Wireless Networks and Multimedia Services Department, TELECOM ... MAC combines both CSMA and TDMA techniques while .... Hard and results have shown that DRAND incurs O(į) running .... An example of collisions that make nodes activate the recovery .... to reuse this slot to transmit data in a collision free manner.
A Free Collision and Distributed Slot Assignment Algorithm for Wireless Sensor Networks Ines Slama, Badii Jouaber and Djamal Zeghlache Wireless Networks and Multimedia Services Department, TELECOM & Management Sudparis 9, rue Charles Fourier - 91011 Evry France {Ines.Slama, Badii.Jouaber, Djamal.Zeghlache}@it-sudparis.eu
Abstract-This paper introduces I-MAC a new hybrid medium access control protocol for wireless sensor networks (WSNs). IMAC combines both CSMA and TDMA techniques while introducing prioritization. It aims to improve energy efficiency as well as channel utilization. I-MAC is composed of two phases: a setup phase and a transmission phase. During the setup phase, several operations such as neighborhood discovery, slot assignment, local framing and global synchronization are run. During the transmission phase, a priority scheme is investigated to control the access to the channel. We focus in this paper on the description and evaluation of the different setup phase operations particularly DNIB, a new TDMA free collision slot assignment algorithm proposed for WSNs.
I. INTRODUCTION A critical design issue for wireless sensor networks is the development of novel communication protocols and algorithms that efficiently tackle the resource constraints and application requirements of wireless sensor networks [1, 2]. Therefore, power aware and power efficient protocols at each layer of the communication are required [2]. The MAC layer is considered as the main responsible of energy consumption. Indeed, it has been identified that the main reasons of energy waste in WSNs are: collisions, overhearing, protocol overhead and finally transmission and reception operations [3]. CSMA and TDMA techniques have been widely used for medium access in wireless sensor networks. CSMA is a contention based access protocol considered as simple, flexible and robust. It is a distributed mechanism that does not require clock synchronization or any knowledge of the network topology. Moreover, nodes can dynamically join or leave the network without additional overhead or operations. However, all this comes at the cost of collisions that occur when two or more nodes (within the same radio range) transmit at the same time. Collisions can cause serious energy waste, high overhead and throughput degradation. CSMA performance decreases then under high contention in terms of both energy consumption and channel utilization. On the contrary, in TDMA, collisions are reduced by scheduling the transmissions of neighboring nodes by minimizing simultaneous transmissions [4]. Thus, TDMA can deliver very good performance especially under high
contention. Moreover, it can provide higher energy saving than CSMA since it allows avoiding wasted energy due to collisions, idle listening and overhearing. However, the challenge with TDMA is to setup an efficient and scalable schedule that ensures collision free transmission times of all nodes in the network. As sensor networks may undergo frequent topology changes caused by battery outage and node failures, the TDMA schedule should gracefully handle these variations and dynamically adapt (frame length and/or time slot assignment) with respect to energy conservation aims. Moreover, under low contention, TDMA gives much lower channel utilization than CSMA since nodes can only transmit during their scheduled slot. TDMA needs also clock synchronization, which incurs high-energy overhead. These difficulties with TDMA and CSMA suggest that a stand-alone TDMA or CSMA scheme cannot be efficient. Moreover, in most of WSNs topologies, some nodes are heavily responsible of forwarding operations because of their central position or because they are closer to the sink and therefore, most of the routes go through them. Obviously, these critical nodes require more bandwidth resources to achieve their forwarding tasks. Besides, since these nodes have to transmit much more packets than other nodes (relaying role), they usually spend more energy and die earlier, which can affect the whole network lifetime. Therefore, controlling access to the channel in wireless sensor networks plays a key role in determining channel utilization, energy consumption and fairness in channel usage [4]. Optimizing channel and energy utilization have been the main focus of significant researches over the last years [310]. Different approaches dealing with this topic have been proposed. In [4], [5], [6], it has been shown that hybrid MAC schemes outperform contention based and schedule based schemes. Indeed, hybrid approaches combine the flexibility of contention-based schemes and the efficiency of schedule based schemes. Unfortunately, most of these hybrid mechanisms equally consider nodes and do not consider the heterogeneity in their loads and roles in the networks. In this paper, we introduce I-MAC, a new Hybrid Medium Access Control Protocol for wireless sensor networks. IMAC targets at improving both channel utilisation and
978-1-4244-2324-8/08/$25.00 © 2008 IEEE. This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2008 proceedings.
energy efficiency while taking into account traffic load for each sensor node according to its role in the network. II.PROPOSED SCHEME I-MAC is a new hybrid MAC protocol designed for wireless sensor networks. It not only combines the strengths of CSMA and TDMA techniques while obviating their weaknesses but also introduces an adaptive priority scheme for channel access. I-MAC operates in two phases: a set-up phase and a transmission phase. During the set-up phase, several operations are successively run: x Neighbours discovery x TDMA Slot assignment x Local framing x Global synchronization These operations run only once during the setup phase and do not run until a significant change in the network topology (such as sensor’s position changes, joining or leaving node…) occurs. Neighbours discovery operation is run by each node and mainly consists in determining the list of its one-hop and two-hop neighbours. This list is then used as an input for the slot assignment operation. The slot assignment problem can be defined as allocating a time slot for each node such that two nodes within two-hop distance should not be allocated the same time slot. To this end, DNIB (Distributed Neighborhood Information Based algorithm), an energy efficient and scalable TDMA channel scheduling algorithm is proposed. DNIB achieves distributed and parallel TDMA scheduling based on 2-hop neighborhood information, which allows it to handle the dynamic topology changes with low complexity and small energy waste. After the slot assignment, each node reuses its assigned slot periodically in every predetermined period called Local Frame. As in [4], we call a node assigned to a time slot an owner of that slot and the others the non-owners of that slot. Once each node computes its Local Frame, Global Synchronization is set-up and the Transmission phase can then be started. During the transmission phase, time is divided into slots. All nodes are allowed to transmit during any slot. Before a node transmits during a slot, it always performs carrier sensing and transmits a packet when the channel is clear. The channel access by the different nodes is controlled by a priority scheme. Indeed, the idea behind I-MAC is to assign different levels of priority to nodes according to their roles. For each node, the priority is implemented by adjusting the initial contention window size according to its priority level. If it has something to transmit, the owner of the slot is guaranteed to get the access first. Whenever the owner has no data to transmit, or when the slot is not owned, non-owners can compete to use the slot. The chance a non-owner node gets a slot depends on its priority level. Assigning a higher priority to forwarding nodes (that have a lot of packets to transmit) ensures that these nodes will be
able to transmit ahead of low priority nodes (in most of the cases) and hence improve the channel utilization. At the same time, and since shorter back off periods are given to higher priority nodes, their energy consumption is reduced [13]. In addition, since competition to access the channel (to get a slot) occurs between nodes within the same priority group, probability of collisions is lowered because the number of competing nodes within each group is lower than the total number of nodes [13]. This energy saving meets the main requirement of WSNs that is lifetime maximization, since it is known that highly loaded nodes usually run out of energy first. In I-MAC, priorities are dynamically attributed during the transmission phase according to the dynamic topology changes and hence to the requirements of the nodes in terms of bandwidth and energy. Resource allocation is statically performed by DNIB during the set up phase and do not consider each node requirements. However, the priority scheme, used in the transmission phase, comes to dynamically and fairly distribute the resources over nodes according to their requirements. In the following, we focus on describing how we implement the different set-up phase operations mentioned above. III. TDMA SLOT ASSIGNMENT ALGORITHM: DNIB TDMA scheduling can be generally defined as assigning transmitting and receiving slots to each node so that collision free transmissions can be achieved [14]. Transmissions may collide in two ways typically referred to as primary conflict and secondary conflict [15]. Primary conflict occurs when a node must do more than one thing at the same time, for instance, transmitting and receiving at the same time or receiving from two nodes at the same time. Secondary conflict occurs when a receiver u, tuned to a particular transmitter v, is within the transmission range of another transmitter whose transmission, though not intended to u, interferes with the transmission of v. Combining these two definitions, we can say that two nodes are in conflict if and only if the simultaneous transmissions from these two nodes causes radio interference at some node. There are two kinds of TDMA scheduling. Broadcast and link [15]. In the broadcast scheduling, the entities scheduled are the nodes whereas, in the link scheduling, they are the links between the nodes. Hence, in the broadcast scheduling, conflict can happen among all the nodes within a two-hop neighbourhood. In the link scheduling, conflict can happen among all the nodes within a one-hop neighbourhood of a transmitter and a receiver. In this work, we focus on broadcast scheduling. From above definitions, we can define the slot assignment problem as finding a time slot for each node, such that no two nodes within two-hop distance should be allocated the same transmitting slot. The performance of a slot assignment algorithm can be determined by the following quantities:
978-1-4244-2324-8/08/$25.00 © 2008 IEEE. This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2008 proceedings.
-
Running time: i.e. the maximum time taken for all the nodes in the network to be assigned a slot for all executions of the algorithm. - Message complexity: i.e. the number of messages exchanged between all the nodes in the network to decide on the time slots. An appropriate slot assignment algorithm should then be as simple as possible with low running time and message complexity. Otherwise, high-energy waste can be expected due to messages transmission and reception operations and overhead. Several TDMA scheduling algorithms were proposed in the literature. Most of them are centralized and therefore not scalable nor efficient for large scale WSNs. Indeed, centralized strategies generally lead to high overhead and high-energy consumption because of the high amount of exchanged messages that are proportional to the network size. Hence, distributed algorithms are more likely to be appropriate for such networks. Recently, some distributed mechanisms were proposed for wireless network. We only consider here the two most relevant to our work and to which our results will be compared. DRAND [12] is a randomized time slot scheduling scheme that implements the centralized scheduling scheme RAND [15] in a probabilistic and distributed way to guarantee both scalability and efficiency. DRAND is the first fully distributed version of RAND and it suits for static wireless ad-hoc networks. In DRAND, the TDMA slot assignment is considered as a direct extension of the graph-coloring problem. The objective is to color the vertices of a graph with a minimum number of colors such that no two adjacent nodes have the same color. This problem is known to be NPHard and results have shown that DRAND incurs O(į) running time and message complexity, where į is the number of two hop neighbors of a node. Deterministic Distributed TDMA scheduling algorithm (DDTDMA) [14] explains how to increase energy efficiency by decreasing the TDMA protocol complexity. It explains that if transmit and receive slots interval is short, then energy consumption can be reduced by keeping nodes in idle state rather than in sleep state. It assumes that switching from sleep mode consumes a non-negligible amount of energy. DD-TDMA claims to outperform DRAND in terms of running time and message complexity. In the following, we present a new time slot assignment algorithm named DNIB (Distributed Neighborhood Information Based algorithm) that allocates transmission opportunities in a free collision manner. DNIB is an efficient scheduling scheme that achieves distributed and parallel TDMA scheduling with lower complexity, fewer exchanged messages and shorter running time compared to existing schemes. A. Assumptions To allow subsequent comparisons with existing mechanisms, we consider here the following assumptions which, are similar to those considered in DD-TDMA and DRAND:
1)Each node has a unique ID. 2)Nodes are synchronized with slot 0. 3)The network is fully connected and each node has information of IDs and hop distances to the sink of its one hop and two-hop neighbors. 4)If a node cannot decode a message, it discards it as collision and does nothing. B. Algorithm specifications As in DD-TDMA and DRAND, the idea behind the DNIB algorithm is to let each node decide its own slot according to the information collected from its one and two-hop neighbor nodes. This information includes neighbors IDs, their distance to the sink and whether they are or not yet scheduled. The DNIB algorithm, running in parallel in each node, is composed of the following procedures: - Slot assignment procedure - Update procedure and - Recovery procedure 1) Slot assignment procedure At the beginning of the algorithm each non-scheduled node computes contender ranks for itself and for its non-scheduled neighbors, according to the ids and to the hop distances to the sink. These contender ranks are computed as follows: Let h be the concatenation of the hop distance (Hop-Dist) and the id (Node_Id) of each non-scheduled node. h = Hop-Dist Ҷ Node_Id Let H be the ordered set of h values corresponding to the non-scheduled one and two-hop neighbors of a given node. The position of the node within H determines its contender rank. A node considers itself as contender 1 (rank = 1) when it has the highest h value. The contender rank is updated each time a neighbor is scheduled. Once a node is scheduled it auto-assigns contender rank 0. We denote by Ci the contender rank of node i. (Ci Ы ͲA node i gets a slot when it becomes contender 1 (Ci=1). It selects the minimum unassigned slot within its one and twohop neighbors. It then informs its neighbors through the Update procedure.
Fig.1. An example of collision during the Update procedure. We assume in the figure above that the numbers in parenthesis are Ci (contender rank i). Consider that node 4 which is Contender 1 got a slot and sends a “one-hop broadcast” message with a “two-hop broadcast schedule” to its one-hop neighbors during slot i. During slot i+1 and according to this schedule, node 2 has to send the two hop broadcast message to its one-hop neighbors. At the same time, node 3 becomes contender 1. If node 3 selects a slot and sends immediately its one-hop broadcast message, a collision will happen for sure
978-1-4244-2324-8/08/$25.00 © 2008 IEEE. This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2008 proceedings.
between node 2 and node 3 messages. In order to avoid these foreseeable collisions, we choose to introduce the Ȧ parameter. This parameter depends on the one-hop neighbors’ number.
2) Update procedure Once a node gets assigned a slot, it sends a “one-hop broadcast” message to update its one-hop neighbors. Those one-hop neighbors send in their turn a “two-hop broadcast” message to update two-hop neighbors. The “one-hop broadcast” message contains: the scheduled node id, the assigned slot and a “two-hop broadcast schedule” that defines for each one-hop neighbor the slot in which it will transmit the two-hop broadcast message. The “two-hop broadcast” message contains all already known scheduling information of one-hop neighbors. During this procedure, two types of collisions may occur: - collisions between “two-hop broadcast” messages. - collisions between “two-hop broadcast” messages and “one-hop broadcast” messages. The “two-hop broadcast schedule” defined above (sent within the “one-hop broadcast” message) is used to avoid the first type of collisions. This is different from the DD-TDMA algorithm where the two-hop broadcast messages are sent randomly. Collisions of the second type can be reduced by delaying one-hop broadcast messages. In DNIB, we propose that, before sending the “one-hop broadcast” message, each scheduled node should wait a predetermined number of slots, denoted by Ȧ. An example is illustrated in Figure 1.
figure 2). In this case, the node can send a “request” message (a specific one hop broadcast message) to its one hop neighbors. This request message has the same format and payload as the “one-hop broadcast” message except the assigned slot values which is set to NULL. It triggers an update procedure forcing its one-hop neighbors to send a “two-hop broadcast” message with all already known scheduling information.
-
Fig.3. An example of recovery procedure. Let’s assume this example of figure 3 where Ȧ = 5. Numbers in parenthesis are Ci. Node 3 is contender 1. Without collisions, node 2 will receive information from 3 (and become contender 1) after 7 slots. Once node 2 is assigned, node 1 will receive information about 2 after 7 other slots. Node 1 can then become contender at the 15th slots. It is then useless that node 2 and node 1 send requests before respectively 7 slots and 15 slots. They should wait the appropriate time before triggering the recovery procedure. The parameter Ri is designed for this purpose. Moreover, and in order to minimize the running time of the algorithm, we chose to set an upper bound for Ri so that nodes do not wait a too long time.
The value of Ri should be appropriately determined in order to: - avoid collisions between nodes making their requests at the same time. To this end, we propose Ri as a function of Ci and a random number. - avoid that a node sends too early/late requests. An illustration is given in Figure 3. We propose here the following expression for Ri:
Ri = 10. (2 + 2. (Ci − 2) + rand);
Fig.2. An example of collisions that make nodes activate the recovery procedure. Consider the sensor network depicted in Figure 2. Numbers in parenthesis are Ci (contender rank i). Let’s assume that during slot i, nodes 3 and 4, recently scheduled, send each a “one-hop broadcast” message to node 2 and node 1 respectively. These messages are aimed to update their onehop neighbors. During the next slot (i.e. slot i+1), node 1 and node 2 send each a “two-hop broadcast” message containing information about the assignments of nodes 4 and 3 (respectively). A collision between these two latter messages occurs and as a consequence, node 1 and 2 don’t get updates about nodes 3 and 4 assignments. Node 2, which is normally the next contender 1, cannot become so, since it lacks information about node 4 assignment. The recovery phase is designed to recover such situations.
3) Recovery procedure This procedure is activated when a node does not succeed to get scheduled (i.e. could not be contender 1) for a predetermined period of time denoted Ri for node i. This may happen for instance because it misses information about some of its one and/or two-hop neighbors (see example of
where: Ci ҆ 2, (otherwise Node i is contender 1) and rand Ы=? If after a request, node i still misses information about its neighbors it can send again another request after Ri time slots. The request operation is repeated until the node gets all the missing information and become able to get assigned a slot. Readers, who are interested in further details about DNIB algorithm pseudo code and mathematic evaluation, are referred to [11]. C. Correctness The correctness of the DNIB algorithm is supported by the following reasoning. Theorem 1: Each two nodes within a two-hop neighborhood are assigned different slots. Proof: This can be proved by following three arguments: (1) a node does not select a slot unless and until it becomes contender 1. (2) At any time, and since all nodes have different IDs, there is one and a unique contender 1 within its
978-1-4244-2324-8/08/$25.00 © 2008 IEEE. This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2008 proceedings.
two hop neighborhood. (3) Contender 1 can only select among unused slots within its two-hop neighborhood. Claim 1: The DNIB algorithm converges even if collisions occur. Justification: With the DNIB algorithm collisions are partially controlled. Indeed, there are no possible collisions between one-hop broadcast messages since only one node is contender 1 within its two-hop neighborhood at a time (see theorem 1). In addition, each node getting a slot (i.e. contender 1) schedules the two-hop broadcast messages within its neighborhood. Therefore, collisions are reduced since they may only happen with messages (one-hop or twohop) coming from outside this neighborhood. Moreover in case of collisions, at least one node (let’s say node i) will not be updated with its neighbors’ information. Nevertheless, this does not stop the DNIB algorithm on that node. Through the recovery procedure, node i will send a “request” message to all its one-hop neighbors after Ri time slots. Since Ri is determined based on Ci and on a random number, the probability of collisions between two “request” messages is low. Furthermore, even if one of the neighbors of node i does not receive the “request” message because of a collision, node i still has a chance to receive reply messages from its other neighbors with the missing information. Finally, a node that misses information will repeatedly send a request message each Ri time slots until it gets fully updated. It then becomes contender 1 and gets scheduled. IV. LOCAL FRAMING Once slot allocation is achieved, the next issue concerns the TDMA frame size of each node. Indeed, after being assigned a slot, a node needs to know in how many slots it will be able to reuse this slot to transmit data in a collision free manner. We call this period time frame and we use the Time Frame rule (TF rule) proposed in [4] to determine its size. According to this rule, each node maintains its proper local time frame that depends on its two-hop neighbourhood size and that avoids any collision with its neighbours. Local frame approach makes the slot assignment scheme adaptive to dynamique topology changes and improves the channel utilization in non-uniform density networks. Readers interested in these issues are referred to [4] for more details. The TF rule is defined as follows: Let ASNmax be the maximum allocated slot number within two-hop neighbourhood of a node. The local frame size of this node is equal to 2a, such that: 2 a -1 d ASN max 2 a . a
Thus, a node can reuse its assigned slot each 2 slot with the guaranty that no other node within its two-hop neighbourhood uses any slot that it uses [4]. V. GLOBAL SYNCHRONIZATION Once each node computes its Local Time Frame, a global synchronization should be set-up i.e., each node starts its time slot 0 at the same time. Like in [4], this is achieved by
setting the beginning of the real time (when the synchronized clock value is 0) to be the beginning of time slot 0. The global synchronization is performed only once and during the set-up phase. However, each node needs to run a local synchronization scheme during the transmission phase. At the end of the setup phase, each node forwards its local frame size and its assigned slot number to its two-hop neighbourhood. Thus, before starting the transmission phase, a node knows at which slots each of its one-hop and two-hop neighbours will transmit. After this phase, all nodes are synchronized to slot 0 and ready to transmit. During the transmission phase, a local synchronization will be run at each node. VI. SIMULATIONS RESULTS AND ANALYSIS We focus in this section on the evaluation of the performances of the slot assignment scheme DNIB. To this end, we implement it using MATALAB. For comparison purposes, the DD-TDMA algorithm is also implemented. However, we do not implement DRAND since we know from [14] that DD-TDMA performs better than DRAND in terms of both running time and message complexity. Two scenarios were investigated: Scenario 1: This first scenario is considered to examine the algorithm scalability. The network topology is uniformly generated and N nodes are deployed into an area of ¥Nx¥N, so that when N varies the node density within the same transmission range remains as constant. We assume in this scenario that, when the transmission range r is set to 1 unit, the maximum one-hop neighborhood size is 4 and the maximum two-hop neighborhood size is 12. The Ȧ parameter is set to 5 slots. Scenario 2: This second scenario is considered to examine the impact of different node densities by varying the nodes’ transmission range. N nodes are randomly deployed into an area of 10x10 units. The number of nodes is kept constant (N = 100) while the transmission range varies (from 2 to 3 units). The parameter Ȧ is equal to 5 slots. The performance results presented here are obtained by averaging the results for 100 different simulations for each scenario. Number of nodes
100
200
300
400
Running time (slots)
473.81
633.89
814.25
891.07
# of msgs per node
9. 2028
10.999
11.976
12.144
Table I. Scalability of DNIB
Table I depicts the running time and the message complexity (i.e. number of transmitted messages per node) for DNIB. Simulations are conducted by varying the number of nodes from 100 to 400. Results show that the DNIB algorithm is scalable since, while the size of the network is increased, the number of messages transmitted per node stays relatively constant and the running time increases linearly. DNIB performances do not depend on the network size, but only on
978-1-4244-2324-8/08/$25.00 © 2008 IEEE. This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2008 proceedings.
the neighborhood size (see figures 4 and5). Indeed, DNIB is a distributed scheduling algorithm that is run in parallel over different neighborhood areas within the network. Further results are obtained by varying the transmission range of the nodes. Figure 4 and Figure 5 show that DNIB outperforms DD-TDMA in terms of running time and message complexity.
nodes that have higher load to get more chances to access the radio resources while saving energy. This is achieved through an adaptive prioritization mechanism embedded within a CSMA/TDMA based scheme. This approach will be investigated in more details and evaluated in a future work. REFERENCES [1]
[2]
[3]
[4] [5]
Fig.4. Running time of DNIB and DD-TDMA (r=2:3).
[6]
[7] [8]
[9]
[10]
Fig.5. Message complexity of DNIB and DD-TDMA (r=2:3).
Figure 4 shows that the DNIB running time is lower than that of DD-TDMA. Moreover, it is less sensitive to the transmission range (and therefore to the number of neighbors per node) since it less expands when the transmission range increases. Same conclusions can be done on message complexity (see Figure5). VII. CONCLUSION AND FUTURE WORK In this paper, we introduced I-MAC, an adaptive hybrid medium access protocol for wireless sensor networks. We mainly focused on the different operations run during the set up phase. We particularly dealt with DNIB, a new TDMA free collision slot assignment algorithm for wireless sensor networks. DNIB is designed as a distributed algorithm that can run in parallel over the network. It is only based on twohop neighborhood information and does not require knowledge of the whole network. Once the setup phase is achieved, the transmission phase can be started. During this latter phase, I-MAC allows sensor
[11]
[12]
[13]
[14]
[15]
ø. F. Akyıldız, W. Su, Y. Sankarasubramaniam and E. Çayırcı, “Wireless Sensor Networks: A Survey,” Computer Networks, No. 38, pp. 393-422, 2002. E. Cayirci, and P. C. Nar, “PCSMAC: A Power Controlled SensorMAC Protocol for Wireless Sensor Networks,” the 2nd European Workshop on Wireless Sensor Network, pp. 81 - 92, 2005. I. Demirkol, C. Ersoy, and F. Alagoz, "MAC protocols for wireless sensor networks: a survey," IEEE Communications Magazine, vol. 44, iss. 4, pp. 115-121, 2006. I. Rhee, A. Warrier, M. Aia, and J. Min, “Z-mac: a hybrid mac for wireless sensor networks,” in Sensys’05, Nov 2-4, 2005. A. D. Myers, Hybrid MAC Protocols For Mobile Ad Hoc Networks. PhD thesis, Computer Science, University of Texas at Dallas, May 2002. I. Chlamtac, A. Farago, A. Myers, V. Syrotiuk, and G.Zaruba, “A performance comparison of hybrid and conventional mac protocols for wireless networks”, in proceedings of VTC 2000, vol. 37, pp. 201-205, 2000. Y.K. Rana, B.H. Ajfandika, S. Jha, “Bandwidth aware slot allocation in hybrid mac”, in proceeding of LCN 2007, pp. 89-96. S. Li, D. Qian, Y. Liu, J. Tong, « Adaptive distributed randomized TDMA scheduling for clustered wireless sensor networks », in proceeding of IEEE WiCom, pp. 2688-2691,2007. W. Ye, J. Heidemann, and D. Estrin. An energy-efficient MAC protocol for wireless sensor networks. In 21st Conference of the IEEE Computer and Communications Societies (INFOCOM), volume 3, pages 1567–1576, June 2002. T. van Dam and K. Langendoen. An adaptive energy-efficient MAC protocol for wireless sensor networks. In 1st ACM Conf. on Embedded Networked Sensor Systems (SenSys 2003), pages 171–180, Los Angeles, CA, November 2003. I. Slama, B. Shrestha, B. Jouaber, D. Zeghlache, T. J. Erke, “DNIB: Distributed Neighborhood Information Based TDMA Scheduling for Wireless Sensor Networks” IEEE Vehicular Technology Conference VTC-fall 08, (Calgary, ALberta, Canada, September 2008) (accepted and to appear). I. Rhee, A. Warrier, J. Min and L. Xu, “DRAND: Distributed Randomized TDMA Scheduling For Wireless Ad-hoc Networks,” ACM MobiHoc 2006. O. Bouattay, T. Chahed, M.Frikha, S. Tabbane, “Improving energy consumption in ad hoc networks through prioritization”, in proceedings of VTC spring 2007, pp. 148-153, 2007. Y. Wang, I. Henning, “A Deterministic Distributed TDMA Scheduling Algorithm for Wireless Sensor Networks,” IEEE WICOM, 2007, pp.2759-2762. S. Ramanathan, “A unified framework and algorithms for (T/F/C)DMA channel assignment in wireless networks,” IEEE INFOCOM, 1997, pp.900-907.
978-1-4244-2324-8/08/$25.00 © 2008 IEEE. This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2008 proceedings.