An Energy-Efficient Data Forwarding Strategy for ... - IEEE Xplore

Received June 14, 2016, accepted July 26, 2016, date of publication September 21, 2016, date of current version November 18, 2016. Digital Object Identifier 10.1109/ACCESS.2016.2611820

An Energy-Efficient Data Forwarding Strategy for Heterogeneous WBANs DAPENG WU1,2 , (Senior Member, IEEE), BORAN YANG1 , HONGGANG WANG2 , (Senior Member, IEEE), DALEI WU3 , AND RUYAN WANG1 1 Chongqing 2 University 3 University

University of Posts and Telecommunications, Chongqing 400065, China of Massachusetts Dartmouth, Dartmouth, MA 02747, USA of Tennessee at Chattanooga, Chattanooga, TN 37403, USA

Corresponding author: H. Wang ([email protected]) This work was supported in part by the National Natural Science Foundation of China under Grant 61371097 and Grant 61271261, in part by the Youth Talents Training Project of Chongqing under Grant CSTC2014KJRC-QNRC40001, and in part by the NSF Awards under Grant IIS1401711, Grant ECCS1407882, and Grant CNS1429120.

ABSTRACT Monitoring human bodies and collecting physiological information are the major healthcare applications for wireless body area networks (WBANs). Due to the limited energy resource of body sensors, their energy depletions will cause severe network performance degradation, such as latency and energy efficiency. The heterogeneity of body sensors can result in different sensor energy consumption rates. Besides, the important degrees of monitored physiological data could vary hugely. Targeting at the abovementioned problems, an energy-efficient data forwarding strategy (EDFS) is proposed in this paper to balance sensor energy consumption and improve network lifetime and collaborative operations of heterogeneous WBANs. Our major contributions include: 1) the original physiological data are processed by compressed sensing to reduce the data size to be transmitted and 2) the remaining energy levels, sampling frequency, and sensor importance are jointly considered by EDFS for the optimal relay sensor selection. With the EDFS, the energy-efficiency and reliability of WBAN data transmission can be improved. In addition, our simulation results demonstrate that the proposed EDFS can effectively deal with the frequently changing WBAN topologies while providing balanced energy consumption and energy efficiency. INDEX TERMS Wireless body area network, data forwarding, energy efficiency, collaborative forwarding.

I. INTRODUCTION

As a special type of Wireless Sensor Networks (WSNs) [1] in biochemical and healthcare related fields, Wireless Body Area Networks (WBANs) have gained considerable attention from academia, industry and governments [2], [3]. A WBAN consists of body sensors with processing and monitoring capabilities, which are wearable or implantable and can collect physiological data such as temperature, pulse, blood oxygen levels, blood pressure, electrocardiogram (ECG), electroencephalogram (EEG), etc [4]. Sensors in WBANs are extremely energy and resource limited and the replacement of implanted sensors is especially costly and complex. Therefore, how to improve the energy efficiency and how to prolong the lifetime of WBANs are the primary challenges of researches and implementations. In terms of network size, WBANs are generally small and their energy consumption on data forwarding is mainly

VOLUME 4, 2016

determined by the volume of transmitted data and the selection of the relay sensors. According to existing work on sensor structures and energy consumption models, data transmission contributes a major part of sensor energy consumption [5]. Therefore, effectively reducing the size of data to be transmitted can directly improve sensor energy efficiency. In real-world health monitoring applications using WBANs, the shadow effect caused by body movements leads to frequently changing wireless link status and network topology, posing challenges on continuous and consistent monitoring of human physiological parameters. As a result, the traditional routing mechanisms for fixed network topologies are infeasible. The work in [6] and [7] showed that multi-hop data transmission through relay nodes is effective in dynamical WBANs implementation scenarios, and can greatly enhance the robustness of inter-node communication. In multi-hop WBANs, the design of routing mechanisms still largely

2169-3536 2016 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

7251

D. Wu et al.: EDFS for Heterogeneous WBANs

remains unexplored. Besides, relay sensors in WBANs can collaboratively perform forwarding [18] operations for other body sensors to achieve successful data transmission. Therefore, the selection of relay sensors [19] can affect the energy consumption rates of body sensors and the according network connectivity [20]. Some of our previous works [8]–[10] are related to WBANs energy consumption and applications. In additions, the data from WBANs can be offloaded to cloud [13], [16], [17] for the further processing in order to save energy. Exploiting the sparsity of sampled signals, Compressed Sensing (CS) [21] theory can compress the original data signal in a lower dimensional space, obtain the discrete sample and recover the original signal from the compressed signal with nonlinear algorithms [22]. Numerous researches indicate that data signals collected by WBANs sensors are sparse [23]. Thus, exploiting CS technology in WBANs can reduce the data size. The work in [24]–[27] employed CS technology for WBANs to exploit the data sparsity and conduct physiological data collection and action recognition. Besides, the computational complexity of CS technology at source sensors is relatively low, which is in line with the restrained computational capability of sensors in the WBANs. Furthermore, the data recovery algorithm is relatively complex, but performed by the sink with abundant resources. Although the above mentioned algorithms successfully reduce the size of transmitted data, the data transmission in actual WBANs applications can be achieved in a collaborative manner [28] with energy cost of relay sensors. In collaborative data transmission, the energy consumption of relay sensors is remarkably high, which inevitably causes unbalanced sensor energy consumption. Some research works related to transmission strategies are reported in [11], [12], [14], and [15]. Due to the heterogeneity of body sensors [29], their energy consumption rates also vary, and as a result, simply employing the initial and residual energy cannot accurately depict the sensor heterogeneity. Besides, the importance degrees of physiological data collected by different body sensors vary hugely, and the residual energy of important sensors has to remain high enough for emergency cases. Based on above analysis, relay sensors have to be properly selected for the data forwarding process of WBANs so that the network energy consumption can be balanced to maximize the network lifetime. However, the body movements are difficult to accurately predict and the positions and relative distance of body sensors are dynamically changing. The changing positions of relay sensors, neighbor sensor detection failure and dynamically changing network topologies can result in data packet loss and interrupted communication. Due to the dynamical features of WBANs, data forwarding protocols should be developed to support frequently changing WBANs topologies and the design of these routing protocols is crucial to the network performance. Targeting at the aforementioned problems, an Energyefficient Data Forwarding Strategy (EDFS) is proposed 7252

in this paper. Once the original physiological data are obtained, WBANs sensors employ the Discrete Cosine Transform (DCT) [30] and sparse binary random measurement matrix [31] to process the original data for data size reduction. Based on the sampling frequencies, the energy consumption rates of body sensors can be determined. During the data forwarding process, the energy consumption rates of body sensors, importance degrees of collected data and proportion of residual energy are jointly considered to select the optimal relay sensor, by which the multi-hop data transmission can be achieved with balanced sensor energy consumption and prolonged network lifetime. Besides, the residual energy thresholds are set based on the importance degrees of body sensors, which prioritizes the transmission of important physiological data. Through effectively dealing with the impacts of body movements on the link status, the multi-hop EDFS can provide enhanced data delivery ratio with reduced energy consumption, according to the simulation results. The major contributions of this paper can be summarized as follows. • The CS technology is employed in this paper to preprocess the collected physiological data to reduce data size and energy consumption. With the sparse binary random measurement matrix and DCT, the matrix multiplication can be transformed into matrix addition and rapid running speed can be achieved. • The impact factors of the sampling frequency, residual energy and sensor importance degree are jointly considered for the optimal relay sensor selection to balance the sensor energy consumption and to prolong the network lifetime. • The EDFS is proposed in this paper to collaboratively exploit the network energy resource, and effectively avoid the energy depletion of important sensors, which makes it robust to the dynamically changing topologies of WBANs applications. The remainder of the paper is organized as follows. The system model is given in Section II. The compressed sensing of original data is introduced in Section III. The relay sensor selection strategy is designed in Section IV. Numerical results are presented in Section V, and Section VI concludes the paper. II. SYSTEM MODEL

Generally, star single-hop and tree multi-hop network topologies are employed in WBANs. In a single-hop network, the sink receives the physiological data collected by body sensors, forwards them to the remote control center and disseminates the feedback message from the remote control center to body sensors, where the two-way data transmissions are both achieved in the single-hop manner. The star topology is commonly employed to provide real-time and rapid data gathering due to the WBANs characteristics of small network size and close inter-node distance. However, low-complexity star topology can incur increased transmission power of body sensors, extra energy consumption and unhealthful heat VOLUME 4, 2016


emission, due to the accuracy and packet loss requirements [32]. In view of the shortcomings of star topology, multi-hop tree topology gradually gains the research attention, where body sensors closer to the sink forward the data packets for farther body sensors. Consequently, the transmission power of these remote body sensors can be notably reduced, the network energy consumption can be effectively balanced, and reliable data transmission can be achieved.

in WBANs, therefore the energy consumption of receiving, processing and transmitting n bits of data can be calculated as Etotal = n × (2Ee + Eamp d α ) + Ep

(3)

III. COMPRESSED SENSING OF COLLECTED DATA

The energy resource and processing capability of WBANs sensors are both constrained while the real-time monitoring of human bodies constantly generates a large volume of data. Thus, transmitting all original physiological data would cause high energy consumption and network congestions. To overcome this issue, collected physiological data are firstly compressed at the body sensors by CS. By exploiting sampling frequencies much lower than the Nyquist frequency [33]–[35], the original physiological data can be represented by the sparse vectors to reduce the size of data to be transmitted and hence energy consumption. The received sparse vectors can be used to accurately recover the original data by the sink. The CS technology adopted in this paper can represent the collected physiological data x = [x1 , x2 , . . . , xN ] with the basis vector ψi , (i = 1, 2, . . . , N ), as shown in Eq. (4), where a = [a1 , a2 , . . . , aN ]T is the coefficient vector, 9 is the sparse basis. Especially, if ||a||0 = k, x has sparsity k in basis 9. x TN ×1 =

N X

ai ψi = 9 N ×N a

(4)

i=1 FIGURE 1. Topology of WBANs.

The design of an energy efficient data forwarding mechanism is crucial for WBANs. To precisely model the actual WBANs of specific application targets, the system structure shown in Fig. 1 is employed to achieve the data transmission for WBANs. In this particular WBANs, multiple body sensors are responsible for collecting various types of data and forwarding data packets for other body sensors, whereas the sink receives these data packets. Obviously, the data packets are transmitted to the sink within several hops and remote body sensors have to properly select the relay for data forwarding. The energy consumption of WBANs sensors mainly consists of three parts as shown in Eq. (1), where Etotal denotes the total energy consumption, Er , Ep and Et denote the energy consumption for receiving, processing and transmitting data packets, respectively. Etotal = Et + Er + Ep

(1)

Let Ee be the energy consumption for transmitting each data bit, Eamp the energy consumption of the amplifying circuit, n the data packet length, d the transmission distance and α the transmission power loss coefficient. Clearly, ( Er = n × Ee (2) Et = n × (Ee + Eamp d α ) The computational cost of employing CS, namely the data processing energy consumption, is non-negligible VOLUME 4, 2016

More specifically, measured by the measurement matrix 8 = [ϕ1 , ϕ2 , . . . , ϕN ], x can be compressed to a lower dimensional space to obtain the M dimensional measurement data, as shown in Eq. (5), which can be further transmitted in the network. yM ×1 = 8M ×N × xNT ×1 = 8 × 9 × a = 2 × a

(5)

The sparsity of original physiological data allows them to be processed by the CS technology. If a in Eq. (5) contains k(k N ) nonzero values, a has sparsity k. For any k-sparsity signal and constant δk ∈ (0, 1), if matrix 2 satisfies Eq. (6), namely the Restricted Isometry Property, we can accurately reconstruct k coefficients from M measurements. 1 − δk ≤

k2xk22 kxk22

≤ 1 + δk

(6)

When obtaining the original physiological data, the body sensor firstly measures the original data with the preset sparse binary random measurement matrix, where the design of the measurement matrix and deriving the sparsity of original data are two major challenges. Firstly, CS measurement matrix 8 must satisfy the Restricted Isometry Property, that is, 8 has to be independent from the sparse basis matrix 9. The Bernoulli, random Fourier, Gaussian random, Hadamard, and Toeplitz matrices are commonly employed by CS technology. However, to simplify algorithm design and reduce implementation difficulty, 7253


the sparse binary random measurement matrix is employed in this paper. The employed measurement matrix can effectively transform the matrix multiplication into matrix addition, which greatly reduces the system energy consumption. Secondly, the sparsity of original physiological data is the prerequisite of employing CS technology. DCT is employed in this paper as the sparse basis due to its rapid running speed and favorable sparse representation performance [36].

TABLE 1. Symbol notation table.

IV. SELECTION OF RELAY SENSORS

According to the established system model, the data packets are transmitted to the sink within several hops [37]. Therefore, the selection of relay sensors directly affects the energy consumption and performance of the data forwarding process. A. IMPACT OF SENSOR STRUCTURE HETEROGENEITY

WBANs sensors have various functions and are responsible for monitoring different physiological information such as temperature, pulse, blood oxygen levels, blood pressure, ECG, EEG, etc. Due to the variety of required precision and numerical range of different physiological data, energy consumption rate of WSN nodes in data sensing are also different. Therefore, the selection of relay sensors should take both the initial and residual energy levels into account. In WBANs, different body sensors monitoring different physiological information have various sampling frequencies. As shown in Table 1, the sampling frequencies of ECG, EEG and temperature are 250, 200 and 0.1 Hz respectively. Different sampling frequencies mean different energy consumption rates. Therefore, body sensors with high sampling frequencies are not suitable for forwarding operations due to their fast energy consumption rates. Obviously, the forwarding capability of a given body sensor is inversely proportional to its sampling frequency. Besides, body sensors with more residual energy can provide more forwarding services, whereas those with less residual energy must reduce the forwarding load to extend their lifetime. The energy consumption rates should be considered along with the sensor residual energy level in selecting the optimal relay sensor. The impact of different sampling frequencies on relay sensor selection is fully considered in this paper. Its impact factor γi is defined as the inverse of the sampling frequency of sensor i, as shown in Eq. (7), where fi is the sampling frequency of sensor i. When the sampling frequency of a given sensor is high, the value of impact factor γi is small and the probability of selecting this sensor as the relay is also small. 1/fi (7) γi = N P 1/fi i=1

In traditional WSNs, the selection of relay sensors is commonly based on the level of sensor residual energy. However, the sensors in WBANs have various energy consumption 7254

rates due to the heterogeneous sensor structures and simply considering the sensor residual energy level cannot accurately select the optimal relay. To address this issue, in this work, residual energy is considered as a factor to balance the energy consumption of each body sensor in the network. When the residual energy reaches the threshold, it will not be used as relay body sensor and the energy attenuation becomes slow. In particular, the cost function of residual energy for traditional WSNs is customized for WBANs to obtain the impact factor of residual energy on relay sensor selection, as shown in Eq. (8), where Erei and Eoi are the residual and initial energy of sensor i. ρi =

Erei Eoi

(8)

During each round of data collecting, the value of impact factor ρi is updated before calculating the residual energy as i Erei = Eoi − Etotal

(9) VOLUME 4, 2016


FIGURE 2. Typical distribution of application-oriented nodes in WBANs.

B. IMPACT OF SENSOR FUNCTION HETEROGENEITY

WBAN is highly application-oriented and consists of various types of multifunctional physiological sensors, which are responsible for monitoring data such as the temperature, pulse, blood oxygen levels, blood pressure, ECG, EEG, etc. As shown in Fig. 2, the types of monitored physiological data vary and their importance degrees are also different. For instance, blood oxygen levels and blood pressure only reflect the health level of a given person, whereas ECG and EEG are the vital signs. Therefore, the energy of body sensors monitoring more important physiological data should be preserved. Considering the application oriented characteristics of WBANs, the impact factor of sensor importance degrees on relay sensor selection can be defined such that N X

ηi = 1

(10)

i=1

According to the application requirements, ηi is assigned for each body sensor and a larger ηi signifies a smaller sensor importance degree and hence a larger probability of being selected as the relay. Especially, ηi of the implanted body sensor equals to zero. To avoid the energy depletion of body sensors with little residual energy, the minimum residual energy thresholds of Eθ of body sensors are set according to their importance degrees. When the residual energy of a given sensor is less than the threshold, it would stop the forwarding services for other sensors to reduce its energy consumption. C. SELECTION OF RELAY SENSORS

In WBANs, the energy consumption rates of sensors are always different. For instance, sensors monitoring ECG signals clearly consume more energy than sensors monitoring blood oxygen level signals. Therefore, the energy of ECG sensors should be preserved and a higher priority should be set to reduce its forwarding load and the corresponding energy consumption. In this paper, we determine the impact factors γi , ρi and ηi and residual energy threshold Eθ according to the specific application requirements. Meanwhile, we assume that the sink knows the sampling frequencies of sensors in the given WBANs. VOLUME 4, 2016

During the initial stage, the sink broadcasts a network initialization message to allocate the impact factor γi , ρi and ηi and residual energy threshold Eθ for each body sensor. Besides, each body sensor can calculate its own distance to the sink according to the location information. In particular, the location information refers to the body sensor’s position on the human body and can be acquired camera or manual measurement after the deployment. During the data collecting phase, after accomplishing the local processing on the collected data, sensor i broadcasts a forwarding request message into the WBANs, which includes the ID and location information of sensor i. Body sensors receiving this message calculate the values of their own residual energy Erei according to Eq. (9), and compare it with its Eθ . If the residual energy of sensor j is less than the preset threshold, namely Erej ≤ Eθ j , sensor j ignores the forwarding request by sensor i. otherwise, sensor j should accept the forwarding request and forward the physiological data for sensor i. The set of body sensors that have accepted the forwarding request of sensor i can be denoted by Fi = 1, 2, ...N , and these sensors can calculate their decision values Decj according to their sampling frequency impact factor γj , importance degree impact factor ηj and residual energy impact factor ρj , as shown in Eq. (11). The calculated Decj can be utilized to reasonably select the relay sensor for sensor i. Decj = γj × ηj × ρj

(11)

Body sensors within Fi then reply to sensor i with feedback messages, which include their IDs, location information and Decj . After receiving these replies, sensor i compares the distance between the sink and sensors in Fi with the distance between the sink and sensor i to select the relay sensor. If the distances between sensor i, j and the sink meet Eq. (12), namely sensor j is between sensor i and the sink, sensor j can forward data for sensor i. In Eq. (12), dis and djs denote the distances from sensor i and j to the sink respectively. Body sensors satisfying Eq. (12) 0 form the set of relay candidates Fi . ( dij < djs (12) djs < dis Eventually, the relay candidate with the maximum Dec 0 0 value in Fi = 1, 2, . . . , N is selected by sensor i as the 0 optimal relay sensor. If Fi is empty, sensor i directly communicates with the sink. With the proposed EDFS, the energy resource of WBAN sensors can be efficiently utilized, the network energy consumption can be accordingly balanced and the network lifetime can also be maximized. Besides, body movements can lead to changed positions and relative distance of body sensors, namely a dynamical network topology. Routing protocols targeting at fixed topologies can result in data packet loss and increased network resource consumption. 7255


With the dynamical relay sensor selection strategy, the proposed EDFS can effectively handle the frequently changing WBANs topologies. V. NUMERICAL RESULTS

In view of the WBANs characteristics, the optimal relay sensor can be selected to assist in data forwarding. Multihop data transmission can effectively reduce the energy consumption and prolong the network lifetime. To evaluate the performance of the proposed EDFS, a MATLAB simulation platform is employed in this paper. TABLE 2. Frequencies of heterogeneous nodes.

1) SIMULATION ENVIRONMENT

The network topology of the simulated WBANs is shown in Fig. 3, where grey dots are the wearable or implantable sensors on the human body and the black dot is the sink. The body sensors are responsible for collecting various physiological data, performing the local CS operations, and reporting the data to the sink. Obviously, the body sensors far away from the sink have to select the optimal relay sensor to achieve the data transmission, and a few transmission paths are denoted by solid lines in Fig. 3. The link status and network topology are constantly changing, and the network topologies of Posture lying (1), standing (2), sitting (3) and walking (4) are shown in Fig. 3, where the locations of body sensors are given in Table 2, as shown in Fig. 1. For instance, sensors 9 and 10 on both thighs are relatively closer to the sink than sensors 5 and 6 on both elbows in Posture 2 and 3. Therefore, sensors 9 and 10 are more likely to be selected as relays. Besides, the importance degree of sensor 2 is much higher than sensors 3 and 4 on both shoulders, while their distances to the sink are almost the same. Therefore, sensors 3 and 4 are more likely to be selected as relays. By comparing the network topologies of Posture 1 and 2, the sampling frequency impact, residual energy impact factor and importance degree impact factor are jointly considered to replace relay sensors 3 and 4 with sensors 5 and 6, by which energy efficient data forwarding can be achieved. As shown in Posture 4, the relative positions of body sensors during walking are continuously changing. For instance, body sensor 7 on the right palm may be in the coverage of sensor 9 and in the coverage of sensor 10, due to the body movements, thus, sensor 7’s communication connection with other nodes is intermittent, and it is necessary to send communication request packages or join request messages to neighbour nodes 7256

FIGURE 3. Network architecture.

at the initialization process. By sending the join request message, body sensor 7 can determine its relay sensor to establish a transmission path. VOLUME 4, 2016


Various multifunctional physiological sensors are deployed in WBANs for different healthcare purposes. For different applications, the data rate and corresponding sampling frequency of certain sensors vary hugely. The parameters of the body sensors adopted in this paper are given in Table 3. For instance, ECG and EEG sensors have relatively higher sampling frequencies and initial energy levels and more importance degrees compared with temperature and blood pressure sensors, because blood oxygen levels and blood pressure only reflect the health of a given person, whereas ECG and EEG are the vital signs. The simulation parameters are given in Table 4. TABLE 3. Locations of body sensors.

TABLE 5. Simulation parameters.

The energy consumption of relay sensors can be analyzed as follows. Assuming there are H sensors in the network area of A × B, the neighbor number of any relay sensor is NRB = π R2 /(AB/H ). Then the receiving and transmitting energy consumption of the relay sensor can be calculated by Z R Ert 0 = 2MEe + M Eamp αx α−1 dx/2 (15) 0

Furthermore, the total energy consumption of the relay sensor can be calculated by Z R Em 0 = ABH [2MEe + M Eamp αx α−1 dx/2]/(2πR2 ) 0

(16)

TABLE 4. Parameters of heterogeneous nodes.

All original physiological data are processed by the CS technology and transmitted to the sink for further data reconstruction. Because the sink is always resource-unrestrained, the energy consumption analysis only needs to consider the data transmitting and processing energy consumption of body sensors. Thus, the total network energy consumption can be calculated as 0 Etotal = Ep 0 + Et 0 + Em 0

(17)

3) DATA RECONSTRUCTION ACCURACY ANALYSIS

2) ENERGY CONSUMPTION ANALYSIS

To evaluate the energy efficiency of the proposed EDFS, the energy consumption of the multi-hop data transmission between body sensors and the sink is analyzed in this paper. The energy consumed by the CS measurement process can be calculated as shown in Eq. (13), where Ec is the CS energy consumption per data bit. √ (13) Ep 0 = 2N × MN × Ec + MNEc Employing CS technology, we reduce the size of transmitted data from N dimentions to M dimentions. The transmission energy consumption can be calculated by Z R 0 Et = MEe + M Eamp αx α−1 dx/2 (14) 0 VOLUME 4, 2016

WBANs sensors employ CS technology to compress the original physiological data to a lower dimensional space and approximately recover the original data signal from the compressed signal. To validate the efficacy of the proposed EDFS, Root Mean Square Error (RMSE) serves as the performance d is the reconparameter, as defined in Eq. (18), where y(t) structed data, y(t) is the original data and N is the number of samples. v u N u1 X d − y(t))2 (y(t) (18) RMSE = t N n=1

The reconstruction performance and residual error are shown in Fig. 4, which indicates that the proposed EDFS can provide favorable reconstruction results. The data errors are between -0.1 and 0.1, and the reconstruction RMSE is 8.803 × 10−3 . Besides, the CS measurement and reconstruction operations on ECG data 100.dat from the MIT/BIH database are shown in Fig. 5. The data error range is 1 × 10−2 , and the 7257


FIGURE 4. Reconstruction performance of random signal after CS.

consumption rates are equal and their residual energy trends are also same. The physiological data collected by sensors 1 and 2 require high sampling frequencies and consume more energy resources. Therefore, the lifetime of sensor 1 or 2 is much shorter 40%-60% than that of sensor 3 or 4 and the network energy consumption is unbalanced. Obviously, the simple star network topology cannot effectively utilize the overall network energy resource and preserve the energy of important sensors. With the proposed EDFS, the multi-hop data transmission can solve this problem and simulation results are given in Fig. 7.

FIGURE 5. Reconstruction performance of 100.dat after CS.

RMSE is 0.013. Therefore, the proposed EDFS can effectively process the physiological data collected by WBANs, while providing high reconstruction accuracy. 4) ENERGY EFFICIENCY ANALYSIS

We used the simulation settings described in Table 5 for energy consumption analysis. For the energy consumption analysis of body sensors under 4 different Postures, we select body sensors 1, 2, 3 and 4 and observe their energy consumption trends. Due to the relatively small network size, WBANs always employ star network topology and body sensors can directly communicate with the sink. Firstly, the energy consumption trends of 4 selected body sensors in the star network topology are shown in Fig. 6. Obviously, in the star network topology, body sensors have constant energy consumption rates and the energy consumption rates of sensors vary hugely. Because sensors 3 and 4 are of the same type, their energy

FIGURE 7. Energy Consumption of the Network Topology With EDFS and Without EDFS in Each Postures.

FIGURE 6. Energy Consumption of Star Network Topology in Each Posture. 7258

The multi-hop data transmission exploits the collaboration between body sensors to improve the network energy efficiency. When employing EDFS, the energy consumption rates of sensors are constantly changing, because body sensors may provide forwarding services while collecting and reporting their own physiological data. For instance, sensor 4 is the closest to the sink and its initial energy consumption rate is high due to the forwarding functions. When its residual energy reaches the threshold value, sensor 4 stops VOLUME 4, 2016


providing forwarding services, thus, its energy consumption rate becomes low. In terms of the sensor importance, sensor 2 is responsible for ECG data collecting which is a relatively more important vital sign. Therefore, the probability of using sensor 2 for forwarding is low. However, in the beginning stage, the physiological data collected by sensor 2 can be forwarded by sensor 4 and the energy consumption rate of sensor 2 is low. As sensor 4 stops providing forwarding services, sensor 2 has to directly communicate with the sink and its energy consumption rate will increase. With the proposed EDFS, the sensor energy resources can be collaboratively utilized while the energy depletion of important sensors can be effectively avoided. We perform the EDFS algorithm, and the relativity between energy consumption and network running time is illustrated in Fig.7. We can observe that, the residual energy gradually approaches zero with the increase of the number of rounds when node ID is either 1 or 3. As shown in Fig. 7, the energy consumption rates of sensor 1 collecting the EEG signal are similar with different body Postures. Due to the uniform relay sensor selecting strategy of EDFS, sensor 3 collecting the ECG signal also has the similar energy consumption rate under different body Postures. Even for body sensors with the same initial energy level, their energy consumption rates vary. For instance, under Posture 1, the energy consumption rates of sensors 1 and 3 are notably different due to the various sampling frequencies of EEG and Blood Oxygen signals. Besides, the energy consumption rates also can be affected by the change of Postures. As shown in the Posture 2, during the early stage, sensor 3 severs as relay, therefore its energy consumption rates is obviously higher than that of sensor 1. Because relay sensor 3 has to forward the received packets along with its own physiological data, their energy resource is depleted fast. At the later stage, when the residual energy of relay sensors reaches the preset threshold, they stop forwarding packets for other sensors. Therefore, their energy consumption rates immediately drop to the normal level and the corresponding energy consumption rates become smooth. Similarly, under other Postures, the corresponding energy consumption rates at the later stage also become smooth, because sensor 3 only transmits its own physiological data. By comparing the energy consumption rates of the proposed EDFS under various Postures with that of the scheme using star network topology, the lifetime of body sensors collecting the EEG and ECG signals is prolonged by EDFS. Besides, the long lifetime of body sensors in the star network topology is reduced due to being selected as relays. Overall, the proposed EDFS can effectively extend the lifetime of important body sensors, balance the energy consumption rates among all sensors and prolong the network lifetime. Additionally, it is noted that the proposed EDFS algorithm is beneficial for preserving energy and then prolonging lifetime of nodes. The factors can be divided into two aspects. On one hand, the original data could be compressed during the procedure of EDFS. On the other hand, the relaying VOLUME 4, 2016

node can be selected by taking into consideration inherent attributes such as sample frequency or data rate in TABLE 1 of node. 5) TOPOLOGY VARIATION ANALYSIS

Due to the dynamical feature of WBAN, the network topology of wearable and implanted body sensors is constantly changing. To evaluate the performance of proposed EDFS, the average hop count is employed as the simulation parameter, namely the average value of the hop distances from the source nodes to the sink. The average hop count can effectively reflect the network energy consumption. Data packets record and update the IDs of its source node and relay nodes during the simulated data transmission process. The average hop counts of four common Postures, including lying (1), standing (2), sitting (3) and walking (4), are simulated and shown in Fig. 8. As can be seen, the average hop counts under four Postures vary; and a small node distance brings reduced average hop count with Posture 4, because the relay body sensors is increased and the frequent encounter of each sensor due to the human mobility. Besides, the average hop counts of Posture 1 and 2 are equal due to the identical network topology, and Posture 3 has a relatively shorter inter-node distance.

FIGURE 8. Average Hop Counts of Various Postures.

Furthermore, the average delay of four Postures are derived and shown in Fig. 9. Apparently, the delay of Posture 1 and 4 is notably higher, due to the shadow effect, increased path loss, affected link status of lying Posture and the constantly changing network topology of walking Posture. Without EDFS scheme, the relation between delay and body postures is similar to that of the above. It is clear that when EDFS is adopted the average delay of packets changes from 66ms to 30ms in terms of posture 1. The value has decreased by almost half which is nearly equivalent to data compression ratio. This observation implies that reducing the volume of data contributes to the decline of average delay. Obviously, the proposed EDFS can effectively deal with the changing 7259


FIGURE 9. Average Delays of Various Postures.

postures of WBAN applications, as suggested by the performances under Posture 1 to 4. VI. CONCLUSION

Due to the limited energy resource of WBAN sensors, their energy depletion will cause severe degradation of the network performance. The EDFS was proposed in this paper to balance sensor energy consumption, improve network lifetime and promote collaborative sensor operations of heterogeneous WBANs. Firstly, the original physiological data are processed by CS technology to reduce data size. Secondly, the initial energy, residual energy, sampling frequency and sensor importance are jointly designed to achieve optimal relay selection. Our studies show that high energy-efficiency and reliability of WBAN data transmission can be achieved by the proposed EDFS. Our future studies include the consideration of energy harvesting in the WBANs. REFERENCES [1] Y. Liu, Y. He, M. Li, J. Wang, K. Liu, and X. Li, ‘‘Does wireless sensor network scale? A measurement study on GreenOrbs,’’ IEEE Trans. Parallel Distrib. Syst., vol. 24, no. 10, pp. 1983–1993, Oct. 2013. [2] M. Chen, S. Gonzalez, A. Vasilakos, H. Cao, and V. C. M. Leung, ‘‘Body area networks: A survey,’’ Mobile Netw. Appl., vol. 16, no. 2, pp. 171–193, 2011. [3] S. Movassaghi, M. Abolhasan, J. Lipman, D. Smith, and A. Jamalipour, ‘‘Wireless body area networks: A survey,’’ IEEE Commun. Surveys Tuts., vol. 16, no. 3, pp. 1658–1686, Jan. 2014. [4] L. Wang et al., ‘‘A wireless biomedical signal interface system-on-chip for body sensor networks,’’ IEEE Trans. Biomed. Circuits Syst., vol. 4, no. 2, pp. 112–117, Apr. 2010. [5] J. Naganawa, K. Wangchuk, M. Kim, T. Aoyagi, and J.-I. Takada, ‘‘Simulation-based scenario-specific channel modeling for WBAN cooperative transmission schemes,’’ IEEE J. Biomed. Health Informat., vol. 19, no. 2, pp. 559–570, Mar. 2015. [6] A. Natarajan, B. de Silva, K.-K. Yap, and M. Motani, ‘‘To hop or not to hop: Network architecture for body sensor networks,’’ in Proc. 6th IEEE Conf. Sensor, Mesh Ad Hoc Commun. Netw. (SECON), Jun. 2009, pp. 1–9. [7] Y. Ge, L. Liang, W. Ni, A. A. P. Wai, and G. Feng, ‘‘A measurement study and implication for architecture design in wireless body area networks,’’ in Proc. IEEE Int. Conf. Pervas. Comput. Commun.(PERCOM) Workshops, Mar. 2012, pp. 799–804. 7260

[8] Z. Zhang, H. Wang, C. Wang, and H. Fang, ‘‘Interference mitigation for cyber-physical wireless body area network system using social networks,’’ IEEE Trans. Emerg. Topics Comput., vol. 1, no. 1, pp. 121–132, Jun. 2013. [9] H. Wang, D. Peng, W. Wang, H. Sharif, H. H. Chen, and A. Khoynezhad, ‘‘Resource-aware secure ecg healthcare monitoring through body sensor networks,’’ IEEE Wireless Commun., vol. 17, no. 1, pp. 12–19, Feb. 2010. [10] Z. Zhang, H. Wang, A. V. Vasilakos, and H. Fang, ‘‘ecg-cryptography and authentication in body area networks,’’ IEEE Trans. Inf. Technol. Biomed., vol. 16, no. 6, pp. 1070–1078, Nov. 2012. [11] J. Shen, H. Tan, J. Wang, J. Wang, and S. Lee, ‘‘A novel routing protocol providing good transmission reliability in underwater sensor networks,’’ J. Internet Technol., vol. 16, no. 1, pp. 171–178, 2015. [12] S. Xie and Y. Wang, ‘‘Construction of tree network with limited delivery latency in homogeneous wireless sensor networks,’’ Wireless Pers. Commun., vol. 78, no. 1, pp. 231–246, 2014. [13] Y. Ren, J. Shen, J. Wang, J. Han, and S. Lee, ‘‘Mutual verifiable provable data auditing in public cloud storage,’’ J. Internet Technol., vol. 16, no. 2, pp. 317–323, 2015. [14] Y. Zhang, X. Sun, and W. Baowei, ‘‘Efficient algorithm for k-barrier coverage based on integer linear programming,’’ China Commun., vol. 13, no. 7, pp. 16–23, Jul. 2016. [15] Z. Fu, K. Ren, J. Shu, X. Sun, and F. Huang, ‘‘Enabling personalized search over encrypted outsourced data with efficiency improvement,’’ IEEE Trans. Parallel Distrib. Syst., vol. 27, no. 9, pp. 2546–2559, Sep. 2016, doi: 10.1109/TPDS.2015.2506573 [16] Z. Xia, X. Wang, L. Zhang, Z. Qin, X. Sun, and K. Ren, ‘‘A privacypreserving and copy-deterrence content-based image retrieval scheme in cloud computing,’’ IEEE Trans. Inf. Forensics Security, vol. 11, no. 11, pp. 2594–2608, Nov. 2016, doi: 10.1109/TIFS.2016.2590944. [17] Z. Xia, X. Wang, X. Sun, and Q. Wang, ‘‘A secure and dynamic multikeyword ranked search scheme over encrypted cloud data,’’ IEEE Trans. Parallel Distrib. Syst., vol. 27, no. 2, pp. 340–352, Feb. 2016. [18] Q. Nadeem, N. Javaid, S. N. Mohammad, M. Y. Khan, S. Sarfraz, and M. Gull, ‘‘Simple: Stable increased-throughput multi-hop protocol for link efficiency in wireless body area networks,’’ in Proc. 8th Int. Conf. Broadband Wireless Comput. Commun. Appl. (BWCCA), Oct. 2013, pp. 221–226. [19] X. Liang et al., ‘‘Exploiting prediction to enable secure and reliable routing in wireless body area networks,’’ in Proc. IEEE INFOCOM, Mar. 2012, pp. 388–396. [20] M. F. Duarte and R. G. Baraniuk, ‘‘Spectral compressive sensing,’’ Appl. Comput. Harmon. Anal., vol. 35, no. 1, pp. 111–129, 2013. [21] D. L. Donoho, ‘‘Compressed sensing,’’ IEEE Trans. Inf. Theory, vol. 52, no. 4, pp. 1289–1306, Apr. 2006. [22] H. Mamaghanian, N. Khaled, D. Atienza, and P. Vandergheynst, ‘‘Compressed Sensing for Real-Time Energy-Efficient ECG Compression on Wireless Body Sensor Nodes,’’ IEEE Trans. Biomed. Eng., vol. 58, no. 9, pp. 2456–2466, Sep. 2011. [23] M. Balouchestani, K. Raahemifar, and S. Krishnan, ‘‘Wireless body area networks with compressed sensing theory,’’ in Proc. ICME Int. Conf. Complex Med. Eng. (CME), Jul. 2012 pp. 364–369. [24] L. Feng et al., ‘‘Golden-angle radial sparse parallel MRI: Combination of compressed sensing, parallel imaging, and golden-angle radial sampling for fast and flexible dynamic volumetric MRI,’’ Magn. Reson. Med., vol. 72, no. 3, pp. 707–717, 2014. [25] D. Gangopadhyay, E. G. Allstot, A. M. R. Dixon, K. Natarajan, S. Gupta, and D. J. Allstot, ‘‘Compressed sensing analog front-end for bio-sensor applications,’’ IEEE J. Solid-State Circuits, vol. 49, no. 2, pp. 426–438, Feb. 2014. [26] F. Chen, F. Lim, O. Abari, A. Chandrakasan, and V. Stojanovic, ‘‘Energyaware design of compressed sensing systems for wireless sensors under performance and reliability constraints,’’ IEEE Trans. Circuits Syst. I, Reg. Papers, vol. 60, no. 3, pp. 650–661, Mar. 2013. [27] S. Yang, J. L. Lu, F. Yang, L. Kong, W. Shu, and M. Y. Wu, ‘‘Poster: Behavior-aware probabilistic routing for wireless body area sensor networks,’’ in Proc. IEEE GLOBECOM, Apr. 2013, pp. 444–449. [28] L. Liang, Y. Ge, G. Feng, W. Ni, and A. A. P. Wai, ‘‘A low overhead treebased energy-efficient routing scheme for multi-hop wireless body area networks,’’ Comput. Netw., vol. 70, no. 18, pp. 45–58, 2014. [29] X. Zhang, H. Wu, and Y. Ma, ‘‘A new auto-focus measure based on medium frequency discrete cosine transform filtering and discrete cosine transform,’’ Appl. Comput. Harmon. Anal., vol. 40, no. 2, pp. 430–437, 2016. VOLUME 4, 2016


[30] Z. Zhang, T.-P. Jung, S. Makeig, and B. D. Rao, ‘‘Compressed sensing of eeg for wireless telemonitoring with low energy consumption and inexpensive hardware,’’ IEEE Trans. Biomed. Eng., vol. 60, no. 1, pp. 221–224, Jan. 2013. [31] E. J. Candès and M. B. Wakin, ‘‘An introduction to compressive sampling,’’ IEEE Signal Process. Mag., vol. 25, no. 2, pp. 21–30, Mar. 2008. [32] C. S. Lin and P. J. Chuang, ‘‘Energy-efficient two-hop extension protocol for wireless body area networks,’’ IET Wireless Sensor Systems, vol. 3, no. 1, pp. 37–56, Mar. 2013. [33] E. J. Candès, J. Romberg, and T. Tao, ‘‘Robust uncertainty principles: Exact signal reconstruction from highly incomplete frequency information,’’ IEEE Trans. Inf. Theory, vol. 52, no. 2, pp. 489–509, Feb. 2006. [34] E. J. Candes and T. Tao, ‘‘Near-optimal signal recovery from random projections: Universal encoding strategies?’’ IEEE Trans. Inf. Theory, vol. 52, no. 12, pp. 5406–5425, Dec. 2006. [35] J. Zou, A. Gilbert, M. Strauss, and I. Daubechies, ‘‘Theoretical and experimental analysis of a randomized algorithm for sparse Fouriertransform analysis,’’ J. Comput. Phys., vol. 211, no. 2, pp. 572–595, 2006. [36] J. Zhang et al., ‘‘Iterative Correction Scheme Based on Discrete Cosine Transform and L1 Regularization for Fluorescence Molecular Tomography With Background Fluorescence,’’ IEEE Trans. Biomed. Eng., vol. 63, no. 6, pp. 1107–1115, Jun. 2016. [37] D. Wu, J. He, H. Wang, C. Wang, and R. Wang, ‘‘A hierarchical packet forwarding mechanism for energy harvesting wireless sensor networks,’’ IEEE Commun. Mag., vol. 53, no. 8, pp. 92–98, Aug. 2015.

DAPENG WU received the M.S. degree in communication and information systems from the Chongqing University of Posts and Telecommunications in 2006, and the Ph.D. degree from the Beijing University of Posts and Telecommunications in 2009. His research interests include ubiquitous networks, IP QoS architecture, network reliability, and performance evaluation in communication systems.

BORAN YANG received the B.S. degree in communication engineering from the Chongqing University of Posts and Telecommunications, Chongqing, China, in 2013, where he is currently pursuing the master’s degree in information and communication engineering. His research interests include opportunistic networks, buffer management, and trust management.

VOLUME 4, 2016

HONGGANG WANG received the Ph.D. degree in computer engineering from the University of Nebraska–Lincoln in 2009. He was a Technical Staff Member with Nokia Bell Labs and Lucent Technologies, China, from 2001 to 2004. He is currently an Associate Professor with the University of Massachusetts (UMass) Dartmouth, and is an affiliated Faculty Member with the Advanced Telecommunications Engineering Laboratory, University of Nebraska–Lincoln. He is also a Faculty Member of the Biomedical Engineering and Biotechnology Ph.D. Program, UMass Dartmouth. His research interests include wireless health, body area networks, cyber security, mobile multimedia and cloud, wireless networks and cyber-physical systems, and big data in mHealth.

DALEI WU received the B.S. and M.Eng. degrees in electrical engineering from Shandong University, Jinan, China, in 2001 and 2004, respectively, and the Ph.D. degree in computer engineering from the University of Nebraska– Lincoln, USA, in 2010. He was a Post-Doctoral Research Associate with the Mechatronics Research Laboratory, Department of Mechanical Engineering, Massachusetts Institute of Technology, Cambridge, USA. He joined The University of Tennessee at Chattanooga in 2014. His research interests include intelligent systems, cyber-physical systems, and complex dynamic system modeling and optimization.

RUYAN WANG received the M.S. degree from the Chongqing University of Posts and Telecommunications (CQUPT), China, in 1997, and the Ph.D. degree from the University of Electronic Science and Technology of China in 2007. Since 2002, he has been a Professor with the Special Research Centre for Optical Internet and Wireless Information Networks, CQUPT. His research interests include network performance analysis and multimedia information processing.

7261