You are on page 1of 5

The 18th Annual IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC'07)

COOPERATION ON DEMAND PROTOCOLS FOR WIRELESS NETWORKS


Jesu's Gomez, Jesu's Alonso-Zairate, Christos Verikoukis Centre Tecnologic de Telecomunicacions de Catalunya Castelldefells, Spain
ABSTRACT
Ana I. Perez-Neira, Luis Alonso Technical University of Catalonia (UPC) Barcelona, Spain

borhood. Orderly in time, those neighbors that overheard the original signal and are likely to have good channel conditions We consider a cooperative network where the communica- with the destination, retransmit the original message so that tion between a single source-destination pair is assisted by the the destination node can properly combine them to decode the transmission of relays. In the considered cooperative model, original message. The diversity gains of this particular strategy, once a node requires cooperation, it broadcasts a claim for co- referred to as Incremental Relaying, have been analyzed in [5]. operation message. Under high mobility conditions or in some Upon cooperation request, all potential relays must decide special scenarios this is the natural procedure to discover the potential relays. However, when there is a large number of whether to transmit or not based only on local information. nodes that can potentially serve as a relay, a Multiple Relay Since they must contend for the channel, efficient Medium ReAccess Control (MRAC) problem turns up and the relay trans- lay Access Control (MRAC) protocols are necessary. The relay missions must be organized in an effective manner. We intro- contention problem could disappear, or at least be minimized, if duce this problematic and propose three different strategies to the relays were identified exclusively. This idea was proposed manage this severe competitive channel. In these strategies ei- in [7] and [8]. In these works all nodes in the network maintain ther the source, the destination or both trigger the cooperative a table that contains the possible relays that can assist with their process. The methodology employed makes the results and retransmissions to a given destination, the Access Point (AP). conclusions valid for a wide range of cooperative scenarios, These tables are updated at each node by listening to the ongoing transmissions both from the rest of nodes to the AP and and hence, useful for a wide range of applications. from the AP to other nodes. As pointed out by the authors, this strategy is required to maintain up-to-date the relay tables unI. INTRODUCTION der high mobility conditions due to the fact that channel state Cooperative communications in wireless networks are mainly information may have changed from one transmission to anmotivated by the broadcast nature of the wireless medium. The other, or even the requested relay may have moved or disapsource input is broadcasted to all nodes in its coverage area, peared. The problem becomes harder when considering more turning them into potential helpers to improve the transmission than one destination, i.e, in the case of ad hoc or sensor netefficiency. As a result, multiple independently faded signals works, where every node must maintain a table of potential can be appropriately combined at the receiver improving the relays for every single destination. reliability and spectral efficiency of the communication. The In order to overcome these storage requirements, we analyze pioneer work of Cover and El Gamal [1] analyzed the achievthree different strategies whose goal is to reduce the multiple able rates for the one-relay network consisting of a source, a contention problem posed by the on demand cooperative scedestination and a single relay node. More recently, cooperanario. The considered scenario employs an Automatic Repeat tive gains have been analyzed for different scenarios in several scheme by which whenever a request for coRequest (ARQ) works [2-6] in terms of diversity and energy savings. is operation sent, only the set of relays accomplishing with a Different cooperative scenarios can be defined depending on certain will try to help out by retransmitting a requirements the specific forwarding strategy implemented at the relays. The of the copy original message. ARQ has been widely adopted design of each of these scenarios poses several challenges that in wireless such as third generation (3G) netexisting systems enclose the whole protocol stack, including for instance, the works or wireless local area networks (WLANs). Since retransphysical transmission-reception schemes, the channel access missions are activated when the receiver fails to decode a only rules, the routing or the application layer. In this work, we in improving systhe is efficient packet, ARQ technique very consider a general cooperative set-up that includes most of the tem and with channel Our analythroughput combating fading. possible cooperative scenarios, hereafter referred to as a coopsis offers novel ideas on how to the that design requirements erative on demand scenarios. Communications are carried out in must fulfil order to potential relays cooperate. over time-division channels and whenever a node requires coThe remainder of the paper is organized as follows. Section operation, it broadcasts a claim for cooperation message. The most intuitive scenario is composed by a source, a destination II is devoted to introduce the cooperative scenario, overview and a set of potential relays. Whenever the destination is un- the relaying MAC strategy (MRAC) and particularize the conable to decode an incoming signal from the source node, it tention resolution time for the target scenario. Sections III, IV, broadcasts a claim for cooperation message to its local neigh- V are each dedicated to the design of one particular MRAC strategy: Flat On-Demand Cooperation, On-Demand CondiThis work was partially supported by the Catalan Government under grant tioned Cooperation and On-Demand Multi-Level Cooperation. SGR2005-00996; by the Spanish Government under projects TEC2005-08122Finally, in Section VI some concluding remarks are given. C03, TEC 2005-07326-C02-01 and TEC 2006-10459/TCM.

1-4244-1144-0/07/$25.00O2007

IEEE

Authorized licensed use limited to: University of Pennsylvania. Downloaded on July 22,2010 at 07:46:17 UTC from IEEE Xplore. Restrictions apply.

The 18th Annual IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC'07)
sou.be
L

DATA|

aIl

WaitThCmem Ibb~rfq
S L . l CS CS _

:Relay 2

4
miFrimWs'dwk

corier seeimi!

LDAXTA1

MAX:f W19(CK,X CMCMa)tCWJ

Figure 1: RCSMA MRAC protocol


II.

COOPERATIVE SCENARIO

We consider an infrastructure-less wireless network such as a wireless ad-hoc or sensor network. Upon selection of a given source-destination pair by a high layer protocol, the scenario is reduced to that of a single source, a single destination and a set of relays. The interference from other systems is not considered and a collision can only occur if more than one relay transmission is received simultaneously in time at the receiver. The optimization of this cooperative transmission scheme involves solving many different problems. However, we focus our analysis on the design of the MRAC protocol that schedules the relays transmissions. We build on the MRAC protocol presented in [9], referred to as Relay Carrier Sensing Multiple Access (RCSMA), and introduce some degree of intelligence by designing the set of conditions that relays need to satisfy in order to cooperate. Although we particularize the design of the RCSMA protocol, the analysis herein presented can be generalized to any MRAC on demand protocol. The objective of these conditions is twofold: firstly, to select the most adequate amount of relays and secondly to help out in self-organizing the relays access to the channel in order to reduce the total contention time.

packet become potential relays and compete for the channel in order to retransmit a copy of the original message transmitted by the source node. Upon correct decoding, the destination node broadcast an acknowledgement packet (ACK) both, to announce the end of the cooperation process and the reception of the message to the source node. B. Contention Resolution Time Within the scope of MAC protocols, there exist in the literature [10] [11] different strategies to derive the contentionresolution time (TCRT) among different sources contenting for the channel. However, when applied to the considered relay scenario the target figure of merit is essentially different. From the MAC point of view the TCRT is defined as the average contention time required to transmit a single data packet but, in the MRAC context it is defined as the average contention time needed to resolve the m required relay transmissions. For the RCSMA protocol, given a number M of contenting stations and assuming that c collisions occur before the m'th relay transmits, the total delay resultant from the cooperation process is given by

TCRT = E[CWm-(M, c, m) + cTcOL + mTCOLAvl, (1)


where E[] denotes the expectation operator. In equation (1) we identify three different contributions to TCRT: 1) Idle time due to back-off CWm(M, c, m): Before attempting to transmit, each relay i initializes a counter by randomly selecting a discrete back-off window CWi within the interval [0, CW]. This back-off counter is decremented by one unit whenever the channel is sensed idle, and frozen otherwise. When the counter expires (gets to zero), a transmission attempt is done. In the case of collision, the CW is doubled up and a new back-off is started. It is worth mentioning that when m relay transmissions are requested, the total idle time due to back-off is equal to the m'th lowest CWm. We assume that this contention time includes the possible contention window resizing procedure whenever a collision occurs. 2) Time wasted due to collisions TCOL: When a collision occurs the channel is busy for the duration of a whole data packet. This time is considered wasted. Note that the time employed to retransmit a data packet, when successful, is not considered wasted and thus not included into TCRT. 3) Time wasted due to collision avoidance TCOLAV: When using the collision avoidance mechanism of the RCSMA protocol (RTS/CTS) for the relay contention phase, the overall spectral efficiency decreases with m due to the increased overhead. We denote by TCOLAV the period of time used in each

A. MRAC Overview In RCSMA, we consider that a node, hereafter referred to as the petitioner, which is typically the destination, but maybe also the source or both cooperatively, broadcasts a Claim for Cooperation (CFC) packet whenever it requires the nodes in its neighborhood to cooperate. Upon reception of the CFC, all potential relays will contend to get access to the channel in order to retransmit a copy of the original message. In Fig. 1 the operation of the RCSMA is represented particularized for the case when the petitioner of cooperation is the destination node and a collision avoidance mechanism based on Request to Send/Clear to Send (RTS/CTS) handshake between sender and receiver is applied. The RCSMA is backward compatible with the Distributed Coordination Function (DCF) of the legacy 802.11 MAC protocol, which is based on CSMA with optional RTS/CTS handshake. Any petitioner must listen to the channel for a Short Inter Frame Space (SIFS) time before transmitting the CFC packet. It is worth mentioning that the operation of the MAC protocol defined in the IEEE 802.1 1, sets a longer IFS (named DIFS) before getting access to the channel for the transmission of a data packet. Therefore, cooperation processes in RCSMA get higher priority than regular transmissions. All those nodes that are able to decode the CFC

Authorized licensed use limited to: University of Pennsylvania. Downloaded on July 22,2010 at 07:46:17 UTC from IEEE Xplore. Restrictions apply.

The 18th Annual IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC'07)
CW=32
60
c)
",

70

;Mm

50 0

m=7 S

F'30 102-0
620

60

>
60

10

20

30

40

low values of M lead to a negligible probability of collision, due to the randomness of the back-off windows values (CW) the bigger the value of M, the lower the m'th back-off window, and therefore, the shorter the contention time. On the other hand, for large values of M the occurrence of collisions is the dominant effect and increasing M only yields higher contention resolution delays. From Fig. 2 it is reasonable to think that by adjusting the size of CW and by controlling the average number of competitors M it would be possible to improve the overall performance. These are the foundations of the strategies hereafter proposed.
IV. ON-DEMAND CONDITIONED COOPERATION

O_
0

10

20

Average Contending Stations (M)

30

40

50

60

70

80

90

100

Figure 2: TCRT, as a function of the average number of nodes M, for different number of required retransmissions m and CW 32 or 64.

RTS/CTS handshake. Note that according to this model, if the collision avoidance mechanism is applied, it holds that TCOL = 0. Since both TCOL and TCOLAV are constant values, they do not provide information about the efficiency of the MRAC protocol. In fact, these values can be properly adjusted when designing the protocol overhead. We will focus the analysis on the contribution of the idle time due to back-off. Next sections are devoted to present three alternative mechanisms to reduce the total contention resolution time. We start off by presenting the problem statement assuming that all nodes that receive the CFC are requested to cooperate (flat ondemand). We then present a conditional on demand mechanism by which only nodes accomplishing with a certain condition will be involved in the cooperation process, and finally, a multilevel on demand mechanism by which potential relays are grouped based on certain criteria.
III.

FLAT ON-DEMAND COOPERATION

In flat on-demand cooperation we consider the CFC packet as a simple bacon that is broadcasted by the petitioner in order to ask the nodes in its transmission range for cooperation. Before designing more sophisticated strategies, this simple scenario is useful to analyze the statistical behavior of the idle time due to back-off. The average idle time due to back-off will be denoted by E[CWm(M, c, m)] = CWm(M, m), where the dependency with the parameter c can be dropped as long as the CWm (M, C, m) is averaged over the collision event. Note that the value of CWm (M, m) depends on both the average number of relays that decide to cooperate M (also referred to as active relays) and will have to contend for the channel, and the number of required retransmissions m. As it will be discussed A. The Average Number of Active Relays M later in Section IV, the distribution of M can be assumed to be Consider, in the more general case, a scenario where two Poisson. In Fig. 2 it is plotted the computed CWm (M, m), in thresholds are included in the CFC message determining lower number of time slots, as a function of M and m for two dif- and upper values for the required SNR of the received CFC ferent initial back-off contention windows CW (32 and 64). packet in order to take part of the cooperative process. Note On the one hand, when a low average number M of relays are that these values can be mapped into two distances from the claimed for cooperation, the more contenting nodes the lower petitioner. Consider the scenario represented in Fig. 3. The the average idle time due to back-off. Despite of the fact that node in the center of the two concentric circles represents the

In on-demand conditioned cooperation, the CFC is not only a simple signalling beacon that triggers the cooperation process, but it could also include additional extra information. Any receiving node (potential relay) could use this information to validate a set of requirements and decide whether to cooperate or not. The proper design of these conditions may allow the protocol to have control on the quantity and characteristics of the relays, and therefore, optimize the performance of the network. The design of the cooperative requirements depends on the specific figure of merit, such as energy efficiency, transmission rate or, as in our case, the TCRT. Without loss of generality in the methodology, the presented results are based on conditions for the estimated distance between the petitioner and the potential relays. This information can be derived from the received Signal to Noise Ratio (SNR) by establishing a cross-layer dialogue between the layers PHY-MAC. As an example, the relays could decide whether to cooperate or not based on the signal strength of the received CFC packet. In order to select the best set of relays with whom the communication may be more efficient and reliable, the decision threshold could be included to the CFC packet. We assume a uniformly distributed user density scenario wherein the channel conditions remain constant from the beginning of the cooperation process until all the relays have finished their transmissions. Next subsection is devoted to show how to design the SNR requirements in order to control the average number of competing relays M. It is worth mentioning that for a given value of M, there exist a residual probability of not activating enough nodes m. Since the cost of such an event depends on the scenario (either repeat the cooperation request or request the retransmission of the packet to the original source node), in subsection B we compute the probability of insufficient responses PIR, and justify that a low value can be warranted for the optimal M.

Authorized licensed use limited to: University of Pennsylvania. Downloaded on July 22,2010 at 07:46:17 UTC from IEEE Xplore. Restrictions apply.

The 18th Annual IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC'07)

petitioner. The circle of radius R represents the ideal transmission range of the CFC broadcast (only considering path loss), being N the total number of contained nodes. The probability for a given node to be at the distance x from the sender of the CFC packet, with d1 < x < d2 is given by

d,
x

PA (x) =p(di <x<d2)

wiR2'

As

(2)
Figure 3:
Distance dependent cooperative scenario.

with As = w (d2- d2) [M2] being the area defined between the circles of radii d1 and d2 respectively. The probability of having n nodes within the given region is given by

f-IR= S;

PIR=1 O' ~~~~~~~~~~~~~~~~~~~~PIR=1

0-'

p(n)=

PA

PA)

N-n

(3)
2
11

The dependency of this expression with N and R makes it difficult to be estimated at run-time and distributively at each node. A possible solution is to use the population node density (p), which can be either previously known by the users or can be easily estimated by sensing the activity of the channel.

Desired requests (m)

0 2 1D 1e

16

AN

wi=?- Nodes/in

(4)

Figure 4: Probability of Insufficient Responses to the CFC message.

Assuming a big transmission radio R in comparison with the number of active users, Eq. (3) can be approximated by a Poisson probability density function with mean and variance M = NPA = pWR2PA = pASD, and therefore, the probability of having n nodes within the area of interest is given by
p (n)

M'e-M

threshold requirements). In that case, the number of active relays (M) is below to the required one (im). This probability will be externally fixed to satisfy certain quality of service criteria. However, here we show how this event can be maintained to occur with low probability. Considering Eq. (5), the probability of this event is given by

PIR (m, M)

PIR (m,T9,m)

= c

m-1 (Tnn nmZ (1 nI

The remaining question is how this information should be used to set the decision required thresholds. For simplicity, we consider the scenario with d1 = U. The relaying region is only defined by one threshold, the distance d2 or equivalently the minimum received SNRTh. If we want to ensure that M nodes, in average, will respond to the CFC packet, the decision threshold should be chosen as

SNRTh

13

:13PT

p7T

-T

where Q refers to the average channel gain normalized to the constant known noise power, PT is the petitioner transmitted power (known in advance at the relays), and M is given by, M
=

pASD

p7d .

(7)

Note that in the case that the values of p, M and PT could be known in advance by all the nodes in the network, the thresholds could be independently computed by each potential relay. Therefore, additional overhead in the CFC packet is avoided.

This probability depends on the ratio between the average number of nodes claimed and the number of desired responses I1 = M. In Fig. 4, the lowest value of 9 achieving a certain PIR is plotted as a function of the desired number of retransmissions m. When m is low (e.g 1) the cooperation request must, in average, activate a large number of nodes (e.g., 10), in order to achieve a low PIR. However, as m increases, the ratio rq tends to be stabilized around 2 or 3. This means that if the petitioner needs m retransmissions to properly decode the original packet, it should ask for at least 2m retransmissions to ensure a negligible probability of insufficient responses. Concerning the joint design of the optimal contention window and M, in Fig. 5 it is shown the minimum value of TCTR and the optimal back-off CW achieving this value, as a function of M for different values of m. Note that this figure shows that, for the optimal window size CW, the best strategy consists in asking for cooperation to all the nodes within the scenario (M --> oc). This result also ensures that the PIR requirements will be satisfied (M > 2m).
V.

n=O

(8)

ON-DEMAND MULTI-LEVEL COOPERATION

B. Probability of insufficient responses As discussed earlier, there is a non-negligible probability that not enough relays respond to the CFC message (satisfy the

The last presented cooperative strategy is based on the Carrier Sensing Prioritization [12]. In this case, the contention windows of the potential relays are chosen as a function of a set of

Authorized licensed use limited to: University of Pennsylvania. Downloaded on July 22,2010 at 07:46:17 UTC from IEEE Xplore. Restrictions apply.

The 18th Annual IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC'07)

35
I

80
.

70 60 50

a:25 20

m5

3: 0

"E

Z5 40

!
,@ 15 5 0
. --l --l
,

30

:1.-T, 20
o 0 0

10

20

30

40

50

60

70

80

90

100

Average Nodes Claimed for Cooperation M


100 1

0.5

Average Nodes Claimed per Category M

1.5

2.5

35 30

C)
B

80
0
_

.
I

25
20

+m=5
m=3

-o

.m
.

1,
Q-

40
20_

Z5 15
2 I 0 5

ol

10

20

Average Nodes Claimed for Cooperation M

30

40

50

60

70

80

90

100

OL. 0

0.5

Average Nodes Claimed per Category M

1.5

2.5

Figure 5: Average Cooperation Time and Optimal Contention Window

Figure 6: On-Demand Multi-Level Cooperation TCRT and the average


number of collisions

conditions that define different categories of relays. For a given category k the contention window is pre-determined; all relays which lay within the same category will collide. In the distance dependent scenario, the CWi of a given node i is set according to CWi = f(d), being d the figure of merit (distance), and f a quantified function with k levels. With this strategy, the best relays can be promoted to transmit first by properly defining the number of levels k and their respective CWk. As aforementioned, this distance can be derived from the estimation of the SNR received from the CFC, i.e., CWi = f (SNR). To simplify the analysis, the function f is designed to include the same average number of relays per level (distance to the petitioner). In Fig. 6 the average level in which the last relay transmits is plotted. This is equivalent to the parameters Wm or TCRT plotted in previous sections as a function of the average number of nodes claimed per category. Note that the higher the number of relays per level, the greater the probability of collision. However, the lower the average number of relays per level, the greater the number of levels and the greater the maximum CW and, hence, the average TCRT. An optimal choice of the average number of nodes per level will be around M = 1. Note that, also in terms of average number of collisions, the performance of this strategy is quite similar to that of the conditional strategy in previous section.

of cooperative scenarios and hence, for a wide range of applications. REFERENCES


[1] T. M. Cover and A. E. Gamal, "Capacity theorems for the relay channel," IEEE Trans. Inform. Theory, vol. 25, no. 5, p. 572, Sept. 1979. [2] A. Sendonaris, E. Erkip, and B. Aazhang, "Increasing uplink capacity via user cooperation diversity," in in Proc IEEE Int. Symp. Information Theory (ISIT'98), Cambridge, MA, Aug. 1998, p. 156. [3] , "User cooperation diversity-part I: System description,," IEEE Trans. Commun., vol. 51, no. 11, pp. 1927-1938, Nov. 2003. [4] , "User cooperation diversity-part II: Implementation aspects and performance analysis," IEEE Trans. Commun., vol. 51, no. 11, pp. 19391948, Nov. 2003. [5] J. N. Laneman, D. N. C. Tse, and G. W. Wornell, "Cooperative diversity in wireless networks: Efficient protocols and outage behavior," IEEE Trans. Inform. Theory, vol. 50, no. 12, Dec. 2004. [6] J. N. Laneman and G. W. Womell, "Distributed space-time-coded protocols for exploiting cooperative diversity in wireless networks," IEEE Trans. Inform. Theory, vol. 49, no. 10, pp. 2415-25, 2003. [7] P. Liu, Z. Tao, and S. S. Panwar, "A cooperative MAC protocol for wireless local area networks," in in Proceedings of the 2005 IEEE International Conference on Communications (ICC), Seoul, Korea, May 2005. [8] T. Korakis, S. Narayanan, A. Bagri, and S. Panwar, "Implementing a cooperative MAC protocol for wireless LANs," in in Proceedings of the 2006 IEEE International Conference on Communications (ICC), 2006. [9] J. Alonso, J. Gomez, C. Verikoukis, L. Alonso, and A. I. Perez-Neira, "Performance evaluation of a cooperative scheme for wireless networks," in Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC), Helsinki (Finland), Sept. 2006. [10] G. Bianchi, "Performance analysis of the IEEE 802.11 distributed coordination function," IEEE J Select. Areas Commun., vol. 18, no. 3, Mar. [11] M. Heusse, F. Rousseau, G. Berger-Sabbatel, and A. Duda, "Performance anomaly of 802.1 lb," in In Proceedings of IEEE INFOCOM 2003, San Francisco, USA, Mar. 2003. [12] Q. Zhao and L. Tong, "Opportunistic carrier sensing for energy-efficient information retrieval in sensor networks," EURASIP J Wirel. Commun. Netw., vol. 2005, no. 2, pp. 231-241, 2005.

VI.

CONCLUSION

In this paper three On-Demand Cooperation strategies have been analyzed from the MRAC point of view; flat cooperation, conditional cooperation and multilevel cooperation. Whenever any node asks for cooperation, all the potential relays are triggered at once. This leads to a severe challenge from the MAC point view since all nodes will try to get access to the channel at the same time setting the network in saturation conditions. We show that for this kind of cooperative scenarios, it is necessary to develop new and efficient MAC protocols tailored for the cooperative specific requirements. The employed methodology makes the results and conclusions valid for a wide range

Authorized licensed use limited to: University of Pennsylvania. Downloaded on July 22,2010 at 07:46:17 UTC from IEEE Xplore. Restrictions apply.

You might also like