and ne;j which can
be obtained from (6) by replacing r with i and j. The parameters Pc;AC½Z;X;i;jand AC½Z;i;j are, respectively, denoted
as the collision probabilities and the events that an AC[Z] transmits its RTS packet in a random slot time. Noted that the subscript X in Pc;AC½Z;X;i;jindicates that the probability
is computed for either Zone A or B with X ¼ A or B. By iteratively computing the relationship from (17) to (24), the set of parameters Pc;AC½Z;X;i;jand AC½Z;i;j can be obtained for
i; j¼ 0 to N. It is noticed that the parameters nAC½1 and nAC½0
in (17)-(24) are, respectively, replaced by ne;i and ne;j with
the consideration of reservation tables. Moreover, the reservation probability Pt;AC½Zcan also be regarded as the
Fig. 7. The Morkov chain model for state transition between backoff slots in two different zones.
probability of successful transmission under the situations that an AC[Z] is either inside or outside of its correspond-ing reservation table TAC½Z, i.e.,
Pt;AC½Z¼
XN i¼0
XN j¼0
CiNit;AC½1ð1 t;AC½1ÞNi
CN j j t;AC½0ð1 t;AC½0Þ NjP t;AC½Z;i;j ð26Þ
for Z ¼ 1; 0. It is noted that Pt;AC½Z;i;j in (26) represents the
probability that an AC[Z] of a WS joins in the reservation table TAC½Z (for Z ¼ 1; 0) given that there are i AC[1]s
and j AC[0]s in the reservation tables TAC½1 and TAC½0,
respectively. Both parameters can be derived as Pt;AC½1;i;j¼ A;i;jð1 PfÞ ne;i N ð1 Pc;AC½1;A;i;jÞ þ N ne;i N þ B;i;jð1 PfÞ ne;i N ð1 Pc;AC½1;B;i;jÞ þ N ne;i N ; ð27Þ Pt;AC½0;i;j¼ ð1 PfÞ ne;j N ð1 Pc;AC½0;B;i;jÞ þ N ne;j N ; ð28Þ where A;i;jand B;i;jare extended from (21) by considering
i AC[1]s and j AC[0]s in their corresponding reservation
tables. As a result, the reservation probability Pt;AC½Zin (26)
and t;AC½Z in (25) for Z ¼ 1; 0 can be acquired by solving
the corresponding nonlinear function, which will be utilized in the computation of throughput performance for the E-ARCR protocol.
7.3 Throughput Performance of the Proposed
E-ARCR Protocol
The analytical model for throughput performance of the proposed E-ARCR protocol can be regarded as an extension of that derived for the ARCR scheme in Section 5.3 with additional consideration of different prioritized traffic. As there are i AC[1]s and j AC[0]s in the reservation tables TAC½1 and TAC½0, respectively, the parameter Ptr;AC½Z;i;j is
defined as the probability that there exists at least one AC[Z] to be transmitted in a slot time; while Ps;AC½Z;X;i;jis
referred as the probability that one AC[Z] successfully transmits its packet in Zone X for X ¼ A; B and Z ¼ 1; 0. Therefore, the corresponding probabilities can be obtained as follows:
Ptr;AC½1;i;j¼ 1 ½1 AC½1;i;j
ne;i;
Ptr;AC½0;i;j¼ 1 ½1 AC½0;i;j
ne;j;
ð29Þ and
Ps;AC½1;A;i;j¼
ne;iAC½1;i;j½1AC½1;i;jne;i1
Ptr;AC½1;i;j ;
Ps;AC½1;B;i;j
¼ne;iAC½1;i;j½1AC½1;i;jne;i1½1AC½0;i;jne;j1
Ptr;AC½1;i;j ;
Ps;AC½0;B;i;j
¼ne;jAC½0;i;j½1AC½1;i;jne;i½1AC½0;i;jne;j1
Ptr;AC½0;i;j ; 8 > > > > > > > > < > > > > > > > > : ð30Þ
where AC½Z;i;j for Z ¼ 1; 0 can be computed from previous
section given that there exists ne;i AC[1]s and ne;j AC[0]s
contending for the channel access. Furthermore, it is required to calculate the average payload size in each transmission for the E-ARCR protocol. Let E½PAC½Z;i;j be
the average payload size of AC[Z]s in a transmission while there are i AC[1]s and j AC[0]s in the reservation tables TAC½1 and TAC½0, respectively. The parameter E½PAC½Z;i;j
can be obtained as E½PAC½1;i;j ¼
h
A;i;jPtr;AC½1;i;jPs;AC½1;A;i;j
þ B;i;jPtr;AC½1;i;jPs;AC½1;B;i;j
i
ð1 PfÞnNe;iE½P ;
E½PAC½0;i;j ¼ B;i;jPtr;AC½0;i;jPs;AC½0;B;i;jð1 PfÞ
N ne;jE½P ; 8 > > > < > > > : ð31Þ
for Z ¼ 1; 0, and E½P denotes the average payload size for both AC[1] and AC[0]. Therefore, the average slot time Tav;i;jfor a transmission can be written as
Tav;i;j¼ A;i;j A;i;jþ B;i;j B;i;j; ð32Þ
where X;i;j is referred as the average slot time utilized for a
transmission in Zone X as
A;i;j ¼ PI;A;i;j þ Ptr;AC½1;i;jPs;AC½1;A;i;j Ts;i
þ ½1 PI;A;i;j Ptr;AC½1;i;jPs;AC½1;A;i;j Tc;
B;i;j ¼ PI;B;i;j þ Ptr;AC½1;i;jPs;AC½1;B;i;j Ts;i
þ Ptr;AC½0;i;jPs;AC½0;B;i;j Ts;j
þ ½1 PI;B;i;j Ptr;AC½1;i;jPs;AC½1;B;i;j
Ptr;AC½0;i;jPs;AC½0;B;i;j Tc:
ð33Þ
It is noted that PI;X;i;jin (33) denotes the probability that the
channel is idle in Zone X, which can be acquired as PI;A;i;j¼ ½1 AC½1;i;j ne;i ; PI;B;i;j¼ ½1 AC½1;i;j ne;i ½1 AC½0;i;j ne;j : ð34Þ The time Tcfor an AC of a WS to sense the collisions can be
obtained similar to Tc in (13) with additional consideration
of QoS requirement, i.e.,
Tc¼ TRT SRþ þ TCT Sþ þ TSIF Sþ TAIF SAC½1: ð35Þ
On the other hand, similar to (14), both the required time Ts;k (for k ¼ i or j) for a successful transmission and the
time duration of failed transmission Tf;k are considered
equal with the same reason as described in Section 5.3. Since Tf;k¼ Ts;k, both values are combined and utilized in (33) as
Ts;k¼
ne;k 1
ne;k
½TRT SRþ TCT Sþ TP HY þ TMACþ TE½P
þ TACKþNT Oþ 3TSIF Sþ 4 þ TAIF SAC½1
þ 1
ne;k
½TRT SRþ TCT Sþ TSIF Sþ 2 þ TAIF SAC½1
þ ðN ne;kþ 1ÞðTP HY þ TMACþ TE½P
þ TACKþNT Oþ 2TSIF Sþ 2Þ:
ð36Þ
By incorporating t;AC½Z in (25), E½PAC½Z;i;j in (31), and
Tav;i;j in (32), the average system throughput SAC½Z
conditioned on i AC[1]s and j AC[0]s within the reserva-tion tables TAC½1 and TAC½0 can be derived as (37) for
Z¼ 1; 0. The throughput performance SAC½Zin (37) for the
proposed E-ARCR scheme will be validated and evaluated in Section 8.2.1.
SAC½Z¼
XN i¼0
XN j¼0
CiNit;AC½1ð1 t;AC½1ÞNi
CjNjt;AC½0ð1 t;AC½0ÞNj E½PAC½Z;i;j
! , XN i¼0 XN j¼0
CiNit;AC½1ð1 t;AC½1ÞNi
CjNjt;AC½0ð1 t;AC½0ÞNj Tav;i;j
! :
ð37Þ
8
P
ERFORMANCEE
VALUATIONIn this section, the performance of the proposed ARCR and E-ARCR protocols will be validated and compared with existing schemes via the well-developed network simulator (NS-2) [32]. All the simulation runs will be conducted for 100 seconds. Performance validation and comparison for the ARCR scheme are conducted in Sections 8.1.1 and 8.1.2; while that for the E-ARCR protocol are shown in Sections 8.2.1 and 8.2.2.
8.1 Performance Validation and Comparison for the
ARCR Protocol
8.1.1 Performance Validation
In order to validate the analytical model for the proposed ARCR scheme, the system throughput S as derived in (15) is compared with simulation results as shown in Figs. 8 and 9. Noted that the legends “ana” and “sim” in both figures represent the results from analytical model and simulations, respectively. The system parameters and MAC configura-tions based on IEEE 802.11b standard are listed in Table 1, and saturation traffic is assumed for each WS to deliver its data packets. Fig. 8 shows the performance validation for throughput performance versus the number of WSs (N) under BER ¼ 0, 105, and 104. It can be intuitively observed
that the system throughput increases as the total number of WSs in the network is augmented. Moreover, Fig. 9 illustrates the throughput versus BER under N ¼ 5; 10, and 20. The throughput performance decreases as the BER values
are increased. It can be seen from both figures that the proposed analytical model can match with the simulation results under different numbers of WSs and BER values. 8.1.2 Performance Comparison
As shown in Figs. 10, 11, 12, and 13, the proposed ARCR protocol is compared with the DCF and GDCF schemes [5], [6] through a series of simulations in terms of both the number of WSs and the BER values. The system parameters in Table 1 are utilized in performance comparison with saturation traffic considered for each WS. It is also assumed that the successful counter c of GDCF is set equal to 2. Fig. 10 shows the performance comparison of system throughput w.r.t. different numbers of WSs under BER ¼ 0 and 105.
It can be observed that the proposed ARCR scheme possesses higher throughput performance than the other two protocols under different numbers of WSs, e.g., around 50 percent gain at N ¼ 10 under error-free channel condi-tion. Noted that the GDCF method is slightly superior to the DCF scheme with more number of WSs in the network. The reason is that the GDCF scheme has higher probability of staying at the stages with larger backoff window sizes compared to the DCF protocol. Less packet collisions will be Fig. 8. Performance validation for the ARCR protocol: system
throughput versus number of WSs. Fig. 9. Performance validation for the ARCR protocol: systemthroughput versus BER.
TABLE 1 System Parameters
incurred by adopting the GDCF scheme especially under larger number of WSs, which results in comparably larger system throughput. Fig. 11 illustrates the comparison of throughput performance versus different BER values under
N ¼ 5 and 50. The proposed ARCR protocol still
outper-forms the other two schemes under various BER values, e.g., around 33 percent gain at BER = 105under N ¼ 5 scenario.
It can also be observed that the system throughput of three schemes decrease and converge with the augmentation of BER values. At higher BER values, the proposed ARCR protocol behaves similar to the DCF scheme since almost all the WSs in the network will be removed from the reservation table due to occurrence of packet error . On the other hand, with higher BER values, the GDCF method is also compar-able to the DCF scheme owing to the reason that its backoff stage will eventually remain at the maximum value.
Moreover, the proposed ARCR protocol is compared with the distributed DCF scheme and the centralized PCF protocol given that the arrival rate is constant bit rate (CBR) and the queue size is equal to 50 in each WS. In order to
illustrate the pure reservation-based system, the PCF scheme is implemented only with the CFP while the CP is not considered in performance comparisons. It is assumed that there are two types of WSs in the network, including the WSs with high packet arrival rate (1 bits/sec or bps)
and with low packet arrival rate (2bps). Note that the unit
of bps is applied to represent packet arrival rate. Let n1and n2 be, respectively, defined as the numbers of WSs with 1 and 2 as the packet arrival rates, the corresponding
average throughput for each WS with 1 and 2 is,
respectively, denoted as 1 and 2 with the unit of bps. It is considered that there are total of 10 WSs in the network for performance comparison, i.e., n1þ n2 ¼ 10.
The performance comparisons of average throughput for each WS (i.e., either 1 or 2) versus the number of WSs with the packets arrive rate equal to 1are shown in Fig. 12.
Noted that the packet arrival rates 1¼ 2 Mbps and 2¼
200Kbps, and the number of WSs with packets arrival rate Fig. 10. Performance comparison for the ARCR protocol: system
throughput versus number of WSs.
Fig. 11. Performance comparison for the ARCR protocol: system throughput versus BER.
Fig. 12. Performance comparison for the ARCR protocol: average throughput of each WS versus number of WSs with 1 (n1)
(n2 ¼ 10 n1, 1¼ 2 Mbps, and 2¼ 0:2 Mbps).
Fig. 13. Performance comparison for the ARCR protocol: average throughput of each WS versus packet arrival rate 2(n1 ¼ 2, n2 ¼ 8,
![Fig. 16. Performance validation for the E-ARCR protocol: system throughput versus number of WSs for the BE (AC[1]s) and BK (AC[0]s) traffic.](https://thumb-ap.123doks.com/thumbv2/9libinfo/7784167.151129/16.850.442.801.72.356/performance-validation-arcr-protocol-throughput-versus-number-traffic.webp)
