Void-hole aware and reliable data forwarding strategy for underwater wireless sensor networks

3.1 Network structure and assumption

The structure of UWSN in the current form typically consists of underwater sensor nodes (CH or cluster member), sinks, underwater acoustic, radio channels, and monitoring center. The sensor nodes are deployed randomly with different depths to collect oceanographic data. We assume a multiple static sinks network architecture to increase network reliability as well as network throughput. This architecture can also minimize energy consumption with respect to sinks nearby nodes. The static sinks are equipped with unlimited radio frequency and energy acoustic modems. Moreover, using acoustic links, the sink communicates with sensor nodes and uses radio links for connection to the monitoring center. All underwater sensor nodes are outfitted with an acoustic modem to communicate; sensor nodes can generate and relay data packets.

The cluster-based architecture for the UWSN is shown in Figure 1, in which CHs manage each of the clusters. Other nodes operate as cluster members and report their sensing results to the CH, which can reach the adjacent CHs in a single-hop manner to forward the collected data toward the sink nodes.

Figure 1

Network structure in an underwater environment.

Because of reflection and refraction; shallow water can impact acoustic communication through ambient noises, surface, and multi-path effect in addition to temperature gradients. Therefore, the cluster’s size has an inverse relation with noise attenuation; the smaller is the attenuation of noise, the greater the cluster’s size and vice versa [31]. For energy balance, we consider that the cluster’s size in deep water is larger than the shallow water.

3.2 Energy consumption model

The underwater environment’s conditions for acoustic communications, including the working frequency range and the signal attenuation, must be considered in establishing the underwater energy model to calculate the specific energy consumption [32]. Equation (1) shows the attenuation of acoustic channel over distance d, for the signal of frequency f.

where k expresses the signal’s spreading factor, and A ₀ is the normalized coefficient. The propagation geometry is described using the spreading factor (1 ≤ k ≤ 2); for a practical scenario, k is given as (k = 1.5). Moreover, based on equation (2) the q is defined as:

where the parameter α is related to the signal frequency f [33], each warrants a different calculation of the α, as shown in equation (3):

Supposed f (kHz) and d (km), by equation (4), the transmitting power consumption P _t can be obtained as:

The factual transmitting power illustrates in equation (4) P t A ( d , f ) if the transmitting power is P t . Equations (5)–(7) show the transmitting, receiving, and fusion energy consumption, respectively. The size of transmitted data is l, r is a constant depending on the receiver, and r ₀ is the energy consumption that compresses every package. The total node energy consumption is expressed as in equation (8):

4.1 Discovery of potential neighboring CHs

The sensor nodes in the VHARD-FS strategy possess a neighbors table to save energy consumption resulting from the over control messages exchanging among network nodes. Therefore, an algorithm for discovering the potential neighboring CHs is required to find the most appropriate neighbor CH as the next forwarder node. Therefore, VHARD-FS could ensure data transferring with high reliability from CHs in the intensive load zone toward the sink node. The procedure for discovering the potential neighboring CHs begins by broadcasting a Hello_Msg from sink node with depth set to 0.

For each CH receiving a Hello_Msg, if the value of Msg_broadcasting is false, the receiver node updates its neighbors table with all neighbors whose packets are received. The neighbors table comprises four particular elements (Node ID, residual energy, node depth, and void-indicator) for each neighbor nodes. This node also updates the Hello_Msg with its own parameter and finally propagates to other neighbors’ nodes within its communication range. Otherwise, that is, the Hello_Msg has already been sent by the node, then the node will only update the neighbors table. The previous steps will be repeated until all CHs in the network are covered. Then, the node performance ranking index (Node-PRI) is calculated for each node in the neighbors table depending on three metrics, namely residual energy (E _r), node depth (d _th), and void-indicator (V _ind). The node performance metric represents the nodes’ status in terms of their energy resource, presence of communication void, and its depth to the sink node. Formula (9) for estimating the Node-PRI value for node i depends on three involved node-related metrics.

where E _r(i) is the residual energy of node i at an instant, and E _ini(i) is the node’s initial battery energy level. The node performance ranking index based on (E _r) has indicated that the node with highly remaining energy corresponds to top Node-PRI (E _r), which reduces the energy exhaustion probability. The d _th(c) is the depth of current node, whereas d _th(i) indicates the potential neighbor node’s depth i. In addition, R _c is the each node’s transmission range. The node with a low depth toward the sink node corresponds to high Node-PRI (d _th), which leads to low energy consumption. The V _ind(i) is an indicator of the number of nodes within the communication range of node i at an instant, and N is the total number of nodes. The node that has a high number of neighbor nodes corresponds to top Node-PRI (V _ind), which leads to void-hole problem handling and minimal packet loss. α, β, and γ are the values used to satisfy the normalizing criteria of importance that range from 0 to 1. The node with the highest performance ranking index is the most suitable candidate as a next forwarder node among all the potential neighboring nodes. The procedure of discovering the potential neighbor CHs within the node radio range is shown in Algorithm 1.

In this phase, each CH is well aware of all its neighboring cluster heads and can use the information in the neighbors' table to send data packets to the sink node.

4.2 Transmission route discovery based on developed Node-PRI

In this process, a group of CHs is considered as the input and the routing path that would transfer the sensed data to be the output of this process. Thus, the proposed VHARD-FS strategy uses the metrics of Node-PRI in the neighbors table, which was created during the first phase for routing path formation. Each CH alerts its neighbors for possible forwarder nodes within its radio range. This condition implies that all nodes can use the neighbors’ information to send data packets to the sink. In this phase, the data from the cluster member nodes are collected by the CH. Then, the data are aggregated and forwarded to its potential next node. The node with the highest performance ranking index (Node-PRI) value will be selected as the most reliable next forwarder node by the current node. However, to even energy dissipation and to handle void-holes problem caused by node failure, VHARD-FS monitors the developed performance ranking index of nodes in routing path. When values are lower than set limits, then new routing path formation is admitted. The routes are created by picking the most desirable neighbor in every step. The proposed procedure for this phase is described in Algorithm 2.

For route discovery or creation in the proposed strategy, each node needs to be aware of its neighbor nodes only. Consequently, the information of the whole route between the source and the sink node is not needed. Therefore, it needs less memory as the routing table at each node contains only the information about the nodes within its communication range. Implementing the HARD-FS strategy is a straightforward process as the algorithm does not require complex calculation or requirements. Figure 3 shows the visual representation of the proposed VHARD-FS.

Figure 3

Flowchart of VHARD-FS.

5.1 Simulation setup

The performance evaluation has been done using the data from the MATLAB simulator with a multiple-sink model of conventional methods. There are three sinks deployed on the water’s surface, and 400 sensor nodes were deployed randomly in a three-dimensional space of 500 m × 500 m × 500 m for 1,000 s of simulation. The simulated work is performed five times and uses the average of the readings to produce graphical results. The main simulation parameters and their values are abstracted in Table 2.

5.2 Experimental results and discussion

In this work, the performance of VHARD-FS, DBR, and DBMR is evaluated in terms of total energy consumption of the network, dead nodes (stability period), throughput, control packets overhead, data delivery delay, and average packet loss. Figure 4 shows the simulation results regarding the network’s total energy consumption per node among VHARD-FS, DBR, and DBMR approach. The total network energy consumption is given by accumulating the energy consumed by each node during the network lifetime. Figure 4 clearly shows that the DBR approach records a higher energy consumption than the proposed VHARD-FS. The highest energy consumption in DBR has resulted from the transmission of more redundant packets. The energy consumption of the proposed VHARD-FS strategy is approximated to DBR and DBMR at the first 200 s because of the need for further energy to the initial phase represented by discovering and establishing a link with a potential neighbor. However, the proposed routing strategy (VHARD-FS) consumes less energy over network simulation time, and such superiority is attributed to considering the node performance ranking index (Node-PRI) including residual energy, node depth, and void-indicator selecting the energy-efficient and reliable paths.

Figure 4

Comparison of total network energy consumption per node.

Figure 5 shows the stability period of VHARD-FS, which effectively increases the network stability period by reducing the number of dead nodes and outperforms both DBR and DBMR during every time step in a simulation. The stability period in both DBR and DBMR is influenced by considering the depth metric only, which leads to the premature death of the low depth nodes because of the high data forwarding rate on these nodes. The superiority of the VHARD-FS strategy in terms of the network stability period has attributed to its consideration of the node performance ranking index. Therefore, it balances energy consumption and lowers the number of dead nodes in the network.

Figure 5

Comparison of total number of dead nodes.

Figure 6 shows the network throughput, which is computed as the total number of packets received successfully at the sink node in a specific period. It is seen that a high throughput was produced by the DBR and DBMR during the first 400 s. This result is because of producing a large number of redundant data by these approaches. On the contrary, the VHARD-FS strategy reduces the number of data packets because of adopting the clustering mechanism, which results in a high aggregate rate by CH nodes. However, with network simulation time passing, the results have proved that the throughput of VHARD-FS is showing stable throughput which is clearly better than DBR and DBMR which are declining with increase in time. However, with network simulation time progression, the results show that throughput of VHARD-FS shows clearly better stable throughput than DBR and DBMR which suffer from failure of some of its nodes and void-holes in the network.

Figure 6

Comparison of network throughput.

Figure 7 shows the number of control packets required to establish and keep the routing structure between the source nodes and sinks. As shown, the experimental results on the first 200 s showed that the numbers of control packets required in DBR, DBMR, and VHARD-FS were relatively similar. As the simulation time progresses, the VHARD-FS requires fewer control packages, the control packets required by VHARD-FS to send the data from the source nodes toward the sinks were 35.22 and 23.51% less of the control packets used by DBR and DBMR, respectively.

Figure 7

Comparison of control packets overhead.

Figure 8 shows that the VHARD-FS has obtained a relatively low end-to-end data delivery delay across the simulation time. The calculation of delay covers the duration from the production of the packet at a sensing node until the packet is delivered successfully at any of the sinks. The void area in the routing path of DBR and DBMR approaches can dramatically increase the end-to-end delay in the network duty. The improved performance of VHARD-FS in terms of end-to-end data delivery delay is mainly attributed to the void-handling metric on next-hop selection, which covers all bridge nodes between the CHs and the sinks.

Figure 8

Comparison of data delivery delay.

The average number of packets lost because of collisions and void-holes occurrence during packets transmissions has been also analyzed to compare VHARD-FS, DBR, and DBMR. In Figure 9, the VHARD-FS generated the minimum average packet loss rates during the network lifetime, which ranged from 0.02% to 0.18%. By contrast, the other approaches have generated 0.09–0.38% average packet loss rates as VHARD-FS avoids the void-holes by continuously tracking the existing path starting from source and ending to the destination.

Figure 9

Comparison of average packet loss.