Content uploaded by Zhiqiang Liu
Author content
All content in this area was uploaded by Zhiqiang Liu on Dec 12, 2017
Content may be subject to copyright.
SPECIAL SECTION ON THE NEW ERA OF SMART CITIES:
SENSORS, COMMUNICATION TECHNOLOGIES AND APPLICATIONS
Received June 22, 2017, accepted September 23, 2017, date of publication October 2, 2017, date of current version October 25, 2017.
Digital Object Identifier 10.1109/ACCESS.2017.2758348
Buffer-Aware Resource Allocation Scheme With
Energy Efficiency and QoS Effectiveness in
Wireless Body Area Networks
ZHIQIANG LIU1, (Student Member, IEEE), BIN LIU 2, (Member, IEEE),
AND CHANG WEN CHEN3, (Fellow, IEEE)
1Department of Electrical Engineering and Information Science, University of Science and Technology of China, Hefei 230027, China
2School of Information and Technology, University of Science and Technology of China, Hefei 230027, China
3Department of Computer Science and Engineering, University at Buffalo, State University of New York, Buffalo, NY 002837 USA
Corresponding author: Bin Liu (flowice@ustc.edu.cn)
This work was supported by the National Natural Science Foundation of China under Grant 61202406.
ABSTRACT Wireless body area network (WBAN) has attracted more and more attention to automatically
and intelligently sense the health data of one person for supporting various health applications in smart cities.
In the energy-constrained and heterogeneous WBAN system, there are three main issues: 1) the dynamic link
characteristics due to the time-varying postures and environments; 2) the high energy efficiency requirement
with considering the limited sensor battery; and 3) the high quality-of-service (QoS) requirement due to the
importance of health data. To provide long service with high quality, the resource allocation scheme becomes
indispensable with considering all these issues. In this paper, a mix-cost parameter is designed to evaluate
the energy efficiency and QoS effectiveness, and a resource allocation problem is formulated to minimize the
total mix-cost with optimizing the transmission rate, the transmission power, and the allocated time slots for
each sensor. Then, a buffer-aware sensor evaluation method with low complexity is introduced to the resource
allocation scheme to evaluate the sensor state in real time and then decide when applying for the resource re-
allocation by the hub for further improving both the short-term and the long-term QoS performance. Finally,
a greedy sub-optimal resource allocation scheme is designed to reduce the time complexity of the resource
allocation scheme. Simulation results are presented to demonstrate the effectiveness of the proposed optimal
buffer-aware resource allocation scheme as well as the greedy sub-optimal resource allocation scheme with
low complexity.
INDEX TERMS Wireless body area network (WBAN), resource allocation scheme, buffer-aware sensor
state evaluation method, energy efficiency, QoS effectiveness.
I. INTRODUCTION
With the increase of the city population, the idea of the
smart city has been proposed to adopt the digital technolo-
gies to enhance the quality and performance of smart ser-
vices, such as the smart healthcare and the smart trans-
portation [1]. As for the smart healthcare, how to improve
the efficiency of smart healthcare systems is one of the
most challenging goals in Smart Cities [2]. To satisfy
the increasing healthcare applications, wireless body area
network (WBAN), as an important wireless networking
technology, has attracted more and more attention in both
healthcare community and engineering industry [3], [4].
Different from the traditional complex and wired healthcare
devices, the body sensors in the WBAN system are able
to continuously monitor the body’s vital signals. A classic
WBAN mainly consists of one hub and several body sensors.
And the hub usually has rich resources, such as energy,
processing and storage buffer, while the body sensors are
energy limited due to the small size. These heterogeneous
body sensors are used to monitor different health attributes.
The physiological data streams are collected by these body
sensors and transmitted to the hub via wireless channels.
Then, the hub can use the existing wireless technologies, such
as the Wifi, 4G technology and so on, to transmit these data to
the medical server of the smart cities, as seen in Fig. 1. These
collected data are gathered when the patients’ activities are in
normal and emergency situations, and they can help doctors
better analysis the health conditions [5]. However, there are
VOLUME 5, 2017
2169-3536 2017 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.
20763
Z. Liu et al.: Buffer-Aware RA Scheme With Energy Efficiency and QoS Effectiveness in WBANs
FIGURE 1. A classical WBAN architecture.
still several issues needed to be solved in the development of
the WBAN [6].
Firstly, the links between the hub and the sensors in the
WBAN have the dynamic characteristics due to the pos-
ture and environment variations [7], [8]. The channel fading
between the body sensors and the hub relates to not only
the distance but also a number of factors such as clothing,
obstructions due to different body segments and so on [9].
When the body posture changes or the environment changes,
the link quality may inevitably change. Therefore, the WBAN
system has to deal with such dynamic link quality. Secondly,
due to the requirement of lightweight, the wireless body
sensors generally have a tiny size and then the resources
such as processing, storage buffer and battery energy supply
are extremely limited compared with other ordinary wire-
less sensors [6]. Finally, the vital physiological data streams
collected by the body sensors should be transmitted reliably
from humans to the hub, and a loss or an excessive delay
of these vital signals may cause a fatal accident [3]. For
example, the heart activity readings, e.g., ECG signals, should
be continuously monitored to detect whether there are some
heart attacks. Once a heart attack is detected, a warning
signal needs to be sent to the medical server for timely assis-
tance. Thus the high quality of service (QoS) metrics, which
includes the packet loss rate (PLR), the throughput and the
delay, should be guaranteed to better support the healthcare
applications when the WBAN system is designed.
To improve the WBAN performance, many strategies have
been widely studied in the literature [6], [10]–[12]. Among
them, the transmission power control (TPC) scheme as a
classic approach has been well studied. Generally, the TPC
schemes are designed to adapt the transmission power to
the dynamic link quality based on the channel estimation,
and thus the short-term QoS performances can be better
improved [13]–[16]. However, the short-term QoS perfor-
mance, which depends on the accuracy of the channel esti-
mation, is difficult to be guaranteed while considering the
highly dynamic link characteristics. In addition, only the
transmission power of each sensor can be adjusted in TPC
schemes to improve the WBAN performance, which lim-
its the effectiveness of TPC schemes. Compared with the
TPC schemes, the resource allocation (RA) schemes can
try to adjust more kinds of resources, such as the transmis-
sion rates, the transmission power, the allocated time slots
channel and so on, for further enhancing the system perfor-
mances [6], [11], [12], [17]. In the traditional resource allo-
cation schemes, the long-term QoS requirements are regarded
as the constraints of the optimization problem, and the
resource allocation strategies for the sensors can be obtained
by solving the optimization problem. However, the long-
term QoS performance cannot always be guaranteed with the
dynamic link characteristics. Considering the importance of
the vital signals, we must try best to avoid some loss or an
excessive delay of these vital signals. Therefore, not only the
average QoS performance in long term should be ensured, but
also the short-term QoS performance should be improved to
provide better service for healthcare applications in WBANs.
In this paper, with a buffer-aware sensor state evalua-
tion method, a buffer-aware resource allocation scheme is
designed to improve the energy efficiency and both the short-
term and the long-term QoS performance. Some preliminary
results have been reported in [18] and [19], and here we
give more technical details and the adequate explanation of
the methodology. The key contributions of this paper are in
three-fold:
Firstly, a mix-cost parameter is designed for each sen-
sor to jointly measure both the energy efficiency and QoS
effectiveness, and then a resource allocation scheme is pro-
posed to optimize the allocation of the transmission power,
the transmission rates and the time slots for each sensor to
minimize the total mix-cost of all sensors. Secondly, a buffer-
aware sensor state evaluation method with low complexity
is designed and imported in the resource allocation scheme
to decide when applying for the resource re-allocation by
the hub. Based on the real-time buffer queue states, it is
executed by each sensor to further improve the short-term
QoS performances with only one more bit in the data frame.
Thirdly, a greedy sub-optimal resource allocation scheme
is proposed to reduce the time complexity of the resource
allocation scheme, while its performance is close to that of
the optimal resource allocation.
The remainder of this paper is organized as follows.
In Section II, we discuss the related work relevant
to this paper. The details of the system model are
presented in Section III. We describe the sensor state evalu-
ation method in Section IV. The design of mix-cost param-
eter is given in Section V. In Section VI, the formulation of
resource allocation problem and the sub-optimal resource
allocation scheme are described and solved. The simulation
results are given in Section VII, and in Section VIII, the con-
clusions are described.
II. RELATED WORKS
The resource allocation methodology has received significant
interest in recent years as a kind of methods to improve the
performance of the WBAN with the dynamic link charac-
teristics and the limited channel resource. The transmission
power control scheme as a simple resource allocation scheme
20764 VOLUME 5, 2017
Z. Liu et al.: Buffer-Aware RA Scheme With Energy Efficiency and QoS Effectiveness in WBANs
adjusts the transmission power of each sensor to improve the
performance of the WBAN system [9], [13]–[16]. To study
the impact of the transmission power in trading off the reli-
ability of time-varying wireless link for energy efficiency,
Xiao et al. [13] adjusted the transmission power when the
real-time mean value of the received signal strength indica-
tion (RSSI) was not in the given up and down thresholds,
which could be tuned to achieve the desired trade-off between
energy savings and reliability. For characterizing the dynamic
link, the dynamic body postures were introduced to the trans-
mission power control. The postural position was inferred
by the observed linear relationship between the transmission
power (TP) and the RSSI, whose parameters were different
between different postural positions [9]. By considering the
partial-periodicity characteristics of the WBAN channels,
the long-term history channel gain values were used by a
predictor to select the most similar part with the latest channel
gain values, then the channel gain after the most similar part
was used to estimate the next channel gain, which was further
mapped to the suitable power for the proper BAN operation
in [20]. In [14], not only the short-term but also the long-term
channel states were estimated to target the RSSI threshold
range, and then the power level was adapted to improve the
energy efficiency and the link reliability. However, according
to the empirical relationship between transmission power and
the RSSI or the partial-periodicity characteristics of WBAN
channels, the accuracy of the channel estimation cannot
always be guaranteed due to many factors, such as different
type of transmitters, high dynamic link characteristics and
the change of the environments. And these TPC schemes
only adjust the transmission power to adapt the dynamic link,
which results in the limited effectiveness.
As for the resource allocation schemes, more parame-
ters can be adjusted to better improve the WBAN perfor-
mances with considering the dynamic link characteristics.
In [12], the transmission power and the transmission rate
were optimized to guarantee the QoS requirements of the data
delivery. However, only the data streams from the hubs to
the base station rather than the nodes to the hub were con-
sidered. In addition, the packet size could also be optimized
to improve the energy efficiency for various communication
scenarios depending on the acknowledgement policy in [10].
To improve the energy efficiency of the WBAN system,
the throughput and the bit loss rate requirements were formu-
lated as the constraints of the resource allocation scheme, and
then the time slots and the transmission power were allocated
to minimize the global energy consumption [17]. However,
the path loss was assumed as a specific value in advance while
the hub broadcast the beacon to reallocate resources for each
sensor at the beginning of every superframe. The assumption
seems not suitable due to the dynamic link characteristics
in the WBAN. In [6], the statistical characteristics of the
channel link for different postures were introduced to obtain
the long-term QoS constraints of the optimal resource alloca-
tion scheme to improve the energy efficiency of the WBAN
system, in which the transmission rate of each sensor was also
adjusted to meet more strict PLR requirements or worse link
quality. However, the long-term QoS performances cannot
satisfy the requirements of the vital signals to monitor some
potential emergent events. In the traditional resource alloca-
tion schemes, the QoS requirements are seen more important
than the energy efficiency, thus they need to be guaranteed
before minimizing the energy consumption of the WBAN
system.
In this paper, both the short-term and long-term QoS per-
formances are considered in the resource allocation scheme
to support transmissions of the vital signals.
FIGURE 2. Scheduled access mechanism in beacon mode with
superframe boundaries.
III. SYSTEM MODEL
A classical WBAN consists of two main parts, one hub
and Nbody sensors as shown in Fig. 1. The hub usu-
ally has rich resources, such as storage, processing and
energy resources, while the body sensors placed in differ-
ent positions of the body are energy-limited and resource-
constrained. Here the set of the body sensors is expressed
as Cn={1,2,· · · ,N}in this paper. At the network layer,
one-hop star-topology is adopted with considering the limited
resources of body sensors and the uncomfortable user expe-
rience with additional relay nodes. At the MAC layer, a Time
Division Multiple Address (TDMA)-based scheduled access
mechanism in beacon mode with superframe boundaries is
adopted to avoid collisions, idle listening, and overhearing
of sensor nodes, as recommended by IEEE 802.15.6 stan-
dard [21]. As seen in Fig. 2, one beacon and Mslots form
a superframe, and the set of these slots can be expressed
as Cs={1,2,· · · ,M}. To allocate the resources for each
sensor, the hub can broadcast beacons to all sensors in the
beacon slot, while each sensor turns active to receive the
beacons for its allocated resources. Therefore, each sen-
sor can save energy by only working in its dedicated slots
to transmit its data streams and sleep in other slots with
low energy cost. In this paper, the optional transmission
rates are assumed to be discrete in narrowband physical
layer as well as the optional transmission power [21], which
can be expressed as Rdev =Rate1,Rate2,· · · ,RateNR
and Pdev =Power1,Power2,· · · ,PowerNP. In addition,
the transmission rates can be specified by adjusting the
parameters of the modulation and coding scheme (MCS) in
the physical (PHY) layer. Each sensor acquires a kind of
physiological signal and then packetizes acquired data in
packets. Finally, these packets are put in the sensor’s packet
queue. We also assume that the First-In-First-Out (FIFO)
queue strategy and the retransmission strategy are adopted by
VOLUME 5, 2017 20765
Z. Liu et al.: Buffer-Aware RA Scheme With Energy Efficiency and QoS Effectiveness in WBANs
each sensor. Each sensor will try to transmit the first packet
in the packet queue to the hub. If the packet is lost during
transmission, it will be retransmitted until its transmission
is successful. Thus, the packets losses will only occur in
the following two situations: the buffer queue overflow and
the packet delay over the preset threshold.The buffer queue
overflow means that there is no more buffer space to store new
packets, thus the arriving packets will be lost [22]. The packet
delay over the preset threshold means that the packet, whose
delay exceeds the preset delay threshold, will be immedi-
ately dropped from the queue with considering the delay
requirements.
For each energy-limited body sensor, most of the energy
is consumed to transmit data packets, while the energy con-
sumption of receiving ACK packets, processing and beacon
listening can be ignored due to the small size of ACK packets
and low-power processing chip. In addition, the transmission
energy consumption can be subdivided into two parts: the
transmit amplifier energy consumption Etx and the circuitry
energy consumption Ect [12]. Here, the energy model can be
expressed as follows [23],
Econ =(1+α)Etx +Ect (1)
where αis the power amplifier inefficiency factor, Etx =Ptx t
and Ect =Pct t. The transmission circuitry power Pct is
a constant depending on the specific transmitter [24], and
Ptx ∈Pdev is the transmission power. tis the packet trans-
mission time.
To support the QoS requirements with the dynamic link
characteristics, the transmission power can be dynamically
adjusted based on the dynamic link quality. When the link
quality becomes worse, the transmission power should be
tuned up to improve the signal to noise ratio (SNR) [18].
In this paper, we focus on the on-body propagation model.
As recommended by IEEE 802.15.6 [21], the path loss
model of both Light-Of-Sight (LOS) and None-Light-Of-
Sight (NLOS) scenarios can be expressed as follows,
PL (d)=PLd0+10nlog10 d
d0+Xσ(2)
where PLd0is the path loss at a reference distance d0, and n
is the path-loss exponent. The shadowing Xσfollows a nor-
mal distribution Nµs, σ 2
s. In addition, the mean value µs
and the standard derivation σsof the shadowing Xσare
various correspondingly with the human postures and the
environments [7], [8].
IV. BUFFER-AWARE SENSOR STATE
EVALUATION METHOD
With considering the dynamic link characteristics, the buffer
queue occupancy fluctuates over time. More resources should
be allocated for one sensor when its buffer queue occupancy
increases to a certain level. Otherwise, the packet losses will
occur due to the buffer queue overflow or the packet delay
over the preset threshold, which may cause a fatal accident.
On the contrary, the additional resources should be released
for other sensors, when there are only a few packets in the
buffer queue. Thus, the sensor buffer states can be applied
to the resource allocation scheme to improve the system
performance.
In this section, the sensor buffer states, which are the queue
buffer state and the packet delay state, are firstly introduced
to evaluate the buffer queue usage and the performance of the
packets in the buffer. Then based on the real-time values of
sensor buffer states for each sensor, we design a low com-
plexity strategy to dynamically calculate the sensor state and
evaluate whether the sensor should be re-allocated resources
for satisfying the performance of the current packets and the
following packets in the queue.
A. SENSOR BUFFER STATES
In this paper, we design two buffer states, the queue state and
the packet delay state, to evaluate the sensor state. When the
short-term link quality becomes worse, the first packet in the
queue buffer may try several times before final successful
transmission. Then the buffer queue will cache some packets
to be transmitted in the following superframe. In addition,
according to the FIFO strategy, the arriving packets in the
following superframe will wait for the dedicated time slots
of the body sensor to be transmitted until all blocked packets
are transmitted, thus all these packets will have a high delay.
At this time, if no more resources are allocated to be used
to transmit these blocked packets, the packets blocked in the
buffer queue will continue to accumulate and the delay of
the blocked packets will increase. Furthermore, the packet
losses will occur. Therefore, the sensor buffer states must be
carefully designed to reflect the packet status in the buffer
queue for enhancing the short-term QoS performance. In this
paper, the queue buffer state is designed to measure the queue
usage and evaluate whether the buffer queue will be overflow.
The packet delay state is formulated to measure the delay of
the packets in the buffer queue and evaluate whether these
packets will be discarded because the delay exceeds the preset
delay threshold.
Definition 1 (Queue Buffer State): In this paper, the queue
buffer state Qiof the sensor iis designed to measure the
occupancy rate of the buffer queue. Based on the queue
buffer state Qi, the sensor can evaluate whether the arriving
packets in the next superframe will be discarded because the
remaining buffer is not enough. Once the buffer queue is
increasing due to the packet loss, more resources should be
allocated to the sensor, otherwise the buffer queue will be
overflow. To evaluate the queue usage, the ratio of the coming
number of packets in the following superframe to the buffer
queue remaining space is used as the queue buffer state Qi,
which is expressed as follows,
Qi=NS,i
Nq,i−1Ni
(3)
where NS,i=lSi·Tframe
Limis the average number of arriving
packets for sensor i,i∈Cnin one superframe. Siis the
20766 VOLUME 5, 2017
Z. Liu et al.: Buffer-Aware RA Scheme With Energy Efficiency and QoS Effectiveness in WBANs
average source rate of sensor i.Tframe is one superframe
length in second, and Liis one packet length in bits. The
mathematical symbol d·eis regarded as the rounding function
which returns the upward-rounded value of one input number.
Nq,irepresents the maximum storage capacity of the buffer
queue to cache packets for sensor i.1Niis the number of
blocked packets in the buffer queue when sensor ihas no time
slots to transmit more packets in the current superframe.
When there are no packets blocked in the queue buffer,
the queue buffer state Qireaches the minimum value
Qi=NS,i
Nq,i. And then the value of the queue buffer state Qi
increases with the blocked packets. When the value of the
queue buffer state Qiis gradually close to 1, it means the
blocked packets are gradually increasing and the remaining
buffer queue will not be enough to cache the arriving packets
in the following superframe. Once the value of the queue
buffer state Qiis greater than 1, it means that the remaining
buffer cannot store all the arriving packets in the following
superframe and thus some packets will be lost due to the
buffer queue overflow. In a word, the larger value of the
queue buffer state Qiwill cause packet losses with a higher
probability due to the buffer queue overflow.
Definition 2 (Packet Delay State): According to the FIFO
strategy, the first packet in the queue will be first transmitted
to the hub and the following packets will not be transmitted
until all packets before they have been transmitted. Corre-
spondingly, the packets in the front of the buffer queue have
waited for a longer time than the packets in the back of the
buffer queue, therefore the first packet in the buffer queue has
the maximum delay. Therefore, if the delay of the first packet
is not larger than the preset delay threshold, all packets in the
buffer queue will not be dropped due to the delay over the
threshold. In this paper, the packet delay state Difor sensor
iis defined to measure the delay of the first packet in the
buffer queue, and then it can be used to evaluate whether the
packet losses will occur which are caused by that the delay
exceeds the preset delay threshold. The packet delay state Di
is expressed as follows,
Di=(ki+1)·Tframe
tD,th,i
(4)
where kiis the number of the superframes that the first packet
has waited from when it is put in the buffer queue to current
superframe. Then plus 1 is to wait until the next superframe
to transmit the first packet. tD,th,iis the preset delay threshold,
which can be represented as the upper bound of the acceptable
delay.
The packet delay state Diincreases with the number kiof
superframe waiting in the queue, and the larger kimeans that
the first packet has been waiting for a longer time to be trans-
mitted to the hub and more packets have been blocked in the
queue. When the packet delay state Diis close to 1, the first
packet and even the following packets will be discarded in
the next superframe due to the delay over the preset threshold
tD,th,i. In addition, the packet delay state Didecreases with
the preset delay threshold tD,th,i. If the delay requirements of
some sensors are not so stricter, the preset delay threshold
tD,th,ican be set to a larger value. In a word, the larger packet
delay state Dimeans the packets in the queue buffer are
more urgent to be transmitted. If these packets cannot be
transmitted in time to the hub, the packets will be dropped
when their delay exceeds the preset delay threshold.
FIGURE 3. First-In-First-Out queue model and buffer-aware sensor state
evaluation method in each sensor.
B. SENSOR STATE EVALUATION METHOD
Considering the dynamic link characteristics, both the queue
buffer state and the packet delay state are time-varying, which
are shown in Fig. 3. When the queue buffer state or the packet
delay state increases to a certain extent, the sensor should
be allocated more resources to transmit the blocked packets
in the buffer queue for avoiding the packet losses. When
the blocked packets have been successfully transmitted to
the hub, additional resources need to be released for other
sensors. Therefore, when the request should be transmitted
by the sensor to the hub for the resource re-allocation is a
key problem. Because frequently applying for the resource
re-allocation will lead to a decrease of the system stability,
meanwhile the complexity of the system will also increase.
Therefore, we design the sensor state evaluation method with
a low complexity to decide when it is proper to send the
request for the resource re-allocation. Generally, the sensor
state is calculated based on the buffer queue state and the
packet delay state. It only changes when the sensor buffer
states become more difficult to be accepted by the sys-
tem or when the sensor buffer states become very good. In this
paper, a low-complexity strategy is designed to evaluate the
sensor state in consideration of the limited processing capac-
ity of body sensors, which is expressed as follows,
Sm,i=
1,if Sm−1,i=0∧Qi≥Qup,i∨Di≥Dup,i
0,if Sm−1,i=0∧Qi<Qup,i∧Di<Dup,i
0,if Sm−1,i=1∧Qi<Qlow,i∧Di<Dlow,i
1,if Sm−1,i=1∧Qi≥Qlow,i∨Di≥Dlow,i
(5)
where mmeans the m-th superframe. Sm,iis the sensor state
parameter in m-th superframe for sensor iwhile Sm−1,imeans
the value of the sensor state in (m−1)th superframe. Qup,iis
the upper threshold of the queue buffer state, and Qlow,i
represents the lower limit of the queue buffer state for sensor i.
VOLUME 5, 2017 20767
Z. Liu et al.: Buffer-Aware RA Scheme With Energy Efficiency and QoS Effectiveness in WBANs
Dup,iand Dlow,iare the upper and lower bounds of the packet
delay state.
The sensor state Sm,iis determined by three parameters,
the last sensor state Sm−1,i, the queue buffer state Qi, and the
packet delay state Di. We should first check the value of
the last sensor state Sm−1,i, and then obtain the sensor state
Sm,iwith the different value of queue buffer state and the
packet delay state. There are total four different situations:
1) When the last sensor state Sm−1,iequals to 0, and the queue
buffer state Qior the packet delay state Diexceeds its preset
upper threshold Qup,ior Dup,i, it means the blocked packets
are increasing and more resources are needed to transmit
the blocked packets. Otherwise, the packet losses will occur.
At this time, the Sm,iwill changes from 0 to 1 to indicate
the requirement of more resources. 2) When the last sensor
state Sm−1,iequals to 0, both the queue buffer state Qiand
the packet delay state Diare lower than the preset upper
thresholds Qup,iand Dup,i, respectively. It means both the
queue buffer state and the packet delay state are within the
acceptable range, thus the sensor state Sm,i=0 remains
unchanged. 3) When the last sensor state Sm−1,iequals to 1,
and both the the queue buffer state Qiand the packet delay
state Diare decreasing to less than the lower thresholds Qlow,i
and Dlow,i, respectively, it means the blocked packets are
reduced to a small value and additional resources can be
released to other sensors. Therefore, the sensor state Sm,i
changes from 1 to 0. 4) When the last sensor state Sm−1,i
equals to 1, and the queue buffer state Qior the packet delay
state Diis still larger than the lower threshold Qlow,ior Dlow,i,
respectively, it means the urgent situation is not dismissed and
the sensor state Sm,iremains unchanged, which equals to 1.
The pseudo codes of the sensor state evaluation method are
illustrated in Algorithm 1.
Meanwhile, the sensor state would not be introduced too
much overhead of the data frame. The parameters of the
queue buffer state Qiand the packet delay state Di, such as
Qup,i,Qlow,i,Dup,iand Dlow,i, can be assumed as the given
constants for the hub. The source rates of each sensor can
be obtained by the hub through the initializing stage, and
then the average number of arriving packets NS,iin one
superframe can also be achieved by simple calculation. Some
other parameters of the sensor can be derived in the hub, and
the initial sensor state S0,iis set to 0 with the initial empty
buffer queue. For each sensor, some simple calculations are
enough to obtain the sensor state, whose time complexity in
the sensor state evaluation method is low and acceptable for
the resource-constrained sensor.
V. DESIGN OF MIX-COST PARAMETER
In this section, we first design a QoS cost to character-
ize the gap between the attainable QoS support and QoS
requirements for evaluating the QoS effectiveness. Then we
formulate an energy cost, which is the equivalent energy con-
sumption per bit with considering the packet losses. Finally,
we define the mix-cost parameter with considering both the
Qos cost and the energy cost.
Algorithm 1 Buffer-Aware Sensor State Evaluation Method
1: if (ki+1) ·Tframe >tD,th,ithen
2: Discard several oldest data packets, whose delay will
exceed the preset threshold in the head of the queue.
3: Update the remaining number 1Niof packets in the
queue.
4: end if
5: if 1Ni+Narr,i>Nq,ithen
6: Discard 1Ni+NS,i−Nq,ioldest data packets in the
head of the queue.
7: Store Narr,inew data packets at the end of the queue.
8: end if
9: Update the remaining number 1Ni=1Ni+Narr,iof
packets in the queue.
10: Obtain queue buffer state Qiand packet delay state Diby
sensor i.
11: Update the sensor state Sm−1,iaccording to Eq. 5.
12: if Sm−1,i== 0then
13: if Qi≥Qup,iand Di≥Dup,ithen
14: Sensor state becomes serious, and set Sm,i=1.
15: Update the sensor state Sm,iin the data frame for
requesting more resources by the Hub.
16: else
17: Sensor state remains unchanged Sm,i=Sm−1,i=0.
18: end if
19: else
20: if Qi<Qup,iand Di<Dup,ithen
21: Sensor state becomes good, and set Sm,i=0.
22: Update the sensor state Sm,iin the data frame for
releasing some resources by the Hub.
23: else
24: Sensor state remains unchanged Sm,i=Sm−1,i=1.
25: end if
26: end if
A. QoS COST
To make the queuing system stable for each sensor,
the throughput condition always needs to be satisfied [25].
However, due to the dynamic body link characteristics,
the short-term throughput condition could not be met, which
may result in an amount of dropped packets. Thus, the sat-
isfaction degree of the throughput condition and the sensor
state can be used to evaluate the QoS performances. Here,
the average packet loss rate PLRave in the physical layer of
the body link can be used to evaluate the long-term PLR
performance of the time-variant body link, which can be
expressed as follows,
PLRave =Z+∞
0
PLR(γ)P(γ|µγdB , σγdB )dγ(6)
where γ=10 Ptx −PL(d)−PN
10 B
Rrepresents the bit signal-to-noise
ratio (SNR). PNis the noise power. Ris the transmission
rate. Bis the bandwidth. The probability density function
P(γ|µγdB , σγdB ) of bit SNR follows a log-normal distribution
20768 VOLUME 5, 2017
Z. Liu et al.: Buffer-Aware RA Scheme With Energy Efficiency and QoS Effectiveness in WBANs
with the mean γand the standard deviation σγdB as the shad-
owing. The packet loss rate PLR(γ) is a decreasing function
of current bit SNR γ, and the details of PLR(γ) are related to
the modulation and coding mode in the PHY layer [26]. In this
paper, PLRave,irepresents the average packet loss rate of sen-
sor i. For sake of readability, when it is not strictly necessary
to distinguish among the sensors, the average packet loss rate
will be simply denoted with PLRave in the remaining paper,
and this same method is also applicable to other parameters.
To analyze the QoS cost, not only the arriving packets in
each superframe but also the blocked packets in the buffer
queue should be taken into consideration. If the throughput
condition can be guaranteed to support the arriving packets,
the packet loss rate requirement can be ensured, however the
delay of the blocked packets will be at a high value with
adopting the FIFO strategy and the retransmission strategy.
Thus both the arrivals of packets and the blocked packets
in the buffer queue should be introduced to the QoS cost.
To ensure the system stability, the throughput requirement
should be guaranteed with considering the path loss for the
queuing system. Then the equivalent number of transmissions
in one superframe can be represented as follows,
Ntran,th,i=&NS,i
1−PLRave,i'(7)
where Ntran,th,irepresents the minimum number of trans-
missions to transmit the arriving packets in one superframe
with considering the average PLR. In addition, the hub can
estimate the number of packets 1Nth,iin the queue of sensor i
with the obtained parameters of the sensor buffer states in the
sensor state evaluation method. The hub can obtain the sensor
state Sm,ithrough received data packets. Once Sm,ichanges
from 0 to 1, the number of packets 1Nth,iin the queue of
sensor ican be obtained by using the following expression,
1Nth,i=min Nq,i−NS,i
Qup,i,
Dup,i·tD,th,i
Tframe
−1·NS,i (8)
where 1Nth,iis assumed as 0 when the sensor state Si
equals to 0. When the sensor iis allocated with Nslot,i
number of slots, the equivalent number of transmissions
is Ntran,i=jRi·Nslot,i·tslot
Lik, where the mathematical sym-
bol b·c is one rounding function which returns the downward-
rounded value of the input number, tslot is the slot length in
second, and Riis the transmission rate of sensor i. The details
of each slot assignment in one superframe can be defined as
the slot assignment variables ρi,j,i∈Cn,j∈Cs, shown as
follows,
ρi,j=1,if slot jis assigned to sensor i
0,otherwise (9)
where Pi∈Cnρi,j≤1 means each slot of the superframe can
only be assigned to one sensor at most. Finally, the satisfac-
tion degree of the throughput condition can be expressed as
a function of Ntran,i,NS,i,Ntran,th,iand 1Nth,i, where 1Nth,i
has different values with different sensor states. The QoS cost
can be obtained in the following situations, respectively.
1) When the throughput condition cannot be guaranteed
without considering both the packs blocked in the
queue and the packet loss due to the dynamic link,
formulated as Ntran,i≤NS,i, it means the system
stability of the queuing system cannot be ensured and
then the buffer queue will be full. Thus the QoS cost
will be set to 1 to indicate the urgent requirements for
additional resources.
2) When the throughput condition with considering both
the arriving packets in one superframe and the blocked
packets can be satisfied, which can be formulated as
Ntran,i≥Ntran,th,i+1Nth,i, it means not only the
arriving packets but also the blocked packets in the
buffer queue can be transmitted with enough resources.
Therefore the QoS cost will be set to 0.
3) Otherwise, the satisfaction degree of the through-
put condition can be formulated as SDi=
Ntran,i−NS,i
Ntran,th,i+1Nth,i−NS,i. The QoS cost is the function of
f(SDi), where the function f(·)should have the fol-
lowing features. Firstly, for the sensor i, the QoS cost
should be the decreasing function of the SDi. Secondly,
when the resources cannot satisfy all the sensors’
requirements due to the dynamic links, the sensor with
the middle value of SDishould have more opportunity
to get more resources for improving the SDi, and then in
the next superframe more resources can be reserved for
other sensors. Because the packets blocked in the queue
will increase shapely when the throughput condition is
still unable to be satisfied.
Finally, the QoS cost can be expressed as following,
CQoS,i=
1,if Ntran,i<NS,i
0,if Ntran,i≥Ntran,th,i+1Nth,i
fNtran,i−NS,i
Ntran,th,i+1Nth,i−NS,i,
if NS,i≤Ntran,i<Ntran,th,i+1Nth,i
(10)
where 1Nth,ihas different values for different sensor states.
When the sensor state Siequals to 0, 1Nth,iis set to 0. And
when the sensor state Siequals to 1, 1Nth,ican be obtained
through Eq.(8). Thus, the QoS cost needs to be recalculated
when the hub receives the packet and observes a change of
the sensor state in the data frame.
In this paper, we carefully design the function f(x)to con-
struct the relationship between the QoS cost and the satisfac-
tion degree of the throughput condition. As shown in Eq. (11),
f(x)is the decreasing function of the satisfaction degree
x∈[0,1] and f(x≤0)=1,f(x≥1)=0, whose curve
is given in Fig. 4. The QoS cost will increase or decrease
sharply when the satisfaction degree xis near 0.5, and the
QoS cost will have the largest amount of reduction with the
same increment of the satisfaction degree. Thus the sensor
with a middle value of satisfaction degree will have more
VOLUME 5, 2017 20769
Z. Liu et al.: Buffer-Aware RA Scheme With Energy Efficiency and QoS Effectiveness in WBANs
opportunity to get more resources in the following resource
allocation scheme to minimize the QoS cost. The f(x)can be
expressed as follows,
f(x)=
0,if x <0
1,if x >1
1
e4−1·e4+1
1+e(8x−4) −1,otherwise
(11)
FIGURE 4. The curve of the function f(x).
B. ENERGY COST
In this paper, we assume the hub is energy and resource
sufficient while the body sensor is constrained by energy.
Thus, only the energy efficiency of the body sensors is con-
sidered in this paper. The energy consumed per transmitting
a bit is generally formulated to evaluate the energy efficiency
of the body sensors. In addition, the dynamic link charac-
teristics should be taken into consideration, and the packet
loss rate should be also introduced to the energy cost for
better evaluating the energy efficiency. Finally, the energy
cost CEis formulated as the equivalent energy consumed
per transmitting a bit with considering the packet loss rate
due to the dynamic link quality. Generally, the energy cost
CEhas a wide range of values, especially some values are
larger than 1. To balance the energy cost and the QoS cost,
proper normalization method should be applied to the energy
cost CE. In this paper, the linear normalization method is
adopted to normalize the energy cost CEinto the range [0,1]
for further better calculating the mix-cost. The energy cost
CE,ican be expressed as follows,
CE,i=
1,if (PLRave,i≥PL Rth,i)
(1+α)Ptx,i+Pct ,i
Ri1−PLRave,i·Rmin 1−PL Rth,i
(1+α)Ptx,max +Pct ,i
,
otherwise
(12)
where the transmission power Ptx,iof sensor ibelongs to the
transmission power set Pdev, and Ptx,max is represented as the
maximum transmission power in Pdev. The transmission rate
Riof sensor iis chosen from the transmission rate set Rdev,
and Rmin is the minimum transmission rate in Rdev.
To evaluate the energy cost, we should first check the
packet loss rate performance. If the link quality becomes
worse due to the dynamic postures and environments,
the attainable average packet loss rate PLRave,iwith current
transmission power and transmission rate may be larger than
the acceptable threshold PLRth,iof the packet loss rate, thus
the energy consumption of the retransmissions for success-
fully transmitting packets will increase sharply. More packets
will be blocked in the buffer queue, thus the delay of these
blocked packets will also have a corresponding increase.
Therefore, the energy cost should be set to the maximum
value 1. Otherwise, the equivalent energy consumption per
bit after normalization is used to evaluate the energy cost with
considering the packet loss rate.
C. MIX-COST PARAMETER
As described above, the energy cost is designed to evaluate
the energy efficiency, while the QoS cost is formulated to
estimate the QoS effectiveness with different sensor states,
and both the energy cost and the QoS cost can reflect the
performances of the WBAN system. In this paper, a mix-cost
parameter is defined and formulated as the combination of
the energy cost CEand the QoS cost CQoS . The mix-cost can
be mathematically expressed as follows,
CMix,i=δ·CE,i+(1 −δ)·CQoS ,i(13)
where δis a weight value used to make the trade-off between
the energy efficiency and the QoS satisfactory, and it is in
the range of [0,1]. Due to the normalization, both the energy
cost and QoS cost are in the range of [0,1], thus the mix-cost
CMix has the same range [0,1]. When sufficient resources are
allocated to the sensor, the energy cost and the QoS cost will
have a small value as the above expressions. Correspondingly
the small value of the mix-cost indicates that the sensor with
currently allocated resources achieves a higher performance,
such as energy efficiency and QoS effectiveness.
VI. RESOURCE ALLOCATION PROBLEM
In this section, we first formulate the optimal resource allo-
cation problem to minimize the total mix-cost of all sensors
by optimizing the resources. Then, a greedy sub-optimal
resource allocation scheme is proposed to decrease the
time complexity of the resource allocation problem. Finally,
we analyze the time complexity and space complexity of both
the optimal and sub-optimal resource allocation schemes.
A. FORMULATION OF RESOURCE ALLOCATION PROBLEM
Considering that the mix-cost combines both the energy
cost and the QoS cost, which are formulated as the func-
tion of the transmission rate, the transmission power and
the allocated time slots, the weighted sum of all sensors’
mix-cost can be regarded as the objective function to be
minimized to obtain the smallest energy cost and QoS cost
for a highest performance of the WBAN system. Thus,
the resource allocation problem is designed to minimize the
weighted sum of all sensors’ mix-cost by optimizing the
20770 VOLUME 5, 2017
Z. Liu et al.: Buffer-Aware RA Scheme With Energy Efficiency and QoS Effectiveness in WBANs
FIGURE 5. The proposed framework of the buffer aware energy-efficient and QoS-effective resource allocation scheme in WBANs.
transmission rates, the transmission power and the allocated
time slots for each sensor. The optimal resource allocation
problem can be expressed as follows,
min
Ri,Ptx,i,Nslot ,iX
i∈Cn
ωi·CMix,i(14)
s.t.Ptx,i∈Pdev ,(14a)
Ri∈Rdev,(14b)
X
i∈Cn
ρi,j≤1,(14c)
ρi,j∈{0,1},∀i∈Cn,∀j∈Cs,(14d)
Nslot,i=X
j∈Cs
ρi,j,(14e)
where ωiis the weight of the sensor iin total sensors’ mix-
cost, which can be used to represent the importance of the
sensor. The larger weight means the sensor is more important
when allocating the resources, it is because that the hub
prefers to allocate the resources to the sensor with a high
weight to minimize the weighted mix-cost when the resources
are not enough for all sensors. The objective function (14) is
to minimize the weighted sum of the mix-costs. In addition,
the resource allocation problem has three constrains: 1) the
transmission power must belong to the transmission power set
Pdev (14a); 2) the transmission rate must be in the transmis-
sion rate set Rdev (14b); 3) each time slot can only be allocated
to one sensor at most (14c)-(14e).
The details of the proposed framework are given in Fig. 5.
Each sensor can calculate its sensor state with adopting the
sensor state evaluation method, and the hub can obtain the
sensor state of each sensor through receiving the data pack-
ets. Once the sensor state changes, the hub will resolve the
optimization problem to reallocate the resources for each
sensor to cope with the corresponding sensor state. These new
resource allocation will be broadcast to all sensors through
the beacons in each superframe.
B. SUB-OPTIMAL RESOURCE ALLOCATION SCHEME
The resource allocation problem in (14) is designed to find
out the optimal allocation of resources for each sensor based
on the sensor state. However, the resource allocation problem
is a mix integer nonlinear programming (MINLP) problem,
which is NP-hard. To find the optimal solution, the exhaustive
search method is a possible way for the resource alloca-
tion problem. However, the exhaustive search method has
a high computational complexity, which is not suitable for
running on the hub. In addition, the complicated calculations
will introduce unacceptable processing delay for the WBAN
system. In this section, a greedy sub-optimal resource allo-
cation algorithm is proposed to allocate the resources for
each sensor. In addition, the sub-optimal resource allocation
algorithm has a much lower complexity than the exhaustive
search method. The details of the greedy resource allocation
algorithm are explained and given in the following.
The algorithm includes two steps. Firstly, we assume each
sensor owns the entire channel, and then calculates the opti-
mal the transmission rate and the transmission power with
different numbers of time slots in the range [1,M]. Then
we find the optimal required number of time slots which
corresponds to the minimum mix-cost for each sensor, respec-
tively. Secondly, we should check whether the total time slots
in one superframe can satisfy all sensors’ optimal number of
time slots. If the slot condition can be satisfied, the optimal
resources for each sensor are the final results. Otherwise,
a greedy approach is adopted to find the sub-optimal allocated
resources for each sensor step by step until the slot condition
can be met.
When we only focus on the optimal resource allocation of
one sensor with the preset number of time slots, the optimal
transmission rate and transmission power for each sensor
i,i∈Cncan be easily obtained by solving the following
optimization problem,
min
Ri,Ptx,i
CMix,i(15)
VOLUME 5, 2017 20771
Z. Liu et al.: Buffer-Aware RA Scheme With Energy Efficiency and QoS Effectiveness in WBANs
s.t.Ptx,i∈Pdev ,(15a)
Ri∈Rdev,(15b)
X
j∈Cs
ρi,j=m,(15c)
where mis a constant which denotes the time slots allocated
for sensor iin one superframe, and its value is in the range
of [1,M]. With different number mof time slots, the optimal
transmission rate R∗
i(m) and the optimal transmission power
P∗
tx,i(m) can be obtained by solving the optimization problem.
Once we obtain the mapping relations between the optimal
transmission rate, the optimal transmission power, and time
slots, the optimal number m∗
iof time slots can be achieved
to find the minimum mix-cost for sensor iwith corre-
sponding optimal transmission rate R∗
i(m∗
i) and transmission
power P∗
tx,i(m∗
i).
The number m∗
iof time slots for each sensor is optimal
without considering the constraint on the total number of time
slots. If the total optimal time slots for all sensors can satisfy
the time slot constraint, which means Pi∈Cnm∗
iis no larger
than the total slot number Mof one superframe, the R∗
i(m∗
i)
and P∗
tx,i(m∗
i) will be regarded as the final optimal resource
allocation results for each sensor. However, when the total
number of one superframe cannot cover the optimal time
slots m∗
ifor all sensors due to the bad link quality, we try
to search for the sub-optimal resource allocation results with
using a greedy algorithm. Once the time slot constraint cannot
be satisfied, some sensors should choose the sub-optimal
number of time slots to reduce the total allocated time slots.
In this paper, the total allocated time slots are reduced step
by step until it is not larger than the total slot number Mof
the superframe. In each iteration, we first reduce one slot for
all sensors and compare the weighted mix-cost increments
of each sensor caused by the time slot reduction, and then
the sensor with the minimum weighted mix-cost increment
is chosen to reduce one allocated slot. If the weighted mix-
cost increments of all sensors are equal to 0, we will add one
slot to the number of slot reduction in the next step until we
find one sensor with minimum weighted mix-cost increment,
which must not be equal to 0, to the reduce allocated time
slots. The details of the proposed greedy resource allocation
algorithm are described in Algorithm 2.
C. COMPLEXITY ANALYSIS
Here we theoretically analyze both the complexity of the opti-
mal resource allocation problem and sub-optimal resource
allocation scheme. For the exhaustive search method of
the optimal resource allocation problem (14), the upper
bound for the time complexity can be derived as Tsearch =
O(PM
n=N
n!
(n−N+1)!·(N−1)!·NRNP), where the time complexity
of calculating one mix-cost parameter is assumed as O(1) and
the constant coefficients and lower order terms are ignored.
Compared to the exhaustive search method, the upper bound
for time complexity of the proposed sub-optimal method is
TsubOpt =O(NNRNP(M−1)+(N−1)MN ). We can find that
the proposed sub-optimal method decreases the complexity
Algorithm 2 Greedy Resource Allocation Algorithm
Require:
1: Calculate P∗
tx,i(m),R∗
i(m),m∈[1,M] for sensor i,i∈
Cnby solving problem (15).
2: Find m∗
i=arg min
P
j∈Cs
ρi,j=m
CMix,ihP∗
tx,i(m),R∗
i(m)iand set
sub-optimal time slot number m†
i=m∗
i.
3: Get the difference of total required optimal slot number
and the superframe length Mby 1total =P
i∈Cn
m∗
i−M.
4: Set initial decrement of slots 1dec =1.
Ensure:
5: while 1total −1dec ≥0do
6: Set candidate time slot number mc
i=m†
i−
1dec,if m†
i>1 for each sensor.
7: Get the weighted mix-cost increment caused by the
reduction of a slot τi=ωi·CMix,ihP∗
tx,i(mc
i),R∗
i(mc
i)i−
ωi·CMix,ihP∗
tx,i(m†
i),R∗
i(m†
i)i.
8: if all τi,i∈Cnequal to 0 then
9: 1dec =1dec +1.
10: else
11: Find the sensor with the minimum mix-cost incre-
ment ind =min
iτi, and update m†
ind =mc
ind .
12: Update the difference 1total =1total −1dec.
13: Update 1dec =1.
14: end if
15: end while
of the resource allocation scheme significantly. Meanwhile,
the proposed sub-optimal method has a larger space complex-
ity, which is given by the O(NNRNP(M−1)). Considering that
the number of sensors in a WBAN is usually small, the time
and space complexity of the proposed sub-optimal resource
allocation scheme is acceptable.
VII. SIMULATIONS
A. SIMULATION SETTING
In this section, an event-driven WBAN system based on
MATLAB is built to implement the proposed algorithms for
a variety of simulations. The application module generates
the preset rate of packets, and the MAC module supports the
scheduled access mechanism in beacon model with super-
frame boundaries as recommended by IEEE 802.15.6 [21].
Besides, the PHY module can set specific transmission rate
by adjusting the parameters of MCS. The channel module
generates the path loss in each slot by using the reference
code in IEEE 802.15.6 standard [27]. A classical WBAN,
which consists of one hub and four sensors, is adopted in the
simulations, and the Table 1 shows the deployed positions,
the sample rates and the corresponding body link parameters
of all sensors. We assume that all sensors have the same 100 J
battery capacities. In the simulations, the variation of the
postures is modeled as a Markov chain, and the probability of
different posture change can be determined from real human
20772 VOLUME 5, 2017
Z. Liu et al.: Buffer-Aware RA Scheme With Energy Efficiency and QoS Effectiveness in WBANs
TABLE 1. Sensor parameters.
posture trace [28]. For convenience, we only consider three
types of body postures, i.e., still, walk and run, and their
steady-state probabilities are set to 0.5, 0.3 and 0.2, respec-
tively. The extension to the case with more body postures is
straightforward. Besides, we assume that these postures can
be identified with high accuracy in real time by hub [29].
In this paper, the standard derivation σsof the shadowing
changes with the postures as shown in Table 2, which are set
based on the measurement results in [7] and [8]. We assume
that the path loss for each sensor remains unchanged during
a superframe period since the duration of the slow fading is
generally larger than the typical length of one superframe,
e.g., 100ms [33]. In addition, the mean value µsof the shad-
owing for all sensor are assumed to be same, and the different
values of µsare used to simulate the different environment for
convenience [8]. The higher value of µsmeans the higher path
loss and the worse link quality. Other simulation parameters
are summarized in Table 3.
TABLE 2. Parameters of the shadowing.
TABLE 3. Simulation parameters.
In order to better evaluate the performances of our pro-
posed algorithms, we compare four different approaches in
the simulations:
FIGURE 6. Relationship between the average packet loss rate and the
mean value µsof the shadowing.
•The proposed Buffer-aware Energy efficient and QoS
effective Resource allocation scheme with applying the
sensor state evaluation method in each sensor, abbrevi-
ated BEQR.
•The proposed Sub-optimal Buffer-aware Energy effi-
cient and QoS effective Resource allocation scheme with
applying the sensor state evaluation method in each
sensor, abbreviated sub-BEQR.
•The proposed Energy efficient and QoS effective
Resource allocation scheme without the buffer evalua-
tion method, abbreviated EQR.
•The Link-State-Estimation-Based transmission power
control (LSEPC) [14] is the enhanced transmission
power control approach, in which not only the short-term
link estimation but also the long-term link estimation
is used to adapt the transmission power, abbreviated
LSEPC.
In this paper, we want to evaluate the role of the sensor
state evaluation method for improving the performance of
the WBAN, so both BEQR and EQR are compared in the
simulations. In addition, the performance of the sub-optimal
resource allocation scheme sub-BEQR is also compared with
BEQR at low complexity.
B. SIMULATION RESULTS
We first evaluate the PLR and delay performances under dif-
ferent mean value µsof the shadowing corresponding to the
different environment. In Fig. 6, we illustrate the relationship
between the average packet loss rate and the mean value µs
of the shadowing, while the relationship between the average
delay and the mean value µsof the shadowing is shown
in Fig. 7. We can see that the proposed BEQR approach
has the best QoS performances, i.e., the PLR performance
in Fig. 6 and the delay performance in Fig. 7. This is because
that not only the energy efficiency and QoS requirements are
introduced to the resource allocation scheme in the BEQR,
but also the sensor buffer state is designed to dynamically
evaluate the real-time sensor state for further improving
VOLUME 5, 2017 20773
Z. Liu et al.: Buffer-Aware RA Scheme With Energy Efficiency and QoS Effectiveness in WBANs
FIGURE 7. Relationship between the average delay of packets and the
mean value µsof the shadowing.
FIGURE 8. Relationship between the system lifetime and the mean value
µsof the shadowing.
the performance. In addition, the QoS performances of the
EQR approach are much worse than the LSEPC when µsis
less than 22dBm. This is because when the EQR approach
takes both the energy and the statistical QoS performances
seriously to allocate resources for each sensor, but it does not
take the real-time buffer state of each sensor into considera-
tion. Compared with the proposed EQR approach, the buffer-
aware sensor state evaluation method is adopted by each
sensor in the BEQR, thus each sensor can adjust the sensor
state according to the buffer states due to the dynamic link
quality, and then it can decide when applying the resource
re-allocation. When the sensor state of one sensor changes
from good to bad, the sensor will apply more resources in
a timely manner to prevent more packets from blocking in
the buffer queue. On the contrary, the sensor will release
additional resources to other sensors for maximizing the
resource utilization. Thus, the proposed buffer aware BEQR
approach can better improve the QoS performances than
the proposed EQR approach, and the QoS performances
of the BEQR as shown in Fig. 6 and Fig. 7 demonstrates
the effectiveness of the buffer-aware sensor state evaluation
method.
To analyze the energy efficiency of the proposed
approaches, the relationship between the system lifetime and
the mean values µsis given in Fig. 8. We can find that
both the proposed BEQR approach and the proposed EQR
approach can achieve much longer system lifetime compared
with the LSEPC approach. This is because that the resource
allocation scheme is carefully designed to minimize both
the energy cost and the QoS cost for each sensor. Thus the
energy efficiency can be improved greatly, meanwhile the
QoS performances are also good as shown in Fig. 6 and Fig. 7.
Besides, the system lifetimes of all approaches decrease with
the increase of the mean values µsof the shadowing. It means
that when the environment becomes stricter and then the
link quality of all sensors becomes worse, all approaches
will allocate more resources and increase the transmission
power to cope with the stricter environment for guaranteeing
the QoS performances. In addition, the proposed sub-BEQR
approach has very similar energy and QoS performances with
the proposed BEQR approach. This is because the greedy
suboptimal resource allocation problem searches the sub-
optimal results step by step with a greedy method, and the
results with the minimum increase of mix-cost are obtained
at each step. When the resources can satisfy the optimal
resource requirements of all sensors, the sub-BEQR and
BEQR will have the same resource allocation results.
FIGURE 9. Changes of the number of packets in the queue buffer over
time in sensor 2.
We also evaluate the effect of the buffer-aware sensor state
evaluation method in the proposed BEQR and sub-BEQR
approaches, and analyze the dynamic queue buffer occupancy
over time. As shown in Fig. 9, the number of packets in the
queue buffer dynamically changes with the increase of the
superframe index. We can see that the number of packets of
the proposed BEQR and sub-BEQR approaches always keep
a lower level than 30 packets, while that of the EQR and the
LSEPC gradually increase to a high value, i.e., more than
120 packets. This is because that the buffer-aware sensor state
20774 VOLUME 5, 2017
Z. Liu et al.: Buffer-Aware RA Scheme With Energy Efficiency and QoS Effectiveness in WBANs
evaluation method can evaluate the sensor state based on the
buffer occupancy and the packet delay state. When the sensor
state becomes worse due to the bad link quality, the number
of packets in the buffer queue will gradually increase to a high
level if there are no more resources allocated for this sensor.
At this time, the sensor with using buffer-aware sensor state
evaluation method will send the re-allocation request to the
hub for more resources with only introducing one bit in the
data frame. Thus the packets in the buffer queue can then be
transmitted to the hub with enough resources, and then the
number of packets in the buffer queue can keep a smaller
value.
FIGURE 10. (a) Average packet loss rate due to the buffer overflow versus
the buffer size of each sensor. (b) Average packet loss rate due to delay
over the preset threshold versus the buffer size of each sensor.
To better analyze the effect of the buffer size, we increase
the buffer size from 10 Kbit to 120 Kbit, and we assume
all sensors have the same buffer size. Besides, the mean
value µsof the shadowing is set to 30dB to simulate the worse
environment for better evaluating the effect of the buffer size
on the system performance. Considering that the packet loss
rate is mainly caused in two situations: the buffer overflow
and the packet delay over the preset threshold, we analyze
the average PLR due to the buffer overflow and the average
PLR due to delay over preset threshold versus the buffer size,
respectively, in Fig. 10. We can see that the PLR due to buffer
overflow decreases with the increase of the buffer size, while
the PLR due to delay over preset threshold increases with
more buffer size when the buffer size is below 100 Kbit.
This is because larger buffer size means longer buffer queue,
which can be used to store more blocked packets, and thus
the packets have a lower probability to overflow. Meanwhile,
more blocked packets will cause more packet losses due to
the delay over the preset threshold with larger buffer queue.
However, the total PLR, which consists of the PLR due to the
buffer overflow and the PLR due to delay over preset thresh-
old, decreases with the increase of the buffer size as shown
in Fig. 11, while the average packet delay correspondingly
increases with larger buffer in Fig. 12. This is because more
buffer resources can be utilized with the resource allocation
scheme to improve the PLR performance. In addition, more
packets will wait longer in the buffer to be successfully
transmitted until more resources can be allocated by the hub
corresponding to the sensor state. Thus, a much larger delay
performance is the price of a smaller PLR performance.
FIGURE 11. Total average packet loss rate, which consists of the PLR due
to the buffer overflow and the PLR due to delay over preset threshold due
to the buffer overflow, versus the buffer size of each sensor.
FIGURE 12. Average packet delay versus the buffer size of each sensor.
VIII. CONCLUSION
In this paper, we design a mix-cost parameter, which consists
of the energy cost and the QoS cost, to jointly measure
both the energy efficiency and the QoS effectiveness of the
WBAN system. Based on the mix-cost parameter, a resource
allocation problem is formulated to optimize the transmission
rates, the transmission power and the time slots for each
sensor to minimize the weighted sum mix-cost for improving
the performances of the WBAN system. Considering the high
complexity of the resource allocation problem, a greedy sub-
optimal resource allocation scheme is carefully designed with
an acceptable time complexity for the hub. In addition, a sen-
sor state evaluation method with low complexity is designed
to dynamically evaluate the sensor state by each sensor, which
is based on both the buffer queue state and the packet delay
state, and it is introduced to the resource allocation scheme
to decide when applying for the resource re-allocation by the
hub to improve both the short-term and the long-term QoS
performance. The simulation results demonstrate that the
resource allocation problem with the sensor state evaluation
method is energy efficient and QoS effective, and the greedy
sub-optimal resource allocation scheme with a lower time
complexity also has a similar performance to the optimal
resource allocation problem.
VOLUME 5, 2017 20775
Z. Liu et al.: Buffer-Aware RA Scheme With Energy Efficiency and QoS Effectiveness in WBANs
REFERENCES
[1] A. Hussain, R. Wenbi, A. L. da Silva, M. Nadher, and M. Mudhish, ‘‘Health
and emergency-care platform for the elderly and disabled people in the
Smart City,’’ J. Syst. Softw., vol. 110, pp. 253–263, Dec. 2015.
[2] L. Catarinucci et al., ‘‘An IoT-aware architecture for smart healthcare
systems,’’ IEEE Internet Things J., vol. 2, no. 6, pp. 515–526, Dec. 2015.
[3] H. Moosavi and F. M. Bui, ‘‘Optimalrelay selection and power control with
quality-of-service provisioning in wireless body area networks,’’ IEEE
Trans. Wireless Commun., vol. 15, no. 8, pp. 5497–5510, Aug. 2016.
[4] B. Liu, Z. Yan, and C. W. Chen, ‘‘Medium access control for wireless body
area networks with QoS provisioning and energy efficient design,’’ IEEE
Trans. Mobile Comput., vol. 16, no. 2, pp. 422–434, Feb. 2017.
[5] M. Salayma, A. Al-Dubai, I. Romdhani, and Y. Nasser, ‘‘Wireless body
area network (WBAN): A survey on reliability, fault tolerance, and tech-
nologies coexistence,’’ ACM Comput. Surv., vol. 50, no. 1, p. 3, 2017.
[6] Z. Liu, B. Liu, C. Chen, and C. W. Chen, ‘‘Energy-efficient resource
allocation with QoS support in wireless body area networks,’’ in Proc.
IEEE Global Commun. Conf. (GLOBECOM), Dec. 2015, pp. 1–6.
[7] E. Reusens et al., ‘‘Characterization of on-body communication channel
and energy efficient topology design for wireless body area networks,’’
IEEE Trans. Inf. Technol. Biomed., vol. 13, no. 6, pp. 933–945, Nov. 2009.
[8] R. D’Errico and L. Ouvry, ‘‘A statistical model for on-body dynamic
channels,’’ Int. J. Wireless Inf. Netw., vol. 17, nos. 3–4, pp. 92–104, 2010.
[9] M. Quwaider, J. Rao, and S. Biswas, ‘‘Body-posture-based dynamic link
power control in wearable sensor networks,’’ IEEE Commun. Mag.,vol. 48,
no. 7, pp. 134–142, Jul. 2010.
[10] K. S. Deepak and A. V. Babu, ‘‘Optimal packet size for energy efficient
WBAN under m-periodic scheduled access mode,’’ in Proc. 20th Nat. Conf.
Commun. (NCC), Feb./Mar. 2014, pp. 1–6.
[11] A. Samanta, S. Bera, and S. Misra, ‘‘Link-quality-aware resource alloca-
tion with load balance in wireless body area networks,’’ IEEE Syst. J., 2015,
doi: 10.1109/JSYST.2015.2458586.
[12] Y. He, W. Zhu, and L. Guan, ‘‘Optimal resource allocation for perva-
sive health monitoring systems with body sensor networks,’’ IEEE Trans.
Mobile Comput., vol. 10, no. 11, pp. 1558–1575, Nov. 2011.
[13] S. Xiao, A. Dhamdhere, V. Sivaraman, and A. Burdett, ‘‘Transmission
power control in body area sensor networks for healthcare monitoring,’’
IEEE J. Sel. Areas Commun., vol. 27, no. 1, pp. 37–48, Jan. 2009.
[14] S. Kim and D.-S. Eom, ‘‘Link-state-estimation-based transmission power
control in wireless body area networks,’’ IEEE J. Biomed. Health Inform.,
vol. 18, no. 4, pp. 1294–1302, Jul. 2014.
[15] Z. Zhang, J. Huang, H. Wang, and H. Fang, ‘‘Power control and local-
ization of wireless body area networks using semidefinite program-
ming,’’ in Proc. 2nd Int. Symp. Future Inf. Commun. Technol. Ubiquitous
HealthCare (Ubi-HealthTech), May 2015, pp. 1–5.
[16] A. H. Sodhro, Y. Li, and M. A. Shah, ‘‘Energy-efficient adaptive trans-
mission power control for wireless body area networks,’’ IET Commun.,
vol. 10, no. 1, pp. 81–90, 2016.
[17] X. Zhou, T. Zhang, L. Song, and Q. Zhang, ‘‘Energy efficiency optimiza-
tion by resource allocation in wireless body area networks,’’ in Proc. IEEE
79th Veh. Technol. Conf. (VTC Spring), May 2014, pp. 1–6.
[18] Z. Liu, B. Liu, C. Chen, and C. W. Chen, ‘‘An energy-efficient and
QoS-effective resource allocation scheme in WBANs,’’ in Proc. IEEE
13th Int. Conf. Wearable Implant. Body Sensor Netw. (BSN), Jun. 2016,
pp. 341–346.
[19] Z. Liu, B. Liu, and C. W. Chen, ‘‘Buffer-aware and QoS-effective resource
allocation scheme in WBANs,’’ in Proc. IEEE 18th Int. Conf. e-Health
Netw., Appl. Services (Healthcom), Sep. 2016, pp. 1–6.
[20] D. B. Smith, L. W. Hanlen, and D. Miniutti, ‘‘Transmit power control
for wireless body area networks using novel channel prediction,’’ in Proc.
IEEE Wireless Commun. Netw. Conf. (WCNC), Apr. 2012, pp. 684–688.
[21] E. Afifi et al.,IEEE Standard for Local and Metropolitan
Area Networks—Part 15.6: Wireless Body Area Networks, IEEE
Standard 802.15.6-2012, IEEE 802.15 Task Group 6, 2012.
[22] A. Boulis, D. Smith, D. Miniutti, L. Libman, and Y. Tselishchev, ‘‘Chal-
lenges in body area networks for healthcare: The MAC,’’ IEEE Commun.
Mag., vol. 50, no. 5, pp. 100–106, May 2012.
[23] L. Lin, C. Yang, K. J. Wong, H. Yan, J. Shen, and S. J. Phee,
‘‘An energy efficient MAC protocol for multi-hop swallowable body sensor
networks,’’ Sensors, vol. 14, no. 10, pp. 19457–19476, 2014.
[24] L. Lin, K.-J. Wong, S.-L. Tan, and S.-J. Phee, ‘‘Asymmetric multi-
hop networks for multi-capsule communications within the gastrointesti-
nal tract,’’ in Proc. 6th Int. Workshop Wearable Implant. Body Sensor
Netw. (BSN), Jun. 2009, pp. 82–86.
[25] D. Gross, J. F. Shortle, J. M. Thompson, and C. M. Harris, Fundamentals
of Queueing Theory. Hoboken, NJ, USA: Wiley, 2013.
[26] A. Goldsmith, Wireless Communications. Cambridge, U.K.: Cambridge
Univ. Press, 2005.
[27] K. Yazdandoost and K. Sayrafian, Channel Model for Body Area Network
(BAN), IEEE Standard P802.15-08-0780-09-0006, IEEE 802.15 Working
Group Document, 2009.
[28] M. Nabi, M. Geilen, and T. Basten, ‘‘MoBAN: A configurable mobility
model for wireless body area networks,’’ in Proc. 4th Int. ICST Conf.
Simulation Tools Techn. (ICST), 2011, pp. 168–177.
[29] Q. Guo, B. Liu, and C. W. Chen, ‘‘A two-layer and multi-strategy frame-
work for human activity recognition using smartphone,’’ in Proc. IEEE Int.
Conf. Commun. (ICC), May 2016, pp. 1–6.
ZHIQIANG LIU received the B.S. degree in elec-
trical engineering from the University of Science
and Technology of China, Hefei, China, in 2013,
where he is currently pursuing the Ph.D. degree
in electrical engineering. His research interests
include resource allocation, energy-saving, and
quality of service guarantee in wireless body area
networks.
BIN LIU received the B.S. and M.S. degrees
in electrical engineering from the University of
Science and Technology of China, Hefei, China,
in 1998 and 2001, respectively, and the Ph.D.
degree in electrical engineering from Syracuse
University, Syracuse, NY, USA, in 2006. He is cur-
rently an Associate Professor with the School of
Information Science and Technology, University
of Science and Technology of China. His research
interests are signal processing and communica-
tions in wireless sensor and body area networks.
CHANG WEN CHEN (F’04) is a Professor of
Computer Science and Engineering with the Uni-
versity at Buffalo, State University of New York,
Buffalo, NY, USA. Previously, he was the Allen S.
Henry Endowed Chair Professor with the Florida
Institute of Technology from 2003 to 2007, a Fac-
ulty Member with the University of Missouri—
Columbia from 1996 to 2003 and with the Uni-
versity of Rochester, Rochester, NY, USA, from
1992 to 1996. He is a Fellow of the International
Society for Optics and Photonics (SPIE). He and his students have received
eight best paper awards or best student paper awards and have been
placed among best paper award finalists many times. He was a recipi-
ent of the Sigma Xi Excellence in Graduate Research Mentoring Award
in 2003, the Alexander von Humboldt Research Award in 2009, and the
SUNY-Buffalo Exceptional Scholar—Sustained Achievements Award
in 2012. He has been an Editor-in-Chief of the IEEE TRANSACTIONS ON
MULTIMEDIA since 2014. He has also served as the Editor-in-Chief of the IEEE
TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY from 2006 to
2009 and an Editor of the PROCEEDINGS of the IEEE, the IEEE TRANSACTIONS
ON MULTIMEDIA, IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, IEEE
JOURNAL ON EMERGING AND SELECTED TOPICS in CIRCUITS AND SYSTEMS, and IEEE
MULTIMEDIA MAGAZINE.
20776 VOLUME 5, 2017
