On the performance evaluation of CSMA/E2CA protocol with ... - UPF

On the performance evaluation of CSMA/E2CA protocol with open loop ARF-based adaptive modulation and coding Gabriel Martorell*, Felip Riera-Palou*, Guillem Femenias*, Jaume Barcelo† and Boris Bellalta† * Mobile Communications Group – University of the Balearic Islands (Spain) Dept. of Information and Communication Technologies – Universitat Pompeu Fabra (Spain) Email: {gabriel.martorell, felip.riera, guillem.femenias}@uib.es, {jaume.barcelo, boris.bellalta}@upf.edu †

Abstract—Carrier sense multiple access with collision avoidance (CSMA/CA) has been the access protocol of choice for IEEE 802.11-based WLANs. In addition to channel sensing before transmission, the probability of a collision in these WLANs is typically reduced by the application of a binary exponential backoff (BEB) algorithm that randomizes the selection of the time slot in which a given station transmits. In a system without adaptive modulation and coding (AMC), the reduction in spectral efficiency caused by BEB is outweighed by the reduction in the number of collisions. In contrast, in systems using the ubiquitous auto rate fallback (ARF) AMC algorithm, which is unable to distinguish a collision from an erroneous transmission, the remaining collisions induce a dramatic drop in system performance. This degradation is caused by the utilization of low-rate transmission modes even when the channel conditions would permit the use of much higher-rate modes. In an attempt to further reduce the number of collisions, a variant of CSMA/CA, called enhanced collision avoidance (CSMA/E2CA), has been recently proposed. In this paper, a model approach to the performance evaluation of both BEB-based CSMA/CA and CSMA/E2CA, used in conjunction with ARF, is presented and validated. Results reveal the synergistic properties of the E2CA and ARF combination, as demonstrated by the superior goodput performance when compared against other strategies. Index Terms—CSMA/E2CA, AMC, ARF, CSMA/CA, DCF, 802.11n.

I. I NTRODUCTION Wireless local area network (WLAN) technology has become prevalent to provide the last-hop connection to the Internet. Many consumer devices have adopted the IEEE 802.11 standard or its amendments as WLAN technology. In IEEE 802.11 systems, the medium access is based on a carrier sense multiple access with a collision avoidance (CSMA/CA) mechanism. This mechanism does not ensure a collision-free operation and, in fact, the number of collisions increases with the number of users. Collisions are one of the main problems of the system performance, leading to a reduction of system goodput and greatly impairing the operation of resource allocation schemes such as adaptive modulation and coding (AMC) algorithms. The distributed coordination function (DCF), the most used medium access control (MAC) mechanism in IEEE 802.11 family, defines the binary exponential backoff (BEB) algorithm as the CSMA/CA backoff technique, to determine how many

slots should be waited by each station (STA) before to transmit. The number of slots is randomly selected, thus leading to collisions when two or more stations chose the same slot. In order to reduce the collisions among long data packets, the standard contemplates the possibility of preceding the data transmission by two short signaling packets: request-tosend and clear-to-send (RTS/CTS). However, these packets represent an additional overhead that leads to a significant information throughput (goodput) reduction. In Basic Access the transmitter directly sends the data packet as soon as the backoff expires without requiring the RTS/CTS packets. Recently, a variant of the CSMA/CA protocol has been presented in [1] and further improved in [2], called CSMA with enhanced collision avoidance (CSMA/E2CA). The key idea is the use of a deterministic backoff after successful transmissions. In particular, in CSMA/E2CA, a deterministic backoff is used for at least two consecutive transmissions after each successful transmission. The use of a deterministic backoff after successes substantially reduces the number of collisions and the system tends to converge to a collisionfree operation. In collision-free operation, the behaviour of the system is completely deterministic and the participating stations transmit in a round-robin fashion. Note that the performance of CSMA/E2CA under realistic channel conditions (AMC mechanisms, channel fading) has not been studied yet. Most IEEE 802.11-based systems employ DCF with Basic Access and adopt open-loop AMC policies such as auto rate fallback (ARF) [3], [4] or one of its variants (e.g. CARA [4], SARA [5]). Owing to its simplicity, ARF is by far the most popular algorithm in use. Remarkably, the DCF with Basic Access scheme does not differentiate between collisions and transmission failures caused by poor channel conditions. Consequently, when the system experiences a high collision probability, ARF tends to use the lowest transmission rate even if the channel conditions would allow the use of much higher transmission modes (see for example, [4], [6], [7], [8]). Other adaptive strategies have been proposed to solve this issue, but they may require frame format changes [9], modifications to the medium access technique [4], or the use of channel quality indicators (e.g. signal strength indicator) [7], [9] and, in fact, none of them has achieved widespread use in current WLAN systems [10]. In contrast, ARF performs remarkably well when

combined with the RTS/CTS access technique although at the cost of an spectral efficiency loss due to the introduced overhead. This paper has a double objective. Firstly, it introduces a semi-analytical framework to model the performance of the different access techniques under consideration (Basic Access, RTS/CTS and CSMA/E2CA) in combination with ARF within the context of the last approved amendment of IEEE 802.11, namely, IEEE 802.11n. The proposed model is generic enough so as to incorporate parameters such as the packet length and number of users. Secondly, the presented results demonstrate that the application of ARF in conjunction with CSMA/E2CA results in a synergistic combination clearly outperforming any of the other schemes. The rest of the paper is structured as follows. Section II describes the system model under consideration. In Section III the main characteristics of CSMA/E2CA are described. Section IV briefly reviews ARF as the AMC algorithm covered in this work. In Section V the analytical framework used to analyze the CSMA/E2CA system goodput is presented. In Section VI the simulation tool features are described and the numerical results comparing the performance of ARF using BEB and E2CA algorithms are presented assuming different configurations and access techniques. Finally, Section VII summarizes the main conclusions of this study. II. S YSTEM OVERVIEW A. Physical layer description Our study focuses on the IEEE 802.11n standard [11], whose PHY layer is based on MIMO-OFDM. The MIMO component exploits multiple transmit and receive antennas to increase the system capacity or reliability, depending on the specific technique employed [12], (e.g., space-time block coding (STBC), space division multiplexing (SDM), cyclic delay diversity (CDD) and/or combinations of them). At the transmitter side, information bits are first encoded with a rate R = 21 convolutional encoder with generator polynomials [133, 171]8 and then punctured to one of the possible coding rates Rm ∈ {1/2, 2/3, 3/4, 5/6}. Depending on the selected MIMO configuration, the resulting bits are demultiplexed into Ns spatial streams. For each stream, the coded bits are interleaved and then mapped to symbols from one of the allowed constellations (BPSK, QPSK, 16-QAM or 64-QAM). In accordance with the selected MIMO configuration, the symbols are then either STBC encoded or antenna mapped on the available NT transmit antennas. The resulting symbols are finally supplied to a conventional OFDM modulator consisting of an IFFT and the addition of a guard interval. For simplicity of exhibition, this paper focuses on a 2 × 2 MIMO system (NT = 2 and NR = 2), implying that MCSs with Ns = 1 and Ns = 2 spatial streams employ STBC [13] and SDM [14], respectively. At the receiver side, Alamouti decoding or Minimum Mean Square Error (MMSE) detection is applied depending on whether STBC or SDM has been employed. In either case, the detector extracts soft information in the form of log-likelihood

ratios (LLRs) that, after suitable de-interleaving/de-parsing, can be exploited by a soft Viterbi decoder [15]. B. MAC layer description The IEEE 802.11 MAC specifies three different medium access control (MAC) mechanisms for WLANs, namely the contention-based distributed coordination function (DCF), the point coordination function (PCF) and the hybrid coordination function (HCF). The DCF is the mandatory MAC mechanism for the IEEE 802.11 standard [16]. It is a random access scheme based on the carrier sense multiple access with collision avoidance (CSMA/CA) protocol that incorporates a binary exponential backoff (BEB) algorithm to manage the retransmission of collided packets. DCF defines both the basic and the RTS/CTS access schemes. The Basic Access technique is the most extensively used [4]. In Basic Access, an STA transmits one data packet at its BEB scheduled slot and waits for its packetacknowledgement (ACK-control frame) from the receiver. If no reply arrives during a predefined time interval, the STA interprets the transmission as erroneous and the packet is either retransmitted in the next backoff scheduled slot or discarded if the number of packet retransmissions exceeds the maximum number of allowed retransmissions, which will be denoted by Rlimit . In contrast to basic access, the implementation of RTS/CTS is optional, yet it is an advisable feature whenever the system operates with packets whose length exceeds a predefined threshold (Sthres ). Unlike Basic Access, prior to the data transmission, RTS/CTS exchanges two control frames, request to send (RTS) and clear to send (CTS), between source and destination. This frame exchange allows the reservation of the channel for the current data transmission and drastically reduces the eventual collision duration to that of two collided RTS frames. However, RTS/CTS induces a considerable increase of the system overhead, lowering in this way the system performance when short packet lengths or high transmission rates are used. Nevertheless, thanks to its reduced collision duration, it outperforms Basic Access in dense user scenarios where the collision probability is high. In this paper, two backoff algorithms are studied: CSMA/CA, as the IEEE 802.11 standard backoff algorithm, and the CSMA/E2CA as a new backoff algorithm that notably improves the CSMA/CA performance. 1) CSMA/CA: In CSMA/CA, the BEB algorithm determines the backoff time prior to the packet transmission. At each packet transmission, the number of waiting slots is randomly chosen in the range (0, CW − 1). The value CW is called contention window and depends on the number of transmission attempts for the considered packet. At the first packet transmission attempt, CW is set to CWmin and then is doubled after each unsuccessful transmission up to a maximum value of CWmax = 2r CWmin . The CWmin and CWmax parameter values are specified in the standard document [16]. 2) CSMA/E2CA: E2CA is an alternative backoff protocol described in [2]. It uses a deterministic backoff for almost

two consecutive times after each successful transmission, and a random backoff in the range [0, CWmin − 1] otherwise. The value of the deterministic backoff after successes is C = CW2min , and the value for CWmax is set equal to the value of CWmin . The value of CWmin is dynamically adjusted to keep the value of busy slots between 18 and 12 as detailed in [2]. This protocol has the advantage of reducing the number of collisions and finally converging to collision-free operation, as it will be explained in Sec. III.

2) RTS/CTS access technique: One of the consequences of the RTS/CTS handshake procedure is that when a collision takes place, its duration is minimized. This is because collisions can only occur during this exchange, which only involves frames of minimal length. The RTS/CTS time elapsed for a successful transmission of an L-bit MPDU using MCS m is Tsrts (m, L) =TRT S (m) + TCT S (m)+ + TT r (m, L) + TACK (m)+

C. Timing of DCF events.

(8)

+ 3 tSIF S + tDIF S ,

1) Basic Access technique: According to the Basic Access scheme of DCF, the elapsed time for a successful transmission of an L-bit MPDU using MCS m is Tsbas (m, L) =TT r (m, L) + tSIF S + + TACK (m) + tDIF S ,

(1)

where tSIF S (short interframe space) and tDIF S (distributed interframe space) are 802.11n time constants defined in [11]. The time elapsed in the MPDU transmission, TT r (m, L), is defined as TT r (m, L) = tP reamble + NSym (m, L) tSym ,

(2)

with tP reamble representing the PLCP preamble duration, tSym denoting the OFDM symbol period and   L + 22 NSym (m, L) = mST BC (3) mST BC NDBP S (m) being the number of OFDM symbols involved in the transmission of a complete MPDU, where NDBP S (m) is the number of bits forming each OFDM symbol as defined by MCS m, and mST BC = 2 if STBC is used and mST BC = 1 otherwise. Similarly, the time required for the transmission of an ACK frame using PHY mode m is given by TACK (m) = tP reamble + NSym (m, 14 × 8) tSym .

(4)

A collision occurs whenever two or more STAs transmit on the same slot, finishing tEIF S (extended interframe space) after the end of the longest transmission of the collided STAs. That is, its duration depends on the MCS and MPDU length corresponding to the longest transmission, denoted by m∗ and L∗ , respectively. Therefore, the collision duration can be mathematically expressed as Tcbas (m∗ , L∗ ) = TT r (m∗ , L∗ ) + tEIF S ,

(5)

tEIF S = tSIF S + TACK (m=0) + tDIF S .

(6)

where

Finally, the MPDU error transmission duration, defined as Te (m, L), is the time elapsed in a transmission that experiences errors without collisions, and it can be expressed as Tebas (m, L) = TT r (m, L) + tEIF S .

(7)

where TRT S (m) = tP reamble + NSym (m, 14 × 8) tSym ,

(9)

TCT S (m) = tP reamble + NSym (m, 20 × 8) tSym ,

(10)

and

are the durations of RTS and CTS frames, respectively. Similarly, the RTS/CTS elapsed time in a transmission error of an L-bit MPDU using MCS m is Terts (m, L) =TRT S (m) + TCT S (m) + TT r (m, L) + 2 tSIF S + tEIF S .

(11)

Finally, the collision duration can be defined as Tcrts (m0 ) = tRT S (m0 ) + tEIF S ,

(12)

where m0 is the lowest MCS value that causes the longest RTS transmission duration from the collided users. In this model, due to its negligible probability of occurrence, we have not considered the possibility of an error in the ACK, RTS and/or CTS transmissions. The ACK transmission takes place under the same system conditions as the packet being acknowledged, i.e., using the same MCS and suffering similar channel conditions. However, its packet size is considerably smaller than that of the information packets and therefore, its error probability can be safely considered insignificant. Similarly, packet sizes for RTS and CTS transmissions are also small and the use of the most reliable MCS warrants a negligible error probability. III. C OLLISION - FREE

OPERATION

The number of collisions in wireless networks can be notably reduced if a deterministic backoff is used after successful transmissions, as in the CSMA/E2CA protocol introduced in the previous section. The deterministic backoff value after successful transmissions has to be the same for all the STAs and we denote it by C. Two STAs that have successfully transmitted in their last transmission attempt have necessarily transmitted in different slots, otherwise their transmission would have resulted in a collision. Then, if those STAs backoff for exactly C slots before their next transmission attempt, they will also choose a different slot in their next transmission attempt. The fundamental principle is that two stations that have successfully transmitted in their last attempt and use a

deterministic backoff will not collide among them in their next transmission slot. If all the STAs have successfully transmitted in their last transmission attempt, this fundamental principle applies to each pair of stations. As a result, all the STAs will use a deterministic backoff C and they will not collide in their next transmission attempt. At this point, it can be stated that collision-free operation has been reached since all the STAs behave deterministically and successfully transmit in a roundrobin fashion. Collision-free operation substantially improves the overall network performance in terms of delay, jitter, throughput and fairness. Before reaching collision-free operation, the network goes through a transient state in which collisions occur. It is highly desirable to reduce the length of this transient state in which collisions penalize the network performance. In [2], it is shown that setting CWmax = CWmin reduces the length of the transient state and thus increases the network performance. If a value CWmax = CWmin is chosen, it is necessary to adjust the value of CWmin to the number of contenders. In order to visualize the execution of the protocol, it is useful to imagine that the stations distributively choose a slot from a set of C slots. If two stations choose the same slot, they collide and they will choose a different slot in their next transmission attempt. When a slot is chosen by a single station, this station succeeds and will use the same slot in their next transmission attempt. In a sense, the station sticks to that slot after a successful transmission. This principle can be easily generalized to consider higher degrees of stickiness. If a stickiness degree of two is used, a STA will use a deterministic backoff for two consecutive transmissions after each successful transmission. When a higher degree of stickiness is used, the STAs have a stronger tendency to stick to a slot in which they have successfully transmitted. Stickiness shortens the duration of the transient state and has the additional benefit of providing extra robustness against channel errors. A STA with a stickiness degree of two will stick to its slot even in the occurrence of a channel error. This STA will only move back to the random behaviour if it suffers two consecutive channel errors in two consecutive transmission attempts, which is highly unlikely if the right MCS is used. The CSMA/E2CA protocol considered in this paper uses a degree of stickiness equal to two. This protocol represents only a subtle modification to the original CSMA/CA and therefore both the new protocol and the legacy one can coexist in the same network. For the new STAs to be fair to the legacy ones, the deterministic backoff should be chosen as C = CW2min , which is the expected backoff after successful transmissions of the legacy STAs. The operation of CSMA/E2CA is illustrated in Fig. 1. The figure shows a timeline which is divided in numbered slots. The transmissions are represented as balls in the slots, and the filling patterns of the balls is used differentiate transmissions by different STAs. Note that in the figure all the slots are depicted as being equal for representation convenience, despite

000 111 00 11 000 111 111 000 0000 1111 00 11 00 11 000 00111 11 000 0000 00 000 111 0 1 2 3 4 5 61111 7 8 11 9 10 11 12 13 14 15 Slot No. 111 000 111 00 11 000 111 00 11 000 111 00 000 111 000 111 00 11 11 00 11 000 00111 11 000 111 00 000 111 16 17 18 19 20 21 22 23 24 11 25 26 27 28 29 30 31 111 000 000 00 11 00111 000 111 000 00111 11 000 111 111 000 111 00 11 11 000 00000 111 11 00111 11 000 111 000 111 00 000 000 32 33 34 35 36 37 38 39 40 11 41 42 43 44 45 46 111 47 00111 111 111 000 000 00 11 11 000 111 000 00 11 00111 11 000 111 000 111 00 000 000 48 49 50 51 52 53 54 55 56 11 57 58 59 60 61 62 111 63 00 11 111 000 000 111 00 11 00 11 000 000 111 00111 11 000 111 000 111 00 11 000 111 000 64 65 66 67 68 69 70 71 72 73 74 75 76 77 79 111 79 Fig. 1.

Example operation of CSMA/E2CA

the fact that in reality busy slots are much longer than empty ones. All the STAs transmit for the first time in the first row (slots from 0 to 15). Those stations that successfully transmit (in slots 1, 9, 11 and 13) use a deterministic backoff equal to 16, which in the figure is equivalent to sticking to the same column. Those STAs that suffer a collision (in slot 7) use a random backoff, which in our figure is equivalent to randomly choosing a new column for the next transmission attempt. Moving to the second row of the figure, we can observe collisions in slots 17 and 27. Now, the station that successfully transmitted in slot 1 and then collided in slot 17 will use a deterministic backoff equal to 16, since a deterministic backoff is used for two consecutive times after each successful transmission. Contrastingly, the STA that collided in slot 7 and in slot 17 will choose a random backoff and will switch columns again. The situation is quite similar in slot 27. There is a collision but one of the STAs will stick to the same column while the other one will switch columns. In the third row of the figure all the STAs successfully transmit. At this point, all the STAs will use a deterministic backoff (which is equivalent to sticking to the same column) and will succeed in their next transmission attempt in the fourth row. It should be clear that collision-free operation has been reached, and therefore the behaviour of the system is completely deterministic and the participating STAs simply transmit in a round-robin fashion. IV. AUTO -R ATE FALLBACK This algorithm adapts the transmission rate according to the number of consecutive transmission failures and successes, both reported by the ACK mechanism. The transmission rate is decreased after two consecutive transmission failures

and increased after either ten consecutive successful packet transmissions or a timeout. This timeout counter is reset after a transmission rate change or after a transmission failure, in order to improve the system adaptation during long intervals of inactivity [3]. An acceptable timeout value lies in the range of 50 − 200 ms [17]. Note that following a rate increase, the next data transmission is deemed as a probing transmission for the new mode. If an ACK is not received for this probing packet, the system falls back to the previous data rate. In order to implement ARF in IEEE 802.11n it is necessary to determine the available rates in the MCS set, denoted by M. In contrast to previous IEEE 802.11 standards, in 802.11n different MCSs ∈ M can provide the same transmission rate, but only one of them can be used by the ARF algorithm. For this reason, the MCSs in M are reordered according to their transmission rate [15]. For those rates that can be attained using either SDM or STBC, only the STBC MCS is kept as it can be shown to be more robust against channel variations [13]. V. G OODPUT ANALYSIS The goodput analysis presented in this paper has been performed assuming that all the STAs in the system are under saturation conditions, i.e., having a packet available to be transmitted at any moment. The goodput analysis has been evaluated in terms of average goodput per slot that can be defined as G=

Average inf ormation per slot . Average slot duration

(13)

Considering ideal conditions, E2CA achieves collision free operation and behaves as a round robin scheme where each user transmits once in every C slots [2]. This leads to a n constant probability of busy and idle slots with values Pb = C C−n and Pi = 1 − Pb = C , respectively, where n is the number of users (or contending STAs) and C is the value of the deterministic backoff. Under ideal conditions, the busy slots are uniquely filled by successful transmissions allowing the goodput to be expressed as G=

Pb Lp Pb Ts + (1 − Pb )σ

(14)

where Ts is the average time elapsed in a system successful transmission and σ is the duration of any idle slot. Similarly, with non-ideal channel conditions for n ≤ C2 with CWmin = CWmax , Pb and Pi can be considered nearly constant when the error rate is not very pronounced [2]. The detailed analysis should also account for transmission errors and collisions. However, observing the weak performance degradation suffered in E2CA with non-ideal channel conditions when n ≤ C2 , a tight upper-bound for the system goodput performance is given by, G≤

Pb (1 − Pe − Pc )Lp   , Pb (1 − Pe − Pc )Ts + Pc Tc + Pe Te + (1 − Pb )σ (15)

where Pe and Pc are the error and collision probabilities per transmission, respectively, and, Te and Tc are the average collision and error duration, respectively. The parameters Pe , Pc , Te and Tc are obtained by simulation, thus making our model semi-analytic. VI. N UMERICAL

RESULTS

A. Scenario configuration In order to validate our semi-analytical model and compare the performance of CSMA/E2CA and CSMA/CA under different system configurations, an IEEE 802.11n system-level Matlab simulator has been implemented using the link-level parameters derived in [15]. In this paper, focus is on the performance evaluation of the uplink scenario where, nevertheless, MAC control frame transmissions from access point (AP) to STA are also accounted for. Different scenarios have been generated by uniformly distributing n static users in a circular area of radius Rmax centered on the AP and then determining the individual channel response from each user to the AP. To this end, the MIMO channel generation tool presented in [18], parameterized with each user’s distance to the AP, has been employed. The maximum radius Rmax has been set to 30 m, a value that ensures the avoidance of the hidden terminal problem. The transmit power for all STAs has been set to 20 dBm and receiver noise power has been fixed to −80 dBm. The physical layer uses only the first 16 MCS modes of IEEE 802.11n (MCS0-MCS15), achieving data rates of up to 130 Mbps [11]. It should be stated that all users in the system use DCF with the same backoff algorithm (either BEB or E2CA). Regarding the access scheme setup, CSMA/CA has been configured using the system parameters defined in the standard [11] (e.g., CWmin = 16) whereas CSMA/E2CA has been set up with a degree of stickiness of two, a maximum contention window CWmax = CWmin and a fixed minimum contention window of CWmin = 2dlog2 (4n)e satisfying n ≤ CW4min . The ARF timeout has been set to 60 ms. In order to obtain an accurate estimate of the average system performance, Nsim = 100 realizations of duration tsim = 22 seconds for each number of users (n) have been simulated using Matlab. B. Results In Fig. 2 the probability that a particular STA transmits in a generic slot (τ ) for CSMA/E2CA and CSMA/CA is shown. According to the characteristics of an ideal round robin scheme, the analytic CSMA/E2CA τ is known to have a constant value of τ = C1 . Although the channel introduces transmission errors, analytic τ of E2CA matches the simulation. Note that τ decreases abruptly each time that CWmin is doubled in order to satisfy that the constraint n ≤ C2 . In contrast, CSMA/CA uses a constant CWmin value for any number of users and its τ performance decreases gradually outperforming E2CA for n > 8. Simulation and analytical results of the probability of busy slot (Pb ) using CSMA/E2CA with Basic Access as a function

CW

=16

min

0.9 0.8 0.7 P CSMA/E2CA Basic Access s

−1

10

CW

Probability

Probability of transmission per station (τ)

1 CSMA/E2CA Basic Access sim. CSMA/E2CA sim. CSMA/E2CA analysis CSMA/CA Basic Access sim. CSMA/CA Basic Access analysis CSMA/CA RTS/CTS sim. CSMA/CA RTS/CTS analysis =32

min

P +P CSMA/E2CA Basic Access

0.6

c

e

Ps CSMA/CA Basic Access

0.5

Pe+Pc CSMA/CA Basic Access P CSMA/CA RTS/CTS

0.4

s

P +P CSMA/CA RTS/CTS e

0.3

c

0.2 CWmin=64

0.1

−2

10

2

4

Fig. 2.

6

8 10 n (number of users)

12

14

16

τ for E2CA and BEB using ARF.

0

2

4

Fig. 4.

6

8 10 n (number of users)

12

14

16

Ps and Pe + Pc for E2CA and BEB using ARF.

0

10

Probability of busy slots (Pb)

CWmin=16

−1

10

CW

=64

min

2

4

Fig. 3.

6

CWmin=32

CSMA/E2CA Basic Access sim. CSMA/E2CA sim. CSMA/E2CA analysis CSMA/CA Basic Access sim. CSMA/CA Basic Access analysis CSMA/CA RTS/CTS sim. CSMA/CA RTS/CTS analysis

8 10 n (number of users)

12

14

16

Pb for E2CA and BEB using ARF.

of the number of users (n) are presented in Fig. 3. For completeness, Pb for CSMA/CA with Basic and RTS/CTS access mechanisms are also presented. A very accurate match between analysis and simulation is obtained for CSMA/E2CA, specially for those cases where n ≤ C2 = CW4min , thus validating the model presented in this paper. Note that, having similar values of τ , CSMA/E2CA attains a higher Pb than CSMA/CA, thus leading to a more efficient channel utilization. The explanation is that in CSMA/CA it is possible that two stations transmit in the same slot and these two transmission attempts count as a single busy slot. By contrast, in CSMA/E2CA the collisions are almost avoided and each transmission attempt results in a single busy slot. A closer look on the busy slots is provided in Figs. 4 and 5. Fig. 4 discriminates between successful and unsuccessful slots and Fig. 5 focuses on unsuccessful transmissions. Despite considering non-ideal channel conditions in E2CA, a collisionfree scheme is obtained, achieving reduced Pc values, lower than that obtained with CSMA/CA, regardless the number

of users or access scheme (see Fig. 5). As a result, most of the busy slots are occupied by successful transmissions (see Fig. 4), thus outperforming the CSMA/CA efficiency in terms of slot occupation probability. Note that Pe has the expected ARF error rate, that is, an average error rate of one error every ten transmissions, because of the probing mechanism executed either after each ten consecutive successful transmissions or after the timeout expiration. It is important to stress that, despite ARF introduces transmission errors, CSMA/E2CA reduces Pc to small values, thus confirming the robustness of CSMA/E2CA. In CSMA/CA, Pe and Pc perform differently depending on the access scheme employed. Using Basic Access, Pe is negligible due to the ARF malfunction that selects the lowest MCS independently of the channel conditions, thus obviously affecting its system goodput. Meanwhile, Pc performs as a BEB scheme under error free channel conditions. In RTS/CTS access scheme, however, Pe fulfills the expected ARF error rate and Pc is slightly lower than that obtained in Basic Access. Nonetheless, their sum, known as the probability of unsuccessful transmission, exceeds its counterpart in Basic Access. In Fig. 6, the ARF goodput performance over DCF as a function of the number of STAs applied under the previous backoff algorithms and access schemes is presented. Additionally, the semi-analytical goodput performance of ARF with CSMA/E2CA is also shown in the figure to demostrate the accuracy of the proposed model. Notice that the semianalytical performance results of CSMA/CA are also presented and their expressions are be obtained from [19]. Regarding the performance results, note that ARF achieves the highest goodput when it is combined with CSMA/E2CA, outperforming CSMA/CA irrespective of the access technique. The utilization of CSMA/E2CA with Basic Access drastically reduces the number of collisions (see Fig. 5) and solves the MCS selection problem of ARF experienced in CSMA/CA with Basic Access. Consequently, the most appropriate transmission mode according to the previous transmission results is selected,

0.35

30 Pe CSMA/E2CA Basic Access Pc CSMA/E2CA Basic Access

0.3

25

P CSMA/CA Basic Access e

P CSMA/CA Basic Access c

0.25

P CSMA/CA RTS/CTS e

20 Goodput (Mbps)

Probability

Pc CSMA/CA RTS/CTS 0.2

0.15

15

10 0.1 5

0.05

0

CSMA/E2CA Basic Access n=2 CSMA/E2CA Basic Access n=3 CSMA/CA Basic Access n=2 CSMA/CA Basic Access n=3 CSMA/CA RTS/CTS n=2 CSMA/CA RTS/CTS n=3

2

4

Fig. 5.

6

8 10 n (number of users)

12

14

0

16

3

10 Average Packet Size (Bytes)

Pe and Pc for E2CA and BEB using ARF.

Fig. 7.

24 35 22 20

30

Goodput (Mbps)

Goodput (Mbps)

18 16 CSMA/E2CA Basic Access sim. CSMA/E2CA Basic Access analysis CSMA/CA Basic Access sim. CSMA/CA Basic Access analysis CSMA/CA RTS/CTS sim. CSMA/CA RTS/CTS analysis

14 12 10

System ARF goodput for n = 2 and n = 3 users

CSMA/E2CA Basic Access n=5 CSMA/E2CA Basic Access n=10 CSMA/E2CA Basic Access n=20 CSMA/CA Basic Access n=5 CSMA/CA Basic Access n=10 CSMA/CA Basic Access n=20 CSMA/CA RTS/CTS n=5 CSMA/CA RTS/CTS n=10 CSMA/CA RTS/CTS n=20

20

15

10

8 6 4

25

4

10

5 2

Fig. 6. Bytes.

4

6

8 10 n (Number of users)

12

14

3

16

System ARF goodput using fixed packet lengths of Lp = 1500

10 Average Packet Size (Bytes)

Fig. 8.

4

10

System ARF goodput for n = 5, n = 10 and n = 20 users

VII. C ONCLUSIONS thus increasing the system goodput. Note that the correct transmission mode selection is achieved without requiring any overhead increase derived of the use of additional packets, such as, the RTS/CTS signaling packets. Therefore, ARF using CSMA/E2CA with basic access also improves the ARF performance of CSMA/CA with RTS/CTS. The system goodput using different average packet lengths (Lp ) for CSMA/CA and CSMA/E2CA is presented in Figs. 7 and 8. The values of Lp have been modelled as a doubly truncated exponential distribution ranging between 40 and 10.000 bytes. Except for very large average packet length and number of users, ARF with CSMA/E2CA using basic access achieves higher goodput than CSMA/CA with either access schemes, specially for shorter Lp where the system goodput is doubled. Although RTS/CTS mechanism shortens the collision duration, the average collision time in CSMA/CA is higher than in CSMA/E2CA, specially for lower Lp values.

This work has presented a semi-analytical framework able to model the performance of various access schemes, namely, Basic Access, RTS/CTS and CSMA/E2CA, in combination with ARF, which is the most common AMC technique currently in use. The accuracy of the proposed model has been validated by contrasting it with simulation results within the context of IEEE 802.11n. Several important conclusions can be drawn from the numerical results. It has been confirmed that the presence of collisions greatly impairs the correct functioning of ARF. This implies that networks based on CSMA/CA with a large number of active users are bound to be largely inefficient. In contrast, when collisions are avoided or drastically reduced, ARF exhibits its full potential in maximising the operating data rate. This effect can be observed to some extent in the throughput attained by RTS/CTS and to an even larger extent in the results obtained for the combination of CSMA/E2CA and ARF. It can be safely stated that this combination, ARF

and E2CA, is the most attractive solution for network deployments where many users are active. ACKNOWLEDGEMENTS This work has been partially funded by MICINN and FEDER through projects TEC2008-02422 (COSMOS), TEC2011-25446 (AM3DIO) and GEPETO (TEC2008-0655), and Conselleria d’Educació, Cultura i Universitats del Govern de les Illes Balears through a PhD grant. R EFERENCES [1] Barcelo, J. and Toledo, A.L. and Cano, C. and Oliver, M., “Fairness and Convergence of CSMA with Enhanced Collision Avoidance (ECA),” in IEEE ICC, may 2010, pp. 1 –6. [2] J. Barcelo, B. Bellalta, C. Cano, A. Sfairopoulou, M. Oliver, and K. Verma, “Towards a Collision-Free WLAN: Dynamic Parameter Adjustment in CSMA/E2CA,” EURASIP Journal on Wireless Communications and Networking, vol. 2011, no. 708617, p. 11, 2011. R a high-performance [3] A. Kamerman and L. Monteban, “WaveLAN -II: wireless LAN for the unlicensed band,” Bell Labs technical journal, vol. 2, no. 3, pp. 118–133, 1997. [4] S. Kim, L. Verma, S. Choi, and D. Qiao, “Collision-Aware Rate Adaptation in multi-rate WLANs: Design and implementation,” Computer Networks, vol. 54, no. 17, pp. 3011 – 3030, 2010. [5] T. Joshi, D. Ahuja, D. Singh, and D. Agrawal, “Sara: stochastic automata rate adaptation for IEEE 802.11 networks,” IEEE Transactions on Parallel and Distributed Systems, vol. 19, no. 11, pp. 1579–1590, 2008. [6] J. He, D. Kaleshi, A. Munro, and J. McGeehan, “Modeling Link Adaptation Algorithm for IEEE 802.11 Wireless LAN Networks,” in IEEE ISWCS, Valencia, Spain, Sept. 2006. [7] J. Zhang, K. Tan, J. Zhao, H. Wu, and Y. Zhang, “A Practical SNRGuided Rate Adaptation,” in IEEE INFOCOM, Phoenix, AZ, April 2008. [8] H. Jung, T. Kwon, Y. Choi, and Y. Seok, “A scalable rate adaptation mechanism for IEEE 802.11e wireless,” in IEEE FGCN, vol. 1, JejuIsland, Korea, Dec. 2007. [9] G. Holland, N. Vaidya, and P. Bahl, “A rate-adaptive MAC protocol for multi-Hop wireless networks,” in ACM MobiCom, 2001, pp. 236–251. [10] J. Choi, J. Na, Y. sup Lim, K. Park, and C. kwon Kim, “Collision-aware design of rate adaptation for multi-rate 802.11 WLANs,” IEEE Journal on Selected Areas in Communications, vol. 26, no. 8, pp. 1366 –1375, 2008. [11] IEEE, “Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specifications Amendment 5: Enhancements for Higher Throughput,” IEEE Std 802.11n-2009, 2009. [12] A. Goldsmith, Wireless Communications. Cambridge University Press, 2005. [13] Y.-S. Choi and S. Alamouti, “A pragmatic PHY abstraction technique for link adaptation and MIMO switching,” IEEE Journal of Selected Areas in Communications, vol. 26, no. 6, pp. 960–971, 2008. [14] G. Foschini, “Layered space-time architecture for wireless communication in a fading environment when using multi-element antennas,” Bell Labs Technical Journal, vol. 1, no. 2, pp. 41–59, 1996. [15] G. Martorell, F. Riera-Palou, and G. Femenias, “Cross-layer fast link adaptation for MIMO-OFDM based WLANs,” Springer Wireless Personal Communications, vol. 56, no. 3, pp. 599–609, 2010. [16] IEEE, “Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specifications,” IEEE Std 802.11-2007 (Revision of IEEE Std 802.11-1999), Dec. 2007. [17] G. Holland, N. Vaidya, and P. Bahl, “A rate-adaptive MAC protocol for multi-hop wireless networks,” in ACM MobiCom, Rome, Italy, 2001. [18] J. Kermoal, L. Schumacher, K. Pedersen, P. Mogensen, and F. Frederiksen, “A stochastic MIMO radio channel model with experimental validation,” IEEE Journal of Selected Areas in Communications, vol. 20, no. 6, pp. 1211–1226, 2002. [19] G. Martorell, F. Riera-Palou, and G. Femenias, “Closed-Loop Adaptive IEEE 802.11n with PHY/MAC Cross-Layer Constraints,” LNCS, Multiple Access Communications, vol. 6886, pp. 63–74, 2011.