Present State-of-the-Art of Continuous Neighbor Discovery in Asynchronous Wireless Sensor Network

In this paper, we reviewed the literature on various techniques found in WSN for neighbor discovery. Neighbor discovery is a fundamental phenomenon in the network in the deployment phase. However, due to the dynamic nature of WSN, and there are possibilities of adding new nodes and removing nodes from the network, continuous neighbour discovery is needed. It does mean that neighbour discovery is not a one-time task. With neighbour discovery, there are many advantages to WSN. The efficiency of the network is improved besides saving the energy of nodes and increasing the lifetime of nodes. The review of the literature made in this paper provides useful insights on the present state of the art of continuous neighbour discovery in WSN. It also provides research gaps found in the literature.


Introduction
Wireless Sensor Network (WSN) is a collection of sensor nodes that form a network. It is meant for sensing and providing the required information. It is used in many realtime applications and became ubiquitous. A sensor is a device that gathers data from the environment or physical conditions like light, heat, pressure, temperature, and so on. Its utility is purely based on the purpose for which sensors are built. An electrical signal is typically an output of a sensor that is transmitted to the controller where it is processed. Due to its widespread usage in a plethora of applications, it became of the important research areas. Multiple sensor nodes are involved in gathering data and forwarding it to the base station, and typical WSN is shown in Figure 1. It has sensor nodes connected to base station or sink node. In turn the network can be integrated with other networks or Internet using a gateway. Each node may act as a transceiver. It can receive data and forward it to the next node or base station. Sensor nodes do have limited resources while the base station is the node which is high in resources.
As the sensor nodes are devices that work with battery power and often deployed in even hostile environments, they have many limitations. They include very little storage and processing power, such as a few hundred KBs and 8MHz, respectively. The nodes do have a short communication range and consume more power for communication. Energy is limited as nodes are batterypowered. The lifetime is finite, and the nodes may exhibit mobility as well. Nevertheless, WSN has many applications. The applications range from monitoring environment to performing sensing activities in the Internet of Things (IoT) integrated smart applications.

Sink
Links to other networks  As shown in Figure 2, WSN has many real-time applications. They are used in surveillance applications. With multimedia streaming capabilities, they are also used in entertainment. WSN is an essential part of IoT based smart applications like smart buildings, smart cities, and so on. They are also used for security and surveillance purposes. WSN is useful in realizing precision agriculture and tracking of animals. It is handy in healthcare applications, as there are wearable sensor devices that are used to monitor the health of patients. WSN is widely used in transportation and logistics. It is also used in civil structure monitoring and urban terrain tracking. Smart grids and energy control systems need the WSN services. Many industrial applications need sensor networks. Environment monitoring can be done using WSN. WSN can be built with many topologies, as shown in Figure 3.
With respect to radio communication networks, WSN can support many topologies such as star, tree, and mesh. In the case of star topology, every sensor node is directly connected to the gateway.
The number of nodes is connected to a single gateway where nodes are not allowed to send data to each other. Thus there is low-latency communication between remote nodes and gateway or base station. In this topology, the base station should be positioned in the radio transmission range of all nodes. It causes less power to be consumed by remote nodes. The tree topology, on the other hand, is a cascaded star topology. Each node is connected to a node positioned higher in the tree. It supports the easier expansion of the network. It also helps in detecting errors with ease. Its limitation is that it relies on but cable if that is collapsed, the entire network will not function. The mesh topology helps nodes to transmit data to other nodes in the communication range. If the destination node is not in the communication, range then an intermediate node is used to forward data. Here error detection becomes easier, but in an extensive network, investment is more.  The importance of neighbor discovery and different approaches for neighbor is reviewed in this paper. Having understood the importance of WSN, its topologies, applications, and deployment models, the remainder of the paper is structured as follows. Section 2 provides the concept of neighbor discovery in WSN. Section 3 provides the neighbor discovery protocols, which are essential. Section 4 presents a summary of neighbor discovery protocols. Section 5 covers performance comparison among different NDP. Section 6 covers the research gaps found in the literature. Section 7 provides conclusions and directions for future work.

Neighbor Discovery in WSN
In WSN, neighbor discovery assumes importance for effective communications. Since WSN supports the dynamic addition of new nodes and the departure of existing nodes, the neighbors of a sensor node are not static, they may change dynamically, and it is essential to discover neighbors. There are concepts of initial neighbor discovery and continuous neighbor discovery. Figure 4 presents the neighbor discovery process and later on we the difference between initial neighbor discovery and continuous neighbor discovery.
As shown in Figure 4, it is evident that the node b sends HELLO, which is not heard by other nodes. However, the nodes a and c discover each other. It in order to discover nodes, the following are essential operations.
Whenever a node wakes up, it has to broadcast the HELLO message. Any node that is already awake can hear that HELLO message to have a connection established to the sender of the message.  As shown in Figure 4, node a can receive HELLO from the node c. That way, they can discover each other. However, the nodes a and c are not aware of the presence of b and vice versa. This indicates the need for neighbour discovery.

Initial neighbor discovery
A sensor node performs initial discovery when it has no information about its neighbors. Before doing this, the sensor node also cannot communicate with the base station. Therefore it is functioning, and usefulness is limited. Therefore, there is a need for initial neighbor discovery and establish connectivity to the base station. Since it is necessary, the energy usage for discovery is justified as it is one only once. However, when the node is operational, it needs to perform continuous discovery from time to time, as long as the node lives in the network. Therefore optimization of continuous neighbor discovery is crucial for improving the lifetime of the network.

Continuous Neighbor Discovery
Before performing continuous neighbor discovery, the node is well aware of its immediate neighbors. Therefore this operation is made together with already known neighbors in order to reduce energy consumption. On the contrary, each node has to execute initial neighbor discovery separately. The purpose of continuous neighbor discovery is to detect all neighbor nodes and also find the shortest path data transfer. As the sensor nodes are randomly deployed in some geographical area, initial neighbor discovery is made once when a sensor is deployed. However, continuous neighbor discovery is needed due to the disruption of wireless connectivity and the loss of local synchronization due to clock drifts.

Neighbor Discovery Schemes
There are many neighbor discovery approaches found in the literature. This section provides a review of them. The approaches discussed in this section include U-connect [9], Disco [8], SearchLight [42], Hedis [29], Todis [29] and Prime Block Diagram (PBD) [33].

U-Connect
Kandhalu et al. [9] proposed U-connect, which is a lowlatency asynchronous neighbor discovery protocol with energy efficiency. U-connect is designed, and its latency is characterized, and its power consumption is analysed. Then it is evaluated with the power-latency metric. U-connect is believed to be a unified protocol that can address neighbor discovery in two settings. They are called symmetric and asymmetric problems. In the process, two nodes can choose m different prime numbers for discovering neighbors. When nodes use the same pair of prime numbers, a worst-case latency performance is analysed.
In order to have better performance, U-connect characterizes network and neighbor discovery schedules. The latency discovery is associated with the following. U-ψ u (m, t) = 1, if [t] p = 0 or 0 ≤ [t] p 2 < 0, otherwise p+1 2 (13) Here the prime number is denoted as p. The case is considered where p>2 for simplicity. The U-connect protocol is designed in such a way that it works well with worst-case latency, which is high. The simulation study revealed that U-connect provide a guarantee for a common active slot. Other advantages of the U-connect include improved latency and energy efficiency.

Disco Approach
This approach is explored by Dutta and Culler [8]. In WSNs, the low-power systems that are awake at different times need to discover neighbors. In such cases, the nodes need to use their radios at low duty cycles. This is the requirement in order to maximize the lifetime of the WSN. It also needs to be vigilant about the emergence of new links and the disappearance of old links. The two activities are not odds, as vigilance and low-power operations are contradicting. In such networks, Disco is the solution provided to have asynchronous neighbor discovery and solve the problem of rendezvous scheduling. The underlying method in Disco chooses two prime numbers in such a way that their reciprocal's sum is equal to the duty cycle of an application in question. Each node maintains a local counter, and it is incremented when the counter is divisible by one of the prime numbers selected. Then the node turns on the radio for the period of one counter. Disco needs an application to select the desired duty cycle and identify a node class. The Disco selects prime number automatically based on the duty cycle matching, and then radio is turned on at every multiple of the selected prime.
When a node is a wake-up, it can listen or beacon or perform both. Disco performs well in terms of rendezvous frequency, discovery latency, and flexibility for applications. Nodes can achieve discovery latency desired by adjusting duty cycles. The flexibility serves interaction patterns, duty cycles, and different needs of the applications. Talking, docking, and flocking are the three common patterns exhibited by Disco. Discovery or rediscovery of neighbors helps two different nodes with different duty cycles in the presence of a lack of current synchronization to

Hedis and Todis
These are the two neighbor discovery protocols proposed by Chen et al. [29]. Hedis stands for heterogeneous discovery as a quorum based protocol while Todis stands for Triple-Odd based discovery as a co-primality based protocol. These two protocols guarantee the process of asynchronous neighbor discovery. They operate in heterogeneous environments with different duty cycles used by each node. The granularity of duty cycles is optimized in order to have better performance. Hedis match actual duty cycles as it is an optimal quorum based approach. According to the design of the Hedis schedule, node a with given duty cycle, the schedule is considered as sa={sta}0≤t<n(n−1), which has n(n-1) time slots. Hedis and Todis are evaluated with different numbers of consecutive odd integers for building a wake-up schedule. There is a requirement of co-prime pair property, which allows a node to choose three consecutive odd integers. From the empirical study, it is understood that both Hedis and Todis can optimize duty cycles in terms of granularity with two approaches for neighbor discovery named quorum based and co-optimality based. Both the protocols can perform well with neighbor discovery latency.

SearchLight
It is a matrix-based neighbor discovery protocol. It is simple to be built. However, it compromises latency and energy efficiency. It uses a matrix to have neighbor discovery schedules. SearchLight [16] is an asynchronous neighbor discovery protocol. It is built based on three ideas that are basis. It improves periodic awake slots and for probing. It facilitates awake slots to cover a large time window. It can use probabilistic techniques.

Prime Block Design Based Neighbor Discovery
Lee et al. [33] proposed a neighbor discovery protocol based on the concept of prime block design (PBD). It provides a near-optimal solution for asynchronous wake-up cycles in WSN. It is an extension to its predecessor known as Balanced Incomplete Block Design (BIBD). The PBD works well with both symmetric and asymmetric duty cycles. It adds less number of duty cycles in excess to that of BIBD for performance improvement. Its advantage is that it is more efficient than other protocols. However, it has a particular limitation. It is the lack of availability of BIBD blocks specific duty cycles.

Summary of Important Techniques Found in Literature
This section provides some of the essential approaches found in the literature. It provides the techniques used by researchers, their advantages, limitations, and the simulation environment used by them.
As presented in Table 1, many neighbor discovery protocols are summarized. From the review of these techniques, the following are the research gaps identified.

Performance Comparison among different NDP
This section provides a comparative study of different NDP with respect to energy efficiency.  Present State-of-the-Art of Continuous Neighbor Discovery in Asynchronous Wireless Sensor Network ____________________________________________________________________________________________________ 5 As shown in Figure 5, the graph shows a comparison of the average energy consumption of different scenarios. The xaxis shows that asymmetric ratio with different values. Yaxis shows that energy consumption. In this graph, DISCO has the highest energy consumption at R=10. PBD has the lowest energy consumption at R=10. Energy consumption at R=5 high for U-connect and low for PBD.TODIS has the highest energy consumption at R=2. Searchlight has the lowest energy consumption at R=10. Energy consumption at R=1 high for DISCO and low for PBD.

Discussion and Research Gap
This section provides a summary of findings with respect to the recent state of the art on continuous neighbour discovery in WSN. The schemes or protocols summarised here include U-Connect, DISCO, SearchLight, Hedis, Todis, and PBD. Each protocol has its advantages and limitations, as provided in Table 1. Figure 5 presented in this section throw light into significant drawbacks of each neighbour discovery scheme. In other words, the research gap associated with each scheme is provided.

Ref
Techniques  Figure 6. Summary of findings As presented in Figure 6, it is evident that the existing neighbor discovery method is known as Prime Block Design (PBD) [12], extends Balanced Incomplete Block Design (BIBD) based protocol [1]. The main problem with PBD is that it is not useful in route discovery when BIBD blocks are not available for specific duty cycles. Moreover, it has limitations in generating discovery schedules for a wide range of duty cycles. This is a challenging problem to be addressed. Our future work focuses on overcoming the drawbacks of the PBD method used for neighbour discovery.

Conclusion and Future Work
Wireless Sensor Network intended to have long term monitoring applications should have an efficient neighbor discovery protocol. During deployment of sensor networks, it is an essential and challenging task to discover neighbors. The neighbor discovery is not a onetime event in the wireless sensor network (WSN) applications since a new batch of sensors can be deployed at any time during the mission. Thus, the latency and energy efficiency of maintaining neighbor nodes are directly related to the lifetime of sensor nodes with tiny batteries, and the main design issue of neighbor discovery protocols (NDPs) has been reducing discovery latency without sacrificing energy efficiency. This paper has made a review of different ND techniques. The existing neighbor discovery method known as Prime Block Design (PBD) [57] extends Balanced Incomplete Block Design (BIBD) based protocol. The main problem with PBD is that it is not useful in route discovery when BIBD blocks are not available for certain duty cycles. Moreover, it has limitations in generating discovery schedules for a wide range of duty cycles. This is a challenging problem to be addressed by considering it for our future work.