Ad Hoc Networks 9 (2011) 164179

Contents lists available at ScienceDirect

Ad Hoc Networks
journal homepage: www.elsevier.com/locate/adhoc

Performance analysis of slotted ALOHA and network coding for single-relay multi-user wireless networks
Daisuke Umehara a,*, Satoshi Denno a, Masahiro Morikura a, Takatoshi Sugiyama b
a b

Graduate School of Informatics, Kyoto University, Kyoto, Japan NTT Access Network Service Systems Laboratories, Nippon Telegraph and Telephone Corporation, Yokosuka, Japan

a r t i c l e

i n f o

a b s t r a c t
Deployment of wireless relay nodes can enhance system capacity, extend wireless service coverage, and reduce energy consumption in wireless networks. Network coding enables us to mix two or more packets into a single coded packet at relay nodes and improve performances in wireless relay networks. In this paper, we succeed in developing analytical models of the throughput and delay on slotted ALOHA (S-ALOHA) and S-ALOHA with network coding (S-ALOHA/NC) for single-relay multi-user wireless networks with bidirectional data ows. The analytical models involve effects of queue saturation and unsaturation at the relay node. The throughput and delay for each user node can be extracted from the total throughput and delay by using the analytical models. One can formulate various optimization problems on trafc control in order to maximize the throughput, minimize the delay, or achieve fairness of the throughput or the delay. In particular, we clarify that the total throughput is enhanced in the S-ALOHA/NC protocol on condition that the transmission probability at the relay node is set at the value on the boundary between queue saturation and unsaturation. Our analysis provides achievable regions in throughput on two directional data ows at the relay node for both the S-ALOHA and S-ALOHA/NC protocols. As a result, we show that the achievable region in throughput can be enhanced by using network coding and trafc control. 2010 Elsevier B.V. All rights reserved.

Article history: Available online 21 July 2010 Keywords: Single-relay multi-user wireless networks Bidirectional data ows Slotted ALOHA Network coding Throughput Delay Trafc control

1. Introduction Wireless relay nodes are highly effective use of connecting nodes located in rural areas [1] and expanding wireless service coverage [2]. The system capacity can be enhanced by substituting one long-distance low-quality link for multiple short-distance high-quality links. This paper considers single-relay multi-user wireless networks in which two multi-user single-hop networks are linked by a single relay node. Network coding allows intermediate nodes in networks to mix two or more received packets into a coded packet and forward it such that destination nodes can

* Corresponding author. Tel.: +81 75 753 5960; fax: +81 75 753 5349. E-mail addresses: umehara@i.kyoto-u.ac.jp (D. Umehara), denno@i. kyoto-u.ac.jp (S. Denno), morikura@i.kyoto-u.ac.jp (M. Morikura), sugiyama. takatoshi@lab.ntt.co.jp (T. Sugiyama). 1570-8705/$ - see front matter 2010 Elsevier B.V. All rights reserved. doi:10.1016/j.adhoc.2010.07.005

decode the original data, not only to replicate and forward received packets. Since the number of packets owed in networks can be reduced by using network coding, network coding enables us to enhance the system throughput, reduce the energy consumption, and reduce the delay from the departure of a packet for source nodes to the arrival of the packet at destination nodes. Network coding was originally proposed by Ahlswede et al. and they proved that the maximum achievable information ow of any single-source multicast networks can be achieved by using network coding in networks [3]. Ho et al. proposed random linear network coding for multi-source multicast networks and showed that it could approach to the multicast capacity as the length of code increases [4]. Wireless network coding allows us to send multiple packets to destination nodes in a single transmission because the wireless medium has a broadcasting nature.

D. Umehara et al. / Ad Hoc Networks 9 (2011) 164179

165

Katti et al. demonstrated that simple XOR-based network coding can achieve high throughput gains for multiple unicast communications in wireless multi-hop networks, where coding and decoding protocols were implemented in IEEE 802.11 stations as an intermediate layer, COPE, between routing layer and medium access control (MAC) layer [5]. COPE is composed of opportunistic listening and opportunistic coding, and does not require any modications to routing and higher layers. Hasegawa et al. proposed a scheme jointly applying packet aggregation and XOR-based network coding, which is called bidirectional packet aggregation and coding (BiPAC), for bidirectional voice over IP (VoIP) ows in wireless linear multi-hop networks and demonstrated the increase of supportable VoIP sessions through the implementation to IEEE 802.11 networks [6]. Many nodes in wireless multi-hop networks will employ random access protocols on the MAC layer. Therefore, one should analyze the performance of network coding in wireless multi-hop networks with random access protocols and clarify system parameters that are crucial for maximizing the achievable throughput or minimizing the delay. Before the pioneering work of network coding, several work has been presented for the analysis of random access protocols in wireless multi-hop networks. Gitman analyzed the capacity of two-hop wireless slotted ALOHA (S-ALOHA) networks for bidirectional data ows, where the frequency channel between the base station and relay stations did not share with that between relay stations and user stations [7]. Tobagi claried the uplink throughput and delay for two-hop wireless access systems in the cases of S-ALOHA [8] and carrier sense multiple access (CSMA) [9], where the frequency channel from user stations to relay stations shared with that from relay stations to the base station. These results provided some important system parameters to maximize the throughput or minimize the delay. Sagduyu and Ephremides showed the achievable throughput region and throughput optimization in random access mode employing network coding over wireless linear multi-hop networks for several source packet transmission schemes in the case of saturated queues [10]. Le et al. proposed a fundamental coding structure and claried the encoding number of packets in that structure taking into consideration physical wireless environments [11]. They also showed that the maximum number of coding ows was fewer than one expected because many of wireless transmissions interfered with each other. Argyriou proposed the use of opportunistic acknowledgements (OACKs) for overheard packets which are obtained from opportunistic listening [12]. OACKs help us to provide the available correct coding opportunities and enhance the throughput. Many studies of random access protocols have an assumption that any wireless nodes always have packets to be transmitted, i.e. their queues are saturated because an efcient and accurate analytical model [13] to evaluate the saturated throughput was proposed for a complicated CSMA with collision avoidance (CSMA/CA) protocol with binary slotted exponential backoff algorithm, which is employed as an MAC protocol in the IEEE 802.11 standard [14]. It is however important to show analytical models on the unsaturated conditions in order to obtain funda-

mental understandings on behavior of random access protocols and provide reference performances of the throughput and delay according to the offered trafc, like the fundamental studies of S-ALOHA [15] and CSMA [16]. We developed analytical models of the throughput and delay for the S-ALOHA protocol with and without XOR-based network coding for simple single-relay two-user wireless networks with symmetric bidirectional ows [17]. This work was extended to two network models, one-to-one single-relay networks with asymmetric bidirectional ows and one-to-many single-relay networks with the same node trafc, in order to evaluate effects of network asymmetry on throughput and delay [18]. It further was extended to more general single-relay multi-user networks and we claried the achievable region in throughput for the S-ALOHA protocol with and without network coding [19], which is an extension of the achievable throughput region for one-hop wireless networks with S-ALOHA [20,21]. More recently, we recognize that the analysis of S-ALOHA protocol with and without network coding substantially helps us to provide fundamental understandings on CSMA protocols with and without network coding [22]. The main contributions of this paper are summarized as follows:  This paper proposes efcient and accurate analytical models of the throughput and delay in the S-ALOHA protocols with and without network coding for general single-relay multi-user networks. One can evaluate the throughput and delay through the proposed analytical models without any computer simulations regardless of whether the queue at the relay node is unsaturated or not. The proposed analytical models provide reference performances of the throughput and delay, which are compared with those obtained from other random access protocols.  This paper succeeds in dividing the total throughput and delay into two directional throughput and delay, respectively, in S-ALOHA with and without network coding for single-relay multi-user wireless networks. One can formulate various optimization problems on trafc control so as to meet system requirements on two directional throughput and delay. In fact, this paper claries achievable regions in two directional throughput with and without network coding. The results show that network coding can enhance the achievable regions in two directional throughput, especially with symmetric network structure.  This paper provides detailed descriptions of proofs in a number of theorems and lemmas with respect to the throughput and delay with and without network coding for single-relay multi-user wireless networks. These detailed descriptions help us to develop new analytical models for other random access protocols and evaluate effects of wireless link qualities or queue capacity on throughput and delay with and without network coding. The rest of this paper is organized as follows: Section 2 describes the S-ALOHA protocol and the S-ALOHA with network coding (S-ALOHA/NC) protocol for single-relay

166

D. Umehara et al. / Ad Hoc Networks 9 (2011) 164179

Opportunistic listening

Node traffic : g0 TX probablity : gb,0 Relay node R

User node group 1 Number of nodes : m1 Node traffic : g1 TX probability : gb,1

Physical obstacle

User node group 2 Number of nodes : m2 Node traffic : g2 TX probability : gb,2

Fig. 1. Single-relay multi-user wireless networks which have two multi-user single-hop wireless networks via the relay node R.

multi-user wireless networks. Section 3 presents the details of our analysis of the throughput with the S-ALOHA and S-ALOHA/NC protocols. Section 4 provides analytical models of the delay with the S-ALOHA and S-ALOHA/NC protocols. Section 5 claries achievable regions in throughput of the S-ALOHA and S-ALOHA/NC protocols and show that the S-ALOHA/NC protocol helps us to enhance the achievable region in throughput as compared with the S-ALOHA protocol. Section 6 concludes this paper. 2. System description This section describes the network model, the trafc model, the S-ALOHA and S-ALOHA/NC protocols, system design parameters, and evaluation parameters. The S-ALOHA and S-ALOHA/NC protocols described in this paper are similar to those described in [18]. 2.1. Single-relay multi-user wireless networks Let us consider single-relay multi-user wireless networks which involve one relay node R and two groups 1 and 2 composed of user nodes, as shown in Fig. 1. For con venience, v 2 f1; 2g is dened as the complementary element of v 2 {1, 2}. Let us consider only bidirectional data ows via relay node R and there is no closed trafc inside groups 1 and 2. The trafc is dened as the average number of new transmissions and retransmissions in a slot. Let us dene the number of user nodes in group v as mv P 1 for any v 2 {1, 2}. It is assumed that all the user nodes in group v have the same trafc which is denoted as 0 < gv 6 1 for any v 2 {1, 2}. The total trafc in group v is described as Gv = mvgv for any v 2 {1, 2}. Let us dene a packet, the source of which is in group v, as a packet v for any v 2 {1, 2}. All the user nodes have one transceiver with a single antenna. A user node is either transmitting or receiving, but not both simultaneously. It is assumed that any user node and relay node R will be in line-of-sight (LOS) and within transmission range but any pair of user nodes between

groups are in non-line-of-sight (NLOS) due to some physical obstacles and out of their transmission range. This means that any user node in a group is hidden from any user node in the other group.1 Any user node in a group can successfully receive a packet transmitted from relay node R as long as all the nodes in the same group are silent during receiving the packet regardless of any transmission in the other group. The received power from relay node R to a group is sufciently larger than the interference from the other group. Also the received signal power is sufciently larger than the thermal noise power at the receiver. The relay node R is a decode-and-forward repeater with an innite storage capacity.2 The transmission queue at any node is rst-in, rst-out (FIFO) and is called output queue. If a node always has any packets in the output queue, the queue at the node is called saturated and otherwise it is called unsaturated.

2.2. S-ALOHA protocol All the nodes are synchronized with slot time and all data packets begin their transmissions only at the beginning of slots and have a constant payload length. A collision at a user node in a group occurs if two or more packets are simultaneously transmitted from the same group or relay node R, whereas that at relay node R occurs if two or more packets are simultaneously transmitted from both the groups or relay node R. A packet is lost only by the collision but not the bit errors.
1 This is known as the hidden node problem. Fortunately, the S-ALOHA protocol will not suffer from the hidden node problem, unlike the CSMA protocol. In some cases, the S-ALOHA protocol may outperform the CSMA protocol both with and without network coding when user nodes in a group are hidden with those in the other group [22]. 2 The innite storage capacity at relay node R will not be real. We however take an information-theoretical approach because we are interested in ultimate bounds for random access protocols. Therefore, the throughput derived in this paper indicates a theoretical upper bound for the S-ALOHA and S-ALOHA/NC protocols.

D. Umehara et al. / Ad Hoc Networks 9 (2011) 164179

167

(1) S-ALOHA protocol

S D
PHY MAC hdr hdr

PAYLOAD PAYLOAD
IFS

ACK ACK

PHY MAC hdr hdr

(2) S-ALOHA/NC protocol (a) Native packet transmission S D

PHY MAC COPE hdr hdr hdr

PAYLOAD PAYLOAD

ACK ACK

PHY MAC COPE hdr hdr hdr

(b) Coded packet transmission

D1 R D2
PHY MAC COPE hdr hdr hdr

PAYLOAD PAYLOAD PAYLOAD

ACK ACK ACK ACK

PHY MAC COPE hdr hdr hdr

PHY MAC COPE hdr hdr hdr

Fig. 2. ACK packet transmissions for the S-ALOHA and S-ALOHA/NC protocols.

If a packet is successfully received by its receiver, then an acknowledgement (ACK) packet is transmitted to the transmitter immediately. Any ACK packets are not lost by the bit errors as well as data packets. A data packet time, an ACK packet time, and their propagation delays are disjointly included in a slot, as shown in Fig. 2(1). The slot time TSLOT is expressed as

(1) User node in group v

Output queue Arrival ga,v TX prob gb,v MAC Traffic gv Traffic g0

Arrival

T SLOT T PHY T MAC T PLD T ACK 2T IFS 2d;

(2) Relay node R

Output queue TX prob gb,0 MAC Traffic g0 Traffic G

where TPHY is the duration of physical (PHY) header, TMAC is the duration of MAC header, TPLD is the duration of payload, TACK is the duration of ACK packet, TIFS is the duration of interframe space (IFS),3 and d is the maximum propagation delay. When a packet is transmitted at the beginning of a slot, the transmitter can observe whether the packet is successfully received or not at the end of that slot. If a collision occurs, the packet is retransmitted until it is received successfully. If a new packet is arrived at a node and any packets do not stay in the output queue, the node transmits the packet immediately. If there are new arrival and retransmission packets in the output queue of a user node in group v, i.e. the node is backlogged, it transmits the packet to send with a constant probability gb,v in a slot. Note that all packets retransmitted from any user node are transmitted with gb,v in a slot. Relay node R transmits any relay packet with a probability gb,0 in a slot when its output queue is nonempty because it generates no new packets. Fig. 3 illustrates the output queue and MAC of user nodes in group v and relay node R. The node trafc gv at a user node in
3 The distributed coordination function (DCF) in the IEEE 802.11 standard has a short IFS (SIFS) time after receiving a data packet and a DCF IFS (DIFS) time after receiving an ACK packet [14]. The DIFS time is longer than the SIFS time because any station performs a clear channel assessment (CCA).

Arrival
Fig. 3. The output queue and MAC in the S-ALOHA and S-ALOHA/NC protocols.

group v will be determined by the arrival rate ga,v of new packets, the transmission probability gb,v, the collision probability, and the queue capacity. Also the node trafc g0 at relay node R will be determined by the arrival rate k of relay packets, the transmission probability gb,0, the collision probability, and the queue capacity. 2.3. S-ALOHA/NC protocol The relay node R XORs two packets 1 and 2 into a coded packet and broadcast it to all the user nodes in groups 1 and 2, where a non-coded packet is called a native packet. A destination node in group v XORs the coded packet and the appropriate packet transmitted from group v into the desired packet. The relay node R needs to know whether the destination nodes can decode the desired packets before its transmission and decoded them after its transmis-

168

D. Umehara et al. / Ad Hoc Networks 9 (2011) 164179

sion. Any user node needs to overhear packets transmitted from the same group and needs to know which packet is encoded into the coded packet. One should design protocols for network coding to work it well in a distributed matter for wireless multi-hop networks. This paper makes use of a part of COPE [5] and a successive ACK transmission scheme [12,18] to perform the functionality of network coding. Note that the asynchronous ACK proposed in [5] is not utilized in this paper. The user nodes in groups 1 and 2 store the packets transmitted from the same group in a packet pool, which is a queue to store undirected packets heard by using opportunistic listening [5].4 The user nodes keeps a hash table, packet info, that is keyed on packet ID, which is a 32-bit hash of the packets IP source address and IP identication number. Each user node transmits reception reports, which include information on stored packets, in the COPE header of data packets or periodically control packets, as shown in Fig. 2(2)a. The relay node R gains the knowledge of which packets a user node stores by gathering the reception reports. The relay node R maintains two virtual queues 1 and 2. The virtual queue v stores packets transmitted from group v. If the MAC permits relay node R to transmit, relay node R checks if it can generate a coded packet and the destination nodes can decode the desired packets. If a coded packet is transmittable, relay node R broadcasts it to both the destination nodes. The packet IDs of two packets XORed into the coded packet are included in its COPE header. Hence the destination nodes obtain information on the XORing packet stored in the packet pool. If it succeeds in decoding the desired packet, the destination node transmits an ACK packet at the beginning in the designated ACK sub-slot. Two ACK transmissions enable relay node R to determine which packet is appropriate in the next transmission. If a coded packet is not transmittable, for example, one of virtual queues is empty but the other is not, relay node R transmits a native packet without waiting a coding opportunity. Fig. 2 shows the additional overhead

Group 1
1 1 1 1 1

R
5

Group 2

Internet backbone

Fig. 4. A test network to evaluate the reference performances (m1 = 5 and m2 = 1).

formance parameters such as throughput Sv and delay Dv for each user node group v. The throughput Sv is dened as the average number of successful transmissions via relay node  R per slot time from groups v to v . The delay Dv from  groups v to v via relay node R is dened as the number of slots from the start slot in which a packet is transmitted by the source node to the slot in which it is received by the destination node. The delay Dv can be expressed as the sum of packet transmission time, retransmission delay from group v to relay node R, queueing delay of a packet v in re lay node R, and access delay from relay node R to group v . The total throughput S is dened as S1 + S2. Fig. 4 illustrates a test network to evaluate the reference performances in Sections 3 and 4. The number of user nodes in group 1 is m1 = 5 and the number of user nodes in group 2 is m2 = 1. It is assumed that the node in group 2 is connected with Internet backbone so that it works as an access point (AP) in the case of IEEE 802.11. The user nodes in group 1 are assumed to exploit some of bidirectional communication services such as VoIP, TV conference, P2P, etc. In this case, the number of ows between relay node R and the user node in group 2 will be m1 times larger than that between relay node R and a user node in group 2. It is one of important issues to achieve fairness between uplink and downlink data ows rather than fairness among all user nodes as discussed in [23] for single-hop wireless IEEE 802.11 networks.

T COPE T ACK T IFS d;

3. Throughput analysis This section analyzes the throughput in S-ALOHA and SALOHA/NC protocols for any single-relay multi-user wireless networks. In this section, let us assume that each user nodes packet transmission is modeled as a sequence of independent Bernoulli trials, that is, the number of intertransmission slots is geometrically distributed with mean of 1/gv (v 2 {1, 2}). This assumption does not reproduce faithful behavior of the protocols but provides excellent approximations with simulation results. It also can be interpreted that the transmission probability gb,v is set at gv when all user nodes in group v are saturated, i.e. some packets always stay in the output queue. The probability in a slot that only one user node in group v transmits is expressed as

as compared with the case of S-ALOHA protocol. This paper however does not evaluate any additional overhead on COPE and successive ACK transmissions because overhead analysis depends on an implementation matter. We rst attempt to evaluate the throughput and delay for random access protocols on ideal conditions and reveal ultimate bounds of the throughput and delay as an information-theoretic approach. 2.4. System design and performance parameters The transmission probability gb,0 at relay node R and gb,v at user nodes in group v are system design parameters to maximize the throughput or minimize the delay for single-relay multi-user wireless networks with the S-ALOHA and S-ALOHA/NC protocols. This paper claries the relationship between such system design parameters and per4 Opportunistic listening can be performed in the promiscuous mode in the case of IEEE 802.11 CSMA/CA.

cv mv g v 1 g v mv 1 Gv 1 g v mv 1 ;

and the probability in a slot that no user nodes in group transmit is expressed as

D. Umehara et al. / Ad Hoc Networks 9 (2011) 164179

169

gv 1 g v mv :

S l1 u1 l2 u2; Sv lv uv ;

10 11

In throughput analysis, we focus on the trafc control to achieve the fairness between the throughput of user node group v for v 2 {1, 2} as mentioned in Section 2.4. The trafc control to achieve per-group fairness can be extended to the case of per-node fairness as discussed later. 3.1. S-ALOHA protocol Let us dene queue states at relay node R as sequences of packets v n v 1 v 2 v n v i 2 f1; 2g in the output 1 queue and the empty queue state is v 0 0. The queue 1 state in k-th slot is denoted as Vk for any k P 0 and let us assume the initial state V0 = 0. Fig. 5 illustrates the Markov chain of Vk, where self-transitions are not drawn. The steady-state probability of a state v n is denoted as uv n and 1 1 the number of packets v in a state v n is denoted as nv v n . 1 1  The state transition that a packet v is transmitted but v is not when the output queue at relay node R is empty corresponds to k0,v in Fig. 5, which is expressed as

respectively, where v stands for the set of queue states, the head of which is a packet v. The following lemma holds for the steady-state probabilities with respect to queue states at relay node R. Lemma 1. The output queue at relay node R is unsaturated if and only if

g b;0 >

c1 c2 1 c1 c2

12

holds. The steady-state probabilities of queue states are expressed as

u0 1 uv n 1

c1 c2 ; g b;0 1 c1 c2 q22
n v n 1

13 n P 1: 14

q11

n v n 1

u0

1 g b;0

k0;v cv gv : 

5
Proof Lemma 1. The proof is an extension of that of [18, Lemma 1] and then is briefed. The sum of steady-state probabilities u(0), u(1), and u(2) is

Similarly by taking into consideration queue behavior at relay node R, the other state transition probabilities illustrated in Fig. 5 are expressed as

kv 1 g b;0 cv gv ; 

6 7

u0 u1 u2 1:

15

lv g b;0 gv : 

The utilization factor with respect to packets v when the output queue is nonempty at relay node R is dened as

The ratio of the steady-state probabilities u(1) to u(2) is proportional to k0,1/k0,2 = k1/k2 and inversely proportional to l1/l2, i.e.

1 g b;0 cv qv ; lv g b;0 kv

u1 k1 l2 c1 g2 g1 c1 : u2 k2 l1 c2 g1 g2 c2
By solving (15) and (16),

16

and the total utilization factor is dened as q = q1 + q2. The trafc on relay node R is given as

uv

g 0 g b;0 1 u0:

cv 1 u0 c1 c2

17

The total throughput and the throughput of user node group v are expressed as

1 1 0,1

111
1

for any v 2 {1, 2}, is obtained. Let us assume that the steady-state probability u(0) is positive, i.e. the output queue at relay node R is unsaturated. The arrival rate kR in relay node R and the departure rate lR from relay node R are balanced in steady-state. The arrival rate kR at relay node R is expressed as

11
1 2 1 2 2

112 121 122 211 212

! g b;0 u0 kR k1 k2 1 1 g b;0 1 g 0 c1 g2 c2 g1 ; g 0 c1 g2 c2 g1 : c1 c2 18

1
1 0,2 2 1

12

and the departure rate lR for relay node R is expressed as

0
2

21
1 1 2 2

lR l1 u1 l2 u2

19

2
2

The balanced equations between (18) and (19) give u(0) expressed as (13) and

22

221 222

uv

cv
g b;0 1 c1 c2

20

Fig. 5. The Markov chain of the queue states process Vk at relay node R in the S-ALOHA protocol, where self-transitions are not drawn.

Eq. (13) and positive u(0) reveal that (12) is a necessary and sufcient condition to unsaturate the output queue at relay node R.

170

D. Umehara et al. / Ad Hoc Networks 9 (2011) 164179

As a similar way to derive (16), u1v n =u2v n c1 =c2 1 1 is obtained for any v n 2 f1; 2gn . Applying the detailed 1 balance equations in ascending order of queue state length gives

uv

qv u0 ; 1 g b;0

uv n qv 1 uv n : 1 2

21

Constraining the sum of all the probabilities to be 1 provides (14). h Lemma 1 reveals the following theorem with respect to the throughput S and the group throughput Sv. Theorem 1. If the output queue at relay node R is unsaturated, the throughput S and the group throughput Sv are expressed as

c1 g2 c2 g1 ; 1 c1 c2 cv gv  ; Sv 1 c1 c2
g b;0 c1 g2 c2 g1 ; c1 c2 g b;0 cv gv  : Sv c1 c2

22 23

respectively. Otherwise, they are expressed as

24 25

Proof Theorem 1. The proof is an extension of that of [18, Theorem 1] and then is briefed. It is sufcient to show the expression of group throughput. Let us assume that the output queue at relay node R is unsaturated, i.e. (12) holds. Substituting u(v) expressed as (20) into Sv = lvu(v) deduces (23). It is assumed that the output queue is saturated, i.e. u(0) = 0. From (17) and the normalized condition,

Fig. 6 illustrates the throughput versus offered trafc for user node group v, where the trafc Gv is controlled to meet the fairness condition between throughput S1 and S2 for the test network shown in Fig. 4. The curves are obtained from the unsaturated group throughput (23) and the symbols are obtained from Monte Carlo simulations when the packet v is randomly transmitted from a user node in group v with probability gv in a slot for 107 slots, in which gb,0 is set at 1. The throughput S1 and S2 achieve the same maximum value 0.126 at node trafc g1 = 0.087 and g2 = 0.324. For the same value of throughput S1 and S2, the trafc G1 is larger than G2. Since the number of user nodes in group 1 is larger that in group 2, the number of collisions in group 1 is larger than that in group 2. Larger trafc G1 in group 1 is required in order to achieve the fair and high throughput between user node groups 1 and 2. The proposed trafc control scheme requires to estimate node trafc in order to calculate (27). It is however difcult to estimate and control node trafc properly because it will be time-varying with controlling it or changing user node requirements. We hence provide a suboptimal scheme to achieve fair and high throughput. The transmission probability gb,v is set at node trafc gv to provide the maximum throughput Sv and satisfy the per-group fairness condition, which is obtained from (23) and (27). The node trafc gv is approximate to the transmission probability gb,v when gv is heavy and hence the fair and high throughput Sv can be achieved. If one attempts to achieve fairness among user nodes, the node trafc gv should be controlled by using the transmission probability gb,v such that the node throughput s1 = S1/m1 and s2 = S2/m2 become identical. Unlike the analysis and discussion described in [18], this paper describes how to achieve per-group and per-node fairness in throughput. This is one of main contributions in this paper.

uv

cv ; c1 c2

26

3.2. S-ALOHA/NC protocol In contrast to the S-ALOHA protocol, the transmission probability gb,0 at relay node R is a system parameter

is derived. Hence, (25) is obtained. h From Theorem 1, if the output queue at relay node R is unsaturated, the throughput S and the group throughput Sv are independent of the transmission probability gb,0 at relay node R. This is because the departure ratios for relay node R, l1 and l2, are proportional to gb,0, whereas the steady-state probability 1 u(0) that the output queue at relay node R is nonempty, i.e. relay node R is transmittable, is inversely proportional to gb,0. On the other hand, if the output queue at relay node R is saturated, the throughput is proportional to gb,0. As a result, relay node R with S-ALOHA protocol should be operated to meet the inequality (12) for any given trafc gv from the aspect of maximizing the throughput and avoiding the queue overow at relay node R. Let us discuss fairness between bidirectional data ows. The fairness between throughput S1 and S2 can be achieved by using trafc controlling such that S1 is equal to S2. The criterion of trafc controlling is expressed as

0.14

Throughput S v [packets/ T SLOT ]

User node group 1 User node group 2 0.12

0.1

0.08

0.06

0.1

Offered traffic G v [packets/ TSLOT ]

Fig. 6. Throughput versus offered trafc for user node groups 1 and 2 in the S-ALOHA protocol. The trafc G1 and G2 are controlled to meet fairness between the throughput S1 and S2 for the test network shown in Fig. 4.

m1 g 1 g2 : m1 g 1 m2 1 g 1

27

D. Umehara et al. / Ad Hoc Networks 9 (2011) 164179

171

which is crucial for maximizing the achievable throughput of networks as in [1719]. Let us dene queue states at relay node R as pairs (n1, n2) of the numbers of packets in the virtual queues 1 and 2. The queue state in k-th slot is dened as Wk = (W1,k, W2,k) for any k P 0 and let us assume the initial state W0 = (0, 0). Fig. 7 illustrates the Markov chain of Wk, where self-transitions are not drawn. The steady-state probability of a state (n1, n2) is denoted as w(n1, n2) and that of a state nv in the virtual queue v is denoted as wv(nv). The state transition probabilities k0,v, kv, and lv in Fig. 7 are expressed as (5)(7), respectively, like the case of the SALOHA protocol. The state transition from Wk = (n1, n2) to Wk+1 = (n1 1,n2 1) in Fig. 7 means that both of packets in a coded packet are successfully received at both of destination nodes. The corresponding state transition probability l is expressed as

respectively. Let us dene an indicator h as

& h

1; c1 P c2 ; 2; c1 < c2 :

33

The following lemma holds for the steady-state probabilities with respect to queue states at relay node R. Lemma 2. Both the virtual queues at relay node R are unsaturated if and only if

g b;0 >

ch
1 ch

34

holds. Then, for any v 2 {1, 2}, the steady-state probabilities of the numbers of packets in virtual queue v are expressed as

wv 0 1 qv cv w0; 0; wv n qn1 1 qv qv cv w0; 0; v

35 n P 1: 36

l g b;0 g1 g2 :

28

The virtual queue h is saturated but  is not if and only if h

The state transition from a k-th state, Wv,k = nv > 0 and W v ;k nv > 0, to a (k + 1)-th state, Wv,k+1 = nv 1 and   W v ;k1 nv in Fig. 7 means that the packet v in    a coded packet is successfully received but v is not. The corresponding state transition probability l0,v is expressed as

c h
1 c h

< g b;0 6

ch
1 ch

37

holds. Then, the steady-state probabilities of the numbers of packets in virtual queue  are expressed as h

l0;v g b;0 1 gv gv : 

w n h

qn  h
g b;0

g b;0 1 g b;0 c ; h

n P 0:

38

29

Note that lv = l0,v + l holds and this means the state transition probability in the virtual queue v from Wk,v = n > 0 to Wk+1,v = n 1. The utilization factor with respect to packets v for the nonempty queue at relay node R is dened as qv of (8) like the case of the S-ALOHA protocol. The trafc on relay node R is given as

Otherwise, both the virtual queues are saturated. Proof Lemma 2. The proof is an extension of that of [18, Lemma 5] and then is briefed. It is assumed that the steady-state probability w(0, 0) is positive, i.e. both the virtual queues are unsaturated. Fig. 8 illustrates the Markov chain with respect to the number of packets in virtual queue v at relay node R. The state transition probability from states 0 to 1 is expressed as

g 0 g b;0 1 w0; 0:

30

The total throughput and the throughput of user node group v are expressed as

S l1 1 w1 0 l2 1 w2 0; Sv lv 1 wv 0;

31 32

  w0; 0 w0; 0 k0;v k1;v kv 1 ; wv 0 wv 0

39

in steady-state from the Markov chain of queue states shown in Fig. 7. The detailed balance equations

wv 1 qv wv 0 cv w0; 0;
1 1 0,1

40 41

3,0
1 2

wv n 1 qv wv n;

n P 1;

2,0
1 2 0,2 1

1,0
1 0,2 0,2 1

2,1
0,1 2

are obtained. For any v 2 {1, 2}, qv is <1. These inequalities reveal that (34) is a necessary and sufcient condition to unsaturate both the virtual queues at relay node R. Summing all the steady-state probabilities wv(n) and the normalized condition enable us to obtain

0,0
2

1,1
0,1 2 0,2 1

1 X n0

w 0 cv w0; 0 wv n v ; 1 qv

42

0,1
2

1,2
0,1 2

and hence (35) and (36) are derived.

1,v v v v

0,2
2

1
v v

2
v

3
v

0,3
Fig. 7. The Markov chain of the queue states process Wk in the S-ALOHA/ NC protocol, where self-transitions are not drawn.

Fig. 8. The Markov chain with respect to the number of packets in virtual queue v at relay node R in the S-ALOHA/NC protocol, where self-transitions are not drawn.

172

D. Umehara et al. / Ad Hoc Networks 9 (2011) 164179

From Fig. 7, the virtual queue h is saturated but  is not h if and only if qh is not <1 and q is <1. This condition is h equivalent to (37) and gives us w(0, 0) = 0 and k1; k h h which is obtained from (39). The detailed balance equations are obtained as

and

wn1 1; n2 q1 wn1 ; n2 ;
for any (n1, n2) (0, 0). (50) and (51) provide
5

wn1 ; n2 1 q2 wn1 ; n2 ; 51

The detailed balance equations

w n 1 q w n; h h h

43

and then (38) is derived by using the normalized condition. h Lemma 2 reveals the following theorem with respect to the throughput S and the group throughput Sv. Theorem 2. If both the virtual queues at relay node R are unsaturated, the throughput S and the group throughput Sv are expressed as

wn1 ; n2

qn1 qn2 w0; 0 1 2

1 g b;0

52

for any (n1, n2) (0, 0). Summing all the steady-state probabilities w(n1, n2), which are functions of w(0, 0), and the normalized condition enable us to obtain

1 1 X X n1 0 n2

w0; 0 wn1 ; n2 1 g b;0 0

1 1 X X n1 0 n2 0

q q g b;0
53

n1 1

n2 2

S c1 g2 c2 g1 1 g b;0 1 w0; 0 ; Sv cv gv 1 g b;0 1 w0; 0 ; 

44 45

w0; 01 g b;0 1 q1 1 q2 ; 1 g b;0 1 q1 1 q2

and then an approximate expression of w(0, 0) is derived as

respectively. If the virtual queue h is saturated but  is not, h they are expressed as

w0; 0

g b;0 1 g b;0 c1 g b;0 1 g b;0 c2 : g b;0 g b;0 1 c1 c2 1 g b;0 c1 c2

54

S g b;0 g 1 g b;0 c gh ; h h ( g b;0 g ; v h; h Sv h: 1 g b;0 c gh ; v  h

46 47

As a result, approximate expressions of the throughput and the group throughput are derived as

If both the virtual queues are saturated, they are expressed as

S g b;0 g1 g2 ; Sv g b;0 gv : 

48 49

g b;0 c1 g2 c2 g1 ; g b;0 1 c1 c2 1 g b;0 c1 c2 g b;0 cv gv  ; Sv g b;0 1 c1 c2 1 g b;0 c1 c2

55 56

Proof Theorem 2. The proof is an extension of that of [18, Theorem 5] and then is briefed. It is sufcient to show the expression of group throughput. Let us assume that both the virtual queues at relay node R are unsaturated, i.e. (34) holds. Substituting wv(0) expressed as (35) into Sv = lv(1 wv(0)) deduces (45). It is assumed that the virtual queue h is saturated but  is not. Substituting wh(0) = h 0 and w 0 expressed as (38) into Sv = lv(1 wv(0)) h deduces (47). Let us assume that both the virtual queues are saturated. Substituting wv(0) = 0 into Sv = lv(1 wv(0)) for any v 2 {1, 2} deduces (49). h Theorem 2 does not clarify the closed-form expressions of the unsaturated throughput S and group throughput Sv because their expressions (44) and (45) still involve the steady-state probability w(0, 0). It will be difcult to solve the exact closed-form expression of w(0, 0). This argument points to a need for some approximation in order to provide the closed-form expression for w(0, 0). Let us relax the detailed balance equations (40) and (41) in the proof of Lemma 2 for the marginal steady-state probabilities into those for the joint steady-state probabilities. This relaxation is based on [24, p. 182] and gives us the detailed balance equations

when both the virtual queues at relay node R are unsaturated. This approximation with respect to the throughput is in good agreement with the throughput obtained from computer simulations [18]. Especially when the transmission probability gb,0 in (55) and (56) approaches to 1, the throughput S and Sv approach to

c1 g2 c2 g1 ; 1 c1 c2 cv gv  ; lim Sv g b;0 !1 1 c1 c2
g b;0 !1

lim S

57 58

that is, the throughput S and Sv with S-ALOHA protocol expressed by (22) and (23), respectively. When the transmission probability gb,0 equals 1 for the S-ALOHA/NC protocol, relay node R will immediately transmit a packet in the next slot after receiving or colliding that packet, that is, relay node R does not have any coding opportunity. The S-ALOHA/NC protocol with gb,0 = 1 is equivalent to the S-ALOHA protocol. Let us dene Cv = cv/(1 + cv) for v 2 {1, 2}. When the transmission probability gb,0 in (55) and (56) approaches to Ch+ and gb,0 in (46) and (47) approaches to Ch , the throughput S and Sv approach to
5 The multidimensional stationary distribution can be expressed in product form as another type of approximations [24, p. 182]. We attempt to apply an approximation w(0, 0) = w1(0)w2(0) into the detailed balance Eqs. (40) and (41) to derive a closed-form expression of throughput but this approximate throughput cannot be in good agreement with the throughput obtained from computer simulations for some transmission probabilities gb,0 at relay node R.

w1; 0

q1
1 g b;0

w0; 0;

w0; 1

q2
1 g b;0

w0; 0;

50

D. Umehara et al. / Ad Hoc Networks 9 (2011) 164179

173

c1 g2 c2 g1 ; 1 ch c g lim Sv lim Sv v v : g b;0 !Ch g b;0 !Ch 1 ch

g b;0 !Ch

lim S

g b;0 !Ch

lim S

59 60

The limitations of approximate throughput S of (55) and Sv of (56) to the boundaries on the unsaturated condition are in exact agreement with the theoretical values. Let us compare the throughput (59) for the S-ALOHA/ NC protocol, which is derived on the boundary in output queue between saturation and unsaturation, with the throughput (22) for the S-ALOHA protocol. The large coding opportunity enables us to remove the term c from h the denominator of (22). In fact, the unsaturated throughput in the S-ALOHA/NC protocol is larger than or equal to that in the S-ALOHA protocol and is smaller than the boundary of throughput between saturation and unsaturation for any given m1, m2, g1, and g2 from Lemma 5 in Section 5. As a result, the upper bounds of unsaturated and stable total throughput and group throughput are obtained as

0.155 at node trafc g1 = 0.097, g2 = 0.348, and transmission probability gb,0 = 0.258 at relay node R. As in the case of the S-ALOHA protocol, the xed transmission probabilities gb,1 = 0.097, gb,2 = 0.348, and gb,0 = 0.258 will be effective for a suboptimal but practical scheme, which give us fair and high bidirectional data ows even when node trafc is heavy. The throughput of user node group v is substantially improved as compared with the S-ALOHA protocol. The capacity gain is 1.234, which is dened as the ratio of the maximum achievable throughput in the S-ALOHA/NC protocol to that in the S-ALOHA protocol. As in the case of S-ALOHA protocol, larger trafc is required for the user node group 1 in order to achieve fairness between bidirectional data ows. If one attempts to achieve fairness among user nodes, the node trafc gv should be controlled such that the node throughput s1 and s2 become identical. 4. Delay analysis Let us assume an individual transmission probability for each user node group unlike the previous work [18]. This enables us to achieve exible control in transmission probability in the backlog mode. The average number of slots from a collision until the collided node retransmits is 1/ gb,v for the user node group v. The system delay at relay node R consists of the queueing delay in the output queue at relay node R and the access delay from relay node R to user nodes can be derived as DR,v = NR,v/ kR,v by applying Littles Theorem [24], where NR,v is the average number of packets v in the output queue at relay node R and kR,v is the arrival rate of packets v at relay node R if the state probabilities in the output queue have a stationary distribution. The delay consists of the packet transmission time, the retransmission delay from user nodes to relay node R, and the system delay at relay node R. As a result, the delay can be in general expressed as

c1 g2 c2 g1 ; 1 ch c g Sv v v : 1 ch

61 62

Note that the boundary of such upper bounds, (61) and (62) are saturated and unstable. From (62), the criterion to achieve fairness between bidirectional data ows can be expressed as (27) as in the case of the S-ALOHA protocol. Fig. 9 illustrates the throughput versus offered trafc for user node groups 1 and 2 in the S-ALOHA/NC protocol. The curves are obtained from the upper bound of unsaturated group throughput (62) and the symbols are obtained from Monte Carlo simulations when the packet v is randomly transmitted from a user node in group v with probability gv in a slot for 107 slots, in which gb,0 is set at ch/(1 + ch). The trafc Gv is controlled to meet fairness between the throughput S1 and S2 for the test network shown in Fig. 4. The throughput S1 and S2 achieve the same maximum value

Dv 1

1 NT;v 1 DR;v ; g b;v

63

0.16

which is normalized by the slot time TSLOT. Note that this paper focuses on the delay related to the wireless transmission region and hence the queueing delay of new arrival packets at the source user node is not included. 4.1. S-ALOHA protocol The steady-state probabilities of the numbers of packets

Throughput Sv [packets/TSLOT]

User node group 2 S-ALOHA/NC

0.14 0.12 0.1 0.08

v in the output queue at relay node R are denoted as /v(n).

S-ALOHA

The following lemma holds for the steady-state probabilities with respect to the numbers of packets v at relay node R. Lemma 3. If the output queue at relay node R is unsaturated, the steady-state probabilities of the numbers of packets v in the output queue at relay node R are expressed as

User node group 1

0.06 0.1 1

Offered traffic Gv [packets/TSLOT]

Fig. 9. Throughput versus offered trafc for user node groups 1 and 2 in the S-ALOHA/NC protocol. The trafc G1 and G2 are controlled to meet the fairness between the throughput S1 and S2 for the test network shown in Fig. 4.

/v 0 /v n

g b;0 1 cv  u0; g b;0 1 g b;0 cv  d qn  v ; n 1 g b;0 n! dqv 1 qv  

64 n P 1: 65

qn u0 v

174

D. Umehara et al. / Ad Hoc Networks 9 (2011) 164179

Proof Lemma 3. This lemma corresponds to [18, Lemma 2] but the results and proofs are different because the total number of packets is divided into the individual numbers of packets 1 and 2 in this paper. From (13) and (14), the steady-state probability /v(0) can be expressed as

imated that node trafc gv is independent of the transmission probability gb,v. However this approximation will be useful for the rough estimation of delay as discussed in Section 4.2 even when the second term in (66) is dominant. 4.2. S-ALOHA/NC protocol This section provides expressions of the delay in the SALOHA/NC protocol. The following theorem is rst proved. Theorem 4. If both the virtual queues are unsaturated, the delay Dv of user node group v is expressed as

/v 0 u0

1  X qnv u0 g b;0 1 cv   v u0: 1 g b;0 g b;0 1 g b;0 cv  n  1

From (14), the steady-state probabilities /v(nv) (nv P 1) can be expanded as

q u0 /v nv v 1 g b;0
nv

nv

1 X  n1 n2  nv 0 

nv

 qnv  v

n 1 X n dv  qvv ; nv qnv   v 1 g b;0 nv ! dqv  n  0

qv u0

Dv 1

  1 Gv 1 1 : g b;v cv gv 1 g 0 gv g b;0 1 g b;0 cv   70

and (65) is derived. h Lemma 3 reveals the following theorem with respect to the delay Dv. Theorem 3. If the output queue at relay node R is unsaturated, the delay Dv of user node group v is expressed as

If the virtual queue h is saturated but  is not, the delay D of h h user node group  is expressed as h

! 1 G 1 h D 1 1 ; h g b; c gh 1 g b;0 gh g b;0 1 g b;0 c h h h 71

and the delay Dh is not nite. Otherwise, the delay Gv is not nite for any v 2 {1, 2}.

Dv 1

  1 Gv 1 c1 c2 1 g b;v cv gv  1 : 66

gv g b;0 1 g b;0 c1 c2 

Otherwise, the delay Dv is not nite. Proof Theorem 3. Let us assume relay node R is unsaturated, i.e. number of packet transmissions successful reception is expressed that the output queue at (12) holds. The average and retransmissions per as

Proof Theorem 4. Let us assume that the virtual queue v at relay node R is unsaturated.  If the virtual queue v at relay node R is unsaturated, the average number of packet transmissions and retransmissions per successful reception for user node group v is expressed as

NT;v

Gv Gv ; Sv cv gv 1 g 0 

72

NT;v

Gv Gv 1 c1 c2 ; Sv cv gv 

67

by substituting (23) for Sv. Applying Lemma 3 enables us to express the average number of packets v staying in the output queue at relay node R as

by substituting (45) for Sv. The average number of packets staying in the virtual queue v at relay node R is expressed as

NR;v

1 X n0

nwv n

1 g 0 c ; g b;0 1 g b;0 cv

73

NR;v

1 X n0

n/v n

qu0
1 g b;0

cv

1 d X q q2 n dqv n0 1

1 c1 c2 g b;0 1 g b;0 c1 c2

68

which is obtained from Lemma 3. The expression (73) of NR,v enables us to evaluate the stability of virtual queue v at relay node R in the S-ALOHA/NC protocol. The system delay at relay node R is expressed as

The expression (68) of NR,v enables us to evaluate the stability at relay node R in the S-ALOHA protocol for single-relay multi-user wireless networks. From (23) and (68), the system delay at relay node R is derived as

DR;v

NR;v 1 : Sv gv g b;0 1 g b;0 cv 

74

DR;v

NR;v 1 : Sv gv g b;0 1 g b;0 c1 c2 

69

As a result, the delay (66) can be obtained by substituting (67) and (69) into (63). h There is a small difference in value between the delay obtained from (66) and the simulation results when the second term in (66) is dominant, i.e. the transmission probability gb,0 at relay node R is high, because it is approx-

As a result, the delay (70) for user node group is obtained by substituting (72) and (74) into (63).  If the virtual queue v at relay node R is saturated, the delay (63) can be obtained by replacing v and g0 in (70) with  and gb,0, respectively, because g0 is equivalent to gb,0 h on the saturated condition of physical queue at relay node R. h Theorem 4 does not clarify the closed-form expressions of delay because (70) involves the steady-state probability w(0, 0). The approximation of w(0, 0) obtained from (54) will be used in order to evaluate the delay. Let us evaluate

D. Umehara et al. / Ad Hoc Networks 9 (2011) 164179

175

the delay Dv of a user node in each group v, which is minimized by conguring the transmission probability gb,0 at relay node on the condition that the throughput Sv achieves fairness between bidirectional data ows for some given transmission probability gb,v in the test network shown in Fig. 4. The transmission probability at a user node in each group is set at gb,1 = 0.087 and gb,2 = 0.324 in the case of S-ALOHA protocol, whereas gb,1 = 0.097 and gb,2 = 0.348 in the case of S-ALOHA/NC protocol. These transmission probabilities provide maximum throughput with fairness between bidirectional data ows as shown in Fig. 9. User node group 1 has a number of user nodes and hence the transmission probability gb,1 should be lower in order to avoid a number of packet collisions. In contrast, user node group 2 has a relatively large value on transmission probability gb,2 because it has a single user node. As a result, a user node in group 2 obtains lower delay than a user node in group 1. Figs. 10 and 11 illustrate the optimized delay D1 and D2 for given throughput S1 and S2, respectively. The curves are obtained from (66) and (70), and the symbols are obtained from Monte Carlo simulations for 107 slots. In the simulations, a packet is arrived at a user node in group v with a probability ga,v, i.e. the new packet arrival process follows Bernoulli trials with mean of ga,v, and the storage capacity of output queue is innite for any user node. The values nearby the symbols in Figs. 10 and 11 stand for transmission probabilities gb,0 at relay node R for the S-ALOHA/NC protocol to minimize the delay, which are estimated from (70). In contrast, the delay can be minimized by substituting the transmission probability gb,0 = 1 at relay node R into (66) in the case of the S-ALOHA protocol because the unsaturated throughput Sv is independent of gb,0. Note that the S-ALOHA/NC protocol is identical with the S-ALOHA protocol if the transmission probability at relay node R is set at gb,0 = 1. The simulation results with gb,0 = 1 also represent those of the S-ALOHA protocol with gb,1 = 0.097, gb,2 = 0.348, and gb,0 = 1. It can be found that there are differences in value between the delay obtained from (70) and computer simulations especially when the transmission probability gb,0 at relay node is equal to 1. This is

20 18 16

Delay D2 [TSLOT]

14 12 10 8

0.377

1.0 1.0 1.0

6 gb,0 = 1.0 1.0 4

2 0 0.08 0.09

0.597 S-ALOHA (gb,0 = 1) S-ALOHA/NC

0.1

0.11

0.12

0.13

0.14

0.15

0.16

Throughput S2 [packets/TSLOT]
Fig. 11. Delay D2 versus throughput S2 in the S-ALOHA and S-ALOHA/NC protocols. The S-ALOHA protocol has gb,1 = 0.087 and gb,2 = 0.324, whereas the S-ALOHA/NC protocol has gb,1 = 0.097 and gb,2 = 0.348.

because it is approximated that node trafc gv is independent of the transmission probability gb,v in order to derive the expressions of delay. However this approximation will be useful for the rough estimation of the delay as shown in Figs. 10 and 11. Notice that the values on transmission probability gb,0 at relay node R are not identical between Figs. 10 and 11 for given throughput Sv, for example gb,0 = 0.390 and 0.597 when Sv = 0.13 for groups 1 and 2, respectively. This indicates that both the delays D1 and D2 are not minimized simultaneously by a single transmission probability gb,0 at relay node R. 5. Achievable region in throughput This section claries the achievable regions in throughput for the S-ALOHA and S-ALOHA/NC protocols. The boundary of achievable region for any given number of nodes (m1, m2) and the ratio of throughput in group 1 to total throughput r = S1/S gives us the upper bound of total throughput S = S1 + S2 such that r = S1/S. The discussions in Sections 3 and 4 are limited to the case of throughput ratio r = 1/2, i.e. fairness between bidirectional data ows at relay node R, and the results will be extended to arbitrary throughput ratio r = S1/S, which will provide various and exible QoS control for single-relay and multi-user wireless networks. 5.1. S-ALOHA protocol

60 50

0.284

Delay D1 [TSLOT]

40 30 20 gb,0 = 1.0 10 0 0.08

0.326 1.0 1.0 1.0 0.390 0.527 S-ALOHA (gb,0 = 1) S-ALOHA/NC

0.12 0.13 0.14 0.15 0.16

The following lemma is obtained from Theorem 1 for the S-ALOHA protocol, where C = (c1 + c2)/(1 + c1 + c2). Lemma 4. When g1, g2, m1, and m2 are given, (a) the throughput S and Sv are constant with varying gb,0 for any C < gb,0 6 1; (b) they increase monotonically with the increase of gb,0 for any 0 6 gb,0 6 C. From Lemma 4, the unsaturated throughput S of (22) and Sv of (23) are larger than the saturated throughput S of (24) and Sv of (25). As a result, the boundary of achievable throughput region is given by the unsaturated

0.09

0.1

0.11

Throughput S1 [packets/TSLOT]
Fig. 10. Delay D1 versus throughput S1 in the S-ALOHA and S-ALOHA/NC protocols. The S-ALOHA protocol has gb,1 = 0.087 and gb,2 = 0.324, whereas the S-ALOHA/NC protocol has gb,1 = 0.097 and gb,2 = 0.348.

176

D. Umehara et al. / Ad Hoc Networks 9 (2011) 164179

Throughput S2 [packets/TSLOT]

throughput S and Sv. For the unsaturated node throughput sv = Sv/mv, the partial differentials of sv with respect to gv and g v are expressed as 
  v  @sv g mv 2 g mv 1 Gv 1 Gv g mv 1  v   v ; 2 @g v 1 c1 c2

0.5
Number of nodes : (m1, m2)

0.4

75

0.3
(1, 1)

and

v  v v  @sv mv g v g mv 1 g mv 1 1 Gv g mv 1 g mv 1  v   ; 2 @g v  1 c1 c2

76

0.2
(, 1)

 where g v 1 g v is the complementary probability of gv for any v 2 {1, 2}. As a result, the Jacobian is derived as

0.1

(, )

J / 1 G1 G2 G1 c2 G2 c1 :

77

The following theorem will be obtained in order to maximize the total throughput. Theorem 5. For any given number of nodes (m1, m2), the maximum achievable throughput (S1, S2) in the S-ALOHA protocol is obtained if and only if

0.1

0.2

0.3

0.4

0.5

Throughput S1 [packets/TSLOT]
Fig. 12. The achievable throughput regions in single-relay multi-user wireless networks with the S-ALOHA protocol for the numbers of nodes (m1, m2) = (1, 1), (1, 1), and (1, 1).

G1 1 c2 G2 1 c1 1;
holds.

78

The boundary of achievable throughput region in the SALOHA protocol can be obtained by assigning the solution (g1, g2) in (78) to the unsaturated throughput (S1, S2) of (23). The solution (g1, g2) in (78) will be obtained by exploiting root-nding algorithms. If both m1 and m2 are equal to 1, i.e. there is a single user node per-group, (78) will reduce to

the number of nodes increases. This will be because a node will interfere all other nodes in the one-hop wireless networks, whereas the interference of a user node is limited to all other user nodes in the same group and relay node R for single-relay multi-user wireless networks. The boundary of achievable throughput region for the test network with (m1, m2) = (5, 1) shown in Fig. 4 is between those with (m1, m2) = (1, 1) and (1, 1). The node trafc g1 and g2 to satisfy the required throughput ratio r = S1/S for any number of nodes (m1, m2) are obtained by solving a system of equations which are composed of

G2

1 G1 ; 1 2G1 1 G1 2 21 G1 G2 1

79

g2

1 rm1 g 1 ; 1 rm1 g 1 rm2 g 2

84

and the group throughput will be expressed as

and (78). Eq. (84) corresponds to (27) when the throughput ratio r is equal to 1/2.

S1

3G2 1 21 G1 G2 1

S2

80

5.2. S-ALOHA/NC Protocol The following lemma is obtained for the S-ALOHA/NC protocol, where Cv = cv/(1 + cv) for any v 2 {1, 2}. Lemma 5. When g1, g2, m1, and m2 are given, (a) the throughput S and Sv satisfy

for any group trafc G1. If group 1 has an innite number of nodes and group 2 has a single node, (78) will reduce to

G2

1 G1 ; 1 G1 G1 eG1

81

and the group throughput will be expressed as

G2 eG1 2 eG1 1 S1 ; 2 G1 eG1 2 G1 G1 eG1 G1 1 G1 e S2 ; 2 G1 eG1 2 G1 G1 eG1

82 83

c1 g2 c2 g1 c g c2 g1 6S< 1 2 ; 1 c1 c2 1 ch cv gv c g  6 Sv < v v ; 1 c1 c2 1 ch

85 86

for any group trafc G1. Fig. 12 illustrates the achievable throughput regions with the S-ALOHA protocol for the number of nodes (m1, m2) = (1, 1), (1, 1), and (1, 1). The maximum achievable throughput, i.e. capacity for single-relay multi-user scheduling wireless networks is exactly half that for onehop scheduling wireless networks. On the other hand, the capacity for single-relay multi-user S-ALOHA wireless networks is more than half that for one-hop S-ALOHA wireless networks, which is obtained from [20,21], especially when

for any Ch < gb,0 6 1; (b) they increase monotonically with the increase of gb,0 for any 0 6 g b;0 6 C . (c) When gb,0, m1, and h m2 are given, the throughput Sh is constant and S decreases h monotonically with the increase of gh for any gh such that C < g b;0 6 Ch . h Proof Lemma 5. The proofs of (b) and (c) are obtained from Theorem 2 straightforwardly. It is assumed that Ch < gb,0 6 1. Let us denote the steadystate probability w(0, 0) as w0. The steady-state probability  that the virtual queue v is nonempty but v is empty, i.e.

D. Umehara et al. / Ad Hoc Networks 9 (2011) 164179

177

Throughput S2 [packets/TSLOT]

native packets v are transmittable is dened as wn,v and the steady-state probability that both the virtual queues are nonempty, i.e. coded packets are transmittable is dened as wc. The sum of these steady-state probabilities is

0.5

Number of nodes : (m1, m2)

0.4

w0 wn;1 wn;2 wc 1;
and

87

0.3

(1, 1)
0.2

wn;v wc qv cv w0 ;

88

is obtained from (35) for any v 2 {1, 2}. From (87) and (88), the probability in a slot that coded packets are transmitted from relay node R is expressed as

(, 1)
0.1

g c g b;0 wc c1 c2 1 c1 c2 g 0 ;

89
0 0

(, )
0.1 0.2 0.3 0.4 Throughput S1 [packets/TSLOT] 0.5

and the probability in a slot that native packets v are transmitted from relay node R is expressed as

g n;v g b;0 wn;v cv 1 cv g 0 :  

90

Since wc is non-negative and wn,v is positive, the trafc on relay node R satises

Fig. 13. The achievable throughput regions in single-relay multi-user wireless networks with the S-ALOHA/NC protocol for the number of nodes (m1, m2) = (1, 1), (1, 1), and (1, 1).

ch
1 ch

< g0 6

c1 c2 : 1 c1 c2

91

Substituting (91) for g0 in expressions (44) and (45) gives us (85) and (86), respectively. h From Lemma 5, the maximum throughput S and Sv expressed as (61) and (62) can be achieved in the S-ALOHA/ NC protocol by controlling the node trafc g1, g2, and the transmission probability gb,0 = ch/(1 + ch) at relay node R. The Jacobian will be derived as

J / 1 G1 G2 ch G ; h

92

and the following theorem will be obtained in order to maximize the total throughput. Theorem 6. For any given number of nodes (m1, m2), the maximum achievable throughput (S1, S2) in the S-ALOHA/NC protocol is obtained if and only if

G1 G2 ch G 1; h
holds.

93

Capacity gain

The boundary of achievable throughput region in the S-ALOHA/NC protocol can be obtained by assigning the solution (g1, g2) in (93) to the throughput (S1, S2) of (62). Note that the boundary of achievable throughput region is always saturated and unstable unlike the S-ALOHA protocol for any pair of throughput (S1,S2). If both m1 and m2 are equal to 1, i.e. there is a single user node per group, (93) will reduce to

The achievable throughput regions in the S-ALOHA/NC protocol shown in Fig. 13 are compared with those in the S-ALOHA protocol shown in Fig. 12 for numbers of nodes (m1, m2) = (1, 1), (1, 1) and (1, 1). The boundary of achievable throughput region for the test network with (m1, m2) = (5, 1) shown in Fig. 4 is between those with (m1, m2) = (1, 1) and (1, 1). It is shown that the maximum achievable throughput, i.e. capacity for single-relay multi-user wireless networks is enhanced due to network coding. A cusp appears when the throughput S1 is approximate to S2 and the higher capacity gain is obtained, where the capacity gain is dened as the ratio of capacity in the SALOHA/NC protocol to that in the S-ALOHA protocol for any given throughput ratio r = S1/S. In order to discuss the capacity gain in more detail, the capacity gain versus throughput ratio r = S1/S is illustrated in Fig. 14. The less the number of nodes is, the higher the capacity gain is. If the number of nodes has symmetry between both the groups, the highest capacity gain is obtained when the throughput ratio r is exactly equal to 0.5. Otherwise, the highest capacity gain can be seen when the throughput in the group with higher number of nodes is higher that in the other group. It is found that the in1.3

(1, 1)
1.25 1.2 1.15 1.1 1.05

(, 1)

(, )

1 G1 G2 ; 1 G1
and the group throughput will be expressed as

94 (

) 1 ; S1 min ; 1 G1 2 2 ( ) 1 1 ; S2 1 G1 2 min ; 1 G1 2 2 2G2 1 1

for any group trafc G1.

95
1 0 0.1 0.2

Number of nodes : (m1, m2)

0.3 0.4 0.5 0.6 0.7 Throughput ratio r = S1 / S 0.8 0.9 1

96

Fig. 14. Capacity gain versus throughput ratio due to network coding for the number of nodes (m1, m2) = (1, 1), (1, 1), and (1, 1).

178

D. Umehara et al. / Ad Hoc Networks 9 (2011) 164179 [11] J. Le, J. Lui, D.M. Chiu, How many packets can we encode? an analysis of practical wireless network coding, in: Proceedings of IEEE Conference on Computer Communications, Phoenix, AZ, USA, 2008, pp. 10401048. [12] A. Argyriou, Wireless network coding with improved opportunistic listening, IEEE Transactions on Wireless Communications 8 (4) (2009) 20142023. [13] G. Bianchi, Performance analysis of the IEEE 802.11 distributed coordination function, IEEE Journal on Selected Areas in Communications 18 (3) (2000) 535547. [14] IEEE Std 802.11-2007, Part 11. Wireless LAN medium access control (MAC) and physical layer (PHY) specications (2007). [15] L.G. Roberts, ALOHA packet system with and without slots and capture, Computer Communication Review (5) (1975) 2842. [16] L. Kleinrock, F.A. Tobagi, Packet switching in radio channels. Part I. Carrier sense multiple-access modes and their throughput-delay characteristics, IEEE Transactions on Communications COM-23 (12) (1975) 14001416. [17] D. Umehara, T. Hirano, S. Denno, M. Morikura, Throughput analysis of wireless relay slotted ALOHA systems with network coding, in: Proceedings of IEEE Global Communications Conference, New Orleans, LA, USA, 2008. [18] D. Umehara, T. Hirano, S. Denno, M. Morikura, T. Sugiyama, Wireless network coding in slotted ALOHA with two-hop unbalanced trafc, IEEE Journal on Selected Areas in Communications 27 (5) (2009) 647661. [19] D. Umehara, S. Denno, M. Morikura, T. Sugiyama, Achievable region in slotted ALOHA throughput for one-relay two-hop wireless network coding, in: Proceedings of International Conference on Ad Hoc Networks, Niagara Falls, Canada, 2009. [20] N. Abramson, The throughput of packet broadcasting channels, IEEE Transactions on Communications COM-25 (1) (1977) 117128. [21] L. Kleinrock, Queueing Systems, Computer Applications, vol. II, John Wiley and Sons, 1976. [22] D. Umehara, S. Denno, M. Morikura, T. Sugiyama, Throughput analysis of two-hop wireless CSMA network coding, in: Proceedings of IEEE International Conference on Communications, Capetown, South Africa, 2010. [23] B.A.H.S. Abeysekera, T. Matsuda, T. Takine, Dynamic contention window control mechanism to achieve fairness between uplink and downlink ows in IEEE 802.11 wireless LANs, IEEE Transactions on Wireless Communications 7 (9) (2008) 35173525. [24] D. Bertsekas, R. Gallager, Data Networks, Second ed., Prentice Hall, 1992.

crease of collision opportunities, i.e. that of number of nodes causes deterioration of capacity gain. The node trafc g1 and g2 to satisfy the required throughput ratio r = S1/S are obtained by solving a system of equations composed of (84) and (93) for any number of nodes (m1, m2). As a result, the trafc control will be effective by using the transmission probabilities in the backlog mode in order to achieve capacity in the S-ALOHA/NC protocol for any requirement in throughput ratio r = S1/S. 6. Conclusion We have developed analytical models of the throughput and delay for the S-ALOHA and S-ALOHA/NC protocols in single-relay multi-user wireless networks with bidirectional data ows. The analytical models have involved the effects of queue saturation and unsaturation at the relay node. We have claried that the node throughput is enhanced in the S-ALOHA/NC protocol on condition that the transmission probability at the relay node is set at a value on the boundary between queue saturation and unsaturation. We have formulated the control of transmission probability at the relay node to minimize the delay. We have shown that the achievable region in throughput can be enhanced by using network coding and trafc control. In the next step analysis, the effects of protocol overhead, nite storage capacity, and bit errors on the throughput and delay will be investigated more closely. Also, these ndings related to network coding and trafc control will be extended to any carrier sense based random access protocols. Acknowledgement This work was supported in part by Grant-in-Aid for Scientic Research B with No. 21360185 and the MRC Foundation. References
[1] J. Ishmael, S. Bury, D. Pezaros, N. Race, Deploying rural community wireless mesh networks, IEEE Internet Computing 12 (4) (2008) 2229. [2] D. Soldani, S. Dixit, Wireless relays for broadband access, IEEE Communications Magazine 46 (3) (2008) 5866. [3] R. Ahlswede, S. Li, R. Yeung, Network information ow, IEEE Transactions on Information Theory 46 (4) (2000) 12041216. [4] T. Ho, M. Mdard, R. Koetter, D.R. Karger, M. Effros, J. Shi, B. Leong, A random linear network coding approach to multicast, IEEE Transactions on Information Theory 52 (10) (2006) 44134430. [5] S. Katti, H. Rahul, W. Hu, D. Katabi, M. Mdard, J. Crowcroft, XORs in the air: practical wireless network coding, IEEE/ACM Transactions on Networking 16 (3) (2008) 497510. [6] J. Hasegawa, H. Yomo, Y. Kondo, P. Davis, K. Sakakibara, R. Miura, S. Obana, Bidirectional packet aggregation and coding for efcient VoIP transmission in wireless multi-hop networks, IEICE Transactions on Communications E92-B (10) (2009) 30603070. [7] I. Gitman, On the capacity of slotted ALOHA networks and some design problems, IEEE Transactions on Communications COM-23 (3) (1975) 305317. [8] F.A. Tobagi, Analysis of a two-hop centralized packet radio network. Part I. Slotted ALOHA, IEEE Transactions on Communications COM28 (2) (1980) 196207. [9] F.A. Tobagi, Analysis of a two-hop centralized packet radio network. Part II. Carrier sense multiple access, IEEE Transactions on Communications COM-28 (2) (1980) 208216. [10] Y.E. Sagduyu, A. Ephremides, Cross-layer optimization of MAC and network coding in wireless queueing tandem networks, IEEE Transactions on Information Theory 54 (2) (2008) 554571.

Daisuke Umehara received his B.S. degree from Nagoya University in 1994, M.I. degree from Japan Advanced Institute of Science and Technology in 1996, and D.E. degree from Tokyo Institute of Technology in 1999. He is currently an assistant professor at Graduate School of Informatics, Kyoto University. He has been engaged in research work on channel modeling, modulation and coding, and media access control protocol. Dr. Umehara is a member of the IEEE and a senior member of the IEICE.

Satoshi Denno received his M.E. and Ph.D. degrees from Kyoto University, Kyoto, Japan, in 1988 and 2000, respectively. He joined NTT Radio Communications Systems Labs, Yokosuka, Japan, in 1988. In 1997, he was seconded to ATR Adaptive Communications Research Laboratories, Kyoto, Japan. From 2000 to 2002, he worked for NTT DoCoMo, Yokosuka, Japan. In 2002, he moved to DoCoMo Communications Laboratories Europe GmbH, Germany. Since 2004, he is an associate professor at Kyoto University. From the beginning of his research career, he has been engaged in the research and development of digital mobile radio communications. In particular, he has considerable interest in channel equalization, adaptive array, STBC, spatial multiplexing, and multimode reception. Dr. Denno is a senior member of the IEICE. He received an excellent paper award from IEICE in 1995.

D. Umehara et al. / Ad Hoc Networks 9 (2011) 164179

179

Masahiro Morikura received his B.E., M.E., and Ph.D. degrees in electronics engineering from Kyoto University, Kyoto, Japan in 1979, 1981, and 1991, respectively. He joined NTT in 1981, where he was engaged in the research and development of TDMA equipment for satellite communications. From 1988 to 1989, he was with the communications Research Centre, Canada, as a guest scientist. From 1997 to 2002, he was active in the standardization of the IEEE802.11a based wireless LAN. He received the Paper Award and the Achievement Award from IEICE in 2000 and 2006, respectively. He also received the Education, Culture, Sports, Science and Technology Minister Award in 2007 and Maejima Award in 2008. Dr. Morikura is now a professor in the Graduate School of Informatics, Kyoto University. He is a member of the IEEE.

Takatoshi Sugiyama received the B.E., M.E. and Ph.D. degrees from Keio University, Japan in 1987, 1989 and 1998, respectively. Since joining NTT in 1989, he had been engaged in the research of interference compensation, CDMA, MIMO-OFDM technologies for satellite, wireless LAN systems. From 2004 to 2007, He was in Wireless Laboratories of NTT DoCoMo, Inc., where he worked for the research of plug-and-play base stations and wireless mesh networks. He is currently a senior research engineer, supervisor, Group Leader in NTT Access Network Service Systems Laboratories responsible for the research and development of intelligent interference compensation technologies and radio propagation modeling for future wireless communication systems. He received the Young Engineers Award from the IEICE of Japan in 1996. He is a member of the IEEE.