0 vues

Transféré par AshwinAnanthapadnabhan

My Journal Paper

- HP Spectre x360 15 User Guide
- IEEE 802.11
- IJETTCS-2014-12-25-157
- Wireless Site Survey Best Practices
- Tl-wr700n v1 User Gudie
- questions and answers assessment
- Ccna Wireless Lans
- e3_7
- CCNA 3 - Chapter 7
- ch7
- CCNA ESwitching Chapter 7 100% 11-02-2011
- DIR-635
- How to Troubleshoot 802.11n Speeds
- 79010533-CW-5350-User-Manual
- 3987285 CCNA3 v4 0 Exam Ch 7 Basic Wireless Concepts and Configuration
- 124019-338870-1-PB
- OW11 Paper Template
- 06167392
- arshad 3
- Smart Antennas

Vous êtes sur la page 1sur 30

IEEE TRANSACTIONS ON MOBILE COMPUTING

SUBMITTED TO TMC 1

Access Protocols with Multi-packet Reception

Yun Han Bae, Bong Dae Choi, Member, IEEE and Attahiru S. Alfa, Member, IEEE

Abstract

This paper considers random access protocols with multi-packet reception (MPR), which include

both slotted-Aloha and slotted -persistent CSMA protocols. For both protocols, each node makes

a transmission attempt in a slot with a given probability. The goals of this paper are to derive the

optimal transmission probability maximizing a system throughput for both protocols and to develop a

simple random access protocol with MPR which achieves a system throughput close to the maximum

value. To this end, we rst obtain the optimal transmission probability of a node in the slotted-Aloha

protocol. The result provides a useful guideline to help us develop a simple distributed algorithm

for estimating the number of active nodes. We then obtain the optimal transmission probability in

the -persistent CSMA protocol. An in-depth study on the relation between the optimal transmission

probabilities in both protocols shows that under certain conditions the optimal transmission probability

in the slotted-Aloha protocol is a good approximation for the -persistent CSMA protocol. Based on

this result, we propose a simple -persistent CSMA protocol with MPR which dynamically adjusts the

transmission probability depending on the estimated number of active nodes and thus can achieve

a system throughput close to the maximum value.

Index Terms

DCF.

Y.H.Bae is with the Department of Mathematics Education, Sangmyung University, 7, Hongji-dong, Jongno-gu, Seoul,

110-743, Republic of Korea (e-mail: yhbae@smu.ac.kr).

B.D.Choi is with Department of Mathematics, Sungkyunkwan University, Suwon, Republic of Korea (e-mail:

queue@korea.ac.kr)

A.S.Alfa is with Department of Electrical and Computer Engineering, University of Manitoba, Winnipeg, MB, Canada.(e-

mail:alfa@ee.umanitoba.ca)

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

IEEE TRANSACTIONS ON MOBILE COMPUTING

SUBMITTED TO TMC 2

I. I NTRODUCTION

A. Motivation

Random access protocols based on carrier sensing or non-carrier sensing have been used for

coordinating the channel access of users in various wireless networks. One of the representative non-

carrier sensing protocols is the slotted-Aloha. Various forms of slotted-Aloha protocols are widely used

in most of the current digital cellular networks, such as the Global System for Mobile communications

(GSM)1 , and in RFID systems as one of the popular anti-collision algorithms. CSMA-based algorithms

constitute a heart of contemporary wireless media access control technology. In the last decades,

various enhancements have been introduced to basic CSMA schemes, rst of all, in order to support

collision avoidance. One of the widely used CSMA algorithms is the -persistent CSMA protocol.

The performance analysis of the -persistent CSMA has gained a renewed interest recently since the

behavior of many CSMA protocols such as IEEE 802.11 DCF [2] and IEEE 802.15.4 [3] might be

studied by a corresponding -persistent CSMA protocol.

With advanced PHY-layer multi-packet reception techniques, it is possible for a receiver to receive

multiple packets transmitted concurrently. This new concept, referred to as multi-packet reception

(MPR), opens up new possibilities for enhancing the capacity of wireless networks. Some recent

efforts for standardization [1] and system design [5] to implement MPR have convinced researchers

that it will not take much time to deploy MPR-based wireless networks in the near future. When the

MPR-capability is employed in the physical layer, it is expected that MAC layer behaves differently

from what is commonly believed in the MAC protocols with conventional single-packet reception.

Therefore, in order to fully exploit the MPR-capability, it is important to nd the optimal transmission

probability of a node to maximize the system throughput in random access protocols with MPR. In

addition, it is strongly required to develop a random access protocol with MPR which achieves a

throughput close to the maximum value.

The performance of random access protocols such as the slotted-ALOHA and the -persistent

CSMA is characterized by the transmission probability of a node. System throughput depends on the

transmission probability of a node. We can expect that there might exist a transmission probability

(called optimal transmission probability) which maximizes the system throughput. Under the assump-

tion of single-packet reception capability, much effort has been devoted to guring out the optimal

1

In the GSM network, the control channels of the TDM channels use slotted-Aloha.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

IEEE TRANSACTIONS ON MOBILE COMPUTING

SUBMITTED TO TMC 3

transmission probability of a node. The performance of random access protocols with single-packet

reception has been extensively investigated and well understood during the last decade. In particular,

since the seminal work of Bianchi [9], there have been extensive efforts to characterize the performance

of 802.11-based WLANs under saturation and non-saturation conditions, respectively, for example see

[11], [10], [12], [13].

In the contexts of study on random access protocols with MPR, Ghez et al. [15], [16] made the rst

attempt to model a general MPR channel in random access based wireless networks, in which stability

properties of conventional slotted-Aloha with MPR were studied under an innite-node assumption.

Their study was extended to CSMA protocols by Chan et al. in [17] and to Aloha systems with MPR

under a nite-node assumption by Naware et al. [18]. The authors [17] focused on the value of the

maximum stable throughput and extended the framework in [15], [16] to grasp the effects of CSMA

with MPR and then found that carrier sensing facilitates an improvement in capacity. Recently, [19]

investigated a system throughput in contention-based MAC protocols with MPR. Chan et al. [14]

investigated the impact of MPR on CSMA protocol and proposed a performance-improving cross-

layer designed CSMA protocol for wireless networks with MPR. Zhang et al. [19] showed that when

the channel has an M -MPR capability, which means that the AP can decode successfully up to M

simultaneous packet transmissions, the throughput scales linearly with M in the slotted-Aloha protocol

with an innite population and a nite population. Specically, their result implies that the achievable

throughput per unit cost (MPR-capability) increases with the MPR capability of the channel. Their

result provides a strong incentive to deploy MPR in the next generation wireless networks. Zhang et

al. extended their investigation in [19] to the non-saturation case [20] and showed that super-linear

scaling also holds for safe-bounded-mean-delay and safe-bounded-delay-jitter throughput which are

dened as the maximum throughput that can be safely sustained with nite mean delay and delay

jitter, respectively.

Among all the works on MPR mentioned above, some ([14], [19], [20], [24]) focused on the

theoretical analysis of the achievable maximum throughput. The others ([15], [17], [18]) dealt with

the stability properties of MAC protocols with MPR. Little attention has been devoted to the study

of the property of the optimal transmission probability which maximizes system throughput. Besides,

the relationship between the optimal transmission probabilities for both the slotted-Aloha and the

-persistent CSMA protocols with MPR has not been addressed so far. To the best of our knowledge,

this paper is the rst attempt to deal with the optimal transmission probability to maximize the system

throughput in random access protocols, including both the slotted-Aloha and the -persistent CSMA

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

IEEE TRANSACTIONS ON MOBILE COMPUTING

SUBMITTED TO TMC 4

The optimal transmission probability clearly depends on the number of active nodes which compete

with each other for the channel access. Apart from the historical works presented above, some

researchers have focused on designing dynamic random access protocols which dynamically adjust

the transmission probability of node ( e.g., via tuning the backoff window size) depending on the

estimated number of active nodes in the network. To perform such a tuning optimally in a way that

the MAC protocol capacity is close to its theoretical bound, it is essential to design an algorithm which

is able to estimate the number of active nodes at runtime. To realize such algorithms under single

packet reception assumption, the main idea adopted in the literature [6], [7], [8] is to use a feedback

from the channel status to tune the transmission probability. The algorithm proposed in [8] computes

an estimate of the collision cost and of the number of active nodes, which are obtained by observing

the three events that may occur on the channel: idle slots, collisions, and successful transmissions. It

seems that the extension of the algorithm to the MPR case is not straightforward. Different from the

algorithm in [8], we introduce a new metric, which turns out to be a quantity related to the length bias

concept, to develop an algorithm for estimating the number of active nodes at runtime. Our proposed

algorithm simply requires each node to obtain the information on the number of nodes involved in

each successful transmission slot. Moreover, under certain conditions the proposed algorithm works

well in both the slotted-Aloha and the -persistent CSMA protocols with MPR.

In this paper, we consider random access protocols with MPR where there are an access point (AP)

and multiple nodes in a wireless network. The AP has an M MPR-capability. The study includes

both the slotted-Aloha and the -persistent CSMA protocols. For both protocols, each node makes a

transmission attempt with a given probability in a generic slot. The goals of this paper are to gure

out the relation between the optimal transmission probabilities for both protocols and, based on the

result, to develop a random access protocol which achieves a throughput close to the maximum value

in the MPR scenario. The main contributions of this work are summarized as follows:

First, we derive the optimal transmission probability in the slotted-Aloha with MPR. The result

reveals that the optimal transmission probability in the slotted-Aloha protocol with MPR is directly

related to a discrete version of the length biasing concept.

Second, based on the derived optimal transmission probability in the slotted-Aloha protocol with

MPR, we provide a simple distributed algorithm for estimating the number of active nodes (i.e.,

nodes that have packets ready for transmission) at runtime. By observing the number of successful

transmissions in a slot, each node can get an estimate of the number of active nodes and use this

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

IEEE TRANSACTIONS ON MOBILE COMPUTING

SUBMITTED TO TMC 5

estimate to tune its transmission probability. The results show that the proposed algorithm can

estimate the real value that we want to nd.

Third, we obtain the optimal transmission probability in the -persistent CSMA protocol with

MPR under the general settings Ti < Ts and Ti < Tc where Ti , Ts and Tc is the length of an idle

slot, a successful transmission slot and a collision slot, respectively. We show that under certain

conditions the optimal transmission probability in the slotted-Aloha protocol with MPR is a good

approximation of the one in -persistent CSMA protocol with MPR. This nding is very useful

because it allows us to directly apply the proposed algorithm for estimating the number of active

nodes in the slotted-Aloha protocol to the case of the -persistent CSMA protocol with MPR.

Finally, to improve the system throughput in the MPR case, we propose a simple -persistent

CSMA protocol with MPR, where the transmission probability of a node is dynamically and

optimally tuned based on the estimated number of active nodes in the network. The proposed

protocol can achieve a system throughput close to the theoretical maximum value.

The rest of this paper is organized as follows. The system model is presented in Section II. In

section III, we derive the optimal transmission probability in the slotted-Aloha and then propose a

new algorithm for estimating the number of active nodes at runtime. Section IV provides the optimal

transmission probability in the -persistent CSMA protocol. In addition, we show that the optimal

transmission probability in the -persistent CSMA protocol is well approximated by the one in the

slotted-Aloha protocol. In Section V, we investigate the performance of IEEE 802.11-based WLAN

with MPR and then show that IEEE 802.11 DCF with MPR gives a poor throughput. In Section VI,

we propose a simple -persistent CSMA protocol which achieves a throughput close to the maximum

value in WLANs with MPR. Some discussion is given in Section VII and nally Section VIII concludes

this paper.

We consider a fully connected one-hop wireless network consisting of N active nodes (nodes that

always have packets ready for transmission) and an AP. For the time being, for both the slotted-Aloha

and the -persistent CSMA protocols we begin with the assumption that the number N of active nodes

is known to all the nodes and then this assumption will be relaxed. The time is divided into slots and

packet transmissions start only at the beginning of a slot. After each transmission, we assume that

the transmitting nodes know the result of the transmission whether it is successful or not. In addition,

all the nodes in the network get the information on how many nodes are involved in each successful

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

IEEE TRANSACTIONS ON MOBILE COMPUTING

SUBMITTED TO TMC 6

transmission by employing an appropriate feedback mechanism. The AP has the capability to decode

up to M simultaneous transmissions. The MPR model used in this paper is described by

X = Y 1{Y M } . (1)

where M is the MPR capability of the AP, Y is the number of nodes which make a transmission

attempt in a time slot, X is the number of packets successfully received by the AP in a time slot

and 1{} is the indicator function. The above model means that the AP can receive up to M packets

simultaneously, and concurrent transmissions more than M packets result in collisions and so all

the packets are lost. Note that the conventional single-packet reception model is the special case of

M = 1. When N M , it is obvious that the optimal transmission probability which enables the

maximal throughput to be achieved is equal to one. In order to obtain non-trivial result, we assume

N > M.

In this section, we rst investigate the optimal transmission probability in the slotted-Aloha protocol

with MPR given that the number of active nodes is known to all the nodes. Based on the result, we

next present an algorithm for estimating the number of active nodes at runtime in a distributed manner.

Combining the algorithm with the result of the optimal transmission probability in the slotted-Aloha

leads to an optimal slotted-Aloha protocol which dynamically and optimally tunes the transmission

probability of node depending on the estimated number of active nodes.

We assume that the number N of active nodes is known a priori. In the slotted-Aloha protocol,

the channel time is divided into time slots of an equal length. We assume that the length of a packet

is constant. A time slot is long enough to accommodate a packet transmission and the corresponding

acknowledgement packet.

Each node attempts to transmit a packet with a probability at the beginning of a time slot,

0 < < 1. Since each node makes a transmission attempt with probability in a time slot, Y follows

a binomial distribution with parameters N and . Thus, its distribution is given by P(Y = k) =

N Ck

k (1 )N k . Let X denote the number of packets received successfully in a time slot. Then,

the probability mass function of X is given by

N

P(X = 0) = P(Y = 0 or Y > M ) = (1 ) + N

N Ck

k

(1 )N k

k=M +1

N k

P(X = k) = P(Y = k) = N Ck (1 ) k

, for 1 k M. (2)

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

IEEE TRANSACTIONS ON MOBILE COMPUTING

SUBMITTED TO TMC 7

Then, for a given , the system throughput Saloha ( ), which is measured in the number of successful

transmissions in a slot, is obtained as

M

M

Saloha ( ) = k P(X = k) = k N Ck k (1 )N k . (3)

k=1 k=1

We want to nd the optimal transmission probability aloha which maximizes a system throughput

Saloha ( ). Differentiating Saloha ( ) with respect to , we have

dSaloha ( )

M

= k N Ck k1 (1 )N k1 (k N ). (4)

d

k=1

dSaloha ( ) dSaloha ( ) dSaloha ( )

It is easy to see d > 0 for 1/N and d < 0 for M/N . Since d

dSaloha ( )

[1/N, M/N ] such that d = 0. The following result provides the properties of solution of

dSaloha ( )

d = 0.

dSaloha ( )

Theorem 1: 1) A solution of equation d = 0 satises the following equation:

E[X 2 ; ]/E[X; ]

= , (5)

N

M M

where E[X; ] = k=1 k N Ck k (1 )N k and E[X 2 ; ] = 2

k=1 k N Ck (1

k )N k .

E[X 2 ; ]/E[X; ] 1

2) Let h( ) = N . Then, lim 0+ h( ) = N, lim 1 h( ) = M

N and h( ) is a monotone

increasing function of .

d h( ) < 1, then Eq.(5) has the unique solution, denoted by aloha , and so Saloha ( ) has the

d

3) If

maximum value at = aloha .

dSaloha ( )

Proof: We prove the rst statement. d = 0 yields

M

M

2 N k1

k N Ck

k1

(1 ) =N k N Ck k (1 )N k1

k=1 k=1

1

M

N

M

k 2 N Ck k (1 )N k = k N Ck k (1 )N k

(1 ) 1

k=1 k=1

1 N

E[X 2 ; ] = E[X; ].

(1 ) 1

We next prove the second statement. Letting ak = N Ck , we have

M 2 N k

k=1 k ak (1 )

k

h( ) =

N M k=1 kak (1 )

k N k

M 2 k

k=1 k ak 1

= M k . (6)

N k=1 kak 1

1

It is easy to see that lim 0+ h( ) = N and lim 1 h( ) = M

N.

1 in Eq.(6), we note that as increases from 0

M 2 k

k=1 k ak x

to 1, x also increases from 0 to . Let q(x) =

N M ka x k . Thus, it is sufcient to show that q(x) is

k=1 k

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

IEEE TRANSACTIONS ON MOBILE COMPUTING

SUBMITTED TO TMC 8

monotone. We claim that q(x) is strictly increasing, namely, q(u + v) > q(u) for any u + v, v (0, 1),

and u > 0 and v > 0. Letting bk = kak , then we have

M M M M

k=1 bk u k=1 kbk (u + v) k=1 kbk u k=1 bk (u + v)

k k k k

q(u + v) q(u) = M . (7)

N M k=1 b k (u + v) k

k=1 b k u k

Since the denominator of (7) is always positive, we only have to show the numerator of (7) is positive.

Denoting the numerator of (7) by c(u, v), simple algebra then yields

M M M M M

M

c(u, v) = bi u

i

bj (u + v)

j

bj u

j

bi (u + v) i

M

M

M

M

M

= bi u

i

bj (u + v) j

bj u

j

bi (u + v) i

M

M

M

M

M

= bi u

i

bj (u + v)j bj u j bi (u + v)i

k=2 i=1 j=k j=k i=1

M

k1 M

M

M

k1

M

= bi u +

i

bi u i

bj (u + v)j bj u j bi (u + v)i + bi (u + v)i

k=2 i=1 i=k j=k j=k i=1 i=k

M

k1

M

M

k1

= bi u i bj (u + v)j bj u j bi (u + v)i

k=2 i=1 j=k j=k i=1

M

k1

M

M

= b i ui bj (u + v)j bi (u + v)i bj u j . (8)

k=2 i=1 j=k j=k

()

Finally, we only have to prove that () of (8) is positive. Rearranging the terms of (), we have

M

() = bi ui bj (u + v)j bi (u + v)i bj uj

j=k

M

= bi bj ui (u + v)i (u + v)ji uji . (9)

j=k

The proof of the third statement can be done trivially from the second result.

d

Remark 1: It seems that

d h( ) < 1 for M = 1, 2, 3, but we cannot prove it for an arbitrary M because it involves extremely

d

d

complex algebraic manipulations. To show the plausibility that

is presented in Fig.1, which plots h( ) for the varying and M when the number N of contending

nodes is 10. To emphasize that h( ) is a function of and M for a given N , let h(, M ) = h( ).

d

We see from the following facts that

First, we note that for a xed , h(, M ) increases with M . This is because E[Z] increases with

M as explained in remark 3.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

IEEE TRANSACTIONS ON MOBILE COMPUTING

SUBMITTED TO TMC 9

N 1 1

Second, when M = 10, i.e., N = M , then h(, 10) = N + N. This is due to the following

reasoning: when M = N , Y and X are identically distributed where Y has binomial distribution

with parameters N and ; therefore, E[X; ] = N , Var[X; ] = N (1 ), and so E[X 2 ; ] =

E[X 2 ; ] N 1 1 N 1

N (1 ) + N 2 2 ; hence, h(, 10) = N E[X ] = N + N. Thus, d h(, 10)

d

N < 1 for

any M .

N 1

d h(, M ) d h(, 10) <1

d d

Third, for any M , N

d h( ) < 1.

d

Hereafter, we assume that

Remark 2: (Fixed-point iteration) Let aloha be the unique solution satisfying Eq.(5). The solution

aloha can be found by the following xed-point iteration method. Dene n = h(n1 ) n = 1, 2,

and 0 (0, 1). Then, with the result d h( )

d

< 1, the well-known Fixed-Point Theorem ensures that

n converges to the solution aloha . For a wide range of 0 (0, 1), we identify that the xed-point

iteration always converges to the unique solution.

Remark 3: (Interpretation of the quantity h( ) as the length bias concept) Note that the quantity

E[X 2 ; ]

E[X; ]

given by Eq.(5) is related to a discrete version of the length biasing concept which is explained

in the following. Let X be the number of nodes involved in a successful transmission slot. Then,

P(X=k)

P(X = k) = P(1XM )

for 1 k M . Choose a packet randomly among the total packets

which have been successfully transmitted. Let Z be the number of nodes involved in a successful

transmission slot in which the tagged packet has been transmitted successfully. Then, the distribution

of Z is given by

k P(X = k) k P(X = k)

P(Z = k) =

= , 1 k M, (10)

E[X ] E[X]

E[X]

where the second equality follows from E[X ] = P(1XM )

and we here denote E[X ; ] and E[X; ]

E[X 2 ]

by E[X ] and E[X], respectively. Thus, we have E[Z] = E[X]

, which explains the fact that the

successful transmission slot including the tagged packet involves more packet transmissions than an

ordinary successful transmission slot.

In the previous section, we have derived the optimal transmission probability in the slotted-Aloha

with MPR, given that the number of active nodes is known a priori. This assumption is restrictive in

the distributed wireless networks. In this section, we present an algorithm for estimating the number

of active nodes. Note that if each node can estimate the number of active nodes, we can obtain the

optimal transmission probability by the xed point iteration as shown in remark 2.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

IEEE TRANSACTIONS ON MOBILE COMPUTING

SUBMITTED TO TMC 10

1

M=10

0.9

M=9

0.8

y=((N1)+1)/N

0.7

0.6

y=h()

0.5

0.4

0.3

M=3

0.2

M=2

0.1 M=1

y=

0

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

E[X 2 ; ]/E[X; ]

Fig. 1. y = h( ) = N

for the varying M and when the number of nodes is 10.

Lets focus on the quantity E[X 2 ; ]/E[X; ]. Note that h(, N ) E[X 2 ; ]/E[X; ] is a function

of and N and, for a xed N , h(, N ) is an increasing function of . In addition, h(, N ) is also

an increasing function of N for xed and M values. The formal proof for that is not presented

here because it involves tedious algebraic manipulations. The reasoning for the result is intuitively

explained as follows: as explained in remark 3 h(, N ) = E[X 2 ; ]/E[X; ] is equal to E[Z] where

Z is the number of nodes involved in a successful transmission slot in which the tagged packet has

been successfully transmitted; note that for xed and M , a larger N implies that more nodes are

likely to attempt to transmit; thus, each successful transmission is likely to involve more nodes and

so it is expected that E[Z] increases with N .

The proposed algorithm regularly updates the necessary estimates of the network status at the end

of every update interval2 with a length of L time slots. The update intervals of all the nodes are

synchronized with each other. During every update interval, each node can obtain the estimates 1

and 2 of E[X] and E[X 2 ], respectively, by overhearing acknowledgment packets broadcasted by the

AP, which contains the information on the number of successful transmissions in every time slot. Let

J be the number of successful transmission slots during the update interval. Then,

J J 2

j=1 xj j=1 (xj )

1 = , 2 = , (11)

J J

where xj is the number of nodes involved in the j -th successful transmission slot during the update

interval.

2

The update interval can be set as the interval between two consecutive beacons, which is usually used in wireless network

for the purpose of time synchronization.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

IEEE TRANSACTIONS ON MOBILE COMPUTING

SUBMITTED TO TMC 11

Suppose that there are N active nodes which we want to estimate. Let Nn be an estimate of N at the

end of the n 1-th update interval (or at the beginning of the n-th update interval). Let aloha (Nn ) be

the optimal transmission probability which satises Eq.(5) for a given Nn , which can be obtained by

a xed point iteration. All the nodes try to transmit packets with probability aloha (Nn ) at every time

slot during the n-th update interval. The proposed algorithm tries to estimate the number of active

nodes while estimating the quantities 1 and 2 in every update interval. The proposed algorithm

simply operates as follows:

(1) Calculation of 1 and 2 : At the end of the n-th update interval (or at the beginning of the

2

(n + 1)-th update interval), calculate 1 and 2 according to Eq.(11), and then obtain 1 .

(2) Updating the estimated number Nn+1 of active node: Let N be the solution which satises

h( (Nn ), ) = 2

1 for given (Nn ) and 2

1 , i.e., N = h1 ( 2 , (Nn )) 3 . Update Nn+1 as

1

follows:

Nn+1 :=

Nn + (1 )N , (12)

x is the least integer not less than x.

(3) Calculation of aloha (Nn+1 ): Based on the estimate Nn+1 , calculate the optimal transmission

probability aloha (Nn+1 ) by applying the xed-point iteration presented in remark 2. During the

(n + 1)-th update interval, all the nodes use aloha (Nn+1 ) as their transmission probability.

Note that in order to obtain reliable estimates by avoiding harmful uctuations, the proposed algorithm

uses a smoothing factor , which is widespread in the network protocols.

Remark 4: The proposed algorithm seems to work well under the following assumption: the number

of active nodes infrequently varies over update intervals and the communication time of an active node

is relatively long, i.e., the time required for an active node to complete the transmission of its whole

packets is not less than an update interval. This assumption is usually satised in the scenario of le

upload of nodes. In such a case, the completion time usually amounts to several seconds to dozens

of seconds.

Now, we present some simulation examples that show the performance of the proposed algorithm.

Specically, we investigate how fast the proposed algorithm nds out the actual number of active nodes

and as well how the estimated value is close to the real value. Results presented in Fig.2 and Fig.3

are obtained under the following assumptions: M = 5; there are 10 active nodes which we want to

estimate, i.e., N = 10; all the nodes initially (at the rst update interval) guess that there are 50 active

3

Note that since h(, N ) is an increasing function of N , the solution N is unique

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

IEEE TRANSACTIONS ON MOBILE COMPUTING

SUBMITTED TO TMC 12

1

nodes in the network4 , i.e., N0 = 50; all the nodes uses 0 = N0 as an initial transmission probability

during the rst update interval. Each sample path in the gure is a result of real estimation procedure

of the number of active nodes during the designated time period. We run 30 independent simulations

and so the lines in each gure display 30 real trajectories. We rst discuss the performance of the

proposed algorithm with a sufciently long update interval. To see the performance of the algorithm

for this ideal case, the length L of an update interval is set as L = 1000 slots and = 0. Note that

= 0 means that the proposed algorithm uses only the latest information for the estimation of the

number of active nodes. As shown in Fig.2, starting from an initial value 50, as the estimation process

goes on, the estimate for N sharply decreases to around 10 within a few update intervals (we call

this period transient period). After the transient period, the estimate keeps being almost same as the

actual number of active nodes (we call this period quasi-stationary period). It is worth noting that the

error between the estimate and actual value is at most one. The reasoning for this result is as follows:

in the case of sufciently long update interval, the law of large number ensures that the estimates 1

and 2 given by Eq.(11) is close to E[X] and E[X 2 ], respectively, and thus are sufcient statistics

to represent E[X] and E[X 2 ]. It is worth noting that in all simulations the transient period of the

proposed algorithm is less than 3 update intervals and thus the convergence speed of the estimation

process is so fast.

We next investigate the performance of the proposed algorithm with L = 200 for the varying .

Fig.3 shows the estimation process of the number of active nodes according to the increase of update

intervals for = 0.3 and = 0.7. In both cases, we see that after a transient period the proposed

algorithm is in quasi-stationary period as similar to Fig.2. Fig.3 indicates that a small leads to a

short transient period and vice versa. The reasoning for that is as follows: if is small, the algorithm

regards the latest information as more important and thus promptly reacts to the latest information. On

the other hand, since with a large the algorithm slowly reacts to the latest information, the estimation

curve smoothly decreases and thus the transient period become long. However, it is worth noting that

when the algorithm is in quasi-stationary period, the error between the estimate and the actual value

is at most less than 2 in all simulations for both cases. Moreover, despite wrongly assuming a highly

congested network initially (N0 =50), the proposed algorithm quickly converges to quasi-stationary

state.

To see how long the transient period is, assume that the length of a time slot is 1 ms, i.e., Ti =

4

Even if an initial guess N0 for the actual value N is different for each node, after an update interval all the nodes have

the same estimate N1 . Moreover, with N0 < N the proposed algorithm still works well.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

IEEE TRANSACTIONS ON MOBILE COMPUTING

SUBMITTED TO TMC 13

Ts = Tc = 1 ms. For each case, the maximum length of transient period, denoted by Ttrans , is as

follows: with L = 200 in Fig.3, Ttrans = 1000 ms ( = 0.3) and Ttrans = 2000 ms ( = 0.7).

Therefore, in both cases the transient period is not larger than 2 seconds. In summary, using large L

and values leads to reliable estimates in quasi-stationary period by avoid harmful uctuations at the

expense of long transient period. Thus, it is desirable to use a long update interval with a small or

a short update interval with a large .

50

45

The estimated number of active nodes

40

35

30

25

20

15

10

5

0 5 10 15 20

The index of update interval

Fig. 2. The estimated number of active nodes versus update interval when the length of an update interval is 1000 slots

and = 0

50 50

45 45

The estimated number of active nodes

40 40

35 35

30 30

25 25

20 20

15 15

10 10

5 5

0 5 10 15 20 0 5 10 15 20

Update interval Update interval

Fig. 3. The estimated number of active nodes versus update interval (The length of an update interval = 200 slots)

-persistent CSMA protocol is an important model in the class of CSMA protocols. It is well-known

that the -persistent CSMA protocol is a good candidate to approximate the performance of IEEE

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

IEEE TRANSACTIONS ON MOBILE COMPUTING

SUBMITTED TO TMC 14

802.11 DCF in both single-packet reception model [9] and as well MPR model [19], [21]. In this

section, we rst obtain the optimal transmission probability in the -persistent CSMA protocol with

MPR with the assumption that the number N of active nodes is known a priori. This assumption will be

relaxed in section VI. We next investigate the relation between the optimal transmission probabilities

for both the slotted-Aloha and the -persistent CSMA protocols with MPR. Based on that, we identify

that under certain conditions the optimal transmission probability in the slotted-Aloha protocol is a

good approximation of the one in the -persistent CSMA protocol.

We assume that the number N of active nodes is known a priori. The time is divided into unit slots

with a length . The length of a data packet is assumed to be constant. In the -persistent CSMA

protocol, each node attempts to transmit a packet with a probability at the beginning of a generic

slot where in CSMA protocols the length of a generic slot is not necessarily xed and may vary

under different contexts as follows: an idle slot with length Ti , a collision slot with length Tc and a

successful slot with length Ts . Thus, according to the employed carrier sensing technique, Ti , Ts and

Tc can be different. For example, in IEEE 802.11 DCF basic access mechanism,

Ti =

where is the time for a node to detect the packet transmission from any other node and is typically

much smaller than Ts and Tc ; H is the transmission time of both PHY and MAC headers; Pdata is

the data amount of a packet; ACK is the transmission time of an ACK packet; is the propagation

delay; and SIF S and DIF S are the interframe space durations; R is the data transmission rate of

a node; EIF S is the extended interframe space time, which includes the transmission time of ACK

frame, SIFS and DIFS.

Let Pi be the probability that a given slot is idle. Since if no node attempts to transmit in a slot, then

the slot is idle, Pi P(Y = 0) = (1 )N where Y is the number of nodes that make a transmission

attempt in a generic slot. Let Ps be the probability that a given slot is a successful transmission slot.

Since with M -MPR capability the AP can successfully decode up to M simultaneous transmissions,

N k where X is the number of packets successfully

we have Ps P(X > 0) = M k=1 N Ck (1 )

k

received by AP in a generic time slot. Let Pc be the probability that a given slot is a collision slot.

N k and clearly P + P + P = 1.

Then, Pc P(Y > M ) = N k=M +1 N Ck (1 )

k

i s c

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

IEEE TRANSACTIONS ON MOBILE COMPUTING

SUBMITTED TO TMC 15

Dene system throughput S( ) as the average amount of bits transmitted successfully per second.

Then, S( ) can be obtained as the ratio between the average payload bits transmitted per generic slot

to the average length of a generic slot as follows:

Pdata M k=1 k P(X = k)

S( ) =

Pi Ti + Ps Ts + Pc Tc

N k

Pdata M k=1 k N Ck (1 )

k

= M N

. (14)

Ti (1 )N + Ts k=1 N Ck k (1 )N k + Tc k=M +1 N Ck k (1 )N k

We want to nd the optimal transmission probability which maximizes the system throughput

S( ) by solving d S( )

d

= 0. The following result gives a functional equation for such .

d

Theorem 2: A solution of

(E[X; ])2

E[X 2 ; ] g( ) (Ts Tc ) E[X 2 ; ]

= N, (15)

E[X; ] Tc

where g( ) = Pi Ti + Ps Ts + Pc Tc .

N k and g( )

Proof: For notational convenience, let ak =N Ck , f ( ) M k=1 kak (1 )

k

N k + T

N N k . Noting that dS( ) = 0

Ti (1 )N + Ts M k=1 ak (1 )

k

c k=M +1 ak (1 )

k

d

df dg

d g( ) f ( ) d = 0, simple algebraic manipulations yield

df dg

gf =0

d d

M

M

2 N k1 N k

g( ) k ak k1

(1 ) +N kak (1 )

k

Ti (1 )N 1

k=1 k=1

M

N

N k1 N k1

+Ts ak (1 )

k

+ Tc ak (1 )

k

k=1 k=M +1

M

M

M

N k1 N k

= N g( ) kak (1 )

k

+ kak (1 )

k

Ts kak k1 (1 )N k1

k=1 k=1 k=1

N

+Tc kak k1 (1 )N k1

k=M +1

N

E[X 2 ; ] N E[X; ] N E[X; ]

g( ) + g( ) = g( ) + E[X; ] Tc kak k1 (1 )N k1

(1 ) 1 1

k=1

M

+(Ts Tc ) kak k1 (1 )N k1 . (16)

k=1

df dg

gf =0

d d

E[X 2 ; ] Tc Ts Tc

g( ) = E[X; ] E[Y ; ] + E[X; ]

(1 ) (1 ) (1 )

E[X 2 ; ]g( ) = E[X; ](Tc N + (Ts Tc )E[X; ]), (17)

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

IEEE TRANSACTIONS ON MOBILE COMPUTING

SUBMITTED TO TMC 16

We discuss about the uniqueness of the solution of Eq.(15) to obtain the optimal transmission

probability which maximizes S( ). Note that if Ti = Tc = Ts , then the -persistent CSMA becomes

the slotted-Aloha and thus Eq.(15) reduces to Eq.(5).

We consider two cases separately; Case (1) Ti < Ts = Tc and Case (2) Ti < Tc < Ts . First, we

consider Case (1). Then, Eq.(15) reduces to

E[X 2 ; ] Ti

= 1 Pi (1 ) N. (18)

E[X; ] Tc

E[X 2 ; ]

Let k( ) = E[X; ] 1 Pi (1 Tc )

Ti

N. The following result shows the uniqueness of the solution

of Eq.(18)

Theorem 3: 1) Eq.(18) has at least one solution and k( ) is a monotone increasing function of

.

d2

2) If d 2 k( ) < 0, which implies that k( ) is concave in , then Eq.(18) has a unique solution and

so S( ) has a maximum value at such .

Proof: We prove the rst statement. We have

E[X 2 ; ] Ti Ti

lim 1 Pi (1 ) N = >0

0 E[X; ] Tc N Tc

E[X 2 ; ] Ti M

lim 1 Pi (1 ) N = < 1. (19)

1 E[X; ] Tc N

Therefore, Eq.(18) has at least one solution by Intermediate value theorem. Note that (1 Pi (1 TTci ))

E[X 2 ; ]

is a monotone increasing function of and E[X; ]

is also a monotone increasing function of .

Therefore, k( ) is a monotone increasing function of .

The proof of the second statement can be done trivially from the fact (19) together with the

assumption.

d2

Remark 5: It seems that d 2 k( ) < 0 holds, which is assumed in theorem 3. We are able to prove

d2

that d 2 k( ) < 0 holds for M = 1, 2. But, we cannot prove it for general M . To show the plausibility

of the assumption, we provide some numerical examples. In Fig.4, k( ) is plotted for the varying M

and when the number N of nodes is 10. First, we see that the solution of Eq.(18) is unique since

E[X 2 ; ]

k(0+) = Ti

N Tc . Second, we observe that for a xed k( ) increases with M . This is because E[X; ]

increases with M and Pi is constant over M . Third, note that, for a given M , k( ) looks concave in

.

d2

Hereafter, we assume that d 2 k( ) < 0.

Remark 6: (Fixed-point iteration) Let be the unique solution of Eq.(18). The solution can

be found using the following xed-point iteration method. Dene n = k(n1 ) for n = 1, 2, and

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

IEEE TRANSACTIONS ON MOBILE COMPUTING

SUBMITTED TO TMC 17

0 ( , 1). If k( ) is concave, then this implies that k( ) < 1 for ( , 1). Thus, the xed-point

iteration theorem ensures that n converges to . We recommend to use the value 1 or a value close

to 1 as an initial value 0 .

1

M=10

0.9

M=9

0.8

M=8

0.7

0.6

y=k()

0.5

y=

0.4

0.3 M=3

0.2

M=2

0.1

M=1

0

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

E[X 2 ; ]

Fig. 4. y = k( ) = E[X; ]

1 Pi (1 Ti

Tc

) N for the varying and M when the number of nodes is 10.

Next, we consider Case (2); Ti < Tc < Ts . By checking many numerical examples, we observe

that Eq.(15) has a unique solution when the difference Ts Tc is small.

We next obtain more useful version (20) which is equivalent to Eq.(15). Applying Chebyshevs

E[(U E[U ])2 ]

version of the second moment method, i.e., P(U = 0) E[U ]2

where U is a nonnegative random

(E[X]) 2

(E[X])2

variable, and keeping in mind that P(X > 0) = Ps , we have E[X 2 ]

Ps . Let ( ) = Ps E[X 2 ]

,

then 0 Ps where depends on a given . Then, Eq.(15) can be rewritten as

(E[X; ])2

E[X 2 ; ] g( ) (Ts Tc ) E[X 2 ; ]

= N

E[X; ] Tc

(E[X; ])2

E[X 2 ; ] Pi Ti + Ps Ts + Pc Tc (Ts Tc ) E[X 2 ; ]

= N

E[X; ] Tc

E[X 2 ; ] Ti Ts Tc

= Ps + Pc + Pi + N

E[X; ] Tc Tc

E[X 2 ; ] Ti Ts Tc

= 1 Pi 1 + N. (20)

E[X; ] Tc Tc

To emphasize that the right-hand side of Eq.(20) is a function of M and N , we let

E[X 2 ; ] Ti Ts Tc

k(M, N ) = 1 Pi 1 + N. (21)

E[X; ] Tc Tc

Using Eq.(20), in the subsequent sections we investigate approximations of the optimal transmission

probabilities for two cases; Case (1) Ti < Ts = Tc and Case (2) Ti < Tc < Ts .

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

IEEE TRANSACTIONS ON MOBILE COMPUTING

SUBMITTED TO TMC 18

B. Approximations of Optimal Transmission Probabilities for Case (1) and Case (2)

In this section, we show that the optimal transmission probability in the -persistent CSMA protocol

with MPR is well approximated by the one in the slotted-Aloha protocol with MPR.

10 10

Approximation Approximation

9 Exact 9 Exact

N=10

8 N=10 8

Maximum system throughput

7 7 N=30

N=30

6 6

5 5

N=50 N=50

4 4

3 3

2 2

1 1

0 0

1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10

MPRcapability M MPRcapability M

Fig. 5. The maximum throughput (measured in the number of packets successfully transmitted per slot) versus MPR-

capability M for different N in case of carrier sensing protocol of Case (1).

0.07

M=4

M=5

0.06 M=7

0.05

The idle probability Pi

0.04

0.03

0.02

0.01

0

5 10 15 20 25 30 35 40 45 50

The number N of nodes

E[X 2 ; ]

Fig. 6. The probability Pi = (1 )N at satisfying the equation k(M, N ) = E[X; ]

1 Pi (1 TTci ) N for a given

M

E[X 2 ; ] Ti

k(M, N ) = 1 Pi (1 ) N. (22)

E[X; ] Tc

be the optimal transmission probability for Case (1) which is a solution of Eq.(22). Note

Let (1)

that aloha is the optimal transmission probability in the slotted-Aloha protocol with MPR satisfying

; ] E[X 2 ; ]

the equation = EE[X

2

[X; ] N . Then, since k(M, N ) E[X; ] N , it is obvious that (1) aloha .

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

IEEE TRANSACTIONS ON MOBILE COMPUTING

SUBMITTED TO TMC 19

=

Note that if M = N , then (1)

aloha = 1. Also, note that (1) increases with M . This implies

)N decreases and is close to 0 for a large M . Thus, is close to

Pi = (1 (1) (1) aloha for a relatively

large M . On the other hand, lets consider the probability Pi for a xed M . We investigate how Pi

varies depending on the number N of nodes. Fig.6 depicts Pi versus N for given M = 4, 5, 7 where

)N . It is obvious to see that P decreases

Pi is evaluated at satisfying Eq.(22), i.e., Pi = (1 (1) i

with M for a given N . It is worth noting that Pi 0 for M 4 regardless of N . This implies that

E[X 2 ; ] . Thus, is close to

the right-hand side of Eq.(22) is well approximated by E[X; ]

near (1) (1) aloha

for M 4 regardless of N . Numerical results show below that as long as M 4 the maximum

) achieved at the optimal = is quite close to the one S(

throughput S((1) (1) aloha ) achieved at

= aloha .

Fig.5(a) and Fig.5(b) display the maximum throughput versus MPR-capability M for the various

number of active nodes in the case of Ts = 500 and Ts = 50, respectively. In gures, approximation

means the maximum throughput S( ) given by Eq.(14) achieved at the = aloha satisfying Eq.(5)

satisfying Eq.(20). As

and exact means the actual maximum value of S( ) achieved at = (1)

shown in Fig.5(a) and Fig.5(b), note that as long as M 4, aloha serves as a good approximation

for (1) .

0.07 0.14

M=4 M=4

M=5 M=5

0.06 M=7 0.12 M=7

0.05 0.1

i

The probability P

0.04 0.08

()

0.03 0.06

0.02 0.04

0.01 0.02

0 0

5 10 15 20 25 30 35 40 45 50 5 10 15 20 25 30 35 40 45 50

The number N of nodes The number N of nodes

N

(a) The probability Pi = (1(2) ) at = (2) satisfying (b) ( ) at = (2) satisfying the Eq.(20) for a given M

the Eq.(20) for a given M

Fig. 7. Pi and ( ) at = (2) satisfying the Eq.(20) with Ts = Tc + 1 and Tc = 30

2) Case (2) Ti < Tc < Ts : We rst note that the relation between Ts and Tc in IEEE 802.11 DCF

basic access mechanism falls on this case, where Ts , Tc and Ti are given by Eq.(13). Specically,

Ti
Ts , Ti
Tc , Ts Tc = , Ti /Tc 0 and (Ts Tc )/Tc =

Tc 0.

the optimal transmission probability for Case (2) which

Let us consider Eq.(20). We denote by (2)

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

IEEE TRANSACTIONS ON MOBILE COMPUTING

SUBMITTED TO TMC 20

E[X 2 ; ] Ti E[X 2 ; ] Ts Tc

1 Pi (1 ) N k(M, N ) 1+ N. (23)

E[X; ] Tc E[X; ] Tc

E[X 2 ; ]

As in Case (1), we claim that the lower bound (the rst term in (23)) of k(M, N ) is close to E[X; ]

N

regardless of N as long as M 4. Fig.7(a) depicts the idle probability

in the neighborhood of (2)

satisfying Eq.(20) with T = T + 15

Pi versus N for given M = 4, 5, 7 where Pi is evaluated at (2) s c

and Tc = 30. Obviously, Pi decreases with M for a given N . From Fig.7(a), we notice that Pi 0

regardless of N as long as M 4. This implies that the lower bound of k(M, N ) in

at = (2)

; ]

(23) is close to EE[X

2

[X; ] N in the neighborhood of (2) . Next, consider the upper bound of K(M, N )

with N for a given

in (23). Note that ( ) Ps . Fig.7(b) displays how ( ) varies at the = (2)

M . We rst notice that ( ) decreases with M for a given N . For a given M , ( ) increases with

Ts Tc

N , but it seems bounded below by a certain value less than Ps as N increases. Moreover, if Tc

c

0. Thus, this implies that the upper bound of K(M, N ) in (23) is

well approximated by EE[X ; ]

2

[X; ] N in the neighborhood of = (2) . From the above two observations,

; ]

k(M, N ) EE[X

2

[X; ] N and as a result (2) aloha regardless of N as long as M 4.

Fig.8(a) and Fig.8(b) show the maximum throughput versus (Ts Tc )/Tc for the varying N and

Tc = 50 in case of M = 5 and M = 9, respectively. In gures, approximation means the maximum

throughput S( ) given by Eq.(14) achieved at = aloha satisfying Eq.(5) and exact means the

satisfying Eq.(20). As (T T )/T increases,

actual maximum value of S( ) achieved at = (2) s c c

which means the difference Ts Tc becomes big, the discrepancy between the approximated and exact

values becomes big as expected. We can observe that as long as (Ts Tc )/Tc < 0.4, the difference is

very small and thus the optimal transmission probability aloha in the slotted-Aloha protocol can be

in the -persistent CSMA

used as a good approximation of the optimal transmission probability (2)

protocol.

With the practical settings of system parameters Ti , Ts and Tc used in IEEE 802.11 DCF basic

access mechanism, we next investigate how well the optimal transmission probability aloha satisfying

in the -persistent CSMA protocol with

Eq.(5) approximates the optimal transmission probability (2)

MPR. The system parameters are presented in Table I and are adopted from IEEE 802.11g. Fig.9

compares the maximum approximated system throughput and the exact system throughput for the

-persistent CSMA protocol with the settings of system parameters Ti , Ts and Tc given by Eq.(13)

for the varying MPR-capability M when N = 10, 40 and packet payload size Pdata =8184 bits,

5

This setting is the case of IEEE 802.11 DCF basic access mechanism.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

IEEE TRANSACTIONS ON MOBILE COMPUTING

SUBMITTED TO TMC 21

3.4

approximation

3.3 exact

7 N=10

3.2

maximum throughput

maximum throughput

3.1 N=10 6.5

3

6 N=20

2.9 N=20

2.8

5.5

N=50 N=50

2.7 approximation

exact

5

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

(T T )/T (T T )/T

s c c s c c

(a) M =5 (b) M =9

Fig. 8. The maximum throughput (measured in the number of packets successfully transmitted per slot) versus (Ts Tc )/Tc

for varying N when Tc = 50 backoff slots

300 180

N=10 (exact) N=10 (exact)

N=10 (approximated) 160 N=10 (approximated)

N=40 (exact) N=40 (exact)

250

N=40 (approximated) N=40 (approximated)

140

System throughput (Mbps)

200 120

100

150

80

100 60

40

50

20

0 0

1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10

MPRcapability M MPRcapability M

(a) Packet payload size = 8184 bits (b) Packet payload size = 4092 bits

Fig. 9. The comparison of the approximated throughput (the value S(aloha ) achieved at = aloha satisfying Eq.(5))

and the exact maximum throughput (the throughput value S((2) ) achieved at = (2) satisfying Eq.(15)) for -persistent

CSMA protocol with the settings of system parameters given by Eq.(13) when packet payload size is set as 8184 bits and

4092 bits, respectively

4092 bits. Specically, in Fig.9,exact denotes the actual maximum throughput of S( ) given by

) achieved at = , and approximated denotes the

Eq.(14), i.e., the throughput value S((2) (2)

throughput value S(aloha

) achieved at = aloha satisfying Eq.(5). As shown in Fig.9, as long as

M 4, with the practical settings of system parameters used in IEEE 802.11g, we again see that

there is a good match between the approximated and actual values. Keeping in mind that aloha serves

as long as (T T )/T 0, the transmission probability

as a good approximation for (2) s c c aloha

satisfying Eq.(5) can be used as a good approximation for the optimal transmission probability (2)

for the -persistent CSMA protocol with system parameters such as IEEE 802.11 DCF basic access

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

IEEE TRANSACTIONS ON MOBILE COMPUTING

SUBMITTED TO TMC 22

mechanism.

Summary: We would like to conclude this section with the brief summary of the obtained results

and the emphasis on its practical meaning.

(A) We have the result that in the MPR case, the optimal transmission probability satisfying

Eq.(15) in the -persistent CSMA protocol is very close to the one aloha satisfying Eq.(5) under

Ts Tc

the conditions; (i) Ti < Tc Ts and Tc 0 and (ii) M 4. This implies that we can use

the aloha satisfying Eq.(5) as a good approximation for the optimal transmission probability ,

instead of resorting to solving Eq.(15), which is very complex6 , to obtain in the -persistent

CSMA protocol with MPR.

(B) The above result (A) gives us an useful implication in terms of practical algorithmic aspects.

Recalling the proposed algorithm presented in section III-B, for the purpose of estimating the

number of nodes the proposed algorithm does not require a node to obtain any additional

E[X 2 ]

information, such as idle slots and collision slots, except the estimates for the quantity E[X]

. It

E[X ]2

is worth noting that E[X]

is a invariant quantity that can be obtained regardless of whether a

protocol is based on carrier sensing or not. Thus, the algorithm proposed in section III-B can

be directly applied to -persistent CSMA protocol with MPR to estimate the number of active

nodes. This issue will be discussed in section VI.

For the conventional IEEE 802.11 DCF with single-packet reception, Bianchi [9] developed an

analytical model to analyze the performance of the IEEE 802.11 DCF and thus obtained the expression

for saturation throughput. His derivation is based on modeling the stochastic behavior of a tagged

node by a bi-dimensional embedded Markov chain represented by its backoff stage s(t) and backoff

counter b(t) at time t. The key approximation made in [9] is that at each transmission attempt, each

packet collides with constant probability p regardless of the number of retransmissions suffered, where

p is referred to as conditional collision probability, meaning that this is the probability of a collision

seen by a packet being transmitted on the channel. Based on this assumption, complex CSMA/CA

protocol with BEB can be approximated as simple -persistent CSMA, where is given by [9], [10]

2(1 2p) 2

= = m1 , (24)

(1 2p)(W + 1) + pW (1 (2p) )

m

W + 1 + pW k=0 (2p)k

where W is the minimum contention window size, m is the maximum backoff stage and p =

(1 )N 1 . The steady-state behavior of IEEE 802.11 DCF with M -MPR capability can also be

6

We use the term complex in the sense that we cannot prove that Eq.(15) for general case has a unique xed solution.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

IEEE TRANSACTIONS ON MOBILE COMPUTING

SUBMITTED TO TMC 23

N 1

approximated by -persistent CSMA [19], where is given by Eq.(24) with p := k=M N 1 Ck

k (1

)N 1k .

Keeping in mind the result given in Eq.(24), for any given minimum contention window W and

maximum backoff stage m in IEEE 802.11 DCF, if we choose satisfying Eq.(24), then the system

throughput for IEEE 802.11 DCF is well approximated by the one for the corresponding -persistent

CSMA protocol. Thus, the system throughput of the optimal -persistent CSMA protocol, where

is the solution satisfying Eq.(15) with the settings of system parameters Ti , Ts and Tc given by

Eq.(13), is a theoretical upper bound for the one of 802.11 DCF.

MAC header 272 bits

PHY overhead 26 s

ACK 112 bits + PHY overhead

Basic rate 6 Mbps

Data rate 54 Mbps

Basic slot time 9 s

SIFS 10 s

TABLE I

S YSTEM PARAMETERS USED IN CARRIER SENSING PROTOCOL (IEEE 802.11 G )

50 50

45 45

The estimated number of active nodes

The estimated number of active nodes

40 40

35 N=30 35

N=30

30 30

25 25

20 20

N=15

N=15

15 15

N=7

10 N=7 10

5 5

0 5 10 15 20 25 30 0 5 10 15 20 25 30

Index of update interval The index of update interval

Fig. 10. The estimated number of active nodes versus update interval

In this section, we develop a simple -persistent CSMA protocol with MPR which achieves the

system throughput close to the maximum value with the conguration of system parameters such as

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

IEEE TRANSACTIONS ON MOBILE COMPUTING

SUBMITTED TO TMC 24

50

45

40

35

5 active nodes

30 are completed

5 active nodes

are injected

25

20

15

10

0 10 20 30 40 50 60 70

The index of update interval

Fig. 11. The performance of the estimation algorithm with L = 200 ms and = 0.6; initially N = 15 and an initial guess

for N is 50, in the middle of simulations 5 new active nodes are injected at the beginning of the 20-th update interval and

then 5 active nodes are terminated at the beginning of the 50-th update interval

Ti , Ts and Tc used in IEEE 802.11. The operation principle of the proposed protocol is to adjust the

transmission probability optimally depending on the number of active nodes in the network. The

proposed protocol requires each node to regularly update the number of active nodes at the end of

every update interval by employing a proper feedback mechanism for each node to get the information7

on the number of nodes involved in each successful transmission slot. The update interval can be set

as a multiple of the interval between two consecutive beacons (for example, a beacon interval is 100

ms as a default value in 802.11 WLAN). Consider the scenario that the number of active nodes in the

network is unknown and varies over time. The proposed algorithm is aloha -persistent CSMA protocol

which tunes the transmission probability aloha depending on the estimate of the number of active

nodes which can be obtained by using the algorithm proposed in section III-B. It is easy to see that

the proposed protocol operates almost optimally if i) the communication time until an active node

completes the transmissions of its whole data packets is relatively long8 (e.g., longer than several

update intervals) and ii) the protocol promptly reacts to the change in the number of active nodes in

the network. Through simulation, we show below that, after the proposed protocol detects the change

in the number of active nodes, it can estimate the number of active nodes promptly.

We next investigate how well the proposed algorithm to estimate the number of active nodes works

under the realistic WLANs environment. The system parameters used in carrier sensing protocols

are presented in Table I and are adopted from IEEE 802.11g standards. In Fig.10, two cases are

7

One of the ways to implement it is to insert the information into the ACK packet.

8

This assumption is usually valid in the scenario of le upload of nodes.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

IEEE TRANSACTIONS ON MOBILE COMPUTING

SUBMITTED TO TMC 25

considered; (a) L = 300 ms and = 0.3, and (b) L = 100 ms and = 0.7. The results are obtained

under the following assumptions: M = 5; all the nodes initially believes there are 50 active nodes

(note that to emphasize the promptness of the proposed algorithm, a relatively large initial value 50

is set.); there are actually 7, 15 and 30 active nodes, respectively, which we want to estimate. Each

gure contains 30 real trajectories to display the results of running 30 independent simulations. In

order to reach a good compromise between accuracy and promptness of the algorithm, as identied

in section III-B, a small = 0.3 is used with a long update interval L = 300 ms and a relatively

large = 0.7 with a short update interval L = 100 ms. In Fig.10, we notice that, starting from the

initial value 50, the estimate for N sharply decreases to the real value. After the transient period,

the algorithm is in quasi-stationary period and the estimate for N oscillates near the real value. It is

worth noting that the error between the estimate and ideal value in the quasi-stationary period is at

most 3 for all cases. Moreover, the length of the transient period is at most 1000 ms (=10 100 ms)

in the case of L = 100 ms and 1200 ms in the case of L = 300 ms, respectively.

To investigate how the proposed algorithm promptly reacts to the change in the number of actual

active nodes, we simulate the proposed algorithm under the following scenario: initially, there are

15 active nodes in the network; all the nodes assume that there might be 50 active nodes initially;

in the middle of simulations, 5 new active nodes are injected into the network at the beginning of

the 20-th update interval and then 5 active nodes are terminated at the beginning of the 50-th update

interval. Fig.11 plots the 30 real trajectories of results of running 30 independent simulations. The

result shows that the proposed algorithm promptly copes with the change in the number of active

nodes. Specically, recalling that L = 200 ms, after the change in the number of actual active nodes,

the transient period (the time which it takes for the proposed algorithm to get accurate estimates close

to the real value) is at most 1000 ms. Besides, the error between the estimates and the real value in

each quasi-stationary period is bounded above by two. We conclude this section with the following

two remarks.

Remark 7: The proposed optimal -persistent CSMA protocol can be also implemented on contention-

window basis. Let W be the contention window size of a node. Then, from the optimal transmission

obtained, the contention window W can be determined by W =

2 1 where

a is the least

integer not less than a.

Remark 8: When M < 4, the approximation of the -persistent CSMA to the slotted-Aloha may

not be accurate as identied in Section IV-B. We therefore need an iterative algorithm to obtain the

optimal transmission probability when M < 4. A heuristic algorithm which may be considered is

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

IEEE TRANSACTIONS ON MOBILE COMPUTING

SUBMITTED TO TMC 26

as follows: As we can estimate E[X 2 ] and E[X] in each update interval, we can also estimate the

probabilities Pi and Ps . Thus, a similar iteration method proposed in Section III-B may be applied

using Eq.(15).

VII. D ISCUSSIONS

200

ideal CSMA (P =8184 bits)

data

180 CSMA DCF (P =8184 bits)

data

SlottedAloha (P =8184 bits)

160 data

ideal CSMA (P =4092 bits)

data

Maximum system throughput

data

SlottedAloha (Pdata=4092 bits)

120

100

80

60

40

20

0

1 2 3 4 5 6 7 8 9 10

The MPRcapability M

Fig. 12. The comparison of the maximum system throughput between slotted-Aloha and -persistent CSMA protocols

when Pdata = 8184 bits and Pdata = 4092 bits, respectively, for N = 20

It is well-known and commonly believed that in the case of single-packet reception scenario

-persistent CSMA protocol outperforms the slotted-Aloha in terms of the achievable maximum

throughput. In this section, we discuss whether this result is still valid in the MPR scenario or not.

To this end, we consider three kinds of protocols; ideal-CSMA, CSMA-DCF and slotted-Aloha as

presented in Fig.12. First, ideal-CSMA is the optimal -persistent CSMA protocol where Ti = ,

Ts = H + Pdata /R + SIF S + ACK and Tc = H + Pdata /R + SIF S , and is the solution

satisfying Eq.(15). Therefore, the ideal-CSMA protocol is able to know immediately whether the

channel is busy or not. Second, CSMA-DCF is the optimal -persistent protocol where Ti = ,

Ts = H + Pdata /R + SIF S + ACK + DIF S and Tc = H + Pdata /R + DIF S , and is the

solution satisfying Eq.(15). The only difference between the ideal-CSMA protocol and the CSMA-

DCF protocol is that the ideal-CSMA protocol does not need DIFS for carrier sensing. Finally,

slotted-Aloha is the slotted-Aloha protocol with the optimal transmission probability aloha where

Ts (= Ti = Tc ) = H + Pdata /R + SIF S + ACK and aloha is the solution satisfying Eq.(5). Fig.12

plots the maximum throughput for the three protocols versus the MPR-capability M when data payload

sizes are 8184 bits and 4092 bits, respectively, for N = 20. We see that the ideal-CSMA protocol

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

IEEE TRANSACTIONS ON MOBILE COMPUTING

SUBMITTED TO TMC 27

gives the best throughput among the three protocols for all cases. On the other hand, it is interesting

to note that when M 3 the CSMA-DCF protocol outperforms the slotted-Aloha protocol, but with

the increase of M the slotted-Aloha protocol gives higher throughput than the CSMA-DCF protocol

regardless of the length of data payload. This result is expected because with the increase of M

the DIFS overhead for carrier sensing more negatively affects the system throughput. In addition, it

is worth noting that with the increase of MPR-capability M the throughput gap between the ideal-

CSMA protocol and the slotted-Aloha becomes less. This nding provides us with a new guideline

to be considered when designing MAC protocols in MPR scenario.

VIII. C ONCLUSIONS

In this paper, we investigate the impact of MPR on the MAC layer behavior where the study

is carried out by considering both slotted-Aloha and -persistent CSMA protocols. Based on the

result, we eventually develop a simple -persistent protocol with MPR which can achieve a system

throughput close to the maximum value. The main results obtained in this paper are summarized as

follows. We rst obtain the optimal transmission probability in slotted-Aloha protocol with MPR. We

show that the optimal transmission probability in slotted-Aloha with MPR is related to the length bias

concept. Second, the result provides us with an useful metric ( EE[X ] 2

involved in each successful transmission slot) which enables each node to estimate the the number

of active nodes in a distributed manner. Third, we obtain the optimal transmission probability in -

persistent CSMA protocol with MPR. We show that under certain conditions the optimal transmission

probability in -persistent CSMA protocol is well approximated by the one in slotted-Aloha protocol.

This nding allows us to directly apply the estimation algorithm developed in slotted-Aloha protocol

to -persistent CSMA protocol. Thus, fourth we develop a -persistent CSMA protocol where the

transmission probability of each node is dynamically and optimally tuned depending on the estimated

number of active nodes. The proposed protocol can achieve a system throughput close to the maximum

value in the case that the completion time of an active node (e.g., le transfer) is relatively long.

As a by-product of our study, we identify that in the MPR scenario -persistent CSMA protocol is

not always superior to slotted-Aloha, contrary to what is known in the single-packet reception model.

This nding is useful in that it provides a new guideline when designing a MAC protocol with MPR.

ACKNOWLEDGEMENT

This research was supported by a grant from the Natural Sciences and Engineering Research Council

of Canada and was supported by National Research Foundation of Korea Grant funded by the Korean

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

IEEE TRANSACTIONS ON MOBILE COMPUTING

SUBMITTED TO TMC 28

R EFERENCES

[1] Task Group 802.11ac timetable, IEEE Std. [Online]. Available: http://grouper.ieee.org/groups/802/11/Reports/tgac

update.htm.

[2] IEEE 802.11 WG, IEEE 802.11-2007, Wireless LAN MAC and PHY Specications, Revised of IEEE 802.11-1999.

IEEE LAN/MAN Standards Committee, June 2007.

[3] IEEE Standard for Information Technology Part 15.4: Wireless Medium Access Control (MAC) and Physical Layer

(PHY) Specications for Low-Rate Wireless Personal Area Networks (LR-WPANs), IEEE Std 802.15.4-2003.

[4] IEEE Std 802.15.1-2005, Wireless Medium Access Control (MAC) and Physical Layer (PHY) Specications for Wireless

Personal Area Networks (WPANs), 14 June 2005. URL http://standards.ieee.org/getieee802/download/802.15.1-2005.pdf.

[5] Sungro Yoon, Injong Rhee, Bang Chul Jung, Babak Daneshrad and Jae H. Kim, Contrabass: Concurrent Transmissions

without Coordination for Ad Hoc Networks, IEEE Infocom 2011.

[6] M. Gerla and L. Kleinrock, Closed loop stability control for S-Aloha satellite communications, in Proc. 5th Data

Commun. Symp., pp2.10-2.19, Sept. 1977.

[7] B. Hajek and T. Van Loon, Decentralized dynamic control of a multiaccess broadcast channel, IEEE Trans. Automat.

Contr., vol.27, pp. 559-569, 1982.

[8] F. Cali, M. Conti and E. Gregori, IEEE 802.11 Protocol: Design and Performance Evaluation of an Adaptive Backoff

Mechanism, IEEE J. Select. Areas Commun., vol.18, no.9, pp.1774-1786, Sept. 2000.

[9] G. Bianchi, Performance analysis of the IEEE 802.11 distributed coordination function, IEEE J. Select. Areas Commun.,

vol.18, no.3, pp. 535-547, Mar. 2000.

[10] A. Kumar, E. Altman, D. Miorandi and M. Goyal,New insights from a xed point analysis of single cell IEEE 802.11

WLANs, IEEE INFOCOM 2005, vol.3, pp.1550-1561, Mar. 2005.

[11] G. Bianchi and I. Tinnirello, Remarks on IEEE 802.11 DCF Performance Analysis, IEEE Comm. Lett., vol. 9, no.

8, pp. 765- 767, Aug. 2005.

[12] Y. Xiao, Performance Analysis of Priority Schemes for IEEE 802.11 and IEEE 802.11e Wireless LANs, IEEE Trans.

Wireless Comm., vol. 4, no. 4, pp. 1506-1515, July 2005.

[13] T. Sakurai and H.L. Vu, MAC Access Delay of IEEE 802.11 DCF, IEEE Trans. Wireless Comm., vol. 6, no. 5, pp.

1702-1710, May 2007.

[14] Douglas S. Chan and Toby Berger, Performance and Cross-Layer Design of CSMA for Wireless Networks with

Multipacket Reception,

[15] S. Ghez, S. Verdu, S. Schwartz, Stability Properties of Slotted ALOHA with Multipacket Reception Capability, IEEE

Trans. Automatic Control, vol. 33, no. 7, pp. 640-649, July 1988.

[16] S. Ghez, S. Verdu, S. Schwartz, Optimal Decentralized Control in the Random Access Multipacket Channel, IEEE

Trans. Automatic Control, vol. 34, no. 11, pp. 1153-1163, Nov. 1989.

[17] D.S. Chan, T. Berger, and L. Tong, On the Stability and Optimal Decentralized Throughput of CSMA with Multipacket

Reception Capability, Proc. Allerton Conf. Comm., Control, and Computing, Sept.-Oct. 2004.

[18] V. Naware, G. Mergen, and L. Tong, Stability and Delay of Finite- User Slotted ALOHA with Multipacket Reception,

IEEE Trans. Information Theory, vol. 51, no. 7, pp. 2636-2656, July 2005.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

IEEE TRANSACTIONS ON MOBILE COMPUTING

SUBMITTED TO TMC 29

[19] Ying Jun (Angela) Zhang, Peng Xuan Zheng and Soung Chang Liew, How Does Multiple-Packet Reception Capability

Scale the Performance of Wireless Local Area Networks?, IEEE Transactions on Mobile Computing, vol.8 no.7, July

2009.

[20] Ying Jun (Angela) Zhang, Soung Chang Liew and Da Rui Chen, Sustainable Throughput of Wireless LANs with

Multipacket Reception Capability under Bounded Delay-Moment Requirements, IEEE TRANSACTIONS ON MOBILE

COMPUTING, VOL. 9, NO. 9, pp.1226-1241, Sep. 2010.

[21] M. H. Mahmood, C. Chang, D. Jung, Z. Mao, H. Lim, H. Lee, Throughput behavior of link adaptive 802.11 DCF

with MUD capable access node, International Journal of Electronics and Communications, vol.64, pp.1031-1041, 2010.

[22] B. J. Kwak, N. O. Song, and L. E. Miller, Performance analysis of exponential backoff, IEEE/ACM Trans.

Networking, vol. 13, no. 2, pp. 343-355, April 2005.

[23] M. Lotnezhad, B. Liang, and E. S. Sousa, Adaptive cluster-based data collection in sensor networks with direct sink

access, IEEE Trans. Mobile Computing, vol. 7, no. 7, pp. 884-897, July 2008.

[24] R.H. Gau and K.-M. Chen, Probability models for the splitting algorithm in wireless access networks with multi-packet

reception and nite nodes, IEEE Trans. Mobile Computing, vol. 7, no. 12, pp. 1519-1535, Dec. 2008.

Yun Han Bae received the B.S., M.S. and Ph.D degrees in Mathematics from Korea University,

Seoul, Korea, in 2003, 2005 and 2009, respectively. He is an assistant professor in the Department of

Mathematics Education, Sangmyung University, Seoul, Korea. His research interests include queueing

theory and its applications to communication systems, and performance analysis of protocols and

wireless networks.

Korea. He received B.S. and M.S. in Mathematics from Kyungpook University, and received Ph.D. in

Mathematics from Ohio State University, 1980. He had worked as a professor at KAIST (1983-1999)

and Korea University (1999-2012), Korea. He received best paper award from IEE in 2000 and he was

awarded Seoul Culture Prize in Science in 2001. He is a fellow of Korea Academy of Science and

Technology. He was an editor of Journal of Communications and Networks, and an associate editor of

Queueing Systems, and is an associate editor of Telecommunication Systems. His areas of interest include queueing theory

and its applications to the communication systems. His recent interests are performance evaluation of IEEE 802.11, 15.4,

16e, power saving scheme, cognitive radio networks and IEEE 802.11p/1609.4 WAVE. He has published about 115 papers

in referred journals. His papers have appeared in Queueing Systems, Journal of Applied Probability, IEEE, IEE, IEICE,

Performance Evaluation, Telecommunication Systems, Computer Networks and others.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

IEEE TRANSACTIONS ON MOBILE COMPUTING

SUBMITTED TO TMC 30

Attahiru S. Alfa is Professor, Department of Electrical and Computer Engineering at the University

of Manitoba, Winnipeg, Manitoba, Canada. He was NSERC Industrial Research Chair of Telecommu-

nications from 2004 to 2012. He carries out research in the areas of queueing and network theories

with applications mostly to telecommunication systems. His current research interests are in the area

of wireless communication networks with recent focus on cognitive radio networks, mobility, Internet

trafc, channel modelling, stochastic models, performance analysis, and teletrafc forecasting models.

He has contributed signicantly in the area of matrix-analytic methods for stochastic models used in telecommunications.

He belongs to the following organizations: APEGM, IEEE, and INFORMS.

- HP Spectre x360 15 User GuideTransféré parenzosingh
- IEEE 802.11Transféré parGine Raj
- IJETTCS-2014-12-25-157Transféré parAnonymous vQrJlEN
- Wireless Site Survey Best PracticesTransféré parjdelacruz010480
- Tl-wr700n v1 User GudieTransféré parcuong955
- questions and answers assessmentTransféré parapi-266658213
- Ccna Wireless LansTransféré parDo Hieu
- e3_7Transféré parscalore
- CCNA 3 - Chapter 7Transféré parMohammad Nour Eddine
- ch7Transféré paralroumy
- CCNA ESwitching Chapter 7 100% 11-02-2011Transféré parVlad Oancea
- DIR-635Transféré parballofcannon
- How to Troubleshoot 802.11n SpeedsTransféré paritfeelssogood
- 79010533-CW-5350-User-ManualTransféré parJeffrey Aquino
- 3987285 CCNA3 v4 0 Exam Ch 7 Basic Wireless Concepts and ConfigurationTransféré parjpataj
- 124019-338870-1-PBTransféré parMustafa Mohammad
- OW11 Paper TemplateTransféré parAlex Atmana
- 06167392Transféré parShahwaiz Afaqui
- arshad 3Transféré parUmair Ali Soomro
- Smart AntennasTransféré parνιηαyαк нєgdє » Nαηdιкαl
- meraki_datasheet_airmarshalTransféré parCarlos Jimenez
- Research PaperTransféré parHusnainLodhi
- Hp v m200 Wifi n Dual Band 1 LanTransféré parRaul Romani Flores
- airOS_UGTransféré parbadbit
- scimakelatex.2041.xcvTransféré parOne TWo
- Ew54aphd ManualTransféré parjmirpay
- Wireless Router ProjectTransféré parAshwini Baskaran
- DS_AP150_0910_v6Transféré parchriss6068
- 802.11Transféré parwityamit
- 3 Steps to Tuning a Cisco WLAN Controller From Default SettingsTransféré parNguyen Cong Thang

- Acer Aspire m1100Transféré parReginald Sosa
- Initial Cell SelectionTransféré parsatishmn
- ejemplo_archimateTransféré parJavier Rosado
- Digital Transmission 4Transféré parSiddiqui Jamil
- Tja 1054Transféré parJoao Silva
- DS-7100HGHI-F1_V3.4.89_20180818Transféré parCarlos E. Salazar
- EPFO Circular on Annexure KTransféré parSunil
- BRKDCT-2218.pdfTransféré parrstoikos
- LANMANWANTransféré parRiajimin
- Inverter Type IDS, IDT-StatronTransféré parrasheed313
- Top Links June 2018Transféré parAndrew Richard Thompson
- Telecom Fraud-Introduction, Types, and Solutions-White Paper.pdfTransféré parSaurabh Jain
- Lab_1_v5_1Transféré parZhen Shen
- x8x_hwmanual_v13_enTransféré parniko67
- An Analysis of Anonymity in the Bitcoin SystemTransféré parjefri
- myftpTransféré parsweetmetalk16
- Cloud ComputingTransféré parSanjib Joshi
- Management EnablementTransféré parWonilLee
- About Peoplesoft Feature PackTransféré parmanedeep
- 96 Catalogue Zaber-A-MCB2 Two-Axis Stepper Motor Controllers WebTransféré pargokonk
- HTC ChaCha User GuideTransféré parEngr. Abdullah
- Mikrotik DUAL WAN Load Balancing Using PCC MethodTransféré pardamian3003
- Programming Approaches in NI LabVIEWTransféré parali_zaries
- Drupalcon Sydney - Show Me the Tests! Writing Automated Tests for DrupalTransféré parAdam Sánchez Ayte
- 14 Chapter 6Transféré parSwati
- Net-Net 4250 SD Hardware Installation GuideTransféré parEric Draven
- Sri_Rudram_Namakam_Devanagari_Large.pdfTransféré parSrilakshmi Maleh
- LTE Site Acceptance Report JED0039 UpdatedTransféré parMoncef Manai
- cut set etcTransféré parpalashkatiyar1234
- 18181-eCall_msd_en_022007Transféré parrahairi