Energy-Efficient Bridge Detection Algorithms for Wireless Sensor Networks

A bridge is a critical edge whose fault disables the data delivery of a WSN component. Because of this, it is important to detect bridges and take preventions before they are corrupted. Since WSNs are battery powered, protocols running on WSN should be energy efficient. In this paper, we propose two distributed energy-efficient bridge detection algorithms for WSNs. The first algorithm is the improved version of Pritchard's algorithm where two phases are merged into a single phase and radio broadcast communication is used instead of unicast in order to remove a downcast operation and remove extra message headers. The second algorithm runs proposed rules on 2-hop neighborhoods of each node and tries to detect all bridges in a breadth-first search (BFS) execution session using O(N) messages with O ( Δ ( log 2 ( N ) ) ) bits where N is the node count and Δ is the maximum node degree. Since BFS is a natural routing algorithm for WSNs, the second algorithm achieves both routing and bridge detections. If the second proposed algorithm is not able to to classify all edges within the BFS phase, improved version of Turau's algorithm is executed as the second phase. We show the operation of the algorithms, analyze them, and provide extensive simulation results on TOSSIM environment. We compare our proposed algorithms with the other bridge detection algorithms and show that our proposed algorithms provide less resource consumption. The energy saving of our algorithms is up to 4.3 times, while it takes less time in most of the situations.


Introduction
Rapid developments in the last decade in wireless and hardware technologies have created low-cost, low-power multifunctional miniature wireless devices.These devices have enabled the use of wireless sensor networks (WSNs) [1].WSNs do not have any fixed infrastructure where hundreds even thousands of sensor nodes cooperate to implement a distributed application.WSNs can be used in various applications including habitat monitoring [1], military [2], and smart home applications [3].Energy consumption of a WSN should be reduced to maximize the application lifetime since sensor nodes are battery powered.The radio component of a sensor node is the dominant energy consumer part.
WSNs are increasingly being used in challenged environments such as underground mines, tunnels, oceans, and the outer space.Wireless communication in challenged environments have channel (edge) failures, mainly as a consequence of direct impact of physical world.In addition to energy constraints and wireless communication problems, tiny sensor motes are prone to failures.In both type of these failures, sensor network can continue its operation without a serious bad effect.On the other side, some edges can have critical tasks in routing operation.These edges are called bridges (cut edges) which its removal breaks connectivity of the network.Bridge detection is an important research area for different types of networks [4][5][6][7][8][9].After bridges are detected, various solutions can be applied [10] in order to neutralize bridges.
A WSN can be modeled with an undirected graph  = (, ), where  is the set of nodes and  is the set of edges.BFS and depth-first search (DFS) are fundamental graph traversal algorithms.DFS starts from sink node, and searches deeper in the graph if possible [11].Like DFS, BFS starts from sink node and search proceeds in a breadth-first manner [11].The edges chosen for BFS are called tree edges, and other edges are called cross edges.In BFS, tree level of sink node is 0, the levels of neighbors of sink node are 1, and levels of other International Journal of Distributed Sensor Networks nodes are their shortest distance to the sink node.From this property, BFS can be used to construct shortest path trees rooted at the sink node.BFS is used widely in sensor network for various purposes such routing and localization [12][13][14][15][16]. Besides, BFS can be modified to detect bridges.In this study, we proposed bridge detection algorithms that use BFS as the basis algorithm.
Distributed bridge detection algorithms proposed so far have some important disadvantages.Although distributed DFS based algorithms [8,[17][18][19][20] are simple and efficient for bridge detection, DFS based applications for WSN are rare in practice.Since then, DFS should be implemented as a separate service where this would be an extra load for batterypowered sensor nodes.Although BFS provides an efficient routing infrastructure for sensor networks, BFS based bridge detection algorithms lack some important design criterions.The message size of the Milic's BFS based algorithm [9] can be as large as O( log 2 ()).Pritchard's BFS based algorithm has two extra phases [21].Because of these reasons, these algorithms may consume significant energy.Regarding these deficiencies, we propose two distributed localization-free and energy-efficient bridge detection algorithms for sensor networks.The contributions of this paper are listed below.
(i) We propose an improved version of Pritchard's algorithm (I-PRITCHARD).In I-PRITCHARD algorithm, radio broadcast communication is used instead of unicast communication, and a downcast operation in Pritchard's algorithm is removed.I-PRITCHARD can be completed within 2 phases; on the other hand, Pritchard's algorithm needs 3 phases.Because of these reasons, I-PRITCHARD consumes less energy than the Pritchard's algorithm.
(ii) We propose the energy-efficient bridge detection algorithm (ENBRIDGE) by using 2-hop neighborhood knowledge and radio broadcast communication.The algorithm uses O() messages with O(Δ log 2 ()) bits, where  is the node count and Δ is the maximum node degree.Besides, the algorithm runs just one phase, that is, integrated with the BFS at the best case.This is a significant improvement over Milic's algorithm.In the worst case, the algorithm runs an improved DFS algorithm where message complexity and message size are asymptotically same with the first phase.ENBRIDGE outperforms its counterparts in the simulations.
The rest of this paper is organized as follows.In Section 2, the network model and bridge detection problem are described, and the related work is surveyed in Section 3. The proposed algorithms are described in Section 4. The detailed analysis of the algorithms are given in Section 5, and the results of performance tests are presented in Section 6. Conclusions are given in Section 7.

Background
In this section, we introduce the network model and the bridge detection problem.

Network Model.
The following assumptions are made about the network as in [8,22].
(i) Each node has distinct node id.
(ii) The nodes are stationary.
(iii) Links between nodes are symmetric.Thus, if there is a link from  to V, there exists a reverse link from V to .
(iv) Nodes do not know their positions.They are not equipped with a position tracker like a GPS receiver.
(v) All nodes are equal in terms of processing capabilities, radio, battery, and memory.
(vi) Each node can send radio broadcast messages to its immediate neighbors in its transmission range.
(vii) Nodes are time synchronized.This can be achieved by implementing a time synchronization protocol as proposed in [23].
Based on these assumptions, the network may be modeled as an undirected graph (, ), where  is the set of vertices and  is the set of the edges.An example of undirected graph model is depicted in Figure 1(a), where the transmission ranges of the nodes are shown with dashed circles.

The Bridge Detection Problem.
Bridges can connect any two nodes on the network.A bridge can connect a leaf node to its parent or connect a whole component to lower layers.An example of sensor network is depicted in Figure 1(b).There are 10 nodes, where node ids are written inside of each node, and the communication edges are shown with the solid lines in this Figure .The edges (, ), (, ), and (, ) are bridges which are depicted with bold lines.If the edge (, ) fails then leaf node  cannot transmit its packet to the lower layers.If the edge (, ) fails then both nodes  and node  cannot relay their data packets where 20% of the network can not transmit its data to the sink node.Node  is the parent of a component consisting of the node , the node , the node , and the node .If the edge (, ) fails, 50% of the total nodes cannot send their data to the sink node.In this paper, our focus is energy-efficient bridge detection for sensor networks, so our objectives are listed below.
(1) Since message transmission is the dominant factor of energy consumption, the bridge detection algorithm should be efficient in terms of message complexity and message size.
(2) Routing is a fundamental operation for sensor networks.It is crucial for data delivery and data aggregation [24].If the bridge detection can be integrated with the routing operation, it can introduce a less total cost to the network.
(3) Sink node may initiate the bridge detection algorithm locally.Hence, these operations should be distributed.
(4) Bridge detection algorithm should be independent from the underlying protocols as much as possible to interface to various MAC and physical layer standards such as in [25][26][27][28][29][30].
(5) The algorithm should not be dependent on localization information.

Related Work
DFS algorithm can be centrally executed by the sink node in order to detect bridges [4,9].Since collecting the whole network information is expensive and not always possible, various distributed implementations are proposed for DFS algorithm [17,20,[31][32][33][34][35][36].Most of these algorithms can be modified to detect bridge in sensor networks by using the rules proposed by Tarjan [4].Cidon [17], Hohberg [18], Chaudhuri [19], Tsin [20], and Turau [8] proposed distributed DFS algorithms for bridge and cut vertex (a vertex whose removal breaks the connectivity of a graph) detection algorithms.Turau's algorithm [8] is an extended and sensor network adopted version of Cidon's [17]  Like DFS, BFS can be centrally executed to detect bridges [9].Although this algorithm is simple to implement, execution of central BFS is an expensive operation in terms of energy consumption caused by message transfers, and it is not suitable for large scale self-organizing distributed sensor networks.Because of these reasons, distributed BFS algorithms are proposed [12][13][14][15][16].For synchronous networks, a well-known greedy algorithm is applied to construct BFS [14].This algorithm consumes O() messages and O() time, where  is the diameter of the network.Although this algorithm is very effective for constructing routing infrastructure, it is not adequate to find bridges without any extension.Liang proposed a BFS algorithm for biconnectivity testing algorithm which runs on permutation graphs and which cannot be generalized [5].Thurimella proposed a BFS biconnectivity testing algorithm in which each processor is assumed to know the whole topology [6].Because of this property, algorithm is not suitable for WSNs.
Milic proposed a bridge detection for wireless ad hoc networks [9].The algorithm uses broadcast communication of wireless nodes, and it is integrated with the BFS operation.The forward phase of the algorithm is nearly the same with the standard BFS operation.In the backward phase of the algorithm, the nodes store a list of cross edges that they found or received, append cross edges to the messages, and send to their parents.Although the algorithm completes its operation within a BFS execution interval, the message size can be very large since it is dependent on the number of cross edges.The message size can be as large as O( log 2 ()).On the other side, our proposed algorithms' message size are O(Δlog 2 ()).Besides, in this paper, we simulate Milic's algorithm against various network topologies and show that our proposed algorithms are favorable.
Pritchard proposed a three-phased algorithm for the distributed bridge detection [21].In the first step, the algorithm finds a spanning tree by implementing distributed greedy BFS tree algorithm.In the second step, the algorithm computes subtree sizes and preorder labels.In the last step, bridges are detected.The time complexity of the algorithm is O(), the message complexity is O(), and the message size is O( log 2 ()) bits.In this paper, we first propose an improved version of this algorithm.Secondly, we propose an energy-efficient bridge detection algorithm which can be integrated with the BFS operation and can finish within the BFS execution.The algorithms covered so far exactly find bridges without localization.In this study, we omit localization-based bridge detection algorithms such as [16].by applying a 3 phase method.Although the algorithm is well designed, it can be further improved for batterypowered sensor nodes.To achieve this, we propose the I-PRITCHARD algorithm which includes the following list of modifications in order to reduce transmitted message counts and transmitted bit counts.

Proposed Algorithms
(1) Phase 1 and Phase 2 can be merged into a single phase.
When the nodes are executing backward operation during BFS execution in Phase 1, each node may calculate its subtree size (#), and the convergecast operation in Phase 1 can be accomplished.To achieve this, each message in backward phase should include # field which is O(log 2 ()) bits size.After backward operation is finished, Phase 2 is not executed, so that O() messages which are flooded by the initiator node are saved by applying this improvement.
(2) Preorder messages in Phase 2 and announcement messages in Phase 3 are sent as broadcast messages instead of unicast.In this case, although the message size increases to O(Δlog 2 ()) bits, header fields for separate messages are not transmitted.So total transmitted bit counts are reduced.
An example operation is given in Figure 2. Since BFS execution is integrated with the preorder labeling in I-PRITCHARD, the algorithm is executed in 2 phases.The first phase is depicted in Figure 2  We propose ENBRIDGE algorithm for detecting bridges in WSN in an energy-efficient manner.The algorithm has two phases.In the first phase, an extended BFS algorithm is executed.The forward phase of the BFS algorithm is the same.At the backward phase of the BFS algorithm, each node broadcasts its edge states to its neighbors where a state of an edge can be one of , , and .Hence, each node knows the edge states of its neighbors.By using these 2-hop information, each node  ∈  runs the ENBRIDGE Classify given in Algorithm 1 to check whether its edge connecting to the parent node  is bridge.Each rule given in Algorithm 1 is executed sequentially by the nodes since they are ordered by considering their computational complexity.The computational complexity of the first three rules is O(Δ), and the computational complexity of the last rule is O(Δ 3 ).
if the node  has only 1 incident edge then the edge (, ) is a bridge (Rule 1).(2) else if the node  has at least one cross edge then the edge (, ) is not a bridge (Rule 2).
(3) else if the edge (, ) is the only edge connecting node  to lower levels and all other edges are bridges then the edge (, ) is a bridge (Rule 3).(4) else if one of node 's children has a cross edge connecting to node  where node 's parent is not  and id of one of node 's children then the edge (, ) is not a bridge (Rule 4).( 5) else if all neighbors of node 's children are also children of node  then the edge (, ) is a bridge (Rule 5).(6) end if.(7) if one of these rules are applied then return true.After executing these rules at the backwards stage of the first phase, each node notifies whether it is able to classify its parent link.To achieve this notification, a 1-bit classified field is transmitted during convergecast operation in backwards stage.If the node  or one of descendants cannot classify its parent link, classified field gets 0, otherwise it gets 1. Inductively, sink node finds whether one of the BFS edges is left unclassified.If the node  sends classified as false, then its parent node  does not execute ENBRIDGE Classify as given in Algorithm 2 in order to save CPU power.If all BFS edges are classified, then sink node stops the execution of the algorithm.Otherwise, the sink node starts the second phase of the ENBRIDGE algorithm.
Although the second phase of the ENBRIDGE algorithm is not always executed, it should be energy-efficient as the first phase.For the second phase, we propose an improved version of the TURAU's DFS based algorithm (I-TURAU).In this improved version,  messages are sent as broadcast messages instead of unicast messages so  messages no longer need to be transmitted.With this improvement, O() messages with O(log 2 ()) bits are transmitted during the second phase of the ENBRIDGE algorithm.The second phase can classify all edges in all situations where the first phase cannot.From this fact, one can claim that the execution of the first phase is redundant.Although this claim can be true for sensor networks which do not use BFS like routing infrastructures, this claim is false for sensor networks where BFS is the dominant routing protocol.The detailed ENBRIDGE algorithm is given in Algorithm 2.

Analysis
In this section, we will analyze proof of correctness, message, time, space, and computational complexities of the I-PRITCHARD and ENBRIDGE algorithms.

Theorem 1. Nodes executing I-PRITCHARD detect bridges and terminate the execution correctly.
Proof.Correctness of merging Phase 1 and Phase 2 into a single phase can be proved by induction.Since the BFS execution is synchronous, each leaf node can calculate its # as 1 during backward phase of BFS as the base case of the induction.Each nonleaf node can calculate its # by aggregating its children's #, where this operation is continued until the sink node's execution inductively.The proof of correctness of using broadcast instead of unicast is trivial since the same information is received by all nodes.

Theorem 2. Each node detects its parent link state correctly after executing the ENBRIDGE algorithm.
Proof.We first prove the correctness of the ENBRIDGE Classify algorithm.To prove the correctness of Rule 1, we assume that the node  has only 1 incident edge.In this case, the node  is a leaf; thus, (, ) is a bridge.To prove the correctness of Rule 2, we assume that the node  has a cross edge , and then  can be one of followings.
To prove the correctness of Rule 3, we assume that the node  does not have any cross edge; the edge (, ) is the only edge connecting node  to lower layers, and all other edges are bridges.In this case, the node  does not have any alternative path connecting it to lower layers which excludes the edge (, ), so the edge (, ) is a bridge.To prove the correctness of Rule 4, we assume that one of the node 's children (node ) has a cross edge connecting to node , where   is not equal to the  and one of node 's children.In this case, an alternative path can be found as (, ), (, ), and (, ) as shown in Figure 4(a).To prove the correctness of Rule 5, we assume that Rule 2 and Rule 4 are not true and all neighbors of the node 's children are also children of the node .In this case, since all edges other than the edge (, ) cannot connect the node  to the lower layers, the edge (, ) is a bridge.An instance of this case is depicted in Figure 4(b).If all of these rules are not applicable, ENBRIDGE uses broadcast-based TURAU to find bridges.Thus, ENBRIDGE detects bridges, and execution of the ENBRIDGE terminates in all nodes.

Theorem 3. The count of sent messages in I-PRITCHARD is 3𝑁-1 at the best case and 4𝑁-3 at the worst case.
Proof.At the best case, the nodes are arranged as a star topology, where 2-1 messages are sent at the first phase, 1 message is sent by the center node at the beginning of the second phase, and -1 messages are sent by the other nodes at the end of the second phase.Thus, 3-1 total messages are sent at the best case.At the worst case, 2-1 messages are sent at the first phase, -1 announcement messages are sent at the beginning of the second phase, and -1 at the end of the second phase; thus, 4-3 messages are sent.

Theorem 4. The count of sent messages in ENBRIDGE is 2𝑁-1 at the best case and 4𝑁-3 at the worst case.
Proof.At the best case, only BFS is executed on the star topology, so 2-1 messages are sent.At the worst case, an extra DFS is executed, where each node uses broadcast instead of unicast, so 2-2 messages are sent.Hence, 4-3 total messages are sent at the worst case.

Theorem 5. The count of received and overheard messages of I-PRITCHARD is O(Δ𝑁).
Proof.At the worst case, each node receives and overhears Δ messages at the first and second phases.Thus, total count for  nodes is O(Δ).

Theorem 6. The count of received and overheard messages of ENBRIDGE is O(Δ𝑁).
Proof.Each node receives and overhears Δ  and  messages during BFS operation at the worst case.This bound is the same for the DFS operation, so that total count of received and overheard messages of ENBRIDGE for  nodes is O(Δ).

Theorem 7. The message size of I-PRITCHARD is O((Δlog 2 (𝑁)) bits.
Proof.In I-PRITCHARD, each parent node announces preorder label of its children by broadcasting a single message.Thus, the message size of the I-PRITCHARD is O(Δlog 2 ()) bits.
Proof.At the backwards stage of ENBRIDGE, each node broadcasts its incident edge states to its neighbors.Because of this, the message size of the ENBRIDGE is O(Δlog 2 ()) bits.

Theorem 9. The time complexity of I-PRITCHARD is O(𝐷).
Proof.Since proposed improvements do no effect on time complexity, time complexity of I-PRITCHARD algorithm depends on the network diameter.Because of this, the time complexity of I-PRITCHARD is O().

Theorem 10. ENBRIDGE takes Ω(𝐷) time at the best case and O(𝑁) time at the worst case.
Proof.At the best case, only Phase 1 is executed, so that the time complexity of the ENBRIDGE algorithm is equal to the time complexity of the BFS operation, so that the lower bound of the time complexity is Ω().At the worst case, Phase 2 is executed with phase 1.In this case, the time complexity is equal to the worst case time of the DFS operation, so that the upper bound of the time complexity is O().
Theorem 11.The space and computational complexities of the I-PRITCHARD algorithm is O(Δ).
Proof.Each node should store its neighbor's state where this table can be at most O(Δ).The algorithm executes on this table, so computational complexity is O(Δ).

Theorem 12. The space complexity of ENBRIDGE is O(Δ 2 ).
Proof.Each node should store its 2-hop neighbor's state, so that this table can be at most O(Δ 2 ).
Theorem 13.The computational complexity of the ENBRIDGE algorithm is O(Δ 3 ).The lower bound for the computational complexity is Ω(Δ).
Proof.At the best case, one of Rule 1, Rule 2, and Rule 3 is executed which results in the Ω(Δ) computational complexity.In order find the computational complexity of Rule 4, we assume that the node  has  cross edges which are represented as (, ) and which are not incident to it but incident to one of its children (lets call it ).In order to find whether  is not equal to  or one of 's children, (Δ 2 − ) computations should be made.If we maximize this equation, then  = Δ/ √ 2, and the computational complexity of Rule 4 becomes O(Δ 5/2 ).Execution of Rule 5 can be O(Δ 3 ) at the worst case since a node may have Δ 2 2-hop neighbor nodes, and these 2-hop neighbor nodes are searched in the list of Δ 1-hop neighbor nodes.
A summary and analytical comparison of central algorithm (CENTRAL), Milic's Algorithm (MILIC), Turau's Algorithm (TURAU), Pritchard's Algorithm (PRITCHARD), I-PRITCHARD, and ENBRIDGE algorithms are given in Table 1.MILIC, I-PRITCHARD, and ENBRIDGE are asymptotically better algorithms in terms of sent and received messages.The message sizes of I-PRITCHARD and ENBRIDGE are O(Δlog 2 ()); on the other side, message size of MILIC is O( log 2 ()).Although TURAU and PRITCHARD algorithm's message sizes are O(log 2 ()) bits, the sent and received message counts are higher than other algorithms.Since ENBRIDGE algorithm's lower bound of sent message count is 2-1, it is analytically the best algorithm among other algorithms in terms of energy consumption caused by sent and received messages.ENBRIDGE is favorable in terms of energy consumption, but its time and computational complexities are worse than I-PRITCHARD as shown in Table 1.Although the worst case time complexity of ENBRIDGE is worse than I-PRITCHARD, its best case time complexity is equal to I-PRITCHARD's time complexity.Besides, ENBRIDGE may terminate just after BFS execution, where I-PRITCHARD has an extra phase.

Performance Evaluations
We implemented I-PRITCHARD and ENBRIDGE algorithms in TOSSIM simulator [37] to evaluate their performance.TOSSIM simulator is developed by the researchers from University of California Berkeley, Intel Research Berkeley, and Harvard University.TOSSIM inherits the TinyOS's structure and provides a simulation environment that is close to the real world.The nesC compiler which is also currently used for the TinyOS applications is modified in order to use the same compiler for both simulation and implementation.
A discrete event queue is used in the execution model.By using this queue, all simulator events are timestamped and processed in order.The hardware parts of the real-world implementations are emulated in TOSSIM.Two radio models are simulated.In the first model, the developers use error-free transmission to test the correctness of their protocols.The second model provides the developers to test the multihop transmissions.
We implemented CENTRAL, MILIC, TURAU, and PRITCHARD algorithms in order to compare them with the proposed algorithms.We generated randomly connected networks varying from 50 to 250 nodes that are uniformly distributed over the sensing area.For the lower layers, we  When the node degree and node count are increased, total received byte count of both algorithms increase linearly.These results conform with theoretical analysis where the complexity of received and overheard messages is O(Δ).These results show that I-PRITCHARD and ENBRIDGE are scalable against varying node degrees and node counts.
Performance comparisons of I-PRITCHARD, ENBRIDGE, and their counterparts are given in Figures 6(a) and 6(b).The received byte count of ENBRIDGE is the smallest, and I-PRITCHARD is the second smallest among the other algorithms.The received byte count performance order of algorithms is ENBRIDGE, I-PRITCHARD, TURAU, PRITCHARD, MILIC, and CENTRAL.These results show that BFS based approaches other than I-PRITCHARD and I-MILIC are worse than Turau's DFS based approach.I-PRITCHARD is approximately 1.7 times better than PRITCHARD on the average.ENBRIDGE is approximately 2 times better than TURAU, 3 times better than PRITCHARD, 4.5 times better than MILIC, and 6 times better than CENTRAL algorithm on the average.The reasons of this significant performance of proposed algorithms are using broadcast messages with at most O(Δlog 2 ()) bits.

Sent Byte
Counts.Secondly, we measured the sent byte counts to evaluate the energy consumption of the algorithms.Total sent byte counts of I-PRITCHARD and ENBRIDGE against node count and node degree are shown in Figures 7(a) and 7(b), respectively.Total received byte count values fluctuate between similar values when the node degree and node count are increased.These results show that I-PRITCHARD and ENBRIDGE are stable and scalable against varying node degrees and node counts.
Total sent byte counts of I-PRITCHARD, ENBRIDGE, and their counterparts are given in Figures 8(a) and 8(b).The sent byte counts of ENBRIDGE are the smallest, and those of I-PRITCHARD are the second smallest among the other algorithms similar to the received byte count performance.The sent byte count performance order of algorithms is ENBRIDGE, I-PRITCHARD, TURAU, PRITCHARD, MILIC, and CENTRAL.The sent byte count of I-PRITCHARD approximately is 1.7 times smaller than PRITCHARD on the average.ENBRIDGE is approximately 1.8 times better than PRITCHARD, 3.6 times better than MILIC, 4.5 times better than MILIC, and 8 times better than CENTRAL algorithm on the average.

Energy Consumption.
Energy efficiency is one of the most important objective for WSN.We measured the energy consumption of the distributed bridge detection algorithms.We assumed that the energy consumption occur mostly by message transfers.Energy consumptions of I-PRITCHARD and ENBRIDGE against node count and node degree are shown in Figures 9(a) and 9(b), respectively.Energy consumptions increase linearly when the node degree and node count are increased.These results show that the energy consumption of I-PRITCHARD and ENBRIDGE are stable and scalable against varying node degrees and node counts.
Performance comparisons of I-PRITCHARD, ENBRIDGE, and their counterparts are given in Figures 10(a      consumes 1.7 times less energy than PRITCHARD algorithm.The energy consumption performance order of algorithms is ENBRIDGE, I-PRITCHARD, TURAU, PRITCHARD, MILIC, and CENTRAL.The energy consumption of ENBRIDGE is approximately 1.6 times better than TURAU, 2.6 times better than PRITCHARD, 4.3 times better than MILIC, and 6.5 times better than CENTRAL algorithm.This is a significant improvement over battery-powered sensor nodes in order to maximize the network lifetime. 6.4.Wallclock Times.Lastly, we measured the wallclock times of the distributed bridge detection algorithms.Firstly, we measured the wallclock times of I-PRITCHARD and ENBRIDGE algorithms against node count and node degree which are shown in Figures 11(a) and 11(b), respectively.Since the network diameter decreases when the degree increases, the wallclock times of both algorithms decrease.A sharp increase in the wallclock time of ENBRIDGE algorithm can be observed in Figure 11(b) when the node count is 150 and the degree is 3.The reason of this sharp increase is the fact that for especially sparse networks, ENBRIDGE runs both of the phases.Since the second phase is improved DFS based cut bridge detection, time consumption increases.
Wallclock times of I-PRITCHARD, ENBRIDGE, and their counterparts are given in Figures 12(a) and 12(a).The wallclock times of ENBRIDGE are the smallest, and those of I-PRITCHARD are the second smallest among the other algorithms.TURAU has the worst performance since it is DFS based and uses unicast for the message transmission.I-PRITCHARD is better than PRITCHARD in all cases since its phase count is 1 less.MILIC algorithm performs well for most of the cases.Although the performance of the ENBRIDGE is not good for some cases as explained in the previous paragraph, the algorithm performs best among other algorithms on the average since it completes its operation within a BFS session in most of the simulations.
In this section, we showed that our analytical results given in Section 5 conform with the simulations results that ENBRIDGE and I-PRITCHARD outperform the previously proposed approaches in terms of energy and time consumptions.

Conclusions
We proposed two distributed energy-efficient bridge detection algorithms I-PRITCHARD and ENBRIDGE.Our first algorithm, is the improved version of the Pritchard's algorithm.In this algorithm, we merged two phases into a single phase which leads to the removal of a downcast operation.Besides, we used radio broadcast communication instead of unicast message transfer that leads to the reduction of message header transmissions.The original idea of the second algorithm is to process proposed rules on 2-hop neighborhood information during a BFS session in order to detect bridges and classify all edges.With the help of these methods, our algorithms have O() sent message complexity, O(Δ) received and overheard message complexity where the largest message is O(Δlog 2 ()) bits.
(a).In this figure, ID of the nodes is written inside of each node, # value and preorder label are written near to each node.The second phase is depicted in Figure2(b).In this figure, low and high values are written near to each node.The edges (1, ), (7, ),(5,9), and (7, 8) are the bridges.

4. 2 .
ENBRIDGE Algorithm.The algorithms covered so far have below listed deficiencies which motivate us to design ENBRIDGE algorithm.(i)Even though the design of an energy-efficient DFS based bridge detection algorithm is possible, DFS applications are rare in real-world applications.Because of that, the DFS based bridge detection module may not be integrated to the other applications.(ii) Although MILIC may be easily integrated to the BFS, transmitted bit count is proportional to the network diameter and cross edge count.(iii) PRITCHARD and I-PRITCHARD are BFS based algorithms, transmitted bit counts are proportional to the average neighbor degree.Although these algorithms are energy-efficient, extra phases are executed after the BFS algorithm.

Figure 5 :
Figure 5: (a) Received bytes of I-PRITCHARD against node degree and node count.(b) received bytes of ENBRIDGE against node degree and node count.

Figure 6 :
Figure 6: (a) Received bytes of algorithms against node degree.(b) Received bytes of algorithms against node count.

Figure 7 :
Figure 7: (a) Sent bytes of I-PRITCHARD against node degree and node count.(b) Sent bytes of ENBRIDGE against node degree and node count.

Figure 8 :
Figure 8: (a) Sent bytes of algorithms against node degree.(b) Sent bytes of algorithms against node count.

Figure 9 :
Figure 9: (a) Energy consumption of I-PRITCHARD against node degree and node count.(b) Energy consumption of ENBRIDGE against node degree and node count.

Figure 10 :Figure 11 :
Figure 10: (a) Energy consumption of algorithms against node degree.(b) Energy consumption of algorithms against node count.

Figure 12 :
Figure 12: Wallclock times of algorithms against node degree.(b) Wallclock times of algorithms against node count.

Table 1 :
Analytical comparison of algorithms.