Location-Based Self-Adaptive Routing Algorithm for Wireless Sensor Networks in Home Automation

Abstract

The use of wireless sensor networks in home automation (WSNHA) is attractive due to their characteristics of self-organization, high sensing fidelity, low cost, and potential for rapid deployment. Although the AODVjr routing algorithm in IEEE 802.15.4/ZigBee and other routing algorithms have been designed for wireless sensor networks, not all are suitable for WSNHA. In this paper, we propose a location-based self-adaptive routing algorithm for WSNHA called WSNHA-LBAR. It confines route discovery flooding to a cylindrical request zone, which reduces the routing overhead and decreases broadcast storm problems in the MAC layer. It also automatically adjusts the size of the request zone using a self-adaptive algorithm based on Bayes' theorem. This makes WSNHA-LBAR more adaptable to the changes of the network state and easier to implement. Simulation results show improved network reliability as well as reduced routing overhead.

1. Introduction

Home automation (HA) systems are increasingly used to increase the safety and comfort of residents and provide distributed control over heating, ventilation, and air conditioning (HVAC), and lighting to save energy cost. Consequently, the home-automation industry has grown remarkably over the last few decades and is still evolving rapidly. Researchers and engineers are increasingly looking at novel technologies to lower the total installation and maintenance cost of HA systems. Wireless technology is a key driver in reaching those goals due to no cost for cabling, easy deployment, good scalability, and easy integration with mobile user devices.

The low-power wireless sensor network (WSN) is a promising network technology that has recently emerged in HA systems. WSNs generally consist of a number of small sensor nodes with sensing, data processing, and wireless communications capabilities [1]. These sensor nodes are inexpensive and have a battery lifetime of several years on at a low-duty cycle. They are suitable for home network settings where smart sensor nodes and actuators may be hidden in appliances such as vacuum cleaners, microwave ovens, refrigerators, and home entertainment devices. These sensor nodes inside devices in the home can interact with each other. They allow residents to manage devices in their homes more easily, both locally and remotely. Therefore, interest has grown in wireless sensor network technology in the field of home automation [2]. We refer to the combination of HA and WSN as wireless sensor networks in home automation (WSNHA).

The most popular standard for WSNHA is the IEEE 802.15.4/ZigBee/HA public application profile, among which IEEE 802.15.4/ZigBee provides general purpose, easy-to-use, and self-organizing wireless communi-cation for low cost, at a low data rate, with low complexity, and using low-power embedded devices [3–5]. The HA public application profile provides standard interfaces and device definitions to allow easy interoperability among ZigBee HA devices produced by various manufacturers of ZigBee HA products. While IEEE 802.15.4 defines the physical (PHY) layer and the medium access control (MAC) layer, ZigBee defines the layers above. IEEE 802.15.4 is considered mainly for sensor networks. Considering the low cost and easy realization in WSN, MAC 802.15.4 reduces the complexity, resulting in a simpler algorithm, but it does not have adequate technology to guarantee reliable transmission in the case of high traffic and high mobility [3–5]. The ZigBee network layer supports AODVjr routing, a variation of ad hoc on-demand distance-vector (AODV) routing [6]. On-demand routing protocol is event-driven, and it searches for a route from the source to the destination only when data packets must be sent. When no data packets are transmitted, the nodes remain silent and eventually enter a sleep status. This type of on-demand routing protocol is most suitable for WSNHA because, unlike proactive routing protocols, it does not maintain a real-time routing table for all nodes. On-demand routing protocols have a lower routing overhead and node storage requirement than do proactive routing protocols. This is the key motivation for ZigBee to adopt AODVjr as the default routing algorithm. A flooding technique is often used for route discovery in on-demand routing protocols. AODVjr [7] also performs route discovery by flooding route request packets (RREQs) to the entire wireless network to guarantee route discovery in the case of HA link instability. However, flooding packets can lead to excessive drain on limited battery power and reduce the packet delivery ratio in WSNHA because MAC 802.15.4 cannot afford heavy routing overhead, which can easily cause a broadcast storm when contention and collision occur in the MAC layer.

In order to save energy and reduce the routing overhead and packet average delay and to ensure reliable data transmission, in this paper we present a new routing algorithm for WSNHA, namely, WSNHA-LBAR (location-based self-adaptive routing for WSNHA). Instead of using flooding technology to search blindly for the route across the entire network, the proposed routing algorithm makes full use of location information of the sensor nodes in WSNHA to confine the flooding route searching space to a smaller estimated cylindrical zone and automatically adjust the radius of the cylindrical zone based on Bayes' theorem. Having a smaller route searching space results in lower routing overhead and reduces broadcast storm in the MAC layer.

The remainder of this paper is structured as follows. Section 2 describes related work, which includes the analysis of the WSNHA characteristics and a survey of the routing protocols for WSNHA. Section 3 highlights the motivation for the current work. Section 4 describes the routing algorithm of the WSNHA-LBAR. Section 5 shows how the performance of WSNHA-LBAR was evaluated by simulation. Section 6 presents the conclusions.

2. Related Works

Many routing, power management, and data dissemination protocols have been specially designed for WSNs, where energy awareness is a central design issue. The focus, however, has been on routing protocols tailored to applications and network architectures. It is therefore necessary for routing designers to meet the requirements of WSNHA systems. This section compares the existing categories of WSN routing protocols based on the characteristics of WSNHA.

2.1. WSNHA Characteristics

HA is now a mature technology, and many articles describe the characteristics of these systems [2, 8]. In general, WSNHA devices can be divided into three categories: sensors, actuators, and controllers. Sensors distributed throughout a house collect physical data such as temperature, humidity, motion, and light level. Actuators are attached to the objects the system controls, such as lamps, refrigerators, and air-conditioners. HA control functions are usually embedded in the actuators. Actuator nodes generally have fixed locations and are powered by a main electricity supply. Controllers are used to control and query the home automation settings. In addition, mobile user interface devices such as PDAs and smart phones are able to access the network for control or monitoring purposes. These handheld devices are usually highly mobile and only communicate sporadically.

Some battery-powered sensor nodes do not easily accommodate battery recharging or frequent battery replacement. This necessitates that the routing algorithm considers energy efficiency. Due to their low cost, sensor nodes usually have limited memory, which requires that the routing algorithm is simple and has low information storage requirements. WSNHA coverage is generally small, and the sensor node distribution depends on the house structure and the application, requiring a routing algorithm that can self-adapt to the node distribution. Link instability can be an issue because signal propagation inside a room encounters greater reflection, diffraction, and dispersion than does that outdoors, especially when the occupants are at home. This requires that the routing algorithm be able to self-adapt to link instability.

Using wireless sensor networks in home automation is prevalent and cost effective. A routing algorithm for WSNHA must meet these requirements to achieve reliability and energy efficiency in data packet delivery.

2.2. Comparisons of Routing Protocols for WSNs

In general, WSN routing protocols can be classified as flat-based routing, hierarchical-based routing, or location-based routing, depending on the network structure [9, 10]. Flat-based routing has low storage requirements and a simple algorithm, and it uses flooding as its main routing technology [9, 10]. Typical common flat-based routing protocols include directed diffusion [11], SPIN [12], rumor routing [13], and GBR [14]. Flooding technology results in considerable delay and needless energy consumption, as data are forwarded to every sensor.

Cluster-based routing is an efficient way to reduce energy consumption and extend the network lifetime within a cluster. The number of messages transmitted to the base station is reduced by data aggregation and fusion. Cluster-based routing is mainly implemented as two-layer routing: one layer is used to select cluster heads, and the other layer is used for routing. High-energy nodes in cluster-based routing can be used to process and send information, whereas low-energy nodes can be used to perform sensing in close proximity to the target. Typical common cluster-based routing protocols include LEACH [15], PEGASIS [16], TEEN [17], and TTDD [18]. The clustering algorithm is based on a distributed algorithm, which incurs extra overhead and is not particularly easy to implement in WSNHA. WSNHA does not require the level of complexity of the cluster formation algorithm.

Location-based routing protocols are less complicated and easier to implement than cluster-based routing protocols and more energy efficient than flat-based routing protocols due to reduced flooding. WSNHA systems are generally small, and most of the nodes are static. Obtaining location information can be easily implemented in WSNHA. The availability of small, low-power global positioning system receivers for calculating relative coordinates makes it possible to apply location-based routing algorithms in WSNHA. The location information of all the sensor nodes in WSNHA can be stored. This makes location-based routing most suitable for WSNHA. Location-based routing makes full use of location information to reduce energy consumption. Typical common location-based routing protocols include GAF [19] and GEAR [20].

2.3. Location-Based Routing

In WSNHA, building an efficient and reliable routing algorithm is a very challenging task due to the limited resources and link instability. We can group location-based routing into three types according to location information usage [21, 22]. The first is the localized routing algorithm in which each node only uses the location of itself, its neighboring nodes, and the destination to forward the packets to the next hop. Typical localized routing protocols include GPSR [23], GEAR [20], and GOAFR [24]. The main component in this type of routing is simple greedy forwarding in which the packet should make progress at each step along the path. Each node forwards the packet to a neighbor closer to the destination than itself, until ultimately the packet reaches the destination. Greedy forwarding easily causes the nodes to end up at a local minimum. In other words, if nodes have consistent location information, greedy forwarding is guaranteed to be loop-free.

The second type of location-based routing is the grid-based routing algorithm, which divides the network into many smaller grids based on the location information of the nodes. All the nodes in the same grid only send the data packet to their grid leader. Grid leaders are responsible for routing data packets by grids. Typical grid-based routing protocols include GAF [19] and GRID [25]. Grid-based routing algorithms are suitable for large and dense networks due to the reduction of routing complexity. However, dividing the network into grids for small systems such as WSNHA is less constructive.

The third type is the location-aided routing algorithm, which uses the location information of nodes for route discovery and limits the route discovery flooding to a geographic area around the destination. Typical location-aided routing protocols include LAR [26], DREAM [27], and LBM [28]. AODVjr in ZigBee also uses flooding for route discovery. So this location-aided routing scheme is promising for the improvement of AODVjr.

3. Motivation for Current Work

Although IEEE 802.15.4/ZigBee, which supports AODVjr as the default routing algorithm, is the popular standard for WSNHA, WSNHA presents certain challenges related to its practical design and implementation. Due to the nonuniform node distribution and link instability in WSNHA, flooding RREQ in AODVjr leads to a high possibility of broadcast storm and collision in MAC 802.15.4, a low packet delivery ratio, and high energy consumption. Therefore, it is desirable to improve the performance of AODVjr as well as to ensure reliable data transmission in WSNHA.

The development of localization work made location-based routing possible. We can make full use of the location information of nodes for route discovery of AODVjr and limit the route discovery flooding to a smaller zone around the destination, a strategy referred to as location-aided routing (the smaller zone is named the "request zone" in this paper). However, two problems remain to be overcome. The first is the definition and calculation of the request zone; the second is self-adaptation of the request zone.

3.1. Definition and Calculation of the Request Zone

LAR [26], DREAM [27], and LBM [28] represent three request zone shapes: rectangle, bar, and fan, respectively. However, LAR and DREAM are designed for Ad Hoc networks, and so the request zones in LAR and DREAM are calculated using the mobile nodes' velocity [26, 27]. The request zone in LBM is not designed for limiting the route discovery flooding, but for data packet transmission [28]. Most of the nodes in WSNHA are static, so the shape of the request zone can derive from the definition in LAR, DREAM, and LBM, but the calculation of the request zone should be appropriate to the task.

3.2. Self-Adaptation of the Request Zone

In general, the smaller the space to be searched is, the smaller the routing overhead and broadcast storm will be. However, too small request zone can lead to no or unstable routing in the request zone, even though a stable route exists outside the request zone. We call this "holes in the request zone." If the request zone has holes, route discovery is likely to be done multiple times, which in turn increases the routing overhead and the route setup time. Expanding the request zone to the entire network when route discovery fails rapidly degrades performance and loses the benefits of an algorithm based on a confined request zone. In addition, expanding the request zone can lead to broadcast storm on the MAC layer and a decrease in the packet delivery ratio. In order for the routing algorithm to meet a relatively high packet delivery ratio while minimizing the size of request zone, which also minimizes the routing overhead, the sensor nodes need to automatically adjust the size of the request zone according to the network state.

This paper focuses on the above problems to develop a routing algorithm that can meet WSNHA requirements while minimizing the routing overhead.

4. Routing Algorithm

In AODVjr routing, when a source node has data to send to a destination node but has no existing route to the destination, it initiates a route discovery process by broadcasting a route request packet (RREQ). An intermediate node, upon receiving the RREQ for the first time, will rebroadcast the RREQ again if it does not know a route to . When the RREQ reaches a node that has a route to (which may be the destination node itself), a route reply packet (RREP) is sent back to . When receives the RREP, it inserts the routing information about into its routing table and uses this routing information to send data to .

Instead of blindly searching for the route in the entire network, WSNHA-LBAR uses the location information of the sensor nodes to confine the flooding route searching space to a smaller estimated request zone (Rzone), which represents the route-searched zone.

4.1. Location-Based Route Discovery

When the Rzone is defined, the addresses of the source node and the destination node are stored in the RREQ. Each intermediate node receives an RREQ and then executes the algorithm of WSNHA-LBAR to forward the RREQ as Algorithm 1 shows.

Algorithm 1: recvRREQ.

Input: RREQ,

Result: how to deal with RREQ

Establish a reverse link to the node from which it

received RREQ

If  RREQ received before  then

 discard RREQ;

else

 if  RREQ.destination==  then

  respond with RREP using the reverse link;

 else

  if  RREQ.destination is the's neighbor  then

   forward RREQ to RREQ.destination;

  else

   if    then

    if    is static  then  broadcast RREQ;

   else

    discard RREQ;

   end

  end

 end

end

In algorithm, the static nodes located in the Rzone are responsible for rebroadcasting an RREQ, but the static nodes outside the Rzone are not responsible for rebroadcasting a RREQ. If a mobile node receives an RREQ and it is not the destination node, it discards the RREQ directly because a route that uses the mobile node as its intermediate node is not stable.

In WSNHA-LBAR, careful choice of the proper Rzone can reduce the number of broadcast RREQs and save bandwidth and energy. So the definition of the Rzone directly influences the performance of WSNHA-LBAR. Because WSNHA is intended for coverage of a small area, a rectangular Rzone does not reduce the routing overhead. If the source and destination nodes are located at the edges of WSNHA, a rectangular Rzone is easily degraded to flooding in the entire network [29]. A fan-shaped Rzone is too narrow for WSNHA and does not include enough nodes to find a route, and it therefore easily leads to the failure of route discovery [29]. In the following, we will introduce the definition of the Rzone and judge whether the sensor nodes are located in the Rzone.

In Figure 1, consider node that needs to find a route to . If no valid path to exists in the routing table of , initiates route discovery to find one. Before route discovery, can establish an Rzone between and . A sphere with as its center and radius describes the transmission range of the radio signal; the transmission range of every node is assumed to be the same. The Rzone is a cylindrical zone, shown as the red dotted line in Figure 1, where it is assumed that the coordinates of , , and are , and , respectively. The distance between and the line is . The condition for determining whether is located in the Rzone is .

Figure 1

Request zone in WSNHA-LBAR.

The calculation of proceeds as follows. Suppose that the equation of a straight line is

(1)

where , , , , , , , and are constants that can be computed from the coordinates of and :

(2)

We can define

(3)

and can be expressed as

(4)

where vector , , and × is the vector cross product.

4.2. Self-Adaptation of the Request Zone

Two cases may lead to a low packet delivery ratio in WSNHA-LBAR. The first is when no route from to is available in the current cylindrical Rzone. In this case, we need to increase the radius of the cylindrical Rzone. The second case involves a heavy collision in the MAC layer, which leads to failure of data packet transmission. In this case, we decrease the radius of the Rzone, as a smaller route-searching space reduces the chance of collision problems in MAC 802.15.4. Furthermore, source-destination pairing in WSNHA is random. If we define the same radius of the Rzone for every source-destination pair, the performance of location-based route discovery cannot reach the optimum because different source-destination pairs maybe subject to different network problems (such as link instability, environment disturbance, and heavy collision in the MAC layer). It is very difficult for the engineer to define the proper radius of the Rzone for every source-destination pair. We proposed a self-adaptive algorithm for the request zone based on Bayes' theorem, which lets the nodes automatically adjust the radius of the Rzone by self-learning.

To realize the automatic adjustment of the radius of the Rzone by self-learning, we need to solve the following two problems.

(i)What kind of information/knowledge the sensor node can learn from route finding?

(ii)How to make full use of the knowledge (the sensor node have learnt) to automatically adjust the radius of cylinder zone?

We can view the number of retransmissions of RREQs as knowledge, which the sensor nodes can learn because the source node will retransmit RREQ when the source node does not receive the RREP. Retransmission of the RREQ implies that the current radius of the Rzone is improper and should be modified. So, we can view successful transmission as receiving an RREP when flooding RREQ in the current Rzone. In a similar way, we can view unsuccessful transmission as not receiving an RREP when flooding RREQ in the current Rzone. The self-learning of the sensor node occurs as it counts the number of successful and unsuccessful transmissions and calculates the probability of successful transmission for different Rzone radii. The sensor node chooses the Rzone radius that corresponds to the highest probability of receiving an RREP.

The above self-learning process can be realized by Bayes' theorem.

4.2.1. Bayes' Theorem

Bayes' theorem [30] shows the way in which conditional probability depends on its inverse. The theorem expresses the posterior probability of a hypothesis in terms of the prior probabilities of and and the probability of given . It implies that evidence has a stronger confirming effect if it was more unlikely before being observed. Bayes' theorem relates the conditional and marginal probabilities of events and , and it is expressed as

(5)

where is the complementary event of , and is the prior probability or marginal probability of . It is "prior" in the sense that it does not take into account any information about . is the conditional probability of , given . It is also called the posterior probability because it is derived from or depends upon the specified value of . is the conditional probability of given . is also the prior probability or marginal probability of . Intuitively, Bayes' theorem describes the way in which one's beliefs about observing "" are updated by having observed "". It implies that evidence has a stronger confirming effect if it was more unlikely before being observed. Bayes' theorem is one of the most important theories in machine learning. Derived from conditional probabilities, we can rewrite Bayes' theorem as

(6)

4.2.2. Mapping Relationships between Bayes' Theorem and Self-Adaptation of the Request Zone

Let be the prior probability of successful transmission and let be the prior probability of unsuccessful transmission. is the conditional probability that the radius of cylindrical Rzone is when we have successful transmission. is the probability that the radius of cylindrical Rzone is and route discovery is successful. is the probability that the radius of cylindrical Rzone is and route discovery is unsuccessful. The conditional probability of successful transmission when the radius of the Rzone is is given by

(7)

4.2.3. Realization of Self-Adaptation of the Request Zone

Data Structures for Realization

We create three tables and two counters for the realization of self-adaptation of cylindrical Rzone based on Bayes' theorem. The functions and descriptions of these data structures are given in Table 1. Here, , , , and are used to calculate the prior probability, and is used to store the posterior probability.

Table 1 The main datastructures: tables and counters.

Before we described the detailed computation, we gave the following nomenclature.

• (i) : it denotes the total number of unsuccessful transmissions when the radius of cylindrical Rzone is , which can be found in table .

• (ii) : it denotes the total number of successful transmissions when the radius of cylindrical Rzone is , which can be found in table success.

The detailed computation is as follows. is calcu-lated from

(8)

where is the total number of unsuccessful transmissions when , which can be found in table . is calculated from

(9)

where is the total number of successful transmissions when , which can be found in table success.

Table probability is used to store the value of , which can be calculated by (7), (8), and (9). is the conditional probability of successful transmission when the radius of the cylindrical Rzone is . is calculated from

(10)

A schematic diagram detailing the calculation is shown in Figure 2.

Figure 2

Realization of Bayes calculation.

Figure 3

Dynamic creation of the tables.

Algorithms for Realization

We modify the location-based routing to realize self-adaptation of the cylindrical Rzone. Two functions must be modified: the function and the function.

Before we analyzed these two revised functions, we gave the following nomenclature.

• (i) : it denotes the number of RREQ retransmission.

: it denotes the optimal .

• (ii) : it denotes the max probability.

: it denotes the probability of successful transmission when the radius of cylindrical Rzone is , which can be found in table .

• (iii) : it denotes whether the value of is tested or not when the radius of cylindrical Rzone is , which can be found in table . When the sensor node sends RREQ for rout finding but it did not receive RREP, it will use another value as the radius of cylindrical Rzone to retransmit RREQ. In order to avoid using the same value as the last time, we marked field try of the used value as "1". Once the sensor node receives RREP, the sensor node will reset field of all the possible radius value to "0".

• (iv) : it denotes the last time radius of the cylindrical Rzone.

Firstly, we analyze . Before the sensor node broadcasts an RREQ for route finding, it must choose the optimal according to the table . Initially, is empty, and the sensor node does not know which is the optimum value; so we set the transmission radius of the sensor node as the initial radius of Rzone, which means that the initial value of equals to the maximum range of transmission of a sender node. Later, as long as the sensor node does not receive an RREP, it will retransmit an RREQ. In other words, the last time radius of the cylindrical Rzone is invalid for route finding. Before the retransmission of an RREQ, the sensor node must update field of corresponding pre_region in table failure and update field probability and try of corresponding pre_region in table probability. So the sensor node sets the field count of the previous to add 1 in table failure, and at the same time, the sensor node increases the by 1. Then, the sensor node uses (10) to recalculate the table probability and set try for the previous to 1 in probability. When it retransmits an RREQ, it can choose the whose probability is highest or one that has not been previously used (the field "try" is initially set to 0, representing the fact that this value of has not been used, and it is reset to 1 when this value is used). This algorithm is shown in Algorithm 2, where the represents the previous , and represents the number of RREQ retransmissions.

Algorithm 2: sendRREQ.

Input:  failure, success, probability, failure_sum

Input:  success_sum, pre_region, req_cnt

/   initialize the max probability to 0     

;

optimal_region = 0;

/   First time to send RREQ        

if  req_cnt == 0  then

 /   Choose the optimal        

 foreach  in probability  do

  if  (probability().try! = 1)

  &&(probability().probability >max)  then

   max = probability().probability;

   optimal_region = probability().;

 end

 /   Table probability is empty        

 if  max == 0  then

  foreach     in probability  do

   if  (probability().try! = 1)

   &&(probability().probability == 0)  then

    optimal_region = probability().;

    break;

  end

 end

/   Retransmit RREQ          

else

 /   Update table probability and failure   

 foreach  in probability  do

  if  probability(). == pre_region  then

   probability().try = 1;

 end

 foreach  in failure  do

  if  failure(). == pre_region  then

   failure().count ++;

 end

 failure_sum ++;

 /   Recalculate the probability      

 foreach  in probability  do

  Probability().probability

  

 end

 /   Choose the new optimal      

 foreach  in probability  do

  if  (probability().try! = 1)

  &&(probability().probability >max)  then

   max = probability().probability;

   optimal_region = probability().;

 end

 if  maxprobability == 0  then

  foreach  in probability do

   if  (probability().try! = 1)

   &&(probability().probability == 0)  then

    optimal_region = probability().;

    break;

  end

 end

end

RREQ.region = optimal_region;

pre_region = optimal_region;

send RREQ;

Second, we analyze the function . This algorithm is shown in Algorithm 3. When the sensor node successfully receives an RREP, it needs to record this successful transmission using current radius value and modify its success table. Because the current radius value has already been recorded by pre_region, so the sensor node adds 1 to pre_region in table success, and at the same time, the sensor node also increases successs_sum by 1. Then, the sensor node uses (10) to recalculate table probability and sets try for all values to 0 in table probability.

Algorithm 3: recvRREP.

Input: failure, success, probability, failure_sum

Input: success_sum, pre_region, req_cnt, RREP

……

/   If RREP for me, update table success    

foreach  in success  do

 if  (success(). == pre_region)  then

  success().count ++;

end

success_sum ++;

/   Recalculate the probability       

foreach  in probability  do

 probability().probability

 

 probability().try = 0;

end

free RREP;

……

Parameters in the Algorithm

In this algorithm, we dynamically create the tables to calculate the probability of successful transmission under the condition of the different . Dynamic creation of those tables depends on two parameters , which represents the grain size about the change of the Rzone, and , which represents the initial radius of the Rzone. It is hard to judge that the failure of RREQ transmission is due to either the collision in MAC layer or the disconnection in Rzone; so we adopt as the center and try the decrease and increase of by the equal probability. Assume that the longest distance of the house is . Using these two parameters, the above three tables can be dynamically created. We create the values of in the following order:

(11)

where and . Figure 4 showed the structures of three tables when and .

Figure 4

Comparison of packet delivery ratio by using WSNHA-LBAR, LAR, and AODVjr in Scenario 1 with 100 nodes, Scenario 2 with 150 nodes, and Scenario 3 with 200 nodes.

Generally, we choose the transmission region of the sensor node as the initial radius. These two parameters can be decided by the engineer. If is increased (or decreased), the variation of the Rzone is increased (or decreased), the accuracy of the adjustment is decreased (or increased), and the size of the three tables is decreased (or increased). The size of table depends on the search_step and the area of the house. Because the coverage of WSNHA is not big, the storage of those tables does not consume much memory.

5. Performance Evaluation

In order to evaluate the performance characteristics of the WSNHA-LBAR protocol, we developed the simulation model using the NS2 simulation tool [31]. Our goal in conducting this evaluation study is to find the advantages of WSNHA-LBAR by comparing the performance of WSNHA-LBAR with other wireless routing protocols. As we know, the popular standard for WSN application is the ZigBee specification. The network layer of ZigBee supports AODVjr routing. So in evaluation study, we used NS2 to compare the performance of WSNHA-LBAR and AODVjr. In addition, in order to find advantages of self-adaptation scheme in WSNHA-LBAR, we also compare the performance of WSNHA-LBAR and LAR in which the cylindrical zone is used as the request zone.

5.1. Performance Measurement

We choose four metrics for analyzing the performance of WSNHA-LBAR and AODVjr.

5.1.1. Packet Delivery Ratio

This is the ratio of the number of data packets received to the number originally sent. This metric indicates the reliability of the routing protocol.

5.1.2. Routing Overhead

This is the number of routing command packets. This metric reflects how much bandwidth is occupied by the routing command packets.

5.1.3. Average Packet Delay

This is the average one-way latency for successfully transmitting a packet from the source to the destination. It reflects the response time of the routing protocol.

5.1.4. Residual Energy Ratio

This is the ratio of the residual energy to the initial energy in the network. It reflects the energy efficiency in the network.

5.2. Simulation Parameters

Apart from the routing algorithm, there are many factors which can influence the final simulation results such as the number of static nodes and mobile nodes, the velocity of the mobile nodes, and the rate of sending packets in application layer. In order to make the simulation environment close to the HA, we consider the following four parameters.

5.2.1. The Number of Mobile Nodes

Generally, there are small number of mobile nodes in WSNHA application; so we do not need to focus on highly mobile nodes. On the other hand, the MAC lay of WSNHA is MAC 802.15.4 [32] which is not suitable for high-mobility network [3–5, 33].

5.2.2. Transmission Range

The transmission range is determined by the characteristics of wireless channel in WSNHA environment and the parameters of the development board we used in HA.

5.2.3. The Rate of Sending Packets

The MAC lay of WSNHA is MAC802.15.4. It has the characteristic of low data throughput application, low power, and low cost. In general, MAC 802.15.4 maintains a high packet delivery ratio for application traffic up to 1 packet per second(pps), but the value decreases quickly as traffic load increases [34–36].

5.2.4. The Size of Packet

On the one hand, application packet size is not very big in most WSNHA applications. On the other hand, application packet size depends on the specification of IEEE802.15.4 since its maximal MAC frame size is 102 bytes. In addition, we must consider the application overhead in application layer and routing overhead in network layer; so in most NS2 simulation, application packet size belongs to the range of 35 bytes to 90 bytes.

In summary, we use the simulation parameters shown in Table 2 to design the simulation scenarios according to the specific application scenarios in WSNHA.

Table 2 Parameters used in simulation.

5.3. Design of Simulation Scenarios

We designed five groups of simulation scenarios according to the HA application. In each group, the basic simulation parameters shown in Table 2 are the same.

5.3.1. The First Group of Simulation Scenarios

In this group simulation scenarios, we fixed the network workload, the number of the mobile nodes, and sensor field size in all simulation scenarios and study the performance measurements as a function of amount of sensor nodes.

Considering that there are few mobile nodes in WSNHA, the number of mobile nodes was limited to 2 in this group of simulation scenarios. Three source/destination pairs were randomly selected from the sensors deployed in a 50 m by 50 m square sensor field. As the size of sensor field was not changed, we gradually increased the number of nodes in the network. The number of sensor nodes was increased from 100 to 200 nodes with an increment interval of 50 nodes.

5.3.2. The Second Group of Simulation Scenarios

In this group of simulation scenarios, we fixed the number of sensor nodes, the number of mobile nodes, and sensor field size in all simulation scenarios and study the performance measurements as a function of the network workload.

The sensor field in this group of simulation scenarios is  m containing 100 nodes. The number of mobile nodes was limited to 2. The number of source/destination pairs was increased from 1 to 4 with an increment interval of 1 pair.

5.3.3. The Third Group of Simulation Scenarios

In this group simulation scenarios, we fixed the number of sensor nodes, the network load, and sensor field size in all simulation scenarios and study the performance measurements as a function of the number of mobile nodes.

The sensor field in this group of simulation scenarios is  m containing 100 nodes. The number of source/destionation pairs was limited to 3. The number of mobile nodes was increased from 1 to 4 with an increment interval of 1 mobile node.

5.3.4. The Fourth Group of Simulation Scenarios

In this group of simulation scenarios, we fixed the network workload and network density in all simulation scenarios and study the performance measurements as a function of sensor nodes number and sensor field size. In other words, we analyzed the performance of AODVjr, LAR, and WSNHA-LBAR in different network coverage. We design this kind of simulation scenarios because the macroscopic connectivity of a sensor field is a function of the average density. If we had kept the sensor field area constant but increased network size, we might have observed performance effects not only due to the larger number of nodes but also due to increased connectivity.

In order to approximately keep the average density of the sensor nodes constant, we designed three simulation scenarios with sensor field dimensions of , , and  m, containing 16, 100, and 256 nodes, respectively. In all simulation scenarios, the number of mobile nodes was limited to 2, and 3 source/destination pairs were randomly selected from the sensors deployed in the sensor fields.

5.3.5. The Fifth Group of Simulation Scenarios

The fifth group of simulation scenarios came from the operational testbed in our HA model. According to the specific application scenarios in this HA model, we design three simulation scenarios with sensor field dimensions of , , and  m, containing 20, 30, and 40 nodes, respectively. In all simulation scenarios, the number of mobile nodes was limited to 1, and 1 source/destination pair was randomly selected from the sensors deployed in the sensor fields.

5.4. Simulation Results and Analysis

5.4.1. The First Group of Simulation Results

Figure 4 shows packet delivery ratios achieved using WSNHA-LBAR, LAR and AODVjr in three scenarios for the first group of simulations. The packet delivery ratios of the three routing algorithms decreased as the number of nodes increased, because this leads to heavy contention in the MAC layer. The packet delivery ratios of the WSNHA-LBAR and LAR were higher than those of AODVjr in all scenarios because the cylindrical Rzone reduced the routing overhead, which in turn reduced the burden on the MAC layer. The packet delivery ratio of the WSNHA-LBAR was higher than that of LAR in all scenarios because WSNHA-LBAR is a self-learning algorithm which lets the sensor node automatically get the optimal by learning the number of the retransmission. WSNHA-LBAR is more flexible than LAR.

Table 3 lists the measurement results of the four performance metrics for WSNHA-LBAR, LAR, and AODVjr in different scenarios. The performance for overhead of WSNHA-LBAR and LAR was better than that of AODVjr when WSNHA-LBAR and LAR maintained a high packet delivery ratio. However, the performance for packetaverage delay of LAR and AODVjr was better than that of WSNHA-LBAR because automatic self-learning in WSNHA-LBAR is exchanged by the decrease of performance for packet average delay. The performance for of WSNHA-LBAR and LAR is very close, and the performance for of three routing algorithms is very close.

Table 3 Performance comparison in different scenarios: WSNHA-LBAR (abbreviated by LBAR) versus LAR versus AODVjr.

5.4.2. The Second Simulation

Figure 5 shows packet delivery ratios achieved using WSNHA-LBAR, LAR, and AODVjr in three scenarios for the second group of simulations. The packet delivery ratios of the three routing algorithms decreased as the number of source/destination pairs increased, because increasing source/destination communication leads to heavy traffic and collision in the MAC layer. The packet delivery ratios of the WSNHA-LBAR and LAR were higher than those of AODVjr in all scenarios because the cylindrical Rzone reduced the routing overhead, which in turn reduced the burden on the MAC layer. The packet delivery ratio of the WSNHA-LBAR was higher than that of LAR in all scenarios because WSNHA-LBAR is a self-adaptive and it can decrease the flooding of the Rzone when traffic is heavy.

Figure 5

Comparison of packet delivery ratio by using WSNHA-LBAR, LAR, and AODVjr in Scenario 1 with 1 pair of source/destionation, Scenario 2 with 2 pair of source/destination, Scenario 3 with 3 pairs of source/destination, and Scenario 4 with 4 pairs of source/destination.

Table 4 lists the measurement results of the four performance metrics for WSNHA-LBAR, LAR, and AODVjr in different scenarios. The performances for overhead of WSNHA-LBAR and LAR was better than that of AODVjr when WSNHA-LBAR and LAR maintained a high packet delivery ratio. However, the performance for packet average delay of LAR and AODVjr was better than that of WSNHA-LBAR because automatic self-learning in WSNHA-LBAR is exchanged by the decrease of performance for packet averagedelay. The performance for overhead of WSNHA-LBAR and LAR is very close, and the performance for residual energy ratio of three routing algorithms is very close.

Table 4 Performance comparison in different scenarios: WSNHA-LBAR (abbreviated by LBAR) versus LAR versus AODVjr.

5.4.3. The Third Simulation

Figure 6 shows packet delivery ratios achieved using WSNHA-LBAR, LAR, and AODVjr in three scenarios for the third group of simulations. MAC 802.15.4 is not designed for a mobile network, and it cannot guarantee reliable transmission when the network topology is frequently changed. The packet delivery ratios of the three routing algorithms decrease as the number of mobile nodes increases. However, the packet delivery ratio of WSNHA-LBAR was higher than that of LAR and AODVjr because WSNHA-LBAR is self-adaptive and it can automatically adjust the Rzone when the network topology changes.

Figure 6

Comparison of packet delivery ratio by using WSNHA-LBAR, LAR, and AODVjr in Scenario 1 with 1 mobile node, Scenario 2 with 2 mobile nodes, Scenario 3 with 3 mobile nodes, and Scenario 4 with 4 mobile nodes.

Table 5 lists the measurement results of the four performance metrics for WSNHA-LBAR, LAR, and AODVjr in different scenarios. The performance for overhead of WSNHA-LBAR and LAR was better than that of AODVjr when WSNHA-LBAR and LAR maintained a high packet delivery ratio. However, the performance for packet average delay of LAR and AODVjr was better than that of WSNHA-LBAR because automatic self-learning in WSNHA-LBAR is exchanged by the decrease of performance for packet average delay. The performance for overhead of WSNHA-LBAR and LAR is very close, and the performance for residual energy ratio of three routing algorithms is very close.

Table 5 Performance comparison in different scenarios: WSNHA-LBAR (abbreviated by LBAR) versus LAR versus AODVjr.

5.4.4. The Fourth Simulation

Figure 7 shows packet delivery ratios achieved using WSNHA-LBAR, LAR, and AODVjr in three scenarios for the fourth group of simulations. The packet delivery ratios of the three routing algorithms decreased as the network coverage and the number of nodes increased, because this leads to heavy contention and collision in the MAC layer. The packet delivery ratio of the WSNHA-LBAR and LAR was higher than that of AODVjr in all scenarios because the cylindrical Rzone reduced the routing overhead, which in turn reduced the burden on the MAC layer. The packet delivery ratio of the WSNHA-LBAR was higher than that of LAR in all scenarios because WSNHA-LBAR is a self-adaptive which results in greater tolerance for changes of the network state.

Figure 7

Comparison of packet delivery ratio by using WSNHA-LBAR, LAR, and AODVjr in Scenario 1, Scenario 2, and Scenario 3.

Table 6 lists the measurement results of the four performance metrics for WSNHA-LBAR, LAR, and AODVjr in different scenarios. We can finds when their performance for packet delivery ratio is very close, their performance for packet average delay is very close. The performances for overhead of WSNHA-LBAR and LAR is very close, and the performances for residual energy ratio of three routing algorithms are very close.

Table 6 Performance comparison in different scenarios: WSNHA-LBAR (abbreviated by LBAR) versus LAR versus AODVjr.

5.4.5. The Fifth Simulation

Figure 8 shows packet delivery ratios achieved using WSNHA-LBAR, LAR and AODVjr in the three scenarios for the fifth group simulations. The packet delivery ratios of the three routing algorithms decreased as the network coverage and the number of nodes increased, because this leads to heavy contention and collision in the MAC layer. The packet delivery ratio of the WSNHA-LBAR and LAR was higher than that of AODVjr in all scenarios because the cylindrical Rzone reduced the routing overhead, which in turn reduced the burden on the MAC layer.

Figure 8

Comparison of packet delivery ratio by using WSNHA-LBAR (abbreviated by LBAR), LAR, and AODVjr in Scenario 1, Scenario 2, and Scenario 3.

Table 7 lists the measurement results of the four performance metrics for WSNHA-LBAR LAR and AODVjr in different scenarios. The performance of WSNHA-LBAR was better than that of AODVjr when WSNHA-LBAR maintained a high packet delivery ratio. The performance of WSNHA-LBAR and LAR is very close when WSNHA-LBAR maintained a high packet delivery ratio.

Table 7 Performance comparison in different scenarios: WSNHA-LBAR (abbreviated by LBAR) versus LAR versus AODVjr.

From the above five groups of simulation results, we can conclude similar characteristics. LBAR shows better performance in packet delivery ratio and routing overhead, but there is no big difference in residual energy ratio, and packet average delay becomes even worse in some case.

Firstly, the packet delivery ratios of the WSNHA-LBAR were higher than those of LAR and AODVjr in all scenarios because the cylindrical Rzone reduced the routing overheads, and self-learning algorithm in WSNHA-LBAR lets the sensor node automatically get the optimal by learning the number of the retransmission.

Secondly, the performance for routing overhead of WSNHA-LBAR and LAR was better than that of AODVjr because the cylindrical Rzone reduced the RREQ transmission. There is no big difference in routing overhead between WSNHA-LBAR and LAR because they use the same cylindrical Rzone in their algorithm except that WSNHA LBAR will adjust size of the cylindrical Rzone when retransmitting RREQ, which leads to a little difference between WSNHA-LBAR and LAR.

Thirdly, let us analyze energy consumption in WSNHA. Energy consumption of transmitting and receiving packets is the main energy consumption in WSNHA. Packets can be divided into two types. One is the command packet, and the other is the data packet. Command packets can be estimated by routing overhead. Data packet can be estimated by packet delivery ratio. From the simulation results, we can find that the performance of routing overhead among those three routing algorithm is close; in other words, energy consumption for command packet transmission is close. The packet delivery ratio of WSNHA-LBAR is the highest. In other words, WSNHA-LBAR transmitted more data packets than LAR and AODVjr; so LBAR should consume more energy than LAR and AODVjr. However, the difference of residual energy ratio among these three routing algorithm is very small. From the simulation results, we will find that their difference does not exceed 2%. In other words, WSNHA-LBAR maintained higher packet delivery ratio without introducing much energy consumption.

Fourthly, let us analyze packet average delay. From the simulation results, we can find that the performance for packet average delay of LAR and AODVjr was better than that of WSNHA-LBAR because automatic self-learning in WSNHA-LBAR is exchanged by the decrease of performance for packet average delay. The process of self-learning and finding the optimal value consumed more time. In addition, we did not count the delay of the packets that were not successfully delivered in this delay analysis. The delay of those packets is considered to be infinite. Because we neglected the undelivered packets that have infinite delay and only counted the packets delivered successfully, the average packet delay of AODVjr is smaller than that of LBAR and LAR. If we count the delay of packets that were not successfully delivered, the difference in delay among LBAR, LAR, and AODVjr is even larger.

6. Conclusions

We have developed a new kind of location-based self-adaptive routing algorithm, called WSNHA-LBAR, based on AODVjr in IEEE 802.15.4/ZigBee and WSNHA. It makes use of location information for the sensor nodes to confine route discovery flooding to a cylindrical request zone instead of searching blindly for a route in the whole network. This reduces the routing overhead and results in fewer broadcast storm problems in the MAC layer. WSNHA-LBAR uses a self-adaptive algorithm based on Bayes' theorem, which can automatically adjust the size of request zone using self-learning to increase the probability of successful route discovery. This results in greater tolerance for changes of the network state and reduces the need for human intervention.

We simulated five typical groups of simulation scenarios to compare the performance of WSNHA-LBAR LAR and AODVjr. When they have the close performance for residual energy ratio, the results for packet delivery ratio showed that WSNHA-LBAR performed better than LAR and AODVjr due to the self-adaptation of Rzone. The increase of performance of packet delivery ratio is exchanged by the decrease of performance for packet average delay. The results for overhead showed that WSNHA-LBAR and LAR performed better than AODVjr due to using cylindrical Rzone to confine route discovery flooding.