Interclass Collision Protection for IEEE e Wireless LANs

Similar documents
Voice capacity of IEEE b WLANs

Analysis of Collaborative Distributed Admission Control in x Networks

THE IEEE standard for wireless local area networks

IEEE E: QOS PROVISIONING AT THE MAC LAYER YANG XIAO, THE UNIVERSITY OF MEMPHIS

Analysis of QoS in WLAN

136 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 6, NO. 1, JANUARY Ching-Ling Huang and Wanjiun Liao, Senior Member, IEEE

USER CLASS BASED QoS DIFFERENTIATION IN e WLAN

Department of Electrical and Computer Systems Engineering

Simulation Based Analysis of FAST TCP using OMNET++

Real-Time Guarantees. Traffic Characteristics. Flow Control

UBICC Publishers 2008 Ubiquitous Computing and Communication Journal

Extension and Comparison of QoS-Enabled Wi-Fi Models in the Presence of Errors

Performance analysis of distributed cluster-based MAC protocol for multiuser MIMO wireless networks

A Fair MAC Algorithm with Dynamic Priority for e WLANs

A Free-Collision MAC Proposal for Networks

A Fair Access Mechanism Based on TXOP in IEEE e Wireless Networks

Efficient QoS Provisioning at the MAC Layer in Heterogeneous Wireless Sensor Networks

Adaptive Network Resource Management in IEEE Wireless Random Access MAC

Efficient Backoff Algorithm in Wireless Multihop Ad Hoc Networks

RAP. Speed/RAP/CODA. Real-time Systems. Modeling the sensor networks. Real-time Systems. Modeling the sensor networks. Real-time systems:

Performance Analysis of Markov Modulated 1-Persistent CSMA/CA Protocols with Exponential Backoff Scheduling

Re-routing Instability in IEEE Multi-hop Ad-hoc Networks *

WIRELESS communication technology has gained widespread

Fast Retransmission of Real-Time Traffic in HIPERLAN/2 Systems

An Analytical Model for IEEE Point-to-Point Link

Research Article A Comparison Performance Analysis of QoS WLANs: Approaches with Enhanced Features

Performance Evaluation of IEEE e based on ON-OFF Traffic Model I. Papapanagiotou PhD. Student

A Game Theory based Contention Window Adjustment for IEEE under Heavy Load

Achievable Bandwidth Estimation for Stations in Multi-Rate IEEE WLAN Cells

Virtual Machine Migration based on Trust Measurement of Computer Node

EFT: a high throughput routing metric for IEEE s wireless mesh networks

Compiler Design. Spring Register Allocation. Sample Exercises and Solutions. Prof. Pedro C. Diniz

Performance Analysis of the IEEE MAC Protocol over a WLAN with Capture Effect

Parallelism for Nested Loops with Non-uniform and Flow Dependences

Video Proxy System for a Large-scale VOD System (DINA)

A Binarization Algorithm specialized on Document Images and Photos

AADL : about scheduling analysis

Efficient Distributed File System (EDFS)

An Energy-Efficient MAC Protocol for WSNs: Game-Theoretic Constraint Optimization with Multiple Objectives

Mobility Based Routing Protocol with MAC Collision Improvement in Vehicular Ad Hoc Networks

Wishing you all a Total Quality New Year!

Impact of the Parameterization of IEEE Medium Access Layer on the Consumption of ZigBee Sensor Motes

IEEE n Aggregation Performance Study for the Multicast

A Wireless MAC Protocol with Efficient Packet Recovery

Experimental Tuning of the AIFSN Parameter to Prioritize Voice Over Data Transmission in E WLAN Networks

Performance Analysis of Beacon-Less IEEE Multi-Hop Networks

Applications and Challenges of the e EDCA Mechanism: An Experimental Study

Comparisons of Packet Scheduling Algorithms for Fair Service among Connections on the Internet

DESIGNING TRANSMISSION SCHEDULES FOR WIRELESS AD HOC NETWORKS TO MAXIMIZE NETWORK THROUGHPUT

VRT012 User s guide V0.1. Address: Žirmūnų g. 27, Vilnius LT-09105, Phone: (370-5) , Fax: (370-5) ,

APPLICATION OF PREDICTION-BASED PARTICLE FILTERS FOR TELEOPERATIONS OVER THE INTERNET

Dynamic Bandwidth Allocation Schemes in Hybrid TDM/WDM Passive Optical Networks

Priority-Based Scheduling Algorithm for Downlink Traffics in IEEE Networks

Load Balancing for Hex-Cell Interconnection Network

An Optimal Algorithm for Prufer Codes *

A General Model of Wireless Interference

A mathematical programming approach to the analysis, design and scheduling of offshore oilfields

CS 268: Lecture 8 Router Support for Congestion Control

A Sub-Critical Deficit Round-Robin Scheduler

APPLICATION OF PREDICTION-BASED PARTICLE FILTERS FOR TELEOPERATIONS OVER THE INTERNET

QoS-aware composite scheduling using fuzzy proactive and reactive controllers

An Approximation to the QoS Aware Throughput Region of a Tree Network under IEEE CSMA/CA with Application to Wireless Sensor Network Design

X- Chart Using ANOM Approach

Aggregated traffic flow weight controlled hierarchical MAC protocol for wireless sensor networks

A fair buffer allocation scheme

Enhanced Markov Chain Model and Throughput Analysis of the Slotted CSMA/CA for IEEE Under Unsaturated Traffic Conditions

Dynamic Bandwidth Provisioning with Fairness and Revenue Considerations for Broadband Wireless Communication

Scheduling and queue management. DigiComm II

Hermite Splines in Lie Groups as Products of Geodesics

Fibre-Optic AWG-based Real-Time Networks

WITH rapid improvements of wireless technologies,

Real-time interactive applications

1978 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 34, NO. 7, JULY 2016

Transmit Power Control Algorithms in IEEE h Based Networks

QoS Bandwidth Estimation Scheme for Delay Sensitive Applications in MANETs

PERFORMANCE ANALYSIS OF ROUTING ALGORITHMS OF RD-C/TDMA PACKET RADIO NETWORKS UNDER DYNAMIC RANDOM TOPOLOGY1

Efficient Content Distribution in Wireless P2P Networks

The Impact of Delayed Acknowledgement on E-TCP Performance In Wireless networks

Combined SINR Based Vertical Handoff Algorithm for Next Generation Heterogeneous Wireless Networks

Avoiding congestion through dynamic load control

Mathematics 256 a course in differential equations for engineering students

A Backoff Algorithm for Improving Saturation Throughput in IEEE DCF

Instantaneous Fairness of TCP in Heterogeneous Traffic Wireless LAN Environments

TIME-DRIVEN ACCESS AND FORWARDING IN IEEE MESH NETWORKS

Utility Constrained Energy Minimization In Aloha Networks

Wireless Communication

Virtual Memory. Background. No. 10. Virtual Memory: concept. Logical Memory Space (review) Demand Paging(1) Virtual Memory

Problem Set 3 Solutions

TECHNIQUE OF FORMATION HOMOGENEOUS SAMPLE SAME OBJECTS. Muradaliyev A.Z.

An Inter-Piconet Scheduling Algorithm for Bluetooth Scatternets

Neural Network Control for TCP Network Congestion

S1 Note. Basis functions.

A Semi-Distributed Load Balancing Architecture and Algorithm for Heterogeneous Wireless Networks

THere are increasing interests and use of mobile ad hoc

Improving Low Density Parity Check Codes Over the Erasure Channel. The Nelder Mead Downhill Simplex Method. Scott Stransky

An Application of the Dulmage-Mendelsohn Decomposition to Sparse Null Space Bases of Full Row Rank Matrices

DEAR: A DEVICE AND ENERGY AWARE ROUTING PROTOCOL FOR MOBILE AD HOC NETWORKS

Quantifying Responsiveness of TCP Aggregates by Using Direct Sequence Spread Spectrum CDMA and Its Application in Congestion Control

Routing in Degree-constrained FSO Mesh Networks

Synchronous Distributed Wireless Network Emulator for High-Speed Mobility: Implementation and Evaluation

Transcription:

Interclass Collson Protecton for IEEE 82.e Wreless LANs Woon Sun Cho, Chae Y. Lee Dstrbuted Coordnaton Functon (DCF) n IEEE 82. and Enhanced Dstrbuted Channel Access (EDCA) n IEEE 82.e are contenton-based access mechansm n wreless LAN. Both DCF and EDCA reduce collsons based on nter-frame space (IFS) and backoff mechansms. However, collsons are unavodable even wth the two mechansms. Especally, n the EDCA, collsons can be classfed nto nterclass and ntra-class collson. To elmnate nterclass collson n wreless LAN, we propose an nterclass collson protecton (ICP) scheme by employng collson protecton perod (CPP) after backoff. Dfferent number of backoff tmeslots s nserted to each class. Hgher class statons are allowed to transmt before lower class statons end backoff perod. Analyss s performed wth one dmensonal dscrete-tme markov chan for the EDCA and proposed ICP-based EDCA. Collson probablty and throughput of each channel access s examned. Throughput ncrease s more than doubled wth reduced collson probablty when the system s saturated. Keywords: wreless LAN, EDCA, nterclass collson, orthogonal backoff sgnal, markov chan I. Introducton Wreless local area network (WLAN) s rapdly growng due to technologcal development n supportng moblty. The last few years have seen growth n the nstallaton of access ponts (APs) based on IEEE 82. WLAN [] to support data communcatons. Its advantages are low cost and smple deployment. Wth the ncreasng expectaton of voce over Internet Protocol (VoIP), WLAN looks forward to provdng real-tme servce n addton to data servce. However, IEEE 82. WLAN s below the level of qualty of servce (QoS) requrements to provde real-tme servces [3]. To support QoS requrements n medum access control (MAC) level, the standardzaton commttee has proposed IEEE 82.e [2]. The IEEE 82.e MAC protocol employs a hybrd coordnaton functon (HCF) that ncludes enhanced dstrbuted channel access (EDCA). Although the EDCA provdes dfferent QoS for each class by dfferentatng the nterframe space and contenton wndow sze, system performance cannot provde pertnent QoS for voce servce. Ths s because the number of collsons ncreases wth the number of contendng statons. That s, collsons are stll an mpedment because EDCA s a contenton-based mechansm

analogous to DCF. The DCF and EDCA mplement a bnary exponental backoff by ncreasng the contenton wndow sze exponentally for each transmsson falure n order to reduce collsons. However, n certan stuatons, ths exponental backoff causes unnecessary dle duraton such that the channel s utlzed neffcently. Due to the aforementoned facts, t s well known that the throughput performance gets severely compromsed when the number of contendng statons ncreases [4, 5, 6]. The performance degradaton due to collsons becomes more severe as the frame sze ncreases snce the bandwdth waste by collsons becomes relatvely large. Moreover, retransmssons due to collsons waste communcaton energy, thus reducng the lfetme of batterypowered wreless devces. In ths paper, we propose a novel contenton-based MAC model by enhancng the 82.e EDCA, followed by analyss. Our motvaton s that nterclass collson can be elmnated by suppressng the transmsson of lower class statons. A collson occurs when two or more statons transmt smultaneously. When a collson occurs among statons n the same class, t s hard to select a staton to transmt because all statons have equvalent transmsson prorty. On the other hand, n the case of nterclass collson, class prorty can play a role of dfferent transmsson opportunty wth each staton. That s, f a staton n the hghest class prorty has transmsson opportunty, nterclass collson does not occur. The proposed contenton-based MAC s nterclass collson protecton (ICP) model. In the ICP, each staton recognzes the class of other statons whch transmt smultaneously by use of control sgnal and only the hghest class staton s allowed to transmt. The remander of ths paper s organzed as follows. Secton II ntroduces IEEE 82.e EDCA and a sgnal-based collson avodance scheme. In Secton III, we present the basc dea and operaton of ICP. Throughput analyss s presented n Secton IV wth numercal results n Secton V. Fnally, Secton VI concludes the paper. II. Related Works. IEEE 82.e EDCA EDCA s desgned to enhance the DCF mechansm by supportng servce dfferentaton among categores and dstrbutng these categores n the channel access. EDCA conssts of four access categores (ACs). Each AC has dfferent prorty wth dfferent ntal contenton wndow sze (CW,mn ), maxmum contenton wndow sze (CW,max ) and arbtraton nterframe space (AIFS ). Dfferent parameter settngs play a role of gvng hgher AC more statstcal opportunty to access the channel. That s, f a class sets smaller parameters than other classes, statons of the class has a better chance to access the wreless medum. The basc operaton of EDCA s smlar to IEEE 82. DCF. If a staton of AC has a frame to transmt, t checks the 2

medum status for dleness. If the medum s sensed to be dle, the staton mmedately proceeds wth ts transmsson after watng AIFS. If the medum s sensed to be busy, the staton defers ts access untl the medum s determned to be dle for AIFS nterval, and then t starts a backoff procedure. A backoff procedure starts by settng ts own backoff tmer by unformly choosng a random value from the range [, CW,j -], where CW,j s the current contenton wndow sze, and determned by AC and backoff stage j. It s an nteger value wthn the range of CW,mn and CW,max. Backoff stage s defned as the number of retransmssons n a staton. The backoff counter s decreased by a slot tme as long as the channel s sensed dle, whle t s frozen when the channel s sensed busy. The backoff counter count-down s resumed after the channel s sensed to be dle for a AIFS nterval. When the backoff counter reaches zero, the staton starts ts data frame transmsson. If the source successfully receves an acknowledgment (ACK) frame after a short nter-frame space (SIFS) dle perod, the transmsson s assumed to be successful. After a successful transmsson, the source resets ts contenton wndow to the mnmum value CW,mn, and performs another backoff process rrespectve of whether t has another frame to transmt or not. It prevents a staton from performng consecutve mmedate access. On the other hand, f a frame transmsson fals, the CW,j s ncreased by CW,j+ wth the maxmum value CW,max. The staton attempts to transmt the frame agan by recalculatng a backoff counter value from contenton wndow ncrease, CW,j+. After the number of falures reaches a retry lmt, the staton drops the frame. The CW,j for prorty class n backoff stage j s determned as follows [5]. j σcw,, for j =,, L, m, f R> m m CW j, = σ CW, = CW,max, for j = m, m +, L,R, f R> m j σcw,, for j =,, L,R, f R m () where σ s the contenton wndow (CW) ncreasng factor of class (for example, the ncreasng factor of DCF s 2), and CW,mn and CW,max are mnmum and maxmum CW sze of class. R s the retry lmt of class m = log ( CW / CW ). Backoff and σ,max,mn perod s randomly chosen n the range of [, CW,j -]. The AIFS s appled to acheve dfferentaton of each class. The AIFS for a gven AC s determned by the followng equaton. AIFS = SIFS + AIFSN δ (2) where AIFSN s AIFS Number of th AC (AC ) determned by the AC and physcal settng of IEEE 82.e standardzaton [2]. δ s the duraton of a tme slot. The AC wth the smallest AIFS has the hghest prorty. 2. Sgnal Based Collson Avodance A collson occurs when two or more statons transmt data smultaneously on the same channel. IEEE 82. standardzaton [] proposed DCF based on carrer sensng 3

multple access wth collson avodance (CSMA/CA) mechansm. The CSMA/CA utlzes the random backoff pror to each frame transmsson attempt. Whle the random backoff can reduce the collson probablty, t cannot completely elmnate the collsons snce two or more statons can fnsh ther backoff procedures smultaneously. Thus the collson s unavodable n a dstrbuted system, even f the hdden termnal problem can be solved wth the Request-To-Send (RTS) and Clear- To-Send (CTS) messages. The DCF mplements a bnary exponental backoff by ncreasng the contenton wndow sze exponentally for each transmsson falure n order to reduce consecutve collsons. However, the performance degradaton due to the collsons becomes more severe as the number of contendng staton ncreases. Many lteratures have studed the method of reducng collson n WLAN [6, 7, 8, 9]. Especally, collson avodance by use of a dummy sgnal acheves practcal results. Blackburst (BB) contenton scheme [7] and busy tone multple access (BTMA) [9] s representatve of a sgnal based collson avodance scheme. Ⅲ. Interclass Collson Protecton. Basc dea of ICP In the enhanced MAC, any collson among ACs n a staton can be handled by a vrtual collson handler. However, the vrtual collson handler cannot elmnate the nterclass collson among statons. The purpose of ICP s the elmnaton of nterclass collson wth an approprate collson protecton perod after the backoff. Fgure (a) llustrates the case of nterclass collson n EDCA. Under EDCA, nterclass collson occurs when the backoff perod of two or more statons belongng to dfferent AC ends smultaneously. Ths nterclass collson can be avoded by usng a control scheme that grants a preferental transmsson rght to the hghest prorty staton among statons of collson as n Fgure (b). The control scheme guarantees to grant greater channel access prorty to the hgher class staton by usng collson protecton perod (CPP). CPP s an extra perod whch s nserted at the end of a backoff perod of each staton. By usng the CPP, only the hghest class staton transmts when backoff perods of dfferent class statons fnsh smultaneously. That s, the hghest class staton transmts a frame and others wat untl the channel s dle. CPP follows a sgnal based collson avodance scheme. CPP uses a specal sgnal to protect nterclass collson, named orthogonal backoff sgnal (OB sgnal). OB sgnal s defned as a busy sgnal that has no effect on any other sgnal to guarantee orthogonalty. It s a busy tone to nform all other statons that the channel s busy as n [9]. For example, let staton A transmt OB sgnal when staton B transmts a data sgnal to staton C and all statons are located n nterference range. In a casual stuaton, a collson occurs because staton C receves sgnal from both statons A and B 4

(a) Interclass collson n EDCA (b) ICP-based EDCA Fgure EDCA vs. ICP-based EDCA Fgure 2 Basc operaton of ICP n four-class model smultaneously. transmt orthogonal backoff sgnal. However, by the defnton of OB sgnal, the Orthogonal tmeslot performs two roles. sgnal of staton A does not affect the sgnal Frst, the hghest AC staton transmts ts data of staton B, so staton C s able to receve the when nterclass collson has occurred. It s sgnal of staton B. deeply related to the defnton of OB sgnal. CPP conssts of two elements; backoff Whle lower class statons transmt OB sgnal, tmeslot and orthogonal tmeslot as n Fgure the hghest class staton transmts data sgnal 2. Backoff tmeslot proceeds as a tmeslot of wth no nterference. Second, when the backoff perod. In other words, each backoff backoff perod of lower AC staton fnshes tmeslot checks whether the channel s busy earler than hgher ACs, OB sgnal from the or not. If the channel s dle at the last lower class staton occupes the channel and backoff tmeslot, a staton transmts at the the hgher class statons backoff counter next tmeslot. Otherwse, a staton freezes ts freezes untl the channel s dle agan. backoff perod untl the channel becomes dle. Fgure 2 llustrates the ICP of each class. Orthogonal tmeslot s the tmeslot to CPP s nserted after the backoff perod. To 5

prevent nterclass collson, each AC has dfferent tme duraton of CPP. Let the total number of AC be N and AC, =,,N has hgher prorty than AC +. CPP of the hghest AC (AC ) does not have an addtonal tmeslot. The hghest class staton transmts data wthout addtonal delay. CPP of AC conssts of one orthogonal tmeslot and backoff tmeslots. Dfferent number of CPP backoff tmeslots allows hgher class statons to transmt before lower class statons end backoff perod. 2. ICP procedure Fgure 3 llustrates the flow dagram of ICP. ICP procedure s smlar to IEEE 82.e EDCA, but ICP adopts the concept of vrtual collson. Vrtual collson occurs when the backoff counters of nterclass statons fnsh smultaneously. Vrtual collson follows the concept of nterclass collson n EDCA. In ICP, when vrtual collson occurs, the hghest class staton transmts, and lower class statons wat untl the hghest staton fnshes transmsson and restarts backoff procedure wthout ncreasng the CW sze. In ths case, a new backoff counter must be provded to lower class statons because the backoff counter of lower class statons has ended. The new backoff counter s calculated on the same backoff stage. It gves lower prorty class statons more transmsson opportunty than n EDCA. In EDCA, when collson occurs, the contenton wndow sze doubles n the backoff procedure by ncreasng backoff stage. The bgger contenton wndow sze causes a larger backoff counter and the staton whch gets the larger backoff counter needs more watng tme to transmt. Therefore, ICP effcently reduces watng tme by not ncreasng the backoff stage when vrtual collson occurs. Fgure 4 shows how ICP avods nterclass collson. In the fgure, staton s AC, staton 2 s AC 2, staton 3 s AC 3, and staton 4 s AC 4. Fgure 4 (a) shows the case of the hghest class staton (staton ) transmttng when the backoff perod of every staton fnshes smultaneously. After the backoff perod, staton transmts the data sgnal and others transmt orthogonal backoff sgnal. The recevng staton or AP receves the data sgnal of staton wth no nterference by the 6

Frame Transmsson Calculate Backoff Frame preparaton Lsten to the channel Wat for IFS Free? Free? Wat for IFS Decrease Backoff counter Lsten to the channel End of Backoff? Hghest Class? Wat for Orthogonal Tmeslot Free? Increase Backoff stage Decrease Backoff tmeslot End of CPP? Collson? Maxmum retransmsson? Intate Transmsson Ack Receved? Transmsson Complete Transmsson falure Fgure 3 Flow dagram of ICP defnton of orthogonal backoff sgnal. After the orthogonal tmeslot, lower class statons sense the busy channel and recognze the occurrence of vrtual collson and freeze backoff counter untl the channel s dle. Fgure 4 (b) shows the data transmsson by staton 3 when the vrtual collson s caused by statons 3 and 4. For staton 3 to transmt data sgnal, hgher class statons should have remanng backoff tmeslot when 7

backoff perod of staton 3 s fnshed. In the fgure, backoff perod of statons and 2 are not fnshed when staton 3 fnshes ts backoff perod. After the last backoff tmeslot, staton 3 transmts an orthogonal backoff sgnal and the remanng backoff tmeslot of hgher class statons are frozen untl channel s dle. In the fgure, staton 4 has a lower class prorty than staton 3, so the number of CPP backoff tmeslot s more than that of staton 3. By the dfferent number of backoff tmeslots, staton 3 transmts before staton 4 fnshes CPP. Therefore, staton 3 transmts and other statons freeze ther backoff perod untl the medum s dle. Ⅳ. Analytcal Model and Throughput Analyss The analytcal model of IEEE 82. DCF proposed by Banch [4] and Banch and Tnnrello [] becomes a motvaton for numerous analyses of 82. DCF and IEEE 82.e EDCA. The man framework of the analytcal model s two dmensonal Markov Chan (MC) organzed by backoff counter (a) Transmsson by Class staton (b) Transmsson by Class 3 staton Fgure 4 Examples of ICP 8

and backoff stage. Ths two dmensonal MC s smplfed to one dmensonal MC [] by separatng the backoff stage and backoff counter.. IEEE 82.e EDCA model To analyze the performance of IEEE82.e EDCA wth one dmensonal MC, we assume that statons transmt n deal condton wth no errors n the channel and no hdden statons. Also, we assume that the channel s saturated and collson probablty of a transmtted packet s constant and ndependent of the retransmssons. Let τ, that s P(TX ), be the probablty that a staton of AC, s transmttng a frame n a tmeslot. Also, let P(s =j) be the probablty that a staton of AC s n backoff stage j. Then, we can obtan Equaton (3) by the defnton of condtonal probablty. P( s = j, TX ) = P( TX ) P( s = j TX ) = P( s = j) P( TX s = j) (, L, N), j (, L, R ) (3) By Equaton (3), P(s =j) can be represented as follows. P( s = j TX) Ps ( = j) = PTX ( ), P ( TX s = j ) (, L, N), j (, L, R ) (4) Because a staton of AC transmts a data frame n stage j, j, L, R ) wth R j= ( Ps ( = j) =, (, L, N), we can obtan transmsson probablty τ as follows. τ = PTX ( ) =, R P( s = j TX) P( TX s = j) j= (, L, N ) (5) In Equaton (5), τ s computed wth the sum of P(s =j TX ) and P(TX s =j). P(s =j TX ) represents the probablty that a staton of AC beng transmttng s found n stage j. Ths probablty s the steady state probablty of a dscrete-tme Markov chan s (k), descrbng the backoff stage durng the staton s transmsson nstant k, whose non-null one-step transton probablty s gven as n Equaton (6). P( s( k + ) = j s( k) = j ) = p P( s( k + ) = s( k) = j) = p Ps ( ( k+ ) = s( k) = R) = j=, L, R j=, L, R j= R (6) where the condtonal collson probablty p s the collson probablty when the packet s transmtted on the channel. From (6) we can get P(s =j TX ). ( p ) j = jtx = R p + Ps ( ) (, L, N), j (, L, R ) (7) P(TX s =j) n Equaton (5) s the probablty that a staton of AC transmts a frame n backoff stage j. From [], gven backoff stage j, P(TX s =j) s obtaned by dvdng the average number of slots spent by the staton durng the nterval of ts backoff counter whch s called a cycle. PTX ( s = j) =, (, L, N), j (, L, R ) + α + Eb [ ] j (8) E[b j ] s the average value of the backoff counter extracted by a staton of AC enterng stage j. α s the dfference between mnmum AIFS (AIFS ) and AIFS. It s used to compensate the dfference of AIFS of each p 9

class. Namely, f all statons have the same length of IFS, whch s same as the sze of AIFS n EDCA, we can thnk of the staton of AC as havng addtonal tmeslots, the sze of α, durng one cycle. By dong ths, the IFS mechansm of DCF enhances AIFS mechansm of EDCA wthout loss. Substtutng Equaton (7) and (8) nto (5), t can be wrtten as. τ = PTX ( ) = =, [ ] R R Ps ( = j TX) p j + α + peb R + j j= PTX ( s = j) p j= (, L, N), j (, L, R ) (9) When a staton of AC transmts, a collson occurs f one or more statons transmt nto the medum. It s the same as the defnton of p, and t s represented as follows. n = h h nh p ( τ ) ( τ ) (, L, N) () where n denotes the number of statons of class. only appled to lower class statons. The hghest class staton transmts frames n vrtual collson, but lower class statons precede new backoff procedure n the same backoff stage. In ICP-based EDCA model, P(s =j TX ) s gven as n Equaton () wth transton dagram n Fgure 5. P( s ( k + ) = j s ( k) = j ) = p P( s ( k + ) = s ( k) = j) = p P( s ( k + ) = s ( k) = R ) = P( s ( k + ) = j s ( k) = j) = q j =,..., R j =,..., R j = R j =,..., R () where q s the vrtual collson probablty. From () we can get P(s =j TX ). p p q q Ps ( = jtx ) =, R + p q (, L, N), j (, L, R ) (2) P(TX s =j) s the same as the EDCA model n Equaton (8). From Equaton (5), (8), and (2) τ s obtaned as follows. j 2. ICP-based EDCA Model Fgure 5 Transton dagram of P(s =j TX ) of ICP An analytcal model of ICP s an enhanced verson of the EDCA model. The major dfference between ICP and EDCA s vrtual collson. As explaned n Secton Ⅲ. 2, the vrtual collson occurs when backoff perod of nterclass statons fnshes smultaneously. Actually, vrtual collson probablty s the probablty τ = PTX ( ) = =, R Ps ( = j TX) p R j ( ) j PTX s j = q p α R + p j= q = + + E[ bj] q (, L, N), j (, L, R ) (3) In ICP, all actual collsons are ntraclass collsons, snce all nterclass collsons are elmnated by usng CPP. Namely, the actual collson occurs n AC when the backoff perod of two or more statons of the same AC end smultaneously but no hgher class staton ends at the same tme. Therefore, the collson probablty s calculated as, n ( τ ) nh p = ( ) ( τh), (, L, N) (4) h

Also, the vrtual collson probablty s computed as n ( ) nh q =, q = ( τ) ( ( τh) ), (, L, N) (5) Equaton (3), (4), and (5) form a nonlnear system wth the same number of varables and equatons. Ths system can be solved by utlzng a numercal method whch has a unque soluton n the range of τ, p, q [,] for (, L, N). 3. Throughput analyss Let the probablty p S, and p C, be successful transmsson and collson probablty of AC respectvely. Also, let p b be the busy channel probablty. Then, we have n nh S, = ( ) ( h) h= p nτ τ τ, (, L, N) (6) nn ( ) p τ τ τ 2 n 2 nh C, = ( ) ( h) 2 h= N b = S, + C, = h, (, L, N) (7) p ( p p ), (, L, N) (8) Let S denote the normalzed throughput of AC. Then the throughput S s gven as follows. E(payload transmsson tme n a slot tme for the class ) S = E(length of a slot tme) p T =, (, L, N) ( p ) + p T + p T S, E( L) N b δ S S C C = (,,,, ) (9) δ, T E(L), T S,, and T C, n the above equaton denote the duraton of empty tmeslot, the tme to transmt an average payload, the average transmsson tme of AC, and the average collson tme of AC, respectvely. T S, and T C, are calculated as TS, = H + TE[ L] + SIFS + ACK + AIFS TS, = H + TEL [ ] + SIFS + ACK + AIFS + ( + ) δ, (2, L, N ) (2) TC, = H + TE[ L*] + E( IFS) TC, = H + TE[ L*] + E( IFS) + E( N, retry) δ, (2, L, N ) (2) If a transmsson successfully ends (or colldes), T S, (or T C, ) requres extra tme to process CPP compared to EDCA model. To get the average collson tme, we need to know the average number of retres, E(N,retry ), whch s gven as j( p + q ) ( p q ), R j retry, = R + j= ( p + q) EN [ ] (, L, N) (22) Ⅴ. Numercal Results We examne system performance through two scenaros. Scenaro llustrates throughput varaton by ncreasng the total number of statons from one to 4. Each staton generates traffcs n four dfferent classes. Scenaro 2 llustrates throughput varaton by ncreasng the number of statons of one class. We fx the total number of statons n the system to 6 whch s consdered to be suffcently saturated. The computatonal result s based on IEEE 82.b [4] and IEEE 82.e

Table Smulaton parameters for IEEE 82.b Parameter Value SIFS μsec DIFS 5 μsec Slot tme 2 μsec apreamblelength 44 μsec aplcpheaderlength 48 μsec SIFS μsec Data transmsson Mbps ACK transmsson 2 Mbps acwmn 64 acwmax 256 Retry lmts {4, 7,, 4} Table 2 EDCA Parameter settngs Access CW mn CW max AIFSN AC_BK (AC 4 ) acwmn acwmax 7 AC_BE (AC 3 ) acwmn acwmax 3 AC_VI (AC 2 ) acwmn/2 acwmn 2 AC_VO (AC ) acwmn/4 acwmn/2 2 standard [2]. The parameters are shown n Table and 2. To compare the performance, IEEE 82.e EDCA and ICP are solved wth MATLAB programmed numercal method, whch s executed at 3. GHz CPU. Fgure 6 shows the result of Scenaro. The throughput gven by IEEE 82.e EDCA s also compared wth ICP. The fgure shows the mprovement of total throughput by the ICP model through ntraclass collson elmnaton. The ncrease of total throughput s more than doubled when the number of statons n each class exceeds 2. Fgure 7 shows the dfference of collson probablty between EDCA and ICP. The collson probablty by ICP s reduced by more than 2% as the channel congeston ncreases. Fgure 6 and 7 clearly show that total throughput and collson probablty are closely related and they are nversely proportonal. In ICP, the most porton of throughput mprovement s contrbuted by AC. It s because ICP provdes hgher prorty to the hghest class when vrtual collson occurs. Another mportant pont proven by the fgures s that the throughput 2

mprovement of AC does not degrade that of the lower class. Fgure 8 and 9 llustrate Scenaro 2. All fgures show total throughput mprovement of ICP compared to EDCA. A large porton of total throughput mprovement occurs n AC whle the throughput of other classes mantans the level of EDCA. It s because ICP enhances system performance by elmnatng nterclass collson and provdes more prorty to AC than other classes. Fgure 8 shows an ncrease n the number of statons of AC whle the total number of statons s fxed at 6. The number of statons of AC ranges [3, 57] as multples of three. In each case the remanng statons are equally dstrbuted nto three other classes. In the Fgure, when the number of statons of AC s smaller than, the throughput of AC ncreases wth the number of AC statons, but the total throughput s decreased by the loss of lower class throughputs. When the number of AC statons s greater than, total throughput s decreased by excessve congeston. Moreover, throughput n Fgure 8 s not better than that n Fgure 6, even f the total number of statons n Fgure 8 s smaller than Fgure 6. The reason s that the ncrease of collson and vrtual collson s serously affected by the number of statons of AC rather than the total number of statons. It s also related to the contenton wndow sze n Table 2, the contenton wndow sze of AC s the smallest. A small contenton wndow sze guarantees large transmsson opportunty, but causes extra collsons when the number of AC statons ncreases. Fgure 9 shows an ncrease n the number of statons of AC 2 whle the total number of statons s fxed at 6. Dfferently from Fgure 8 total throughput n Fgure 9 mantans almost same level. In the fgure, throughput of AC by ICP and EDCA shows dfferent movements. In ICP, when AC 2 s smaller than 45, the throughput mantans about.35 regardless of the change n the number of statons, whereas, throughput of EDCA s decreased accordng to the rmalzed throughput.9.8.7.6.5.4.3 AC e AC2 e AC3 e AC4 e total e AC ICP AC2 ICP AC3 ICP AC4 ICP total ICP.2. 5 5 2 25 3 35 4 Number of statons Fgure 6 rmalzed throughput 3

.9 Collson probablty 82.e Collson probablty ICP.8.7 probablty.6.5.4.3.2. 5 5 2 25 3 35 4 Number of statons Fgure 7 Collson probablty rmalzed throughput.9.8.7.6.5.4.3 AC e AC2 e AC3 e AC4 e total e AC ICP AC2 ICP AC3 ICP AC4 ICP total ICP.2. 2 3 4 5 6 Number of statons of AC Fgure 8 Throughput varaton by change of AC rmalzed throughput.9.8.7.6.5.4.3 AC e AC2 e AC3 e AC4 e total e AC ICP AC2 ICP AC3 ICP AC4 ICP total ICP.2. 2 3 4 5 6 Number of statons of AC2 Fgure 9 Throughput varaton by change of AC 2 4

ncreasng number of statons of AC 2. It s because ICP provdes the hghest class wth more prorty to transmt than EDCA. Ⅵ. Concluson In ths paper, we have proposed a novel dstrbuted contenton-based MAC based on IEEE 82.e and analyzed the model through analytcal model. To elmnate nterclass collson, we have proposed the collson protecton perod. When backoff perod of two or more nterclass statons fnsh smultaneously, the hghest staton transmts and the others wat untl the next dle perod. We have made an analytcal model of ICP and compared t wth IEEE 82.e EDCA. The numercal results show that the ICP model can provde better performance compared to the exstng EDCA model. The proposed collson protecton scheme based on the orthogonal backoff sgnals ncreases overall throughput wthout degradng the performance of lower classes. The ncrease of throughput by the proposed ICP model s more than doubled compared to the EDCA model, when the number of statons n each class exceeds 2. Also, the collson probablty by ICP s reduced by more than 2% as the channel congeston ncreases. VII. References [] IEEE 82., Part : Wreless LAN Medum Access Control (MAC) and Physcal Layer (PHY) Specfcatons, Aug. 999. [2] IEEE Std 82.e/D3., Part : Wreless LAN Medum Access Control (MAC) and Physcal Layer (PHY) Specfcatons: Amendment: Medum Access Control (MAC) Qualty of Servce (QoS) Enhancements, Jan. 25. [3] S. Kumar, V. S. Raghavan, and J. Deng, Medum Access Control protocols for ad hoc wreless networks: A survey, Ad hoc Networks 4, 26, pp. 326-358. [4] G. Banch, Performance Analyss of the IEEE 82. dstrbuted coordnaton functon, IEEE J. Select. Areas. Commun., vol.8, no. 3, Mar. 2, pp. 535-547. [5] Y. Xao, Performance Analyss of Prorty Schemes for IEEE 82. and IEEE 82.e Wreless LANs, IEEE Trans. Wreless Commun., vol. 4, no. 4, Jul. 25, pp. 56-55. [6] J. Cho, J. Yoo, S. Cho, and C. Km, EBA: An Enhancement of the IEEE 82. DCF va Dstrbuted Reservaton, IEEE Trans. On Moble Computng, vol. 4, no. 4, Jul./Aug. 25, pp. 378-39. [7] J. L. Sobrnho and A. S. Krshnakumar, Qualty-of-Servce n Ad Hoc Carrer Sense Multple Access Wreless Networks, IEEE J. Select. Areas. Commun., Vol. 7,. 8, Aug. 999. [8] W. T. Chen, B. B. Jan, and S. C. Lo, An adaptve retransmsson scheme 5

wth QoS support for the IEEE 82. MAC enhancement, VTC Sprng 22. IEEE 55th, Vol., 22, pp. 7-74. [9] F. A. Tobag and L. Klenrock, Packet Swtchng n Rado Channels: Part II- The hdden termnal problem n carrer sense multple-access and the busytone soluton, IEEE Trans. On Commun., vol. com-23,. 2, December 975. [] E. Zouva and T. Antonakopoulos, CSMA/CA performance under hgh traffc condtons: Throughput and delay analyss, Comp. Commun., vol. 25, no. 3, Feb. 22, pp. 33-32. [] G. Banch and I. Tnnrello, Remarks on IEEE 82. DCF Performance Analyss, IEEE Commun. Lett., vol. 9, no. 8, Aug. 25, pp. 765-767. [2] Y. Kwon, Y. Fang, and H. Latchman, A vel MAC Protocol wth Fast Collson Resoluton for Wreless LANs, n Proc. IEEE INFOCOM 3, Apr. 23. [3] H. Wu, Y. Peng, K. Long, S. Cheng, and J. Ma, Performance of Relable Transport Protocol over IEEE 82. Wreless LAN: Analyss and Enhancement, n Proc. IEEE INFOCOM 2, vol. 2, Jun. 22. [4] S. Cho, PCF vs. DCF Lmtatons and Trends, IEEE 82.-/54, Jan. 2. [5] Z. Kong, D. H. K. Tsang, B. Bensaoum and D. Gao, Performance Analyss of IEEE 82.e Contenton-Based Channel Access, IEEE J. Select. Areas. Commun., vol. 22, no., Dec. 24. 6