You are on page 1of 9

Biomedical Signal Processing and Control 41 (2018) 100–108

Contents lists available at ScienceDirect

Biomedical Signal Processing and Control


journal homepage: www.elsevier.com/locate/bspc

A new priority-based dynamic protocol for multiple access control


improves data transfer rates in WBANs
Sérgio Ricardo de Jesus Oliveira ∗ , Alcimar Barbosa Soares
Biomedical Engineering Lab, Faculty of Electric Engineering, Federal University of Uberlândia, Uberlândia, MG, Brazil

a r t i c l e i n f o a b s t r a c t

Article history: Even with the advances made in the so-called Body Area Networks (BAN), there is still a distinct and
Received 9 February 2017 growing demand for technology that allows for the development of applications involving high sensor
Received in revised form 28 October 2017 density and signals gathered at very distinct sampling rates, which impose non-trivial demands on the
Accepted 19 November 2017
Medium Access Control (MAC) protocol for the management of wireless systems (WBANs). In this article,
the authors present a new medium access protocol for WBANs, which is based on a priority model
Keywords:
that dynamically alters the reading sequence of the network nodes. The proposed protocol, referred
Wireless body area networks
to as MAR-PC (Multiple Access with Reserve and Priority Control), was developed by taking as a base
WBAN
MAC
the Polling protocol. It is capable of defining the priority of the sensors in the network based on their
Priority based protocol sampling rates as well as on an estimate of the volume of pending data to be read in each sensor. Through
this strategy, the central node, even though it is not in constant communication with the sensors, is
capable of dynamically evaluating the time limit for the interrogation of a node, thus avoiding loss of
data. The performance of the MAR-PC was compared to that of a standard Polling protocol and to other
MACs published in recent literature. The results show a significant improvement in the throughput of
the WBAN in comparison with competing protocols, especially when the sensor nodes operate at distinct
and considerably different sampling rates.
© 2017 Elsevier Ltd. All rights reserved.

1. Introduction Multiple Access with Collision Avoidance) access method, defined


in the MAC layer, is a multi-network access method in which the
Wireless Body Area Networks (WBANs) [1,2] have been used in nodes seek to avoid collisions by transmitting data only when the
a variety of applications and scenarios found in the areas of medi- channel is free. Such a strategy generates a dispute between the
cal healthcare, health information and sports, to name a few [3–5]. network nodes over the RF (Radio Frequency) channel [3]. As a
Despite the benefits, depending on the protocol used to control result, in situations such as those previously described, the sensors
data flow and on the demands and the complexity of the network, that operate with a higher sampling rate contribute to an increased
the efficiency of data transfer between the sensors and the control probability of data collision between the nodes, thus degrading the
node can be severely compromised. For instance, in networks with performance of data transfer in proportion to the demand of each
sensor nodes acquiring data at widely different sampling rates, it is sensor [13]. Schedule-based protocols, such as Polling and TDMA
important that the central node deals with those nodes operating at (Time Division Multiple Access), would also have their efficiency
higher sampling rates more often than the other nodes, in order to compromised in similar situations. Because the TDMA protocol
prevent information loss [3,6,7]. Such a situation generates impor- allocates time slots for each node to transmit its data to the cen-
tant consequences that compromise the efficiency of the main tral node [14], nodes with higher sampling rate will have greater
MAC (Medium Access Control) protocols used by the IEEE stan- difficulty to transmit their information (once they complete a trans-
dards 802.15.1 [8,9], 802.15.4 [10] and 802.15.6 [11,12], commonly mission, they must wait until all nodes use their slots to receive the
employed in WBAN implementations. The CSMA/CA (Carrier Sense next transmission slot) [15]. A limitation of that type is also seen
in the Polling protocol, in which the central node interrogates, in a
previously defined sequence, each sensor node from the WBAN for
data reading [16,17].
∗ Corresponding author at: Faculty of Electrical Engineering Federal University
As an alternative to problems of a similar nature, in which
of Uberlândia Avenida João Naves de Ávila, 2121 Bairro Santa Mônica 38408-144,
Uberlândia, MG, Brazil.
specific nodes require intense medium access, while others only
E-mail address: srjesus@ufu.br (S.R.d.J. Oliveira). demand sporadic access and are idle most of the time, Sharon et al.

https://doi.org/10.1016/j.bspc.2017.11.013
1746-8094/© 2017 Elsevier Ltd. All rights reserved.
S.R.d.J. Oliveira, A.B. Soares / Biomedical Signal Processing and Control 41 (2018) 100–108 101

Fig. 1. Diagram of the network topology used for the implementation of the MAR-PC protocol.

[16] proposed a MAC protocol based on two RF communication


channels and where a sensor node resides either in an active or in
an idle ring, depending on if it has data to be transmitted or not. The
strategy implemented by the authors explores the so-called Cap-
ture Effect [18,19], which eventually defines a certain priority of
the nodes in the active ring over those in the idle ring. This has the
effect of rendering higher transmission rates to active nodes and, as
such, shortening medium access delays. However, an essential pre-
requisite for this protocol is that the relative distance between the
central node and the furthest node must to be fixed, otherwise the
performance of the MAC is highly compromised. Such restriction
imposes severe limitations when one tries to apply this protocol to
the design of WBANs, since the distances between the sensors and
central nodes are not necessarily fixed.
Rezvani et al. [20] proposed a MAC protocol based on TDMA,
where the nodes are kept in idle state while there is no data to be
transmitted and are awaken before the arrival of the next pack from Fig. 2. State diagram for the main application running on the host computer.
the central node. The authors report an increase in the transfer-
ence rate, when compared to the IEEE 802.15.4 protocol. However,
each sensor based on its sampling rate and the volume of data wait-
when high sampling rates and real-time monitoring are required,
ing to be read on each one. We hypothesize that such a strategy can
the inclusion of idle states would be severely restricted, increasing
reduce the probability of data loss in situations where the various
the possibility of data collision.
nodes of a network operate at diverse and very different sampling
Other authors proposed WBAN models that are based on RF
rates.
transceivers, such as the nRF24L01+ and nRF24E1 (Nordic Semi-
conductor − [21,22]), which operate outside of the IEEE standards
[23]. Although they bring an interesting approach to the problem, 2. Materials and methods
most of the attempts using such a strategy find similar obstacles
regarding the occurrence of collisions, with the same limitations as 2.1. WBAN topology
for the CSMA/CA, described earlier.
In general, the techniques developed to improve throughput in Fig. 1 presents the network model, in star topology (currently,
WBANs have been focused on strategies to improve the efficiency of one of the most used network topologies for WBANs [1,16,17]),
communication between the nodes of a WBAN [7,14,16,17,23–25]. onto which the MAR-PC was implemented for performance evalua-
However, the diversity of signals and sampling rates in more tion and comparisons with other MACs. The RF communication will
sophisticated WBANs that are dedicated, for instance, to health- occur exclusively between the central node and the sensor nodes,
care and sports, are usually not taken into account when designing where the latter will only transmit to answer a request from the
communication protocols, leading to failure or data loss. central node.
In this article, we propose a new medium access protocol for
WBANs based on a priority model, which dynamically alters the 2.1.1. Computer
sequence for reading the nodes from the network (sensor nodes). A standard IBM-PC (Intel i7, 8GBytes of Ram, Windows 7 64bit)
Our model, hereinafter referred to as MAR-PC (Multiple Access with was used to run an application for controlling the entire operation
Reserve and Priority Control), is capable of defining the priority of of the WBAN and to read the data produced by the sensor nodes.
Fig. 2 shows the state diagram for the application.
102 S.R.d.J. Oliveira, A.B. Soares / Biomedical Signal Processing and Control 41 (2018) 100–108

Fig. 4. State diagram of the tasks executed by the sensor nodes.


Fig. 3. State diagram of the tasks executed by the central node.

Table 1
The application remains in the standby, waiting for a request The format of the communication packages between the central node and the sensor
from the user, who can choose from: configuring the network nodes, for the MAR-PC protocol.
(number of sensor nodes, sampling rates and the amount of data Communication direction Package format
to be collected); reset the WBAN (resets the central node and all
Central node to Sensor’s address Command
sensors, cancelling any process in execution); and start data acqui- sensor nodes 8 bits 8 bits
sition. During data acquisition all the data generated by the sensor
nodes will be read by the application (‘Read Data WBAN’), shown Sensor nodes to Sensor’s address FIFOstep #Bytes Data
on the interface and stored for later use. Data acquisition will stop central node 1 Byte 4 Bits 1 Byte n Bytes
(‘Stop Acquisition’) when all data required has been collected or
upon user’s request.
(data is stored in a FIFO (First In, First Out) until transmitted to
2.1.2. Central node the central node); ‘Stop Acquisition’ – stops the data acquisition
The central node was implemented combining the nRF24L01+ process; ‘Send Payload’ – send data to the central node (payload
RF transceiver (Nordic Semiconductor – ultra low power 2Mbps is assembled and transmitted with a new data from the FIFO or
RF transceiver for the 2.4 GHz ISM (Industrial, Scientific and Med- with the same data as sent in the previous transmission, in case of
ical) band – [21]) and a board equipped with the microcontroller communication failure).
Atmel SAM3 × 8E ARM Cortex-M3 (32bit data bus, clock frequency
of 84 MHz, SPI port for communication with the RF transceiver and 2.2. Multiple access with reserve and priority control – MAR-PC
two USB ports for communication with the central computer). The
state diagram of the main tasks executed by the microcontroller is The main feature of MAR-PC protocol is a dynamic alteration of
shown in Fig. 3. the sequence for querying the sensor nodes, according to the vol-
The operation of the central node is divided into ‘Command ume of data that needs to be transferred. The sensor that contains
Mode’ (execute commands sent by the host application/computer: the highest volume of data will receive the highest reading priority
‘Config Sensor’, ‘Reset Sensor’, ‘Start Acquisition’ and ‘Stop Acquisi- from the central node. The quantity of available data for reading on
tion’) and ‘Scan Mode’ (read data generated by the sensors – ‘Read each sensor will depend upon the sampling rate of the signal and
Sensor’, and send them to the host application). In ‘Scan Mode’ the the possible communication errors with the central node. The data
central node continuously accesses all sensor nodes to request and generated on the sensor node is stored in a FIFO. At each interroga-
read the data, according to the sequence defined by the MAC proto- tion from the central node, the sensor node should send a data pack
col. Data read from the sensor nodes are sent to the host computer from the FIFO, together with its fill level. A vector, maintained in the
via USB port (not shown in the diagram for simplicity). central node, will be used to store the priorities of all nodes in the
network. The value contained within each element of this vector
2.1.3. Node i – sensor node represents an estimate for the time remaining until the overflow
The sensor nodes of the WBAN were constructed using the of the FIFO from the corresponding sensor node. The overflow time
transceiver nRF24LE1 (Nordic Semiconductor). This device is sim- estimate is based on the sensor’s sampling rate and on the FIFO fill
ilar to the nRF24L01+, used for the central node, but includes a level.
microcontroller compatible with the 8051 from the Intel Corpo-
ration [22]. The control of the biomedical sensor, as well as the 2.2.1. Package formats
tasks associated with communication with the central node, will The information that flows between the central node and the
be implemented onto this device. The state diagram of the main sensors is organized into two different package formats, where one
tasks executed by the nRF24LE1 is shown in Fig. 4. is for sending interrogations from the central node and the other
Initially, the sensor node is in standby waiting the arrival of a for responses from the sensors, as shown in Table 1.
command from the central node. Upon receiving the command, In the interrogation process, the central node sends the address
the sensor node will execute the tasks required, as follows: ‘Reset of the sensor node and the command that it should execute. The
Sensor’ – restarts the application software embedded in the sensor sensor node will send its own address, followed by the fill level of
node’s microcontroller; ‘Config Sensor’ – sets the new sampling the FIFO (FIFOstep – sent in a quantity of 4 bits, with the objective of
rate for the sensor, according to the value received from the central minimizing the volume of data transmitted in the RF package), the
node; ‘Start Acquisition’ – starts a new data acquisition process number of data bytes that are being sent (#Bytes) and the actual
S.R.d.J. Oliveira, A.B. Soares / Biomedical Signal Processing and Control 41 (2018) 100–108 103

data requested by the central node. The number of data bytes to be


sent by the sensor is variable, as it depends on the quantity of data
available for transmission.

2.2.2. Network scan and priority management


The selection of the sensor node to be consulted by the cen-
tral node will be given through the priority level of each node.
As described previously, for the MAR-PC, a priority vector (PRT)
is maintained by the central node and contains an estimate of the
time remaining until the overflow of the FIFO of each sensor node.
Eqs. (1–4) describe the models for estimating the priority level of
each sensor to be interrogated by central node.
During the network configuration, the initial priority values for
each sensor are defined according to Eq. (1).

PRT (x)0 = sizeFIFO(x) ∗ T s (x) (1)

where,

• 0 ≤ × ≤ number of sensor nodes − 1;


• sizeFIFO(x): size of the FIFO of sensor x (one unit corresponds to
the number of bytes necessary for storing a sample of the signal
digitalized by the sensor);
• Ts : sampling period of the sensor x, in milliseconds;
• PRT(x)0 : priority level of the sensor node x at the start of the
network operation, in milliseconds.

Note that, in this manner, the level of priority for each sensor is
in fact an estimate of the remaining time in which a FIFO overflow
can occur. Therefore, the sensor with the least PTR(x) will be given
highest priority and successively downwards. Fig. 5. Flowchart showing the main steps for reading data and dynamically update
During the network operation, the priority of a sensor can be the priorities of the sensor nodes in the WBAN.
adjusted after it is read by the central node and immediately before
the interrogation of any sensor node of the network, whichever it • x = 0, 1, . . ., number of sensor nodes − 1;
may be. • PRT(x)previous : current priority level of the sensor node x, in mil-
Every time the central node interrogates and finishes the data liseconds.
reading of a sensor node x, its priority will be adjusted according to • Tquery : elapsed time since the last interrogation by the central
the new quantity of free positions in the FIFO, according to Eq. (2). node, in milliseconds.
• PRT(x)new : new priority level of the sensor node x, in milliseconds,
PRT (x)new = newGap(x) ∗ T s (x) (2)
updated immediately before the selection of a new sensor node,
where, whichever that may be.

• x = 0, 1, . . ., number of sensor nodes − 1; Fig. 5 summarizes the main steps for reading data and updat-
• newGap(x): an estimate of the number of free positions in the ing the priorities of the sensor nodes. As previously described, at
FIFO of the sensor x, calculated based on the fill level of the FIFO, any moment, the highest priority will be given to that node with
sent by the sensor during its interrogation (FIFOstep ); the lowest value on the priority vector PRT, since it is that node
• Ts : sampling period of the sensor x, in milliseconds: which has the highest risk of data loss through the overflow of its
• PRT(x)new : new priority level of the sensor node x, in milliseconds, FIFO. Once the highest priority node has been selected, the central
updated after its reading. node interrogates it, to send the data contained within its FIFO. The
volume of data removed from the FIFO will correspond to the maxi-
As seen from Table 1, only 4 bits are used to transfer the fill level mum capacity of the payload for the RF package. Should the central
of the FIFO (FIFOstep ) in order to minimize the volume of data trans- node receive the data correctly, these will be sent to the computer.
mitted in the RF package. Hence, FIFOstep contains values between Following on, the priority level of the sensor that has been dealt
0 and 15, representing the filling of the FIFO in 16 levels. Therefore, with will be redefined in the PRT vector, as in Eq. (2), once the fill
each FIFOstep level will represent a quantity of occupied positions level of its FIFO has been modified. After completion of this process,
in the FIFO equal to sizeFIFO(x)/(24 -1). Thus, the actual quantity of or in the case where the reading operation has failed, the central
free positions in the FIFO can be estimated by the central node, node will adjust the priority of every sensor in the network, as in
according to Eq. (3). Eq. (4). The procedure can therefore be reinitiated for the selection
and reading of the next sensor.
newGap(x) = sizeFIFO(x) − (FIFOstep ∗(sizeFIFO(x)/15)) (3)

In addition, during the operation of the network and immedi- 2.3. Configuration and execution
ately before the interrogation of a sensor, whichever that may be,
all the PRT elements will be updated, according to Eq. (4). The configuration of the network is defined on the host applica-
bPRT (x)new = PRT (x)previous − Tquery (4) tion running on the computer. The information that refers to each
sensor node to be used in the WBAN (size of FIFO and sampling rate)
where, is sent to the central node, which puts together its internal list of
104 S.R.d.J. Oliveira, A.B. Soares / Biomedical Signal Processing and Control 41 (2018) 100–108

nodes to be consulted. In addition, the central node will define the implemented for these experiments), while for the remaining six
initial priority of each sensor node, as previously described, and sensors the sampling rate was set at 10 Hz. The value of 10 Hz was
then send the sampling rate specified for each one of them. As of chosen for the sake of convenience, as a means to characterize an
this point, the central node waits for the “Start Acquisition” com- expressive difference between the sampling rates. The number of
mand that must be sent by the host application. Once the command sensors operating at their maximum rate was defined empirically,
is received, the central node will send a broadcast message to all in a manner that none of the protocols under test presented oper-
the sensor nodes, requesting that these immediately start signal ational failures through overflow of the FIFO, on any of the sensor
sampling. A broadcast message is required so that data collection nodes.
from the various sensors occurs in a synchronous fashion. Experiment 3: Point of overflow failure
This experiment aims to compare the performance of the proto-
2.4. Experimental protocol cols, through the systematic increase in the number of sensor nodes
operating at their maximum sampling rates (TSmax = 500 Hz) – with
2.4.1. WBAN implementation the remaining nodes operating at TS = 10 Hz, until the moment that
The WBAN model presented in Fig. 1 was implemented with the protocols present a failure through overflow of the FIFO from
9 sensor nodes, 1 central node and a computer. The sensor nodes one of the sensor nodes.
(containing the transceiver nRF24LE1) and the central node (con- The results will be presented in graph form, by pointing out the
taining the transceiver nRF24L01 + ) were configured to operate on temporal evolution of the fill percentage of the FIFO from the sensor
the RF channel 90 (2.490 GHz) in the ISM band, in order to avoid nodes, according to Eq. (5).
conflicts with WiFi networks [26].
The FIFO memory of each sensor node was implemented and FIFOstep% = trunc(FIFOstep ∗ (100/15)) (5)
managed by the software executed on the microcontroller embed-
Experiment 4: Network throughput with different number of
ded on the IC nRF24LE1. Due to the limitations of the device, the
active sensors, under the control of the MAR-PC protocol.
capacity of the FIFO was set to 120 bytes allowing for the storage
This experiment aims to evaluate the performance of the MAR-
of up to 60 two-byte data samples (sizeFIFO(x)).
PC in terms of network throughput, under heavy traffic and variable
number of active sensor nodes. To do so, the following tests were
2.4.2. Experiments and performance evaluation
performed:
One of the most used topologies for the implementation of
WBANs is the star topology, on which the sensors transmit their
1) The sampling rate of all nodes was set to the same and maximum
data to a central node, as shown in Fig. 1. For this type of network,
possible value, so that they could all operated simultaneously
the Polling protocol presents itself as a natural and adequate option
and without loss of data. Next, we gradually increased the num-
[8,12,23,24,25,27]. Therefore, in this study, the performance of the
ber of active sensors and measured the total network throughput
MAR-PC was first compared to that of the standard Polling protocol,
at each step;
which has also been chosen in various publications as the start-
2) Similarly as above, the number of active sensor nodes was also
ing point and benchmarking for new strategies involving WBAN
gradually increased in this test. However, the sampling rate of
protocols [8,12,16,17,24,25,28]. The standard Polling protocol was
the sensors was not constant, but set, in each step, to the maxi-
implemented on the same physical network as that on which the
mum possible value so that the network would operate without
MAR-PC was implemented. Next, the performance of the MAR-PC
loss of data (maximum throughput limit as a function of the
was also compared against the results obtained from recent studies
number of sensor nodes).
published in the literature [29–33].
The experiments were designed to test the performance of the
protocols under the following conditions: i) Sensor nodes operating The results will also be compared with those reported in recent
with sampling rates such that the operational limit of the network relevant studies [29–33].
was not reached, or be it, there was no data loss caused by over- All the experiments were performed by taking into considera-
flow from any of the sensor FIFOs; ii) Taking the network beyond tion a fixed acquisition time, equal to 10 s, which was also defined
its operational limit, which thus causes data loss through overflow for the sake of convenience and considered satisfactory for the
of the sensor FIFOs; iii) Maximum data transference rate (through- evaluations required for this work.
put) with the network operating with a variable number of active
sensors, under the control of the MAR-PC protocol. 3. Results
Experiment 1: All the sensor nodes operating with the same
sampling rate, at the network operational limit Experiment 1: All the sensor nodes operating with the same sam-
This first experiment was designed to evaluate the differences pling rate, at the network operational limit
between the Polling and MAR-PC protocols with the network oper- The maximum sampling rate (TSmax ) was defined empirically, by
ating at full capacity, where all the nodes are sampling signals at increasing the sampling rate of every sensor equally, until the point
the same, and maximum possible, rate (TSmax ). The maximum sam- data loss was detected. The limit encountered was TSmax = 280 Hz.
pling rate for the sensors was defined empirically, as described in Fig. 6 shows the percentage changes in the FIFO fill level for sensor
the Results section. node 1 throughout the experiment (the behavior of all the remain-
Experiment 2: Sensor nodes operating at distinct frequencies and ing nodes was similar).
within the network operational limit In this situation, the FIFO fill level of the sensor nodes reached
In this experiment, we intend to evaluate the performance of a maximum of 20% of their maximum capacity when the network
the protocols when controlling a WBAN composed of sensor nodes operated under the control of both MAR-PC and Polling. The MAR-
working at different and very distinct sampling rates. For instance, PC resulted in oscillations in the FIFO fill level between 13% and 20%.
some nodes would be sampling signals at much higher rates than Once the acquisition was brought to close, the fill level of the FIFOs
the others, thus contributing to the degradation of the transference returned to zero, as they were emptied by the last interrogations
rate of the network, in proportion to the demand of each sensor. from the central node.
To this end, the sampling rate of three of the sensors (1, 3 and Experiment 2: Sensor nodes operating at distinct frequencies and
5) was adjusted to its maximum value (500 Hz, for the hardware within the operational limit of the network
S.R.d.J. Oliveira, A.B. Soares / Biomedical Signal Processing and Control 41 (2018) 100–108 105

Fig. 6. FIFO fill level, in percent, for sensor node 1. All sensor nodes operating at Fig. 8. FIFO fill level, in percent, for sensors 1 and 2. Four sensor nodes (1, 3, 5 and
TSmax = 280 Hz. 7) operating at TSmax = 500 Hz and five (2, 4, 6, 8 and 9) operating at TS = 10 Hz.

Fig. 7. FIFO fill level, in percent, for sensor nodes 1 and 2. Sensor nodes 1, 3 and 5
operating at TSmax = 500 Hz; Sensor nodes 2, 4, 6, 7, 8 and 9 operating at TS = 10 Hz.

Fig. 7 presents the percentage changes in the fill levels of the


FIFOs for the sensor nodes 1 (TSmax = 500 Hz) and 2 (TS = 10 Hz) for
both protocols under test (Polling and MAR-PC). Nodes 5 and 9,
also operating at TSmax = 500 Hz showed behavior similar to that of
node 1. In the same manner, all the remaining nodes operating at
TS = 10 Hz showed a similar behavior to that found for node 2.
In this situation, with the network operating under the Polling
protocol, the FIFO fill level of the nodes operating at TSmax = 500 Hz
reached peaks that surpassed 30% (green line), while the FIFO of
those operating at TS = 10 Hz remained practically empty during the
whole operation (blue line). In terms of the MAR-PC protocol, the
fill level of the nodes operating at TSmax = 500 Hz was maintained
below 10% over most of the time, showing some peaks of 13%, 20%
and 33% (red line), while the occupation level of the FIFO of the
nodes operating at TS = 10 Hz remained at around 13% (dashed black Fig. 9. FIFO fill level, in percent, for the sensor nodes 1 and 2. MAR-PC: (a) Five sensor
line). nodes (1, 3, 5, 7 and 9) operating at TSmax = 500 Hz and four (2, 4, 6 and 8) operating
Experiment 3: Point of overflow failure at TS = 10 Hz; (b) Six sensor nodes (1, 3, 5, 6, 7 and 9) operating at TSmax = 500 Hz and
three (2, 4 and 8) operating at TS = 10 Hz.
As previously described, this experiment aims at taking the
Polling and MAR-PC protocols to their point of failure, in order to
evaluate the conditions where this occurs for each. With this in
mind, the sensor nodes initiated operation at TS = 10 Hz, except the the sensor nodes. For simplicity, only the curves referring to sensors
sensor node 1, which operated at TSmax = 500 Hz. In this situation, 1 and 2 are shown. The remaining nodes, which operated at equal
as noted through experiment 2, neither of the protocols presented frequencies, presented similar behavior.
data loss by overflow of the FIFO. As of this point, the number As noted, the Polling protocol was the first to fail, as the demand
of nodes operating at TSmax = 500 Hz was gradually increased, one from the network increased. Under control of this protocol, and
node at a time, until the point where one of the protocols under with four of the nine sensors operating at their maximum rate, the
analysis presented a failure (Fig. 8). Following on, the quantity of network broke down after about one second of operation, due to
sensor nodes operating at TSmax = 500 Hz continued to be increased FIFO overflow of the sensor nodes with the highest sampling rate.
to evaluate the behavior of the other protocol, until the point that Furthermore, the failed sensor nodes were unable to recuperate
this too presented a failure (Fig. 9a–b). while the data collecting process was still running. On the other
Fig. 8 presents the first of the experiments from this stage, in hand, the FIFOs of the remaining sensors operating at TS = 10 Hz
which the protocols failed due to FIFO overflow in at least one of remained practically empty during the whole acquisition period.
106 S.R.d.J. Oliveira, A.B. Soares / Biomedical Signal Processing and Control 41 (2018) 100–108

and IEEE 812.15.4 (black line) protocols. To ensure a fair compari-


son with regard to the behavior of the throughput as the number of
sensor nodes increased, we adjusted the operational parameters of
our tests so that the initial throughput of the MAR-PC matched that
of the TraPy-MAC, when only one sensor node was used (4kbps).
This was achieved by setting the sampling rate of the sensors at
250 Hz (2 bytes per sample at 250 Hz).
As seen in Fig. 10, the IEEE 802.15.4 protocol shows an increase in
throughput as the number of sensor nodes of the network increases
up to 7 nodes. From this point on, the throughput will not increase.
On the other hand, the MAR-PC and TraPy-MAC protocols show a
steady increase in throughput as the number of sensors increase,
with a more pronounced increase for the MAR-PC (with nine active
sensors, the MAR-PC shows a throughput of 35.4 kbps, while for the
Fig. 10. Throughput vs number of sensor nodes for the MAR-PC, TraPy-MAC [29] TraPy-MAC it is around 26 kbps).
and IEEE 812.15.4 [32]. Still in Fig. 10, the cyan dashed line shows the maximum
throughput that the MAR-PC can achieve as a function of the num-
For the same situation, the MAR-PC protocol was capable of ber of sensor nodes. It can be noticed an increase in throughput up
safely avoid any possibility of FIFO overflow. The fill level of the to six nodes, after which it remains fairly stable.
FIFOs of the sensor nodes operating at TSmax = 500 Hz remained
around 6%, with some peaks reaching a little more than 30% and
reaching 0% at other moments. The fill level of the FIFO for the 4. Discussion
remaining sensors operating at TS = 10 Hz remained stable at 13%
during the whole acquisition period. According to Rathee et al. [4], one of the most important aspects
As of this point, we tested the limits of the MAR-PC by increasing to be considered in WBAN design is the data transference rate
the number of sensors operating at maximum sampling rate until between the sensor nodes and the central node. To reach this
break down occurred. end, new hardware models [11,28], along with new strategies
Fig. 9a shows the performance of MAR-PC with five nodes oper- for optimizing the communication processes between the net-
ating at TSmax = 500 Hz and four operating at TS = 10 Hz. In the same work nodes [16,17,20,29,30], have been proposed. In this sense,
manner, as with the other experiments, only the curves for two sen- the MAR-PC protocol was designed with the purpose of increasing
sors are shown, since the remaining nodes operating at the same the throughput of WBANs, through the adoption of a model that
sampling rates showed similar behavior. As noted, even with the adjusts priorities dynamically for the interrogation of the sensor
greater number of the sensor nodes operating at their maximum nodes. The results from our experiments demonstrate that such a
rate, the MAR-PC protocol is capable of sustaining the operation strategy, when compared to the Polling protocol and to other rel-
without failure. This is noted even when the fill levels reach peaks of evant studies published in recent literature, presents considerable
around 52% for those nodes operating at maximum rate, while the advantages for WBANs, especially when the sensor nodes operate
FIFO of the remaining nodes, operating at TS = 10 Hz, are maintained at variable and very different sampling rates from one another.
at a stable fill level of 13%. Experiments 1 and 2 were designed to evaluate the performance
In Fig. 9b, the performance of the MAR-PC is shown with six of the MAR-PC and Polling protocols under ideal operation (with-
nodes operating at 500 Hz and three at 10 Hz. The behavior of the out data loss due to FIFO overflow on the sensor nodes). The results
remaining sensors, all operating at the same sampling rates are highlighted in Figs. 6 and 7 demonstrate that, in these situations,
similar to those shown in the figure for nodes 1 and 2. As seen, when both MAR-PC and Polling protocols present a similar and satisfac-
six of the nine sensor nodes operate at sampling rates of 500 Hz, the tory performance. The oscillations in the FIFO fill level when the
operational limit of the MAR-PC is reached for this network. In the network operates under control of the MAR-PC (Fig. 6) illustrate the
figure, the nodes that operate at TS = 10 Hz are far from their limits consequences of the proposed priority model. At each reading of the
(fill level of the FIFO around 13%), while those nodes operating at sensor node, the quantity of data in its FIFO decreased around 13%
the maximum rate show overflow of their FIFO after only 0.8 s, with on average. When the reading is finished, the central node alters
oscillations between 80% and 100% during the acquisition period. the priority for that node in proportion to the quantity of free posi-
Experiment 4: Network throughput with different number of tions of the FIFO (approximately 67%) and the sampling rate of that
active sensors, under the control of MAR-PC protocol. sensor (Eq. (2)). However, while other sensor nodes with higher
The network throughput as a function of the number of active priority on the network are dealt with, the data acquisition pro-
sensor nodes, under the control of the MAR-PC protocol, is shown cess on that initial sensor continues to load its FIFO with new data.
in Fig. 10: i) the sampling rate of all sensors was set to a fixed and As the strategy adopted by the MAR-PC does not interrogate the
maximum possible value, yet allowing for operation without loss nodes systematically, the central node should make an estimation
of data, in any of the nine steps of the test (blue line); ii) in every in order to know which sensor node has the highest FIFO overflow
step, the sampling rate of the active nodes was adjusted to the max- probability. Such estimate is, in fact, an update of the priority value
imum value that allowed for operation without loss of data, for that of all the sensor nodes, which occurs immediately before the selec-
specific step (cyan dashed line). tion of the next node to be interrogated (see Eqs. (4–5)). Therefore,
We also compare the throughput of the MAR-PC against that of when that initial sensor node reaches highest priority, it will once
other MACs recently published in the literature. We have chosen a again be selected for reading. At this moment, it is expected that
well-cited MAC proposed by Ullah et al. [29] as the basis for com- the amount of data in its FIFO has increased in proportion to its
parison. In their paper, the authors compared the performance of sampling rate and to the time that it has run since the last reading.
their proposed protocol (TraPy-MAC) with that of other MACs (PLA- In Fig. 7, one notes that, for the Polling protocol, the sensor
MAC [30], MAC Scheme-1 [31] and IEEE 812.15.4 [32]), showing a nodes operating at higher sampling rates present a higher FIFO fill
superior performance in all cases. As such, in Fig. 10, we superim- level than those operating at lower sampling rates. This is expected,
posed the results described in [29] for the TraPy-MAC (red line) since the Polling protocol always uses the same sequence for read-
S.R.d.J. Oliveira, A.B. Soares / Biomedical Signal Processing and Control 41 (2018) 100–108 107

ing the sensor nodes, independent of their FIFO fill level. In the sors operating at 500 Hz and 6 sensors operating at 10 Hz – hence,
case of the MAR/PC, as the nodes operating under higher sampling this protocol was capable of maintaining average data traffic equal
rates received higher reading priority, their average FIFO fill level to at least 24960 bits/second [(3 sensors*500 Hz + 6 sensors*10 Hz)
was much lower than that presented for the same nodes operat- * 16bits per sample]. However, the MAR-PC was capable of sus-
ing under the control of the Polling protocol. On the other hand, taining the operation of the network with 5 sensors operating at
the nodes operating at lower sampling rates (illustrated by the 500 Hz and 4 sensors operating at 10 Hz. In other words, the MAR-
response of sensor 2) present a stable fill level, at around 13%, PC is capable of maintaining average data traffic equal to at least,
over the whole acquisition period. This occurs due to the condi- 40640 bits/second [(5 sensors*500 Hz + 4 sensors*10 Hz) * 16bits
tions established for this experiment and to the low sampling rate per sample] − an expressive increase of around 62.82%, when com-
of the sensors on those nodes (10 Hz). At each new reading, the pared to the Polling protocol. It needs to be highlighted, once again,
percentage of occupation of their FIFOs was found to be on the that such an improvement only occurs in cases where the net-
same level as the previous reading. Furthermore, we can observe work operates with the sensors sampling signals at distinct rates
that the FIFO fill level of the nodes with the highest sampling rates, between themselves. This in fact is an everyday situation on WBANs
for both protocols, presented oscillations during the data collection − thus attesting to the suitability of the proposed method for the
period. However, the oscillations did not show growth tendencies control of medium access on this type of network.
in the direction of overflow of the FIFOs. This means that, for this Through further analysis of the results obtained from the exper-
experiment, the data production rate from the sensors is always iments, one notes another important advantage of the MAR-PC in
lower than the data download rate by the central node. However, relation to Polling. In a real application for WBAN, there can occur
the differences in behavior between Polling and MAR-PC, with the interferences in RF communication between the nodes, which
MAR-PC showing much lower oscillations, in average, than the impedes the central node from successfully reading the data of a
Polling, is already an indication of the possible benefits of the adop- particular sensor node [15]. In the case of Polling, the sensor node
tion of a priority scheme for defining the reading sequence of the will have to wait until all the nodes have been dealt with, in order
sensor nodes on a WBAN. for the central node effectuate a new reading − request the retrans-
Experiment 3 was designed to investigate the limits of operation mission of the lost data pack. This situation can increase the FIFO
for the protocols under evaluation, and as a consequence, evaluate fill level of the failed sensor, possibly to the point of overflow [9].
which of these would support higher demands from the nodes for In the case of MAR-PC, this situation also causes an increase in
medium access. To this end, all sensor nodes were initiated operat- the FIFO fill level of the sensor node, for which the reading was
ing at TS = 10 Hz, with the exception of only one, which operated at not performed successfully. However, as the central node was not
TSmax = 500 Hz. Gradually, the number of sensor nodes operating at capable of collecting information from the sensor, the updating of
TSmax was increased, until the point where the protocols presented its priority that would happen after the reading will not occur (as
a failure due to FIFO overflow of at least one of the sensors. Fig. 8 shown in Fig. 5). Nevertheless, the time estimate to the overflow
illustrates the point at which the first protocol is unable to manage of the FIFO of that specific node will continue to be reduced every
the data traffic from the network. The Polling protocol failed when time the central node defines the next sensor node to be dealt with
four of the nine sensors of our experimental WBAN operated at (Eq. (4)). As the protocol implemented on the central node also
500 Hz. In this situation, the central note was not able to deal with waits for the reading status of each sensor, it is able, at the moment
all the data produced by the sensor nodes, and started losing data when a particular failed node assumes maximum reading priority,
due to FIFO overflow. In this same situation, the MAR-PC continued to request the resending of data still not received. This is possible,
operating normally and far from a condition that would cause FIFO as the protocol implemented on the sensor nodes maintains all the
overflow, which demonstrates, once again, the higher robustness transmitted data until the receiving of the next reading request. In
of the priority model, when compared to the Polling protocol. other words, reading failures caused by some sort of interference
Fig. 9 illustrates the behavior of the MAR-PC protocol as the in RF communication can also be minimized through the MAR-PC
demands from the sensors for medium access are increased to lev- strategy.
els above those that caused failure of the Polling protocol. Fig. 9a Finally, experiment 4 was designed to evaluate the performance
shows that, even under conditions where the majority of the sen- of the MAR-PC in terms of network throughput and to compare it
sors operates at maximum rate, the MAR-PC is capable of sustaining against other relevant studies reported in recent literature. Fig. 10
the operation without failures, even with FIFO fill levels reaching shows the throughput (kbps) of the MAR-PC as a function of the
peaks of around 52% for the nodes with greater demand. The point number of active nodes in the WBAN. The dashed line represents
of failure of the MAR-PC occurred with the network operating with the maximum throughput that can be achieved with different num-
6 nodes at their maximum sampling rate (Fig. 9b). ber of active nodes. To do so, in each step, the sampling rate of the
Although those results are specific to the implemented condi- active nodes was adjusted to the maximum value that allowed for
tions and hardware, in a general sense, one notes that when all operation without loss of data. As it can be noticed, the maximum
the sensor nodes operate at the same sampling frequency, the two throughput of the MAR-PC increases up to six nodes, remaining
protocols present a similar performance. However, when the net- fairly stable afterwards at around 35 kbps. This means that the
work operates with the sensor nodes at very different sampling smaller the number of nodes in the network, the greater the maxi-
rates, the advantages of the proposed protocol become evident, mum sampling rate that can be used by the sensor nodes.
when compared to the Polling protocol. In this situation, as well Fig. 10 also shows a comparison of the performance of the MAR-
as for the specific conditions presented in this study, the MAR-PC PC with that of other MACs (TraPy-MAC [29] and IEEE 802.15.4
was capable of sustaining the network operation with a higher [10]), under the same experimental condition (the sampling rate
number of sensors operating at high sampling rates of 500 Hz, for all sensors was set to a fixed and maximum possible value
while the other nodes operated at much lower sampling rates. that allowed for operation without loss of data). Because the IEEE
This effectively means a significant improvement to the medium 802.15.4 has only 7 slots for data transfer from the sensor nodes
access management. If we evaluate the operational limit of the to the central node, its throughput reaches a limit at that number
developed network, it is noted that the MAR-PC has the capac- of active nodes. Although not shown in the figure, the TraPy-MAC
ity to sustain the network operation with a significant average would reach its throughput limit when the WBAN operates with 14
traffic and without failures for a considerable period of time. The nodes [29]. However, we see in Fig. 10 a consistent superior per-
limit of operation for Polling protocol was reached with 3 sen- formance of the MAR-PC over the TraPy-MAC for operation with
108 S.R.d.J. Oliveira, A.B. Soares / Biomedical Signal Processing and Control 41 (2018) 100–108

up to 9 sensor nodes. We believe that one of the main reasons for References
that lies in the careful design of the MAR-PC RF packets, from both
the central node to the sensor node and from the sensor node to [1] J. Ahmad, F. Zafar, Review of body area network technology & wireless
medical monitoring, Int. J. Inf. Commun. Technol. 2 (2) (2012) 186–188.
the central node. Those packets were designed to have the small- [2] T. Hayajneh, G. Almashaqbeh, S. Ullah, et al., A survey of wireless technologies
est possible number of bits, in an attempt to optimize throughput. coexistence in WBAN: analysis and open research issues, Wireless Netw. 20
In the TraPy-MAC protocol, the RF packets are fixed and equal to (8) (2014) 2165–2199.
[3] J.Y. Khan, M.R. Yuce, G. Bulger, B. Harding, Wireless body area network
127 bytes or 1016 bits. For the MAR-PC, RF packets from the central (WBAN) design techniques and performance evaluation, J. Med. Syst. 33 (2)
node to the sensor nodes have 64 bits, while the size of the RF pack- (2010) 1441–1457.
ets from the sensor node to the central node would range from 88 [4] D. Rathee, S. Rangi, S.K. Chakarvarti, V.R. Singh, Recent trends in Wireless
Body Area Network (WBAN) research and cognition based adaptive WBAN
to 280 bits (according to the volume of data available in the sensor architecture for healthcare, Health Technol. 4 (3) (2014) 239–244.
node). This strategy would make better use of the RF channel for [5] A. Malathi, P.K. Kannan, Analysis on the performance of wireless protocols,
exchanging data between all nodes of the WBAN. Int. J. Comput. Trends Technol. (IJCTT) 6 (2) (2013).
[6] N. Bradai, L.C. Fourati, L. Kamo, Investigation and performance analysis of MAC
protocols for WBAN networks, J. Netw. Comput. Appl. 46 (2014) 362–373.
5. Study limitations and future work [7] B. Latre, B. Braem, I. Moerman, C. Blondia, P. Demeester, A survey on wireless
body area networks, Wireless Netw. 17 (1) (2011) 1–18.
[8] LAN/MAN Standards Committee, 802.15.1, IEEE, (2005).
Instead of performing the experiments on a simulated environ-
[9] B. Yu, L. YangS, C. Chong, Egg Monitoring over Bluetooth: Data Compression
ment and based only on mathematical models, the authors wanted and Transmission, IEEE, 2010.
to test their strategy on a real-world system – similar to what has [10] IEEE 802.15.4, IEEE standard for Information Technology 802.15.4, Wireless
been reported by other researchers [9,11,23,28]. This has the advan- Medium Access Control (MAC) and Physical layer (PHY) specifications for
low-rate wireless personal area networks (LR-WPANs), 2006.
tage of testing the limits of the proposed MAC in situations where [11] W. Xiaoyan, N. Kiyani, M. Vidojkovic, K. Philips, H.S. de Groot, A. Masui, Wban
random artifacts, such as variable electromagnetic interference and 802. 15.6 compliant multi-band re-configurable transceiver for medical
nearby WiFi sources, are common and could have an adverse effect applications, Bodynets (2014) 50–53.
[12] LAN/MAN Standards Committee, 802.15.6, IEEE, (2012).
on data communication. The results show that, in such a scenario, [13] S. Ullah, M. Chen, K.S. Kwak, Throughput and delay analysis of IEEE
the MAR-PC can provide a robust solution, as described earlier. 802.15.6-based CSMA/CA protocol, IEEE WCNC 36 (6) (2012) 3875–3891.
However, a more detailed mathematical model could also produce [14] T. Hayajneh, G. Almashaqbeh, S. Ullah, A.V. Vasilakos, A survey of wireless
technologies coexistence in WBAN: analysis and open research issues,
further evidence to the robustness of our MAC under different con- Wireless Netw. 20 (8) (2014) 2165–2199.
ditions, such as when the WBAN operates with a greater number [15] J.S. Choi, J.G. Kim, An improved MAC protocol for WBAN through modified
of sensor nodes or variable over-the-air data rates. Such model is frame structure, IJSH 8 (2) (2014) 65–76.
[16] O. Sharon, E. Altman, An efficient polling MAC for wireless LANs, IEEE/ACM
under development and will be the subject of future publication. Trans. Netw. 9 (4) (2001) 439–451.
Finally, for the proposed protocol, we did not include a strat- [17] S. Motoyama, A hierarchical polling-based MAC scheme for wireless body
egy to allow for reading data when blockage (due to obstacles, sensor network, Int. J. Emerg. Technol. Adv. Eng. 4 (8) (2014) 396–405.
[18] M. Ameen, A. Nessa, K.S. Kwak, QoS issues with focus on wireless body area
body movement, etc.) impedes the streaming of data from a sensor
networks, Third 2008 International Conference on Convergence and Hybrid
node to the central node, and vice-versa. In this situation, there is Information Technology (2008) 801–807.
a probability of data loss due to overflow of the sensor’s FIFO. As [19] A. Kochut, A. Vasan, A.U.A. Shankar Agrawala, Sniffing out the correct physical
such, the authors are working on a new model in which the sensor layer capture model in 802.11b, in: 12th I C Network Protocol, 2004.
[20] S. Rezvani, S.A. Ghorashi, A novel WBAN MAC protocol with improved energy
nodes, besides capturing and transmitting their data to the central consumption and data rate, KSII Trans. Internet Inf. Syst. 6 (9) (2012)
node, can also act as routers, and establish communication between 2302–2322.
neighboring hidden nodes and the central node. [21] Nordic Semiconductor, nRF24L01 Single Chip 2.4 GHz Transceiver – Product
Specification v1.0 (2010).
[22] Nordic Semicondutor, nRF24LE1 – Ultra-low Power Wireless System On-Chip
6. Conclusion Solution, v1.6 (2010).
[23] Z. Chen, C. Hu, J. Liao, S. Liu, Protocol architecture for wireless body area
network based on nRF24L01, International Conference on Automation and
A number of possible applications involving WBANs have a great Logistics (2008) 3050–3054.
demand for technologies that allow for a large quantity of biomed- [24] A.K. Jacob, G.S.L. Kishore Jacob, Contention versus polling access in IEEE
ical sensors sampling signals at different sampling rates, which 802.15.6: delay and lifetime analysis, Twenty First National Conference on
Communications (NCC) (2015).
places non-trivial demands on the medium access protocol in net- [25] A. Boulis, Y. Tselishchev, Contention vs. polling: a study in body area networks
work management. The MAR-PC protocol has met this demand by MAC design, BodyNets ‘10 Proceedings of the Fifth International Conference
incorporating a strategy for a more efficient use of the RF channel, on Body Area Networks (2010).
[26] Radio-Electronics.com, Wi-Fi/WLAN Channels, Frequencies, Bands &
through the use of a dynamic priority model for medium access and Bandwidths, 2017 (Accessed June 2 17).
for data reading by the central node. [27] G. Ragesh, K. Baskaran, A survey on futuristic health care system: wBANs,
In situations where the WBAN contains sensors operating at International Conference on Communication Technology and System Design
30 (2012) 889–896.
very different sampling rates among themselves, the MAR-PC pro-
[28] J. Ok Ha, S. Hwan Jung, M.C. Park, L. Kyung Hak, Y.S. Eo, A fully integrated
tocol presents important advantages when compared to the Polling 3–5 GHz UWB RF transceiver for WBAN applications, IEEE MTT-S
and other protocols. In our experiments, the MAR-PC was capable International Microwave Workshop Series on RF and Wireless Technologies
of sustaining the operation of the network with an average traffic for Biomedical and Healthcare Applications (IMWS-BIO) (2013).
[29] F. Ullah, A.H. Abdullah, O.Y. Kaiwartya Cao, TraPy-MAC: traffic priority aware
of 62.82% higher than that supported by Polling and at least 36% medium access control protocol for wireless body area network, Mobile
higher than that shown in other studies described in the related Wirel. Health vol. 41 (93) (2017) 1–18.
literature. [30] I. Anjum, N. Alam, M.A. Razzaque, M. Mehedi Hassan, A. Alamri, Traffic
priority and load adaptive MAC protocol for QoS provisioning in body sensor
networks, Int. J. Distrib. Sensor Netw. 2013 (2013) 1–9.
Acknowledgment [31] Z.A. Khan, M.B. Rasheed, N. Javaid, B. Robertson, Effect of inter-arrival time on
the energy consumption of beacon enabledMAC protocol for body area
networks, Proc. Comput. Sci. 32 (2014) 579–586.
We thank CAPES, CNPq and FAPEMIG, Brazil, for the financial [32] A. Nepal, A. Shin, Fast channel assignment scheme for emergency handling in
support. wireless body area networks, Sensors 16 (2017).
[33] A. Sevin, C. Bayilmis, E.I. Kirbas, Design and implementation of a new quality
of service-aware cross-layer medium access protocol for wireless body area
networks, Comput. Electr. Eng. 56 (2016) 145–156.

You might also like