Multiple-optimizing Dynamic Sensor Networks with MIMO Technology

Multiple-Optimizing Wireless Sensor Networks with MIMO Technology* Wei Chen, Miao He, Fletch Gregory Department of Computer Science, Tennessee State University 3500 John A. Merritt Blvd, Nashville, TN 37209, USA

Abstract: A Multiple-Input Multiple-Output (MIMO) transceiver provides extremely high spectral efficiencies by simultaneously transmitting multiple data streams in the same channel. MIMO links with MIMO transceiver at each end have been used to raise data rate or reduce energy consumption at links in a wireless sensor network (WSN). In this paper, we investigate the approach of cross-layer design for leveraging the advantage of the MIMO technology from local links to the whole network. We propose a novel MIMOaware cluster-based network architecture which enables efficient routing and inexpensive network reconfiguration. We also propose a MAC protocol which can improve data rate and energy usage at the same time in MIMO links by optimally assigning transmission antennas based on the conditions at the physical layer. Our routing algorithms provide robust and shortest paths on the proposed architecture. The performance evaluation shows that design can fully realize the potential of the MIMO technology and multiple optimize network throughput, network lifetime and network reconfiguration cost. Keywords: wireless sensor network, MIMO, cross-layer design, network architecture, MAC and routing protocol 1. Introduction A wireless sensor network (WSN) with a large number of inexpensive micro sensor nodes can be used for real time applications such as environment monitoring and vehicle tracking. In many scenarios, the wireless nodes must operate without battery replacement for years. High networking performance and low energy consumption are two of the most important issues. A hierarchically organized WSN induces efficient routing and offers higher networking performance. However, it requires extra time and energy for architecture which occurs when the network geographical topology changes. A sensor node leaves the network when its battery voltage is low, and a mobile node can move into and move out from the network at any time. The reconfiguration can be very expensive if the whole network has to be reconstructed. It is very challenge to design a WSN which can achieve high network throughput, long network lifetime, and low network reconfiguration cost at same time. In this paper, we introduce a new technology into WSNs. MIMO transceivers employ digital adaptive transmission and *This work was performed under contract with AFRL/RYWPAFB-08-0030.

reception antenna arrays which provide extremely high spectral efficiencies by simultaneously transmitting multiple data streams in the same channel. We assume that in our MIMO WSN each node has a MIMO transceiver. A MIMO link with MIMO transceivers at both ends can provide two types of gain: diversity gain and spatial multiplexing gain. Diversity gain primarily provides range extension, while spatial multiplexing gain primarily provides higher data rate. Theoretically, both gains can be obtained with the same channel without using extra energy, though practically, the circuit and the coding for a MIMO transceiver are more complicated and need more energy. So far, most research works have focused on raising data rate, or reducing energy consumption at MIMO links. In literatures [2, 3], energy management is embedded into the MIMO transceiver for minimizing the energy consumption at MIMO links. In literatures [16], the authors investigated the tradeoffs of diversity gain and multiplexing gain. They concluded that by dividing the antenna arrays both types of gains can be achieved simultaneously. Recently, some teams have been working on improving network throughput. In literatures [10, 12, 13, 15], MAC protocols are designed jointly with MIMO transceivers, where network throughput is improved by controlling the data streams in antenna arrays at MIMO links. In literature [11], routing protocols are designed jointly with MIMO transceivers, where the routing is improved by extending transmission range by using diversity gain. We have not seen any research work that can multiple-optimize network throughput, network lifetime, and network reconfiguration cost. In this paper, we jointly design the network architecture, MAC protocol and routing algorithms with underlying hardware conditions by using the MIMO technology for achieving multiple-optimization in the global network. We first figure out the tradeoffs among the constraints in a MIMO transceiver and establish a model for physical layer. We propose a MIMO-aware cluster-based network architecture which has a number of nice properties: (1) it contains a small number of clusters, therefore, can provide efficient routing, (2) it supports inexpensive reconfiguration: node-joining and node-leaving operations can be performed in a local single hop, and (3) it extremely extends network lifetime: the network can be still alive even most nodes died or moved out. Head rotating and link jumping enabled are main techniques by the MIMO technology used for architecture maintenance, where head rotation is used for balancing workload inside a cluster

and link jumping is used for keeping the network connected even when all nodes of a cluster died or moved out. We also design a MIMO-aware MAC. Our MAC protocol is CSMA/CA based and works as a local optimizer. By optimally using diversity gain and multiplexing gain based on the trade-offs provided by the physical model, it simultaneously minimizes energy consumption and maximize data rate at MIMO links. Finally, our routing algorithms perfectly combine the proposed cluster-based architecture and MAC protocol: routes are short and robust and the data rate and energy consumption at each link of the routes are locally optimized by the MAC protocol. We evaluate the performance of the proposed network architecture, MAC protocols and routing algorithms on various types of MIMO links, filed sizes, numbers of sensor nodes, communication ranges, battery energy and other constraints. We conclude that our joint design can fully exploit the advantage of MIMO technology and significantly improve the network throughput, network lifetime and network reconfiguration cost. We also compare the performance of our cluster-based MIMO WSN with that of a flat MIMO WSN, and conclude that the proposed clusterbased architecture plays a crucial role in our design.

2. MIMO Technology and Its Physical Model A MIMO transceiver employs digital adaptive transmitting and receiving antenna arrays. It can be used for increasing the amount of diversity or the number of degrees of freedom in wireless communication systems. Traditionally, multiple antennas have been used to increase diversity to combat channel fading. Each pair of transmitter and receive antennas provides a signal path from the transmitter to the receiver. By sending signals that carry the same information streams through different paths, multiple independently faded replicas of data symbol can be obtained at the receiver end; hence, more reliable reception is achieved. The diversity gain can be used to provide range extension or reduce error rate. On the other hand, if the path gains between individual transmitreceive antenna pairs fade independently, the channel matrix is well conditioned with high probability, multiple parallel spatial channels are created. By transmitting independent information streams in parallel through the spatial channels, the data rate can be incurred. This effect is called as special multiplexing gain. The maximum diversity order afforded by an M × N MIMO link with M transmit antennas and N receive antennas is MN; meanwhile, assuming that M = N, the capacity grows linearly with M from spatial multiplexing gain [10, 11, 12]. In Fig.1, assuming that the route from node a to node d is abcd, node a can use spatial multiplexing gain to get higher data rate in link ab. Meanwhile, when node b ran out of the battery or moved out, node a can use diversity gain to extend its communication range to node c. According to [16], a MIMO link can achieve both types of gain at same time by dividing antenna arrays. For example, an 8×8 MIMO link can be used as four 2×2 MIMO virtual links: each 2×2 MIMO link transmits coherent data signal for getting

diversity gain and different virtual links transmit independent data signal for getting multiplexing gain. MIMO transceiver T× 1

R× 1

T× 2

R× 2

MIMO sensor network

c

b T× M

d

a

R× M N

diversity gain multiplexing gain

Fig. 1 MIMO transceiver and its gain: diversity gain and multiplexing gain Theoretically, diversity gain and multiplexing gain can be obtained in the same channel without using extra energy. However, the transceiver circuit with more transmitter and receiver antennas consumes a bit more energy, and the encoding/decoding the transceiver consumes extra energy too. In this paper, we assume that each node in a WSN is equipped with an M × M MIMO transceiver. The following physical model is achieved by simplifying the models in [2, 3]. We use the same assumption that the energy for encoding/decoding is small enough to be ignored. The total energy consumption per bit, denoted as Eb , for a fixed-rate system can be estimated approximately as follows:

Eb = ( PPA + Pc ) / Rb ------------------------- (a)

PPA is the power consumption for transmission and reception, Pc is the power consumption for circuit and Rb is the bit rate for each antenna, respectively. Furthermore, PPA and Pc can be estimated as follows: where

PPA = (1 + α )

MRb N 0 (4πD) 2 × Ml N f 1/ M Gt Gr λ pb

Pc = φM + ϕ where D, M , pb are the transmission distance, number of transmitter and receiver-antennas in the MIMO transceiver, and average bit error rate, respectively. Others are system constants. Furthermore, the spatial multiplexing gain obtained from a M × M link can be explained as follows:

RbM = εMRb -------------------------------------- (b) where Rb is the bit rate of a M × M MIMO link when all M

antennas are used for spatial multiplexing gain, and ε ≤ 1 is a system constant that can be 1 if the multiplexing fading on the different antenna elements is completely uncorrelated. The physical model provides the tradeoffs in physical layer and gives a foundation for our cross-layer design.

3. MIMO Aware Cluster-based Network Architecture Clustering has been used to induce hierarchical structure over a flat (unstructured) WSN which minimizes communication overhead, and facilitates time and energy efficient networking functions. Several reconfigurable cluster-based architectures have been proposed. In [8], reconfigurable overlays are designed for multi-scale communication. In [14], the architecture reconfiguration is enabled by using node-joining and node-leaving operations. A node-joining operation can be executed in a single hop; however, a node-leaving operation can cause a complete reconstruction of the architecture, which is an unavoidable problem for most reconfigurable architectures. Also, the reconfiguration is possible only when the network is connected. Some literatures use head-rotation to reduce the workload of cluster head nodes [4, 5, 9]. The technique has the same problem, that is, it does not work if all nodes in some cluster died or moved out. In this section, we show how the MIMO technology can be used solve the problem.

3.1 Cluster-based Structure

We assume that each node in a WSN has an M × M MIMO transceiver operating a single channel. Since transmission range at a node can be extended by using diversity gain, we call the initial transmission distance without any extension to be primary transmission distance, denoted as d. A flat WSN can be represented by an undirected graph G = (V, E), where V is the set of sensor nodes, and nodes u and v have an edge between them iff they are in the primary transmission distance with each other. We have the following assumptions: 1. Sensor nodes in G work repeatedly in rounds. One round consists of one transmission or one reception, and some local computation. 2. When a node receives information signals from more than one node, collision happens, and the node can not get any information. 3. When the nodes are deployed in a field, each node knows only its ID, i.e., they do not know any network knowledge such as neighbors and number of nodes in the network. We first define a primary cluster-based structure on G. In a primary clustering, the nodes of G are separated into nodedisjoint clusters, where each cluster is a complete subgraph of G (Fig.1 (a)). The primary cluster-based structure of G is a tree formed by the nodes of G with two types of edges: cluster edges in E connecting cluster heads and their members (thin edges in Fig. 2(b)); backbone edges not in E and they are added for connecting cluster heads (thick edges in Fig. (b)). The transmission distance for backbone edges is 2d.

(a)

(b)

Fig. 2 (a) A primary clustering: each cluster is a complete subgraph of G; (b) A primary cluster-based structure formed by cluster-edges (thin ones) with transmission distance d and backbone-edges (thick ones) with transmission distance 2d A general cluster-based structure of G is the same as that of a primary cluster-based structure except the transmission distance for a backbone edge can be hd ( h ≥ 2) , where h can be different for each edge.

Definition 1 (1) A cluster-based structure of a graph G = (V, E) is a tree, denoted as CNet( G ) = (V , E CNet ) , where E CNT consists of cluster edges and backbone edges. In CNet( G ), the cluster-edges have the same primitive transmission distance d, and the backbone edges can have different transmission distance hd ( h ≥ 2) . (2) The backbone tree of G, denoted as BT(G), is the subtree of CNet(G) formed by cluster heads and backbone edges. In Fig 2 (b), the backbone of G consists of the cluster heads and the backbone edges (thick edges). In our cluster-based structure, we assume that each node v has a unique ID, v.id. v also keeps the following one-hop knowledge: (1) node v knows its status as a cluster member or a cluster head, denoted as v.status; (2) if v is a cluster member, it knows its cluster head’s ID, denoted as v.head; and (3) if v is a cluster head (i) it knows the knowledge of its cluster members, denoted as v.members. The knowledge includes member’ ID and battery energy level, and (ii) it knows the knowledge of its backbone neighbors (neighbors in BT(G)) and transmission distance d(v,w) from v to each neighbor w. Assume that G0 is the flat sensor network formed by n sensor nodes which are deployed in a field randomly. A CNet( G0 ) can be self-initialized as follows. One node invokes a gossip. All nodes will get the knowledge of G0 when the gossip completed. Then, each node forms the same CNet( G0 ) with the knowledge of G0 by using node-joining operation (it will be described in Section 3.2). A gossip can be completed in O(n) rounds [1]. The construction of the CNet( G0 ) at each node is local computation. Therefore, the initialization can be completed in O(n) rounds.

3.2 Head-Rotation and Link-Jumping Operations head-rotation and link-jumping are designed for maximizing network lifetime. They will be used in network reconfiguration, MAC protocol and routing algorithms later. Head-rotation is used for evenly using the nodes in a cluster. Link-jumping is invoked when a cluster-head u is the final node in the cluster and u is going to die. In the backbone tree BT(G), u’s neighbors are either u’s children or u’s parent. In the link-jumpting, u’s children extend their transmission distances so that they can connect to u’s parent. Head-Rotation Algorithm (u: head, e: energy threshold) 1. u checks its battery energy; 2. if u’s energy level is lower than e then if there is no member whose energy level is larger than e, then u invokes link-jumping operation; else u selects a member v from its cluster with the largest energy level; u transmits a head-rotation request “new head is v” with its member list using primary transmission distance d; when v received the request from u, v changes its status to be head, and delete u from its member list; when u’s any other member w received the request from u, w changes v to be its head. Link-Jumping Algorithms (u: head) 1. u transmits a link-jumping request with the following knowledge using transmission distance max{d(u,v), where v is u’s backbone neighbor}: u’s parent w in BT(G), and transmission distance d(u,w) from u to w. 2. When any u’s backbone neighbor v receives the request with the knowledge, v deletes u from its backbone neighbor list, adds w into its backbone neighbor list as the parent, and sets the transmission distance from v and w to be d(v,w) = d(v,u)+d(u,w). In the above link-jumping, in order to reach every backbone neighbor, u sends the request using the largest transmission distance among u and its neighbors.

range, then new becomes the member of that cluster; otherwise, new becomes the cluster-head of a new cluster and new sets the cluster-head of w ( it is w when w is the clusterhead) to be its parent in the backbone tree. Node-Joining Algorithm (new: node for joining) 1. new invokes a numbering algorithm to give an order to all its neighbors in G (when this step finishes, new has a list of ordered neighbors in G); 2. new transmits a request “I want to join” with a list of the ordered neighbors using primary transmission distance d; 3. When new’s neighbors receive the request, they set new to be their neighbor in G and send their knowledge back to new one by one in turn; 4. if new finds a head u in its neighbors and all u’s members are new’s neighbors (i.e., new and u’s members form a complete graph), then new sets u to be its head; new transmits a request “join to u’s cluster” to u using primary transmission distance d; when u received the request, u adds new into its member list; else new sets itself to be the cluster-head of a new cluster, then it selects a node w from its neighbors and set the cluster-head of w (it is w if w is a cluster-head), say x, to be its backbone neighbor and the parent in BT(G); new transmits a request “I am x’s backbone neighbor” using transmission distance d(new,x) = d(new,w) + d(w,x); when x received new’s request, x adds new into its backbone list and set the transmission distance to be d(x, new);

Theorem 2 Give graph G and CNet(G), a node-joining operation can be completed in O(δ new ), where δ new is new’s degree in G. Proof: In Step 1, the numbering algorithm can be completed in O(δ new ) rounds [14]. In Step 3, δ new rounds are needed for each neighbor sending its knowledge to new in turn. Other steps use at most 3 rounds.

Theorem 1 A head-rotation and a link-jumping can be completed in O(1) rounds, respectively. Proof: In both operations, u transmits a request with some knowledge once only, and u’s cluster members or its backbone neighbors correct their local knowledge when they receive the request.

3.3 Network Reconfiguration In this section, we define node-joining and node-leaving operations for network reconfiguration. Let G = (V, E) be a flat WSN, CNet( G ) = (V , E CNet ) be G’s cluster-based WSN, and new are the node who wants to join G. We assume that at least one node w of G is in new’s primary transmission range d. In the node-joining operation, if new can find a cluster whose members are all in new’s primary transmission

Node-leaving is a very tough task in a cluster-based WSN if not using MIMO technology. However, with the headrotation which including link-jumping enabled by MIMO technology, the operation needs only O(1) rounds. Assume that a node lev wants to leave G. The following algorithm reconfigures the cluster-based structure when old leaves G. Node-Leaving Algorithm(lev: node for leaving) if lev is a cluster-member, then lev transmits a request “lev is leaving” using primary transmission distance d, and lev leaves (move out or turn off the power); when lev’s head receives the request, it deletes old from its member list; else lev invokes head-rotating operation (either a cluster-

member will replace lev if lev has alive cluster-members or lev’s backbone children will connect to lev’s backbone parent by invoke a link-jumping), and lev leaves.

Theorem 3 Give graph G and CNet(G), a node-leaving operation can be completed in O (1) rounds. Proof: Step 1 needs 1 round. Step 2 invokes a head-rotating operation which can be completed in O (1) rounds according to Theorem 1.

4.MAC Protocol and Routing Algorithms 4.1 MAC Protocol as Local Multiple-Optimizer Carrier Sense Multiple Access with Collision Avoidance (CSMA/CA) has been used for MAC protocol for ad-hoc network environments in many cases. A simple extension of CSMA/CA for a M × M MIMO link, denoted as CSMA/CA(M), can be realized that can provide a k fold improvement in throughput performance through spatial multiplexing gain comparing to a pure omni-directional environment [14]. According to formula (b) in Section 2, the bit rate gain from a M × M MIMO link with the M CSMA/CA(M) protocols is Rb = εMRb , where ε ≤ 1 is a system constant. CAMA/CA(M) allows only fixed transmission to take place in a give time slot. In literature [11, 12], a new CSMA/CA(M)-based MAC protocol is proposed that achieves higher data rate by controlling the data streams at transmission antennas. Our MAC protocol, called as MultiOptimal-MAC, is a CSMA/CA(M)-based multioptimizer. It improves the data rate and energy usage at the same time by optimally using MIMO antenna arrays for achieving diversity gain and spatial multiplexing gain. Assume that an SISO link uses energy e to transmit one bit in the primary transmission distance d with data rate r. As a default, the antenna channels in a MIMO link are assigned according to the current transmission distance h at the link. If h = kd, the M × M MIMO link is separated into t = M/k groups and each group is a k × k MIMO link. The antennas in the same group transmit coherent data stream to get diversity gain for extending the transmission range. Different groups transmit independent data stream in order to get multiplexing gain for increasing data rate. In this way, a bit is transmitted k times farther and t times faster. The energy usage can be calculated by formula (a) and it should be almost the same as that the SISO uses. The MultiOptimalMAC can be customized and used in various ways. For example, if a date rate of r times higher is required, the M × M MIMO link can be separated into r groups and each group is a M / r × M / r MIMO link. The antenna array can be assigned also based on the required error rate and other issues.

4.2 Routing Algorithms

In this section, we show the routing algorithms for broadcast, data gathering and node-to-node communication. The routes are scheduled on the back bone tree BT(G). In our algorithms, we use a procedure Eulerian(T, s, m) which relays a massage m from a source node s to all other nodes in a tree T by traveling a Eulerian tour of T. In a directed graph, an Eulerian Tour is a tour which passes each edge of the graph exactly once. Assuming that T is undirected tree, by replacing each edge of T with two edges in opposite directions, T can be changed into a directed graph. Procedure Eulerian(T, s, m) can be completed in 2× | T | rounds [14], where |T| is the number of edges in T. This procedure can be used for broadcast. Algorithm Broadcast(BT(G), s, m) If source node s is a cluster-head, s invokes Eulerian(BT(G), s, m); otherwise s sends message m to its head u and u invokes Eulerian(BT(G), u, m). BT(G) consists of all cluster heads in G.. In the algorithm Broadcast(BT(G), s, m), whenever a node of BT(G) relay/transmit the message m, all its cluster-members receive m. Therefore, when the algorithm finishes, all nodes of G receive the message m. The algorithm can be completed in 2|BT(G)| rounds that is less than the twice of the clusters. Since the number of clusters is much smaller than the number of nodes in G, the broadcast can be completed very fast. In the following algorithm, source node s collects required data from all node in G by using CNet(G). Algorithm DataGathering(CNet(G), s) If s is a cluster head, s sets s to be u; otherwise s sets s’s head to be u and s sends message m to u. Then u invokes Eulerian(CNet(G), u, m), where m is the message for collecting required type of data. In the Eulerian tour, each node receives data from the previous node, adds its data together, and then transmits it to the next node. When the tour comes back to node u, u receives all required data. If u is not source node s, u sends the data to its member s. In the data gathering algorithm, the size of the data can be larger and larger. If the data has k packets, a node needs k rounds to transmit the data to the next one. Algorithm DataGathering(CNet(G), s) can be completed in 2g|CNet(G)| rounds, where g is the number of packets for the required data, and |CNet(G)| is the number of edges in CNet(G) that is n-1. Therefore, the algorithm can be completed in O(gn) rounds. The following algorithm finds a route for the communication between nodes s and v. Algorithm RouteBetweenNodes (BT(G), s, v) If s is a head, it sets u to be s; otherwise it sets u to be s’head. Node u invokes Eularlian(BT(G), u, m), where m is message “finding a route from u to v”. In the tour, when the message comes to a node w who is v or it has a member who is v, w stops the broadcast and sends a message “set the route from u to v” back to its parent x in the tour. When x received the massage, it sets w to be its successor in the route from u to v,

It is easy to see that above algorithm builds a route between s and v in O(|BT(G)|) rounds and the route is the shortest path between s and v in BT(G).

5. Performance Evaluation

Network Throughput

16 Number of packets per second

and then x sends the setting route message back to its parent y in the tour. When y received the massage, y will do the same thing to set up x to be its successor in the route. The route from u to v is completely set up when the route setting massage relays back to u. If u is not s, then u sends the message back to s, and s sets u to be its successor in the route from s to v.

14 12 10 8 6 4 2 0

We developed a simulator, Neptune 2.0, for testing the average performance of the proposed cluster-based architecture, MIMO protocol and routing algorithms. The field size is set to be 210m × 210m .The number of nodes varies from 400 to 800. The primary transmission range is 30m. It can be extended up to 120m by link-jumping operation. The energy at each node is initialized to be 843.75mJ, about 1 /( 4 × 100 ) of the energy in an AAAA battery. A packet is set to have 38 byte (payload 29 byte +

9.6

SISO (C) SISO (F)

Also, in order to see how a critical role of the MIMOaware cluster-based architecture plays, we compare the performance with that of a flat MIMO WSN by using the same MultiOptimal MAC protocol. The broadcast in a flat MIMO WSN is a simplified flooding: each node transmits the received packet only once. The simplification can avoid broadcast storm problem [7], however, it can not promise all nodes receive the broadcast packet. We only use statistics from the broadcasts whose success rate is larger than 0.95 (i.e., more than 95% nodes in G received the broadcast packet). The broadcast on our cluster-based architecture is deterministic and success rate is 100%. Because of the space limitation, we only show the results for the WSN with 800 nodes. In Fig 3, C and F represent the proposed cluster-based structure and the flat structure, respectively. As we described in Section 2, an M × M MIMO link can only speed-up the data rate up to M times even all the antenna arrays are used for spatial multiplexing gain. However, through Fig 3 we observe that our cross-layer design can improve network throughput more than M times; meanwhile, it significantly extends network lifetime and reduces network reconfiguration cost. Therefore, our design fully realized the potential and leveraged the advantage of MIMO technology from local links into the global network.

4X4 (C) 4X4 (F)

Energy Consumption

Energy (mJ)/ per packet

45000 40000 35000 30000 25000 20000 15000 10000 5000 0 9.6

19.2 Data Rate (kbps) SISO (C) 2X2 (C) SISO (F) 2X2 (F)

57.6 4X4 (C) 4X4 (F)

Network Lifetime 2000 1800 1600 Time (second)

header 9 byte). The average error rate is set to be 10 . Data rate can be 57.6kbps, 19.2 kbps and 9.6 kbps, respectively. We compare the network throughput, network lifetime and reconfiguration cost of SISO WSNs, 2×2 MIMO WSNs, and 4×4 MIMO WSNs by repeatedly broadcasting packets until the WSNs die (a WSN dies if some link on the route has a transmission distance larger than 120 m).

2X2 (C) 2X2 (F)

57.6

50000

3

−3

19.2 Data Rate (kbps)

1400 1200 1000 800 600 400 200 0 9.6

19.2 Data Rate (kbps) SISO (C) 2X2 (C) SISO (F) 2X2 (F)

57.6 4X4 (C) 4X4 (F)

Time for Network Reconfiguration in Whole Network Lifetime

50

5.

45

Time (second)

40 35 30

6.

25 20 15

7.

10 5 0 9.6

19.2 Data Rate (kbps) SISO (C) SISO (F)

57.6

2X2 (C) 2X2 (F)

8.

4X4 (C) 4X4 (F)

9. 70000

Energy for Network Reconfiguration in Whole Network Lifetime

10.

60000

Energy (mJ)

50000 40000

11.

30000 20000

12. 10000 0 9.6

19.2

57.6

13.

Data Rate (kbps) SISO (C) SISO (F)

2X2 (C) 2X2 (F)

4X4 (C) 4X4 (F)

Fig 3. Performance Evaluation of MIMO Sensor Networks

14.

References 1.

2.

3.

4.

B. S. Chlebus, L. Gasieniec, A. M. Gibbons, A. Pelc, and W. Rytter, “Deterministic broadcasting in ad hoc radio networks”, Distributed Computing 15, pp. 27-38, 2002. S. Cui, A. J. Goldsmith, “Energy-efficiency of MMIO and cooperative MOMO techniques in sensor networks,” IEEE Journal on Selected Areas in Communication, pp. 1-9, 2004 S. Cui, A. J. Goldsmith, A. Bahai, “Energy-constrained modulation opetimization,” IEEE Transactions A dynamic sensor network is a reconfigurable network which supports two functions: node-joining and nodeleaving. on Wireless Communications, pp. 2349-2360, Vol. 4, No. 5, 2005. W. Ke, P. Basu, A. Ayyash, T.E.C. Little, “Attribute

15.

16.

based hierarchical clustering in wireless sensor networks,” MCL Technical Report TR-03-24-2003, 2003. B. V. Nadimpalli, P. Mulukutla, R. Garimella, Srinivias MB, “Energy-aware routing in sensor networks using dual membership clusters and data highways,” TENCON 2004, 2004 IEEE Region 19 Conference Vol C, Issues 21-24, pp. 184-187, 2004. K. Nakano and S. Olariu, “Randomized initialization protocols for radio networks,” Handbook of Wireless Networks and Mobile Computing, pp.195-218, 2002. S. Y. Ni, Y. C. Chen, J. P. Sheu, “The broadcast storm problem in a mobile ad hoc network,” Proceedings of the 5th ACM/IEEE International Conference on Mobile Computing and Networking, pp. 151-162, 1999. S. PalChaudhuri, R. Kumar, R. G. Baraniuk, “Desing of adaptive overlays for multi-scale communication in sensor networks,” DCOSS, LNCS 3560, pp. 173-190, 2005. H. Su, X. Zhang, “Energy-efficient clustering system model and reconfiguration schemes for wireless sensor networks,” Annual Conference on Information Science and Systems, pp. 99-104, 2006. K. Sundaresan, R. Sivakumar, “A unified MAC layer framework for ad-hoc networks with smart antennas,” Proceedings of ACM International Symposium on Mobile Ad-Hoc Networking and Computing, 2005. K. Sundaresan, R. Sivakumar, “Routing in ad-hoc networks with MMO Links,” Proceedings of IEEE International Conference on Network Protocols, 2005. K. Sundaresan, R. Sivakumar, “Medium access control in ad-hoc networks with MIMO links: optimization considerations and algorithms,” IEEE Transactions on Mobile Computing, 2004. T. Tang, M. Park, R. W. Heath, Jr. and Scott M. Nettles, “A joint MIMO-ODF transceiver and MAC design for mobile ad hoc networking,” pp. 315-319, Proceedings of International workshop on Wireless Ad-Hoc Networks, 2004. J. Uchida, I. Muzahidul, Y. Katayama, W. Chen, K. Wada, “Construction and Maintenance of A Clusterbased Architecture for Dynamic Radio Networks,” 39th Hawaii International Conference on System Sciences, 2006. D. Wang, U. Tureli, “Joint MIMO-OFDM and MAC design for broadband miltihop ad hoc networks,” EURASIP Journal on Wireless Communication and Networking, pp. 1-9, Vol. 2006, 2006. L. Zheng, David N. C. Tse, “Diversity and multiplexing: a fundamental tradeoff in multiple-antenna channels,” IEEE Transactions on Information Theory, pp. 10731096, Vol. 49, No. 5, 2003.