Abstract
Energy-efficient data gathering techniques play a crucial role in promoting the development of smart portable devices as well as smart sensor devices based Internet of Things (IoT). For data gathering, different applications require different delay constraints; therefore, a delay Differentiated Services based Data Routing (DSDR) scheme is creatively proposed to improve the delay differentiated services constraint that is missed from previous data gathering studies. The DSDR scheme has three advantages: first, DSDR greatly reduces transmission delay by establishing energy-efficient routing paths (E2RPs). Multiple E2RPs are established in different locations of the network to forward data, and the duty cycles of nodes on E2RPs are increased to 1, so the data is forwarded by E2RPs without the existence of sleeping delay, which greatly reduces transmission latency. Secondly, DSDR intelligently chooses transmission method according to data urgency: the direct-forwarding strategy is adopted for delay-sensitive data to ensure minimum end-to-end delay, while wait-forwarding method is adopted for delay-tolerant data to perform data fusion for reducing energy consumption. Finally, DSDR make full use of the residual energy and improve the effective energy utilization. The E2RPs are built in the region with adequate residual energy and they are periodically rotated to equalize the energy consumption of the network. A comprehensive performance analysis demonstrates that the DSDR scheme has obvious advantages in improving network performance compared to previous studies: it reduces transmission latency of delay-sensitive data by 44.31%, reduces transmission latency of delay-tolerant data by 25.65%, and improves network energy utilization by 30.61%, while also guaranteeing the network lifetime is not lower than previous studies.
1. Introduction
Internet of Things (IoT) [1–3] and cloud computing [4–6] leverage the ubiquity of sensor-equipped devices (e.g., smart portable devices, smart sensor nodes) to collect information at low cost and provide a new paradigm for solving the complex sensing applications from the significant demands of critical infrastructure such as surveillance systems [7–10], remote patient care systems in healthcare [11, 12], intelligent traffic management [13], automated vehicles in transportation environmental [14], and weather monitoring systems. Data gathering is one of the important functions for such sensor based networks, in which each smart sensor based device is sensing information from the surrounding environment and sends data to the sink via multihop routing. After data have been acquired, the sink can make decisions about perceived events, thus bringing smart life to people [8, 10, 11]. But, due to the limited energy of sensor devices, how to ensure that data is gathered to the sink quickly as well as in an energy-efficient way is an important issue [15–17].
Although the smart sensor based devices (e.g., wireless sensor nodes, smart phone, and tablets) have powerful computing and communication capabilities, the data produced are getting more and more as the scope of application expands and the variety increases, so the energy of the device is drained out much faster than before. Green communications play a critical role in or are expected to be an indispensable part of the smart sensor based networks [1, 10, 13, 18]. Due to the diversity and complexity of applications running over the networks, green communications under delay or quality of service guarantee in such networks gain increasing attention in the research community [19, 20] and face a great challenge.(1)The challenge of green communication: communication consumption is the largest energy consumption in sensor based networks, so the key to green communications is to effectively reduce the energy consumption in communication. Overall, the main ways to reduce energy consumption are as follows:(a)Using asynchronous sleep/waking working mode can effectively reduce the energy consumption of nodes. In this mode, the sensor nodes periodically rotate between sleeping and waking states, and the ratio of the duration that the nodes are in the waking state to the unit cycle is called the duty cycle [11, 21]. In general, the energy consumption of the node in the sleeping state is 1/100–1/1000 of that in waking state [8, 12, 21]; thus, in order to save energy, the node should be in sleeping state as much as possible. However, the node can communicate only when it is in waking state, so the small duty cycle of node can reduce its energy consumption effectively, but when it has data to send, it needs to wait for the waking up of the receiver to communicate. And the time required from the preparation of data at the sender to the waking up at the receiver is called sleeping delay. This means that to reduce the data transmission delay it is necessary to increase the duty cycle, but at the same time it will increase the energy consumption of the node, so how to balance the delay and energy consumption is a tricky problem.(b)Another way to effectively reduce energy consumption is data fusion. Reference [22] points out that event information has relevance both in time and in space, so data fusion can be used to fuse multiple event information, thus greatly reducing the amount of data and reducing energy consumption. From the perspective of reducing energy consumption, more data packets should be fused so that the redundancy is minimized. Although a lot of routing strategies for data fusion have been proposed, most of them have not been applied to sensor networks. What is more, applications often require that the delay of the data’s arrival is less than a specified threshold. For example, in a fire monitoring system, “alarm” message is required to be transmitted to the control center as soon as possible; otherwise it will cause significant losses. The same situation exists in the medical systems, the sensor node attached to the patient is generally used to record the patient’s daily vital signs, but, in an emergency where the patient’s heart rate changes dramatically, the “alarm” packet of the data needs to be transmitted to the doctor in time to take the appropriate first-aid measures [5, 12, 23]. Obviously, in such cases, the strategies proposed for maximizing data fusion are not able to ensure that the delay satisfies the requirements. Therefore, it is another important issue to make the data fusion under the precondition of guaranteeing the application delay and reduce the network energy consumption effectively.(2)The delay or quality of service (QoS) guarantees for sensor networks: as an important part of information infrastructure, sensor based networks often collect data for a variety of applications. Different applications have different requirements for QoS, and even the same applications may have different requirements due to the different events [20, 22], where delay is the basic metrics in data gathering. Generally speaking, delay-sensitive data requires a small delay to the sink, and delay-tolerant data can tolerate a certain delay. The best way to optimize network performance is to use a differentiated service to distinguish different transmissions to meet application requirements. But, to the best of our knowledge, there is no research that is able to provide differentiated services in green communications.
In summary, how to combine the settings of duty cycles with data fusion technologies and differentiated services to establish efficient routings to ensure energy-efficient and delay-guaranteed data gathering is of great significance and also has broad application prospects. Thus, a delay Differentiated Services based Data Routing (DSDR) scheme is creatively proposed, in which the energy consumption can be reduced and delay can be guaranteed for applications. The contributions of the DSDR scheme are as follows.
(1) The DSDR scheme has high energy efficiency and reduces transmission delay greatly by establishing energy-efficient routing paths (E2RPs). First, multiple E2RPs are established in different locations of the network to forward data, and the duty cycles of nodes on E2RPs are increased to 1, so the data is forwarded by E2RPs without the existence of sleeping delay, which greatly reduces transmission latency. What is more, the E2RPs are built in the region with adequate residual energy, and the residual energy of nodes on E2RPs is periodically measured, and once the remaining energy is below the threshold, the E2RP where the node located is immediately removed. Therefore, the DSDR scheme enhances the energy utilization of network effectively without affecting the network lifetime.
(2) The DSDR scheme adopts delay differentiated services, which makes the scheme have a better performance under the premise of guaranteeing the service quality. In DSDR, for delay-sensitive data, the direct-forwarding strategy is used to ensure the minimal end-to-end transmission delay, while, for delay-tolerant data, the wait-forwarding method and data fusion are adopted to reduce energy consumption.
(3) The DSDR scheme has the capability of global event data fusion, which greatly reduces the data amount needed to be transmitted and improves the energy efficiency of the network. Firstly, E2RPs are naturally data fusion paths, which can greatly increase the probability of data fusion. Secondly, the data are all routed through E2RPs, making the data fusion under DSDR scheme global and the largest possible, which greatly reduces data redundancy and energy consumption. To the best of our knowledge, the DSDR is the first more perfect combination of data fusion, differentiated services, and duty cycle setting and integrates an efficient data gathering in a number of performances far better than the previous studies.
(4) Through our extensive theoretical analysis, we demonstrate that the DSDR scheme proposed in this paper has better performances. It realizes synchronous optimization in latency and energy, reduces the transmission delay of delay-sensitive data by 44.31%, reduces the transmission delay of delay-tolerant data by 25.65%, and improves network energy utilization by 30.61%. Compared with the previous schemes, DSDR has obvious advantages in delay optimization and energy utilization.
The rest of this paper is organized as follows: in Section 2, a literature review related to this work is introduced. Then the system model and problem statement are described in Section 3. In Section 4, we proposed an efficient DSDR scheme. The performance analysis of the DSDR scheme is provided in Section 5. Finally, Section 6 presented the conclusion and future perspectives of our work.
2. Related Work
Different applications running on the same network platform may have different requirements on the service of quality; the basic requirement is network delay [19, 21, 24, 25]. In this section, we introduce some state-of-the-art researches for delay differentiated service.
Considering the network delay simply, the duty cycle is the key to reduce delay [26–29]. In WSNs, the node is awake periodically; the longer the node in the waking state, the smaller its sleeping delay, and the smaller the network latency. Based on this idea, the authors of [14] proposed a distributed duty cycle allocation algorithm. The algorithm uses the network residual energy to gradually increase the duty cycles of nodes in the far-sink region based on the distance between the node and the sink and makes the delay of the network external nodes decrease. In [16], the authors also consider the adjustment of the node duty cycle. However, the adjustment under this study is based on the load of the entire network. This kind of duty cycle adjustment scheme has some innovation, but it increases the complexity of the network system, especially when the network is large and the parameter configuration will be tricky.
In addition, since the data are transmitted to the sink through multihop relays, reducing the hops of relay can effectively reduce the end-to-end delay. The authors of [15] realize this and propose a communication scheme called T-LOGF to select the next hop based on the shortest distance. The scheme uses one-step-look-ahead data transmission principle, each time selects the node with the smallest distance from the sink to forward data, so that the total relay hops can be minimized, and reduces the delay, but the scheme ignores the waiting delay caused by the sleep/waking mode.
In [30], the authors studied the issue of network latency requirements for differential services in content-centric networking (CCN), while optimizing the content overall transmission performance. In this paper, in order to guarantee the network delay, the authors first proposed a simple and comprehensive network model; the network delay of routing content is characterized to different locations of the client. Ensure that each content provider has optimized content and latency by combining network locations with content popularity. Then, the content distribution model is incorporated into the proposed global network model, and the delay is addressed as the nonlinear integer programming (NIP) problem in the given network resource and traffic access mode.
The above studies have been considered unilaterally from the network latency, but in fact the data integrity is valued by most applications. In [3], a multipath dynamic routing algorithm called IDDR is proposed to satisfy these two requirements simultaneously. In IDDR, according to the sensitivity of the data to delay and integrity, on the basis of weights of each packet and different QoS requirements, the data packets in the network are separated, and different data are transferred to the sink through different paths to improve data fidelity and integrity and minimize the end-to-end transmission latency. This dynamic routing for data integrity and delay differentiated service can meet the quality requirements of different data, but it is necessary to create a route for each data transfer, increasing the complexity of the network.
Similar to [3], authors of [31] evaluate the quality of the network from the timeliness and reliability. The authors proposed a packet delivery mechanism called Multipath and Multi-SPEED Routing Protocol (MMSPEED) to improve data latency and reliability. For timeliness, MMSPEED provides multiple QoS levels to ensure the transmission speed of multiple packets. For reliability, probabilistic multipath forwarding supports the various reliability requirements of data. These mechanisms for QoS provisioning are realized in a localized way without global information by adopting localized geographic packet forwarding augmented with dynamic compensation, which compensates for local decision inaccuracies as a packet travels towards the destination.
In a more complex system, the quality of service of the network may involve multiple parameters [32–36]. In [20], the authors proposed RPL for the application of traffic prioritization and differentiation for bandwidth, delay, reliability, and security. RPL addresses QoS channel access through queue scheduling mechanism or the priority of media access control (MAC). Specifically, multiple instances of RPL implement the validity of QoS differences based on different objective functions. At the same time, three variants of RPL: standard RPL, multi-instance RPL (RPL-M), and multi-instance RPL with prioritized channel backoffs (RPL-M+) along with two distinct traffic classes have been examined as data traffic rate and composition was varied.
A three-layer QoS scheduling model is proposed for service-oriented IoT in [37]; it selects different scheduling schemes from the application layer, network layer, and sensing layer. In application layer, it uses the service of each component to explore the best QoS-aware service composition. In network layer, the model mainly deals with the scheduling of a heterogeneous network environment. In sensing layer, it is responsible for different service information collection and resource allocation scheduling. The scheme can reduce the costs of network resources in a certain extent.
WSNs are highly complex systems, and energy consumption in resource-constrained WSNs is also an important parameter to consider when reducing network latency. In-network data fusion and clustering are important technologies to reduce energy consumption in wireless sensor networks; the authors of [38] proposed a delay-aware network structure for WSNS with in-network data fusion by utilizing this technique, which divides the sensor nodes into different size clusters, so that each cluster can communicate with the fusion center in an interleaved manner. At the same time, an optimization process is proposed to optimize the communication distance within the group. The proposed network can reduce the latency of the data aggregation process, provided that the data can only be partially fused.
Villas et al. of [22] also use data fusion to achieve the performance requirements of the network and propose an in-network data aggregation routing strategy. The strategy establishes routing paths in the nonhotspot area of the network, and then a number of events that occur in different area are merged into the path, and when the packets of these events go through the same node, data fusion is performed to reduce redundancy. However, this strategy can only be part of the local event information together, but no global network event information fusion capabilities.
In relatively differentiated services, the network traffic is divided into a small number of service classes based on their packets transmission quality, in terms of per-hop metrics for queuing delays and packet losses. In [39], the authors improved and quantified relatively differentiated services and proposed proportional differentiation model for network operators providing “tuning knobs” to adjust the quality interval between classes. The model has a strict priority or capacity differentiation and does not consider the class load and loss differences.
It can be seen from the above that duty cycle and data fusion are all key technologies to improve network performance. Therefore, in the DSDR scheme, we first reduce the transmission delay of delay-sensitive data by increasing duty cycles of relay nodes and then decrease the energy consumption of delay-tolerant data by data fusion.
3. The System Model and Problem Statement
3.1. System Model
The network model of this paper is a typical planar periodic data gathering wireless sensor network, which is similar to [14, 40]; its model structure is as follows.
(1) homogeneous sensor nodes are randomly deployed in a two-dimensional planar network, the sink is the center of the entire network, the network radius is , the node communication radius is , and the node density is . Each node in the network continuously monitors the surrounding environment, and once the event of interest is detected, the event data is immediately sent to the sink via multihop relays.
(2) Sensor nodes sleep/waking work periodically, nodes monitor the target and transmit the data in the waking state. All the monitoring targets are randomly distributed in different positions of the network, so the probability that each node monitors target and generates data is equal.
(3) The lossless step-by-step multihop aggregation model is used in this paper [22]. In this model, the result of data fusion of nodes and is denoted by , which can be calculated as
where is the correlation coefficient of and and can be calculated as is the distance between and and is the constant of correlation coefficient, while and , indicating there is no correlation between and , and the larger is, the higher data correlation of them is.
3.2. Network Parameters
A sensor node is composed of a sensing unit and a communication unit [17], the sensing unit is responsible for sensing the monitoring event and invoking the communication unit after sensing the event, and then the communication unit initiates its internal communication mechanism to transmit data to the sink. Due to the fact that the energy of nodes is strictly limited, the sensor nodes use sleep/waking cycling mode to save energy [15]. In one unit cycle, the nodes are in the sleeping and waking state periodically, and only when the node is in the waking state can it perceive events and transmit data. In one unit cycle, the ratio of the node in the waking state to the whole cycle is called duty cycle; suppose that is the sensing duty cycle and is the communication duty cycle, and thenwhere is the sensing cycle, is the communication cycle, is the time that the node is in the waking state during the sensing cycle, is the time that the node is in the sleeping state during the sensing cycle, is the time that the node is in the waking state during the communication cycle, and is the time that the node is in the sleeping state during the communication cycle.
The energy consumption model of this paper is similar to [14, 41], and the energy consumption of nodes is composed of event sensing, data transmission, data receiving, and low power listening. Therefore, the energy consumption model can be expressed as
Table 1 and the Parameters Related to Calculation list the relevant parameters used in this paper, and the parameters values are derived from the internal data tables of the prototype sensor nodes as in [7].
3.3. Problem Statements
The main objective of this paper is to design an efficient communication scheme suitable for wireless sensor networks. In the quality of service, the scheme can meet different delay requirements. In terms of network performance, the scheme is able to maintain the network lifetime while strengthening the effective use of energy. It can be expressed as follows.
(1) Minimize End-to-End Delay for Delay-Sensitive Data. The end-to-end transmission delay of data refers to the time required from the preparation of data at the sender to its arrival at the sink [17, 42]. Assuming that the transmission delay of kth hop is , and the number of relay hops is , then the minimized end-to-end delay is expressed as follows:
(2) Balance Energy Consumption and Maximize Energy Utilization. All data in the network are transmitted to the sink via multihop relays. This sink-centered data gathering model results in a large amount of data and energy consumption in the area that is relatively close to the sink [11, 43, 44], while, in the area far away from the sink, the energy consumption is very small, which contributed to an imbalance in the energy consumption. Assuming that there are N nodes in the network, the energy consumption of node is , and the equilibrium energy consumption of the network is to minimize the energy difference between the maximum energy consumption and the minimum energy consumption; that is,
Energy utilization is the ratio of the energy consumed by the network to the initial energy. Assuming that the initial energy of the node is , the maximized energy utilization can be expressed as follows:
(3) Maximize Network Lifetime. Network lifetime is the data transmission rounds that have been performed before the death of the network, which is defined in most studies as the death time of the first node in the network [12, 45, 46]. In wireless sensor networks, the node is powered by the battery, and when the energy of the node is exhausted, the node dies [47]. Therefore, the network lifetime depends on the maximum energy consumption of the network. Assuming that the energy consumption of the node is , its initial energy is , and there are N nodes in the network. Therefore, the maximized lifetime of the network is to maximize the lifetime of the node with the largest energy consumption; that is,
In summary, the research objectives are as follows:
4. DSDR Scheme
4.1. Research Motivation
Applications running on the same network platform may have different latency requirements due to the differences in events [3, 48], and most existing schemes cannot intelligently select appropriate data routing based on the latency sensitivity of the events, thus failing to meet the requirements of different data. For example, in a network applied to biological monitoring, the sensor nodes are usually deployed to record the daily behaviors of the organism, but if the organism gets into danger and its vital signs change dramatically, it is required to transfer the urgent data to decision-makers as soon as possible in order to take corresponding rescue measures in time. Therefore, for such networks, the difference of its service quality is embodied in the following: for delay-sensitive data, it is required to transmit information to the sink as soon as possible through a reliable transmission mechanism, and for delay-tolerant data only the normal transmission is required. As shown in Figure 1, and are two nodes in the network, where senses normal events that carry delay-tolerant data, while senses urgent incidents which have strict requirements for the transmission delay. According to the urgency of the packets, the data sent by should be transmitted to the sink earlier than . But, in the wireless sensor networks, the packets are relayed to the sink through multi hops, and the maximum forwarding distance at each hop for nodes is , where is the communication radius of nodes. Therefore, the farther the distance between the sending node and the sink is, the more relay hops the data passes, and the greater the transmission delay is, so, in the previous studies, the data of in Figure 1 is much more delayed than that of , which conflicts with the latency requirements of the data.
In a sparse network with , , and , suppose nodes adopt 0-1 communication mode and data transmission adopts multihop routing; then the network performance under different parameter values is shown in Figures 2–6. When the node communication duty cycle is 0.2–0.5, respectively, the end-to-end delay at different distances of the network is shown in Figure 2. It can be seen from Figure 2 that the transmission delay of data gradually increases with the distance from the sink, and the delay of the far-sink region is much higher than that of the near-sink region, which is about 4 times (). And the near-sink region refers to the region within one-hop range of the sink, in a network where the network radius is 500 and the node communication radius is 80, the area whose distance from the sink is 0–80 is near-sink region, and nodes located in the near-sink area are near-sink nodes. Similarly, the far-sink region refers to the area outside one-hop range of the sink, which is an area whose distance from the sink is 80–500, and the nodes in this area are far-sink nodes. It is clear that the far-sink area occupies a large area of the network, so the probability that the delay-sensitive events are distributed in the far-sink area is much higher than that of the near-sink area, and this raises the requirements for the service quality of the network, especially the transmission delay in the far-sink area. In addition, to overcome the limited energy of the nodes and the difficult replacement of batteries, the sleep/waking working mode is used to save energy [15], and the node can forward data only when it is in the waking state. Obviously, prolonging the time of the node in the waking state can effectively reduce the waiting delay, thus improving the transmission efficiency.
This can also be seen from Figure 2, compared with the lower duty cycle 0.2, when the duty cycle is 0.5, the network delay is greatly reduced, and the maximum delay in far-sink area is shortened from 167.25 ms to 69.75 ms.
Increasing the duty cycle can effectively reduce the transmission delay, but this is at the expense of energy consumption. Because the energy consumption of nodes in the waking state is 100–1000 times that in the sleeping state [5, 12, 23], therefore, if the duty cycle of all nodes in the network increases, it will greatly exacerbate the network energy consumption and shorten the network lifetime. According to the energy consumption model of (4), the energy consumption of nodes is caused by event sensing, data transmission, data receiving, and low power listening, where energy consumption in sensing and low power listening are related to the node’s sensing and communication cycle, and the cycles are constant. The energy consumption in data transmission and receiving are affected by data amount. When the node communication duty cycle is 0.3, 0.4, and 0.5, respectively, the energy consumption of the network under different data amount is given in Figure 3. As shown in Figure 3, the energy consumption increases linearly as the amount of data increases, and the larger the data amount is, the more energy it consumes, and, at the same time, the larger the duty cycle is, the faster the energy consumption increases.
It can be seen from Figure 3 that reducing the amount of data is the key to saving energy. Numerous studies have shown that network data are correlated with each other in time and space, and there is redundancy among the data detected by multiple nodes of the event [3, 22]. Therefore, data fusion technology can be used to aggregate the information of multiple events and then send it to the sink, thus greatly reducing the amount of data and energy consumption. When the probability of generating data is 0.1, the data fusion rate is between 0.1 and 0.3, and the comparison of data amount before and after data fusion is shown in Figure 4. Since the fusion rate between any two data packets is random and uncertain, the amount of data after fusion for different data varies. However, in general, compared with the original data, the data amount is greatly reduced after fusion.
As mentioned in Figure 4, improving the node duty cycle can effectively reduce delay, while using data fusion can alleviate the energy consumption caused by the increase of duty cycle. However, from the perspective of data generation, in general, the data transmitted in the network are delay-tolerant data; that is, the probability of generating delay-sensitive data is relatively small, and for these small probability data improving duty cycles of all nodes, obviously, is not the most reasonable solution. Therefore, it can only increase the duty cycles of a small part of nodes and establish energy-efficient routing paths by these nodes and finally relay the delay-sensitive data to the sink via these paths; as for delay-tolerant data, data fusion is performed at each hop on the path to save energy, and we called this communication scheme DSDR (delay Differentiated Services for Data Routing).
In the same network scenario as Figure 2, if the DSDR scheme is adopted, then the comparison between DSDR scheme and DR-GA scheme in terms of latency and network lifetime is given in Figures 5 and 6, where the DR-GA (General Data Routing Approach) represents the routing scheme widely used in other studies, the data transmission under this scheme also uses multihop relaying, and the relay node selection is based on the shortest distance. As can be seen from Figure 5, compared with the DR-GA scheme, the transmission delay under the DSDR scheme is greatly reduced, and the latency of delay-sensitive data is smaller than that of delay-tolerant data, which means that DSDR can provide differentiated services while ensuring low transmission latency. More importantly, although the DSDR scheme can effectively reduce the network delay, as shown in Figure 6 it has a longer network lifetime than the DR-GA scheme, which also shows the advantage of the DSDR scheme.
4.2. General Design of DSDR
The main idea of DSDR is to establish energy-efficient routing paths (E2RPs) at different locations of the network, and the communication duty cycles of nodes on E2RPs are increased to 1 (except for the nodes in the near-sink area), and then transmit data to the sink through these E2RPs by adopting different routing methods for different data.
In general, the data communication under DSDR includes the following phases.
Phase 1 (establishment of energy-efficient routing paths (E2RPs)). Each node in the network is required to store two values: and , where is the minimum hops to the sink and is the minimum hops to E2RPs. Before the establishment of E2RPs, the values of and are POSITIVE_INFINITY, and the values of them are constantly updated with the constant broadcast in the network. The general process is as follows: firstly, starting from the sink, it broadcasts the latest to the nodes in the communication range, which is valued 1, and then the nodes that receive the broadcast message compare in the broadcast with its stored value and update its to the smaller value and then repeat the same process, and then each node determines its by continuous diffusion of broadcasting. Then, starting from a node in the network, for each hop, select a node whose is smaller than its own in the communication range to establish a path to the sink (i.e., an E2RP), and then configure the duty cycle and of the nodes on this E2RP. Finally, of each node is determined by the same broadcast diffusion as in the center of each node on the E2RP. In a similar way, a number of E2RPs in other areas are established.
Phase 2 (delay differentiated services for different data). After E2RPs are established, all the data is transferred to the sink through E2RPs, and different transmission methods are adopted according to different delay requirements of the data. First, for the delay-sensitive data, a direct-forwarding routing method is adopted, the specific process is as follows: starting from the sending node, each time the node selects a node whose is smaller than its own in the communication range as the next-hop node. This process continues until of the next node becomes 0, which means that the data has reached the nearest E2RP and then use the E2RP to transmit data to the sink. During the entire transmission process, the data is not stopped. Then, for delay-tolerant data, a waiting-forwarding routing method is adopted, and its data routing process is the same as delay-sensitive data, which is also transmitted to the sink through the nearest E2RP and under the guidance of , but delay-tolerant data stays for a certain amount of time at each hop for maximum data fusion, and the specific dwell time is determined by the delay requirements of the data.
Phase 3 (data fusion for delay-tolerant data). Starting from the sending node, data fusion is performed at each relay hop. Firstly, the maximum waiting time at each hop is obtained according to the total waiting time of the data and the distance from the sending node to the sink and the E2RP. Then, the data stops for corresponding time at each relay to merge with other packets received by the relay. After the waiting duration is reached, the fused data are transmitted to the next hop. And the above process is repeated until the data are transferred to the sink.
Figure 7 is the data routings for different data in DSDR. The data routing of delay-tolerant data and delay-sensitive data is described with an E2RP as an example. Compared to delay-tolerant data, delay-sensitive data has a higher transmission priority. When the delay-sensitive data are perceived, the sensing node immediately transmitted it to the sink via E2RP, and there is no stop during the transmission. And for delay-tolerant data, such as data in Area 1 and Area 2, after the node senses the data, it is transmitted to the next hop, and then the next hop node waits for more packets around it to arrive. When the corresponding waiting time for the data is up, the relay node merges all the data it receives with the packet to remove the redundant information and then forwards it to the next hop. As we can see, the delay-tolerant data is fused at each hop it passes.
4.3. Implementation of DSDR
In DSDR, each E2RP forwards events in its vicinity; therefore, the transmission quality of the path directly affects the performance of the entire network. In addition, the amount of data carried by the E2RP is very large, and the nodes on the path are awake all the time; therefore, the energy consumption of E2RPs is very serious, so the establishment and switching of the E2RPs are an important issue. On the other hand, how to determine the urgency of data, define its data type, and set the maximum waiting time at each hop is essential for data routing. And, for the delay-tolerant data in the network, the specific calculation of data fusion is also an important issue, and we mainly discuss the implementation of these in this section.
(1) Establishment and Switching of E2RP. Not every node that generates a packet will establish an E2RP, and the establishment of E2RPs only happens in two phases: (a) network initialization and (b) when an E2RP is removed. The specific establishment of an E2RP is divided into three scenarios: Confirmation of Minimum Hops to the Sink. The confirmation of is based on the acquisitions of neighbors’ information by broadcast diffusion; when the network is initialized, the sink sets its to 0, and the other nodes set their to ∞. Then, the sink sends a broadcast message to nodes in its communication range, informing them of the latest , which is . Each node that received the broadcast information compares the broadcast value with its stored . If the value is smaller than the previously stored , of the node is reassigned to the broadcast value. Then, the node that has updated broadcasts the latest to other nodes, which is of this node plus 1. The whole process continues until all nodes in the network are not continually updated to ensure that each node confirms its . Establishment of an E2RP. After each node in the network has confirmed its , starting from the sending node, for each hop, select a next hop whose is smaller than its own from the neighboring nodes in the communication range. This process continues until of the relay node becomes 0, which means that the data has reached sink. Then of all relay nodes on the path are set to 0 and the communication duty cycles of these nodes are set to 1, except for nodes in near-sink region. Confirmation of the Minimum Hops to E2RPs. Making each node on the E2RPs as the center, determine of each node by the same broadcast diffusion of . In a similar way, a number of E2RPs in other areas are established.
As shown in Figure 8, there are four E2RPs in the network: Paths 1–4. of each node in the figure is centered on the sink and incremented from inside to outside. of the node is not marked in this figure, but it can be seen from the data routing of nodes, whether delay-tolerant data or delay-sensitive data, once the sending node senses the data, the node will transmit it through the nearest E2RP, and, in fact, it is completed under the guidance of . Assuming that in each broadcast message is , in each broadcast message is , the communication duty cycle is , and the Nodes.br is used to identify whether nodes have received or sent broadcast messages before, and then the establishment and configuration of an E2RP are described by Algorithm 1.
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
The energy consumption of the E2RPs is serious. Because the lifetime of network is determined by the maximum energy consumption of nodes, in order to avoid the energy consumption of E2RPs shortening the network lifetime, we measure the residual energy of the E2RPs periodically and switch it at appropriate time. The principle of switching and reconstructing the E2RPs is as follows: for each E2RP in the network, the residual energy of its nodes is measured at intervals of time, and once the energy is lower than the threshold, which is set to 20%–30% of the initial energy, the path is immediately removed and the parameters of nodes on it are restored to the system initial values, and all nodes on the path no longer act as E2RPs nodes. Then find new relay nodes which are located in the far-sink area and have a large amount of residual energy. And the establishment of the new E2RP is similar to Scenario 2 in Algorithm 1, but the remaining energy is taken into account when selecting relay nodes on E2RPs. If the residual energy of the node is below the threshold, then it will not serve as a relay node on E2RPs.
(2) Setting of Data Urgency and Waiting Time. There are two types of data under the DSDR scheme, one is delay-sensitive data, which is denoted as , and the other is delay-tolerant data, which is denoted as . Each packet has a parameter to identify the urgency of its data; when is true, it means that the data are delay-sensitive data; and when is false, it represents delay-tolerant data. The waiting time for delay-sensitive data is 0 and for delay-tolerant data is , and the value of varies with the specific situation of data. Depending on of data and the distance between the sensing node and the sink, the waiting duration at each hop is determined, as shown in Theorem 1.
Theorem 1. Suppose that is a delay-tolerant data packet whose distance from sink is , the waiting time of is , its hops to E2RP are , and the hops from E2RP to the sink are , and then the waiting duration at each hop for can be expressed as follows:
Proof. The total waiting time of is , and the transmission from the sensing node to the sink is composed of two phases: one is the data transmission from the sensing node to the E2RP, and the number of relay hops in this phase is ; the other is the transmission from the E2RP to the sink, and the number of relay hops in this phase is . Therefore, the total number of relay hops from the sensing node to the sink is . Then the waiting duration at each hop is obtained by dividing the total waiting time by total relay hops, that is, . However, it should be noted that since is the maximum waiting time of the data and there are some other delays in the transmission, therefore, the actual waiting duration at each hop is , where is the other delays that may be encountered on the transmission path, including queuing and possible retransmission delays. In a sparse network with good link quality, can be ignored.
After the data type and one-hop waiting duration are confirmed, the data can be routed as in Figure 9. The waiting time of delay-sensitive data in the network is 0, and once the data are sensed, they are quickly transmitted to the sink via E2RP, such as Data 1 and other delay-sensitive packets on E2RP in the figure. When delay-tolerant data are sensed, they are waiting for time at each hop to fuse with other data received by the relay, such as Data 2 in the figure. Three delay-tolerant packets in Area 1 are waiting to be merged with other data because of their one-hop waiting time, while the waiting time for Data 3 is reached and it is forwarded to the next hop through E2RP.
(3) Related Implementation of Data Fusion. Data fusion rules of the delay-tolerant data are as follows: data stay time at each hop of the data routing. In , the data wait for the other packets received by the relay node to fuse, and once time is reached, the merged data are sent to the next hop.
In this paper, the nodes have different data amount due to the different positions, and the trend of data amount is as follows: (A) the closer the node is to the sink, the larger its data amount is, the data amount of the nodes in the near-sink area is large, and that of far-sink area is small. (B) The closer the node is to the E2RP, the larger its data amount is, and the data amount of nodes that are far away from the E2RP and the sink is small.
According to [14], suppose that the network radius is , the node communication radius is , the probability of generating data is , is the node whose distance to the sink is , and is the largest positive integer that makes the expression true, and then the number of packets received by can be expressed as follows:
Theorem 2. In DSDR, suppose that the network radius is R, the node communication radius is r, the probability of generating data is , is the node whose distance to the sink is i, the distance from to its nearest E2RP is , and then the number of packets received by in this paper can be expressed as follows:
Proof. Equation (11) gives the data packets of nodes in a network with the sink as the center: , where is the largest positive integer that makes the expression true. In this paper, the data of the far-sink region are all transmitted through the E2RPs, so the data packets of the far-sink nodes in the network decrease gradually with the distance from the E2RP. When the distance from to its nearest E2RP is , according to [14], the data packets of nodes in a network with E2RPs as the center are . However, the regions in the network are overlapped, and the data packets in regions closer to the sink are much larger when it is centered on the sink than centered on E2RPs and vice versa, so the data packets actually assumed by the nodes in the network should be the larger of the two cases, which is (12).
In DSDR, the data fusion is performed after each relay node located on the transmission path receives a delay-tolerant packet. Suppose that the number of delay-tolerant packets received by is , the packet length is , and , where is the correlation coefficient of data, and, according to (1) and [24], the amount of data received by can be expressed as follows:
According to (13), suppose that the number of delay-tolerant packets is , the packet length is , and , where is the correlation coefficient of data, and then the amount of data after fusing k times can be expressed as follows:
In a circular network with a radius of 500 and a node communication radius of 100, a data fusion rate of 0.01–0.15, the data amount of the network under different fusion times is given in Figure 10. It can be seen from the figure that data fusion can effectively reduce data amount, but, because of the uncertainty of the data correlation of packets in the network, the data fusion rate between packets is also different. Generally, data relevance for the packets sensed at the same time and space is higher. Observing the circumstances under different data fusion times, it is found that the data amount decreases with the increase of the fusion times. Since the data redundancy becomes less and less in the process of continuous fusion, when the number of fusions reaches a certain level, the reduction in the amount of data is less obvious.
After the E2RPs are established and the data fusion rules are confirmed and the routing rules are shown in Figure 11, consider the following.
(1) For data in the near-sink area, they are sent directly to the sink regardless of the data type, as Data 1 in Figure 11.
(2) For delay-sensitive data in far-sink area, they are transmitted to the sink via the E2RP, as Data 2 in Figure 11.
(3) For the delay-tolerant data in the far-sink area, they are also transmitted to the sink through the E2RP, but the time is reserved at each hop to fuse with other data received by the node. After the waiting time is reached, the merged data are transferred to the next hop, as Data 3 and Data 4 in Figure 11.
Suppose that the node that senses the data is , the relay hops from to the sink are , and the relay hops from to E2RP are ; if the data are delay-tolerant data, is false, if the data are delay-sensitive data, is true, and the waiting duration for delay-tolerant data at each hop is , and then the data routing under the DSDR scheme can be described by Algorithm 2.
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
5. Performance Analysis of DSDR
In this section, the effectiveness of DSDR scheme is analyzed from three aspects: network delay, energy efficiency, and network lifetime. At the same time, the performance of DSDR scheme is compared with the General Data Routing Approach (DR-GA), which is a routing scheme based on the shortest distance and multihop relaying. Without additional descriptions, the network parameters are set as follows: , , and , and for values for other parameters please see Table 1.
5.1. Transmission Delay
According to [17], suppose that is the node whose distance to the sink is i, its communication duty cycle is , communication cycle is , and then the one-hop transmission delay of can be expressed as follows:
In DR-GA, suppose that is the node whose distance to the sink is i, its communication radius is , and then the relay hops from to the sink are . Therefore, the end-to-end transmission delay from to the sink is the product of the relay hops and one-hop delay; that is,
In DSDR, one-hop delay varies depending on the different types of data, suppose that are delay-sensitive data and are delay-tolerant data, and then the one-hop delay of them can be calculated by Theorems 3 and 4.
Theorem 3. In DSDR, suppose that is the node whose distance to the sink is i, the data it transferred is , its communication duty cycle is , communication cycle is , if is the far-sink node on E2RPs then is true, and then the one-hop delay of can be expressed as follows:
Proof. Nodes have different one-hop delay due to different types. When is an ordinary node of the far-sink area or a near-sink node, its parameter values are not adjusted. Its one-hop delay is the same as that of the DR-GA. However, if is the far-sink node on E2RPs, because the communication duty cycles of nodes on E2RPs are 1, they have no delay. The one-hop delay is . is the packet preamble receiving delay, is the data communication delay, and is the acknowledgment message transmission delay.
Theorem 4. In DSDR, suppose that is the node whose distance to the sink is i, the data it transferred is , its waiting duration at each relay is , if is the far-sink node on E2RPs then is true, and then the one-hop delay of can be expressed as follows:
Proof. When the data transmitted is , its one-hop delay is divided into two cases: is an ordinary node of the far-sink area or a near-sink node. is the far-sink node on E2RPs. The delay of the two cases is similar to (17). However, delay-tolerant data are delayed time at each relay hop, and the delay of delay-tolerant data is more than that of delay-sensitive data. Therefore, (18) is proved.
Theorem 5. In DSDR, suppose that is the node whose distance to the sink is i, its minimum hops to the sink are , its hops to nearest E2RP are , and the hops from to the sink are . If is the node in near-sink area then is 0, if is the node in far-sink area and its transmitted data is then is 1, if is the node in far-sink area and its transmitted data is then is , and then the end-to-end delay from to the sink can be expressed as follows:where
Proof. In DSDR, the transmission delay of is divided into three situations depending on the network locations and data types. When is a node in near-sink area, then the data is sent directly to the sink regardless of the data type. Therefore, its end-to-end delay is the product of the one-hop delay and relay hops: . When is a node in far-sink area and the data it transferred is , its delay is divided into two phases: delay from to E2RP and delay from E2RP to the sink. Delay in the previous phase is the product of one-hop delay and hops from to E2RP: . Delay in the latter part is also divided into two cases. Because the duty cycles of far-sink nodes on the E2RP are 1 and the duty cycles of near-sink nodes remain unchanged, the delay in this phase is divided into a partial delay from E2RP to near-sink node (the duty cycle is 1) and the one-hop delay from near-sink node to the sink (the original duty cycle). The delay of the previous part is , followed by a delay of . When is a node in far-sink area and the data it transferred is , its delay is divided into two cases similar to , but delay-tolerant data wait for at each relay, so its delay is more than delay-sensitive data.
Definition 6. Total network delay (denoted as ) refers to the sum of transmission delays of the entire network. Since the transmission delay is distinct for different distances from the sink, hence, is determined by integrating the transmission delays of different regions. Suppose that is the node whose distance to the sink is , its end-to-end delay is , taking an arbitrary annular area with the distance from the sink , the width of , and the angle of , then the delay of this region is , and, therefore, the delay of the entire network can be expressed as
Figures 12–16 show the transmission delays in DR-GA and DSDR. Figure 12 is the one-hop delay of them. In DR-GA, the one-hop delay of all nodes is the same due to the constant parameter values. In DSDR, the one-hop delay of nodes that transmit delay-sensitive data is divided into two cases. After the data reach the E2RP, the communication duty cycles of relay nodes are 1 and the delay is small; otherwise the one-hop delay of the nodes is the same as that in DR-GA. The delay of delay-tolerant data in DSDR is similar to that of delay-sensitive data, but the nodes stay for a short period of time for data fusion at each relay, so the delay is larger than the delay-sensitive data.
Figure 13 shows the end-to-end delay under DR-GA scheme. It can be seen that the transmission delay increases rapidly with the distance from the sink, and the maximum delay in the network is about 4–6 times the minimum delay.
Figure 14 shows the end-to-end delay for different types of data under DSDR scheme. As can be seen from the figure, transmission delay of the delay-sensitive data is much smaller than that of delay-tolerant data and delay in the far-sink area is greatly reduced by transmitting data via E2RPs, and the maximum delay in the network is about 2.5 times the minimum delay. Combining Figures 13 and 14 it can be seen that DSDR scheme has a significant advantage in reducing transmission delay when comparing with DR-GA scheme.
Figures 15 and 16 show the transmission delay at different duty cycles and communication radius under DR-GA and DSDR scheme. As can be seen from Figure 15, when the node duty cycle is 0.2, the transmission delay under the DSDR scheme is much lower than the DR-GA, which means that the DSDR can improve the transmission speed with less energy cost. And as shown in Figure 16, obviously, compared with the DR-GA scheme, the DSDR scheme proposed in this paper has high efficiency in transmitting both delay-sensitive data and delay-tolerant data, especially in the transmission of delay-sensitive data, and it can reduce the delay by 44.31%.
5.2. Energy Utilization
According to (11), in DR-GA, suppose that is the node whose distance to the sink is i, its communication radius is r, the probability of generating data is , and then the number of packets received by is , where Z is the largest positive integer that makes the expression true. The number of packets transmitted by is the sum of the packets received and the packets it generated, and it can be expressed as follows:
Similarly, according to (12), in DSDR, suppose that is the node whose distance to the sink is i, the distance from to its nearest E2RP is , and the number of packets received by is , and then the number of packets transmitted by can be expressed as follows:
Theorem 7. In DSDR, the direct-forwarding is adopted for delay-sensitive data and wait-forwarding is adopted for delay-tolerant data, and suppose that is the node whose distance to the sink is i, and the number of packets received by is , where stands for delay-sensitive data, the packet length of delay-sensitive data is and the packet length of delay-tolerant data is , and , where is the correlation coefficient of data, and the data has passed k hops before relaying to , and then the amount of data received by can be expressed as follows:
Proof. The amount of data received by is divided into the amount of delay-sensitive data and the amount of delay-tolerant data. Since of is delay-sensitive data, the number of delay-sensitive packets is , the length of each delay-sensitive packet is , and, therefore, the amount of delay-sensitive data received by is . And delay-tolerant data are fused at each hop, suppose that the data has passed k hops before it relays to , then the data packets after data fusion can be obtained according to (14), and the amount of delay-tolerant data received by is the product of data packets after data fusion and the length of each packet.
The amount of data transmitted by can be obtained by fusing the received delay-tolerant data once again:
Definition 8. Average data amount of an E2RP (denoted as ) refers to the average data amount borne by each E2RP, including the receiving data amount and transmitting data amount. Since E2RPs only forward the data of far-sink area, therefore, is the quotient of the total data amount of the far-sink area and the total number of E2RPs. Suppose that the node from the sink i is , the data amount of is , and there are E2RPs in the network, taking an arbitrary annular area with the distance from the sink , the width of , and the angle of , and then the data amount of the far-sink area can be obtained by integrating the area with the radius of r-R and the angle of 0-2. Consequently, the average data amount of an E2RP can be expressed as
Figure 17 shows the data amount of different nodes in the far-sink area under the DSDR scheme. As we can see from the figure, because the nodes on E2RPs are expected to forward data from the nearby far-sink area, their data amount is much higher than ordinary nodes’, which are about 5 times. The data amount of each node in the network increases gradually with the increase of the duty cycle, especially the nodes on E2RPs.
Equation (4) is the energy consumption model: , where is the energy consumption in event sensing, is the energy consumption in data transmission, is the energy consumption in data receiving, and is the energy consumption in low power listening. Energy consumption in data transmission and data receiving is closely related to the data amount, and when the data amount is large, the energy consumption of nodes is serious.
According to [17], suppose that the node communication duty cycle is , communication cycle is , and then the energy consumption to transmit a packet can be expressed as follows:
The first item is the transmission consumption to notify the receiving node of the periodic preamble when the packet arrives, and is the data transmission consumption.
With the same duty cycle and communication cycle, the energy consumption to receive a packet can be expressed as follows:where is the energy consumption to receive sequences, is the energy consumption to receive data, and is the energy consumption to transmit the acknowledgment messages to the sending node.
The energy consumption of event sensing is composed of its sleep consumption and sensing consumption. Suppose that the sensing duty cycles of nodes are , the power consumption of event sensing is , and the power consumption of sleeping is , and then the energy consumption of nodes in a sensing cycle of can be expressed as follows:
The energy consumption of low power listening is relatively small, according to [14], suppose that is the node whose distance to the sink is i, its communication duty cycle is , communication cycle is , and then the energy consumption of in low power listening can be expressed as follows:
The energy consumption of nodes in low power listening consists of listening consumption and sleeping consumption, that is, , but it is required to subtract the consumption of and that have been calculated in data transmission and receiving.
The calculations for and in (29) are given in [14, 43], as shown in
At this point, the energy consumption of each part for the node has been given. According to the energy consumption model of this paper, suppose that is the node whose distance to the sink is i, and then the energy consumption of under the DSDR scheme can be expressed as follows:
Figures 18–20 show the energy consumption under DR-GA and DSDR. Among them, Figures 18 and 19 describe the energy consumption corresponding to different duty cycles under DR-GA and DSDR, respectively. As shown in Figure 18, the energy consumption under DR-GA decreases gradually with the distance from the sink, and the energy consumption of near-sink area is serious, while the consumption of far-sink are is slight, and the maximum energy consumption in the network is about 3 times the minimum energy consumption, which indicates that the energy consumption under DR-GA is not balanced. While, in the DSDR scheme shown in Figure 19, the energy consumption of the entire network is much more balanced, the maximum energy consumption in the network is less than twice the minimum, and the energy consumption of the entire network presents such a law: the energy consumption of regions near the sink or E2RPs is serious, and it decreases with the distance from them. However, whether DR-GA or DSDR, the node with the most serious energy consumption is still the node closest to the sink, and the maximum energy consumption in DSDR is slightly smaller than DR-GA due to the reduction of redundancy by data fusion, which also shows the efficiency of DSDR.
Figure 20 shows the energy consumption of different data relevance under the DSDR scheme, where is the correlation coefficient of data. Since directly affects the data fusion rate, when is larger, the data can be fused to a greater degree, then the data amount needs to be transmitted is reduced, and the energy consumed by data forwarding is less. As shown in Figure 20, when is between 0.2 and 0.3, its energy consumption is much smaller than it is between 0.1 and 0.2.
Definition 9. Energy difference (denoted as ) refers to the difference of the maximum energy consumption and the minimum energy consumption in the network. Suppose there are N nodes in the network, the kth node is denoted as , and the energy consumed by is , and then the calculation of is depicted as
Definition 10. Energy utilization (denoted as ) refers to the ratio of the total energy consumed by the network when the network dies to the total initial energy of the network, which is determined by the quotient of the energy consumed by all nodes and the initial energy of all nodes. Suppose there are N nodes in the network, the kth node is denoted as , and the energy consumed by is , the initial energy of is , and then the calculation of is depicted as
Figure 21 is the energy consumption difference under DSDR and DR-GA, and the smaller the difference is, the more balanced the energy consumption is; otherwise it shows that the energy of some area in the network is not fully utilized. It is clear that compared with DR-GA the DSDR proposed in this paper has a lower energy difference, which shows that the energy consumption under DSDR is more balanced. Figure 22 shows the energy utilization of DSDR and DR-GA. Compared with DR-GA, the DSDR can improve the energy utilization by 30.61%, and especially when the duty cycle is large it has significant advantages in energy utilization.
5.3. Network Lifetime
Definition 11. Network lifetime (denoted as ) refers to the time intervals from the time when the network is established to the time when the network is dead, which is defined as the death time of the first node in the network, where the death of nodes resulted from exhausted battery energy. Suppose there are N nodes in the network, the kth node is denoted as , and the energy consumed by is , the initial energy of is , and then the is depicted as
Figure 23 shows the network lifetime under different data fusion rate, and the higher the data fusion rate is, the less data are needed to transmit, the less energy the nodes consume, and the longer the network lifetime is.
Figure 24 is the lifetime comparison of DSDR and DR-GA, although the energy utilization greatly improved under DSDR scheme, the increasing consumption has no impact on network lifetime. What is more, it makes the lifetime have a certain improvement. This is explained by the fact that, under the DSDR proposed in this paper, we choose different routings for different areas. In the far-sink area we use to transmit data and switch the E2RPs intelligently, so the energy consumption of E2RPs in far-sink area will only enhance energy utilization and do not affect the maximum energy consumption of the network. For the near-sink area we retain the original data routing mode, and the near-sink nodes are often the nodes that affect the entire network lifetime, under the condition that the energy consumption of these nodes is constant, and the lifetime of the network remains unchanged. In addition, data redundancy is reduced by data fusion, so the maximum energy consumption under DSDR is smaller than that of DR-GA.
5.4. Effect of Other Parameters on the Performance
(1) The Proportion of Delay-Sensitive Data
Theorem 12. Suppose that there are nodes in the far-sink area of the network, the kth node is denoted as , and the data amount of is , where stands for delay-sensitive data, the transmission delay for delay-sensitive data is , and the transmission delay for delay-tolerant data is , and then the average transmission delay of far-sink area can be expressed as follows:
Proof. The transmission delay of is composed of the transmission delay of delay-sensitive data and the transmission delay of delay-tolerant data. Because the delay for the delay-sensitive packet is , and the proportion of the delay-sensitive data to the data amount is , then the delay of delay-sensitive data is . Correspondingly, the delay of delay-tolerant data is . The average transmission delay of far-sink area is obtained by dividing the total delays into the number of nodes.
Similar to Theorem 12, suppose that there are nodes in the far-sink area of the network, the kth node is denoted as , and the data amount of is , where stands for delay-sensitive data, the energy consumption to transmit a delay-sensitive packet is , the energy consumption to transmit a delay-tolerant packet is , and then the average energy consumption of far-sink area can be expressed as follows:
Figures 25 and 26 show the transmission delay and energy consumption at different proportion of delay-sensitive data under DSDR scheme. In Figure 25, because the delay of delay-sensitive data is less than delay-tolerant data, then the higher the proportion of delay-sensitive data is, the smaller the average transmission delay of the network is. In Figure 26, the greater the proportion of delay-sensitive data is, the more the serious energy the network consumes, because the delay-tolerant data are transmitted after data fusion and the data redundancy is greatly reduced, so energy consumption of delay-sensitive data is greater than delay-tolerant data. From Figures 25 and 26, we can see the effectiveness of DSDR: when there are more delay-sensitive data in the network, the DSDR can transfer data to the sink quickly and has a high transmission efficiency; when there are more delay-tolerant data in the network, the DSDR can reduce energy consumption and improve network lifetime.
(2) The Probability of Data Generation. Figures 27 and 28 show the effect of the probability of data generation on network performance. When the probability is higher, the network needs to transmit more data, and the data amount each node needs to forward is greater, thus the energy consumption is more serious, and the network energy utilization is higher, but this also shortens the network lifetime. From Figure 27, we can see that the energy utilization increases with the increase of data generation rate and duty cycle. Contrary to the energy utilization, it can be seen from Figure 28 that the network lifetime is shortened as the data generation rate increases.
(3) Network Radius and Node Communication Radius. The relationship between communication radius of nodes and network radius is also an important factor that affects the performance of the network. When the communication radius of the node is larger, it means that the node can forward farther at each hop, and the number of relay hops is less.
Figures 29 and 30 show the effect of the node communication radius on the transmission delay. It can be seen that when the communication radius is greater, the number of relay hops is less and the transmission delay is smaller. Similarly, when the network radius is small, the node needs to be relayed multiple hops before its arrival to the sink, the energy consumption is more serious, and the energy consumption directly affects the energy utilization, and also it has an impact on the network lifetime, as shown in Figures 31 and 32.
6. Conclusions and Future Work
Applications running on the same network may have different service requirements depending on the different backgrounds of their events, and the basic requirement is the transmission delay. Traditional wireless sensor networks lack intelligence and cannot dynamically select appropriate data routing according to the delay sensitivity of the data, so it cannot meet the requirements of the events. Then, a delay Differentiated Services for Data Routing scheme (DSDR) is creatively proposed in this paper, which establishes energy-efficient routing paths (E2RPs) at different locations of the network to forward data. In DSDR, data routing is intelligently selected according to the delay requirements of data. For delay-sensitive data it provides direct-forwarding method to minimize the transmission delay. For delay-tolerant data it uses wait-forwarding method and reduces redundancy by data fusion, thus to reduce network energy consumption. A comprehensive performance analysis shows that DSDR has significant advantages in terms of transmission delay and energy utilization. Compared with the general approaches, DSDR can reduce the transmission delay of delay-sensitive data by 44.31%, reduce transmission delay of delay-tolerant data by 25.65%, and improve the energy utilization by 30.61%, while also enabling the network to have a lifetime that is not less than the previous studies.
With the depth of the relevant research, the content of this paper can be extended. In this paper, we mainly study the delay differentiated services of data; in fact, the integrity of the data can also be differentiated by establishing different transmission routings in the network according to the delay and integrity requirements. Therefore, in the next working plan, we consider data integrity in the guarantee of transmission latency.
Parameters Related to Calculation
| : | Sensing duty cycle |
| : | Communication duty cycle |
| : | Energy consumption in low power listening |
| : | Energy consumption in data transmission |
| : | Energy consumption in event sensing |
| : | Energy consumption in data receiving |
| : | Receiving data packets |
| : | Sending data packets |
| : | Receiving data amount |
| : | Transmitting data amount |
| : | End-to-end transmission delay |
| : | Energy consumption difference |
| : | Energy utilization rate |
| : | Network lifetime |
| : | Energy consumed by node |
| : | Total number of sensor nodes |
| : | The th node in the network |
| : | The node whose distance to the sink is . |
Conflicts of Interest
The authors declare that there are no conflicts of interest regarding the publication of this article.
Acknowledgments
This work was supported in part by the National Natural Science Foundation of China (61772554, 61379110, 61370229, and 61370178); the National Basic Research Program of China (973 Program) (2014CB046305); the National Key Technology R&D Program of China under Grant no. 2014BAH28F02; the S&T Projects of Guangdong Province under Grant nos. 2014B010103004, 2014B010117007, 2015B010110002, and 2016B010109008; and the S&T Project of Guangzhou Municipality under Grant no. 201604010054.