Energy-aware determination of compression for low latency in solar-powered wireless sensor networks

Abstract

There have been many studies performed about increasing network lifetime in wireless sensor networks that involve reducing data size, since the data transmission process takes up a large part of energy consumption. However, reducing data size results in increased delay time due to not only the compression computation time but also the waiting time to gather a sufficient amount of data for compression. Meanwhile, in solar-powered wireless sensor networks, the harvested energy may be surplus to the basic operations of sensor nodes. In this study, such surplus energy is utilized to reduce the delay time between nodes. Nodes with residual energy less than a certain threshold transfer data with compression in order to reduce energy consumption, and nodes with residual energy over the threshold (which means there is surplus energy) transfer data without compression to reduce the delay time between nodes by using the surplus energy. Simulation-based performance verifications show that the technique proposed in this study exhibits optimal performance in terms of both energy and delay times compared with traditional methods.

Keywords

Wireless sensor network solar energy latency compression energy-aware

Introduction

A wireless sensor network (WSN) can be defined as a network of devices, denoted as nodes, which can sense the environment and communicate information gathered from the monitored field through wireless links. The data are forwarded, usually via multiple hops, to a sink. Monitoring and communication are performed cooperatively by the nodes. WSNs have become an increasingly active field of research in recent years because it proposes the novel and practical idea of making many small objects with limited capabilities (the sensors) collaborate to create a very versatile and powerful system (a WSN). In fact, WSNs are used in various specific applications in fields as diverse as the military, medicine, ecological monitoring, and smart homes.

Generally, in WSNs, a node has limited hardware resources, and the lifetime of a network is usually limited by the availability of energy from the batteries or other sources. Attempts to extend network lifetime have involved techniques such as duty-cycle control, data transmission power level control, and energy-efficient routing. Alternatively, energy-collecting sensor nodes with rechargeable batteries can be used to extend network lifetime indefinitely.¹ The use of solar energy is preferred because of its periodicity, predictability, and density. It is notable that the amount of harvested energy may be surplus to the amount of energy necessary to the basic operations of the sensor node. Yang et al.² have suggested a method for determining the amount of surplus energy during a given period in their research.

Furthermore, there have been many studies performed about increasing network lifetime in WSNs involving reducing data size, since the data transmission process takes up a large proportion of the energy required by a sensor node. However, reducing the data size results in increased delay time due to not only the compression computation time but also the waiting time to gather a sufficient amount of data for compression, which means there is a trade-off between the energy saved by reducing the amount of transmissions, and the time required for compression.³

In this study, we introduce an energy-adaptive selective compression scheme (EASCS). This scheme involves determining an energy threshold used to decide whether there is any surplus energy. If the residual energy of a node is below this energy threshold (which means there is no surplus energy), then a node compresses data before transmission in order to reduce its energy consumption; otherwise, the node transfers data without compression to reduce delay. This technique helps reduce end-to-end delay without causing blackout time, during which sensor nodes stop working without warning.

The remainder of this article is organized as follows: in section “Related work,” we explain energy-harvesting WSNs and data compression schemes for WSNs. In section “Experimental study on energy and delay for compression,” we describe the experimental results of energy consumed by the compression algorithm and the delay that it incurs. In section “The proposed EASCS,” we introduce our EASCS with a threshold determination algorithm. In section “Performance evaluation,” performance verification through simulation is presented. Finally, we draw conclusions and discuss the scope for future work.

Related work

Amount of harvested solar energy

Recent studies on energy-harvesting sensor nodes have focused on finding fundamental solutions for energy limitation. Energy-harvesting sensor nodes collect energy using various nearby environmental-friendly energy sources and store and consume the harvested energy using rechargeable batteries. Such sources of environmental energy include sunlight, wind, temperature difference, vibration, sound, and the movements generated by the user of a wearable device. Their respective power densities are listed in Table 1,⁴ and we see that solar energy has by far the highest density.

Table 1.

Power density of various environmental energies.

Power source	Power density
Sunlight (outdoors at noon)	15 mW/cm²
Wearer movement (shoe inserts)	330 µW/cm²
Vibration (microwave oven)	116 µW/cm³
Temperature difference (10°C)	40 µW/cm³
Sound (100 dB)	960 nW/cm³

Table 2⁵ shows the power requirement of various commercial sensors. Considering the duty cycle of a sensor node is usually less than 0.1, it is confirmed that the solar energy–harvesting system can meet the power demand of most sensor systems. For example, the Heliomote,⁶ one of the solar-powered sensor nodes developed at UCLA, draws an average of 25 mW, and it has a solar panel with an area of 60 cm². X Jiang et al.⁷ review the characteristics of solar energy as a power source for sensor systems in more detail.

Table 2.

Power consumption of various sensor devices.

Sensor	Producer	Sensing	Power demand (mW)
STCN77	STM	Temperature	0.4
QST108KT6	STM	Touch	7
SG-LINK (1000 Ω)	MicroStrain	Strain	9
SG-LINK (350 Ω)	MicroStrain	Strain	24
iMEMS	ADI	Acceleration	30
2200/2600 Series	Gems	Pressure	50
T150	Gefran	Humidity	90
LUC-M10	Pepperl+Fuchs	Level	300
CP18, VL18, GM60	Visolux	Proximity	350
TDAA0161	STM	Proximity	420
FCS-GL1/2A4-AP8X	Turck	Flow	1250

Energy-harvesting sensor system

Since solar energy has a high-power density and a periodic charge cycle, it is one of the most attractive energy sources. Therefore, many solar-powered WSNs^8–10 were designed as prototypes. Most of these schemes focused on a node-level design such as power controls or hardware structures and did not address performance optimization from the application or network perspective, such as data throughput or data reliability. For example, a study by Kansal et al.⁸ explained the relationship between design factors and important issues that should be considered during the design of solar-powered nodes and implemented a prototype called Heliomote. A further study by Alippi and Galperti⁹ designed a low-power maximum power point tracker (MPPT) circuit to effectively transmit the solar energy harvested by solar panels to a rechargeable battery. And J Taneja et al.¹⁰ proposed a systematic approach (e.g. types of solar panel, regulator, energy storage, and capacitor) to design a micro solar-powered system.

In recent years, research on the performance at network-level of solar-powered WSNs has started.^11,12 For example, Noh et al.¹² studied end-to-end delay of data as the quality of service (QoS) and proposed a low-delayed data transmission technique that considered geographical locations, the amount of solar energy harvested, and the duty cycle of neighbor nodes.

In Yang et al.,² effective data distribution and resource allocation techniques to minimize data loss for solar-powered WSNs were proposed. A node with a low amount of residual energy just sensed and stored data locally. When this remaining energy was above a certain threshold, the node started extra functions in order to minimize loss of data, such as data replication and distribution.

Data compression for WSNs

Generally, sensor nodes have limited energy, bandwidth for communication, processing speed, and memory space. Therefore, many studies conducted so far focused on how to achieve the maximum utilization of these resource-constrained nodes while minimizing energy consumption. One subset of these studies focused on data compression. Compressing data is expected to reduce power consumption involved in transmitting data to neighbors and thus to extend the network lifetime. In addition, by reducing the data size, less bandwidth is required for sending and receiving data. However, one point we have to address in data compression is that compression itself requires some amount of energy. This is the reason why typical compression algorithms for general desktop computers are inappropriate for sensor nodes, and they generally require too many system overheads, including energy. Therefore, researchers have sought optimal ways to compress the data on a sensor node and have suggested several compression schemes tailored for WSNs^13–17 that consume much less energy for compression than previous schemes did. Especially, some studies^15–17 suggest efficient schemes based on lossy compression algorithm.

Yao et al.¹⁵ proposed a data collection protocol called EDAL, an energy-efficient delay-aware lifetime-balancing data collection protocol in WSNs, which is inspired by recent techniques developed for open vehicle routing problems with time deadlines (OVRP-TD) in operational research. The goal of EDAL is to generate routes that connect all source nodes with minimal total path cost, under the constraints of packet delay requirements and load balancing needs. The lifetime of the deployed sensor network is also balanced by assigning weights to links based on the remaining power level of individual nodes. They proved that the problem formulated by EDAL is NP-hard. Therefore, they developed a centralized heuristic to reduce computational overhead and a distributed heuristic to make the algorithm scalable for large-scale network operations.

Xu et al.¹⁶ have investigated a novel power-aware data collection scheme, hierarchical data aggregation using compressive sensing (HDACS), for large-scale dense WSNs. They addressed this problem by incorporating compressive sensing (CS) in a multilevel data aggregation hierarchy. They proved theoretically the advantages of the proposed HDACS over the existing work, in terms of the amount of data communicated in each cluster head, the total data volume for transmission, and the data compression ratio. Furthermore, they constructed a novel energy model by taking the computation cost in the processor into account, and they showed that computation cost is insignificant compared to communication cost with the aid of real hardware specs.

The previous two studies^15,16 are distributed data compression approaches, while Zordan et al.¹⁷ proposed a local data compression approach which is similar to our compression scheme. Zordan et al. introduced optimal compression/transmission policies through a Lagrangian relaxation approach combined with a dichotomic search for the Lagrangian multiplier. With that, they characterized the optimal policies theoretically and numerically. To propose the optimal policy, they studied the problem of designing efficient policies to control the process of data compression for wireless transmission over fading channels in the presence of a stochastic energy input process and a replenishable energy buffer. They first modeled the transmission and energy dynamics of a sensor node implementing practical lossy compression methods as a constrained Markov decision problem. They proved that, under realistic assumptions, the optimal policy is non-decreasing in each of the system state components, that is, energy buffer state, channel state, and energy source state. This study¹⁷ agrees with the context of our research in that it uses harvesting energy efficiently to improve network performance. However, while Zordan et al.¹⁷ aim to utilize energy to reduce data loss, our scheme aims to reduce latency for soft real-time applications.

We observed that the data latency (end-to-end delay) could still increase when the compression schemes tailored for WSNs are used, due to not only the compression computation time but also the waiting time to gather a sufficient amount of data for compression. In order to solve this problem, this article explores the trade-off between end-to-end delay and energy consumption and suggests a new selective compression scheme.

Experimental study on energy and delay for compression

S-LZW algorithm

Sensor Lempel–Ziv–Welch (S-LZW)¹⁸ is one of the typical directory-based lossless compression schemes used in WSNs due to its high compression ratio and light weight. S-LZW is a re-designed compression method for the WSN environment by reducing the weight of the LZW compression algorithm that has been used widely in desktop PC environments.

To adapt LZW to a sensor node, S-LZW balance three major inter-related points: the dictionary size, the size of the data to be compressed, and the protocol to follow when the dictionary fills. First, the memory constraint requires that LZW uses the dictionary size as small as possible. Additionally, to decode a dictionary entry, the decoder must have received all previous entries in the block. Unfortunately, however, sensor nodes never deliver 100% of their data to the source. For these reasons, S-LZW separates the data stream into small, independent blocks so that if a packet is lost it only affects the data that follows it in its own block, as shown in Figure 1.

Figure 1.

Separation of the data into individual blocks before compression.¹⁸

Experimental results for compression

From now on, we present the experimental results of both the energy consumption and processing delay of the compression algorithm. The compression algorithm of S-LZW¹⁸ was tested with a dataset of SensorScope¹⁹ gathered by 20 mica2 motes, which have the radio and processing modules CC1000 and TIMSP430X1611, respectively. The mote senses information about light, temperature, and sound, and the sampling rate was 48 bytes/s. The compression process starts to run when 528 bytes or more of sensory data are gathered, since this amount leads to the best performance of the S-LZW.

When compressing 528 bytes of data with the S-LZW, the data size reduced to 201 bytes as shown in Figure 2(a). Energy consumption and processing delay are also shown in Figure 2(b) and (c), respectively. Owing to the smaller data size, the entire energy consumption when compressing data with the S-LZW was 58% less than with the non-compression scheme, although some amount of energy (0.33 mW) was consumed for the compression processing. However, we noted that compression incurred the additional delay of 4.52 s including time of gathering and compressing data, as shown in Figure 2(c), which is critical to the delay-sensitive applications.

Figure 2.

Energy consumption in the case of non-compression and compression: (a) comparison of data size, (b) comparison of energy consumption, and (c) comparison of delay.

The proposed EASCS

In this section, the energy-threshold determination scheme and the EASCS are explained.

Overview of the proposed EASCS

In the EASCS, the node operates in either CompressionMode or LatencyMode depending on the energy status of itself and the next-hop node on a route to the sink. If the amount of residual energy of both nodes exceeds a certain threshold, then the node runs in LatencyMode where it transfers data without compression in order to decrease the delay. This mode requires a relatively large amount of energy as explained in section “Experimental study on energy and delay for compression,” but this does not matter since there is surplus energy over the threshold and the node only uses this surplus energy to reduce the end-to-end delay. Otherwise, the node operates in CompressionMode where the data are transferred after being compressed by the S-LZW algorithm. This mode is more efficient than LatencyMode from the energy-saving aspect, but it incurs a high latency. Since there is a deficient amount of energy, operating a node in CompressionMode can help to prevent blackout time.

By dynamically selecting these two modes depending on residual energy, the EASCS controls the trade-off between delay and energy consumption efficiently. More specifically, in the EASCS, the node tries to reduce the delay only when surplus energy comes along, and thus blackout time does not increase at all.

Determination of node operation mode using the energy threshold

The energy threshold, which determines the two operating modes described in the above section, is related to the energy-consumption rate, solar energy–harvesting rate, and residual energy in the battery of the target node. The solar energy–harvesting rate is affected by the types and location of solar cell panels, weather, and season; the energy-consumption rate of the node is affected by its data sensing ratio, data transfer ratio, and duty cycle. However, these factors cannot be accurately estimated. In this study, therefore, a simple but efficient energy model that does not require accurate estimation of these factors was used.² Given that the average energy-harvesting rate of Node $n_{i}$ is $P_{solar} (i)$ and its average energy-consumption rate is $P_{sys} (i)$ , the value of $P_{sys} (i)$ can be derived by profiling the amount of energy consumed while the node is in operation. In addition, the value of $P_{solar} (i)$ can also be obtained by profiling the amount of energy harvested. When the residual energy currently available in the battery in Node $n_{i}$ is $E_{residual} (i)$ , the amount of time expected for the battery to be fully charged in the node is represented as follows

T_{full} (E_{residual} (i)) = \frac{C (i) - E_{residual} (i)}{P_{solar} (i) - P_{sys} (i)}

(1)

where $C (i)$ is the overall battery capacity in the node. It should be noted that the battery is charged only when the condition $P_{solar} (i) > P_{sys} (i)$ is satisfied. The value of $P_{solar} (i)$ cannot be adjusted, whereas the desired value of $P_{sys} (i)$ can be controlled by changing the duty cycle of Node $n_{i}$ . Thus, the condition $P_{solar} (i) > P_{sys} (i)$ can be fulfilled by adjusting the value of $P_{sys} (i)$ .

Meanwhile, the amount of solar energy harvested cannot be estimated accurately because the amount of solar energy charged constantly varies according to the weather and time. However, when the amount of residual energy in the battery satisfies the condition in equation (2), blackout time from the current time to the time when the battery is fully charged remains zero

E_{residual} (i) \geq P_{sys} (i) \cdot T_{full} (E_{residual} (i))

(2)

In other words, a system that satisfies equation (2) operates without blackout time until the subsequent time when the battery is fully charged, even in the worst situation. $E_{residual} (i) \geq (P_{sys} (i)) / (P_{solar} (i)) \cdot C (i)$ is derived through equations (1) and (2). It means that a system operates properly even in unexpected situations such as changes in the weather as long as its battery has energy greater than $(P_{sys} (i)) / (P_{solar} (i)) \cdot C (i)$ . In this study, such this minimum required energy is defined as an energy threshold and is represented in by the following equation

E_{threshold} (i) = \frac{P_{sys} (i)}{P_{solar} (i)} \cdot C (i)

(3)

When $E_{residual} (i) < E_{threshold} (i)$ , blackout time may exist for Node $n_{i}$ . Thus, the node operates in CompressionMode to minimize energy consumption. In contrast, the condition $E_{residual} (i) > E_{threshold} (i)$ implies that there is enough residual energy for a node to operate ordinarily. Accordingly, the node operates in LatencyMode to minimize latency by using the surplus energy. However, it should be noted that equation (1) is satisfied only when $P_{solar} (i) > P_{sys} (i)$ . For this reason, equation (3), derived through equation (1), is also valid only when the same condition $P_{solar} (i) > P_{sys} (i)$ is fulfilled. When $P_{solar} (i) < P_{sys} (i)$ , the value of $E_{threshold} (i)$ is higher than that of the battery capacity $C (i)$ . As the amount $E_{residual} (i)$ of residual energy in a battery cannot exceed the value of $C (i)$ , this node always operates in CompressionMode. Therefore, the value of $P_{sys} (i)$ constantly decreases over time, and the condition $P_{solar} (i) > P_{sys} (i)$ becomes satisfied. Subsequently, the node can dynamically select either mode based on equation (3). Figure 3 shows the relationship between the energy model proposed in this study and system mode.

Figure 3.

Relation between residual energy, threshold, and operating modes.

In addition, for a prudent determination of the node operation mode, our scheme considers the energy status of its neighbor nodes as well as its own energy status. This is because the operation in LatencyMode can increase the energy consumption of its next-hop node on the route to the sink. Therefore, assuming that node $n_{j}$ is the next routing target of node $n_{i}$ , node $n_{i}$ operates in LatencyMode only when both conditions of $E_{residual} (i) > E_{threshold} (i)$ and $E_{residual} (j) > E_{threshold} (j)$ are satisfied.

Practical algorithm of the EASCS

We would use the symbols defined in Table 3 to represent the practical code of the EASCS. EASCS(i) is invoked and periodically conducted in Node $n_{i}$ . In the meantime, unnecessary changes can frequently occur during the process of adjusting the modes of the node. For example, the following case can be assumed. Suppose a node is operating in CompressionMode and the value of $E_{residual} (i)$ is about to exceed that of $E_{threshold} (i)$ , resulting in the mode of Node $n_{i}$ being changed to LatencyMode. At this time, the value of $E_{residual} (i)$ is likely to decrease below the threshold again because Node $n_{i}$ now has insufficient energy for stable operation. Thus, the mode of Node $n_{i}$ is swiftly changed to CompressionMode, and the node may repeatedly change its operating modes. Such frequent mode adjustment leads to reduced system stability and performance. Thus, an impact attenuation level (ψ) is applied to prevent frequent mode adjustment in this study, as shown in Figure 3.

Table 3.

Notations for EASCS-Init(i) and EASCS(i).

Symbol	Meaning
$E_{residual} (i)$	Amount of residual energy of node $n_{i}$
$E_{threshold} (i)$	Energy threshold for changing mode, $(P_{sys} (i) / P_{solar} (i)) C$
$m_{i}$	1 if sensor node $n_{i}$ is in LatencyMode, otherwise 0
ψ	Energy window to prevent frequent change of the mode
S-LZW	S-LZW compression algorithm
period	Periodic invocation cycle of the $EASCS (i)$

S-LZW: Sensor Lempel–Ziv–Welch; EASCS: energy-adaptive selective compression scheme.

In addition, our algorithm considers the energy state of the neighboring nodes, that is, the operating mode of the neighboring node, when determining the operation mode of the node. In order to obtain this information without additional overhead, the reserved bits of the multiple access control (MAC) frame header are used as shown in Figure 4 to exchange the information of operation mode between nodes.

Figure 4.

Operating mode information in reserved bits of MAC header.

Performance evaluation

We measured the blackout time and end-to-end latency over time in order to evaluate the performance of the proposed EASCS and compared them with non-compression scheme, compression scheme, temporal compression strategies (TCS),¹⁷ and aggregation scheme. In the compression scheme, nodes always compress data gathered for a while before transmission. In contrast, in the non-compression scheme, nodes do not compress the data and transmit it immediately. TCS uses selective compression according to energy state like the proposed EASCS, but the method of determining node’s operation mode is different from our scheme. EASCS mathematically calculates the optimal threshold, and based on this it determines whether to compress or not according to the state of the remaining energy, while the TCS technique determines it according to the presence or absence of current energy input. Finally, the aggregation scheme is a technique of collecting data for a certain period of time to remove redundant data (but not to compress) and transmit them at a time. All of these schemes were measured in a SolarCastalia simulator,²⁰ which was designed to simulate solar-powered WSNs.

Test environment

The proposed technique is designed for solar-powered WSNs in which each sensor node periodically harvests energy from environments. We applied a sample harvested energy data to the simulation, which is measured with ez430-RF2500-SEH²¹ mote for 1 month (May 2014) in Seoul, Korea, and provided by SolarCastalia. The average amount of harvested energy during 1 day is about 8.417 mAh.²²

In this simulation, the WSN was composed of 20–60 solar-powered sensor nodes and it used the energy-aware location-based routing scheme.²³Table 4 summarizes the important parameters used in this simulation.

Table 4.

Simulation parameters.

Parameter	Value
Number of nodes	20, 30, 40, 50, 60
Node placement	Random
Size of field area	10 × 10 (m)
Duration	30 days
Duty cycle	0.1
Transmission range	3.5 m
TX energy	0.021 mW/byte
RX energy	0.015 mW/byte
Compression energy	0.00165 mW/byte (S-LZW)
Routing scheme	Energy-adaptive location-based routing

S-LZW: Sensor Lempel–Ziv–Welch.

Analysis of the blackout time

In order to analyze the energy efficiency of the EASCS, we measured the blackout time of all the nodes in each scheme. Figure 5 shows the sum of blackout time as the number of nodes was varied from 20 to 60, and Figure 6 shows the number of blackout nodes with time during 2 days when the number of nodes is 60. As shown in Figures 5 and 6, the blackout time of the proposed scheme is shorter than both the non-compression scheme and aggregation scheme. This means that our scheme consumes less energy for transmitting data both of them, as transmitting compressed data consumes less energy. They also show that the blackout time of our scheme is almost the same as the compression scheme. This means that the proposed scheme only utilizes excessive energy to transmit non-compressed data, and it has no influence on the number of blackouts of nodes. TCS uses selective compression depending on the energy state as in EASCS, so it shows blackout time similar to our scheme. In addition, the result in Figure 5 in which the EASCS shows a stable performance, even when the number of nodes increases, verifies the good scalability of the EASCS.

Figure 5.

Comparison of blackout time for various numbers of nodes.

Figure 6.

Number of blackout nodes changes with time during two sample days.

We also confirm that the average values of energy harvest and consumption (per day at each node) are about 390 and 249 J, respectively. In addition, the standard deviation of energy consumption among each node is about 75.7. This deviation is due to the difference in energy consumption depending on the distance to the sink. The closer to the sink, the more data is transmitted, so the more energy is consumed.

Analysis of end-to-end delay

We compared the end-to-end delay of our scheme with the compression, non-compression, TCS, and aggregation schemes. Figures 7 and 8 show the end-to-end delay for each scheme by number of nodes and by time, respectively. The end-to-end delay of the proposed scheme is much shorter than the compression and aggregation scheme. That is because the end-to-end delay increases in cases of the compression and aggregation scheme since the node should aggregate a certain amount of data for a period, while each node in our scheme selects CompressionMode or LatencyMode depending on its energy level.

Figure 7.

Comparison of average end-to-end latency for various numbers of nodes.

Figure 8.

Latency changes with time during two sample days.

Although the non-compression scheme has the lowest latency, as shown in Figure 5, the non-compression scheme has the problem that the total amount of data collected in the sink is too small because the blackout time is too large.

In the case of the TCS, as described in section “Analysis of the blackout time,” the blackout time is similar to our scheme, but as you can check in Figures 7 and 8, the delay time is not as good as our technique. This is because the algorithm for determining node’s operation mode performed in the TCS is somewhat naive, so the node may not operate in LatencyMode despite the sufficient energy. However, since TCS considers the state of the channel as well as the energy input when determining the operation mode of the node, the TCS may perform better than the proposed EASCS in a bad channel condition. These results confirm that the proposed scheme can reduce latency effectively without causing any additional blackout nodes compared to the compression scheme.

Conclusion

Solar-powered WSNs are preferred for optimizing energy usage because energy is charged periodically but unpredictably, unlike battery-based networks. The proposed energy-aware compression scheme in this study selectively uses either CompressionMode or LatencyMode to utilize the surplus energy for optimization between latency and energy consumption. LatencyMode follows a non-compression method with large energy consumption and short end-to-end delay, and CompressionMode follows small energy consumption and high end-to-end delay. By doing this, the proposed technique can achieve two goals, low blackout time and delay time, which exist in a trade-off relationship.

Footnotes

Academic Editor: Jose Pelegri-Sebastia

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was supported by the MSIP (Ministry of Science, ICT and Future Planning), Korea, under the ITRC (Information Technology Research Center) support program (IITP-2016-H8501-16-1008) supervised by the IITP (Institute for Information & communications Technology Promotion.

References

Sudevalayam

Kulkarni

Energy harvesting sensor nodes: survey and implications. IEEE Commun Surv Tutor 2010; 13(3): 443–461.

Yang

Wang

Noh

. SolarStore: enhancing data reliability in solar-powered storage-centric sensor networks. In: Proceedings of the 7th annual international conference on mobile systems, applications, and services, Kraków, 22–25 June 2009, pp.333–346. New York: ACM.

Lee

Kim

Chang

IJ.

CPAC: energy-efficient data collection through adaptive selection of compression algorithms for sensor networks. Sensors 2014; 14(4): 6419–6442.

Stojmenovic

Handbook of sensor networks. Hoboken, NJ: Wiley, 2005.

Alippi

Anastasi

Francesco

. Energy management in wireless sensor networks with energy-hungry sensors. IEEE Instru Meas Mag 2009; 12(2): 16–23.

Kansal

Hsu

Zahedi

. Power management in energy harvesting sensor networks. J ACM Trans Embed Comput Syst 2007; 6(4): 1–38.

Jiang

Polastre

Culler

Perpetual environmentally powered sensor networks. In: Proceedings of the forth international symposium on information processing in sensor networks, Los Angeles, CA, 24–27 April 2005, pp.463–468. New York: ACM.

Kansal

Hsu

Srivastava

. Harvesting aware power management for sensor networks. In: Proceedings of the 43rd annual design automation conference, San Francisco, CA, 24–28 July 2006, pp.651–656. New York: ACM.

Alippi

Galperti

An adaptive system for optimal solar energy harvesting in wireless sensor network nodes. IEEE T Circuits Syst 2008; 55(6): 1742–1750.

10.

Taneja

Jeong

Culler

. Design, modeling, and capacity planning for micro-solar power sensor networks. In: Proceedings of the 7th international conference on information processing in sensor networks, St. Louis, MO, 22–24 April 2008, pp.407–418. New York: ACM.

11.

Noh

Lee

Shin

QoS-aware geographic routing for solar-powered wireless sensor networks. J Inst Electron Inf Commun Eng Trans Commun 2007; 90-B(12): 3373–3382.

12.

Noh

Yoon

Shin

Low-latency geographic routing for asynchronous energy-harvesting WSNs. J Networks 2008; 3(1): 78–85.

13.

Srisooksai

Keamarungsi

Lamsrichan

. Practical data compression in wireless sensor networks: a survey. J Netw Comput Appl 2012; 35(1): 37–59.

14.

Petrović

Shah

Ramchandran

. Data funneling: routing with aggregation and compression for wireless sensor networks. In: Proceedings of the first IEEE international workshop on sensor network protocols and applications, Anchorage, AK, 11 May 2003, pp.156–162. New York: ACM.

15.

Yao

Cao

Vasilakos

AV.

EDAL: an energy-efficient, delay-aware, and lifetime-balancing data collection protocol for heterogeneous wireless sensor networks. IEEE/ACM T Network 2013; 23(3): 810–823.

16.

Ansari

Khokhar

. Hierarchical data aggregation using compressive sensing (HDACS) in WSNs. ACM T Sensor Network 2015; 15(2): 1336–1352.

17.

Zordan

Melodia

Rossi

On the design of temporal compression strategies for energy harvesting sensor networks. IEEE T Wirel Commun 2016; 15(2): 1336–1352.

18.

Sadler

Martonosi

. Data compression algorithms for energy-constrained devices in delay tolerant networks. In: Proceedings of the 4th international conference on embedded networked sensor systems, Boulder, CO, 1–3 November 2006, pp.265–278. New York: ACM.

19.

Schmid

Dubois-Ferriere

Vetteri

SensorScope: experiences with a wireless building monitoring sensor network. In: Proceedings of the workshop on real-world wireless sensor networks (RealWSN), Stockholm, 20–21 June 2005.

20.

Kang

Noh

SolarCastalia: solar energy harvesting wireless sensor network simulator. Int J Distrib Sens N 2015; 11(6): 1–10.

21.

Texas instrument EZ430-RF2500 solar energy harvesting kit, http://ti.com/docs/toolsw/folders/print/ez430-rf2500-seh.html

22.

Kang

Kim

Yang

. Energy-aware transmission power control for solar energy harvesting wireless sensor system and its effects on network-wide performance. In: Proceedings of the 34th KIICE fall conference, Jeju, Korea, 15–16 November 2013, p.316 (in Korean). Seoul: KIICE.

23.

Chen

. A novel location-based routing algorithm or energy balance in wireless sensor networks. In: Proceedings of the IEEE communications and mobile computing, Kunming, China, 6–8 January 2009, pp.568–572. New York: IEEE.