Abstract
Wireless sensor network (WSN) has become part of human life as it is used in several applications including healthcare, environment and agricultural, public safety, military, transportation as well as in the industry. In spite of its usefulness, it is challenging to maintain long-term operations due to limited battery life. Several energy efficient protocols have been designed to prolong the network lifetime. The integration of mobility technology with the conventional static sensor network, described as hybrid WSN, promises a new solution that balances energy consumption among sensor nodes and extends the network lifetime. To the best of our knowledge, there has not been as yet an evaluation of the energy-efficiency of the data collection approaches in terms of the energy conservation techniques adopted. In this paper, the architecture of data collection approaches in WSN is discussed. Then, we propose and discuss a taxonomy of types of data collection in WSN. We further present and discuss in details a thematic taxonomy of energy conservation techniques adopted in the various hybrid WSN data collection approaches. Consequently, we compare the different energy conservation approaches that minimize energy consumption in hybrid WSN, highlighting their pros and cons. In conclusion, we point out open research challenges and future directions in the field.
Similar content being viewed by others
1 Introduction
Wireless sensor networks (WSN) refers to a collection of several tiny sensor nodes that are distributed across a region of interest, to autonomously sense, process and transmit required data to sink or base station. The transmission link can be direct or in multi-hop structure. The advances in research into microelectronics, sensors and wireless communications technologies have enabled more wide-ranging applications of WSN. Areas WSN can be applied include military surveillance, precision agriculture, environmental and disaster monitoring. It may also be applied in wildlife animal protection, structural monitoring, target tracking, home automation and health care [2, 3, 8, 100].
Sensor nodes are low cost, energy constraint devices with low memory and bandwidth whose resource consumption is application dependent. The nodes are usually deployed in remote locations and thus, difficult if not impossible to recharge their power sources. Meanwhile, depending on the application, sensor nodes may be required to work from several weeks to years. The need to design energy-efficient data collection protocols to prolong the lifetime of WSN has been a thought-provoking issue. It has led to the evolution of Hybrid WSN [95, 98] or Mixed WSN [48]. It involves the combination of mobile elements and static sensor nodes to cooperatively collect information within a network field [85, 92]. The mobile element, such as a sink, tours around network field to gather data from static sensor nodes. By this, static nodes are relieved of their routing duties to base station, which consumes a significant amount of the nodes’ energy [33]. It also solves the hot spot energy-hole problems in WSNs. In this case, the static nodes can concentrate on sensing for physical parameters.
There are several pieces of research that use mobile sinks to enhance data collection performance in WSN. Khan et al. [42], have presented a general survey of schemes that utilize mobile sinks for data collection in WSN. The study compared the schemes based on their goals, but it did not do a thorough evaluation of schemes in terms of their energy-conserving techniques. Francesco et al. [20], present a valuable taxonomy of mobile elements data collection architecture based on their rolls. However, the study does not analyze the energy saving methods of the data collections schemes.
Our aim is to review energy-efficient data collection approaches in hybrid WSN, evaluate the different energy conservation techniques that each approach have adopted to minimize the energy consumption. Then, we analyze and point out the type of data collection that each of the approaches can best fit. We believe this will help guide users to select the appropriate energy conservation type that works well for their type of application. We also describe the types of data collection approaches in hybrid WSN. As part of our contribution, we present a taxonomy of WSN data collection approaches. Then, we reviewed and summarized recent data collection approaches in hybrid WSN, based on their energy conservation technique. Furthermore, we highlight the strengths and weaknesses of the different approaches in the light of energy conservation technique adopted. Where possible, we give suggestions as to address the observed weaknesses. Our study is original in that we focus on critical evaluation of energy conservation technique used by recent energy-efficient hybrid WSN data collection approaches.
The rest of this paper is organized as follows: Sect. 2 presents data collection architecture in WSN. Section 3 presents taxonomy of energy conservation approaches with an in-depth discussion of data collection approaches that utilize them. Then, in Sect. 4, a qualitative comparison of the different energy conservation approaches is made. In Sect. 5, open research challenges and future directions are represented. Finally, Sect. 6 concludes the paper.
2 Data collection architecture in wireless sensor network
Data collection in WSN involves the deployment of a large set of sensor nodes that have the ability to sense specific events around their neighborhood and to communicate with adjacent neighbor nodes for onward transmission to a base station. The acquired data is transmitted in multi-hop routing structure to a sink via wireless communication links. Then the sink also forwards the data to a base station for processing and analysis [50, 79]. The sink may be another sensor node (a super node or computer) while the base station is always a computer located close to the monitoring area or located at a remote place. Because of the ability to deploy sensors nodes at remote locations, the WSN technology has been used in a wide range of applications. Some of the application areas include monitoring the environment [49], health [11], smart agriculture [26, 52] and radiation detection [29]. Figure 1 shows a classical sketch of data collection in WSN.
Depending on the application requirement, the sink can be static, in which case nodes will transmit their data to it. There can also be mobile sinks in which case the sink visits nodes to collect data. The sink may undergo one of three different kinds of mobility, i.e., random, deterministic and controlled mobility; any of which impacts the phase of the data collection [20]. In Fig. 2, we present a high-level classification of data collection approaches under different kinds of data collection in WSN and sink mobility.
2.1 Data collection with static sink
When both sensor and sink nodes have fix topology locations, the sensed data packets are transmitted in multi-hop paths to reach the static sink as in [7, 9, 17, 37]. The nearest nodes to the sink do not only send their data to the sink, but serve as relay nodes to transmit packets of distant sensor nodes the sink [76]. Consequently, such relay nodes deplete their energy faster and can lead to their early death [66]. Depending on the topology of the network, a single or multiple static sink(s) may be used.
2.1.1 Data collection with single static sinks
Depending on the application, a single sink whose location is known, may be used to collect data. The sensor nodes transmit data to the sink via two ways. Firstly, by a direct communication protocol, whereby each sensor node transmits its data directly to a base station [65]. This approach is, however, energy intensive if the transmission distance between the sensor node and base station is long, which can lead to fast node depletion [21]. The second approach involves sensor nodes acting as routers to relay data from other nodes [65]. This approach is good for small scale purpose and can achieve faster data collection as locations of sensor nodes can be easily computed to enable data forwarding to the sink.
2.1.2 Data collection with multiple static sinks
There are some data collection approaches involving multiple static sinks. Usually, in such cases, the network system is in a large scale, and the sensing field is divided into zones. Sensor nodes in each zone transmit their data to the sink that is associated with them. Each sink gathers all the data from the sensor nodes around its neighborhood; a concept called data aggregation or clustering [13, 14, 72, 93, 99] as well as transmitting the received packets to the base station. When multiple sinks are used, data collection rate is faster, and energy of nodes is conserved since packets are not required to propagate over long distances to reach the sink. Figure 1 demonstrates multiple sinks that cluster data in the data collection process.
2.2 Data collection with mobile sink in WSN
Mobility has been found to be useful when incorporated in WSN for data collection [5, 27, 39, 53, 71, 81]. Mobile sinks are exploited in hybrid WSN data collection approaches for various purposes depending on the application. The main goals include, but not limited to, (1) increase lifetime of sensor nodes—mobile sinks move around to collect data from different locations, thereby (1) balancing the energy consumption among nodes [20, 51, 74, 88], (2) Improve network coverage [4, 18, 64, 103], (3) Connectivity—mobile sinks can maintain connectivity in sparse and partitioned sub-networks [20], (4) increase reliability of data reporting—mobile sinks can visit nodes and collect data directly which reduces contention, collision and message loss [6, 67].
In this paper, since our goal is to evaluate the data collection approaches based on their energy conservation methods. In Sect. 3, prior to describing Energy-efficient mobile-sink based data collection approaches, energy conservation techniques in WSNs are discussed. Then, in Sect. 4, the energy-conserving approaches that seek to prolong network lifetime are described.
Available studies such as in [38, 78, 80], show that there are three strategies for data collection in hybrid WSN. These strategies are as follows:
-
I.
One-hop based data collection strategy [75, 102]. In this strategy, mobile sinks traverse randomly or in planned paths in the sensing field and directly collect data from sensor nodes. The sink can be a bus, mobile robot, autonomous land or aerial vehicles. Such a data collection is possible when both mobile sink and static nodes are within each other’s communication range. Moreover, one-hop based strategy can work well if the network size is small, and the transmission power of sensor nodes is powerful enough to reach the sink. The one-hop based strategy has the advantage of reducing communication traffic by the use of mobile sink for the data collection. In Fig. 3, the mobile sink collects data from the transmitting node using the one-hop based data collection strategy.
-
II.
Multi-hop based data collection strategy [15, 58, 77]. Multi-hop based data collection strategy involves the transmission of data from the source node to the mobile sink through intermediary nodes, instead of a long direct pathway. The function of the intermediary nodes is only to relay cooperatively, the data towards the mobile sink. The strategy is used for data collection if the transmission range of sensor nodes is limited, and the sensor nodes are densely populated in the network. The intermediate nodes only relay the data of distant nodes to the mobile sink without the need for any alterations to the size and form of data. Figure 4 shows the Multi-hop data collection strategy implementation. The source node communicates with the mobile sink via intermediary nodes. That is, \((\hbox {D} \rightarrow \hbox {A})\), \((\hbox {E}\rightarrow \hbox {B})\) or \((\hbox {G}\rightarrow \hbox {F}\rightarrow \hbox {C})\). In each route, the sensor nodes do not make changes to the data, but only relay it to the next node towards the mobile sink.
-
III.
Hierarchy-based data collection strategy [23, 54]. The hierarchy-based data collection is mostly used in applications that involve the use of hundreds of thousands of sensor nodes spread over wide fields. The network is divided into multiple layers and each layer may be organized into clusters or grid [57, 84]. By this arrangement, data gathered at the lower layers are repeatedly clustered in the upper layers until it reaches the mobile sink [16]. Many criteria are available to select the cluster head (CH) such as residual energy, proximity to bottom-level CH, centrality and maximum node degree. However, the CHs may only function as relay nodes, where they forward the gathered data to the upper layers without any alterations to the data. On the other hand, the CH may perform special tasks of fusion, aggregation and compression on the data, before transmitting it to the upper layers. Such data alteration usually decreases the total number of relay packet [61]. Of course, hierarchized data transmission inherently implements Multi-hop communication between the CHs, while the intra-cluster transmission is usually single-hop. However, the reverse is not always true. A “pure” Multi-hop data transmission as described in (II) above, is not necessarily hierarchical, and therefore, does not include data clustering. The hierarchical strategy reduces both total energy consumption and the maximum energy consumption and can better balance the energy consumption among nodes by the use of mobile aggregators [16]. Figure 5 is an example of hierarchical tree data collection strategy to show the logical relationship among clusters at multiple layers in the hierarchy. There are n nodes at layer i where \(i\ge 2\). For instance, nodes at layer 1 gather the required data and transmit it to the mobile sink through layers 2, 3 and 4 nodes. Depending on the application, the upper layers may aggregate the data or not.
2.2.1 Data collection with single mobile sink
It is an approach in which only one mobile sink traverses a sensing area periodically, checks every sensor node it comes close to and collects data from it in direct communication for onward transmission to a base station [43, 44, 83]. The approach allows direct transmissions of data packets to the mobile sink, which prolongs the lifetime of the network. This approach is good for small-scale data collection. However, in large-scale applications, the single sink takes a long time to traverse the entire region of interest. There is a kind of trade-off between energy savings of nodes and data transmission latency resulting in packet loss due to a buffer overflow. Also, the high data delivery latency makes this approach unsuitable for event-driven and time sensitive applications.
The advantage in using single mobile sink is the minimization of the financial cost associated with additional hardware. The disadvantage is that it supports only limited network sizes [86]. Besides, it is not suitable for event-based data collection WSN. Since an event may require attention while the mobile node is located somewhere else. Thus, the single mobile sink is suitable for query-base data collection response time can be compromised.
2.2.2 Data collection with multiple mobile sinks
This approach involves the deployment of two or more mobile sinks in the sensing field. It will move around and collect data directly from sensor nodes. It can also do that via short multi-hop relays. The sinks can be located in different areas of the field to shorten the traversal distance to event locations, as in Yin et al. [97]. Mobile sinks, therefore, need not travel over long geographic distances in large fields to collect data. The approach results in high data throughput [56] and minimizes data collection latency [36] in large scale WSN data collection [94]. Also, it drastically conserves energy of the nodes as they will not have to transmit data over long distances.
The disadvantages of using multiple mobile sinks are the additional financial cost incurred and the additional hardware maintenance to the network system. However, the benefits gained by the use of multiple sinks are huge; the lives and properties saved in such critical mission applications outweigh the cost limitations. An example is using multiple sinks in earthquake disasters and avalanches. In such situations, large areas are covered and time is of the essence. The use of multiple sinks in such situations is highly recommended. Therefore, we think that multiple sinks can and should be used when the application requires them.
3 Energy conservation approaches in hybrid WSN
In this section, we reviewed the existing energy conservation techniques proposed in WSN. Then, a level taxonomy of the techniques is summarized in Fig. 6.
In WSN applications, energy consumption is a fundamental issue owing to limited battery power and long lifetime requirement. A large amount of energy is consumed by the node functional modules: Micro-processor, Transceiver, Sensor, and power supply modules [7]. However, the energy consumption rate depends on the power characteristics of the particular node. For example, the power characteristics of Waspmote by Libelium are different from Crossbow Mica2 motes. Typically, the common issues that run across all nodes are that, the communication module consumes the most energy than the rest [74]. The energy consumed to transmit one bit of data is about the same as that required to process a thousand operations in a typical WSN [69]. Moreover, depending on the particular application, nodes may also contain power-hungry components such as, GPS - for finding their positions, camera - for image capture and so on. Usually, energy-saving techniques target two subsystems: the networking subsystem and sensing subsystem. In the first case, every node tries to manage its energy in its operations, as well as ensuring that networking protocol design is efficient. In the second case, the technique tries to reduce the amount of energy-costly samples [7]. Nevertheless, nodes also lose some amount of energy through its hardware components (i.e., CPU, RF circuitry, radio etc.) regardless of their state. Power management protocols are thus used to activate or deactivate nodes [45], subject to whether the node is in communication state or not. Generally we classified energy conservation approaches into two broad areas, namely energy consumption reduction and battery recharging. We further identified techniques such as clustering, overhead reduction and transmission power control that each approach uses to minimize the energy consumption.
3.1 Energy consumption reduction
Minimizing energy consumption is an important consideration in the design of energy constrained hybrid WSN. The energy consumption approaches reduce the average energy consumption of network by using techniques such as reducing the number of data packets, clustering and cooperative communication [10, 28, 68]. Since the conventional batteries are non-rechargeable, the energy consumption reduction techniques are designed to minimize the energy consumption rate of the sensor nodes and prolong the network lifetime. One such technique is the Topology control (TC). It organizes sensor nodes into a network with guaranteeing connectivity among them [40]. In WSN, according to the energy constrain nature of sensor nodes, using max transmission power for any transmission is not efficient and results in energy depletion of the node. TC thereby allows each sensor nodes to adjust its transmission power to reduce the energy consumption while guaranteeing the original connectivity among them. By integrating TC in hybrid WSN, static sensor nodes conserve significant energy as they do not have to send data to long destinations that would require the use of maximum transmission power. Instead, they only wait for the mobile nodes to come closer before data exchange occurs. Thus, the mobile sink approach enhances the TC technique to conserve energy better and to prolong the network lifetime more. The three main ways to form TC include clustering, sleep wake-up and transmission power control. We have explained in details those three techniques and other energy consumption reduction techniques in the subsequent sections.
In general, energy consumption reduction techniques in mobile WSNs can be classified as in the following sections.
3.1.1 Transmission power control
Reducing the transmission power of a node is believed to reduce the energy cost of transmission and hence conserve energy. Xu et al. [96], have established that reducing the transmission power of sensor nodes reduces the total power consumption of the network. However, this compromise coverage as the transmission signal cannot reach long distance nodes. The reverse is true when the transmission power increases. Thus, adjusting transmission power rightly will improve the energy efficiency of the sensor nodes prolong the network lifetime.
The authors in Madani et al. [60] have proposed power-control routing (PCR) scheme in which ordinary nodes adjust their transmission power to send data packets to a CH. PCR first forms the clusters in which nodes adjusted in a linear combination of remaining energy (E) and identification (I) of the sensor node expressed in Eq. 1.
where \(0<\alpha <\beta <1\). Then when the clustering ends, ordinary nodes send their data packets to the CHs by adjusting their transmission power based on their distance to the CH defined as in Eq. 2.
where n is the signal propagation constant, A is received signal’s strength at a distance of 1m and \(\upsigma _{\mathrm{offset}}\) is a random value of the measured RSSI, which ranges from 0 to 1dBm. Thus, nodes closer to the CH only transmit with reduced value that increases the network lifetime as well as the throughput.
3.1.2 Single-hop communication
Some approaches used single-hop communication in their attempts to reduce energy consumption. For instance, a single-hop transmission from CHs to the mobile sink reduces energy consumption of the network since only a fraction of sensor nodes communicate with the sink. In transmitting data over a large number of hops, intermediate nodes consume energy for transmitting data that they have not generated [30] which reduces the lifetime of the network. Moreover, in Single-hop mode, the nodes do not have to transmit data over several hops to reach the sink. Thus, energy consumption by intermediate nodes is avoided. Furthermore, nodes do not recalculate routes updates to changing positions of mobile sinks, which consumes the energy of nodes. Communication overhead and cost required for routing information is also avoided. Some approaches such as [88], have adopted the single-hop communication technique in their energy efficient data collection method. This type of data collection is useful when the sensor nodes are close to the sink and can reach it directly. It can be noted that the single-hop communication is suitable only in small scale networks.
Multiple mobile sink-based routing (MMSR) algorithm is proposed in Wang et al. [90] to mitigate hot-spots problems during data collection to prolong network lifetime of WSN. Three mobile sinks travel separately along predetermined routes in a circular network as shown in Fig. 7. The network is partitioned into two, an inner concentric circular region of radius r (labeled A) and an outer circular region (labeled B). The outer circle is further sub-portioned into B1, B2,..., B8. One mobile sink travels over the diameter and the others travel over arc trails to collect data from the nodes. Each sensor node maintains two tables: sink table and route table. The first table keeps track of the mobile sink and flag to NEIGHBOR_OF_SINK = TRUE when the mobile sink is within node transmission range, else, FALSE. The second table takes accounts of sensor node’s neighbor information to reach the sink. As the mobile sinks travel back and forth along their tracks, they sojourn at some fixed points and broadcast hello messages to neighbor nodes. Each node that receives the hello message transmits its data to the mobile sink.
In the MMSR approach, three mobile sinks collect data simultaneously from different areas that reduce the data collection latency. Also, as mobile sink travels to sojourn areas to collect data, the transmission distance of sensors shortens which reduces the energy dissipation rate. The two activities show that the approach may be energy efficient. However, mobile sinks broadcast HELLO messages to nodes; this could introduce unnecessary packet redundancy that consumes energy. Again, the method may not be suitable for time-sensitive applications, as nodes keep data until mobile sink comes by for the collection.
An Energy-Efficient Bounded Relay Hop Mobile data gathering (EEBRHM) algorithm is proposed by Chen et al. [12] to prolong the lifetime of WSN. The EEBRHM algorithm builds a shortest path tree (SPT) that is rooted at the sink and cover nodes with minimum hops in the WSN. In the case of large WSN in which partitions may occur, due to death of some nodes or sparse deployment, the algorithm creates virtual roots at the center of each partition. Also, nodes are assigned with ID numbers before deployment, and these numbers are used to calculate separate delay time slots for each node. Then, the number of nodes’ children is used as metric to balance degrees of nodes. Meanwhile, the root (i.e., sink) is assumed to be energy-rich and does not care about balancing its degrees. Each node checks if its parent is not a root, then it looks for a neighbor that has the minimum number of children and at the same height as its parent in its own SPT. The node then joins such a neighbor as its new parent. Figure 8a shows a non-uniform and randomly distributed network whose degrees are not balanced. The algorithm balances this network to degrees of 2 for each node in Fig. 8b. After this stage, EEBRHM decides the nodes to represent as the local data aggregation nodes (LNs), preferring nodes that are closer to the sink due to the shorter distance to the sink. To select the LNs, each sensor node is assigned to one of three large time slots (in seconds) depending on the distance to the sink. Then a further small time (in milliseconds), using the formula \((T_n +ID\times {t^\prime }),\hbox {n}=0,1,2\), is assigned to nodes, in order to prevent them from interference during execution of the algorithm. The nearest node to the sink is assigned an initial larger time slot and has the high chance of selection as LN. Finally, based on TSP algorithm an approximate shortest tour traveling from one node to all LNs is built. A mathematical model for the energy consumption of the LNs is given in (3).
where \(\hbox {D}(\hbox {v}_{\mathrm{i}})\) is the degree of a node in the sub-tree. \(\hbox {c}= (\hbox {E}_{\mathrm{t}}-\hbox {E}_{\mathrm{r}})/\hbox {E}_{\mathrm{r}}\), \(\hbox {E}(\hbox {v}_{\mathrm{i}})\) is the residual energy of one node, \(\hbox {v}_{\mathrm{i}}\) is a sensor node, \(\hbox {E}_{\mathrm{t}}\) is transmission energy, \(\hbox {E}_{\mathrm{r}}\) is receiving energy. Base on the Eq. (20), the network lifetime is maximized by minimizing node degree of each node of a sub-tree.
When a node degree is balanced, load per node is uniform and energy consumption is evenly distributed among the nodes, which prolongs the network lifetime. However, the closest nodes to sink are selected as LNs from whom mobile collector visits to collect data. Depending on node density in the sub-network, the LN may endure a lot of data from its associated nodes that may lead to packet drops and retransmissions. Besides, LN could be stressed to a critical energy point if it always relays data from other nodes to MS. This will deplete its energy and reduce the network lifetime. We believe the approach may be energy-efficient in small and loosely dense WSNs but cannot say same for large and denser WSNs.
A heuristic tour-planning algorithm (TPA) is proposed in Ma et al. [59] for a single mobile collector (M-collector) to traverse and collect data directly from sensors. As the M-collector tours, the algorithm selects a set of points among candidate polling points that match some neighbor sensors. The algorithm estimates the shortest tour on the selected points and the sink and terminates when all sensors are covered. M-collector then moves via the shortest distance to collect data, thereby improving the data collection latency. The algorithm also prolongs the network lifetime, by directly collecting data from the sensor nodes. Furthermore, multiple M-collectors are utilized to solve single-hop data-gathering problem mechanism. In this mechanism, only a single M-collector is employed to visit the transmission range of the data sink. The whole network is partitioned into sub-networks. In each sub-network, an M-collector gathers data from close neighbors in the sub-area. Occasionally, the M-collector transmits its sensed data to another M-collector when both are within reach. Furthermore, data is transmitted to the M-collector that goes to stationery data sink through the relays of other M-collectors. The proposed mechanism minimizes tour length to collect data in single-hops. Such a way conserves the energy of sensor nodes as they avoid long transmissions. Moreover, the authors claim that the algorithm greatly prolonged the network lifetime when compared with schemes that use static data collector. Again, they claim the proposed mechanism performs more than the schemes in which mobile data collector traverse only along straight lines. Based on the above analysis, the proposed data collection mechanism may be energy-efficient.
3.1.3 Multi-hop communication
The view under this technique is that the single-hop data collection may not be practical. The reason being that nodes at a distance greater than one-hop from the sink do not have enough power to send data directly to the sink [2]. Moreover, in the Single-hop mode, distant nodes tend to deplete their battery power faster than other nodes in the network. The reason being that those nodes have higher energy burden due to the long-range transmission, which can result in their early death [47]. Thus, the multi-hop communication is used to conserve energy by transmitting the data at short-range hop-by-hop fashion. In the end, each sensor node only consumes a small amount of energy to transmit via a short range, instead of utilizing maximum energy to transmit across long distance. By this way, the overall energy consumption of the network is balanced. In other words, with multi-hop communication, energy dissipation is distributed among all nodes along the data travel path. Thus, some particular nodes are not burdened in transmitting over long range. Researchers such as [19, 46, 70] have adopted the Multi-hop communication technique in their data collection approaches.
A mobility aided cooperative MIMO, called MACO MIMO model, is proposed in Medhi and sharma [62]. The model ensures balanced and reduced rate of energy consumption of sensor nodes, by consuming minimum transmission energy. The network is dynamically clustered, comprising listener sensors, supervisor sensors and a sink. The approach made some assumptions: both listeners and supervisors are location aware, mobile supervisors are in controlled movement and the sink is aware of locations of both sensors. The sink calculates the dynamic cluster size and assigns a pair of supervisors to each cluster. To start a data collection, the sink sends out a query to each supervisor. Clearly stating their IDs, locations and IDs of listeners the supervisor should contact. When supervisors receive the query, they traverse to their assigned locations and send ACK message \(\langle S_{id},\,Ack\rangle \) to sink to inform about their arrival. At their specific locations, supervisor broadcasts a message enclosing its ID and assigned listeners IDs (\(\langle Lid(1\,to\,n),\,S_{id}\rangle \)) and report this to the sink. A listener, upon receiving the messages selects the supervisor it is associated with and waits to respond to queries only from that supervisor. Listeners adopt sleep/wake-up mechanisms; \(t_{wait}\) and \(t_{sleep}\) times and will stop hearing broadcast messages when it receives the message with its ID. During data gathering, each listener node is within the transmission range of its supervisor and transmits sensed data to supervisor at different time slots and the same rate. Meanwhile, only one of the pair of supervisors in a cluster gathers data in each round, alternatively. Then it correlates, compresses and transmits the fused data to its co-supervisor node. The two adjust the distance between them and cooperatively transmit data to sink or through intermediate supervisors.
The proposed MACO MIMO model has implemented enough energy saving techniques - clustering, the mobility of supervisor nodes, data compression and multi-hops to make it energy efficient. Moreover, simulation results [62] shows that in terms of death rate, almost all the sensor nodes die out evenly in MACO MIMO while nodes die out irregularly in its counterpart approaches. Again, in terms of network lifetime, MACO MIMO runs for almost 4000 rounds while comparative approaches run only for 1400 rounds. Against this backdrop, the MACO MIMO is energy efficient.
A cluster-based virtual MINO and multi-hop technologies are combined to form a transmission scheme for energy-constrained WSN. The scheme called MIHOP (MIMO and Multi-hop) is proposed in Dampu et al. [19]. The multi-hop mode is used to transmit data when the sensor node is within a specific hop count from a sink, and the other nodes transmit using the virtual MIMO mode. The network consists of sensors nodes deployed in a square field and a mobile sink traverses along a fixed crossed paths in a controlled mobility. Mobile sink stops at some data collection locations, where it announces BEACON messages and gather data from the sensor nodes. Each sensor keeps its hop count \(N_{hop}\), which denotes its shortest hop count to the mobile sink. \(N_{hop}\) is initialized to infinity and fixed to 0 on the sink. Then, the sink stops at data collection locations and broadcasts BEACON messages with a maximum number of hops \(M_{H}\) and K. Where K is initialize to 0. Every sensor node that receives BEACON packet adds 1 to K, modernizes its \(N_{hop}\) to \(N_{hop} =min\left\{ {N_{hop} ,K} \right\} \) before retransmitting the BEACON with the new K. Iteration of the process lasts till all nodes are covered. Then, the sink moves to next location where it does broadcast again. Furthermore, sensor nodes whose \(N_{hop}\) is higher than the given \(M_{H}\) will form clusters and transmit their data to the mobile sink via the virtual MIMO mechanism. Other nodes with hop N value not more than \(M_{H}\), transmit via the multi-hop transmission technology. Some sensor nodes that may not be in either of the two networks will use the long-distance SISO transmission mechanism. Simulation results [19] shows that in terms of energy conservation, MIHOP outperforms virtual MIMO by 12.98 %, the multi-hop scheme by 47.55 % and double-string networks by 48.30 %. Such performance makes the scheme energy efficient. However, the mobile sink adopts a broadcast method to announce its presence at every stopping point. Such action is basic, introduces packet redundancy in the network and consumes energy.
Multiple mobile sinks are exploited to present an energy-efficient distance-aware routing (EDAR) algorithm for data collection in Wang et al. [88]. Initially, all mobile sinks are located at one departure point. Figure 9, shows one mobile sink \(MS_{1}\) first leaves the departure point, and then a second one \(MS_{2}\) follows. There are sojourn locations \(SL_{i}\), at some parking positions in a PP_TABLE. Each mobile node, from its sojourn location, will choose where it will stop to collect the data. Except those points, the mobile sink will not collect any data while it is moving. When \(MS_{i}\) is on its way to a parking position, it floods the network with a broadcast message. The message is made up of ARRIVAL_MSG and a NEXT_POSITION_MSG messages. Nodes who receive the message communicate with mobile sink directly if it can reach it and also act as relay nodes of distant nodes to the mobile node. A sensor node i, that cannot reach the mobile sink selects a neighbor node j that has a minimum link cost to it (node i), defined in Eq. (4), to relay the data to mobile sink.
where \(\omega \) is a value between 0 and 1. \(D_f\) and \(E_f\) are distance and energy factor metrics which are also defined as in Eqs. (5) and (6).
where d is the distance between the two nodes, E(j) is the residual energy of sensor nodes and the maximum residual energy can be received by flooding. When mobile sink arrives at the parking station, the surrounding nodes compute their distance to the mobile sink to establish if they are in transmission range of the mobile sink. The nodes then broadcast NEIGHBOR_STATUS message and wait for JOIN_REQUEST reply from other nodes. The neighbor nodes then forward their monitored data directly to mobile sink using a time division multiple access (TDMA) schedules and also stay alive to receive JOIN_REQUEST. Else, if the mobile sink does not receive any message, then it waits for next neighbor node announcement at the next parking station. Then it repeats the process of judging its neighbor nodes and sending JOIN_REQUEST messages. The approach suggests transmission power control, data fusion and unique spreading code mechanism to reduce energy consumption. Handshaking between nodes occurs for reliable data transmission. After the handshake, sensor nodes with enough residual energy transmit data. Else, a REJECT_MSG is sent to the peripheral node so that adjacent node with the highest residual energy transmits the data while the drained node abstains from participating in the next data transmission process.
The EDAR algorithm suggests data-fusion, transmission control and TDMA for sensor nodes before transmitting data to mobile sinks. These clearly reduce the energy consumption of the sensor nodes. However, mobile sinks send broadcast messages to sensor nodes to announce their arrival at the parking stations. Such broadcasts introduce redundant packets that consume the energy of the sensor nodes. Nodes receive ARRIVAL_MSG and NEXT_POSITION_MSG messages, before being allowed to forward data to mobile sinks. Moreover, sensor nodes negotiate handshake to ensure reliable transmission link. These introduce communication overheads that consume the energy of nodes. Given the many energy consumptions characteristics of the method, we hold the view that the algorithm may not be energy efficient.
3.1.4 Clustering
In WSN is organizing sensor nodes into local groups. Each cluster is organized into higher level CHs and several lower layer cluster members. Cluster members transmit sensed data to corresponding CHs that forward packets to base station, either directly or through inter-CHs communication [82]. The method reduces the number of transmissions as well as transmission distance of each node to base station. When implemented, clustering attains scalability, energy efficiency and long-lasting network lifetime in large scale multi-hop WSN environments. Clustering has therefore been widely adopted in several data gathering approaches such as [25].
An energy-efficient multi-sink clustering algorithm (EMCA) and a mobile-sink based energy-efficient clustering algorithm (MECA) are cluster based algorithms proposed in Jin et al. [41]. Two sink mobility based energy-efficient clustering algorithms are studied under home network performance to measure energy consumption and network lifetime, as well as to solve the hot spot problem in WSN. The network is grouped into clusters; each contains a CH (CH). The CH is selected by considering the sensor node with the highest residual energy and smallest ID.
In the EMCA, each CH transmits its aggregated data to the sink that it considers best to save its energy. It considers minimum distance to the sink, for CH to minimize its energy consumption. In this regard, Eq. (7) is used.
The energy consumption that is incurred due to long transmissions can be avoided when a multi-hop routing approach is used. The energy consumption associated with a member node, \(S_i\) transmitting data directly to its \(CH_{s_i}\) is modeled in Eq. (8).
Also, for sensor nodes \(S_i\) that must transmit their data to \(CH_{s_i}\) via relay node \(S_j\), the energy consumption is formulated as:
In the case of MECA, sink node broadcasts its location message once to all sensor nodes, to initialize the operation process. Nodes keep track of the sink’s original location \(P_0\), and after a period interval \(\Delta t\), the nodes can reduce the change angle \(\theta \), using Eq. (10).
From Fig. 10, the new mobile sink location \(P_{\Delta t}\) is determined with the knowledge of initial position \(P_0\), angular velocity v, and angle of displacement \(\theta \). Immediately after the broadcast phase ends, the data collection starts. In this case, the mobile sink sojourns at one location and complete a round of data collection before proceeding to the next position.
Simulation results [41] as claimed by authors, shows that MECA performs better than LEACH in terms of number of nodes alive and average of residual energy. Also, the lifetime of MECA is twice than LEACH. Additionally, the increase in the number of sinks greatly decreases the total energy consumption. Given the analysis made, we agreed that MECA approach is energy efficient.
The uneven clustering (UC) algorithm and mobile sink deterministic mobility strategy are explored in Wang et al. [89]. The network consists of a large static sensor nodes and a single mobile node. The mobile node moves along predefined routes and sojourn at some locations to collect data from CHs. Generated data from sensor nodes are forwarded to the CHs for onward transmission to the base station. These CHs are formed out of competitive distance range \(R_{i}\) to the sink, which is calculated in equation in Eq. (11).
where d is distance to mobile sink.
The calculation may create several candidate CH nodes, but the final ones are selected based on the ones with highest remaining energy and smaller ID. The nodes that are not selected go into sleep mode to save energy. The selected CHs automatically track the sink’s sojourn locations to transmit aggregate data from members to the sink within the sojourn time. When the mobile sink reaches a scheduled sojourn location, it broadcasts arrival message to surrounding neighbors. Then the neighbors will forward their fused data in return within the sojourn time. After the sojourn time, mobile sink moves to the next location until it finishes around.
The UC strategy puts nodes in sleep mode until they are ready to transmit data. Such a strategy significantly reduces energy consumption. The clustering technique also avoids collisions and retransmissions. However, the mobile node broadcasts arrival time to neighbors around the sojourn area before receiving packets from them. Uncontrolled broadcasts introduce serious redundancy problem, which result in energy consumption. In general, except the uncontrolled broadcasts problem, the method appears to be energy efficient.
The MobiCluster protocol, proposed by Konstantopoulos et al. [46] is to reduce network overhead and energy consumption in data collection. MobiCluster maintains balanced energy consumption among sensor nodes and extends lifetime of the network. Mobile sinks (MS) are mounted on buses that traverse along predetermined routes, with a periodic schedule that may pass along perimeter of some isolated “sensor islands”, as illustrated in Fig. 11. The protocol starts with a setup phase, in which the mobile sink makes a trip and periodically broadcast BEACON messages. Sensor nodes who receive the BEACON are organized into clusters. The clusters consist of cluster members and CH, which aggregates and filters the received data from their associated members. Then in the steady phase, sensor nodes, through CHs, send data to assigned local rendezvous nodes (RN) that are located close to route of MS for onward submission to the mobile sink. During this second phase, there could be reselection of a RN and/or CH, if any of the current nodes are in critical energy condition. Before RNs deliver data to the MS, there must be a communication link between them. In this regard, an acknowledgement-based protocol is setup in which the MS periodically broadcasts a POLL packet to announce its arrival and request for data as it travels along its path. Upon receiving the POLL, RN transmits its data to the MS. MS acknowledges each received data packet to inform RN that the data was reliably received. By this way, RN clears its buffer to receive other packets.
The MobiCluster protocol has demonstrated enough energy conservation techniques to make it energy efficient. Firstly, clustering is used to gather data from sensor nodes; which is a proven technique to achieving energy efficiency, reducing channel contention and packet collision [46, 61]. Secondly, it adopts energy balancing strategy by changing RNs and CHs when current ones are at critical energy threshold. However, we strongly argue that the broadcast of BEACON messages by MS during its initialization trip and the subsequent traversals, introduce redundancy and communications overheads into the networks. Such activity compromises with the gains of the energy conservation strategies.
3.1.5 Overhead reduction
Usually, data collection protocols may themselves require information about the network system such as route discovery, route request and route acknowledgement before packet transmission. They may also check for errors during transmission or retransmission due to packet drops or in the multi-hop and multipath packet transmission. All these add overhead to the data collection protocol and consume energy. The multi-hop transmissions of data can as well increase the communication overhead. However, single-hop clustering rather reduces the overhead. The mobile sinks in hybrid WSN relieve the static sensor nodes of their routing duties by coming closer to their locations for direct data collection or within few hops way, thereby conserving energy of the sensor nodes. However, the mobile sinks have to update sensor nodes of their locations information for ease of data transmission.
Many researchers have addressed this issue to reduce the update overhead. For example, the authors in Wang et al. [87] proposed the local update-based routing protocol (LURP) to limit the location updates of mobile sinks within a confine area and not the entire network field. The aim is to minimize the communication overhead. In the protocol, a mobile sink marks a circular area around its present location with its present position and update its range. Then the position and update range are broadcast over the network field. The mobile sink updates only nodes within the destination area that it meets. Nodes outside the destination area transmit the packet via geographical forwarding while the inside nodes transmit their packet via topology-based routing manner. When the sink travels outside its present destination area, it redefines a new area and broadcasts it over the whole network. The authors further suggest an Adaptive LURP to limit the updates inside the destination area that is similar to the LURP except that the distance between the virtual center and present position are the update range, but smaller than the range in LURP. It must be noted that large update range increases the overhead for local updating, and a small update range increases the frequency of global updating.
In Khan et al. [43], the authors tried to reduce route reconstruction cost of sensor nodes by suggesting a virtual grid-based dynamic routes adjustment (VGDRA) scheme for periodic data collection. VGDRA scheme divides the sensing field into a virtual grid of equal sides and forms virtual backbone network and elects cell leaders among the nodes in each cell. The cell leaders collect and transmit data along the backbone paths towards boundary cells. Then the cell leaders interact with the constantly moving mobile sink around a predetermined path at the boundary of the network field. To solve the route overhead due to changing locations of the mobile sink, a set of rules are formulated to guide the route re-adjustment of the sensor nodes. Communication paths are also formed to reduce the latency and energy cost n transmitting data to the mobile sink. In the end, VGDRA achieves near optimal paths the current positions of the mobile sink without much network overhead. The authors only evaluated the efficiency of VGDRA, in hybrid WSN with a single sink.
3.1.6 Data fusion
This technique reduces energy consumption by minimizing the amount of transmitted data across networks to the sink. Various methods exist for this purpose among them includes data aggregation, where only the maximum, average or minimum collected data is transmitted to the sink. There can also be network coding, where broadcast transmissions send a linear combination of some packets and not a copy of each packet to reduce traffic in broadcast scenarios. Thus, unnecessary data transfer is avoided which saves the energy of nodes [73]. The Network Coding over Connected Dominating Set (NCDS) scheme is proposed by Wang et al. [91] to reduce energy consumption of sensor nodes. NCDS first constructs a minimum connected dominating set together then adds network coding to control broadcasting of the nodes. By such combination, NCDS achieves considerable energy gains.
3.1.7 Sleep/wake-up
The technique adjusts the radio state of the node depending to reduce energy consumption of sensor nodes. The radio is switched OFF during inactive state and comes ON when communication is required. Duty cycling, passive wake-up radio and topology control are implementation techniques of the sleep/wake-up scheme. The method is quite effective in conserve energy of sensor node; however, it impacts on the latency of data collection, since receiving node must first be woken up before communication can be established. Also, it is difficult for a sending node to connect to neighbors at the same time since they have different wake up times. Some researchers have tried to improve the performance in terms of memory overflow and latency, by adjusting the active states online.
The connected k-neighbourhood (CKN) algorithm is a duty-cycle sleep/wake-up based technique presented by Han et al. [34]. CKN synchronizes and regulates a duty cycled sleep/wake-up schedule of sensor nodes to prolong network lifetime and increase data collection. In the CKN mechanism, a node chooses a random rank, \(rank_{u}\), announces the \(rank_{u}\) and accepts its neighbors’ ranks \(\hbox {R}_{\mathrm{u}}\). Node u further announces \(\hbox {R}_{\mathrm{u}}\) and receives \(\hbox {R}_{\mathrm{v}}\) from surrounding nodes. If nodes u or its neighbors have less than k neighbors, node u stays awake. Else, node u calculates a subset \(\hbox {C}_{\mathrm{u}}\) of its neighbors that have rank \(<\hbox {rank}_{\mathrm{u}}\). Node u only sleeps when all nodes in \(\hbox {C}_{\mathrm{u}}\) have to be connected by nodes that have rank \(<\hbox {rank}_{\mathrm{u}}\), and every neighbor has at least k neighbors from \(\hbox {C}_{\mathrm{u}}\). Simulation results shows adjusting network parameters, CKN mechanism achieves maximum data collection and extends network lifetime.
3.2 Battery recharging
This category is a recent breakthrough that keeps the battery continually charged to allow endless operations of sensor nodes. This category effectively sustains the liveliness of the node’s battery and improves the network lifetime considerably [24, 31, 55]. So far, the implantation of recharging can be done in two ways. So far, the implantation of recharging can be done in two ways.
3.2.1 Renewable energy harvesting
This process attempts to capture energy from the ambient environment that otherwise would be lost, and convert them to electrical form. It integrates an energy harvesting module to the normal sensor node. The technique relieves sensor nodes of their energy limitation by charging them through power sources such as heat, light, vibrations, wind or electromagnetic energy [1, 63]. It can effectively improve network performance and prolong network lifetime. Many researchers believe this could provide a self-sustainable WSN for the several applications.
The authors in Eu et al. [22] have proposed a multi-hop energy harvesting opportunistic routing (EHOR) for WSN that are powered by energy harvesting devices. In EHOR, nodes are partitioned into regions to minimize latency while increasing throughput. Then transmission priorities are assigned to the partitioned regions based on nearness to sink and remaining energy of the nodes in that region. Simulation indicates EHOR outperforms its peers in terms of efficiency, data delivery, and fairness.
3.2.2 Wireless energy transmission
The technique involves the transmission of energy from a power source to a sensor node by the use of a wireless medium. By this technique, sensor nodes can be recharged without the need for conventional wire or plug. Many techniques abound on the operations of wireless energy transmissions, and interested readers can check [1] for details. In recent times, the wireless energy transmission techniques have been exploited in Hybrid WSN data collection approaches as in [30, 35, 86, 101].
A Joint Mobile Energy Replacement and Data Gathering (J-MERDG) framework in rechargeable WSN is proposed in Zhao et al. [101].The authors exploit mobility for energy replenishment and data gathering to deliver steady and high recharging rates, to achieve energy efficient data gathering from sensors. A mobile element, SenCar, is employed to perform two functions. Firstly, it runs a controlled mobility Re charging tour (R-tour), in which it selects and visits nodes with least remaining energy to recharge them within a given time. It ensures that nodes are protected from complete energy exhaustion. Secondly, it runs data gathering tour (d-tour) in applications where energy is less consumed by nodes. In this situation, SenCar visits anchor points that yield optimal data gathering. While at the data gathering points, nearby nodes are charged.
4 Comparison of energy conservation approaches in hybrid WSN
The need to support the limited battery capacity of sensor nodes to prolong network lifetime of WSN applications is crucial. It has prompted the design of several energy saving schemes which this paper categorized into two. It is worth to note that each technique fits well for a particular application. In the following, we analyze and compare the approaches in line with the type of data collection they best suit.
Table 1 compares the data collection approaches discussed in Sect. 3 based on their important design features. The last two columns show the energy conservation category a protocol is associated with, as well as the particular strategy adopted to reduce energy consumption. Table 1 is interpreted for example; EEBRHM protocol uses the transmission power control technique to reduce energy consumption of sensor nodes and improves the network lifetime.
In the energy consumption reduction approach, sensor nodes consume energy evenly across the network voiding hot-spot problem prolonging network lifetime. A good number of approaches such as [19, 32, 41], have adopted the clustering strategy to reduce energy consumption in their data collection approaches. The clustering technique reduces intra-cluster communication range—which consumes less transmission power. It also aggregates and transmits fused data to sink—which limits the number of transmissions. Furthermore, it facilitates nodes’ sleep/wake-up and allows CHs to take control of data forwarding. Last but not the least, it alternates CHs among homogenous nodes, which balances the energy consumption [73]. The radios of sensor nodes can be in different states; transmitting, receiving, listening, sleeping and idle. In each state, energy consumption is different and considerable amount of energy can be conserved when the radio transceiver is switched off when no data transmission or reception is expected. However, the method interrupts packet transmission process since all nodes may not be ON at the same time. Selecting the appropriate radio switch-off period makes the implementation of this technique difficult. Depending on the application area, the sleep/wake-up cycles of the nodes can be scheduled to be synchronous. In this way, all the nodes can be scheduled to wake-up at the same time, exchange data packets and go back to sleep simultaneously. Nodes can also be scheduled asynchronously, in which case each node wake-up independently, but can still be able to communicate with other nodes. The main issue with this method is that nodes that have data to send will have to wait until receiver nodes wake-up before the communication occurs. Such an action delays the data collection process, and not suitable event-based data collection that requires real-time data delivery. The challenge is to schedule the sleep/wake-up periods that can reduce the energy consumption while it meets the application needs which is a research challenge.
Some different approaches use both single hop and multi-hop communications depending on requirement needs. Multi-hop communication saves energy when nodes engage in short-range, hop-by-hop transmissions. It consumes less energy than long direct hop transmission [70]. Moreover, energy dissipation is distributed among all nodes along the routing path and does not burden one particular node to death. Approaches that have employed this method in their data collection include [19, 46]. However, in a large scale data collection, sensor nodes may have to communicate their packets over several hops towards the mobile sink. Besides, nodes may always need to recalculate their routes to reach the changing locations of the mobile sinks, thereby dissipating more energy and may render the amount of conserved energy ineffective. The transmission power control technique effectively improves the energy efficiency and significantly prolongs the network lifetime of WSN. Nodes do not have the same transmission power; instead each node determines its transmission power according to its relative position to its neighbors. It behavior is like topology control protocol and chooses that neighbor it can communicate with to spend less energy. The technique reduces the overall transmission power consumption and can serve well in both event-driven and query driven data collection.
The use of rechargeable batteries as energy reservoirs for sensor nodes operations has been commonly used and promises an end to the energy constraints in WSN. However, there are issues that pose challenges to their use.
In the first place, integrating renewable energy harvesting requires fixing an energy harvesting module, which increases cost and bulkiness of tiny sensor nodes. Besides, it requires considerable research effort. Recharging via light that is commonly used depends on the solar panel that limits its usage in indoor applications where there is no direct sunlight. Even in the case of outdoor applications, the presence of sunlight is not always available. The method is not suitable for power-hungry applications, such as surveillance cameras. The effectiveness of sensor nodes’ output power relies on the size of the solar panel and increasing the size will compromise with the tiny nature of the sensor node. Efficient power conversion solar panels are expensive and defeat the very theme of sensor nodes as low cost. Similarly, the other recharging methods also have similar challenges that render their usage difficult in WSN. Secondly, a wireless energy transmissions recharging method, such as inductive coupling, appears to lack power transmission range and its improved form, the magnetic coupling, also performs at low efficiency with increasing distance [1]. Using wireless energy transmissions recharging technique in large scale WSN application is a challenging issue. The use of a mobile element for recharging of the nodes is a better alternative to providing reliable and better solution. However, the locomotion energy of the mobile entity is also a challenge since energy expensed due to the locomotion may render the energy conserved in the network operation negligible.
In spite of the challenges of the mobile element recharging technique, this technique remains a sustainable alternative. Using mobile recharging entities that can traverse the locations of the sensor nodes to recharge them will provide an easier and efficient way. Even though the sensor recharging is the best way forward, reducing the energy consumption remains fundamental to the energy efficiency of WSN. The method can support all forms of data collection, be it event-driven or query driven.
We also observe, In Table 1, that most of the data collection approaches have adopted one type of energy conservation technique or another in their protocols. For instance, while the EEBRHM, MMSR, and MMSR have adopted Single-hop transmission, the EDARA and MIHOP have utilized the Multi-hop, and the UC and MobiCluster have considered the Clustering technique.
5 Open research challenges
Hybrid WSN data collection brings enormous relive to the energy consumption of sensor nodes. But for effective hybrid WSN data collection, it is crucial for sensor nodes to send data to moving nodes without incurring localization overhead that is an important challenge and thus open up for future research.
Most of the approaches apply well in query-based or periodic data collection application. Selecting appropriate approaches that can work well in event-based data collection application is challenging, and more so, dealing with the overhead costs associated with it.
Incorporating renewable energy into hybrid WSN is still at the very early stages and opens up for more research efforts. The size of a solar panel is important for the amount of energy harvested, yet the appropriate solar panel sizes that will fit the current tiny sensor nodes impose a bigger challenge that requires huge research effort. Furthermore, the additional hardware to sensor nodes make the low-cost sensor nodes expensive and so more research work is needed to bring the lower the cost.
Although the wireless energy charging appears the best solution provider, it requires a close contact between the charger and the sensor nodes, which limits the strategy from being in large scale and randomly deployed WSN. Long and non-line of sight charging of nodes is a challenging issue that requires more research effort.
We observe that the broadcasting of packets is very common in hybrid data collection approaches. However, it contributes huge overhead to the network system as all nodes tend to relay packets frequently and redundantly across the entire network. Again, the movement of mobile sinks introduces frequent update of their positions to the sensor nodes. An efficient data collection protocol in WSN must avoid broadcasting packet across the network, or at least effectively control it. Such a capability saves both the energy as well as the bandwidth that are valuable resources in WSN. Also, transmitting smaller packet sizes across short-range routes can reduce overhead in the network. If the only nodes with higher residual energy above a certain threshold level are allowed to forward the data packets, the protocol can balance the energy consumption among the nodes. Last but not the least; the protocol can reduce the energy consumption due to the overhead associated with the location update of mobile sinks. To design a protocol that incorporates all these capabilities is an important challenge worth addressing.
6 Conclusions
The integration of static and mobile sensors nodes to form hybrid WSN is a recent frontier of improving energy-efficient data collection to prolong network lifetime of WSN applications. It balances traffic load, minimizes transmission range and allow static sensor nodes to concentrate on their main function of sensing. We have provided a review of hybrid WSN data collection approaches and their architecture. Then we developed taxonomy of the different types of data collection approaches. Furthermore, we have introduced an original taxonomy of energy conservation approaches in hybrid WSN. Then we reviewed and summarized hybrid WSN data collections approaches that integrate the different techniques in their operations. Again, we presented a qualitative comparison of the various energy conservation approaches and highlighted the pros and cons of each. Additionally, we have presented an evaluation of energy-efficiency of the various data collection approaches and remark on their strengths and weakness to prolonging the lifetime of hybrid WSN.
References
Akhtar, F., & Rehmani, M. H. (2015). Energy replenishment using renewable and traditional energy resources for sustainable wireless sensor networks: A review. Renewable & Sustainable Energy Reviews, 45, 769–784.
Akkaya, K., & Younis, M. (2005). A survey on routing protocols for wireless sensor networks. Ad Hoc Networks, 3(3), 325–349.
Akyildiz, I. F., Su, W., Sankarasubramaniam, Y., & Cayirci, E. (2002). Wireless sensor networks: A survey. Computer Networks, 38(4), 393–422.
Ammari, H. M. (2012). On the problem of K-coverage in mission-oriented mobile wireless sensor networks. Computer Networks, 56(7), 1935–1950.
Ammari, H. M. (2013). Joint K-coverage and data gathering in sparsely deployed sensor networks: Impact of purposeful mobility and heterogeneity. ACM Transactions on Sensor Networks, 10(1), 8.
Anastasi, G., Borgia, E., Conti, M., & Di Francesco, M. (2011). Reliable data delivery in sparse wsns with multiple mobile sinks: An experimental analysis. In 2011 IEEE Symposium on Computers and Communications (ISCC) (pp. 698–705).
Anastasi, G., Conti, M., Di Francesco, M., & Passarella, A. (2009). Energy conservation in wireless sensor networks: A survey. Ad Hoc Networks, 7(3), 537–568.
Anisi, M. H., Abdul-Salaam, G., & Abdullah, A. H. (2015). A survey of wireless sensor network approaches and their energy consumption for monitoring farm fields in precision agriculture. Precision Agriculture, 16(2), 216–238.
Anisi, M. H., Abdullah, A. H., Coulibaly, Y., & Razak, S. A. (2013). Edr: Efficient data routing in wireless sensor networks. International Journal of Ad Hoc and Ubiquitous Computing, 12(1), 46–55.
Anisi, M. H., Abdullah, A. H., Razak, S. A., & Ngadi, M. A. (2012). An overview of data routing approaches for wireless sensor networks. Sensors, 12(4), 3964–3996.
Benharref, A., & Serhani, M. A. (2014). Novel cloud and SOA-based framework for E-health monitoring using wireless biosensors. IEEE Journal of Biomedical and Health Informatics, 18(1), 46–55.
Chen, L., Wang, J., Peng, X., & Kui, X. (2015). An energy-efficient and relay hop bounded mobile data gathering algorithm in wireless sensor networks. International Journal of Distributed Sensor Networks.
Chen, S., Coolbeth, M., Dinh, H., Kim, Y.-A., & Wang, B. (2009). Data collection with multiple sinks in wireless sensor networks. In B. Liu, A. Bestavros, D.-Z. Du, & J. Wang (Eds.), Wireless algorithms, systems, and applications (pp. 284–294). Berlin/Heidelberg: Springer.
Chen, Y.-L., Wang, N.-C., Shih, Y.-N., & Lin, J.-S. (2013). Improving low-energy adaptive clustering hierarchy architectures with sleep mode for wireless sensor networks. Wireless Personal Communications, 75, 1–20.
Chen, Y., Tang, Y., Xu, G., Qian, H., & Xu, Y. (2011). A data gathering algorithm based on swarm intelligence and load balancing strategy for mobile sink. In 2011 9th World Congress on Intelligent Control and Automation (WCICA) (pp. 1002–1007).
Chen, Y. P., Liestman, A. L., & Liu, J. (2006). A hierarchical energy-efficient framework for data aggregation in wireless sensor networks. IEEE Transactions on Vehicular Technology, 55(3), 789–796.
Cheng, C.-T., & Tse, C. K. (2013). A delay-aware network structure for wireless sensor networks with consecutive data collection processes. IEEE Sensors Journal, 13(6), 2413–2422.
Costa, D. G., & Guedes, L. A. (2010). The coverage problem in video-based wireless sensor networks: A survey. Sensors, 10(9), 8215–8247.
Danpu, L., Kailin, Z., & Jie, D. (2013). Energy-efficient transmission scheme for mobile data gathering in wireless sensor networks. Communications, China, 10(3), 114–123.
Di Francesco, M., Das, S. K., & Anastasi, G. (2011). Data collection in wireless sensor networks with mobile elements: A survey. ACM Transactions on Sensor Networks, 8(1), 7.
Ebrahimi, D., & Assi, C. (2014). Compressive data gathering using random projection for energy efficient wireless sensor networks. Ad Hoc Networks, 16, 105–119.
Eu, Z. A., Tan, H.-P., & Seah, W. K. (2010). Opportunistic routing in wireless sensor networks powered by ambient energy harvesting. Computer Networks, 54(17), 2943–2966.
Faheem, Y., Boudjit, S., & Chen, K. (2011). Dynamic sink location update scope control mechanism for mobile sink wireless sensor networks. In 2011 8th International Conference on Wireless On-Demand Network Systems and Services (WONS) (pp. 171–178).
Fan, K.-W., Zheng, Z., & Sinha, P. (2008). Steady and fair rate allocation for rechargeable sensors in perpetual sensor networks. In Proceedings of the 6th ACM Conference on Embedded Network Sensor Systems (pp. 239–252).
Farris, I., Militano, L., Iera, A., Molinaro, A., & Spinella, S. C. (2015). Tag-based cooperative data gathering and energy recharging in wide area rfid sensor networks. Ad Hoc Networks.
Fernandes, M. A., Matos, S. G., Peres, E., Cunha, C. R., Lopez, J. A., Ferreira, P. J. S. G., et al. (2013). A framework for wireless sensor networks management for precision viticulture and agriculture based on IEEE 1451 standard. Computers and Electronics in Agriculture, 95, 19–30.
Ganeriwal, S., Kansal, A., & Srivastava, M. B. (2004). Self aware actuation for fault repair in sensor networks. robotics and automation, 2004. In IEEE International Conference on Proceedings ICRA’04. 2004 (pp. 5244–5249).
Garcia, M., Sendra, S., Lloret, J., & Canovas, A. (2013). Saving energy and improving communications using cooperative group-based wireless sensor networks. Telecommunication Systems, 52(4), 2489–2502.
Gomaa, R., Adly, I., Sharshar, K., Safwat, A., & Ragai, H. (2013). Zigbee wireless sensor network for radiation monitoring at nuclear facilities. In 2013 6th Joint IFIP Wireless and Mobile Networking Conference (WMNC) (pp. 1–4).
Gu, Y., Bozdag, D., Ekici, E., Özgüner, F., & Lee, C.-G. (2005). Partitioning based mobile element scheduling in wireless sensor networks. In SECON (pp. 386–395).
Guo, S., Wang, C., & Yang, Y. (2014). Joint mobile data gathering and energy provisioning in wireless rechargeable sensor networks. IEEE Transactions on Mobile Computing, 13(12), 2836–2852.
Ha, I., Djuraev, M., & Ahn, B. (2014). An energy-efficient data collection method for wireless multimedia sensor networks. International Journal of Distributed Sensor Networks.
Halder, S., & Das, B. S. (2014). Enhancement of wireless sensor network lifetime by deploying heterogeneous nodes. Journal of Network and Computer Applications, 38, 106–124.
Han, G., Guo, H., Zhang, C., & Shu, L. (2014). Parameter optimisation in duty-cycled wireless sensor networks under expected network lifetime. International Journal of Ad Hoc and Ubiquitous Computing, 15(1), 57–67.
He, L., Fu, L., Zheng, L., Gu, Y., Cheng, P., Chen, J., & Pan, J. (2014). Esync: An energy synchronized charging protocol for rechargeable wireless sensor networks. In Proceedings of the 15th ACM International Symposium on Mobile Ad Hoc Networking and Computing (pp. 247–256).
Hu, Y., Ding, Y., Hao, K., Ren, L., & Han, H. (2014). An immune orthogonal learning particle swarm optimisation algorithm for routing recovery of wireless sensor networks with mobile sink. International Journal of Systems Science, 45(3), 337–350.
Incel, O. D., Ghosh, A., Krishnamachari, B., & Chintalapudi, K. (2012). Fast data collection in tree-based wireless sensor networks. IEEE Transactions on Mobile Computing, 11(1), 86–99.
Iwanicki, K., & Van Steen, M. (2009). Multi-hop cluster hierarchy maintenance in wireless sensor networks: A case for Gossip-rased protocols. In U. Roedig & C. J. Sreenan (Eds.), Wireless sensor networks, proceedings (pp. 102–117).
Jawhar, I., Mohamed, N., Al-Jaroodi, J., & Zhang, S. (2014). A framework for using unmanned aerial vehicles for data collection in linear wireless sensor networks. Journal of Intelligent and Robotic Systems: Theory and Applications, 74(1–2), 437–453.
Jia, L., Rajaraman, R., & Scheideler, C. (2003). On local algorithms for topology control and routing in ad hoc networks. In Proceedings of the 15th Annual ACM Symposium on Parallel Algorithms and Architectures (pp. 220–229).
Jin, W., Yue, Y., Jianwei, Z., Sungyoung, L., & Sherratt, R. S. (2013). Mobility based energy efficient and multi-sink algorithms for consumer home networks. IEEE Transactions on Consumer Electronics, 59(1), 77–84.
Khan, A. W., Abdullah, A. H., Anisi, M. H., & Bangash, J. I. (2014). A comprehensive study of data collection schemes using mobile sinks in wireless sensor networks. Sensors, 14(2), 2510–2548.
Khan, A. W., Abdullah, A. H., Razzaque, M. A., & Bangash, J. I. (2015a). Vgdra: A virtual grid-based dynamic routes adjustment scheme for mobile sink-based wireless sensor networks. IEEE Sensors Journal, 15(1), 526–534.
Khan, A. W., Abdullah, A. H., Razzaque, M. A., Bangash, J. I., & Altameem, A. (2015b). Vgdd: A virtual grid based data dissemination scheme for wireless sensor networks with mobile sink. International Journal of Distributed Sensor Networks.
Khan, J. A., Qureshi, H. K., & Iqbal, A. (2015c). Energy management in wireless sensor networks: A survey. Computers & Electrical Engineering, 41, 159–176.
Konstantopoulos, C., Pantziou, G., Gavalas, D., Mpitziopoulos, A., & Mamalis, B. (2012). A Rendezvous-based approach enabling energy-efficient sensory data collection with mobile sinks. IEEE Transactions on Parallel and Distributed Systems, 23(5), 809–817.
Kumar, D., & Patel, R. (2011). Multi-hop data communication algorithm for clustered wireless sensor networks. International Journal of Distributed Sensor Networks.
Lambrou, T. P., & Panayiotou, C. G. (2013). Collaborative path planning for event search and exploration in mixed sensor networks. The International Journal of Robotics Research, 32(12), 1424–1437.
Larios, D. F., Barbancho, J., Sevillano, J. L., Rodriguez, G., Molina, F. J., Gasull, V. G., et al. (2013). Five years of designing wireless sensor networks in the Donana biological reserve (Spain): An applications approach. Sensors, 13(9), 12044–12069.
Levin, L., Efrat, A., & Segal, M. (2014). Collecting data in ad-hoc networks with reduced uncertainty. Ad Hoc Networks, 17, 71–81.
Li, J., & Mohapatra, P. (2007). Analytical modeling and mitigation techniques for the energy hole problem in sensor networks. Pervasive and Mobile Computing, 3(3), 233–254.
Li, Z., Wang, N., Franzen, A., Taher, P., Godsey, C., Zhang, H., et al. (2014). Practical deployment of an in-field soil property wireless sensor network. Computer Standards & Interfaces, 36(2), 278–287.
Liang, W., Luo, J., & Xu, X. (2013). Network lifetime maximization for time-sensitive data gathering in wireless sensor networks with a mobile sink. Wireless Communications & Mobile Computing, 13(14), 1263–1280.
Lin, C.-J., Chou, P.-L., & Chou, C.-F. (2006). Hcdd: Hierarchical cluster-based data dissemination in wireless sensor networks with mobile sink. In Proceedings of the 2006 International Conference on Wireless Communications and Mobile Computing (pp. 1189–1194).
Liu, R.-S., Fan, K.-W., Zheng, Z., & Sinha, P. (2011). Perpetual and fair data collection for environmental energy harvesting sensor networks. IEEE/ACM Transactions on Networking, 19(4), 947–960.
Liu, W., Lu, K., Wang, J., Huang, L., & Wu, D. O. (2012). On the throughput capacity of wireless sensor networks with mobile relays. IEEE Transactions on Vehicular Technology, 61(4), 1801–1809.
Lu, K.-H., Hwang, S.-F., Su, Y.-Y., Chang, H.-N., & Dow, C.-R. (2012). Hierarchical ring-based data gathering for dense wireless sensor networks. Wireless Personal Communications, 64(2), 347–367.
Luo, H., Ye, F., Cheng, J., Lu, S., & Zhang, L. (2005). Ttdd: Two-tier data dissemination in large-scale wireless sensor networks. Wireless Networks, 11(1–2), 161–175.
Ma, M., Yang, Y., & Zhao, M. (2013). Tour planning for mobile data-gathering mechanisms in wireless sensor networks. IEEE Transactions on Vehicular Technology, 62(4), 1472–1483.
Madani, S. A., Hayat, K., & Khan, S. U. (2012). Clustering-based power-controlled routing for mobile wireless sensor networks. International Journal of Communication Systems, 25(4), 529–542.
Mamalis, B., Gavalas, D., Konstantopoulos, C., & Pantziou, G. (2009). Clustering in wireless sensor networks. In Y. Zhang, L. T. Yang, & J. Chen (Eds.), RFID and sensor networks: architectures, protocols, security and integrations (pp. 324–353).
Medhi, N., & Sarma, N. (2012). Mobility aided cooperative mimo transmission in wireless sensor networks. In S. K. Jena & B. Majhi (Eds.), 2nd International Conference on Communication, Computing & Security (ICCCS-2012) (pp. 362–370).
Moon, J., & Leeb, S. B. (2015). Analysis model for magnetic energy harvesters. IEEE Transactions on Power Electronics, 30(8), 4302–4311.
Munir, S. A., Ren, B., Jiao, W., Wang, B., Xie, D. and Ma, J. (2007). Mobile wireless sensor network: architecture and enabling technologies for ubiquitous computing. In 21st International Conference on Advanced Information Networking and Applications Workshops, 2007, AINAW’07 (pp. 113–120).
Narendra, K., & Varun, V. (2014). A comparative analysis of energy-efficient routing protocols in wireless sensor networks. In Emerging Research in Electronics, Computer Science and Technology (pp. 399–405). Springer.
Nazir, B., & Hasbullah, H. (2013). Energy efficient and Qos aware routing protocol for clustered wireless sensor network. Computers & Electrical Engineering, 39(8), 2425–2441.
Ngai, E.-H., Zhou, Y., Lyu, M. R., & Liu, J. (2006). Reliable reporting of delay-sensitive events in wireless sensor-actuator networks. In 2006 IEEE International Conference on Mobile Adhoc and Sensor Systems (MASS) (pp. 101–108).
Ortiz, A. M., Royo, F., Olivares, T., Castillo, J. C., Orozco-Barbosa, L., & Marron, P. J. (2013). Fuzzy-logic based routing for dense wireless sensor networks. Telecommunication Systems, 52(4), 2687–2697.
Pottie, G. J., & Kaiser, W. J. (2000). Wireless integrated network sensors. Communications of the ACM, 43(5), 51–58.
Qiuling, T., Changyin, S., Huan, W., & Ye, L. (2010). Cross-layer energy efficiency analysis and optimization in WSN. In 2010 International Conference on Networking, Sensing and Control (ICNSC) (pp. 138–142). Accessed 10–12 Apr 2010.
Rahimi, M., Shah, H., Sukhatme, G., Heideman, J., & Estrin, D. (2003). Studying the feasibility of energy harvesting in a mobile sensor network. In Proceedings of ICRA ’03. IEEE International Conference on Robotics and Automation, 2003 (vol. 11, pp. 19–24). Accessed 14–19 Sept 2003.
Ranjani, S. S., Krishnan, S. R., Thangaraj, C., & Devi, K. V. (2013). Achieving energy conservation by cluster based data aggregation in wireless sensor networks. Wireless Personal Communications, 73(3), 731–751.
Rault, T., Bouabdallah, A., & Challal, Y. (2014). Energy efficiency in wireless sensor networks: A top-down survey. Computer Networks, 67, 104–122.
Sara, G. S., & Sridharan, D. (2014). Routing in mobile wireless sensor network: A survey. Telecommunication Systems, 57(1), 51–79.
Shah, R. C., Roy, S., Jain, S., & Brunette, W. (2003). Data Mules: Modeling and analysis of a three-tier architecture for sparse sensor networks. Ad Hoc Networks, 1(2–3), 215–233.
Shankar, T., & Shanmugavel, S. (2014). Energy optimization in cluster based wireless sensor networks. Journal of Engineering Science and Technology, 9(2), 246–260.
Shi, L., Zhang, B., Huang, K., & Ma, J. (2011). An efficient data-driven routing protocol for wireless sensor networks with mobile sinks. In 2011 IEEE International Conference on Communications (ICC) (pp. 1–5).
Shin, I., Kim, M., Mutka, M. W., Choo, H., & Lee, T.-J. (2009). Mcbt: Multi-hop cluster based stable backbone trees for data collection and dissemination in WSNs. Sensors, 9(8), 6028–6045.
Shrivastava, P., & Pokle, S. B. (2014). Energy efficient scheduling strategy for data collection in wireless sensor networks, pp. 170–173.
Swami, A., Zhao, Q., Hong, Y.-W., & Tong, L. (2007). Wireless sensor networks: Signal processing and communications. New York: Wiley.
Tseng, Y.-C., Wu, F.-J., & Lai, W.-T. (2013). Opportunistic data collection for disconnected wireless sensor networks by mobile Mules. Ad Hoc Networks, 11(3), 1150–1164.
Tyagi, S., & Kumar, N. (2013). A systematic review on clustering and routing techniques based upon leach protocol for wireless sensor networks. Journal of Network and Computer Applications, 36(2), 623–645.
Tzung-Cheng, C., Tzung-Shi, C., & Ping-Wen, W. (2011). On data collection using mobile robot in wireless sensor networks. IEEE Transactions on Systems, Man and Cybernetics, Part A: Systems and Humans, 41(6), 1213–1224.
Van Le, D., Oh, H., & Yoon, S. (2014). Hicodg: A hierarchical data-gathering scheme using cooperative multiple mobile elements. Sensors, 14(12), 24278–24304.
Verma, A., Sawant, H., & Tan, J. (2006). Selection and navigation of mobile sensor nodes using a sensor network. Pervasive and Mobile Computing, 2(1), 65–84.
Wang, C., Li, J., Ye, F. and Yang, Y. (2013a). Multi-vehicle coordination for wireless energy replenishment in sensor networks. In 2013 IEEE 27th International Symposium on Parallel & Distributed Processing (IPDPS) (pp. 1101–1111).
Wang, G., Wang, T., Jia, W., Guo, M., & Li, J. (2009). Adaptive location updates for mobile sinks in wireless sensor networks. The Journal of Supercomputing, 47(2), 127–145.
Wang, J., Li, B., Xia, F., Kim, C.-S., & Kim, J.-U. (2014a). An energy efficient distance-aware routing algorithm with multiple mobile sinks for wireless sensor networks. Sensors, 14(8), 15163–15181.
Wang, J., Yang, X., Li, B., Lee, S., & Jeon, S. (2013b). A mobile sink based uneven clustering algorithm for wireless sensor networks. Journal of Internet Technology, 14(6), 895–902.
Wang, J., Zuo, L., Shen, J., Li, B., & Lee, S. (2014b). Multiple mobile sink-based routing algorithm for data dissemination in wireless sensor networks. Concurrency and Computation: Practice and Experience.
Wang, S., Vasilakos, A., Jiang, H., Ma, X., Liu, W., Peng, K., Liu, B., & Dong, Y. (2011). Energy efficient broadcasting using network coding aware protocol in wireless ad hoc network. In 2011 IEEE International Conference on Communications (ICC) (pp. 1–5).
Wang, Y.-C. (2014). Mobile sensor networks: System hardware and dispatch software. ACM Computing Surveys, 47(1), 12.
Wen, Y. F., Anderson, T. A., & Powers, D. M. (2014). On energy-efficient aggregation routing and scheduling in IEEE 802.15. 4-based wireless sensor networks. Wireless Communications and Mobile Computing, 14(2), 232–253.
Wichmann, A., Chester, J., & Korkmaz, T. (2012). Smooth path construction for data mule tours in wireless sensor networks. In 2012 IEEE Global Communications Conference (GLOBECOM) (pp. 86–92).
Wimalajeewa, T., & Jayaweera, S. K. (2010). Impact of mobile node density on detection performance measures in a hybrid sensor network. IEEE Transactions on Wireless Communications, 9(5), 1760–1769.
Xu, L., Delaney, D. T., O’hare, G. M., & Collier, R. (2013). The impact of transmission power control in wireless sensor networks. In 2013 12th IEEE International Symposium on Network Computing and Applications (NCA) (pp. 255–258).
Yin, F., Li, Z., & Wang, H. (2013). Energy-efficient data collection in multiple mobile gateways WSN-MCN convergence system. pp. 271–276.
You-Chiun, W., Wen-Chih, P., & Yu-Chee, T. (2010). Energy-balanced dispatch of mobile sensors in a hybrid wireless sensor network. IEEE Transactions on Parallel and Distributed Systems, 21(12), 1836–1850.
Yuan, F., Zhan, Y., & Wang, Y. (2014). Data density correlation degree clustering method for data aggregation in WSN. IEEE Sensors Journal, 14(4), 1089–1098.
Zhang, J., & Varadharajan, V. (2010). Wireless sensor network key management survey and taxonomy. Journal of Network and Computer Applications, 33(2), 63–75.
Zhao, M., Li, J., & Yang, Y. (2014). A framework of joint mobile energy replenishment and data gathering in wireless rechargeable sensor networks. IEEE Transactions on Mobile Computing, 13(12), 2689–2705.
Zhao, M., Ma, M., & Yang, Y. (2011). Efficient data gathering with mobile collectors and space-division multiple access technique in wireless sensor networks. IEEE Transactions on Computers, 60(3), 400–417.
Zhong, M., & Cassandras, C. G. (2011). Distributed coverage control and data collection with mobile sensor networks. IEEE Transactions on Automatic Control, 56(10), 2445–2455.
Acknowledgments
The authors thank the Islamic Development Bank (IDB) Scholarship Division, for supporting this work. The research is also supported by the Ministry of Education Malaysia (MOE) and conducted in collaboration with Research Management Center (RMC) at Universiti Teknologi Malaysia (UTM) under VOT NUMBER: Q.J130000.2528.06H00. They also thank the University of Malaya for the financial assistance (UMRG Grant RG325-15AFR). Lastly, the authors extend their appreciation to the Deanship of Scientific Research at King Saud University for supporting this work through the research group project No RGP-VPP-318.
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflict of interest
The authors declare no conflict of interest.
Rights and permissions
About this article
Cite this article
Abdul-Salaam, G., Abdullah, A.H., Anisi, M.H. et al. A comparative analysis of energy conservation approaches in hybrid wireless sensor networks data collection protocols. Telecommun Syst 61, 159–179 (2016). https://doi.org/10.1007/s11235-015-0092-8
Published:
Issue Date:
DOI: https://doi.org/10.1007/s11235-015-0092-8