Energy-Balanced Rate Assignment and Routing ...

1

Energy-Balanced Rate Assignment and Routing Protocol for Body Area Networks Nedal Ababneh, Nicholas Timmons, Jim Morrison and David Tracey WiSAR lab, School of Engineering Letterkenny Institute of Technology Port Road, Co. Donegal, Ireland E-mail:{nedal, nick, jim, david}@wisar.org

Abstract—Body Area Networks (BANs) are a key component of healthcare solutions, in which many different physiological parameters can be monitored and collected continuously by wearable wireless sensors with high accuracy and low power consumption. Abnormal events indicating coronary problems such as high heartbeat, body temperature or blood pressure can be detected and reported to a doctor for immediate medical actions. In BANs, quality of service is needed to provide reliable data communication over prioritized data streams coming from energy constrained sensors attached to, or possibly implanted in, patients. In this work, we focus on BAN for real-time data streaming applications, where the real-time nature of data streams is of critical importance for providing a useful and efficient sensorial feedback for the user while system lifetime should be maximized. Thus, bandwidth, throughput and energy efficiency of the communication protocol must be carefully optimized. In this paper, we present an Energy-Balanced Rate Assignment and Routing Protocol (EBRAR). EBRAR selects routes based on the residual energy, thus, instead of continuously routing data through an optimized (energy efficient) fixed path, the data is transported more intelligently and the burden of forwarding the data is more equally spread among the nodes. Another unique property of EBRAR is its ability to provide adaptive resource allocation. Our experimental results show that the proposed protocol performs well in terms of energy balance, network lifetime and utility. Index Terms—Body Area Networks, Balancing Energy Consumption, Energy-Aware Rate Assignment, Optimal, Lifetime.

I. I NTRODUCTION A Wireless Body Area Network (WBAN, widely simplifies as BAN) consists of lightweight, small-size, miniaturized, ultra-low-power, and intelligent monitoring wearable sensors deployed around or embedded in the human body [1], [2], [3]. It can be used to stream biological information from the human body and transmit it over a short distance to a control device (i.e., the sink), which in turn can relay the information wirelessly to an out-of-body remote server to get relevant diagnostic response. BAN sensor devices work autonomously and may interface with other wireless technologies, such as Wireless Local Area Network (WLAN), 3G network or Internet to reliably transmit data to a remote server [4] enabling a variety of medical and non-medical applications. One such important application is health monitoring, which has growing interest from hospitals as well as from healthy individuals who have potential health risks. BAN is particularly suitable for postoperative care in hospitals and for treatment of chronically ill or aged patients at home.

It is well-known that energy is an extremely critical resource for battery-powered BANs, thus making energy-efficient protocol design a key challenging problem [5]. Most of the existing energy-efficient routing protocols always choose a static optimal path (minimum energy path) to the sink to merely minimize energy consumption, which readily causes an unbalanced distribution of residual energy among sensor nodes since the energy at the nodes on the optimal path is quickly depleted. Such imbalance of energy consumption is definitely undesirable for the long-term health of the network as it reduces the network lifetime and causes network partition [6]. If the sensor nodes consume their energy more evenly, the connectivity between them and the sink can be maintained for a longer time, thus postponing the network partition. This more graceful degradation of the network connectivity extends its operable lifetime. In general, we consider an asymmetric architecture, in which most processing is done on a resource rich aggregator, minimizing the load on resource limited sensor nodes. All data in the BAN is generated at the nodes and continuously fed as a stream to the sink. Most of the data will flow to the sink, while a limited amount of control traffic will go in the other direction, which is neglected in this work. For these applications, it is essential to be able to reliably collect physiological readings from humans via sensor nodes. Such networks could benefit from Quality of Service (QoS) mechanisms that support prioritized data streams, especially when the channel is impaired by interference or fading [7]. For example, heart activity readings (e.g., Electrocardiogram (ECG) data) are often considered more important than body temperature readings, and hence can be assigned a higher priority in the system. QoS support is needed to ensure reliable data collection for high priority data streams and to dynamically re-allocate bandwidth as conditions change, especially when the effective channel bandwidth is reduced. In the standard system, when the effective throughput is scarce, data rate for all the nodes drops equally and the utility of the whole monitoring system drops significantly. In this case, to guarantee the QoS we need to re-allocate resources from lower priority streams to higher priority streams. Adaptive QoS resource allocation is thus needed to provide bandwidth guarantees, which are essential for reliable data collection in BANs. In this paper, we propose Energy-Balanced Rate Assignment

2

and Routing Protocol (EBRAR). In EBRAR, we estimate the utility for different nodes (data streams) and put some low utility nodes offline (i.e., stop data streaming) and thereby maintain high utility (the term utility is defined later in Section III-C2) of the monitoring system, utility of high priority nodes is particularly improved. We select the optimal data rate for each node with efficient link utilization such that the network utility is maximized without violating the QoS constraints. EBRAR constructs routing paths in terms of depth and residual energy. The goal of this basic approach is to force packets to move towards the sink through high energy nodes so as to protect the nodes with relatively low residual energy. The remainder of this paper is organized as follows. Section II summarizes related work. In Section III, we detail the EBRAR protocol. Simulation results are shown in Section IV. Finally, Section V concludes this paper and highlights our future work. II. R ELATED W ORK Many protocols have been previously proposed for traditional wireless sensor networks (WSNs) in the literature [5]. However, they are not always well suited to support a BAN because of the unique properties and application requirements of BAN. The differences between BAN and WSNs are discussed in details in [1], [3]. Recently, several protocols in the area of routing in BANs have been proposed [2], mostly validated either using theoretical analysis or simulation, and involve MAC or power control mechanisms. These protocols deal with limitations and unique characteristics of BANs, such as communication range or irregular traffic patterns. However, research on routing protocols for BANs is still at its infancy and can be categorized as follows: • Cluster Based Routing: This type of protocols uses clustering to reduce the number of direct transmissions to the remote base station. Hybrid Indirect Transmissions (HIT) [8] is a data gathering protocol based on LEACH [9], it combines clustering with forming chains. In doing so, the energy efficiency and as a consequence network lifetime is improved, specifically in sparse BAN. Reliability, however, is not considered. Another problem with HIT is the conflicting interaction between its communication routes and those desirable for the application. As a result, HIT requires more communication energy in a dense network. • Temperature Routing [10], [11], [12]: When considering wireless transmission around and in the body, important issues are radiation absorption and heating effects on the human body. In this approach, they try to route data away from high temperature areas due to focusing data communications, defined as hot spots. Temperature routing protocols can be sub-divided further into two categories: (a) adjusting radio transmission power and traffic control algorithms, and (b) balancing the communication over the sensor nodes, which is used by Thermal Aware Routing Algorithm (TARA), Least Temperature Routing (LTR) and Adaptive Least Temperature Routing (ALTR). Temperature routing protocols may suffer from short network lifetime, low delivery ratio, low reliability,

inefficient energy usage, increased routing overhead and high end-to-end delay, which are problematic for BAN. • Cross-layer protocols: Cross-layer design is a way to improve the efficiency of and interaction between the protocols in the network by combining two or more layers from the protocol stack. Researchers have developed cross-layer techniques for leveraging postural mobility for optimizing energy usage by designing collision-free medium access and routing protocols in [13], [14]. This allows nodes to save energy by sleeping when they are neither transmitting nor receiving. Another important issue in BAN, from an application point of view, is offering QoS. A virtual MAC is developed in BodyQoS [7] to make it radio-agnostic, allowing the protocol to schedule wireless resources and to provide adaptive resource scheduling using aggregator nodes. When the effective bandwidth of the wireless channel degrades due to RF interference or RF fading, the bandwidth is adaptively scheduled to meet the necessary QoS requirements. However, integration into existing MAC and especially routing protocols is not optimal [2]. When tackling the specific QoS requirements of BANs, integration with all layers should be strong. Given the difficult environment, where the path loss is high and as a result bandwidth is low, QoS protocols should expect and use all information about the network that is available. III. EBRAR: E NERGY-BALANCED J OINT R ATE A SSIGNMENT AND ROUTING P ROTOCOL In this section, we describe the design and implementation issues of the proposed protocol in depth. Our proposed solution consists of four phases described as follows: A. Topology Discovery One of the essential aspects of any BAN is the topology discovery. Several topology discovery and traffic information exchange algorithms have been proposed for BANs and traditional WSNs [15], [16], [17]. In our network scenario, the nodes use similar techniques to discover neighboring sensor nodes as well as to detect transient channel variations and topology changes (e.g., node or link failure, add new nodes, etc.). B. Tree Construction To achieve high network performance in BAN, the route construction within the network is a crucial task. Constructing a routing tree is equivalent to finding each node’s routing path towards the sink. To this end, a tree-based routing scheme easily allows traffic aggregation and ensures an optimal utilization of the limited available bandwidth. To maximize network operable lifetime and throughput capacity, we use the following metrics to construct the routing tree: 1) Energy: In this algorithm, we use node’s residual energy as a selection metric. The sum of the residual energy of all nodes that reside in the path of a node, say node a, towards the sink is termed the path-energy (PE), Penergy (a). For each node to choose its parent (selecting node’s parent is equivalent

3

to build routing paths), it chooses the neighbor which renders the largest path-energy among all other potential parents (i.e., neighbor nodes). In fact, the path-energy metric is the sum of the residual energy metric of the node’s predecessors. For example, assume sink-c-b-a is a path in the routing tree, and Penergy (a) is the sum of node’s residual energy through the path from node a whose parent node is b to the sink. Thus, Penergy (a) is calculated as follows:

Penergy (a) = energy(a) + energy(b) + energy(c)

(1)

We suppose all the nodes enter the network one by one. When a node, say node a, enters, all its neighbors are eligible to be a’s parent. In order to improve the network performance, node a selects a parent node, say node b, with largest value of Penergy (b). The parent node is then selected such that: P rnt(a) = maxi∈neighbor(a) Penergy (i)

(2)

2) Degree : We define for each node, a degree function which shows the total number of neighboring nodes. The sum of the degree functions of all nodes that reside in the path of a node, say node a, is termed the path-degree (PD), Pdegree (a). For each node to choose its parent, it chooses the neighbor which renders the smallest path-degree among all other neighbor nodes, as described above in the first algorithm (i.e., Energy). 3) Hop-Count: Using the hop-count metric enables rapid convergence, however, simply constructing a routing tree by shortest path is not sufficient to maximize network throughput as the link on the shortest path between two nodes may have bad quality [19]. In this paper, we show that using shortest path routing is not always harmful and may perform as good as other routing algorithms in some scenarios.

C. Rate Assignment We solve the rate assignment problem based on the tree constructed above in phase (III-B). After constructing the routing tree, it is up to the rate assignment phase to provide nodes with an appropriate data rate. The performance of the rate assignment model benefits from a well-structured routing tree, however, constructing optimal routing trees is beyond the scope of this paper. The rate assignment process is subject to the following QoS constraints: (1) a lower bound and upper bound on node data rate, (2) an upper bound on the uplink physical capacity, and (3) an upper bound on channel capacity at each relay node in the tree. In this paper, for simplicity, we consider a single-radio, multi-channel BAN network, where interfering wireless links operate on different channels, enabling multiple parallel transmissions as described in [18]. The problem therefore consists of increasing the number of accepted high priority nodes in the routing tree(s) at higher data rate while balancing out energy consumption (choosing better energy paths to forward the traffic).

1) The Problem Definition : The proposed BAN network is modeled as undirected graph G = (V, E), called a connectivity graph, where V is the set of vertices representing the nodes in the network, and is composed of a group of nodes denoted as Vn and one or more sinks denoted as Vs . E is the set of edges that represents the communication network topology, edge(vi , vj ) ∈ E iff vi , vj are within each other’s communication range. Hence, nodes form a multi-hop ad-hoc network among themselves to relay traffic to the sink(s). Also, each edge (vi , vj ) has a physical capacity Lij , which represents the maximum amount of traffic that could pass through this particular link, and each node v ∈ Vn represents a node, and all nodes in the network are assumed to work on the same fixed transmission power with circular transmission range RT . At any given time, a node may either transmit or listen to a single wireless channel, with channel capacity C. The resulting routing tree T = (V ′ , E ′ ) is a subgraph of G, where E ′ represents the communication links in the final tree, and V ′ is the set of nodes and sinks included in the resulting routing tree. Ui represents the utility of streaming data from node v i . W min is the minimum acceptable data rate by node vi , and W max represents the desired data rate (i.e., the maximum possible data rate of a node). P i represents the priority (i.e., either high or low) of the node vi based on the equipped sensor type, while ri is the current rate (i.e., ratio to W max ) of received data sent from node v i . The utility of streaming data decreases with decreasing received data rate and available energy level, and the utility becomes insignificant beyond a certain value (i.e., Rmin and E min in this case). In cases where the data rate and residual energy level for a given node falls below the acceptable level (i.e., Rmin and E min ), we propose to stop live data streaming and store the data from this offline node(s) so that it can be transmitted once the available bandwidth permits. This will prevent flooding the channel and reducing the data quality below the acceptable level for data streams from all other nodes. 2) Integer Linear Program Formulation: Let zi be a 0-1 integer variable for each node vi ∈ Vn , such that zi = 1 if the node vi is included (i.e., accepted) in the resulting tree vi ∈ Vn′ . Let ri be a positive real variable for each vi ∈ Vn , representing the effective data rate of vi such that ri = 0 if vi is not included in the resulting routing tree (i.e., z i = 0). Let Xij be a 0-1 integer parameter for each edge (vi , vj ) ∈ E, Xij = 1 if the edge is included the resulting tree (i.e., edge (vi , vj ) ∈ E ′ ). Furthermore, let yij be a positive integer variable for each edge (vi , vj ) ∈ E ′ , showing the amount of data transmitted from node vi to node vj (i.e., uplink effective data rate), the receiver could be a sink node. The ILP for the rate assignment problem can thus be stated as follows: Objective (Utility) function: P max i∈Vn (zi × Pi × Umin + P e ∀j∈V :(i,j)∈E zi × (ei − (Emin × Xij )) × Ustep + r Pi × (ri − (Rmin × zi )) × Ustep )

The first term of the utility function is the minimum utility for each node in the network, the second and third terms

4

X X ∀j∈V :(i,j)∈E

yji 6 Ci , ∀i ∈ V

(4)

∀j∈Vn :(j,i)∈E

yij −

X

3

1

Reset Topology discovery

200

Continuous data streaming

Reset

2

1

Determine and Broadcast data rate and node status (final routing tree)

(3)

0

Construct initial routing tree

yij 6 Lij × Xij , ∀i ∈ Vn , ∀j ∈ V : (i, j) ∈ E

200 seconds application cycle

Topology discovery

denote the utility evolution with energy and rate, respectively. e and the Multiplying the second term by the coefficient Ustep r third term by the coefficientUstep ensures utility evolution with energy and rate. Also, multiplying each term by zi guarantees the consideration of the included vertices nodes in the resulting tree only. Each accepted node vi ∈ Vn′ is assigned rate ri > Rmin . The coefficient U min is the minimum utility of each accepted node vi , and must be set to any positive value greater than zero (i.e., U min = 1 in this work). Constraints:

Figure 1: The different phases of the network cycle with EBRAR.

yji = ri × Wmax , ∀i ∈ Vn (5)

∀j∈V :(j,i)∈E

zi > ri , ∀i ∈ Vn

(6)

ri > Rmin × zi , ∀i ∈ Vn

(7)

Ei > Emin × Xij , ∀i ∈ Vn , ∀j ∈ V : (i, j) ∈ E

(8)

Constraint (3) ensures that the uplink effective rate of each included edge in the resulting tree is bounded by the maximum physical link capacity. Constraint (4) provides an upper bound (i.e., the cell capacity) on the relay load constraint, it ensures that the incoming flow is always less than cell capacity Ci . Constraint (5) is for flow conservation. It implies that the difference between the outgoing traffic and the incoming traffic at node vi is the volume of traffic generated by node vi itself. Since all data flows are originated from nodes and do not return to the nodes, it will not lead to cycles in our solution. All data flows will eventually reach the sink. Constraint (6) ensures that node data rate is assigned to accepted nodes in the final routing tree only, i.e., nodes not included in the resulting tree have rate equal to zero. Constraint (7) ensures that each accepted node vi in the resulting tree has to be assigned rate ri > Rmin , this is to satisfy the QoS requirements. Finally, Constraint (8) ensures that available energy for each accepted node vi in the resulting tree has to be > Emin , as Emin is the residual energy value where a node is still able to send/receive messages properly. D. EBRAR state transitions In EBRAR, nodes are in one of three states: sleep, listen and active. Initially, all nodes start out in the listen state. When in the listen state, a node turns on its radio(s) and exchanges discovery messages to gather information about its neighborhood. Neighborhood information gathered across the network is also sent to the sink, so that the sink will have a knowledge about the topology and network characteristics. In addition, when a node enters listen state, it sets up a timer Tlisten . When the timer fires, the sink calculates rates to the nodes in the network and floods the allocated rate information in ANNOUNCEMENT message. Upon receiving this ANNOUNCEMENT message, a node finds out its allocated data rate (i.e., bandwidth) as well as its status information (i.e.,

whether active (on) or sleep (off)), that is if the rate is zero then the node moves to the sleep state and to transit to active state otherwise (rate r must be > Rmin ). When transiting to sleep state, a node sets the sleep timer Tsleep and turns off its radio(s). A node in sleep state returns to the listen state after an application dependent sleep time Tsleep . If the node is accepted as a traffic source (i.e., assigned rate r > Rmin ), the node enters the active state and sets a timeout value Tactive to define how long this node can stay in active state. When Tactive expires, the node returns to listen state. While active, the node continuously sends streaming data and, possibly, forwards other nodes’ data up towards the sink. The choice of the timers’ values (i.e., Tsleep and Tactive ) in EBRAR is application dependent and these timers have the same value (i.e., round time) in order to synchronize the network, Figure 1 illustrates the application cycle of EBRAR. Setting a small value for these timers implies rotating the roles of the nodes often and ensures prolonged network lifetime and allows fast reaction to topology changes and network conditions while causing more packet loss. Packets destined to the active nodes may get lost because these nodes transit to sleep state and the routing protocol still consider them as active nodes (i.e., traffic forwarder). Therefore, if the application requires a high delivery rate we should set higher value for these timers and reset the sensor nodes less often. This stabilizes the network and improves system performance. In this case we sacrifice the longer network lifetime to get better delivery rate. Since the nodes have a finite amount of energy represented by battery capacity, these energy savings obtained from turning off unaccepted traffic source nodes in the network directly correspond to the same relative increase in the lifetime of a node, which ultimately results in a prolonged lifetime of the BAN. E. Load balancing energy usage EBRAR employs a load balancing strategy so that all nodes remain up and running together for as long as possible. To ensure fairness, the role of active nodes is rotated periodically to ensure energy-balanced operations. EBRAR uses the following load balancing strategy (as in ECTC [20], LEACH [9]

5

and GAF [21]). After a node remains in active state for time Tactive , it changes its state to listen state to give a chance to other nodes in the neighborhood to become active. Recall that, nodes are ranked according to their importance (type of equipped sensors) and remaining energy levels. When the active node changes its state to discovery, it is more likely that it has less remaining energy in its battery than its neighbors because presumably the neighbor nodes were in the sleep state conserving energy during the node’s active time (i.e., active time is equal to round time as nodes keep their state for the round time period). Consequently, the node that was active is less likely to remain active after the listening phase. It is worth noting that another reason why we chose to run EBRAR in rounds is to react to topology changes (e.g., node or link failure, adding new nodes, etc.) and dynamic network conditions. IV. P ERFORMANCE E VALUATION A. Simulation Environment For the evaluation, we have conducted an extensive set of experiments using a VC++ coded simulator. To be accurate, we solve the integer linear program presented in Section III-C by using AMPL and CPLEX. The wireless channel capacity is C = 2 Mbps, and link capacity L = 1 Mbps. The minimum acceptable data rate generated by each node Rmin is 128 Kbps and Rmax is fixed at 512 kbps. The minimum utility of each e r accepted node U min = 1 and Ustep = Ustep = 1/5. Our goal of implementing EBRAR protocol is to minimize the total energy spent in the network to communicate the information gathered by sensor nodes to the information-processing center (i.e., the sink), while achieving higher network utility. To evaluate EBRAR, we simulated 16 node network illustrated in Figure 2(a), using different high priority nodes in each run. To force the packets to go through multi-hop on route to the sink, we use small radio range with average node degree set to 7. All nodes are assumed to have the same radio range. Two nodes are connected if they are within each other’s transmission range. Unless otherwise stated, all the following investigations adopt these values. For all simulation results in this paper, each point in the plots is averaged over 10 runs, and the simulation lasts for 2000 s. B. Simulation Results and Discussions Results are presented from simulations of the three different routing algorithms proposed in Section III-B, namely best Path-Energy routing (PE), minimum Path-Degree routing (PD) and Shortest-Path routing (SP). Varying number of high priority nodes and sink location, by mean of simulations, we are interested in the following issues: 1) How many traffic sources are accepted in the network and how does that affect the network utility? 2) How is the distribution of residual energy of nodes? How much can the network’s operational lifetime be extended? 3) How the sink placement affects the network energy behavior and network performance? One of the commonly made assumptions in BAN literature is that the sink node

(a)

(b)

(c)

(d)

Figure 2: Example of BAN optimal routing tree formation based on (b) EBRAR-SP; (c) EBRAR-PD and (d) EBRARPE. A red filled circle represents a high priority node, a blue filled circle represents a regular node and a green filled square represents the sink. Solid line indicates communication link, and (a) presents nodes positions on the body and original communication graph, each node is sending a Constant Bit Rate (CBR) stream to the sink with radios capable of transmitting up to 0.5 Mbps.

is located at the waist [2], [1], [3]. To investigate how the sink placement affects the network performance, we repositioned the sink to the ankle. To get initial answers to these questions, we focused our simulations on the tree construction and rate assignment process, which are mainly responsible for network utility and energy consumption. 1) Utility: To study the impact of the number of high priority nodes on the network performance we vary the number of high priority nodes from 2 to 14. The utility of a data transmission rate below Rmin (i.e., 128 kbps) is considered to be insignificant, hence, the node is putted offline. Utility for each offline node is assumed to be zero. Other settings are the same as above. We can clearly see in Figure 3 that all schemes experience a linear utility increase with the increase of high priority nodes. This is because as more high priority nodes are deployed, all proposed schemes will try to accommodate as many of them as possible at the best possible data rate, which results in a better utility. It is interesting to note that, our protocol consistently yields a better performance when the sink is located at ankle than at waist. This is due to more energy conservation being achieved (selecting better energy paths) as depicted in Figures 4(a) and (b) resulting in higher network utility. EBRAR-PE (sink at ankle) experiences a drop in its performance when number of high priority nodes is over 10 as it is not able to accept as many high priority (and even low priority) nodes in the network as the other algorithms, see Figure 7. On the other hand, the three routing algorithms achieve a more or less equal utility when the sink is placed at waist. This is because the three algorithms select routes with minimal possible traffic congested links (as relay load is considered in selecting nodes rate and status in the

6

1

3200

EBRAR‐SP

Normalized residual energy

0.9 3000

Utility

2800

EBRAR‐SP‐waist EBRAR‐PD‐waist

2600

EBRAR‐PE‐waist 2400 EBRAR‐SP‐ankle 2200

EBRAR‐PD‐ankle

2000

EBRAR‐PE‐ankle 2

4

6

8

10

12

0.8

EBRAR‐PD

0.7

EBRAR‐PE

0.6 0.5 0.4 0.3 0.2 0.1 0

14

1

2

3

4

5

6

7

Number of high priority nodes

8

9

10 11 12 13 14 15

Node ID

(a) Sink at waist

Figure 3: Performance comparison of network utility as a function of number of high priority nodes. Normalized residual energy

model), thus the packets are dispersed widely and concurrent transmission can be fully utilized. In addition, we can see that although sometimes the SP routing does not necessarily indicate the higher network throughput, which is true in most cases, it performs similar to PD in terms of network utility in this experiment. 2) Load Distribution: In order to observe how well EBRAR promotes load balancing among the nodes, we ran a simulation for 2000 s. At the outset, each node had 2 J battery energy. Figure 4 shows relative residual energy at the end of simulation across nodes of EBRAR-SP, -PD and -PE, respectively. The plots are separated to avoid clutter. It is obvious that EBRARPE achieves the best performance by maintaining a higher value of residual energy. As expected, EBRAR-SP and -PD did not perform as well mainly because it does not consider energy level when constructing the routing tree even though it is taken into account in the rate assignment and final routing tree construction step (i.e., Section III-C). It can also be observed in Figures 4(a) and (b) that the proposed protocol is able to achieve better load distribution when sink is placed at waist, as shorter (fewer hops) and yet energy efficient paths are available to route data to the sink. This confirms that depth has a negative impact on network energy distribution as the sink will have smaller node degree and network congestion may occur in this case. It is interesting to note that when sink is placed at ankle, nodes have more residual energy, this is because the routing tree size (i.e., number of nodes that forms the routing backbone) is smaller as depicted in Figures 6(a) and (b). 3) Network Lifetime: In this experiment, we use simulations to quantitatively find how network lifetime changes with different number of high priority nodes and with sink location. We changed high priority nodes number from 2 to 14. In order to have enough time to drain node energy, we extend the simulation time to 5000 s. Our goal is to extend the network lifetime as a whole through energy conservation. In the simulations, streaming traffic is constant. We therefore can measure how long the network information by periodically monitoring the fraction of survived nodes for every fixed time frame, i.e., round time. If the monitored fraction of survived nodes drops after time t, we can say that the network lifetime ends at time t. In our simulations, we found that network

1 0.9

EBRAR‐SP

0.8

EBRAR‐PD

0.7

EBRAR‐PE

0.6 0.5 0.4 0.3 0.2 0.1 0 0

1

2

3

4

5

6

7

8

9

10 11 12 13 14

Node ID

(b) Sink at ankle

Figure 4: Residual energy distribution of nodes after 2000 s simulation of EBRAR: The nodes are initially equipped with 2 J battery energy.

throughput drops dramatically when fraction of surviving nodes is less than 70 %, and that is the reason to chose this value as t. It is important to note that the decrease of the number of surviving nodes affects routing fidelity. Normally, the fewer surviving nodes, the worse the packet delivery ratio. In Figure 5, the improvement gained through EBRAR-PE is further exemplified by the network lifetime graph. For this investigation, we set initial battery energy at only 1 J to let the nodes die sooner. This, however, does not change the pattern of behavior of the studied algorithms. It is evident that EBRAR-PE exhibits the longest lifetime with having more nodes remaining fully functional. It is found that EBRAR promotes good load balancing across the entire network to sustain the network for its longest possible use. As shown in Figure 5(b), the high priority nodes density has a negative effect on network lifetime. Initial increases in the number of high priority nodes significantly reduce the network lifetime. This is mainly due to the fact that EBRAR will try to accept as many high priority nodes in the network as possible and keep them up all the time until their energy is drained. Thus, any further introduction of nodes into the network has a negative impact on the network lifetime. However, because of the clever energy balancing strategy in EBRAR, nodes in the network consume energy evenly (and thus die together), which ultimately results in a prolonged lifetime of the BAN. Figures 5(a) and (b) also indicate that placing the sink in the center of the body (i.e., the waist) has smaller improvement in

7

5000 EBRAR‐PD

4100

EBRAR‐PE

3800

15

15

13

13

11

11

3500

Routing tree size

Network lifetime (s)

4400

Routing tree size

EBRAR‐SP

4700

9 7 EBRAR‐SP

5

EBRAR‐PD

3

9 7 EBRAR‐SP

5

EBRAR‐PD

3

EBRAR‐PE

3200

EBRAR‐PE

1

1 2

4

6

2900

8

10

12

2

14

4

6


8

10

12

14


2600

(a) Sink at waist

2300 2000 2

4

6

8

10

12

(b) Sink at ankle

Figure 6: Routing tree size as a function of number of high priority nodes.

14


Network lifetime (s)

5000 4700

EBRAR‐SP

4400

EBRAR‐PD

4100

EBRAR‐PE

3800

15 13 11 9 7 EBRAR‐SP

5

EBRAR‐PD

3

EBRAR‐PE 1 2

3500

Number of accepted traffic sources in the network

Number of accepted traffic sources in the network

(a) Sink at waist

4

6

8

10

12

14

15 13 11 9 7 EBRAR‐SP

5

EBRAR‐PD

3

EBRAR‐PE 1 2


4

6

8

10

12

14


3200

(a) Sink at waist

2900 2600

(b) Sink at ankle

Figure 7: Number of accepted traffic sources as a function of number of high priority nodes.

2300 2000 2

4

6

8

10

12

14


(b) Sink at ankle

Figure 5: Performance comparison of network lifetime as a function of number of high priority nodes.

the lifetime especially at higher level, consistent with the result reported in Figure 4. In Figure 5(a), all algorithms experience an increase in their lifetime when only 4 high priority nodes are deployed in the network as EBRAR is able to accept all four of them at highest possible rate (i.e., 512 Kbps), and the routing tree size is 4 nodes only as the nodes have direct link with the sink, see Figures 6 and 7. 4) Routing Tree Size : Figure 6 shows the impact of the number of high priority nodes on the total size of the routing trees (i.e., total number of nodes in the resulting routing backbone) that each algorithm obtains. We note that all curves depict a larger routing tree for larger number of high priority nodes, which suggests that when increasing number of high priority nodes in the network the proposed protocol tries to accept as many nodes in the network as possible and thus increases the routing tree size. Tree size influences the total consumed energy and thus network operable lifetime. The larger the tree is the more energy consumed in the network. 5) Accepted Traffic Sources : Figures 7(a) and (b) depict the number of accepted nodes in each routing algorithm for different number of high priority nodes. As can be seen from the graphs, the proposed algorithms react smoothly and consistently as the number of high priority nodes increases. It is clear that, as the number of high priority nodes increases, all schemes experience an increase in the number of accepted nodes. In other words, as we increase the number high priority nodes deployed in the network, the proposed solutions will try

to accept as many of them as possible with best data rate as discussed above. When there is only 2 high priority nodes deployed in the network, EBRAR is able to accept them at the highest data rate and, furthermore, accept a number of low priority nodes at the minimum data rate, see Figure 8. 6) Quality of Received Data: The influence of number of high priority nodes on the data quality of the accepted high priority nodes is also evaluated. Figures 8(a)-(f) show the effect of different number of high priority nodes on the streaming data quality coming from the accepted high priority nodes for different sink placement, while the sensory data is the actual information transferred across the wireless links. We classified the data quality based on the allocated data rate into three levels: poor (128 - 256 kbps), good (257 - 384 kbps) and excellent (385 - 512 kbps). From the figures, we can see that the percentage of high priority nodes accepted at better quality, and thus higher utility, decreases with the the number of high priority nodes in all solutions. However, the three algorithms consistently perform better when the sink is placed at waist. This is because more routes are available in the network with possibly better capacity, and EBRAR (regardless the adopted routing algorithm) is able to select the best capacity path (advantageous route) and thus the best node data rates are allocated to high priority nodes. On the other hand, the performance is degraded when the sink is placed at ankle, this is because of the presence of traffic congestion. This is especially true for nodes closer to the sink as they are continuously used to forward the whole network traffic towards the sink, where relay capacity of each node is upper bounded by channel capacity. Thus, for small number of high priority nodes in the network, all high priority nodes will be accepted at highest possible data rate. On the other hand, when the number of high priority nodes is increased, more high priority

8

errors. Future work will be necessary to characterize the performance under non-ideal channels. In addition, since the running time required to obtain the optimal solution for the rate assignment problem increases exponentially with the number of nodes in the network, optimal results are only possible when the problem scale is small. Thus, we are interested in designing Number of HP nodes a near optimal rate assignment algorithm to scale for, possibly, (b) EBRAR-SP, sink placed at ankle larger network size. 1

Poor

Good

Excellent

Accpeted HP nodes (%)


1 0.8 0.6 0.4 0.2 0 4

6

8

10

12

14

Excellent

0.6 0.4 0.2

2

Number of HP nodes

(a) EBRAR-SP, sink placed at waist

4

6

8

10

Good

Excellent


1

Poor

12

14

1


Good

0

2

0.8

Poor

0.8

Poor

Good

Excellent

R EFERENCES

0.8



[1] B. Latre, B. Braem, I. Moerman, C. Blondia, and P. Demeester, “A 0.6 0.6 survey on wireless body area networks,” Wireless Networks, vol. 17, pp. 0.4 0.4 1–18, January 2011. 0.2 0.2 [2] S. Ullah, H. Higgins, B. Braem, B. Latre, C. Blondia, I. Moerman, 0 0 S. Saleem, Z. Rahman, and K. Kwak, “A comprehensive survey of 2 4 6 8 10 12 14 2 4 6 8 10 12 14 wireless body area networks,” Journal of Medical Systems, pp. 1–30, Number of HP nodes Number of HP nodes 2010. (c) EBRAR-PD, sink placed at waist (d) EBRAR-PD, sink placed at ankle [3] M. Chen, S. Gonzalez, A. Vasilakos, H. Cao, and V. C. M. Leung, “Body area networks: A survey,” Mobile Networks and Applications, vol. 15, 1 1 2010. Poor Good Excellent Poor Good Excellent [4] S. Ullah and K. S. Kwak, “An ultra low-power and traffic-adaptive 0.8 0.8 medium access control protocol for wireless body area network,” Journal 0.6 0.6 of Medical Systems, vol. 34, 2010. 0.4 0.4 [5] I. F. Akyildiz, W. Su, Y. Sankarasubramaniam, and E. Cayirci, “A survey 0.2 0.2 on sensor networks,” IEEE Communications Magazine, vol. 40, no. 8, 0 0 pp. 102–114, 2002. 2 4 6 8 10 12 14 2 4 6 8 10 12 14 [6] F. Ren, J. Zhang, T. He, C. Lin, and S. Ren, “Ebrp: Energy-balanced Number of HP nodes Number of HP nodes routing protocol for data gathering in wireless sensor networks,” IEEE Transactions on Parallel and Distributed Systems, no. 99, 2011. (e) EBRAR-PE, sink placed at waist (f) EBRAR-PE, sink placed at ankle [7] G. Zhou, J. Lu, C.-Y. Wan, M. D. Yarvis, and J. A. Stankovic, “Bodyqos: Adaptive and radio-agnostic qos for body sensor networks,” in Proc. of Figure 8: Received data quality of accepted high priority nodes the IEEE INFOCOM, 2008, pp. 565–573. as a function of number of high priority (HP) nodes. [8] M. Moh, B. J. Culpepper, L. Dung, T.-S. Moh, T. Hamada, and C.-F. Su, “On data gathering protocols for in-body biomedical sensor networks,” in Proc. of IEEE GLOBECOM, 2005. [9] W. B. Heinzelman, A. P. Chandrakasan, and H. Balakrishnan, “An nodes will be accepted at lower data rate. application-specific protocol architecture for wireless microsensor networks,” vol. 1, no. 4, pp. 660–670, 2002. V. C ONCLUSION AND F UTURE WORK [10] H. Ren and M. Q.-H. Meng, “Rate control to reduce bioeffects in wireless biomedical sensor networks,” in Proc. of MobiQuitous, 2006. This paper proposes a joint rate assignment and routing protocol, termed EBRAR, that performs real-time monitoring [11] Q. Tang, N. Tummala, S. K. S. Gupta, and L. Schwiebert, “Communication scheduling to minimize thermal effects of implanted biosensor of complex conditions on streaming data from various body networks in homogeneous tissue,” IEEE Trans. Biomed. Eng., vol. 52, no. 7, pp. 1285–1294, 2005. sensors within a Body Area Network (BAN). Our solution is built on the notion that since the available throughput in the [12] A. Bag and M. A. Bassiouni, “Energy efficient thermal aware routing algorithms for embedded biomedical sensor networks,” in Proc. of IEEE BAN may not be sufficient to accommodate the streaming data MASS, 2006. from all sensor nodes, some low priority streams (nodes) may [13] B. Braem, B. Latre, I. Moerman, C. Blondia, and P. Demeester, “The wireless autonomous spanning tree protocol for multihop wireless body be put offline (stop streaming) to conserve system bandwidth. area networks,” in Proc. of MobiQuitous, 2006. We proposed a solution that calculates the utility of different [14] B. Latre, B. Braem, I. Moerman, C. Blondia, E. Reusens, W. Joseph, and P. Demeester, “A low-delay protocol for multihop wireless body streams taking into account stream priority and number of area networks,” in Proc. of MobiQuitous, 2007. flows passing through each node, and based on the estimated [15] I. Georgeff, “A distributed topology discovery algorithm for wireless utility, decides which stream(s) to put offline so that overall sensor networks,” The University of Western Australia, Tech. Rep., 2004. utility of the whole monitoring system is improved. BANs [16] T. Watteyne, I. Augé-Blum, M. Dohler, and D. Barthel, “Anybody: a self-organization protocol for body area networks,” in Proc. of BodyNets, for healthcare put more emphasis on adaptation to changes in 2007. context and application requirement, thus, EBRAR adopts an [17] B. Deb, S. Bhatnagar, and B. Nath, “A topology discovery algorithm for sensor networks with applications to network management,” in Proc. of adaptive resource allocation policy to continuously meet QoS IEEE Sensors, 2002. requirements. Furthermore, the proposed solution no longer [18] B. Aoun, R. Boutaba, and G. W. Kenward, “Gateway placement opticonsiders the naive approach of only sending data to the mization in wireless mesh networks with qos constraints,” IEEE JSAC, vol. 24, pp. 2127–2136, 2006. direct neighbors of a node. Instead, the data is transported more intelligently and the burden of forwarding the data [19] D. S. J. De Couto, D. Aguayo, B. A. Chambers, and R. Morris, “Performance of multihop wireless networks: Shortest path is not enough,” in is more equally spread among the nodes. It is shown in Proc. of HotNets-I. Princeton, New Jersey: ACM SIGCOMM, October 2002. the paper that the EBRAR is energy efficient for streaming communication while balancing energy consumption across [20] N. Ababneh, A. Viglas, H. Labiod, and N. Boukhatem, “Ectc: Energy efficient topology control algorithm for wireless sensor networks,” in the BAN guarantees longer lifetime. Proc. of IEEE WOWMOM, 2009. The results presented in this paper indicate the the gains [21] Y. Xu, J. Heidemann, and D. Estrin, “Geography-informed energy conservation for ad hoc routing,” in Proc. of ACM MOBICOM, 2001. achievable under ideal channel conditions with no transmission